diff --git a/UPDATING b/UPDATING
index e71403a..1c9f863 100644
--- a/UPDATING
+++ b/UPDATING
@@ -32,7 +32,7 @@
 Example:
 $ cd external/libvpx/armv7a
 $ ../libvpx/configure --target=armv7-android-gcc --disable-runtime-cpu-detect \
-  --disable-neon --sdk-path=$ANDROID_NDK_ROOT --disable-vp8-encoder \
+  --disable-neon --sdk-path=$ANDROID_NDK_ROOT --disable-vp9-encoder \
   --disable-examples --disable-docs
 
 Run 'make libvpx_srcs.txt'
@@ -43,5 +43,7 @@
 libvpx_srcs.txt
 vpx_config.c
 vpx_config.h
-vpx_rtcd.h
+vpx_scale_rtcd.h
+vp8_rtcd.h
+vp9_rtcd.h
 vpx_version.h
diff --git a/armv7a-neon/libvpx_srcs.txt b/armv7a-neon/libvpx_srcs.txt
index 494fe37..15973e2 100644
--- a/armv7a-neon/libvpx_srcs.txt
+++ b/armv7a-neon/libvpx_srcs.txt
@@ -63,7 +63,6 @@
 vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm.s
 vp8/common/arm/reconintra_arm.c
 vp8/common/arm/variance_arm.c
-vp8/common/asm_com_offsets.c
 vp8/common/blockd.c
 vp8/common/blockd.h
 vp8/common/coefupdateprobs.h
@@ -120,8 +119,8 @@
 vp8/common/treecoder.h
 vp8/common/variance_c.c
 vp8/common/variance.h
+vp8/common/vp8_asm_com_offsets.c
 vp8/common/vp8_entropymodedata.h
-vp8/decoder/asm_dec_offsets.c
 vp8/decoder/dboolhuff.c
 vp8/decoder/dboolhuff.h
 vp8/decoder/decodemv.c
@@ -134,6 +133,7 @@
 vp8/decoder/onyxd_int.h
 vp8/decoder/threading.c
 vp8/decoder/treereader.h
+vp8/decoder/vp8_asm_dec_offsets.c
 vp8/encoder/arm/armv5te/boolhuff_armv5te.asm.s
 vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm.s
 vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm.s
@@ -153,7 +153,6 @@
 vp8/encoder/arm/neon/vp8_mse16x16_neon.asm.s
 vp8/encoder/arm/neon/vp8_shortwalsh4x4_neon.asm.s
 vp8/encoder/arm/quantize_arm.c
-vp8/encoder/asm_enc_offsets.c
 vp8/encoder/bitstream.c
 vp8/encoder/bitstream.h
 vp8/encoder/block.h
@@ -199,12 +198,92 @@
 vp8/encoder/tokenize.h
 vp8/encoder/treewriter.c
 vp8/encoder/treewriter.h
+vp8/encoder/vp8_asm_enc_offsets.c
 vp8/vp8_common.mk
 vp8/vp8cx_arm.mk
 vp8/vp8_cx_iface.c
 vp8/vp8cx.mk
 vp8/vp8_dx_iface.c
 vp8/vp8dx.mk
+vp9/common/generic/vp9_systemdependent.c
+vp9/common/vp9_alloccommon.c
+vp9/common/vp9_alloccommon.h
+vp9/common/vp9_asm_com_offsets.c
+vp9/common/vp9_blockd.h
+vp9/common/vp9_common.h
+vp9/common/vp9_convolve.c
+vp9/common/vp9_convolve.h
+vp9/common/vp9_debugmodes.c
+vp9/common/vp9_default_coef_probs.h
+vp9/common/vp9_entropy.c
+vp9/common/vp9_entropy.h
+vp9/common/vp9_entropymode.c
+vp9/common/vp9_entropymode.h
+vp9/common/vp9_entropymv.c
+vp9/common/vp9_entropymv.h
+vp9/common/vp9_enums.h
+vp9/common/vp9_extend.c
+vp9/common/vp9_extend.h
+vp9/common/vp9_filter.c
+vp9/common/vp9_filter.h
+vp9/common/vp9_findnearmv.c
+vp9/common/vp9_findnearmv.h
+vp9/common/vp9_idct.c
+vp9/common/vp9_idct.h
+vp9/common/vp9_loopfilter.c
+vp9/common/vp9_loopfilter_filters.c
+vp9/common/vp9_loopfilter.h
+vp9/common/vp9_mbpitch.c
+vp9/common/vp9_modecont.c
+vp9/common/vp9_modecontext.c
+vp9/common/vp9_modecont.h
+vp9/common/vp9_mv.h
+vp9/common/vp9_mvref_common.c
+vp9/common/vp9_mvref_common.h
+vp9/common/vp9_onyxc_int.h
+vp9/common/vp9_onyx.h
+vp9/common/vp9_ppflags.h
+vp9/common/vp9_pragmas.h
+vp9/common/vp9_pred_common.c
+vp9/common/vp9_pred_common.h
+vp9/common/vp9_quant_common.c
+vp9/common/vp9_quant_common.h
+vp9/common/vp9_reconinter.c
+vp9/common/vp9_reconinter.h
+vp9/common/vp9_reconintra.c
+vp9/common/vp9_reconintra.h
+vp9/common/vp9_rtcd.c
+vp9/common/vp9_rtcd_defs.sh
+vp9/common/vp9_sadmxn.h
+vp9/common/vp9_seg_common.c
+vp9/common/vp9_seg_common.h
+vp9/common/vp9_subpelvar.h
+vp9/common/vp9_systemdependent.h
+vp9/common/vp9_textblit.h
+vp9/common/vp9_tile_common.c
+vp9/common/vp9_tile_common.h
+vp9/common/vp9_treecoder.c
+vp9/common/vp9_treecoder.h
+vp9/decoder/vp9_asm_dec_offsets.c
+vp9/decoder/vp9_dboolhuff.c
+vp9/decoder/vp9_dboolhuff.h
+vp9/decoder/vp9_decodemv.c
+vp9/decoder/vp9_decodemv.h
+vp9/decoder/vp9_decodframe.c
+vp9/decoder/vp9_decodframe.h
+vp9/decoder/vp9_detokenize.c
+vp9/decoder/vp9_detokenize.h
+vp9/decoder/vp9_idct_blk.c
+vp9/decoder/vp9_idct_blk.h
+vp9/decoder/vp9_onyxd.h
+vp9/decoder/vp9_onyxd_if.c
+vp9/decoder/vp9_onyxd_int.h
+vp9/decoder/vp9_read_bit_buffer.h
+vp9/decoder/vp9_treereader.h
+vp9/vp9_common.mk
+vp9/vp9_dx_iface.c
+vp9/vp9dx.mk
+vp9/vp9_iface_common.h
 vpx_config.c
 vpx/internal/vpx_codec_internal.h
 vpx_mem/include/vpx_mem_intrnl.h
@@ -214,7 +293,9 @@
 vpx_ports/arm_cpudetect.c
 vpx_ports/arm.h
 vpx_ports/asm_offsets.h
+vpx_ports/emmintrin_compat.h
 vpx_ports/mem.h
+vpx_ports/vpx_once.h
 vpx_ports/vpx_ports.mk
 vpx_ports/vpx_timer.h
 vpx_scale/arm/neon/vp8_vpxyv12_copyframe_func_neon.asm.s
@@ -223,13 +304,14 @@
 vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm.s
 vpx_scale/arm/neon/yv12extend_arm.c
 vpx_scale/generic/gen_scalers.c
-vpx_scale/generic/vpxscale.c
+vpx_scale/generic/vpx_scale.c
 vpx_scale/generic/yv12config.c
 vpx_scale/generic/yv12extend.c
-vpx_scale/generic/yv12extend_generic.h
-vpx_scale/scale_mode.h
-vpx_scale/vpxscale.h
+vpx_scale/vpx_scale_asm_offsets.c
+vpx_scale/vpx_scale.h
 vpx_scale/vpx_scale.mk
+vpx_scale/vpx_scale_rtcd.c
+vpx_scale/vpx_scale_rtcd.sh
 vpx_scale/yv12config.h
 vpx/src/vpx_codec.c
 vpx/src/vpx_decoder.c
diff --git a/armv7a-neon/vpx_rtcd.h b/armv7a-neon/vp8_rtcd.h
similarity index 83%
rename from armv7a-neon/vpx_rtcd.h
rename to armv7a-neon/vp8_rtcd.h
index 914b08d..9cad64a 100644
--- a/armv7a-neon/vpx_rtcd.h
+++ b/armv7a-neon/vp8_rtcd.h
@@ -1,5 +1,5 @@
-#ifndef VPX_RTCD_
-#define VPX_RTCD_
+#ifndef VP8_RTCD_H_
+#define VP8_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
@@ -7,7 +7,9 @@
 #define RTCD_EXTERN extern
 #endif
 
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -20,6 +22,9 @@
 union int_mv;
 struct yv12_buffer_config;
 
+void vp8_clear_system_state_c();
+#define vp8_clear_system_state vp8_clear_system_state_c
+
 void vp8_dequantize_b_c(struct blockd*, short *dqc);
 void vp8_dequantize_b_v6(struct blockd*, short *dqc);
 void vp8_dequantize_b_neon(struct blockd*, short *dqc);
@@ -119,8 +124,8 @@
 void vp8_build_intra_predictors_mbuv_s_c(struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride);
 #define vp8_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s_c
 
-void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
-void vp8_intra4x4_predict_armv6(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_armv6(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
 #define vp8_intra4x4_predict vp8_intra4x4_predict_armv6
 
 void vp8_sixtap_predict16x16_c(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch);
@@ -381,85 +386,7 @@
 int vp8_denoiser_filter_c(struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset);
 #define vp8_denoiser_filter vp8_denoiser_filter_c
 
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_4_5_scale vp8_horizontal_line_4_5_scale_c
-
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_4_5_scale vp8_vertical_band_4_5_scale_c
-
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_4_5_scale vp8_last_vertical_band_4_5_scale_c
-
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_3_scale vp8_horizontal_line_2_3_scale_c
-
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_3_scale vp8_vertical_band_2_3_scale_c
-
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_2_3_scale vp8_last_vertical_band_2_3_scale_c
-
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_5_scale vp8_horizontal_line_3_5_scale_c
-
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_5_scale vp8_vertical_band_3_5_scale_c
-
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_5_scale vp8_last_vertical_band_3_5_scale_c
-
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_4_scale vp8_horizontal_line_3_4_scale_c
-
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_4_scale vp8_vertical_band_3_4_scale_c
-
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_4_scale vp8_last_vertical_band_3_4_scale_c
-
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_1_2_scale vp8_horizontal_line_1_2_scale_c
-
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_1_2_scale vp8_vertical_band_1_2_scale_c
-
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_1_2_scale vp8_last_vertical_band_1_2_scale_c
-
-void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
-
-void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
-
-void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
-
-void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
-
-void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
-
-void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
-void vp8_yv12_extend_frame_borders_neon(struct yv12_buffer_config *ybf);
-#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_neon
-
-void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-void vp8_yv12_copy_frame_neon(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_frame vp8_yv12_copy_frame_neon
-
-void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-void vp8_yv12_copy_y_neon(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_y vp8_yv12_copy_y_neon
-
-void vpx_rtcd(void);
+void vp8_rtcd(void);
 #include "vpx_config.h"
 
 #ifdef RTCD_C
diff --git a/armv7a-neon/vp9_rtcd.h b/armv7a-neon/vp9_rtcd.h
new file mode 100644
index 0000000..cc3c834
--- /dev/null
+++ b/armv7a-neon/vp9_rtcd.h
@@ -0,0 +1,190 @@
+#ifndef VP9_RTCD_H_
+#define VP9_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_16x16 vp9_idct_add_16x16_c
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_8x8 vp9_idct_add_8x8_c
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add vp9_idct_add_c
+
+void vp9_idct_add_32x32_c(int16_t *q, uint8_t *dst, int stride, int eob);
+#define vp9_idct_add_32x32 vp9_idct_add_32x32_c
+
+void vp9_copy_mem16x16_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem16x16 vp9_copy_mem16x16_c
+
+void vp9_copy_mem8x8_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x8 vp9_copy_mem8x8_c
+
+void vp9_copy_mem8x4_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x4 vp9_copy_mem8x4_c
+
+void vp9_build_intra_predictors_c(uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available);
+#define vp9_build_intra_predictors vp9_build_intra_predictors_c
+
+void vp9_build_intra_predictors_sby_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sby_s vp9_build_intra_predictors_sby_s_c
+
+void vp9_build_intra_predictors_sbuv_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sbuv_s vp9_build_intra_predictors_sbuv_s_c
+
+void vp9_intra4x4_predict_c(struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride);
+#define vp9_intra4x4_predict vp9_intra4x4_predict_c
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_8x8 vp9_add_constant_residual_8x8_c
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_16x16 vp9_add_constant_residual_16x16_c
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_32x32 vp9_add_constant_residual_32x32_c
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_vertical_edge_w vp9_mb_lpf_vertical_edge_w_c
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_vertical_edge vp9_mbloop_filter_vertical_edge_c
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_vertical_edge vp9_loop_filter_vertical_edge_c
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_horizontal_edge_w vp9_mb_lpf_horizontal_edge_w_c
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_horizontal_edge vp9_mbloop_filter_horizontal_edge_c
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_horizontal_edge vp9_loop_filter_horizontal_edge_c
+
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_inner vp9_blend_mb_inner_c
+
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_outer vp9_blend_mb_outer_c
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_b vp9_blend_b_c
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8 vp9_convolve8_c
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_horiz vp9_convolve8_horiz_c
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_vert vp9_convolve8_vert_c
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg vp9_convolve8_avg_c
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_horiz vp9_convolve8_avg_horiz_c
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_vert vp9_convolve8_avg_vert_c
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_1_add vp9_short_idct4x4_1_add_c
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_add vp9_short_idct4x4_add_c
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct8x8_add vp9_short_idct8x8_add_c
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_8x8_add vp9_short_idct10_8x8_add_c
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_8x8 vp9_short_idct1_8x8_c
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct16x16_add vp9_short_idct16x16_add_c
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_16x16_add vp9_short_idct10_16x16_add_c
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_16x16 vp9_short_idct1_16x16_c
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct32x32_add vp9_short_idct32x32_add_c
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_32x32 vp9_short_idct1_32x32_c
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_32x32_add vp9_short_idct10_32x32_add_c
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht4x4_add vp9_short_iht4x4_add_c
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht8x8_add vp9_short_iht8x8_add_c
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *output, int pitch, int tx_type);
+#define vp9_short_iht16x16_add vp9_short_iht16x16_add_c
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output);
+#define vp9_idct4_1d vp9_idct4_1d_c
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride);
+#define vp9_dc_only_idct_add vp9_dc_only_idct_add_c
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_1_add vp9_short_iwalsh4x4_1_add_c
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_add vp9_short_iwalsh4x4_add_c
+
+unsigned int vp9_sad32x3_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad32x3 vp9_sad32x3_c
+
+unsigned int vp9_sad3x32_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad3x32 vp9_sad3x32_c
+
+void vp9_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+#include "vpx_ports/arm.h"
+static void setup_rtcd_internal(void)
+{
+    int flags = arm_cpu_caps();
+
+    (void)flags;
+
+
+}
+#endif
+#endif
diff --git a/armv7a-neon/vpx_config.c b/armv7a-neon/vpx_config.c
index 863c84c..77be6fb 100644
--- a/armv7a-neon/vpx_config.c
+++ b/armv7a-neon/vpx_config.c
@@ -5,5 +5,5 @@
 /* tree. An additional intellectual property rights grant can be found */
 /* in the file PATENTS.  All contributing project authors may */
 /* be found in the AUTHORS file in the root of the source tree. */
-static const char* const cfg = "--force-target=armv7-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-examples --disable-docs --enable-realtime-only";
+static const char* const cfg = "--target=armv7-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-vp9-encoder --disable-examples --disable-docs --enable-realtime-only";
 const char *vpx_codec_build_config(void) {return cfg;}
diff --git a/armv7a-neon/vpx_config.h b/armv7a-neon/vpx_config.h
index b3179e5..a808f7c 100644
--- a/armv7a-neon/vpx_config.h
+++ b/armv7a-neon/vpx_config.h
@@ -9,6 +9,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    
+#define INLINE      __inline__ __attribute__((always_inline))
 #define ARCH_ARM 1
 #define ARCH_MIPS 0
 #define ARCH_X86 0
@@ -34,7 +35,7 @@
 #define HAVE_SYS_MMAN_H 1
 #define HAVE_UNISTD_H 1
 #define CONFIG_EXTERNAL_BUILD 0
-#define CONFIG_INSTALL_DOCS 1
+#define CONFIG_INSTALL_DOCS 0
 #define CONFIG_INSTALL_BINS 1
 #define CONFIG_INSTALL_LIBS 1
 #define CONFIG_INSTALL_SRCS 0
@@ -61,7 +62,10 @@
 #define CONFIG_INTERNAL_STATS 0
 #define CONFIG_VP8_ENCODER 1
 #define CONFIG_VP8_DECODER 1
+#define CONFIG_VP9_ENCODER 0
+#define CONFIG_VP9_DECODER 1
 #define CONFIG_VP8 1
+#define CONFIG_VP9 1
 #define CONFIG_ENCODERS 1
 #define CONFIG_DECODERS 1
 #define CONFIG_STATIC_MSVCRT 0
@@ -77,4 +81,11 @@
 #define CONFIG_UNIT_TESTS 0
 #define CONFIG_MULTI_RES_ENCODING 0
 #define CONFIG_TEMPORAL_DENOISING 1
+#define CONFIG_EXPERIMENTAL 0
+#define CONFIG_DECRYPT 0
+#define CONFIG_ONESHOTQ 0
+#define CONFIG_MULTIPLE_ARF 0
+#define CONFIG_NON420 0
+#define CONFIG_ALPHA 0
+#define CONFIG_BALANCED_COEFTREE 0
 #endif /* VPX_CONFIG_H */
diff --git a/armv7a-neon/vpx_scale_rtcd.h b/armv7a-neon/vpx_scale_rtcd.h
new file mode 100644
index 0000000..ed84626
--- /dev/null
+++ b/armv7a-neon/vpx_scale_rtcd.h
@@ -0,0 +1,62 @@
+#ifndef VPX_SCALE_RTCD_H_
+#define VPX_SCALE_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+struct yv12_buffer_config;
+
+void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
+
+void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
+
+void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
+
+void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
+
+void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
+
+void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
+void vp8_yv12_extend_frame_borders_neon(struct yv12_buffer_config *ybf);
+#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_neon
+
+void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+void vp8_yv12_copy_frame_neon(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_frame vp8_yv12_copy_frame_neon
+
+void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+void vp8_yv12_copy_y_neon(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_y vp8_yv12_copy_y_neon
+
+void vp9_extend_frame_borders_c(struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y);
+#define vp9_extend_frame_borders vp9_extend_frame_borders_c
+
+void vpx_scale_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+#include "vpx_ports/arm.h"
+static void setup_rtcd_internal(void)
+{
+    int flags = arm_cpu_caps();
+
+    (void)flags;
+
+
+}
+#endif
+#endif
diff --git a/armv7a-neon/vpx_version.h b/armv7a-neon/vpx_version.h
index 663dd49..512851c 100644
--- a/armv7a-neon/vpx_version.h
+++ b/armv7a-neon/vpx_version.h
@@ -1,7 +1,7 @@
 #define VERSION_MAJOR  1
-#define VERSION_MINOR  1
+#define VERSION_MINOR  2
 #define VERSION_PATCH  0
 #define VERSION_EXTRA  ""
 #define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH))
-#define VERSION_STRING_NOSP "v1.1.0"
-#define VERSION_STRING      " v1.1.0"
+#define VERSION_STRING_NOSP "v1.2.0"
+#define VERSION_STRING      " v1.2.0"
diff --git a/armv7a/libvpx_srcs.txt b/armv7a/libvpx_srcs.txt
index e63834d..bab4901 100644
--- a/armv7a/libvpx_srcs.txt
+++ b/armv7a/libvpx_srcs.txt
@@ -32,7 +32,6 @@
 vp8/common/arm/loopfilter_arm.c
 vp8/common/arm/reconintra_arm.c
 vp8/common/arm/variance_arm.c
-vp8/common/asm_com_offsets.c
 vp8/common/blockd.c
 vp8/common/blockd.h
 vp8/common/coefupdateprobs.h
@@ -89,8 +88,8 @@
 vp8/common/treecoder.h
 vp8/common/variance_c.c
 vp8/common/variance.h
+vp8/common/vp8_asm_com_offsets.c
 vp8/common/vp8_entropymodedata.h
-vp8/decoder/asm_dec_offsets.c
 vp8/decoder/dboolhuff.c
 vp8/decoder/dboolhuff.h
 vp8/decoder/decodemv.c
@@ -103,6 +102,7 @@
 vp8/decoder/onyxd_int.h
 vp8/decoder/threading.c
 vp8/decoder/treereader.h
+vp8/decoder/vp8_asm_dec_offsets.c
 vp8/encoder/arm/armv5te/boolhuff_armv5te.asm.s
 vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm.s
 vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm.s
@@ -115,7 +115,6 @@
 vp8/encoder/arm/boolhuff_arm.c
 vp8/encoder/arm/dct_arm.c
 vp8/encoder/arm/quantize_arm.c
-vp8/encoder/asm_enc_offsets.c
 vp8/encoder/bitstream.c
 vp8/encoder/bitstream.h
 vp8/encoder/block.h
@@ -161,12 +160,92 @@
 vp8/encoder/tokenize.h
 vp8/encoder/treewriter.c
 vp8/encoder/treewriter.h
+vp8/encoder/vp8_asm_enc_offsets.c
 vp8/vp8_common.mk
 vp8/vp8cx_arm.mk
 vp8/vp8_cx_iface.c
 vp8/vp8cx.mk
 vp8/vp8_dx_iface.c
 vp8/vp8dx.mk
+vp9/common/generic/vp9_systemdependent.c
+vp9/common/vp9_alloccommon.c
+vp9/common/vp9_alloccommon.h
+vp9/common/vp9_asm_com_offsets.c
+vp9/common/vp9_blockd.h
+vp9/common/vp9_common.h
+vp9/common/vp9_convolve.c
+vp9/common/vp9_convolve.h
+vp9/common/vp9_debugmodes.c
+vp9/common/vp9_default_coef_probs.h
+vp9/common/vp9_entropy.c
+vp9/common/vp9_entropy.h
+vp9/common/vp9_entropymode.c
+vp9/common/vp9_entropymode.h
+vp9/common/vp9_entropymv.c
+vp9/common/vp9_entropymv.h
+vp9/common/vp9_enums.h
+vp9/common/vp9_extend.c
+vp9/common/vp9_extend.h
+vp9/common/vp9_filter.c
+vp9/common/vp9_filter.h
+vp9/common/vp9_findnearmv.c
+vp9/common/vp9_findnearmv.h
+vp9/common/vp9_idct.c
+vp9/common/vp9_idct.h
+vp9/common/vp9_loopfilter.c
+vp9/common/vp9_loopfilter_filters.c
+vp9/common/vp9_loopfilter.h
+vp9/common/vp9_mbpitch.c
+vp9/common/vp9_modecont.c
+vp9/common/vp9_modecontext.c
+vp9/common/vp9_modecont.h
+vp9/common/vp9_mv.h
+vp9/common/vp9_mvref_common.c
+vp9/common/vp9_mvref_common.h
+vp9/common/vp9_onyxc_int.h
+vp9/common/vp9_onyx.h
+vp9/common/vp9_ppflags.h
+vp9/common/vp9_pragmas.h
+vp9/common/vp9_pred_common.c
+vp9/common/vp9_pred_common.h
+vp9/common/vp9_quant_common.c
+vp9/common/vp9_quant_common.h
+vp9/common/vp9_reconinter.c
+vp9/common/vp9_reconinter.h
+vp9/common/vp9_reconintra.c
+vp9/common/vp9_reconintra.h
+vp9/common/vp9_rtcd.c
+vp9/common/vp9_rtcd_defs.sh
+vp9/common/vp9_sadmxn.h
+vp9/common/vp9_seg_common.c
+vp9/common/vp9_seg_common.h
+vp9/common/vp9_subpelvar.h
+vp9/common/vp9_systemdependent.h
+vp9/common/vp9_textblit.h
+vp9/common/vp9_tile_common.c
+vp9/common/vp9_tile_common.h
+vp9/common/vp9_treecoder.c
+vp9/common/vp9_treecoder.h
+vp9/decoder/vp9_asm_dec_offsets.c
+vp9/decoder/vp9_dboolhuff.c
+vp9/decoder/vp9_dboolhuff.h
+vp9/decoder/vp9_decodemv.c
+vp9/decoder/vp9_decodemv.h
+vp9/decoder/vp9_decodframe.c
+vp9/decoder/vp9_decodframe.h
+vp9/decoder/vp9_detokenize.c
+vp9/decoder/vp9_detokenize.h
+vp9/decoder/vp9_idct_blk.c
+vp9/decoder/vp9_idct_blk.h
+vp9/decoder/vp9_onyxd.h
+vp9/decoder/vp9_onyxd_if.c
+vp9/decoder/vp9_onyxd_int.h
+vp9/decoder/vp9_read_bit_buffer.h
+vp9/decoder/vp9_treereader.h
+vp9/vp9_common.mk
+vp9/vp9_dx_iface.c
+vp9/vp9dx.mk
+vp9/vp9_iface_common.h
 vpx_config.c
 vpx/internal/vpx_codec_internal.h
 vpx_mem/include/vpx_mem_intrnl.h
@@ -176,17 +255,20 @@
 vpx_ports/arm_cpudetect.c
 vpx_ports/arm.h
 vpx_ports/asm_offsets.h
+vpx_ports/emmintrin_compat.h
 vpx_ports/mem.h
+vpx_ports/vpx_once.h
 vpx_ports/vpx_ports.mk
 vpx_ports/vpx_timer.h
 vpx_scale/generic/gen_scalers.c
-vpx_scale/generic/vpxscale.c
+vpx_scale/generic/vpx_scale.c
 vpx_scale/generic/yv12config.c
 vpx_scale/generic/yv12extend.c
-vpx_scale/generic/yv12extend_generic.h
-vpx_scale/scale_mode.h
-vpx_scale/vpxscale.h
+vpx_scale/vpx_scale_asm_offsets.c
+vpx_scale/vpx_scale.h
 vpx_scale/vpx_scale.mk
+vpx_scale/vpx_scale_rtcd.c
+vpx_scale/vpx_scale_rtcd.sh
 vpx_scale/yv12config.h
 vpx/src/vpx_codec.c
 vpx/src/vpx_decoder.c
diff --git a/armv7a/vpx_rtcd.h b/armv7a/vp8_rtcd.h
similarity index 80%
rename from armv7a/vpx_rtcd.h
rename to armv7a/vp8_rtcd.h
index 6553876..fa79b13 100644
--- a/armv7a/vpx_rtcd.h
+++ b/armv7a/vp8_rtcd.h
@@ -1,5 +1,5 @@
-#ifndef VPX_RTCD_
-#define VPX_RTCD_
+#ifndef VP8_RTCD_H_
+#define VP8_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
@@ -7,7 +7,9 @@
 #define RTCD_EXTERN extern
 #endif
 
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -20,6 +22,9 @@
 union int_mv;
 struct yv12_buffer_config;
 
+void vp8_clear_system_state_c();
+#define vp8_clear_system_state vp8_clear_system_state_c
+
 void vp8_dequantize_b_c(struct blockd*, short *dqc);
 void vp8_dequantize_b_v6(struct blockd*, short *dqc);
 #define vp8_dequantize_b vp8_dequantize_b_v6
@@ -101,8 +106,8 @@
 void vp8_build_intra_predictors_mbuv_s_c(struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride);
 #define vp8_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s_c
 
-void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
-void vp8_intra4x4_predict_armv6(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_armv6(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
 #define vp8_intra4x4_predict vp8_intra4x4_predict_armv6
 
 void vp8_sixtap_predict16x16_c(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch);
@@ -327,82 +332,7 @@
 int vp8_denoiser_filter_c(struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset);
 #define vp8_denoiser_filter vp8_denoiser_filter_c
 
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_4_5_scale vp8_horizontal_line_4_5_scale_c
-
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_4_5_scale vp8_vertical_band_4_5_scale_c
-
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_4_5_scale vp8_last_vertical_band_4_5_scale_c
-
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_3_scale vp8_horizontal_line_2_3_scale_c
-
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_3_scale vp8_vertical_band_2_3_scale_c
-
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_2_3_scale vp8_last_vertical_band_2_3_scale_c
-
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_5_scale vp8_horizontal_line_3_5_scale_c
-
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_5_scale vp8_vertical_band_3_5_scale_c
-
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_5_scale vp8_last_vertical_band_3_5_scale_c
-
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_4_scale vp8_horizontal_line_3_4_scale_c
-
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_4_scale vp8_vertical_band_3_4_scale_c
-
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_4_scale vp8_last_vertical_band_3_4_scale_c
-
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_1_2_scale vp8_horizontal_line_1_2_scale_c
-
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_1_2_scale vp8_vertical_band_1_2_scale_c
-
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_1_2_scale vp8_last_vertical_band_1_2_scale_c
-
-void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
-
-void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
-
-void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
-
-void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
-
-void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
-
-void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
-#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
-
-void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
-
-void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_y vp8_yv12_copy_y_c
-
-void vpx_rtcd(void);
+void vp8_rtcd(void);
 #include "vpx_config.h"
 
 #ifdef RTCD_C
diff --git a/armv7a/vp9_rtcd.h b/armv7a/vp9_rtcd.h
new file mode 100644
index 0000000..cc3c834
--- /dev/null
+++ b/armv7a/vp9_rtcd.h
@@ -0,0 +1,190 @@
+#ifndef VP9_RTCD_H_
+#define VP9_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_16x16 vp9_idct_add_16x16_c
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_8x8 vp9_idct_add_8x8_c
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add vp9_idct_add_c
+
+void vp9_idct_add_32x32_c(int16_t *q, uint8_t *dst, int stride, int eob);
+#define vp9_idct_add_32x32 vp9_idct_add_32x32_c
+
+void vp9_copy_mem16x16_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem16x16 vp9_copy_mem16x16_c
+
+void vp9_copy_mem8x8_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x8 vp9_copy_mem8x8_c
+
+void vp9_copy_mem8x4_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x4 vp9_copy_mem8x4_c
+
+void vp9_build_intra_predictors_c(uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available);
+#define vp9_build_intra_predictors vp9_build_intra_predictors_c
+
+void vp9_build_intra_predictors_sby_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sby_s vp9_build_intra_predictors_sby_s_c
+
+void vp9_build_intra_predictors_sbuv_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sbuv_s vp9_build_intra_predictors_sbuv_s_c
+
+void vp9_intra4x4_predict_c(struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride);
+#define vp9_intra4x4_predict vp9_intra4x4_predict_c
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_8x8 vp9_add_constant_residual_8x8_c
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_16x16 vp9_add_constant_residual_16x16_c
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_32x32 vp9_add_constant_residual_32x32_c
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_vertical_edge_w vp9_mb_lpf_vertical_edge_w_c
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_vertical_edge vp9_mbloop_filter_vertical_edge_c
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_vertical_edge vp9_loop_filter_vertical_edge_c
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_horizontal_edge_w vp9_mb_lpf_horizontal_edge_w_c
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_horizontal_edge vp9_mbloop_filter_horizontal_edge_c
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_horizontal_edge vp9_loop_filter_horizontal_edge_c
+
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_inner vp9_blend_mb_inner_c
+
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_outer vp9_blend_mb_outer_c
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_b vp9_blend_b_c
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8 vp9_convolve8_c
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_horiz vp9_convolve8_horiz_c
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_vert vp9_convolve8_vert_c
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg vp9_convolve8_avg_c
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_horiz vp9_convolve8_avg_horiz_c
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_vert vp9_convolve8_avg_vert_c
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_1_add vp9_short_idct4x4_1_add_c
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_add vp9_short_idct4x4_add_c
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct8x8_add vp9_short_idct8x8_add_c
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_8x8_add vp9_short_idct10_8x8_add_c
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_8x8 vp9_short_idct1_8x8_c
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct16x16_add vp9_short_idct16x16_add_c
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_16x16_add vp9_short_idct10_16x16_add_c
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_16x16 vp9_short_idct1_16x16_c
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct32x32_add vp9_short_idct32x32_add_c
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_32x32 vp9_short_idct1_32x32_c
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_32x32_add vp9_short_idct10_32x32_add_c
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht4x4_add vp9_short_iht4x4_add_c
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht8x8_add vp9_short_iht8x8_add_c
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *output, int pitch, int tx_type);
+#define vp9_short_iht16x16_add vp9_short_iht16x16_add_c
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output);
+#define vp9_idct4_1d vp9_idct4_1d_c
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride);
+#define vp9_dc_only_idct_add vp9_dc_only_idct_add_c
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_1_add vp9_short_iwalsh4x4_1_add_c
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_add vp9_short_iwalsh4x4_add_c
+
+unsigned int vp9_sad32x3_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad32x3 vp9_sad32x3_c
+
+unsigned int vp9_sad3x32_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad3x32 vp9_sad3x32_c
+
+void vp9_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+#include "vpx_ports/arm.h"
+static void setup_rtcd_internal(void)
+{
+    int flags = arm_cpu_caps();
+
+    (void)flags;
+
+
+}
+#endif
+#endif
diff --git a/armv7a/vpx_config.c b/armv7a/vpx_config.c
index 559f9b0..a246c39 100644
--- a/armv7a/vpx_config.c
+++ b/armv7a/vpx_config.c
@@ -5,5 +5,5 @@
 /* tree. An additional intellectual property rights grant can be found */
 /* in the file PATENTS.  All contributing project authors may */
 /* be found in the AUTHORS file in the root of the source tree. */
-static const char* const cfg = "--force-target=armv7-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-examples --disable-docs --disable-neon --enable-realtime-only";
+static const char* const cfg = "--target=armv7-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-vp9-encoder --disable-neon --disable-examples --disable-docs --enable-realtime-only";
 const char *vpx_codec_build_config(void) {return cfg;}
diff --git a/armv7a/vpx_config.h b/armv7a/vpx_config.h
index ddb331f..e04f103 100644
--- a/armv7a/vpx_config.h
+++ b/armv7a/vpx_config.h
@@ -9,6 +9,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    
+#define INLINE      __inline__ __attribute__((always_inline))
 #define ARCH_ARM 1
 #define ARCH_MIPS 0
 #define ARCH_X86 0
@@ -34,7 +35,7 @@
 #define HAVE_SYS_MMAN_H 1
 #define HAVE_UNISTD_H 1
 #define CONFIG_EXTERNAL_BUILD 0
-#define CONFIG_INSTALL_DOCS 1
+#define CONFIG_INSTALL_DOCS 0
 #define CONFIG_INSTALL_BINS 1
 #define CONFIG_INSTALL_LIBS 1
 #define CONFIG_INSTALL_SRCS 0
@@ -61,7 +62,10 @@
 #define CONFIG_INTERNAL_STATS 0
 #define CONFIG_VP8_ENCODER 1
 #define CONFIG_VP8_DECODER 1
+#define CONFIG_VP9_ENCODER 0
+#define CONFIG_VP9_DECODER 1
 #define CONFIG_VP8 1
+#define CONFIG_VP9 1
 #define CONFIG_ENCODERS 1
 #define CONFIG_DECODERS 1
 #define CONFIG_STATIC_MSVCRT 0
@@ -77,4 +81,11 @@
 #define CONFIG_UNIT_TESTS 0
 #define CONFIG_MULTI_RES_ENCODING 0
 #define CONFIG_TEMPORAL_DENOISING 1
+#define CONFIG_EXPERIMENTAL 0
+#define CONFIG_DECRYPT 0
+#define CONFIG_ONESHOTQ 0
+#define CONFIG_MULTIPLE_ARF 0
+#define CONFIG_NON420 0
+#define CONFIG_ALPHA 0
+#define CONFIG_BALANCED_COEFTREE 0
 #endif /* VPX_CONFIG_H */
diff --git a/armv7a/vpx_scale_rtcd.h b/armv7a/vpx_scale_rtcd.h
new file mode 100644
index 0000000..3f25632
--- /dev/null
+++ b/armv7a/vpx_scale_rtcd.h
@@ -0,0 +1,59 @@
+#ifndef VPX_SCALE_RTCD_H_
+#define VPX_SCALE_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+struct yv12_buffer_config;
+
+void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
+
+void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
+
+void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
+
+void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
+
+void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
+
+void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
+#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
+
+void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
+
+void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_y vp8_yv12_copy_y_c
+
+void vp9_extend_frame_borders_c(struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y);
+#define vp9_extend_frame_borders vp9_extend_frame_borders_c
+
+void vpx_scale_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+#include "vpx_ports/arm.h"
+static void setup_rtcd_internal(void)
+{
+    int flags = arm_cpu_caps();
+
+    (void)flags;
+
+
+}
+#endif
+#endif
diff --git a/armv7a/vpx_version.h b/armv7a/vpx_version.h
index 663dd49..512851c 100644
--- a/armv7a/vpx_version.h
+++ b/armv7a/vpx_version.h
@@ -1,7 +1,7 @@
 #define VERSION_MAJOR  1
-#define VERSION_MINOR  1
+#define VERSION_MINOR  2
 #define VERSION_PATCH  0
 #define VERSION_EXTRA  ""
 #define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH))
-#define VERSION_STRING_NOSP "v1.1.0"
-#define VERSION_STRING      " v1.1.0"
+#define VERSION_STRING_NOSP "v1.2.0"
+#define VERSION_STRING      " v1.2.0"
diff --git a/generic/libvpx_srcs.txt b/generic/libvpx_srcs.txt
index 5756427..8c1ec80 100644
--- a/generic/libvpx_srcs.txt
+++ b/generic/libvpx_srcs.txt
@@ -4,7 +4,6 @@
 libs.mk
 vp8/common/alloccommon.c
 vp8/common/alloccommon.h
-vp8/common/asm_com_offsets.c
 vp8/common/blockd.c
 vp8/common/blockd.h
 vp8/common/coefupdateprobs.h
@@ -61,8 +60,8 @@
 vp8/common/treecoder.h
 vp8/common/variance_c.c
 vp8/common/variance.h
+vp8/common/vp8_asm_com_offsets.c
 vp8/common/vp8_entropymodedata.h
-vp8/decoder/asm_dec_offsets.c
 vp8/decoder/dboolhuff.c
 vp8/decoder/dboolhuff.h
 vp8/decoder/decodemv.c
@@ -75,7 +74,7 @@
 vp8/decoder/onyxd_int.h
 vp8/decoder/threading.c
 vp8/decoder/treereader.h
-vp8/encoder/asm_enc_offsets.c
+vp8/decoder/vp8_asm_dec_offsets.c
 vp8/encoder/bitstream.c
 vp8/encoder/bitstream.h
 vp8/encoder/block.h
@@ -122,11 +121,91 @@
 vp8/encoder/tokenize.h
 vp8/encoder/treewriter.c
 vp8/encoder/treewriter.h
+vp8/encoder/vp8_asm_enc_offsets.c
 vp8/vp8_common.mk
 vp8/vp8_cx_iface.c
 vp8/vp8cx.mk
 vp8/vp8_dx_iface.c
 vp8/vp8dx.mk
+vp9/common/generic/vp9_systemdependent.c
+vp9/common/vp9_alloccommon.c
+vp9/common/vp9_alloccommon.h
+vp9/common/vp9_asm_com_offsets.c
+vp9/common/vp9_blockd.h
+vp9/common/vp9_common.h
+vp9/common/vp9_convolve.c
+vp9/common/vp9_convolve.h
+vp9/common/vp9_debugmodes.c
+vp9/common/vp9_default_coef_probs.h
+vp9/common/vp9_entropy.c
+vp9/common/vp9_entropy.h
+vp9/common/vp9_entropymode.c
+vp9/common/vp9_entropymode.h
+vp9/common/vp9_entropymv.c
+vp9/common/vp9_entropymv.h
+vp9/common/vp9_enums.h
+vp9/common/vp9_extend.c
+vp9/common/vp9_extend.h
+vp9/common/vp9_filter.c
+vp9/common/vp9_filter.h
+vp9/common/vp9_findnearmv.c
+vp9/common/vp9_findnearmv.h
+vp9/common/vp9_idct.c
+vp9/common/vp9_idct.h
+vp9/common/vp9_loopfilter.c
+vp9/common/vp9_loopfilter_filters.c
+vp9/common/vp9_loopfilter.h
+vp9/common/vp9_mbpitch.c
+vp9/common/vp9_modecont.c
+vp9/common/vp9_modecontext.c
+vp9/common/vp9_modecont.h
+vp9/common/vp9_mv.h
+vp9/common/vp9_mvref_common.c
+vp9/common/vp9_mvref_common.h
+vp9/common/vp9_onyxc_int.h
+vp9/common/vp9_onyx.h
+vp9/common/vp9_ppflags.h
+vp9/common/vp9_pragmas.h
+vp9/common/vp9_pred_common.c
+vp9/common/vp9_pred_common.h
+vp9/common/vp9_quant_common.c
+vp9/common/vp9_quant_common.h
+vp9/common/vp9_reconinter.c
+vp9/common/vp9_reconinter.h
+vp9/common/vp9_reconintra.c
+vp9/common/vp9_reconintra.h
+vp9/common/vp9_rtcd.c
+vp9/common/vp9_rtcd_defs.sh
+vp9/common/vp9_sadmxn.h
+vp9/common/vp9_seg_common.c
+vp9/common/vp9_seg_common.h
+vp9/common/vp9_subpelvar.h
+vp9/common/vp9_systemdependent.h
+vp9/common/vp9_textblit.h
+vp9/common/vp9_tile_common.c
+vp9/common/vp9_tile_common.h
+vp9/common/vp9_treecoder.c
+vp9/common/vp9_treecoder.h
+vp9/decoder/vp9_asm_dec_offsets.c
+vp9/decoder/vp9_dboolhuff.c
+vp9/decoder/vp9_dboolhuff.h
+vp9/decoder/vp9_decodemv.c
+vp9/decoder/vp9_decodemv.h
+vp9/decoder/vp9_decodframe.c
+vp9/decoder/vp9_decodframe.h
+vp9/decoder/vp9_detokenize.c
+vp9/decoder/vp9_detokenize.h
+vp9/decoder/vp9_idct_blk.c
+vp9/decoder/vp9_idct_blk.h
+vp9/decoder/vp9_onyxd.h
+vp9/decoder/vp9_onyxd_if.c
+vp9/decoder/vp9_onyxd_int.h
+vp9/decoder/vp9_read_bit_buffer.h
+vp9/decoder/vp9_treereader.h
+vp9/vp9_common.mk
+vp9/vp9_dx_iface.c
+vp9/vp9dx.mk
+vp9/vp9_iface_common.h
 vpx_config.c
 vpx/internal/vpx_codec_internal.h
 vpx_mem/include/vpx_mem_intrnl.h
@@ -134,17 +213,20 @@
 vpx_mem/vpx_mem.h
 vpx_mem/vpx_mem.mk
 vpx_ports/asm_offsets.h
+vpx_ports/emmintrin_compat.h
 vpx_ports/mem.h
+vpx_ports/vpx_once.h
 vpx_ports/vpx_ports.mk
 vpx_ports/vpx_timer.h
 vpx_scale/generic/gen_scalers.c
-vpx_scale/generic/vpxscale.c
+vpx_scale/generic/vpx_scale.c
 vpx_scale/generic/yv12config.c
 vpx_scale/generic/yv12extend.c
-vpx_scale/generic/yv12extend_generic.h
-vpx_scale/scale_mode.h
-vpx_scale/vpxscale.h
+vpx_scale/vpx_scale_asm_offsets.c
+vpx_scale/vpx_scale.h
 vpx_scale/vpx_scale.mk
+vpx_scale/vpx_scale_rtcd.c
+vpx_scale/vpx_scale_rtcd.sh
 vpx_scale/yv12config.h
 vpx/src/vpx_codec.c
 vpx/src/vpx_decoder.c
diff --git a/generic/vpx_rtcd.h b/generic/vp8_rtcd.h
similarity index 76%
rename from generic/vpx_rtcd.h
rename to generic/vp8_rtcd.h
index ab83978..97ef714 100644
--- a/generic/vpx_rtcd.h
+++ b/generic/vp8_rtcd.h
@@ -1,5 +1,5 @@
-#ifndef VPX_RTCD_
-#define VPX_RTCD_
+#ifndef VP8_RTCD_H_
+#define VP8_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
@@ -7,7 +7,9 @@
 #define RTCD_EXTERN extern
 #endif
 
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -20,6 +22,9 @@
 union int_mv;
 struct yv12_buffer_config;
 
+void vp8_clear_system_state_c();
+#define vp8_clear_system_state vp8_clear_system_state_c
+
 void vp8_dequantize_b_c(struct blockd*, short *dqc);
 #define vp8_dequantize_b vp8_dequantize_b_c
 
@@ -83,7 +88,7 @@
 void vp8_build_intra_predictors_mbuv_s_c(struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride);
 #define vp8_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s_c
 
-void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
 #define vp8_intra4x4_predict vp8_intra4x4_predict_c
 
 void vp8_sixtap_predict16x16_c(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch);
@@ -284,82 +289,7 @@
 int vp8_denoiser_filter_c(struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset);
 #define vp8_denoiser_filter vp8_denoiser_filter_c
 
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_4_5_scale vp8_horizontal_line_4_5_scale_c
-
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_4_5_scale vp8_vertical_band_4_5_scale_c
-
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_4_5_scale vp8_last_vertical_band_4_5_scale_c
-
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_3_scale vp8_horizontal_line_2_3_scale_c
-
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_3_scale vp8_vertical_band_2_3_scale_c
-
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_2_3_scale vp8_last_vertical_band_2_3_scale_c
-
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_5_scale vp8_horizontal_line_3_5_scale_c
-
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_5_scale vp8_vertical_band_3_5_scale_c
-
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_5_scale vp8_last_vertical_band_3_5_scale_c
-
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_4_scale vp8_horizontal_line_3_4_scale_c
-
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_4_scale vp8_vertical_band_3_4_scale_c
-
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_4_scale vp8_last_vertical_band_3_4_scale_c
-
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_1_2_scale vp8_horizontal_line_1_2_scale_c
-
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_1_2_scale vp8_vertical_band_1_2_scale_c
-
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_1_2_scale vp8_last_vertical_band_1_2_scale_c
-
-void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
-
-void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
-
-void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
-
-void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
-
-void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
-
-void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
-#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
-
-void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
-
-void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_y vp8_yv12_copy_y_c
-
-void vpx_rtcd(void);
+void vp8_rtcd(void);
 #include "vpx_config.h"
 
 #ifdef RTCD_C
diff --git a/generic/vp9_rtcd.h b/generic/vp9_rtcd.h
new file mode 100644
index 0000000..dee08d4
--- /dev/null
+++ b/generic/vp9_rtcd.h
@@ -0,0 +1,185 @@
+#ifndef VP9_RTCD_H_
+#define VP9_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_16x16 vp9_idct_add_16x16_c
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_8x8 vp9_idct_add_8x8_c
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add vp9_idct_add_c
+
+void vp9_idct_add_32x32_c(int16_t *q, uint8_t *dst, int stride, int eob);
+#define vp9_idct_add_32x32 vp9_idct_add_32x32_c
+
+void vp9_copy_mem16x16_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem16x16 vp9_copy_mem16x16_c
+
+void vp9_copy_mem8x8_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x8 vp9_copy_mem8x8_c
+
+void vp9_copy_mem8x4_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x4 vp9_copy_mem8x4_c
+
+void vp9_build_intra_predictors_c(uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available);
+#define vp9_build_intra_predictors vp9_build_intra_predictors_c
+
+void vp9_build_intra_predictors_sby_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sby_s vp9_build_intra_predictors_sby_s_c
+
+void vp9_build_intra_predictors_sbuv_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sbuv_s vp9_build_intra_predictors_sbuv_s_c
+
+void vp9_intra4x4_predict_c(struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride);
+#define vp9_intra4x4_predict vp9_intra4x4_predict_c
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_8x8 vp9_add_constant_residual_8x8_c
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_16x16 vp9_add_constant_residual_16x16_c
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_32x32 vp9_add_constant_residual_32x32_c
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_vertical_edge_w vp9_mb_lpf_vertical_edge_w_c
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_vertical_edge vp9_mbloop_filter_vertical_edge_c
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_vertical_edge vp9_loop_filter_vertical_edge_c
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_horizontal_edge_w vp9_mb_lpf_horizontal_edge_w_c
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_horizontal_edge vp9_mbloop_filter_horizontal_edge_c
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_horizontal_edge vp9_loop_filter_horizontal_edge_c
+
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_inner vp9_blend_mb_inner_c
+
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_outer vp9_blend_mb_outer_c
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_b vp9_blend_b_c
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8 vp9_convolve8_c
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_horiz vp9_convolve8_horiz_c
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_vert vp9_convolve8_vert_c
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg vp9_convolve8_avg_c
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_horiz vp9_convolve8_avg_horiz_c
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_vert vp9_convolve8_avg_vert_c
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_1_add vp9_short_idct4x4_1_add_c
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_add vp9_short_idct4x4_add_c
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct8x8_add vp9_short_idct8x8_add_c
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_8x8_add vp9_short_idct10_8x8_add_c
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_8x8 vp9_short_idct1_8x8_c
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct16x16_add vp9_short_idct16x16_add_c
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_16x16_add vp9_short_idct10_16x16_add_c
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_16x16 vp9_short_idct1_16x16_c
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct32x32_add vp9_short_idct32x32_add_c
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_32x32 vp9_short_idct1_32x32_c
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_32x32_add vp9_short_idct10_32x32_add_c
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht4x4_add vp9_short_iht4x4_add_c
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht8x8_add vp9_short_iht8x8_add_c
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *output, int pitch, int tx_type);
+#define vp9_short_iht16x16_add vp9_short_iht16x16_add_c
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output);
+#define vp9_idct4_1d vp9_idct4_1d_c
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride);
+#define vp9_dc_only_idct_add vp9_dc_only_idct_add_c
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_1_add vp9_short_iwalsh4x4_1_add_c
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_add vp9_short_iwalsh4x4_add_c
+
+unsigned int vp9_sad32x3_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad32x3 vp9_sad32x3_c
+
+unsigned int vp9_sad3x32_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad3x32 vp9_sad3x32_c
+
+void vp9_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+}
+#endif
+#endif
diff --git a/generic/vpx_config.c b/generic/vpx_config.c
index 61f4fd8..c87cb91 100644
--- a/generic/vpx_config.c
+++ b/generic/vpx_config.c
@@ -5,5 +5,5 @@
 /* tree. An additional intellectual property rights grant can be found */
 /* in the file PATENTS.  All contributing project authors may */
 /* be found in the AUTHORS file in the root of the source tree. */
-static const char* const cfg = "--force-target=generic-gnu --disable-examples --disable-docs --enable-realtime-only";
+static const char* const cfg = "--force-target=generic-gnu --disable-vp9-encoder --disable-examples --disable-docs --enable-realtime-only";
 const char *vpx_codec_build_config(void) {return cfg;}
diff --git a/generic/vpx_config.h b/generic/vpx_config.h
index 9c6da62..44e6842 100644
--- a/generic/vpx_config.h
+++ b/generic/vpx_config.h
@@ -9,6 +9,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    
+#define INLINE      __inline__ __attribute__((always_inline))
 #define ARCH_ARM 0
 #define ARCH_MIPS 0
 #define ARCH_X86 0
@@ -34,7 +35,7 @@
 #define HAVE_SYS_MMAN_H 1
 #define HAVE_UNISTD_H 1
 #define CONFIG_EXTERNAL_BUILD 0
-#define CONFIG_INSTALL_DOCS 1
+#define CONFIG_INSTALL_DOCS 0
 #define CONFIG_INSTALL_BINS 1
 #define CONFIG_INSTALL_LIBS 1
 #define CONFIG_INSTALL_SRCS 0
@@ -61,7 +62,10 @@
 #define CONFIG_INTERNAL_STATS 0
 #define CONFIG_VP8_ENCODER 1
 #define CONFIG_VP8_DECODER 1
+#define CONFIG_VP9_ENCODER 0
+#define CONFIG_VP9_DECODER 1
 #define CONFIG_VP8 1
+#define CONFIG_VP9 1
 #define CONFIG_ENCODERS 1
 #define CONFIG_DECODERS 1
 #define CONFIG_STATIC_MSVCRT 0
@@ -77,4 +81,11 @@
 #define CONFIG_UNIT_TESTS 1
 #define CONFIG_MULTI_RES_ENCODING 0
 #define CONFIG_TEMPORAL_DENOISING 1
+#define CONFIG_EXPERIMENTAL 0
+#define CONFIG_DECRYPT 0
+#define CONFIG_ONESHOTQ 0
+#define CONFIG_MULTIPLE_ARF 0
+#define CONFIG_NON420 0
+#define CONFIG_ALPHA 0
+#define CONFIG_BALANCED_COEFTREE 0
 #endif /* VPX_CONFIG_H */
diff --git a/generic/vpx_scale_rtcd.h b/generic/vpx_scale_rtcd.h
new file mode 100644
index 0000000..3a1db05
--- /dev/null
+++ b/generic/vpx_scale_rtcd.h
@@ -0,0 +1,54 @@
+#ifndef VPX_SCALE_RTCD_H_
+#define VPX_SCALE_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+struct yv12_buffer_config;
+
+void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
+
+void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
+
+void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
+
+void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
+
+void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
+
+void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
+#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
+
+void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
+
+void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_y vp8_yv12_copy_y_c
+
+void vp9_extend_frame_borders_c(struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y);
+#define vp9_extend_frame_borders vp9_extend_frame_borders_c
+
+void vpx_scale_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+}
+#endif
+#endif
diff --git a/generic/vpx_version.h b/generic/vpx_version.h
index 663dd49..512851c 100644
--- a/generic/vpx_version.h
+++ b/generic/vpx_version.h
@@ -1,7 +1,7 @@
 #define VERSION_MAJOR  1
-#define VERSION_MINOR  1
+#define VERSION_MINOR  2
 #define VERSION_PATCH  0
 #define VERSION_EXTRA  ""
 #define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH))
-#define VERSION_STRING_NOSP "v1.1.0"
-#define VERSION_STRING      " v1.1.0"
+#define VERSION_STRING_NOSP "v1.2.0"
+#define VERSION_STRING      " v1.2.0"
diff --git a/libvpx.mk b/libvpx.mk
index 5cb7820..197ed75 100644
--- a/libvpx.mk
+++ b/libvpx.mk
@@ -60,14 +60,16 @@
 # used yet but are included in the comments for future reference.
 
 libvpx_asm_offsets_intermediates := \
-    vp8/common/asm_com_offsets.intermediate \
-    vp8/decoder/asm_dec_offsets.intermediate \
-    vp8/encoder/asm_enc_offsets.intermediate \
+    vp8/common/vp8_asm_com_offsets.intermediate \
+    vp8/decoder/vp8_asm_dec_offsets.intermediate \
+    vp8/encoder/vp8_asm_enc_offsets.intermediate \
+    vpx_scale/vpx_scale_asm_offsets.intermediate \
 
 libvpx_asm_offsets_files := \
-    vp8/common/asm_com_offsets.asm \
-    vp8/decoder/asm_dec_offsets.asm \
-    vp8/encoder/asm_enc_offsets.asm \
+    vp8/common/vp8_asm_com_offsets.asm \
+    vp8/decoder/vp8_asm_dec_offsets.asm \
+    vp8/encoder/vp8_asm_enc_offsets.asm \
+    vpx_scale/vpx_scale_asm_offsets.asm \
 
 # Build the S files with inline assembly.
 COMPILE_TO_S := $(addprefix $(libvpx_intermediates)/, $(libvpx_asm_offsets_intermediates))
@@ -107,6 +109,7 @@
     $(libvpx_intermediates)/vp8/common \
     $(libvpx_intermediates)/vp8/decoder \
     $(libvpx_intermediates)/vp8/encoder \
+    $(libvpx_intermediates)/vpx_scale \
 
 libvpx_target :=
 libvpx_asm :=
diff --git a/libvpx/CHANGELOG b/libvpx/CHANGELOG
index dcb9f73..ef64a96 100644
--- a/libvpx/CHANGELOG
+++ b/libvpx/CHANGELOG
@@ -1,3 +1,32 @@
+2012-12-21 v1.2.0
+  This release acts as a checkpoint for a large amount of internal refactoring
+  and testing. It also contains a number of small bugfixes, so all users are
+  encouraged to upgrade.
+
+  - Upgrading:
+    This release is ABI and API compatible with Duclair (v1.0.0). Users
+    of older releases should refer to the Upgrading notes in this
+    document for that release.
+
+  - Enhancements:
+      VP8 optimizations for MIPS dspr2
+      vpxenc: add -quiet option
+
+  - Speed:
+      Encoder and decoder speed is consistent with the Eider release.
+
+  - Quality:
+      In general, quality is consistent with the Eider release.
+
+      Minor tweaks to ARNR filtering
+      Minor improvements to real time encoding with multiple temporal layers
+
+  - Bug Fixes:
+      Fixes multithreaded encoder race condition in loopfilter
+      Fixes multi-resolution threaded encoding
+      Fix potential encoder dead-lock after picture resize
+
+
 2012-05-09 v1.1.0 "Eider"
   This introduces a number of enhancements, mostly focused on real-time
   encoding. In addition, it fixes a decoder bug (first introduced in
diff --git a/libvpx/args.c b/libvpx/args.c
index 37ba778..9dabc9b 100644
--- a/libvpx/args.c
+++ b/libvpx/args.c
@@ -25,241 +25,214 @@
 #endif
 
 
-struct arg arg_init(char **argv)
-{
-    struct arg a;
+struct arg arg_init(char **argv) {
+  struct arg a;
 
-    a.argv      = argv;
-    a.argv_step = 1;
-    a.name      = NULL;
-    a.val       = NULL;
-    a.def       = NULL;
-    return a;
+  a.argv      = argv;
+  a.argv_step = 1;
+  a.name      = NULL;
+  a.val       = NULL;
+  a.def       = NULL;
+  return a;
 }
 
-int arg_match(struct arg *arg_, const struct arg_def *def, char **argv)
-{
-    struct arg arg;
+int arg_match(struct arg *arg_, const struct arg_def *def, char **argv) {
+  struct arg arg;
 
-    if (!argv[0] || argv[0][0] != '-')
-        return 0;
+  if (!argv[0] || argv[0][0] != '-')
+    return 0;
 
-    arg = arg_init(argv);
+  arg = arg_init(argv);
 
-    if (def->short_name
-        && strlen(arg.argv[0]) == strlen(def->short_name) + 1
-        && !strcmp(arg.argv[0] + 1, def->short_name))
-    {
+  if (def->short_name
+      && strlen(arg.argv[0]) == strlen(def->short_name) + 1
+      && !strcmp(arg.argv[0] + 1, def->short_name)) {
 
-        arg.name = arg.argv[0] + 1;
-        arg.val = def->has_val ? arg.argv[1] : NULL;
-        arg.argv_step = def->has_val ? 2 : 1;
+    arg.name = arg.argv[0] + 1;
+    arg.val = def->has_val ? arg.argv[1] : NULL;
+    arg.argv_step = def->has_val ? 2 : 1;
+  } else if (def->long_name) {
+    const size_t name_len = strlen(def->long_name);
+
+    if (strlen(arg.argv[0]) >= name_len + 2
+        && arg.argv[0][1] == '-'
+        && !strncmp(arg.argv[0] + 2, def->long_name, name_len)
+        && (arg.argv[0][name_len + 2] == '='
+            || arg.argv[0][name_len + 2] == '\0')) {
+
+      arg.name = arg.argv[0] + 2;
+      arg.val = arg.name[name_len] == '=' ? arg.name + name_len + 1 : NULL;
+      arg.argv_step = 1;
     }
+  }
+
+  if (arg.name && !arg.val && def->has_val)
+    die("Error: option %s requires argument.\n", arg.name);
+
+  if (arg.name && arg.val && !def->has_val)
+    die("Error: option %s requires no argument.\n", arg.name);
+
+  if (arg.name
+      && (arg.val || !def->has_val)) {
+    arg.def = def;
+    *arg_ = arg;
+    return 1;
+  }
+
+  return 0;
+}
+
+
+const char *arg_next(struct arg *arg) {
+  if (arg->argv[0])
+    arg->argv += arg->argv_step;
+
+  return *arg->argv;
+}
+
+
+char **argv_dup(int argc, const char **argv) {
+  char **new_argv = malloc((argc + 1) * sizeof(*argv));
+
+  memcpy(new_argv, argv, argc * sizeof(*argv));
+  new_argv[argc] = NULL;
+  return new_argv;
+}
+
+
+void arg_show_usage(FILE *fp, const struct arg_def *const *defs) {
+  char option_text[40] = {0};
+
+  for (; *defs; defs++) {
+    const struct arg_def *def = *defs;
+    char *short_val = def->has_val ? " <arg>" : "";
+    char *long_val = def->has_val ? "=<arg>" : "";
+
+    if (def->short_name && def->long_name) {
+      char *comma = def->has_val ? "," : ",      ";
+
+      snprintf(option_text, 37, "-%s%s%s --%s%6s",
+               def->short_name, short_val, comma,
+               def->long_name, long_val);
+    } else if (def->short_name)
+      snprintf(option_text, 37, "-%s%s",
+               def->short_name, short_val);
     else if (def->long_name)
-    {
-        const size_t name_len = strlen(def->long_name);
+      snprintf(option_text, 37, "          --%s%s",
+               def->long_name, long_val);
 
-        if (strlen(arg.argv[0]) >= name_len + 2
-            && arg.argv[0][1] == '-'
-            && !strncmp(arg.argv[0] + 2, def->long_name, name_len)
-            && (arg.argv[0][name_len+2] == '='
-                || arg.argv[0][name_len+2] == '\0'))
-        {
+    fprintf(fp, "  %-37s\t%s\n", option_text, def->desc);
 
-            arg.name = arg.argv[0] + 2;
-            arg.val = arg.name[name_len] == '=' ? arg.name + name_len + 1 : NULL;
-            arg.argv_step = 1;
-        }
+    if (def->enums) {
+      const struct arg_enum_list *listptr;
+
+      fprintf(fp, "  %-37s\t  ", "");
+
+      for (listptr = def->enums; listptr->name; listptr++)
+        fprintf(fp, "%s%s", listptr->name,
+                listptr[1].name ? ", " : "\n");
     }
-
-    if (arg.name && !arg.val && def->has_val)
-        die("Error: option %s requires argument.\n", arg.name);
-
-    if (arg.name && arg.val && !def->has_val)
-        die("Error: option %s requires no argument.\n", arg.name);
-
-    if (arg.name
-        && (arg.val || !def->has_val))
-    {
-        arg.def = def;
-        *arg_ = arg;
-        return 1;
-    }
-
-    return 0;
+  }
 }
 
 
-const char *arg_next(struct arg *arg)
-{
-    if (arg->argv[0])
-        arg->argv += arg->argv_step;
+unsigned int arg_parse_uint(const struct arg *arg) {
+  long int   rawval;
+  char      *endptr;
 
-    return *arg->argv;
+  rawval = strtol(arg->val, &endptr, 10);
+
+  if (arg->val[0] != '\0' && endptr[0] == '\0') {
+    if (rawval >= 0 && rawval <= UINT_MAX)
+      return rawval;
+
+    die("Option %s: Value %ld out of range for unsigned int\n",
+        arg->name, rawval);
+  }
+
+  die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
+  return 0;
 }
 
 
-char **argv_dup(int argc, const char **argv)
-{
-    char **new_argv = malloc((argc + 1) * sizeof(*argv));
+int arg_parse_int(const struct arg *arg) {
+  long int   rawval;
+  char      *endptr;
 
-    memcpy(new_argv, argv, argc * sizeof(*argv));
-    new_argv[argc] = NULL;
-    return new_argv;
+  rawval = strtol(arg->val, &endptr, 10);
+
+  if (arg->val[0] != '\0' && endptr[0] == '\0') {
+    if (rawval >= INT_MIN && rawval <= INT_MAX)
+      return rawval;
+
+    die("Option %s: Value %ld out of range for signed int\n",
+        arg->name, rawval);
+  }
+
+  die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
+  return 0;
 }
 
 
-void arg_show_usage(FILE *fp, const struct arg_def *const *defs)
-{
-    char option_text[40] = {0};
-
-    for (; *defs; defs++)
-    {
-        const struct arg_def *def = *defs;
-        char *short_val = def->has_val ? " <arg>" : "";
-        char *long_val = def->has_val ? "=<arg>" : "";
-
-        if (def->short_name && def->long_name)
-        {
-            char *comma = def->has_val ? "," : ",      ";
-
-            snprintf(option_text, 37, "-%s%s%s --%s%6s",
-                     def->short_name, short_val, comma,
-                     def->long_name, long_val);
-        }
-        else if (def->short_name)
-            snprintf(option_text, 37, "-%s%s",
-                     def->short_name, short_val);
-        else if (def->long_name)
-            snprintf(option_text, 37, "          --%s%s",
-                     def->long_name, long_val);
-
-        fprintf(fp, "  %-37s\t%s\n", option_text, def->desc);
-
-        if(def->enums)
-        {
-            const struct arg_enum_list *listptr;
-
-            fprintf(fp, "  %-37s\t  ", "");
-
-            for(listptr = def->enums; listptr->name; listptr++)
-                fprintf(fp, "%s%s", listptr->name,
-                        listptr[1].name ? ", " : "\n");
-        }
-    }
-}
-
-
-unsigned int arg_parse_uint(const struct arg *arg)
-{
-    long int   rawval;
-    char      *endptr;
-
-    rawval = strtol(arg->val, &endptr, 10);
-
-    if (arg->val[0] != '\0' && endptr[0] == '\0')
-    {
-        if (rawval >= 0 && rawval <= UINT_MAX)
-            return rawval;
-
-        die("Option %s: Value %ld out of range for unsigned int\n",
-            arg->name, rawval);
-    }
-
-    die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
-    return 0;
-}
-
-
-int arg_parse_int(const struct arg *arg)
-{
-    long int   rawval;
-    char      *endptr;
-
-    rawval = strtol(arg->val, &endptr, 10);
-
-    if (arg->val[0] != '\0' && endptr[0] == '\0')
-    {
-        if (rawval >= INT_MIN && rawval <= INT_MAX)
-            return rawval;
-
-        die("Option %s: Value %ld out of range for signed int\n",
-            arg->name, rawval);
-    }
-
-    die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
-    return 0;
-}
-
-
-struct vpx_rational
-{
-    int num; /**< fraction numerator */
-    int den; /**< fraction denominator */
+struct vpx_rational {
+  int num; /**< fraction numerator */
+  int den; /**< fraction denominator */
 };
-struct vpx_rational arg_parse_rational(const struct arg *arg)
-{
-    long int             rawval;
-    char                *endptr;
-    struct vpx_rational  rat;
+struct vpx_rational arg_parse_rational(const struct arg *arg) {
+  long int             rawval;
+  char                *endptr;
+  struct vpx_rational  rat;
 
-    /* parse numerator */
-    rawval = strtol(arg->val, &endptr, 10);
+  /* parse numerator */
+  rawval = strtol(arg->val, &endptr, 10);
 
-    if (arg->val[0] != '\0' && endptr[0] == '/')
-    {
-        if (rawval >= INT_MIN && rawval <= INT_MAX)
-            rat.num = rawval;
-        else die("Option %s: Value %ld out of range for signed int\n",
-                     arg->name, rawval);
-    }
-    else die("Option %s: Expected / at '%c'\n", arg->name, *endptr);
+  if (arg->val[0] != '\0' && endptr[0] == '/') {
+    if (rawval >= INT_MIN && rawval <= INT_MAX)
+      rat.num = rawval;
+    else die("Option %s: Value %ld out of range for signed int\n",
+               arg->name, rawval);
+  } else die("Option %s: Expected / at '%c'\n", arg->name, *endptr);
 
-    /* parse denominator */
-    rawval = strtol(endptr + 1, &endptr, 10);
+  /* parse denominator */
+  rawval = strtol(endptr + 1, &endptr, 10);
 
-    if (arg->val[0] != '\0' && endptr[0] == '\0')
-    {
-        if (rawval >= INT_MIN && rawval <= INT_MAX)
-            rat.den = rawval;
-        else die("Option %s: Value %ld out of range for signed int\n",
-                     arg->name, rawval);
-    }
-    else die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
+  if (arg->val[0] != '\0' && endptr[0] == '\0') {
+    if (rawval >= INT_MIN && rawval <= INT_MAX)
+      rat.den = rawval;
+    else die("Option %s: Value %ld out of range for signed int\n",
+               arg->name, rawval);
+  } else die("Option %s: Invalid character '%c'\n", arg->name, *endptr);
 
-    return rat;
+  return rat;
 }
 
 
-int arg_parse_enum(const struct arg *arg)
-{
-    const struct arg_enum_list *listptr;
-    long int                    rawval;
-    char                       *endptr;
+int arg_parse_enum(const struct arg *arg) {
+  const struct arg_enum_list *listptr;
+  long int                    rawval;
+  char                       *endptr;
 
-    /* First see if the value can be parsed as a raw value */
-    rawval = strtol(arg->val, &endptr, 10);
-    if (arg->val[0] != '\0' && endptr[0] == '\0')
-    {
-        /* Got a raw value, make sure it's valid */
-        for(listptr = arg->def->enums; listptr->name; listptr++)
-            if(listptr->val == rawval)
-                return rawval;
-    }
+  /* First see if the value can be parsed as a raw value */
+  rawval = strtol(arg->val, &endptr, 10);
+  if (arg->val[0] != '\0' && endptr[0] == '\0') {
+    /* Got a raw value, make sure it's valid */
+    for (listptr = arg->def->enums; listptr->name; listptr++)
+      if (listptr->val == rawval)
+        return rawval;
+  }
 
-    /* Next see if it can be parsed as a string */
-    for(listptr = arg->def->enums; listptr->name; listptr++)
-        if(!strcmp(arg->val, listptr->name))
-            return listptr->val;
+  /* Next see if it can be parsed as a string */
+  for (listptr = arg->def->enums; listptr->name; listptr++)
+    if (!strcmp(arg->val, listptr->name))
+      return listptr->val;
 
-    die("Option %s: Invalid value '%s'\n", arg->name, arg->val);
-    return 0;
+  die("Option %s: Invalid value '%s'\n", arg->name, arg->val);
+  return 0;
 }
 
 
-int arg_parse_enum_or_int(const struct arg *arg)
-{
-    if(arg->def->enums)
-        return arg_parse_enum(arg);
-    return arg_parse_int(arg);
+int arg_parse_enum_or_int(const struct arg *arg) {
+  if (arg->def->enums)
+    return arg_parse_enum(arg);
+  return arg_parse_int(arg);
 }
diff --git a/libvpx/args.h b/libvpx/args.h
index 7963fa6..ad591af 100644
--- a/libvpx/args.h
+++ b/libvpx/args.h
@@ -13,29 +13,26 @@
 #define ARGS_H
 #include <stdio.h>
 
-struct arg
-{
-    char                 **argv;
-    const char            *name;
-    const char            *val;
-    unsigned int           argv_step;
-    const struct arg_def  *def;
+struct arg {
+  char                 **argv;
+  const char            *name;
+  const char            *val;
+  unsigned int           argv_step;
+  const struct arg_def  *def;
 };
 
-struct arg_enum_list
-{
-    const char *name;
-    int         val;
+struct arg_enum_list {
+  const char *name;
+  int         val;
 };
 #define ARG_ENUM_LIST_END {0}
 
-typedef struct arg_def
-{
-    const char *short_name;
-    const char *long_name;
-    int         has_val;
-    const char *desc;
-    const struct arg_enum_list *enums;
+typedef struct arg_def {
+  const char *short_name;
+  const char *long_name;
+  int         has_val;
+  const char *desc;
+  const struct arg_enum_list *enums;
 } arg_def_t;
 #define ARG_DEF(s,l,v,d) {s,l,v,d, NULL}
 #define ARG_DEF_ENUM(s,l,v,d,e) {s,l,v,d,e}
diff --git a/libvpx/build/arm-msvs/obj_int_extract.bat b/libvpx/build/arm-msvs/obj_int_extract.bat
new file mode 100644
index 0000000..147342d
--- /dev/null
+++ b/libvpx/build/arm-msvs/obj_int_extract.bat
@@ -0,0 +1,25 @@
+REM   Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+REM
+REM   Use of this source code is governed by a BSD-style license
+REM   that can be found in the LICENSE file in the root of the source
+REM   tree. An additional intellectual property rights grant can be found
+REM   in the file PATENTS.  All contributing project authors may
+REM   be found in the AUTHORS file in the root of the source tree.
+echo on
+
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp9/common/vp9_asm_com_offsets.c"
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp9/decoder/vp9_asm_dec_offsets.c"
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp9/encoder/vp9_asm_enc_offsets.c"
+obj_int_extract.exe rvds "vp9_asm_com_offsets.obj" > "vp9_asm_com_offsets.asm"
+obj_int_extract.exe rvds "vp9_asm_dec_offsets.obj" > "vp9_asm_dec_offsets.asm"
+obj_int_extract.exe rvds "vp9_asm_enc_offsets.obj" > "vp9_asm_enc_offsets.asm"
+
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp8/common/vp8_asm_com_offsets.c"
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp8/decoder/vp8_asm_dec_offsets.c"
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vp8/encoder/vp8_asm_enc_offsets.c"
+obj_int_extract.exe rvds "vp8_asm_com_offsets.obj" > "vp8_asm_com_offsets.asm"
+obj_int_extract.exe rvds "vp8_asm_dec_offsets.obj" > "vp8_asm_dec_offsets.asm"
+obj_int_extract.exe rvds "vp8_asm_enc_offsets.obj" > "vp8_asm_enc_offsets.asm"
+
+cl /I "./" /I "%1" /nologo /c /DWINAPI_FAMILY=WINAPI_FAMILY_PHONE_APP "%1/vpx_scale/vpx_scale_asm_offsets.c"
+obj_int_extract.exe rvds "vpx_scale_asm_offsets.obj" > "vpx_scale_asm_offsets.asm"
diff --git a/libvpx/build/make/Android.mk b/libvpx/build/make/Android.mk
index d54639a..1ff0884 100644
--- a/libvpx/build/make/Android.mk
+++ b/libvpx/build/make/Android.mk
@@ -27,7 +27,7 @@
 # Android.mk file in the libvpx directory:
 # LOCAL_PATH := $(call my-dir)
 # include $(CLEAR_VARS)
-# include libvpx/build/make/Android.mk
+# include jni/libvpx/build/make/Android.mk
 #
 # There are currently two TARGET_ARCH_ABI targets for ARM.
 # armeabi and armeabi-v7a.  armeabi-v7a is selected by creating an
@@ -48,7 +48,7 @@
 # Running ndk-build will build libvpx and include it in your project.
 #
 
-CONFIG_DIR := $(LOCAL_PATH)
+CONFIG_DIR := $(LOCAL_PATH)/
 LIBVPX_PATH := $(LOCAL_PATH)/libvpx
 ASM_CNV_PATH_LOCAL := $(TARGET_ARCH_ABI)/ads2gas
 ASM_CNV_PATH := $(LOCAL_PATH)/$(ASM_CNV_PATH_LOCAL)
@@ -56,9 +56,9 @@
 # Makefiles created by the libvpx configure process
 # This will need to be fixed to handle x86.
 ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
-  include $(CONFIG_DIR)/libs-armv7-android-gcc.mk
+  include $(CONFIG_DIR)libs-armv7-android-gcc.mk
 else
-  include $(CONFIG_DIR)/libs-armv5te-android-gcc.mk
+  include $(CONFIG_DIR)libs-armv5te-android-gcc.mk
 endif
 
 # Rule that is normally in Makefile created by libvpx
@@ -106,26 +106,25 @@
 
 $(1) : $$(_OBJ) $(2)
 	@mkdir -p $$(dir $$@)
-	@grep $(OFFSET_PATTERN) $$< | tr -d '\#' | $(CONFIG_DIR)/$(ASM_CONVERSION) > $$@
+	@grep $(OFFSET_PATTERN) $$< | tr -d '\#' | $(CONFIG_DIR)$(ASM_CONVERSION) > $$@
 endef
 
 # Use ads2gas script to convert from RVCT format to GAS format.  This passes
 #  puts the processed file under $(ASM_CNV_PATH).  Local clean rule
 #  to handle removing these
-ASM_CNV_OFFSETS_DEPEND = $(ASM_CNV_PATH)/asm_com_offsets.asm
-ifeq ($(CONFIG_VP8_DECODER), yes)
-  ASM_CNV_OFFSETS_DEPEND += $(ASM_CNV_PATH)/asm_dec_offsets.asm
-endif
 ifeq ($(CONFIG_VP8_ENCODER), yes)
-  ASM_CNV_OFFSETS_DEPEND += $(ASM_CNV_PATH)/asm_enc_offsets.asm
+  ASM_CNV_OFFSETS_DEPEND += $(ASM_CNV_PATH)/vp8_asm_enc_offsets.asm
+endif
+ifeq ($(HAVE_NEON), yes)
+  ASM_CNV_OFFSETS_DEPEND += $(ASM_CNV_PATH)/vpx_scale_asm_offsets.asm
 endif
 
 .PRECIOUS: %.asm.s
 $(ASM_CNV_PATH)/libvpx/%.asm.s: $(LIBVPX_PATH)/%.asm $(ASM_CNV_OFFSETS_DEPEND)
 	@mkdir -p $(dir $@)
-	@$(CONFIG_DIR)/$(ASM_CONVERSION) <$< > $@
+	@$(CONFIG_DIR)$(ASM_CONVERSION) <$< > $@
 
-# For building vpx_rtcd.h, which has a rule in libs.mk
+# For building *_rtcd.h, which have rules in libs.mk
 TGT_ISA:=$(word 1, $(subst -, ,$(TOOLCHAIN)))
 target := libs
 
@@ -177,7 +176,14 @@
   LOCAL_STATIC_LIBRARIES := cpufeatures
 endif
 
-$(foreach file, $(LOCAL_SRC_FILES), $(LOCAL_PATH)/$(file)): vpx_rtcd.h
+# Add a dependency to force generation of the RTCD files.
+ifeq ($(CONFIG_VP8), yes)
+$(foreach file, $(LOCAL_SRC_FILES), $(LOCAL_PATH)/$(file)): vp8_rtcd.h
+endif
+ifeq ($(CONFIG_VP9), yes)
+$(foreach file, $(LOCAL_SRC_FILES), $(LOCAL_PATH)/$(file)): vp9_rtcd.h
+endif
+$(foreach file, $(LOCAL_SRC_FILES), $(LOCAL_PATH)/$(file)): vpx_scale_rtcd.h
 
 .PHONY: clean
 clean:
@@ -189,23 +195,18 @@
 
 include $(BUILD_SHARED_LIBRARY)
 
-$(eval $(call asm_offsets_template,\
-    $(ASM_CNV_PATH)/asm_com_offsets.asm, \
-    $(LIBVPX_PATH)/vp8/common/asm_com_offsets.c))
-
-ifeq ($(CONFIG_VP8_DECODER), yes)
+ifeq ($(HAVE_NEON), yes)
   $(eval $(call asm_offsets_template,\
-    $(ASM_CNV_PATH)/asm_dec_offsets.asm, \
-    $(LIBVPX_PATH)/vp8/decoder/asm_dec_offsets.c))
+    $(ASM_CNV_PATH)/vpx_scale_asm_offsets.asm, \
+    $(LIBVPX_PATH)/vpx_scale/vpx_scale_asm_offsets.c))
 endif
 
 ifeq ($(CONFIG_VP8_ENCODER), yes)
   $(eval $(call asm_offsets_template,\
-    $(ASM_CNV_PATH)/asm_enc_offsets.asm, \
-    $(LIBVPX_PATH)/vp8/encoder/asm_enc_offsets.c))
+    $(ASM_CNV_PATH)/vp8_asm_enc_offsets.asm, \
+    $(LIBVPX_PATH)/vp8/encoder/vp8_asm_enc_offsets.c))
 endif
 
 ifeq ($(CONFIG_RUNTIME_CPU_DETECT),yes)
 $(call import-module,cpufeatures)
 endif
-
diff --git a/libvpx/build/make/Makefile b/libvpx/build/make/Makefile
index 1088c84..7a25239 100644
--- a/libvpx/build/make/Makefile
+++ b/libvpx/build/make/Makefile
@@ -74,7 +74,7 @@
 TGT_ISA:=$(word 1, $(subst -, ,$(TOOLCHAIN)))
 TGT_OS:=$(word 2, $(subst -, ,$(TOOLCHAIN)))
 TGT_CC:=$(word 3, $(subst -, ,$(TOOLCHAIN)))
-quiet:=$(if $(verbose),,yes)
+quiet:=$(if $(or $(verbose), $(V)),, yes)
 qexec=$(if $(quiet),@)
 
 # Cancel built-in implicit rules
@@ -103,6 +103,18 @@
 .PHONY: testdata
 testdata::
 
+# Add compiler flags for intrinsic files
+$(BUILD_PFX)%_mmx.c.d: CFLAGS += -mmmx
+$(BUILD_PFX)%_mmx.c.o: CFLAGS += -mmmx
+$(BUILD_PFX)%_sse2.c.d: CFLAGS += -msse2
+$(BUILD_PFX)%_sse2.c.o: CFLAGS += -msse2
+$(BUILD_PFX)%_sse3.c.d: CFLAGS += -msse3
+$(BUILD_PFX)%_sse3.c.o: CFLAGS += -msse3
+$(BUILD_PFX)%_ssse3.c.d: CFLAGS += -mssse3
+$(BUILD_PFX)%_ssse3.c.o: CFLAGS += -mssse3
+$(BUILD_PFX)%_sse4.c.d: CFLAGS += -msse4.1
+$(BUILD_PFX)%_sse4.c.o: CFLAGS += -msse4.1
+
 $(BUILD_PFX)%.c.d: %.c
 	$(if $(quiet),@echo "    [DEP] $@")
 	$(qexec)mkdir -p $(dir $@)
@@ -253,10 +265,25 @@
 	$(if $(quiet),@echo "    [LD] $$@")
 	$(qexec)$$(LD) -shared $$(LDFLAGS) \
             -Wl,--no-undefined -Wl,-soname,$$(SONAME) \
-            -Wl,--version-script,$$(SO_VERSION_SCRIPT) -o $$@ \
-            $$(filter %.o,$$?) $$(extralibs)
+            -Wl,--version-script,$$(EXPORTS_FILE) -o $$@ \
+            $$(filter %.o,$$^) $$(extralibs)
 endef
 
+define dl_template
+# Not using a pattern rule here because we don't want to generate empty
+# archives when they are listed as a dependency in files not responsible
+# for creating them.
+$(1):
+	$(if $(quiet),@echo "    [LD] $$@")
+	$(qexec)$$(LD) -dynamiclib $$(LDFLAGS) \
+	    -exported_symbols_list $$(EXPORTS_FILE) \
+        -Wl,-headerpad_max_install_names,-compatibility_version,1.0,-current_version,$$(VERSION_MAJOR) \
+        -o $$@ \
+        $$(filter %.o,$$^) $$(extralibs)
+endef
+
+
+
 define lipo_lib_template
 $(1): $(addsuffix /$(1),$(FAT_ARCHS))
 	$(if $(quiet),@echo "    [LIPO] $$@")
@@ -321,6 +348,7 @@
 	@touch $@
 $(foreach lib,$(filter %_g.a,$(LIBS)),$(eval $(call archive_template,$(lib))))
 $(foreach lib,$(filter %so.$(VERSION_MAJOR).$(VERSION_MINOR).$(VERSION_PATCH),$(LIBS)),$(eval $(call so_template,$(lib))))
+$(foreach lib,$(filter %$(VERSION_MAJOR).dylib,$(LIBS)),$(eval $(call dl_template,$(lib))))
 
 INSTALL-LIBS=$(call cond_enabled,CONFIG_INSTALL_LIBS,INSTALL-LIBS)
 ifeq ($(MAKECMDGOALS),dist)
@@ -360,10 +388,14 @@
     DIST-SRCS-$(CONFIG_MSVS)  += build/make/gen_msvs_sln.sh
     DIST-SRCS-$(CONFIG_MSVS)  += build/x86-msvs/yasm.rules
     DIST-SRCS-$(CONFIG_MSVS)  += build/x86-msvs/obj_int_extract.bat
+    DIST-SRCS-$(CONFIG_MSVS)  += build/arm-msvs/obj_int_extract.bat
     DIST-SRCS-$(CONFIG_RVCT) += build/make/armlink_adapter.sh
-    # Include obj_int_extract if we use offsets from asm_*_offsets
+    # Include obj_int_extract if we use offsets from *_asm_*_offsets
     DIST-SRCS-$(ARCH_ARM)$(ARCH_X86)$(ARCH_X86_64)    += build/make/obj_int_extract.c
     DIST-SRCS-$(ARCH_ARM)    += build/make/ads2gas.pl
+    DIST-SRCS-$(ARCH_ARM)    += build/make/ads2gas_apple.pl
+    DIST-SRCS-$(ARCH_ARM)    += build/make/ads2armasm_ms.pl
+    DIST-SRCS-$(ARCH_ARM)    += build/make/thumb.pm
     DIST-SRCS-yes            += $(target:-$(TOOLCHAIN)=).mk
 endif
 INSTALL-SRCS := $(call cond_enabled,CONFIG_INSTALL_SRCS,INSTALL-SRCS)
diff --git a/libvpx/build/make/ads2armasm_ms.pl b/libvpx/build/make/ads2armasm_ms.pl
new file mode 100755
index 0000000..1def539
--- /dev/null
+++ b/libvpx/build/make/ads2armasm_ms.pl
@@ -0,0 +1,38 @@
+#!/usr/bin/perl
+##
+##  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+use FindBin;
+use lib $FindBin::Bin;
+use thumb;
+
+print "; This file was created from a .asm file\n";
+print ";  using the ads2armasm_ms.pl script.\n";
+
+while (<STDIN>)
+{
+    undef $comment;
+    undef $line;
+
+    s/REQUIRE8//;
+    s/PRESERVE8//;
+    s/^\s*ARM\s*$//;
+    s/AREA\s+\|\|(.*)\|\|/AREA |$1|/;
+    s/qsubaddx/qsax/i;
+    s/qaddsubx/qasx/i;
+
+    thumb::FixThumbInstructions($_, 1);
+
+    s/ldrneb/ldrbne/i;
+    s/ldrneh/ldrhne/i;
+
+    print;
+}
+
diff --git a/libvpx/build/make/ads2gas.pl b/libvpx/build/make/ads2gas.pl
index 95be467..9c41901 100755
--- a/libvpx/build/make/ads2gas.pl
+++ b/libvpx/build/make/ads2gas.pl
@@ -17,9 +17,24 @@
 #
 # Usage: cat inputfile | perl ads2gas.pl > outputfile
 #
+
+use FindBin;
+use lib $FindBin::Bin;
+use thumb;
+
+my $thumb = 0;
+
+foreach my $arg (@ARGV) {
+    $thumb = 1 if ($arg eq "-thumb");
+}
+
 print "@ This file was created from a .asm file\n";
 print "@  using the ads2gas.pl script.\n";
 print "\t.equ DO1STROUNDING, 0\n";
+if ($thumb) {
+    print "\t.syntax unified\n";
+    print "\t.thumb\n";
+}
 
 # Stack of procedure names.
 @proc_stack = ();
@@ -151,8 +166,13 @@
     # ALIGN directive
     s/\bALIGN\b/.balign/g;
 
-    # ARM code
-    s/\sARM/.arm/g;
+    if ($thumb) {
+        # ARM code - we force everything to thumb with the declaration in the header
+        s/\sARM//g;
+    } else {
+        # ARM code
+        s/\sARM/.arm/g;
+    }
 
     # push/pop
     s/(push\s+)(r\d+)/stmdb sp\!, \{$2\}/g;
@@ -162,6 +182,10 @@
     s/(vld1.\d+\s+)(q\d+)/$1\{$2\}/g;
     s/(vtbl.\d+\s+[^,]+),([^,]+)/$1,\{$2\}/g;
 
+    if ($thumb) {
+        thumb::FixThumbInstructions($_, 0);
+    }
+
     # eabi_attributes numerical equivalents can be found in the
     # "ARM IHI 0045C" document.
 
diff --git a/libvpx/build/make/ads2gas_apple.pl b/libvpx/build/make/ads2gas_apple.pl
index 81280bf..51e6fbc 100755
--- a/libvpx/build/make/ads2gas_apple.pl
+++ b/libvpx/build/make/ads2gas_apple.pl
@@ -10,12 +10,12 @@
 ##
 
 
-# ads2gas.pl
+# ads2gas_apple.pl
 # Author: Eric Fung (efung (at) acm.org)
 #
 # Convert ARM Developer Suite 1.0.1 syntax assembly source to GNU as format
 #
-# Usage: cat inputfile | perl ads2gas.pl > outputfile
+# Usage: cat inputfile | perl ads2gas_apple.pl > outputfile
 #
 print "@ This file was created from a .asm file\n";
 print "@  using the ads2gas_apple.pl script.\n\n";
diff --git a/libvpx/build/make/configure.sh b/libvpx/build/make/configure.sh
index 05bbabe..ee4493d 100755
--- a/libvpx/build/make/configure.sh
+++ b/libvpx/build/make/configure.sh
@@ -88,6 +88,7 @@
   ${toggle_debug}             enable/disable debug mode
   ${toggle_gprof}             enable/disable gprof profiling instrumentation
   ${toggle_gcov}              enable/disable gcov coverage instrumentation
+  ${toggle_thumb}             enable/disable building arm assembly in thumb mode
 
 Install options:
   ${toggle_install_docs}      control whether docs are installed
@@ -265,18 +266,20 @@
 fi
 TMP_H="${TMPDIRx}/vpx-conf-$$-${RANDOM}.h"
 TMP_C="${TMPDIRx}/vpx-conf-$$-${RANDOM}.c"
+TMP_CC="${TMPDIRx}/vpx-conf-$$-${RANDOM}.cc"
 TMP_O="${TMPDIRx}/vpx-conf-$$-${RANDOM}.o"
 TMP_X="${TMPDIRx}/vpx-conf-$$-${RANDOM}.x"
 TMP_ASM="${TMPDIRx}/vpx-conf-$$-${RANDOM}.asm"
 
 clean_temp_files() {
-    rm -f ${TMP_C} ${TMP_H} ${TMP_O} ${TMP_X} ${TMP_ASM}
+    rm -f ${TMP_C} ${TMP_CC} ${TMP_H} ${TMP_O} ${TMP_X} ${TMP_ASM}
 }
 
 #
 # Toolchain Check Functions
 #
 check_cmd() {
+    enabled external_build && return
     log "$@"
     "$@" >>${logfile} 2>&1
 }
@@ -290,9 +293,9 @@
 
 check_cxx() {
     log check_cxx "$@"
-    cat >${TMP_C}
-    log_file ${TMP_C}
-    check_cmd ${CXX} ${CXXFLAGS} "$@" -c -o ${TMP_O} ${TMP_C}
+    cat >${TMP_CC}
+    log_file ${TMP_CC}
+    check_cmd ${CXX} ${CXXFLAGS} "$@" -c -o ${TMP_O} ${TMP_CC}
 }
 
 check_cpp() {
@@ -414,6 +417,8 @@
 BUILD_PFX=${BUILD_PFX}
 TOOLCHAIN=${toolchain}
 ASM_CONVERSION=${asm_conversion_cmd:-${source_path}/build/make/ads2gas.pl}
+GEN_VCPROJ=${gen_vcproj_cmd}
+MSVS_ARCH_DIR=${msvs_arch_dir}
 
 CC=${CC}
 CXX=${CXX}
@@ -431,14 +436,15 @@
 extralibs = ${extralibs}
 AS_SFX    = ${AS_SFX:-.asm}
 EXE_SFX   = ${EXE_SFX}
+VCPROJ_SFX = ${VCPROJ_SFX}
 RTCD_OPTIONS = ${RTCD_OPTIONS}
 EOF
 
     if enabled rvct; then cat >> $1 << EOF
-fmt_deps = sed -e 's;^__image.axf;\$(dir \$@)\$(notdir \$<).o \$@;' #hide
+fmt_deps = sed -e 's;^__image.axf;\${@:.d=.o} \$@;' #hide
 EOF
     else cat >> $1 << EOF
-fmt_deps = sed -e 's;^\([a-zA-Z0-9_]*\)\.o;\$(dir \$@)\1\$(suffix \$<).o \$@;'
+fmt_deps = sed -e 's;^\([a-zA-Z0-9_]*\)\.o;\${@:.d=.o} \$@;'
 EOF
     fi
 
@@ -459,6 +465,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    ${RESTRICT}
+#define INLINE      ${INLINE}
 EOF
     print_config_h ARCH   "${TMP_H}" ${ARCH_LIST}
     print_config_h HAVE   "${TMP_H}" ${HAVE_LIST}
@@ -596,8 +603,13 @@
             armv6*)
                 tgt_isa=armv6
                 ;;
+            armv7*-hardfloat*)
+                tgt_isa=armv7
+                float_abi=hard
+                ;;
             armv7*)
                 tgt_isa=armv7
+                float_abi=softfp
                 ;;
             armv5te*)
                 tgt_isa=armv5te
@@ -641,6 +653,9 @@
                 tgt_isa=x86_64
                 tgt_os=darwin12
                 ;;
+            x86_64*mingw32*)
+                tgt_os=win64
+                ;;
             *mingw32*|*cygwin*)
                 [ -z "$tgt_isa" ] && tgt_isa=x86
                 tgt_os=win32
@@ -767,6 +782,7 @@
             ;;
         armv5te)
             soft_enable edsp
+            disable fast_unaligned
             ;;
         esac
 
@@ -782,8 +798,15 @@
             check_add_asflags --defsym ARCHITECTURE=${arch_int}
             tune_cflags="-mtune="
             if [ ${tgt_isa} == "armv7" ]; then
-                check_add_cflags  -march=armv7-a -mfloat-abi=softfp
-                check_add_asflags -march=armv7-a -mfloat-abi=softfp
+                if [ -z "${float_abi}" ]; then
+                    check_cpp <<EOF && float_abi=hard || float_abi=softfp
+#ifndef __ARM_PCS_VFP
+#error "not hardfp"
+#endif
+EOF
+                fi
+                check_add_cflags  -march=armv7-a -mfloat-abi=${float_abi}
+                check_add_asflags -march=armv7-a -mfloat-abi=${float_abi}
 
                 if enabled neon
                 then
@@ -801,6 +824,18 @@
 
             enabled debug && add_asflags -g
             asm_conversion_cmd="${source_path}/build/make/ads2gas.pl"
+            if enabled thumb; then
+                asm_conversion_cmd="$asm_conversion_cmd -thumb"
+                check_add_cflags -mthumb
+                check_add_asflags -mthumb -mimplicit-it=always
+            fi
+            ;;
+        vs*)
+            asm_conversion_cmd="${source_path}/build/make/ads2armasm_ms.pl"
+            AS_SFX=.s
+            msvs_arch_dir=arm-msvs
+            disable multithread
+            disable unit_tests
             ;;
         rvct)
             CC=armcc
@@ -906,7 +941,7 @@
             add_ldflags -arch_only ${tgt_isa}
 
             if [ -z "${alt_libc}" ]; then
-                alt_libc=${SDK_PATH}/SDKs/iPhoneOS5.1.sdk
+                alt_libc=${SDK_PATH}/SDKs/iPhoneOS6.0.sdk
             fi
 
             add_cflags  "-isysroot ${alt_libc}"
@@ -994,13 +1029,6 @@
 #error "not x32"
 #endif
 EOF
-        soft_enable runtime_cpu_detect
-        soft_enable mmx
-        soft_enable sse
-        soft_enable sse2
-        soft_enable sse3
-        soft_enable ssse3
-        soft_enable sse4_1
 
         case  ${tgt_os} in
             win*)
@@ -1042,18 +1070,33 @@
                 add_ldflags -m${bits}
                 link_with_cc=gcc
                 tune_cflags="-march="
-            setup_gnu_toolchain
+                setup_gnu_toolchain
                 #for 32 bit x86 builds, -O3 did not turn on this flag
-                enabled optimizations && check_add_cflags -fomit-frame-pointer
+                enabled optimizations && disabled gprof && check_add_cflags -fomit-frame-pointer
             ;;
             vs*)
                 # When building with Microsoft Visual Studio the assembler is
                 # invoked directly. Checking at configure time is unnecessary.
                 # Skip the check by setting AS arbitrarily
                 AS=msvs
+                msvs_arch_dir=x86-msvs
             ;;
         esac
 
+        soft_enable runtime_cpu_detect
+        soft_enable mmx
+        soft_enable sse
+        soft_enable sse2
+        soft_enable sse3
+        soft_enable ssse3
+        # We can't use 'check_cflags' until the compiler is configured and CC is
+        # populated.
+        if enabled gcc && ! disabled sse4_1 && ! check_cflags -msse4; then
+            RTCD_OPTIONS="${RTCD_OPTIONS}--disable-sse4_1 "
+        else
+            soft_enable sse4_1
+        fi
+
         case "${AS}" in
             auto|"")
                 which nasm >/dev/null 2>&1 && AS=nasm
@@ -1069,12 +1112,14 @@
             win32)
                 add_asflags -f win32
                 enabled debug && add_asflags -g cv8
+                EXE_SFX=.exe
             ;;
             win64)
                 add_asflags -f x64
                 enabled debug && add_asflags -g cv8
+                EXE_SFX=.exe
             ;;
-            linux*|solaris*)
+            linux*|solaris*|android*)
                 add_asflags -f elf${bits}
                 enabled debug && [ "${AS}" = yasm ] && add_asflags -g dwarf2
                 enabled debug && [ "${AS}" = nasm ] && add_asflags -g
@@ -1154,6 +1199,14 @@
     [ -f "${TMP_O}" ] && od -A n -t x1 "${TMP_O}" | tr -d '\n' |
         grep '4f *32 *42 *45' >/dev/null 2>&1 && enable big_endian
 
+    # Try to find which inline keywords are supported
+    check_cc <<EOF && INLINE="inline"
+    static inline function() {}
+EOF
+    check_cc <<EOF && INLINE="__inline__ __attribute__((always_inline))"
+    static __attribute__((always_inline)) function() {}
+EOF
+
     # Almost every platform uses pthreads.
     if enabled multithread; then
         case ${toolchain} in
@@ -1175,9 +1228,6 @@
         ;;
     esac
 
-    # for sysconf(3) and friends.
-    check_header unistd.h
-
     # glibc needs these
     if enabled linux; then
         add_cflags -D_LARGEFILE_SOURCE
diff --git a/libvpx/build/make/gen_msvs_proj.sh b/libvpx/build/make/gen_msvs_proj.sh
index 6d42941..cff27c8 100755
--- a/libvpx/build/make/gen_msvs_proj.sh
+++ b/libvpx/build/make/gen_msvs_proj.sh
@@ -26,6 +26,7 @@
     --help                      Print this message
     --exe                       Generate a project for building an Application
     --lib                       Generate a project for creating a static library
+    --dll                       Generate a project for creating a dll
     --static-crt                Use the static C runtime (/MT)
     --target=isa-os-cc          Target specifier (required)
     --out=filename              Write output to a file [stdout]
@@ -142,7 +143,9 @@
             if [ "${f##*.}" == "$pat" ]; then
                 unset file_list[i]
 
+                objf=$(echo ${f%.*}.obj | sed -e 's/^[\./]\+//g' -e 's,/,_,g')
                 open_tag File RelativePath="./$f"
+
                 if [ "$pat" == "asm" ] && $asm_use_custom_step; then
                     for plat in "${platforms[@]}"; do
                         for cfg in Debug Release; do
@@ -152,14 +155,27 @@
                             tag Tool \
                                 Name="VCCustomBuildTool" \
                                 Description="Assembling \$(InputFileName)" \
-                                CommandLine="$(eval echo \$asm_${cfg}_cmdline)" \
-                                Outputs="\$(InputName).obj" \
+                                CommandLine="$(eval echo \$asm_${cfg}_cmdline) -o \$(IntDir)$objf" \
+                                Outputs="\$(IntDir)$objf" \
 
                             close_tag FileConfiguration
                         done
                     done
                 fi
+                if [ "$pat" == "c" ] || [ "$pat" == "cc" ] ; then
+                    for plat in "${platforms[@]}"; do
+                        for cfg in Debug Release; do
+                            open_tag FileConfiguration \
+                                Name="${cfg}|${plat}" \
 
+                            tag Tool \
+                                Name="VCCLCompilerTool" \
+                                ObjectFile="\$(IntDir)$objf" \
+
+                            close_tag FileConfiguration
+                        done
+                    done
+                fi
                 close_tag File
 
                 break
@@ -190,6 +206,8 @@
         ;;
         --exe) proj_kind="exe"
         ;;
+        --dll) proj_kind="dll"
+        ;;
         --lib) proj_kind="lib"
         ;;
         --src-path-bare=*) src_path_bare="$optval"
@@ -242,10 +260,15 @@
 case "${vs_ver:-8}" in
     7) vs_ver_id="7.10"
        asm_use_custom_step=$uses_asm
+       warn_64bit='Detect64BitPortabilityProblems=true'
     ;;
     8) vs_ver_id="8.00"
+       asm_use_custom_step=$uses_asm
+       warn_64bit='Detect64BitPortabilityProblems=true'
     ;;
     9) vs_ver_id="9.00"
+       asm_use_custom_step=$uses_asm
+       warn_64bit='Detect64BitPortabilityProblems=false'
     ;;
 esac
 
@@ -284,10 +307,11 @@
 case "$target" in
     x86_64*)
         platforms[0]="x64"
+        asm_Debug_cmdline="yasm -Xvc -g cv8 -f \$(PlatformName) ${yasmincs} &quot;\$(InputPath)&quot;"
+        asm_Release_cmdline="yasm -Xvc -f \$(PlatformName) ${yasmincs} &quot;\$(InputPath)&quot;"
     ;;
     x86*)
         platforms[0]="Win32"
-        # these are only used by vs7
         asm_Debug_cmdline="yasm -Xvc -g cv8 -f \$(PlatformName) ${yasmincs} &quot;\$(InputPath)&quot;"
         asm_Release_cmdline="yasm -Xvc -f \$(PlatformName) ${yasmincs} &quot;\$(InputPath)&quot;"
     ;;
@@ -299,6 +323,8 @@
     case "$proj_kind" in
         exe) vs_ConfigurationType=1
         ;;
+        dll) vs_ConfigurationType=2
+        ;;
         *)   vs_ConfigurationType=4
         ;;
     esac
@@ -318,13 +344,6 @@
     done
     close_tag Platforms
 
-    open_tag ToolFiles
-    case "$target" in
-        x86*) $uses_asm && tag ToolFile RelativePath="$self_dirname/../x86-msvs/yasm.rules"
-        ;;
-    esac
-    close_tag ToolFiles
-
     open_tag Configurations
     for plat in "${platforms[@]}"; do
         plat_no_ws=`echo $plat | sed 's/[^A-Za-z0-9_]/_/g'`
@@ -346,8 +365,8 @@
                             PreprocessorDefinitions="WIN32;DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE" \
                             RuntimeLibrary="$debug_runtime" \
                             WarningLevel="3" \
-                            Detect64BitPortabilityProblems="true" \
                             DebugInformationFormat="1" \
+                            $warn_64bit \
                     ;;
                     vpx)
                         tag Tool \
@@ -363,7 +382,7 @@
                             UsePrecompiledHeader="0" \
                             WarningLevel="3" \
                             DebugInformationFormat="1" \
-                            Detect64BitPortabilityProblems="true" \
+                            $warn_64bit \
 
                         $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="true"
                     ;;
@@ -377,7 +396,7 @@
                             UsePrecompiledHeader="0" \
                             WarningLevel="3" \
                             DebugInformationFormat="1" \
-                            Detect64BitPortabilityProblems="true" \
+                            $warn_64bit \
 
                         $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="true"
                     ;;
@@ -454,8 +473,8 @@
                             RuntimeLibrary="$release_runtime" \
                             UsePrecompiledHeader="0" \
                             WarningLevel="3" \
-                            Detect64BitPortabilityProblems="true" \
                             DebugInformationFormat="0" \
+                            $warn_64bit \
                     ;;
                     vpx)
                         tag Tool \
@@ -472,7 +491,7 @@
                             UsePrecompiledHeader="0" \
                             WarningLevel="3" \
                             DebugInformationFormat="0" \
-                            Detect64BitPortabilityProblems="true" \
+                            $warn_64bit \
 
                         $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs"
                     ;;
@@ -487,7 +506,7 @@
                             UsePrecompiledHeader="0" \
                             WarningLevel="3" \
                             DebugInformationFormat="0" \
-                            Detect64BitPortabilityProblems="true" \
+                            $warn_64bit \
 
                         $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs"
                     ;;
diff --git a/libvpx/build/make/gen_msvs_sln.sh b/libvpx/build/make/gen_msvs_sln.sh
index 240678b..5a8c793 100755
--- a/libvpx/build/make/gen_msvs_sln.sh
+++ b/libvpx/build/make/gen_msvs_sln.sh
@@ -25,7 +25,7 @@
 Options:
     --help                      Print this message
     --out=outfile               Redirect output to a file
-    --ver=version               Version (7,8,9) of visual studio to generate for
+    --ver=version               Version (7,8,9,10,11) of visual studio to generate for
     --target=isa-os-cc          Target specifier
 EOF
     exit 1
@@ -55,14 +55,19 @@
 
 parse_project() {
     local file=$1
-    local name=`grep Name "$file" | awk 'BEGIN {FS="\""}{if (NR==1) print $2}'`
-    local guid=`grep ProjectGUID "$file" | awk 'BEGIN {FS="\""}{if (NR==1) print $2}'`
+    if [ "$sfx" = "vcproj" ]; then
+        local name=`grep Name "$file" | awk 'BEGIN {FS="\""}{if (NR==1) print $2}'`
+        local guid=`grep ProjectGUID "$file" | awk 'BEGIN {FS="\""}{if (NR==1) print $2}'`
+    else
+        local name=`grep RootNamespace "$file" | sed 's,.*<.*>\(.*\)</.*>.*,\1,'`
+        local guid=`grep ProjectGuid "$file" | sed 's,.*<.*>\(.*\)</.*>.*,\1,'`
+    fi
 
     # save the project GUID to a varaible, normalizing to the basename of the
     # vcproj file without the extension
     local var
     var=${file##*/}
-    var=${var%%.vcproj}
+    var=${var%%.${sfx}}
     eval "${var}_file=\"$1\""
     eval "${var}_name=$name"
     eval "${var}_guid=$guid"
@@ -83,14 +88,14 @@
     # vcproj file without the extension
     local var
     var=${file##*/}
-    var=${var%%.vcproj}
+    var=${var%%.${sfx}}
     eval "${var}_guid=$guid"
 
     echo "Project(\"{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}\") = \"$name\", \"$file\", \"$guid\""
     indent_push
 
     eval "local deps=\"\${${var}_deps}\""
-    if [ -n "$deps" ]; then
+    if [ -n "$deps" ] && [ "$sfx" = "vcproj" ]; then
         echo "${indent}ProjectSection(ProjectDependencies) = postProject"
         indent_push
 
@@ -221,7 +226,7 @@
     ;;
     --ver=*) vs_ver="$optval"
              case $optval in
-             [789])
+             [789]|10|11)
              ;;
              *) die Unrecognized Visual Studio Version in $opt
              ;;
@@ -257,6 +262,20 @@
     9) sln_vers="10.00"
        sln_vers_str="Visual Studio 2008"
     ;;
+    10) sln_vers="11.00"
+       sln_vers_str="Visual Studio 2010"
+    ;;
+    11) sln_vers="12.00"
+       sln_vers_str="Visual Studio 2012"
+    ;;
+esac
+case "${vs_ver:-8}" in
+    [789])
+    sfx=vcproj
+    ;;
+    10|11)
+    sfx=vcxproj
+    ;;
 esac
 
 for f in "${file_list[@]}"; do
diff --git a/libvpx/build/make/gen_msvs_vcxproj.sh b/libvpx/build/make/gen_msvs_vcxproj.sh
new file mode 100755
index 0000000..4875915
--- /dev/null
+++ b/libvpx/build/make/gen_msvs_vcxproj.sh
@@ -0,0 +1,530 @@
+#!/bin/bash
+##
+##  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+
+self=$0
+self_basename=${self##*/}
+self_dirname=$(dirname "$0")
+EOL=$'\n'
+
+show_help() {
+    cat <<EOF
+Usage: ${self_basename} --name=projname [options] file1 [file2 ...]
+
+This script generates a Visual Studio project file from a list of source
+code files.
+
+Options:
+    --help                      Print this message
+    --exe                       Generate a project for building an Application
+    --lib                       Generate a project for creating a static library
+    --dll                       Generate a project for creating a dll
+    --static-crt                Use the static C runtime (/MT)
+    --target=isa-os-cc          Target specifier (required)
+    --out=filename              Write output to a file [stdout]
+    --name=project_name         Name of the project (required)
+    --proj-guid=GUID            GUID to use for the project
+    --module-def=filename       File containing export definitions (for DLLs)
+    --ver=version               Version (10,11) of visual studio to generate for
+    --src-path-bare=dir         Path to root of source tree
+    -Ipath/to/include           Additional include directories
+    -DFLAG[=value]              Preprocessor macros to define
+    -Lpath/to/lib               Additional library search paths
+    -llibname                   Library to link against
+EOF
+    exit 1
+}
+
+die() {
+    echo "${self_basename}: $@" >&2
+    exit 1
+}
+
+die_unknown(){
+    echo "Unknown option \"$1\"." >&2
+    echo "See ${self_basename} --help for available options." >&2
+    exit 1
+}
+
+generate_uuid() {
+    local hex="0123456789ABCDEF"
+    local i
+    local uuid=""
+    local j
+    #93995380-89BD-4b04-88EB-625FBE52EBFB
+    for ((i=0; i<32; i++)); do
+        (( j = $RANDOM % 16 ))
+        uuid="${uuid}${hex:$j:1}"
+    done
+    echo "${uuid:0:8}-${uuid:8:4}-${uuid:12:4}-${uuid:16:4}-${uuid:20:12}"
+}
+
+indent1="    "
+indent=""
+indent_push() {
+    indent="${indent}${indent1}"
+}
+indent_pop() {
+    indent="${indent%${indent1}}"
+}
+
+tag_attributes() {
+    for opt in "$@"; do
+        optval="${opt#*=}"
+        [ -n "${optval}" ] ||
+            die "Missing attribute value in '$opt' while generating $tag tag"
+        echo "${indent}${opt%%=*}=\"${optval}\""
+    done
+}
+
+open_tag() {
+    local tag=$1
+    shift
+    if [ $# -ne 0 ]; then
+        echo "${indent}<${tag}"
+        indent_push
+        tag_attributes "$@"
+        echo "${indent}>"
+    else
+        echo "${indent}<${tag}>"
+        indent_push
+    fi
+}
+
+close_tag() {
+    local tag=$1
+    indent_pop
+    echo "${indent}</${tag}>"
+}
+
+tag() {
+    local tag=$1
+    shift
+    if [ $# -ne 0 ]; then
+        echo "${indent}<${tag}"
+        indent_push
+        tag_attributes "$@"
+        indent_pop
+        echo "${indent}/>"
+    else
+        echo "${indent}<${tag}/>"
+    fi
+}
+
+tag_content() {
+    local tag=$1
+    local content=$2
+    shift
+    shift
+    if [ $# -ne 0 ]; then
+        echo "${indent}<${tag}"
+        indent_push
+        tag_attributes "$@"
+        echo "${indent}>${content}</${tag}>"
+        indent_pop
+    else
+        echo "${indent}<${tag}>${content}</${tag}>"
+    fi
+}
+
+generate_filter() {
+    local name=$1
+    local pats=$2
+    local file_list_sz
+    local i
+    local f
+    local saveIFS="$IFS"
+    local pack
+    echo "generating filter '$name' from ${#file_list[@]} files" >&2
+    IFS=*
+
+    file_list_sz=${#file_list[@]}
+    for i in ${!file_list[@]}; do
+        f=${file_list[i]}
+        for pat in ${pats//;/$IFS}; do
+            if [ "${f##*.}" == "$pat" ]; then
+                unset file_list[i]
+
+                objf=$(echo ${f%.*}.obj | sed -e 's/^[\./]\+//g' -e 's,/,_,g')
+
+                if ([ "$pat" == "asm" ] || [ "$pat" == "s" ]) && $asm_use_custom_step; then
+                    open_tag CustomBuild \
+                        Include=".\\$f"
+                    for plat in "${platforms[@]}"; do
+                        for cfg in Debug Release; do
+                            tag_content Message "Assembling %(Filename)%(Extension)" \
+                                Condition="'\$(Configuration)|\$(Platform)'=='$cfg|$plat'"
+                            tag_content Command "$(eval echo \$asm_${cfg}_cmdline) -o \$(IntDir)$objf" \
+                                Condition="'\$(Configuration)|\$(Platform)'=='$cfg|$plat'"
+                            tag_content Outputs "\$(IntDir)$objf" \
+                                Condition="'\$(Configuration)|\$(Platform)'=='$cfg|$plat'"
+                        done
+                    done
+                    close_tag CustomBuild
+                elif [ "$pat" == "c" ] || [ "$pat" == "cc" ] ; then
+                    open_tag ClCompile \
+                        Include=".\\$f"
+                    # Separate file names with Condition?
+                    tag_content ObjectFileName "\$(IntDir)$objf"
+                    close_tag ClCompile
+                elif [ "$pat" == "h" ] ; then
+                    tag ClInclude \
+                        Include=".\\$f"
+                elif [ "$pat" == "vcxproj" ] ; then
+                    open_tag ProjectReference \
+                        Include="$f"
+                    depguid=`grep ProjectGuid "$f" | sed 's,.*<.*>\(.*\)</.*>.*,\1,'`
+                    tag_content Project "$depguid"
+                    tag_content ReferenceOutputAssembly false
+                    close_tag ProjectReference
+                else
+                    tag None \
+                        Include=".\\$f"
+                fi
+
+                break
+            fi
+        done
+    done
+
+    IFS="$saveIFS"
+}
+
+# Process command line
+unset target
+for opt in "$@"; do
+    optval="${opt#*=}"
+    case "$opt" in
+        --help|-h) show_help
+        ;;
+        --target=*) target="${optval}"
+        ;;
+        --out=*) outfile="$optval"
+        ;;
+        --name=*) name="${optval}"
+        ;;
+        --proj-guid=*) guid="${optval}"
+        ;;
+        --module-def=*) module_def="${optval}"
+        ;;
+        --exe) proj_kind="exe"
+        ;;
+        --dll) proj_kind="dll"
+        ;;
+        --lib) proj_kind="lib"
+        ;;
+        --src-path-bare=*) src_path_bare="$optval"
+        ;;
+        --static-crt) use_static_runtime=true
+        ;;
+        --ver=*)
+            vs_ver="$optval"
+            case "$optval" in
+                10|11)
+                ;;
+                *) die Unrecognized Visual Studio Version in $opt
+                ;;
+            esac
+        ;;
+        -I*)
+            opt="${opt%/}"
+            incs="${incs}${incs:+;}${opt##-I}"
+            yasmincs="${yasmincs} ${opt}"
+        ;;
+        -D*) defines="${defines}${defines:+;}${opt##-D}"
+        ;;
+        -L*) # fudge . to $(OutDir)
+            if [ "${opt##-L}" == "." ]; then
+                libdirs="${libdirs}${libdirs:+;}\$(OutDir)"
+            else
+                 # Also try directories for this platform/configuration
+                 libdirs="${libdirs}${libdirs:+;}${opt##-L}"
+                 libdirs="${libdirs}${libdirs:+;}${opt##-L}/\$(PlatformName)/\$(Configuration)"
+                 libdirs="${libdirs}${libdirs:+;}${opt##-L}/\$(PlatformName)"
+            fi
+        ;;
+        -l*) libs="${libs}${libs:+ }${opt##-l}.lib"
+        ;;
+        -*) die_unknown $opt
+        ;;
+        *)
+            file_list[${#file_list[@]}]="$opt"
+            case "$opt" in
+                 *.asm|*.s) uses_asm=true
+                 ;;
+            esac
+        ;;
+    esac
+done
+outfile=${outfile:-/dev/stdout}
+guid=${guid:-`generate_uuid`}
+asm_use_custom_step=false
+uses_asm=${uses_asm:-false}
+case "${vs_ver:-11}" in
+    10|11)
+       asm_use_custom_step=$uses_asm
+    ;;
+esac
+
+[ -n "$name" ] || die "Project name (--name) must be specified!"
+[ -n "$target" ] || die "Target (--target) must be specified!"
+
+if ${use_static_runtime:-false}; then
+    release_runtime=MultiThreaded
+    debug_runtime=MultiThreadedDebug
+    lib_sfx=mt
+else
+    release_runtime=MultiThreadedDLL
+    debug_runtime=MultiThreadedDebugDLL
+    lib_sfx=md
+fi
+
+# Calculate debug lib names: If a lib ends in ${lib_sfx}.lib, then rename
+# it to ${lib_sfx}d.lib. This precludes linking to release libs from a
+# debug exe, so this may need to be refactored later.
+for lib in ${libs}; do
+    if [ "$lib" != "${lib%${lib_sfx}.lib}" ]; then
+        lib=${lib%.lib}d.lib
+    fi
+    debug_libs="${debug_libs}${debug_libs:+ }${lib}"
+done
+debug_libs=${debug_libs// /;}
+libs=${libs// /;}
+
+
+# List of all platforms supported for this target
+case "$target" in
+    x86_64*)
+        platforms[0]="x64"
+        asm_Debug_cmdline="yasm -Xvc -g cv8 -f \$(PlatformName) ${yasmincs} &quot;%(FullPath)&quot;"
+        asm_Release_cmdline="yasm -Xvc -f \$(PlatformName) ${yasmincs} &quot;%(FullPath)&quot;"
+    ;;
+    x86*)
+        platforms[0]="Win32"
+        asm_Debug_cmdline="yasm -Xvc -g cv8 -f \$(PlatformName) ${yasmincs} &quot;%(FullPath)&quot;"
+        asm_Release_cmdline="yasm -Xvc -f \$(PlatformName) ${yasmincs} &quot;%(FullPath)&quot;"
+    ;;
+    arm*)
+        asm_Debug_cmdline="armasm -nologo &quot;%(FullPath)&quot;"
+        asm_Release_cmdline="armasm -nologo &quot;%(FullPath)&quot;"
+        if [ "$name" = "obj_int_extract" ]; then
+            # We don't want to build this tool for the target architecture,
+            # but for an architecture we can run locally during the build.
+            platforms[0]="Win32"
+        else
+            platforms[0]="ARM"
+        fi
+    ;;
+    *) die "Unsupported target $target!"
+    ;;
+esac
+
+generate_vcxproj() {
+    echo "<?xml version=\"1.0\" encoding=\"utf-8\"?>"
+    open_tag Project \
+        DefaultTargets="Build" \
+        ToolsVersion="4.0" \
+        xmlns="http://schemas.microsoft.com/developer/msbuild/2003" \
+
+    open_tag ItemGroup \
+        Label="ProjectConfigurations"
+    for plat in "${platforms[@]}"; do
+        for config in Debug Release; do
+            open_tag ProjectConfiguration \
+                Include="$config|$plat"
+            tag_content Configuration $config
+            tag_content Platform $plat
+            close_tag ProjectConfiguration
+        done
+    done
+    close_tag ItemGroup
+
+    open_tag PropertyGroup \
+        Label="Globals"
+        tag_content ProjectGuid "{${guid}}"
+        tag_content RootNamespace ${name}
+        tag_content Keyword ManagedCProj
+    close_tag PropertyGroup
+
+    tag Import \
+        Project="\$(VCTargetsPath)\\Microsoft.Cpp.Default.props"
+
+    for plat in "${platforms[@]}"; do
+        for config in Release Debug; do
+            open_tag PropertyGroup \
+                Condition="'\$(Configuration)|\$(Platform)'=='$config|$plat'" \
+                Label="Configuration"
+            if [ "$proj_kind" = "exe" ]; then
+                tag_content ConfigurationType Application
+            elif [ "$proj_kind" = "dll" ]; then
+                tag_content ConfigurationType DynamicLibrary
+            else
+                tag_content ConfigurationType StaticLibrary
+            fi
+            if [ "$vs_ver" = "11" ]; then
+                if [ "$plat" = "ARM" ]; then
+                    # Setting the wp80 toolchain automatically sets the
+                    # WINAPI_FAMILY define, which is required for building
+                    # code for arm with the windows headers. Alternatively,
+                    # one could add AppContainerApplication=true in the Globals
+                    # section and add PrecompiledHeader=NotUsing and
+                    # CompileAsWinRT=false in ClCompile and SubSystem=Console
+                    # in Link.
+                    tag_content PlatformToolset v110_wp80
+                else
+                    tag_content PlatformToolset v110
+                fi
+            fi
+            tag_content CharacterSet Unicode
+            if [ "$config" = "Release" ]; then
+                tag_content WholeProgramOptimization true
+            fi
+            close_tag PropertyGroup
+        done
+    done
+
+    tag Import \
+        Project="\$(VCTargetsPath)\\Microsoft.Cpp.props"
+
+    open_tag ImportGroup \
+        Label="PropertySheets"
+        tag Import \
+            Project="\$(UserRootDir)\\Microsoft.Cpp.\$(Platform).user.props" \
+            Condition="exists('\$(UserRootDir)\\Microsoft.Cpp.\$(Platform).user.props')" \
+            Label="LocalAppDataPlatform"
+    close_tag ImportGroup
+
+    tag PropertyGroup \
+        Label="UserMacros"
+
+    for plat in "${platforms[@]}"; do
+        plat_no_ws=`echo $plat | sed 's/[^A-Za-z0-9_]/_/g'`
+        for config in Debug Release; do
+            open_tag PropertyGroup \
+                Condition="'\$(Configuration)|\$(Platform)'=='$config|$plat'"
+            tag_content OutDir "\$(SolutionDir)$plat_no_ws\\\$(Configuration)\\"
+            tag_content IntDir "$plat_no_ws\\\$(Configuration)\\${name}\\"
+            close_tag PropertyGroup
+        done
+    done
+
+    for plat in "${platforms[@]}"; do
+        for config in Debug Release; do
+            open_tag ItemDefinitionGroup \
+                Condition="'\$(Configuration)|\$(Platform)'=='$config|$plat'"
+            if [ "$name" = "vpx" ]; then
+                open_tag PreBuildEvent
+                tag_content Command "call obj_int_extract.bat $src_path_bare"
+                close_tag PreBuildEvent
+            fi
+            open_tag ClCompile
+            if [ "$config" = "Debug" ]; then
+                opt=Disabled
+                runtime=$debug_runtime
+                curlibs=$debug_libs
+                confsuffix=d
+                case "$name" in
+                obj_int_extract)
+                    debug=DEBUG
+                    ;;
+                *)
+                    debug=_DEBUG
+                    ;;
+                esac
+            else
+                opt=MaxSpeed
+                runtime=$release_runtime
+                curlibs=$libs
+                confsuffix=""
+                tag_content FavorSizeOrSpeed Speed
+                debug=NDEBUG
+            fi
+            case "$name" in
+            obj_int_extract)
+                extradefines=";_CONSOLE"
+                ;;
+            *)
+                extradefines=";$defines"
+                ;;
+            esac
+            tag_content Optimization $opt
+            tag_content AdditionalIncludeDirectories "$incs;%(AdditionalIncludeDirectories)"
+            tag_content PreprocessorDefinitions "WIN32;$debug;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE$extradefines;%(PreprocessorDefinitions)"
+            tag_content RuntimeLibrary $runtime
+            tag_content WarningLevel Level3
+            # DebugInformationFormat
+            close_tag ClCompile
+            case "$proj_kind" in
+            exe)
+                open_tag Link
+                if [ "$name" = "obj_int_extract" ]; then
+                    tag_content OutputFile "${name}.exe"
+                else
+                    tag_content AdditionalDependencies "$curlibs"
+                    tag_content AdditionalLibraryDirectories "$libdirs;%(AdditionalLibraryDirectories)"
+                fi
+                tag_content GenerateDebugInformation true
+                close_tag Link
+                ;;
+            dll)
+                open_tag Link
+                tag_content GenerateDebugInformation true
+                tag_content ModuleDefinitionFile $module_def
+                close_tag Link
+                ;;
+            lib)
+                open_tag Lib
+                tag_content OutputFile "\$(OutDir)${name}${lib_sfx}${confsuffix}.lib"
+                close_tag Lib
+                ;;
+            esac
+            close_tag ItemDefinitionGroup
+        done
+
+    done
+
+    open_tag ItemGroup
+    generate_filter "Source Files"   "c;cc;def;odl;idl;hpj;bat;asm;asmx;s"
+    close_tag ItemGroup
+    open_tag ItemGroup
+    generate_filter "Header Files"   "h;hm;inl;inc;xsd"
+    close_tag ItemGroup
+    open_tag ItemGroup
+    generate_filter "Build Files"    "mk"
+    close_tag ItemGroup
+    open_tag ItemGroup
+    generate_filter "References"     "vcxproj"
+    close_tag ItemGroup
+
+    tag Import \
+        Project="\$(VCTargetsPath)\\Microsoft.Cpp.targets"
+
+    open_tag ImportGroup \
+        Label="ExtensionTargets"
+    close_tag ImportGroup
+
+    close_tag Project
+
+    # This must be done from within the {} subshell
+    echo "Ignored files list (${#file_list[@]} items) is:" >&2
+    for f in "${file_list[@]}"; do
+        echo "    $f" >&2
+    done
+}
+
+# This regexp doesn't catch most of the strings in the vcxproj format,
+# since they're like <tag>path</tag> instead of <tag attr="path" />
+# as previously. It still seems to work ok despite this.
+generate_vcxproj |
+    sed  -e '/"/s;\([^ "]\)/;\1\\;g' |
+    sed  -e '/xmlns/s;\\;/;g' > ${outfile}
+
+exit
diff --git a/libvpx/build/make/obj_int_extract.c b/libvpx/build/make/obj_int_extract.c
index bf317bd..1604b5e 100644
--- a/libvpx/build/make/obj_int_extract.c
+++ b/libvpx/build/make/obj_int_extract.c
@@ -17,21 +17,19 @@
 #include "vpx_config.h"
 #include "vpx/vpx_integer.h"
 
-typedef enum
-{
-    OUTPUT_FMT_PLAIN,
-    OUTPUT_FMT_RVDS,
-    OUTPUT_FMT_GAS,
+typedef enum {
+  OUTPUT_FMT_PLAIN,
+  OUTPUT_FMT_RVDS,
+  OUTPUT_FMT_GAS,
 } output_fmt_t;
 
-int log_msg(const char *fmt, ...)
-{
-    int res;
-    va_list ap;
-    va_start(ap, fmt);
-    res = vfprintf(stderr, fmt, ap);
-    va_end(ap);
-    return res;
+int log_msg(const char *fmt, ...) {
+  int res;
+  va_list ap;
+  va_start(ap, fmt);
+  res = vfprintf(stderr, fmt, ap);
+  va_end(ap);
+  return res;
 }
 
 #if defined(__GNUC__) && __GNUC__
@@ -40,175 +38,148 @@
 #include <mach-o/loader.h>
 #include <mach-o/nlist.h>
 
-int parse_macho(uint8_t *base_buf, size_t sz)
-{
-    int i, j;
-    struct mach_header header;
-    uint8_t *buf = base_buf;
-    int base_data_section = 0;
-    int bits = 0;
+int parse_macho(uint8_t *base_buf, size_t sz) {
+  int i, j;
+  struct mach_header header;
+  uint8_t *buf = base_buf;
+  int base_data_section = 0;
+  int bits = 0;
 
-    /* We can read in mach_header for 32 and 64 bit architectures
-     * because it's identical to mach_header_64 except for the last
-     * element (uint32_t reserved), which we don't use. Then, when
-     * we know which architecture we're looking at, increment buf
-     * appropriately.
-     */
-    memcpy(&header, buf, sizeof(struct mach_header));
+  /* We can read in mach_header for 32 and 64 bit architectures
+   * because it's identical to mach_header_64 except for the last
+   * element (uint32_t reserved), which we don't use. Then, when
+   * we know which architecture we're looking at, increment buf
+   * appropriately.
+   */
+  memcpy(&header, buf, sizeof(struct mach_header));
 
-    if (header.magic == MH_MAGIC)
-    {
-        if (header.cputype == CPU_TYPE_ARM
-            || header.cputype == CPU_TYPE_X86)
-        {
-            bits = 32;
-            buf += sizeof(struct mach_header);
-        }
-        else
-        {
-            log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_[ARM|X86].\n");
-            goto bail;
-        }
+  if (header.magic == MH_MAGIC) {
+    if (header.cputype == CPU_TYPE_ARM
+        || header.cputype == CPU_TYPE_X86) {
+      bits = 32;
+      buf += sizeof(struct mach_header);
+    } else {
+      log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_[ARM|X86].\n");
+      goto bail;
     }
-    else if (header.magic == MH_MAGIC_64)
-    {
-        if (header.cputype == CPU_TYPE_X86_64)
-        {
-            bits = 64;
-            buf += sizeof(struct mach_header_64);
-        }
-        else
-        {
-            log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_X86_64.\n");
-            goto bail;
-        }
+  } else if (header.magic == MH_MAGIC_64) {
+    if (header.cputype == CPU_TYPE_X86_64) {
+      bits = 64;
+      buf += sizeof(struct mach_header_64);
+    } else {
+      log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_X86_64.\n");
+      goto bail;
     }
-    else
-    {
-        log_msg("Bad magic number for object file. 0x%x or 0x%x expected, 0x%x found.\n",
-                MH_MAGIC, MH_MAGIC_64, header.magic);
+  } else {
+    log_msg("Bad magic number for object file. 0x%x or 0x%x expected, 0x%x found.\n",
+            MH_MAGIC, MH_MAGIC_64, header.magic);
+    goto bail;
+  }
+
+  if (header.filetype != MH_OBJECT) {
+    log_msg("Bad filetype for object file. Currently only tested for MH_OBJECT.\n");
+    goto bail;
+  }
+
+  for (i = 0; i < header.ncmds; i++) {
+    struct load_command lc;
+
+    memcpy(&lc, buf, sizeof(struct load_command));
+
+    if (lc.cmd == LC_SEGMENT) {
+      uint8_t *seg_buf = buf;
+      struct section s;
+      struct segment_command seg_c;
+
+      memcpy(&seg_c, seg_buf, sizeof(struct segment_command));
+      seg_buf += sizeof(struct segment_command);
+
+      /* Although each section is given it's own offset, nlist.n_value
+       * references the offset of the first section. This isn't
+       * apparent without debug information because the offset of the
+       * data section is the same as the first section. However, with
+       * debug sections mixed in, the offset of the debug section
+       * increases but n_value still references the first section.
+       */
+      if (seg_c.nsects < 1) {
+        log_msg("Not enough sections\n");
         goto bail;
-    }
+      }
 
-    if (header.filetype != MH_OBJECT)
-    {
-        log_msg("Bad filetype for object file. Currently only tested for MH_OBJECT.\n");
+      memcpy(&s, seg_buf, sizeof(struct section));
+      base_data_section = s.offset;
+    } else if (lc.cmd == LC_SEGMENT_64) {
+      uint8_t *seg_buf = buf;
+      struct section_64 s;
+      struct segment_command_64 seg_c;
+
+      memcpy(&seg_c, seg_buf, sizeof(struct segment_command_64));
+      seg_buf += sizeof(struct segment_command_64);
+
+      /* Explanation in LG_SEGMENT */
+      if (seg_c.nsects < 1) {
+        log_msg("Not enough sections\n");
         goto bail;
+      }
+
+      memcpy(&s, seg_buf, sizeof(struct section_64));
+      base_data_section = s.offset;
+    } else if (lc.cmd == LC_SYMTAB) {
+      if (base_data_section != 0) {
+        struct symtab_command sc;
+        uint8_t *sym_buf = base_buf;
+        uint8_t *str_buf = base_buf;
+
+        memcpy(&sc, buf, sizeof(struct symtab_command));
+
+        if (sc.cmdsize != sizeof(struct symtab_command)) {
+          log_msg("Can't find symbol table!\n");
+          goto bail;
+        }
+
+        sym_buf += sc.symoff;
+        str_buf += sc.stroff;
+
+        for (j = 0; j < sc.nsyms; j++) {
+          /* Location of string is cacluated each time from the
+           * start of the string buffer.  On darwin the symbols
+           * are prefixed by "_", so we bump the pointer by 1.
+           * The target value is defined as an int in *_asm_*_offsets.c,
+           * which is 4 bytes on all targets we currently use.
+           */
+          if (bits == 32) {
+            struct nlist nl;
+            int val;
+
+            memcpy(&nl, sym_buf, sizeof(struct nlist));
+            sym_buf += sizeof(struct nlist);
+
+            memcpy(&val, base_buf + base_data_section + nl.n_value,
+                   sizeof(val));
+            printf("%-40s EQU %5d\n",
+                   str_buf + nl.n_un.n_strx + 1, val);
+          } else { /* if (bits == 64) */
+            struct nlist_64 nl;
+            int val;
+
+            memcpy(&nl, sym_buf, sizeof(struct nlist_64));
+            sym_buf += sizeof(struct nlist_64);
+
+            memcpy(&val, base_buf + base_data_section + nl.n_value,
+                   sizeof(val));
+            printf("%-40s EQU %5d\n",
+                   str_buf + nl.n_un.n_strx + 1, val);
+          }
+        }
+      }
     }
 
-    for (i = 0; i < header.ncmds; i++)
-    {
-        struct load_command lc;
+    buf += lc.cmdsize;
+  }
 
-        memcpy(&lc, buf, sizeof(struct load_command));
-
-        if (lc.cmd == LC_SEGMENT)
-        {
-            uint8_t *seg_buf = buf;
-            struct section s;
-            struct segment_command seg_c;
-
-            memcpy(&seg_c, seg_buf, sizeof(struct segment_command));
-            seg_buf += sizeof(struct segment_command);
-
-            /* Although each section is given it's own offset, nlist.n_value
-             * references the offset of the first section. This isn't
-             * apparent without debug information because the offset of the
-             * data section is the same as the first section. However, with
-             * debug sections mixed in, the offset of the debug section
-             * increases but n_value still references the first section.
-             */
-            if (seg_c.nsects < 1)
-            {
-                log_msg("Not enough sections\n");
-                goto bail;
-            }
-
-            memcpy(&s, seg_buf, sizeof(struct section));
-            base_data_section = s.offset;
-        }
-        else if (lc.cmd == LC_SEGMENT_64)
-        {
-            uint8_t *seg_buf = buf;
-            struct section_64 s;
-            struct segment_command_64 seg_c;
-
-            memcpy(&seg_c, seg_buf, sizeof(struct segment_command_64));
-            seg_buf += sizeof(struct segment_command_64);
-
-            /* Explanation in LG_SEGMENT */
-            if (seg_c.nsects < 1)
-            {
-                log_msg("Not enough sections\n");
-                goto bail;
-            }
-
-            memcpy(&s, seg_buf, sizeof(struct section_64));
-            base_data_section = s.offset;
-        }
-        else if (lc.cmd == LC_SYMTAB)
-        {
-            if (base_data_section != 0)
-            {
-                struct symtab_command sc;
-                uint8_t *sym_buf = base_buf;
-                uint8_t *str_buf = base_buf;
-
-                memcpy(&sc, buf, sizeof(struct symtab_command));
-
-                if (sc.cmdsize != sizeof(struct symtab_command))
-                {
-                    log_msg("Can't find symbol table!\n");
-                    goto bail;
-                }
-
-                sym_buf += sc.symoff;
-                str_buf += sc.stroff;
-
-                for (j = 0; j < sc.nsyms; j++)
-                {
-                    /* Location of string is cacluated each time from the
-                     * start of the string buffer.  On darwin the symbols
-                     * are prefixed by "_", so we bump the pointer by 1.
-                     * The target value is defined as an int in asm_*_offsets.c,
-                     * which is 4 bytes on all targets we currently use.
-                     */
-                    if (bits == 32)
-                    {
-                        struct nlist nl;
-                        int val;
-
-                        memcpy(&nl, sym_buf, sizeof(struct nlist));
-                        sym_buf += sizeof(struct nlist);
-
-                        memcpy(&val, base_buf + base_data_section + nl.n_value,
-                               sizeof(val));
-                        printf("%-40s EQU %5d\n",
-                               str_buf + nl.n_un.n_strx + 1, val);
-                    }
-                    else /* if (bits == 64) */
-                    {
-                        struct nlist_64 nl;
-                        int val;
-
-                        memcpy(&nl, sym_buf, sizeof(struct nlist_64));
-                        sym_buf += sizeof(struct nlist_64);
-
-                        memcpy(&val, base_buf + base_data_section + nl.n_value,
-                               sizeof(val));
-                        printf("%-40s EQU %5d\n",
-                               str_buf + nl.n_un.n_strx + 1, val);
-                    }
-                }
-            }
-        }
-
-        buf += lc.cmdsize;
-    }
-
-    return 0;
+  return 0;
 bail:
-    return 1;
+  return 1;
 
 }
 
@@ -216,448 +187,400 @@
 #include "elf.h"
 
 #define COPY_STRUCT(dst, buf, ofst, sz) do {\
-        if(ofst + sizeof((*(dst))) > sz) goto bail;\
-        memcpy(dst, buf+ofst, sizeof((*(dst))));\
-    } while(0)
+    if(ofst + sizeof((*(dst))) > sz) goto bail;\
+    memcpy(dst, buf+ofst, sizeof((*(dst))));\
+  } while(0)
 
 #define ENDIAN_ASSIGN(val, memb) do {\
-        if(!elf->le_data) {log_msg("Big Endian data not supported yet!\n");goto bail;}\
-        (val) = (memb);\
-    } while(0)
+    if(!elf->le_data) {log_msg("Big Endian data not supported yet!\n");goto bail;}\
+    (val) = (memb);\
+  } while(0)
 
 #define ENDIAN_ASSIGN_IN_PLACE(memb) do {\
-        ENDIAN_ASSIGN(memb, memb);\
-    } while(0)
+    ENDIAN_ASSIGN(memb, memb);\
+  } while(0)
 
-typedef struct
-{
-    uint8_t      *buf; /* Buffer containing ELF data */
-    size_t        sz;  /* Buffer size */
-    int           le_data; /* Data is little-endian */
-    unsigned char e_ident[EI_NIDENT]; /* Magic number and other info */
-    int           bits; /* 32 or 64 */
-    Elf32_Ehdr    hdr32;
-    Elf64_Ehdr    hdr64;
+typedef struct {
+  uint8_t      *buf; /* Buffer containing ELF data */
+  size_t        sz;  /* Buffer size */
+  int           le_data; /* Data is little-endian */
+  unsigned char e_ident[EI_NIDENT]; /* Magic number and other info */
+  int           bits; /* 32 or 64 */
+  Elf32_Ehdr    hdr32;
+  Elf64_Ehdr    hdr64;
 } elf_obj_t;
 
-int parse_elf_header(elf_obj_t *elf)
-{
-    int res;
-    /* Verify ELF Magic numbers */
-    COPY_STRUCT(&elf->e_ident, elf->buf, 0, elf->sz);
-    res = elf->e_ident[EI_MAG0] == ELFMAG0;
-    res &= elf->e_ident[EI_MAG1] == ELFMAG1;
-    res &= elf->e_ident[EI_MAG2] == ELFMAG2;
-    res &= elf->e_ident[EI_MAG3] == ELFMAG3;
-    res &= elf->e_ident[EI_CLASS] == ELFCLASS32
-        || elf->e_ident[EI_CLASS] == ELFCLASS64;
-    res &= elf->e_ident[EI_DATA] == ELFDATA2LSB;
+int parse_elf_header(elf_obj_t *elf) {
+  int res;
+  /* Verify ELF Magic numbers */
+  COPY_STRUCT(&elf->e_ident, elf->buf, 0, elf->sz);
+  res = elf->e_ident[EI_MAG0] == ELFMAG0;
+  res &= elf->e_ident[EI_MAG1] == ELFMAG1;
+  res &= elf->e_ident[EI_MAG2] == ELFMAG2;
+  res &= elf->e_ident[EI_MAG3] == ELFMAG3;
+  res &= elf->e_ident[EI_CLASS] == ELFCLASS32
+         || elf->e_ident[EI_CLASS] == ELFCLASS64;
+  res &= elf->e_ident[EI_DATA] == ELFDATA2LSB;
 
-    if (!res) goto bail;
+  if (!res) goto bail;
 
-    elf->le_data = elf->e_ident[EI_DATA] == ELFDATA2LSB;
+  elf->le_data = elf->e_ident[EI_DATA] == ELFDATA2LSB;
 
-    /* Read in relevant values */
-    if (elf->e_ident[EI_CLASS] == ELFCLASS32)
-    {
-        elf->bits = 32;
-        COPY_STRUCT(&elf->hdr32, elf->buf, 0, elf->sz);
+  /* Read in relevant values */
+  if (elf->e_ident[EI_CLASS] == ELFCLASS32) {
+    elf->bits = 32;
+    COPY_STRUCT(&elf->hdr32, elf->buf, 0, elf->sz);
 
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_type);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_machine);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_version);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_entry);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phoff);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shoff);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_flags);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_ehsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phentsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phnum);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shentsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shnum);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shstrndx);
-    }
-    else /* if (elf->e_ident[EI_CLASS] == ELFCLASS64) */
-    {
-        elf->bits = 64;
-        COPY_STRUCT(&elf->hdr64, elf->buf, 0, elf->sz);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_type);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_machine);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_version);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_entry);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phoff);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shoff);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_flags);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_ehsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phentsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phnum);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shentsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shnum);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shstrndx);
+  } else { /* if (elf->e_ident[EI_CLASS] == ELFCLASS64) */
+    elf->bits = 64;
+    COPY_STRUCT(&elf->hdr64, elf->buf, 0, elf->sz);
 
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_type);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_machine);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_version);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_entry);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phoff);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shoff);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_flags);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_ehsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phentsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phnum);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shentsize);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shnum);
-        ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shstrndx);
-    }
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_type);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_machine);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_version);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_entry);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phoff);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shoff);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_flags);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_ehsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phentsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phnum);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shentsize);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shnum);
+    ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shstrndx);
+  }
 
-    return 0;
+  return 0;
 bail:
-    log_msg("Failed to parse ELF file header");
-    return 1;
+  log_msg("Failed to parse ELF file header");
+  return 1;
 }
 
-int parse_elf_section(elf_obj_t *elf, int idx, Elf32_Shdr *hdr32, Elf64_Shdr *hdr64)
-{
-    if (hdr32)
-    {
-        if (idx >= elf->hdr32.e_shnum)
-            goto bail;
-
-        COPY_STRUCT(hdr32, elf->buf, elf->hdr32.e_shoff + idx * elf->hdr32.e_shentsize,
-                    elf->sz);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_name);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_type);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_flags);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addr);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_offset);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_size);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_link);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_info);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addralign);
-        ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_entsize);
-    }
-    else /* if (hdr64) */
-    {
-        if (idx >= elf->hdr64.e_shnum)
-            goto bail;
-
-        COPY_STRUCT(hdr64, elf->buf, elf->hdr64.e_shoff + idx * elf->hdr64.e_shentsize,
-                    elf->sz);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_name);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_type);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_flags);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addr);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_offset);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_size);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_link);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_info);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addralign);
-        ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_entsize);
-    }
-
-    return 0;
-bail:
-    return 1;
-}
-
-char *parse_elf_string_table(elf_obj_t *elf, int s_idx, int idx)
-{
-    if (elf->bits == 32)
-    {
-        Elf32_Shdr shdr;
-
-        if (parse_elf_section(elf, s_idx, &shdr, NULL))
-        {
-            log_msg("Failed to parse ELF string table: section %d, index %d\n",
-                    s_idx, idx);
-            return "";
-        }
-
-        return (char *)(elf->buf + shdr.sh_offset + idx);
-    }
-    else /* if (elf->bits == 64) */
-    {
-        Elf64_Shdr shdr;
-
-        if (parse_elf_section(elf, s_idx, NULL, &shdr))
-        {
-            log_msg("Failed to parse ELF string table: section %d, index %d\n",
-                    s_idx, idx);
-            return "";
-        }
-
-        return (char *)(elf->buf + shdr.sh_offset + idx);
-    }
-}
-
-int parse_elf_symbol(elf_obj_t *elf, unsigned int ofst, Elf32_Sym *sym32, Elf64_Sym *sym64)
-{
-    if (sym32)
-    {
-        COPY_STRUCT(sym32, elf->buf, ofst, elf->sz);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_name);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_value);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_size);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_info);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_other);
-        ENDIAN_ASSIGN_IN_PLACE(sym32->st_shndx);
-    }
-    else /* if (sym64) */
-    {
-        COPY_STRUCT(sym64, elf->buf, ofst, elf->sz);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_name);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_value);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_size);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_info);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_other);
-        ENDIAN_ASSIGN_IN_PLACE(sym64->st_shndx);
-    }
-    return 0;
-bail:
-    return 1;
-}
-
-int parse_elf(uint8_t *buf, size_t sz, output_fmt_t mode)
-{
-    elf_obj_t    elf;
-    unsigned int ofst;
-    int          i;
-    Elf32_Off    strtab_off32;
-    Elf64_Off    strtab_off64; /* save String Table offset for later use */
-
-    memset(&elf, 0, sizeof(elf));
-    elf.buf = buf;
-    elf.sz = sz;
-
-    /* Parse Header */
-    if (parse_elf_header(&elf))
+int parse_elf_section(elf_obj_t *elf, int idx, Elf32_Shdr *hdr32, Elf64_Shdr *hdr64) {
+  if (hdr32) {
+    if (idx >= elf->hdr32.e_shnum)
       goto bail;
 
-    if (elf.bits == 32)
-    {
-        Elf32_Shdr shdr;
-        for (i = 0; i < elf.hdr32.e_shnum; i++)
-        {
-            parse_elf_section(&elf, i, &shdr, NULL);
+    COPY_STRUCT(hdr32, elf->buf, elf->hdr32.e_shoff + idx * elf->hdr32.e_shentsize,
+                elf->sz);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_name);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_type);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_flags);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addr);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_offset);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_size);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_link);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_info);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addralign);
+    ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_entsize);
+  } else { /* if (hdr64) */
+    if (idx >= elf->hdr64.e_shnum)
+      goto bail;
 
-            if (shdr.sh_type == SHT_STRTAB)
-            {
-                char strtsb_name[128];
+    COPY_STRUCT(hdr64, elf->buf, elf->hdr64.e_shoff + idx * elf->hdr64.e_shentsize,
+                elf->sz);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_name);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_type);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_flags);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addr);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_offset);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_size);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_link);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_info);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addralign);
+    ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_entsize);
+  }
 
-                strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name));
-
-                if (!(strcmp(strtsb_name, ".shstrtab")))
-                {
-                    /* log_msg("found section: %s\n", strtsb_name); */
-                    strtab_off32 = shdr.sh_offset;
-                    break;
-                }
-            }
-        }
-    }
-    else /* if (elf.bits == 64) */
-    {
-        Elf64_Shdr shdr;
-        for (i = 0; i < elf.hdr64.e_shnum; i++)
-        {
-            parse_elf_section(&elf, i, NULL, &shdr);
-
-            if (shdr.sh_type == SHT_STRTAB)
-            {
-                char strtsb_name[128];
-
-                strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name));
-
-                if (!(strcmp(strtsb_name, ".shstrtab")))
-                {
-                    /* log_msg("found section: %s\n", strtsb_name); */
-                    strtab_off64 = shdr.sh_offset;
-                    break;
-                }
-            }
-        }
-    }
-
-    /* Parse all Symbol Tables */
-    if (elf.bits == 32)
-    {
-        Elf32_Shdr shdr;
-        for (i = 0; i < elf.hdr32.e_shnum; i++)
-        {
-            parse_elf_section(&elf, i, &shdr, NULL);
-
-            if (shdr.sh_type == SHT_SYMTAB)
-            {
-                for (ofst = shdr.sh_offset;
-                     ofst < shdr.sh_offset + shdr.sh_size;
-                     ofst += shdr.sh_entsize)
-                {
-                    Elf32_Sym sym;
-
-                    parse_elf_symbol(&elf, ofst, &sym, NULL);
-
-                    /* For all OBJECTS (data objects), extract the value from the
-                     * proper data segment.
-                     */
-                    /* if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name)
-                        log_msg("found data object %s\n",
-                                parse_elf_string_table(&elf,
-                                                       shdr.sh_link,
-                                                       sym.st_name));
-                     */
-
-                    if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT
-                        && sym.st_size == 4)
-                    {
-                        Elf32_Shdr dhdr;
-                        int val = 0;
-                        char section_name[128];
-
-                        parse_elf_section(&elf, sym.st_shndx, &dhdr, NULL);
-
-                        /* For explanition - refer to _MSC_VER version of code */
-                        strcpy(section_name, (char *)(elf.buf + strtab_off32 + dhdr.sh_name));
-                        /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */
-
-                        if (strcmp(section_name, ".bss"))
-                        {
-                            if (sizeof(val) != sym.st_size)
-                            {
-                                /* The target value is declared as an int in
-                                 * asm_*_offsets.c, which is 4 bytes on all
-                                 * targets we currently use. Complain loudly if
-                                 * this is not true.
-                                 */
-                                log_msg("Symbol size is wrong\n");
-                                goto bail;
-                            }
-
-                            memcpy(&val,
-                                   elf.buf + dhdr.sh_offset + sym.st_value,
-                                   sym.st_size);
-                        }
-
-                        if (!elf.le_data)
-                        {
-                            log_msg("Big Endian data not supported yet!\n");
-                            goto bail;
-                        }
-
-                        switch (mode)
-                        {
-                            case OUTPUT_FMT_RVDS:
-                                printf("%-40s EQU %5d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                                break;
-                            case OUTPUT_FMT_GAS:
-                                printf(".equ %-40s, %5d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                                break;
-                            default:
-                                printf("%s = %d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                        }
-                    }
-                }
-            }
-        }
-    }
-    else /* if (elf.bits == 64) */
-    {
-        Elf64_Shdr shdr;
-        for (i = 0; i < elf.hdr64.e_shnum; i++)
-        {
-            parse_elf_section(&elf, i, NULL, &shdr);
-
-            if (shdr.sh_type == SHT_SYMTAB)
-            {
-                for (ofst = shdr.sh_offset;
-                     ofst < shdr.sh_offset + shdr.sh_size;
-                     ofst += shdr.sh_entsize)
-                {
-                    Elf64_Sym sym;
-
-                    parse_elf_symbol(&elf, ofst, NULL, &sym);
-
-                    /* For all OBJECTS (data objects), extract the value from the
-                     * proper data segment.
-                     */
-                    /* if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name)
-                        log_msg("found data object %s\n",
-                                parse_elf_string_table(&elf,
-                                                       shdr.sh_link,
-                                                       sym.st_name));
-                     */
-
-                    if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT
-                        && sym.st_size == 4)
-                    {
-                        Elf64_Shdr dhdr;
-                        int val = 0;
-                        char section_name[128];
-
-                        parse_elf_section(&elf, sym.st_shndx, NULL, &dhdr);
-
-                        /* For explanition - refer to _MSC_VER version of code */
-                        strcpy(section_name, (char *)(elf.buf + strtab_off64 + dhdr.sh_name));
-                        /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */
-
-                        if ((strcmp(section_name, ".bss")))
-                        {
-                            if (sizeof(val) != sym.st_size)
-                            {
-                                /* The target value is declared as an int in
-                                 * asm_*_offsets.c, which is 4 bytes on all
-                                 * targets we currently use. Complain loudly if
-                                 * this is not true.
-                                 */
-                                log_msg("Symbol size is wrong\n");
-                                goto bail;
-                            }
-
-                            memcpy(&val,
-                                   elf.buf + dhdr.sh_offset + sym.st_value,
-                                   sym.st_size);
-                        }
-
-                        if (!elf.le_data)
-                        {
-                            log_msg("Big Endian data not supported yet!\n");
-                            goto bail;
-                        }
-
-                        switch (mode)
-                        {
-                            case OUTPUT_FMT_RVDS:
-                                printf("%-40s EQU %5d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                                break;
-                            case OUTPUT_FMT_GAS:
-                                printf(".equ %-40s, %5d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                                break;
-                            default:
-                                printf("%s = %d\n",
-                                       parse_elf_string_table(&elf,
-                                                              shdr.sh_link,
-                                                              sym.st_name),
-                                       val);
-                        }
-                    }
-                }
-            }
-        }
-    }
-
-    if (mode == OUTPUT_FMT_RVDS)
-        printf("    END\n");
-
-    return 0;
+  return 0;
 bail:
-    log_msg("Parse error: File does not appear to be valid ELF32 or ELF64\n");
-    return 1;
+  return 1;
+}
+
+char *parse_elf_string_table(elf_obj_t *elf, int s_idx, int idx) {
+  if (elf->bits == 32) {
+    Elf32_Shdr shdr;
+
+    if (parse_elf_section(elf, s_idx, &shdr, NULL)) {
+      log_msg("Failed to parse ELF string table: section %d, index %d\n",
+              s_idx, idx);
+      return "";
+    }
+
+    return (char *)(elf->buf + shdr.sh_offset + idx);
+  } else { /* if (elf->bits == 64) */
+    Elf64_Shdr shdr;
+
+    if (parse_elf_section(elf, s_idx, NULL, &shdr)) {
+      log_msg("Failed to parse ELF string table: section %d, index %d\n",
+              s_idx, idx);
+      return "";
+    }
+
+    return (char *)(elf->buf + shdr.sh_offset + idx);
+  }
+}
+
+int parse_elf_symbol(elf_obj_t *elf, unsigned int ofst, Elf32_Sym *sym32, Elf64_Sym *sym64) {
+  if (sym32) {
+    COPY_STRUCT(sym32, elf->buf, ofst, elf->sz);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_name);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_value);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_size);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_info);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_other);
+    ENDIAN_ASSIGN_IN_PLACE(sym32->st_shndx);
+  } else { /* if (sym64) */
+    COPY_STRUCT(sym64, elf->buf, ofst, elf->sz);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_name);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_value);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_size);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_info);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_other);
+    ENDIAN_ASSIGN_IN_PLACE(sym64->st_shndx);
+  }
+  return 0;
+bail:
+  return 1;
+}
+
+int parse_elf(uint8_t *buf, size_t sz, output_fmt_t mode) {
+  elf_obj_t    elf;
+  unsigned int ofst;
+  int          i;
+  Elf32_Off    strtab_off32;
+  Elf64_Off    strtab_off64; /* save String Table offset for later use */
+
+  memset(&elf, 0, sizeof(elf));
+  elf.buf = buf;
+  elf.sz = sz;
+
+  /* Parse Header */
+  if (parse_elf_header(&elf))
+    goto bail;
+
+  if (elf.bits == 32) {
+    Elf32_Shdr shdr;
+    for (i = 0; i < elf.hdr32.e_shnum; i++) {
+      parse_elf_section(&elf, i, &shdr, NULL);
+
+      if (shdr.sh_type == SHT_STRTAB) {
+        char strtsb_name[128];
+
+        strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name));
+
+        if (!(strcmp(strtsb_name, ".shstrtab"))) {
+          /* log_msg("found section: %s\n", strtsb_name); */
+          strtab_off32 = shdr.sh_offset;
+          break;
+        }
+      }
+    }
+  } else { /* if (elf.bits == 64) */
+    Elf64_Shdr shdr;
+    for (i = 0; i < elf.hdr64.e_shnum; i++) {
+      parse_elf_section(&elf, i, NULL, &shdr);
+
+      if (shdr.sh_type == SHT_STRTAB) {
+        char strtsb_name[128];
+
+        strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name));
+
+        if (!(strcmp(strtsb_name, ".shstrtab"))) {
+          /* log_msg("found section: %s\n", strtsb_name); */
+          strtab_off64 = shdr.sh_offset;
+          break;
+        }
+      }
+    }
+  }
+
+  /* Parse all Symbol Tables */
+  if (elf.bits == 32) {
+    Elf32_Shdr shdr;
+    for (i = 0; i < elf.hdr32.e_shnum; i++) {
+      parse_elf_section(&elf, i, &shdr, NULL);
+
+      if (shdr.sh_type == SHT_SYMTAB) {
+        for (ofst = shdr.sh_offset;
+             ofst < shdr.sh_offset + shdr.sh_size;
+             ofst += shdr.sh_entsize) {
+          Elf32_Sym sym;
+
+          parse_elf_symbol(&elf, ofst, &sym, NULL);
+
+          /* For all OBJECTS (data objects), extract the value from the
+           * proper data segment.
+           */
+          /* if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name)
+              log_msg("found data object %s\n",
+                      parse_elf_string_table(&elf,
+                                             shdr.sh_link,
+                                             sym.st_name));
+           */
+
+          if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT
+              && sym.st_size == 4) {
+            Elf32_Shdr dhdr;
+            int val = 0;
+            char section_name[128];
+
+            parse_elf_section(&elf, sym.st_shndx, &dhdr, NULL);
+
+            /* For explanition - refer to _MSC_VER version of code */
+            strcpy(section_name, (char *)(elf.buf + strtab_off32 + dhdr.sh_name));
+            /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */
+
+            if (strcmp(section_name, ".bss")) {
+              if (sizeof(val) != sym.st_size) {
+                /* The target value is declared as an int in
+                 * *_asm_*_offsets.c, which is 4 bytes on all
+                 * targets we currently use. Complain loudly if
+                 * this is not true.
+                 */
+                log_msg("Symbol size is wrong\n");
+                goto bail;
+              }
+
+              memcpy(&val,
+                     elf.buf + dhdr.sh_offset + sym.st_value,
+                     sym.st_size);
+            }
+
+            if (!elf.le_data) {
+              log_msg("Big Endian data not supported yet!\n");
+              goto bail;
+            }
+
+            switch (mode) {
+              case OUTPUT_FMT_RVDS:
+                printf("%-40s EQU %5d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+                break;
+              case OUTPUT_FMT_GAS:
+                printf(".equ %-40s, %5d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+                break;
+              default:
+                printf("%s = %d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+            }
+          }
+        }
+      }
+    }
+  } else { /* if (elf.bits == 64) */
+    Elf64_Shdr shdr;
+    for (i = 0; i < elf.hdr64.e_shnum; i++) {
+      parse_elf_section(&elf, i, NULL, &shdr);
+
+      if (shdr.sh_type == SHT_SYMTAB) {
+        for (ofst = shdr.sh_offset;
+             ofst < shdr.sh_offset + shdr.sh_size;
+             ofst += shdr.sh_entsize) {
+          Elf64_Sym sym;
+
+          parse_elf_symbol(&elf, ofst, NULL, &sym);
+
+          /* For all OBJECTS (data objects), extract the value from the
+           * proper data segment.
+           */
+          /* if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name)
+              log_msg("found data object %s\n",
+                      parse_elf_string_table(&elf,
+                                             shdr.sh_link,
+                                             sym.st_name));
+           */
+
+          if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT
+              && sym.st_size == 4) {
+            Elf64_Shdr dhdr;
+            int val = 0;
+            char section_name[128];
+
+            parse_elf_section(&elf, sym.st_shndx, NULL, &dhdr);
+
+            /* For explanition - refer to _MSC_VER version of code */
+            strcpy(section_name, (char *)(elf.buf + strtab_off64 + dhdr.sh_name));
+            /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */
+
+            if ((strcmp(section_name, ".bss"))) {
+              if (sizeof(val) != sym.st_size) {
+                /* The target value is declared as an int in
+                 * *_asm_*_offsets.c, which is 4 bytes on all
+                 * targets we currently use. Complain loudly if
+                 * this is not true.
+                 */
+                log_msg("Symbol size is wrong\n");
+                goto bail;
+              }
+
+              memcpy(&val,
+                     elf.buf + dhdr.sh_offset + sym.st_value,
+                     sym.st_size);
+            }
+
+            if (!elf.le_data) {
+              log_msg("Big Endian data not supported yet!\n");
+              goto bail;
+            }
+
+            switch (mode) {
+              case OUTPUT_FMT_RVDS:
+                printf("%-40s EQU %5d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+                break;
+              case OUTPUT_FMT_GAS:
+                printf(".equ %-40s, %5d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+                break;
+              default:
+                printf("%s = %d\n",
+                       parse_elf_string_table(&elf,
+                                              shdr.sh_link,
+                                              sym.st_name),
+                       val);
+            }
+          }
+        }
+      }
+    }
+  }
+
+  if (mode == OUTPUT_FMT_RVDS)
+    printf("    END\n");
+
+  return 0;
+bail:
+  log_msg("Parse error: File does not appear to be valid ELF32 or ELF64\n");
+  return 1;
 }
 
 #endif
@@ -671,244 +594,222 @@
 #define get_le32(x) ((*(x)) | (*(x+1)) << 8 |(*(x+2)) << 16 | (*(x+3)) << 24 )
 #define get_le16(x) ((*(x)) | (*(x+1)) << 8)
 
-int parse_coff(uint8_t *buf, size_t sz)
-{
-    unsigned int nsections, symtab_ptr, symtab_sz, strtab_ptr;
-    unsigned int sectionrawdata_ptr;
-    unsigned int i;
-    uint8_t *ptr;
-    uint32_t symoffset;
+int parse_coff(uint8_t *buf, size_t sz) {
+  unsigned int nsections, symtab_ptr, symtab_sz, strtab_ptr;
+  unsigned int sectionrawdata_ptr;
+  unsigned int i;
+  uint8_t *ptr;
+  uint32_t symoffset;
 
-    char **sectionlist;  //this array holds all section names in their correct order.
-    //it is used to check if the symbol is in .bss or .rdata section.
+  char **sectionlist;  // this array holds all section names in their correct order.
+  // it is used to check if the symbol is in .bss or .rdata section.
 
-    nsections = get_le16(buf + 2);
-    symtab_ptr = get_le32(buf + 8);
-    symtab_sz = get_le32(buf + 12);
-    strtab_ptr = symtab_ptr + symtab_sz * 18;
+  nsections = get_le16(buf + 2);
+  symtab_ptr = get_le32(buf + 8);
+  symtab_sz = get_le32(buf + 12);
+  strtab_ptr = symtab_ptr + symtab_sz * 18;
 
-    if (nsections > 96)
-    {
-        log_msg("Too many sections\n");
-        return 1;
+  if (nsections > 96) {
+    log_msg("Too many sections\n");
+    return 1;
+  }
+
+  sectionlist = malloc(nsections * sizeof(sectionlist));
+
+  if (sectionlist == NULL) {
+    log_msg("Allocating first level of section list failed\n");
+    return 1;
+  }
+
+  // log_msg("COFF: Found %u symbols in %u sections.\n", symtab_sz, nsections);
+
+  /*
+  The size of optional header is always zero for an obj file. So, the section header
+  follows the file header immediately.
+  */
+
+  ptr = buf + 20;     // section header
+
+  for (i = 0; i < nsections; i++) {
+    char sectionname[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+    strncpy(sectionname, ptr, 8);
+    // log_msg("COFF: Parsing section %s\n",sectionname);
+
+    sectionlist[i] = malloc(strlen(sectionname) + 1);
+
+    if (sectionlist[i] == NULL) {
+      log_msg("Allocating storage for %s failed\n", sectionname);
+      goto bail;
+    }
+    strcpy(sectionlist[i], sectionname);
+
+    if (!strcmp(sectionname, ".rdata")) sectionrawdata_ptr = get_le32(ptr + 20);
+
+    ptr += 40;
+  }
+
+  // log_msg("COFF: Symbol table at offset %u\n", symtab_ptr);
+  // log_msg("COFF: raw data pointer ofset for section .rdata is %u\n", sectionrawdata_ptr);
+
+  /*  The compiler puts the data with non-zero offset in .rdata section, but puts the data with
+      zero offset in .bss section. So, if the data in in .bss section, set offset=0.
+      Note from Wiki: In an object module compiled from C, the bss section contains
+      the local variables (but not functions) that were declared with the static keyword,
+      except for those with non-zero initial values. (In C, static variables are initialized
+      to zero by default.) It also contains the non-local (both extern and static) variables
+      that are also initialized to zero (either explicitly or by default).
+      */
+  // move to symbol table
+  /* COFF symbol table:
+      offset      field
+      0           Name(*)
+      8           Value
+      12          SectionNumber
+      14          Type
+      16          StorageClass
+      17          NumberOfAuxSymbols
+      */
+  ptr = buf + symtab_ptr;
+
+  for (i = 0; i < symtab_sz; i++) {
+    int16_t section = get_le16(ptr + 12); // section number
+
+    if (section > 0 && ptr[16] == 2) {
+      // if(section > 0 && ptr[16] == 3 && get_le32(ptr+8)) {
+
+      if (get_le32(ptr)) {
+        char name[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+        strncpy(name, ptr, 8);
+        // log_msg("COFF: Parsing symbol %s\n",name);
+        /* The 64bit Windows compiler doesn't prefix with an _.
+         * Check what's there, and bump if necessary
+         */
+        if (name[0] == '_')
+          printf("%-40s EQU ", name + 1);
+        else
+          printf("%-40s EQU ", name);
+      } else {
+        // log_msg("COFF: Parsing symbol %s\n",
+        //        buf + strtab_ptr + get_le32(ptr+4));
+        if ((buf + strtab_ptr + get_le32(ptr + 4))[0] == '_')
+          printf("%-40s EQU ",
+                 buf + strtab_ptr + get_le32(ptr + 4) + 1);
+        else
+          printf("%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4));
+      }
+
+      if (!(strcmp(sectionlist[section - 1], ".bss"))) {
+        symoffset = 0;
+      } else {
+        symoffset = get_le32(buf + sectionrawdata_ptr + get_le32(ptr + 8));
+      }
+
+      // log_msg("      Section: %d\n",section);
+      // log_msg("      Class:   %d\n",ptr[16]);
+      // log_msg("      Address: %u\n",get_le32(ptr+8));
+      // log_msg("      Offset: %u\n", symoffset);
+
+      printf("%5d\n", symoffset);
     }
 
-    sectionlist = malloc(nsections * sizeof(sectionlist));
+    ptr += 18;
+  }
 
-    if (sectionlist == NULL)
-    {
-        log_msg("Allocating first level of section list failed\n");
-        return 1;
-    }
+  printf("    END\n");
 
-    //log_msg("COFF: Found %u symbols in %u sections.\n", symtab_sz, nsections);
+  for (i = 0; i < nsections; i++) {
+    free(sectionlist[i]);
+  }
 
-    /*
-    The size of optional header is always zero for an obj file. So, the section header
-    follows the file header immediately.
-    */
+  free(sectionlist);
 
-    ptr = buf + 20;     //section header
-
-    for (i = 0; i < nsections; i++)
-    {
-        char sectionname[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
-        strncpy(sectionname, ptr, 8);
-        //log_msg("COFF: Parsing section %s\n",sectionname);
-
-        sectionlist[i] = malloc(strlen(sectionname) + 1);
-
-        if (sectionlist[i] == NULL)
-        {
-            log_msg("Allocating storage for %s failed\n", sectionname);
-            goto bail;
-        }
-        strcpy(sectionlist[i], sectionname);
-
-        if (!strcmp(sectionname, ".rdata")) sectionrawdata_ptr = get_le32(ptr + 20);
-
-        ptr += 40;
-    }
-
-    //log_msg("COFF: Symbol table at offset %u\n", symtab_ptr);
-    //log_msg("COFF: raw data pointer ofset for section .rdata is %u\n", sectionrawdata_ptr);
-
-    /*  The compiler puts the data with non-zero offset in .rdata section, but puts the data with
-        zero offset in .bss section. So, if the data in in .bss section, set offset=0.
-        Note from Wiki: In an object module compiled from C, the bss section contains
-        the local variables (but not functions) that were declared with the static keyword,
-        except for those with non-zero initial values. (In C, static variables are initialized
-        to zero by default.) It also contains the non-local (both extern and static) variables
-        that are also initialized to zero (either explicitly or by default).
-        */
-    //move to symbol table
-    /* COFF symbol table:
-        offset      field
-        0           Name(*)
-        8           Value
-        12          SectionNumber
-        14          Type
-        16          StorageClass
-        17          NumberOfAuxSymbols
-        */
-    ptr = buf + symtab_ptr;
-
-    for (i = 0; i < symtab_sz; i++)
-    {
-        int16_t section = get_le16(ptr + 12); //section number
-
-        if (section > 0 && ptr[16] == 2)
-        {
-            //if(section > 0 && ptr[16] == 3 && get_le32(ptr+8)) {
-
-            if (get_le32(ptr))
-            {
-                char name[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
-                strncpy(name, ptr, 8);
-                //log_msg("COFF: Parsing symbol %s\n",name);
-                /* The 64bit Windows compiler doesn't prefix with an _.
-                 * Check what's there, and bump if necessary
-                 */
-                if (name[0] == '_')
-                    printf("%-40s EQU ", name + 1);
-                else
-                    printf("%-40s EQU ", name);
-            }
-            else
-            {
-                //log_msg("COFF: Parsing symbol %s\n",
-                //        buf + strtab_ptr + get_le32(ptr+4));
-                if ((buf + strtab_ptr + get_le32(ptr + 4))[0] == '_')
-                    printf("%-40s EQU ",
-                           buf + strtab_ptr + get_le32(ptr + 4) + 1);
-                else
-                    printf("%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4));
-            }
-
-            if (!(strcmp(sectionlist[section-1], ".bss")))
-            {
-                symoffset = 0;
-            }
-            else
-            {
-                symoffset = get_le32(buf + sectionrawdata_ptr + get_le32(ptr + 8));
-            }
-
-            //log_msg("      Section: %d\n",section);
-            //log_msg("      Class:   %d\n",ptr[16]);
-            //log_msg("      Address: %u\n",get_le32(ptr+8));
-            //log_msg("      Offset: %u\n", symoffset);
-
-            printf("%5d\n", symoffset);
-        }
-
-        ptr += 18;
-    }
-
-    printf("    END\n");
-
-    for (i = 0; i < nsections; i++)
-    {
-        free(sectionlist[i]);
-    }
-
-    free(sectionlist);
-
-    return 0;
+  return 0;
 bail:
 
-    for (i = 0; i < nsections; i++)
-    {
-        free(sectionlist[i]);
-    }
+  for (i = 0; i < nsections; i++) {
+    free(sectionlist[i]);
+  }
 
-    free(sectionlist);
+  free(sectionlist);
 
-    return 1;
+  return 1;
 }
 #endif /* defined(_MSC_VER) || defined(__MINGW32__) || defined(__CYGWIN__) */
 
-int main(int argc, char **argv)
-{
-    output_fmt_t mode = OUTPUT_FMT_PLAIN;
-    const char *f;
-    uint8_t *file_buf;
-    int res;
-    FILE *fp;
-    long int file_size;
+int main(int argc, char **argv) {
+  output_fmt_t mode = OUTPUT_FMT_PLAIN;
+  const char *f;
+  uint8_t *file_buf;
+  int res;
+  FILE *fp;
+  long int file_size;
 
-    if (argc < 2 || argc > 3)
-    {
-        fprintf(stderr, "Usage: %s [output format] <obj file>\n\n", argv[0]);
-        fprintf(stderr, "  <obj file>\tobject file to parse\n");
-        fprintf(stderr, "Output Formats:\n");
-        fprintf(stderr, "  gas  - compatible with GNU assembler\n");
-        fprintf(stderr, "  rvds - compatible with armasm\n");
-        goto bail;
-    }
+  if (argc < 2 || argc > 3) {
+    fprintf(stderr, "Usage: %s [output format] <obj file>\n\n", argv[0]);
+    fprintf(stderr, "  <obj file>\tobject file to parse\n");
+    fprintf(stderr, "Output Formats:\n");
+    fprintf(stderr, "  gas  - compatible with GNU assembler\n");
+    fprintf(stderr, "  rvds - compatible with armasm\n");
+    goto bail;
+  }
 
-    f = argv[2];
+  f = argv[2];
 
-    if (!strcmp(argv[1], "rvds"))
-        mode = OUTPUT_FMT_RVDS;
-    else if (!strcmp(argv[1], "gas"))
-        mode = OUTPUT_FMT_GAS;
-    else
-        f = argv[1];
+  if (!strcmp(argv[1], "rvds"))
+    mode = OUTPUT_FMT_RVDS;
+  else if (!strcmp(argv[1], "gas"))
+    mode = OUTPUT_FMT_GAS;
+  else
+    f = argv[1];
 
-    fp = fopen(f, "rb");
+  fp = fopen(f, "rb");
 
-    if (!fp)
-    {
-        perror("Unable to open file");
-        goto bail;
-    }
+  if (!fp) {
+    perror("Unable to open file");
+    goto bail;
+  }
 
-    if (fseek(fp, 0, SEEK_END))
-    {
-        perror("stat");
-        goto bail;
-    }
+  if (fseek(fp, 0, SEEK_END)) {
+    perror("stat");
+    goto bail;
+  }
 
-    file_size = ftell(fp);
-    file_buf = malloc(file_size);
+  file_size = ftell(fp);
+  file_buf = malloc(file_size);
 
-    if (!file_buf)
-    {
-        perror("malloc");
-        goto bail;
-    }
+  if (!file_buf) {
+    perror("malloc");
+    goto bail;
+  }
 
-    rewind(fp);
+  rewind(fp);
 
-    if (fread(file_buf, sizeof(char), file_size, fp) != file_size)
-    {
-        perror("read");
-        goto bail;
-    }
+  if (fread(file_buf, sizeof(char), file_size, fp) != file_size) {
+    perror("read");
+    goto bail;
+  }
 
-    if (fclose(fp))
-    {
-        perror("close");
-        goto bail;
-    }
+  if (fclose(fp)) {
+    perror("close");
+    goto bail;
+  }
 
 #if defined(__GNUC__) && __GNUC__
 #if defined(__MACH__)
-    res = parse_macho(file_buf, file_size);
+  res = parse_macho(file_buf, file_size);
 #elif defined(__ELF__)
-    res = parse_elf(file_buf, file_size, mode);
+  res = parse_elf(file_buf, file_size, mode);
 #endif
 #endif
 #if defined(_MSC_VER) || defined(__MINGW32__) || defined(__CYGWIN__)
-    res = parse_coff(file_buf, file_size);
+  res = parse_coff(file_buf, file_size);
 #endif
 
-    free(file_buf);
+  free(file_buf);
 
-    if (!res)
-        return EXIT_SUCCESS;
+  if (!res)
+    return EXIT_SUCCESS;
 
 bail:
-    return EXIT_FAILURE;
+  return EXIT_FAILURE;
 }
diff --git a/libvpx/build/make/rtcd.sh b/libvpx/build/make/rtcd.sh
index ddf9e09..6cc3684 100755
--- a/libvpx/build/make/rtcd.sh
+++ b/libvpx/build/make/rtcd.sh
@@ -59,13 +59,13 @@
 # Routines for the RTCD DSL to call
 #
 prototype() {
-  local rtyp
+  rtyp=""
   case "$1" in
     unsigned) rtyp="$1 "; shift;;
   esac
   rtyp="${rtyp}$1"
-  local fn="$2"
-  local args="$3"
+  fn="$2"
+  args="$3"
 
   eval "${2}_rtyp='$rtyp'"
   eval "${2}_args='$3'"
@@ -74,7 +74,7 @@
 }
 
 specialize() {
-  local fn="$1"
+  fn="$1"
   shift
   for opt in "$@"; do
     eval "${fn}_${opt}=${fn}_${opt}"
@@ -84,13 +84,13 @@
 require() {
   for fn in $ALL_FUNCS; do
     for opt in "$@"; do
-      local ofn=$(eval "echo \$${fn}_${opt}")
+      ofn=$(eval "echo \$${fn}_${opt}")
       [ -z "$ofn" ] && continue
 
       # if we already have a default, then we can disable it, as we know
       # we can do better.
-      local best=$(eval "echo \$${fn}_default")
-      local best_ofn=$(eval "echo \$${best}")
+      best=$(eval "echo \$${fn}_default")
+      best_ofn=$(eval "echo \$${best}")
       [ -n "$best" ] && [ "$best_ofn" != "$ofn" ] && eval "${best}_link=false"
       eval "${fn}_default=${fn}_${opt}"
       eval "${fn}_${opt}_link=true"
@@ -121,15 +121,15 @@
 determine_indirection() {
   [ "$CONFIG_RUNTIME_CPU_DETECT" = "yes" ] || require $ALL_ARCHS
   for fn in $ALL_FUNCS; do
-    local n=""
-    local rtyp="$(eval "echo \$${fn}_rtyp")"
-    local args="$(eval "echo \"\$${fn}_args\"")"
-    local dfn="$(eval "echo \$${fn}_default")"
+    n=""
+    rtyp="$(eval "echo \$${fn}_rtyp")"
+    args="$(eval "echo \"\$${fn}_args\"")"
+    dfn="$(eval "echo \$${fn}_default")"
     dfn=$(eval "echo \$${dfn}")
     for opt in "$@"; do
-      local ofn=$(eval "echo \$${fn}_${opt}")
+      ofn=$(eval "echo \$${fn}_${opt}")
       [ -z "$ofn" ] && continue
-      local link=$(eval "echo \$${fn}_${opt}_link")
+      link=$(eval "echo \$${fn}_${opt}_link")
       [ "$link" = "false" ] && continue
       n="${n}x"
     done
@@ -143,12 +143,12 @@
 
 declare_function_pointers() {
   for fn in $ALL_FUNCS; do
-    local rtyp="$(eval "echo \$${fn}_rtyp")"
-    local args="$(eval "echo \"\$${fn}_args\"")"
-    local dfn="$(eval "echo \$${fn}_default")"
+    rtyp="$(eval "echo \$${fn}_rtyp")"
+    args="$(eval "echo \"\$${fn}_args\"")"
+    dfn="$(eval "echo \$${fn}_default")"
     dfn=$(eval "echo \$${dfn}")
     for opt in "$@"; do
-      local ofn=$(eval "echo \$${fn}_${opt}")
+      ofn=$(eval "echo \$${fn}_${opt}")
       [ -z "$ofn" ] && continue
       echo "$rtyp ${ofn}($args);"
     done
@@ -163,20 +163,20 @@
 
 set_function_pointers() {
   for fn in $ALL_FUNCS; do
-    local n=""
-    local rtyp="$(eval "echo \$${fn}_rtyp")"
-    local args="$(eval "echo \"\$${fn}_args\"")"
-    local dfn="$(eval "echo \$${fn}_default")"
+    n=""
+    rtyp="$(eval "echo \$${fn}_rtyp")"
+    args="$(eval "echo \"\$${fn}_args\"")"
+    dfn="$(eval "echo \$${fn}_default")"
     dfn=$(eval "echo \$${dfn}")
     if $(eval "echo \$${fn}_indirect"); then
       echo "    $fn = $dfn;"
       for opt in "$@"; do
-        local ofn=$(eval "echo \$${fn}_${opt}")
+        ofn=$(eval "echo \$${fn}_${opt}")
         [ -z "$ofn" ] && continue
         [ "$ofn" = "$dfn" ] && continue;
-        local link=$(eval "echo \$${fn}_${opt}_link")
+        link=$(eval "echo \$${fn}_${opt}_link")
         [ "$link" = "false" ] && continue
-        local cond="$(eval "echo \$have_${opt}")"
+        cond="$(eval "echo \$have_${opt}")"
         echo "    if (${cond}) $fn = $ofn;"
       done
     fi
@@ -185,7 +185,7 @@
 }
 
 filter() {
-  local filtered
+  filtered=""
   for opt in "$@"; do
     [ -z $(eval "echo \$disable_${opt}") ] && filtered="$filtered $opt"
   done
@@ -196,8 +196,9 @@
 # Helper functions for generating the arch specific RTCD files
 #
 common_top() {
-  local outfile_basename=$(basename ${symbol:-rtcd.h})
-  local include_guard=$(echo $outfile_basename | tr '[a-z]' '[A-Z]' | tr -c '[A-Z]' _)
+  outfile_basename=$(basename ${symbol:-rtcd})
+  include_guard=$(echo $outfile_basename | tr '[a-z]' '[A-Z]' | \
+    tr -c '[A-Z0-9]' _)H_
   cat <<EOF
 #ifndef ${include_guard}
 #define ${include_guard}
@@ -227,7 +228,7 @@
 
   # Assign the helper variable for each enabled extension
   for opt in $ALL_ARCHS; do
-    local uc=$(echo $opt | tr '[a-z]' '[A-Z]')
+    uc=$(echo $opt | tr '[a-z]' '[A-Z]')
     eval "have_${opt}=\"flags & HAS_${uc}\""
   done
 
@@ -254,7 +255,7 @@
 
   # Assign the helper variable for each enabled extension
   for opt in $ALL_ARCHS; do
-    local uc=$(echo $opt | tr '[a-z]' '[A-Z]')
+    uc=$(echo $opt | tr '[a-z]' '[A-Z]')
     eval "have_${opt}=\"flags & HAS_${uc}\""
   done
 
diff --git a/libvpx/build/make/thumb.pm b/libvpx/build/make/thumb.pm
new file mode 100644
index 0000000..f347287
--- /dev/null
+++ b/libvpx/build/make/thumb.pm
@@ -0,0 +1,70 @@
+#!/usr/bin/perl
+##
+##  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+package thumb;
+
+sub FixThumbInstructions($$)
+{
+    my $short_branches = $_[1];
+    my $branch_shift_offset = $short_branches ? 1 : 0;
+
+    # Write additions with shifts, such as "add r10, r11, lsl #8",
+    # in three operand form, "add r10, r10, r11, lsl #8".
+    s/(add\s+)(r\d+),\s*(r\d+),\s*(lsl #\d+)/$1$2, $2, $3, $4/g;
+
+    # Convert additions with a non-constant shift into a sequence
+    # with left shift, addition and a right shift (to restore the
+    # register to the original value). Currently the right shift
+    # isn't necessary in the code base since the values in these
+    # registers aren't used, but doing the shift for consitency.
+    # This converts instructions such as "add r12, r12, r5, lsl r4"
+    # into the sequence "lsl r5, r4", "add r12, r12, r5", "lsr r5, r4".
+    s/^(\s*)(add)(\s+)(r\d+),\s*(r\d+),\s*(r\d+),\s*lsl (r\d+)/$1lsl$3$6, $7\n$1$2$3$4, $5, $6\n$1lsr$3$6, $7/g;
+
+    # Convert loads with right shifts in the indexing into a
+    # sequence of an add, load and sub. This converts
+    # "ldrb r4, [r9, lr, asr #1]" into "add r9, r9, lr, asr #1",
+    # "ldrb r9, [r9]", "sub r9, r9, lr, asr #1".
+    s/^(\s*)(ldrb)(\s+)(r\d+),\s*\[(\w+),\s*(\w+),\s*(asr #\d+)\]/$1add $3$5, $5, $6, $7\n$1$2$3$4, [$5]\n$1sub $3$5, $5, $6, $7/g;
+
+    # Convert register indexing with writeback into a separate add
+    # instruction. This converts "ldrb r12, [r1, r2]!" into
+    # "ldrb r12, [r1, r2]", "add r1, r1, r2".
+    s/^(\s*)(ldrb)(\s+)(r\d+),\s*\[(\w+),\s*(\w+)\]!/$1$2$3$4, [$5, $6]\n$1add $3$5, $6/g;
+
+    # Convert negative register indexing into separate sub/add instructions.
+    # This converts "ldrne r4, [src, -pstep, lsl #1]" into
+    # "subne src, src, pstep, lsl #1", "ldrne r4, [src]",
+    # "addne src, src, pstep, lsl #1". In a couple of cases where
+    # this is used, it's used for two subsequent load instructions,
+    # where a hand-written version of it could merge two subsequent
+    # add and sub instructions.
+    s/^(\s*)((ldr|str)(ne)?)(\s+)(r\d+),\s*\[(\w+), -([^\]]+)\]/$1sub$4$5$7, $7, $8\n$1$2$5$6, [$7]\n$1add$4$5$7, $7, $8/g;
+
+    # Convert register post indexing to a separate add instruction.
+    # This converts "ldrneb r9, [r0], r2" into "ldrneb r9, [r0]",
+    # "add r0, r2".
+    s/^(\s*)((ldr|str)(ne)?[bhd]?)(\s+)(\w+),(\s*\w+,)?\s*\[(\w+)\],\s*(\w+)/$1$2$5$6,$7 [$8]\n$1add$4$5$8, $8, $9/g;
+
+    # Convert a conditional addition to the pc register into a series of
+    # instructions. This converts "addlt pc, pc, r3, lsl #2" into
+    # "itttt lt", "movlt.n r12, pc", "addlt.w r12, #12",
+    # "addlt.w r12, r12, r3, lsl #2", "movlt.n pc, r12".
+    # This assumes that r12 is free at this point.
+    s/^(\s*)addlt(\s+)pc,\s*pc,\s*(\w+),\s*lsl\s*#(\d+)/$1itttt$2lt\n$1movlt.n$2r12, pc\n$1addlt.w$2r12, #12\n$1addlt.w$2r12, r12, $3, lsl #($4-$branch_shift_offset)\n$1movlt.n$2pc, r12/g;
+
+    # Convert "mov pc, lr" into "bx lr", since the former only works
+    # for switching from arm to thumb (and only in armv7), but not
+    # from thumb to arm.
+    s/mov(\s*)pc\s*,\s*lr/bx$1lr/g;
+}
+
+1;
diff --git a/libvpx/build/x86-msvs/obj_int_extract.bat b/libvpx/build/x86-msvs/obj_int_extract.bat
index 1bb8653..47fef97 100644
--- a/libvpx/build/x86-msvs/obj_int_extract.bat
+++ b/libvpx/build/x86-msvs/obj_int_extract.bat
@@ -7,9 +7,17 @@
 REM   be found in the AUTHORS file in the root of the source tree.
 echo on
 
-cl /I "./" /I "%1" /nologo /c "%1/vp8/common/asm_com_offsets.c"
-cl /I "./" /I "%1" /nologo /c "%1/vp8/decoder/asm_dec_offsets.c"
-cl /I "./" /I "%1" /nologo /c "%1/vp8/encoder/asm_enc_offsets.c"
-obj_int_extract.exe rvds "asm_com_offsets.obj" > "asm_com_offsets.asm"
-obj_int_extract.exe rvds "asm_dec_offsets.obj" > "asm_dec_offsets.asm"
-obj_int_extract.exe rvds "asm_enc_offsets.obj" > "asm_enc_offsets.asm"
+cl /I "./" /I "%1" /nologo /c "%1/vp9/common/vp9_asm_com_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp9/decoder/vp9_asm_dec_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp9/encoder/vp9_asm_enc_offsets.c"
+obj_int_extract.exe rvds "vp9_asm_com_offsets.obj" > "vp9_asm_com_offsets.asm"
+obj_int_extract.exe rvds "vp9_asm_dec_offsets.obj" > "vp9_asm_dec_offsets.asm"
+obj_int_extract.exe rvds "vp9_asm_enc_offsets.obj" > "vp9_asm_enc_offsets.asm"
+
+cl /I "./" /I "%1" /nologo /c "%1/vp8/common/vp8_asm_com_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp8/decoder/vp8_asm_dec_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp8/encoder/vp8_asm_enc_offsets.c"
+obj_int_extract.exe rvds "vp8_asm_com_offsets.obj" > "vp8_asm_com_offsets.asm"
+obj_int_extract.exe rvds "vp8_asm_dec_offsets.obj" > "vp8_asm_dec_offsets.asm"
+obj_int_extract.exe rvds "vp8_asm_enc_offsets.obj" > "vp8_asm_enc_offsets.asm"
+
diff --git a/libvpx/build/x86-msvs/yasm.rules b/libvpx/build/x86-msvs/yasm.rules
deleted file mode 100644
index ee1fefb..0000000
--- a/libvpx/build/x86-msvs/yasm.rules
+++ /dev/null
@@ -1,115 +0,0 @@
-﻿<?xml version="1.0" encoding="utf-8"?>
-<VisualStudioToolFile
-	Name="Yasm"
-	Version="8.00"
-	>
-  <Rules>
-    <CustomBuildRule
-			Name="YASM"
-			DisplayName="Yasm Assembler"
-			CommandLine="yasm -Xvc -f $(PlatformName) [AllOptions] [AdditionalOptions] [Inputs]"
-			Outputs="[$ObjectFileName]"
-			FileExtensions="*.asm"
-			ExecutionDescription="Assembling $(InputFileName)"
-			ShowOnlyRuleProperties="false"
-			>
-      <Properties>
-        <StringProperty
-					Name="Defines"
-					DisplayName="Definitions"
-					Category="Pre-Defined Symbols"
-					Description="Specify pre-defined symbols (&apos;symbol&apos; or &apos;symbol = value&apos;) "
-					Switch="-D [value]"
-					Delimited="true"
-					Inheritable="true"
-				/>
-        <StringProperty
-					Name="IncludePaths"
-					DisplayName="Include Paths"
-					Category="Configuration"
-					Description="Set the paths for any additional include files"
-					Switch="-I [value]"
-					Delimited="true"
-					Inheritable="true"
-				/>
-        <StringProperty
-					Name="UnDefines"
-					DisplayName="Remove Definitions"
-					Category="Pre-Defined Symbols"
-					Description="Remove pre-defined symbols "
-					Switch="-U [value]"
-					Delimited="true"
-					Inheritable="true"
-				/>
-        <StringProperty
-					Name="ObjectFileName"
-					DisplayName="Object File Name"
-					Category="Output"
-					Description="Select the output file name"
-					Switch="-o [value]"
-					DefaultValue="$(IntDir)\$(InputName).obj"
-				/>
-        <StringProperty
-					Name="ListFileName"
-					DisplayName="List File Name"
-					Category="Output"
-					Description="Select an output listing by setting its file name"
-					Switch="-l [value]"
-				/>
-        <StringProperty
-					Name="PreIncludeFile"
-					DisplayName="Pre Include File"
-					Category="Configuration"
-					Description="Select a pre-included file by setting its name"
-					Switch="-P [value]"
-				/>
-        <BooleanProperty
-					Name="Debug"
-					DisplayName="Debug Information"
-					Category="Output"
-					Description="Generate debugging information"
-					Switch="-g cv8"
-				/>
-        <EnumProperty
-					Name="PreProc"
-					DisplayName="Pre-Processor"
-					Category="Configuration"
-					Description="Select the pre-processor (&apos;nasm&apos; or &apos;raw&apos;)"
-					>
-          <Values>
-            <EnumValue
-							Value="0"
-							Switch="-rnasm"
-							DisplayName="Nasm "
-						/>
-            <EnumValue
-							Value="1"
-							Switch="-rraw"
-							DisplayName="Raw"
-						/>
-          </Values>
-        </EnumProperty>
-        <EnumProperty
-					Name="Parser"
-					DisplayName="Parser"
-					Category="Configuration"
-					Description="Select the parser for Intel (&apos;nasm&apos;) or AT&amp;T ( &apos;gas&apos;) syntax"
-					>
-          <Values>
-            <EnumValue
-							Value="0"
-							Switch="-pnasm"
-							DisplayName="Nasm"
-						/>
-            <EnumValue
-							Value="1"
-							Switch="-pgas"
-							DisplayName="Gas"
-						/>
-          </Values>
-        </EnumProperty>
-      </Properties>
-    </CustomBuildRule>
-  </Rules>
-</VisualStudioToolFile>
-
diff --git a/libvpx/configure b/libvpx/configure
index b3c5fe9..28676fb 100755
--- a/libvpx/configure
+++ b/libvpx/configure
@@ -34,6 +34,7 @@
   ${toggle_md5}                   support for output of checksum data
   ${toggle_static_msvcrt}         use static MSVCRT (VS builds only)
   ${toggle_vp8}                   VP8 codec support
+  ${toggle_vp9}                   VP9 codec support
   ${toggle_internal_stats}        output of encoder internal stats for debug, if supported (encoders)
   ${toggle_mem_tracker}           track memory usage
   ${toggle_postproc}              postprocessing
@@ -97,6 +98,7 @@
 all_platforms="${all_platforms} armv7-linux-rvct"    #neon Cortex-A8
 all_platforms="${all_platforms} armv7-linux-gcc"     #neon Cortex-A8
 all_platforms="${all_platforms} armv7-none-rvct"     #neon Cortex-A8
+all_platforms="${all_platforms} armv7-win32-vs11"
 all_platforms="${all_platforms} mips32-linux-gcc"
 all_platforms="${all_platforms} ppc32-darwin8-gcc"
 all_platforms="${all_platforms} ppc32-darwin9-gcc"
@@ -105,6 +107,7 @@
 all_platforms="${all_platforms} ppc64-darwin9-gcc"
 all_platforms="${all_platforms} ppc64-linux-gcc"
 all_platforms="${all_platforms} sparc-solaris-gcc"
+all_platforms="${all_platforms} x86-android-gcc"
 all_platforms="${all_platforms} x86-darwin8-gcc"
 all_platforms="${all_platforms} x86-darwin8-icc"
 all_platforms="${all_platforms} x86-darwin9-gcc"
@@ -120,6 +123,8 @@
 all_platforms="${all_platforms} x86-win32-vs7"
 all_platforms="${all_platforms} x86-win32-vs8"
 all_platforms="${all_platforms} x86-win32-vs9"
+all_platforms="${all_platforms} x86-win32-vs10"
+all_platforms="${all_platforms} x86-win32-vs11"
 all_platforms="${all_platforms} x86_64-darwin9-gcc"
 all_platforms="${all_platforms} x86_64-darwin10-gcc"
 all_platforms="${all_platforms} x86_64-darwin11-gcc"
@@ -130,6 +135,8 @@
 all_platforms="${all_platforms} x86_64-win64-gcc"
 all_platforms="${all_platforms} x86_64-win64-vs8"
 all_platforms="${all_platforms} x86_64-win64-vs9"
+all_platforms="${all_platforms} x86_64-win64-vs10"
+all_platforms="${all_platforms} x86_64-win64-vs11"
 all_platforms="${all_platforms} universal-darwin8-gcc"
 all_platforms="${all_platforms} universal-darwin9-gcc"
 all_platforms="${all_platforms} universal-darwin10-gcc"
@@ -176,19 +183,24 @@
 enable temporal_denoising
 
 [ -d ${source_path}/../include ] && enable alt_tree_layout
-for d in vp8; do
+for d in vp8 vp9; do
     [ -d ${source_path}/${d} ] && disable alt_tree_layout;
 done
 
 if ! enabled alt_tree_layout; then
 # development environment
 [ -d ${source_path}/vp8 ] && CODECS="${CODECS} vp8_encoder vp8_decoder"
+[ -d ${source_path}/vp9 ] && CODECS="${CODECS} vp9_encoder vp9_decoder"
 else
 # customer environment
 [ -f ${source_path}/../include/vpx/vp8cx.h ] && CODECS="${CODECS} vp8_encoder"
 [ -f ${source_path}/../include/vpx/vp8dx.h ] && CODECS="${CODECS} vp8_decoder"
+[ -f ${source_path}/../include/vpx/vp9cx.h ] && CODECS="${CODECS} vp9_encoder"
+[ -f ${source_path}/../include/vpx/vp9dx.h ] && CODECS="${CODECS} vp9_decoder"
 [ -f ${source_path}/../include/vpx/vp8cx.h ] || disable vp8_encoder
 [ -f ${source_path}/../include/vpx/vp8dx.h ] || disable vp8_decoder
+[ -f ${source_path}/../include/vpx/vp9cx.h ] || disable vp9_encoder
+[ -f ${source_path}/../include/vpx/vp9dx.h ] || disable vp9_decoder
 
 [ -f ${source_path}/../lib/*/*mt.lib ] && soft_enable static_msvcrt
 fi
@@ -230,6 +242,13 @@
     sys_mman_h
     unistd_h
 "
+EXPERIMENT_LIST="
+    oneshotq
+    multiple_arf
+    non420
+    alpha
+    balanced_coeftree
+"
 CONFIG_LIST="
     external_build
     install_docs
@@ -276,8 +295,12 @@
     unit_tests
     multi_res_encoding
     temporal_denoising
+    experimental
+    decrypt
+    ${EXPERIMENT_LIST}
 "
 CMDLINE_SELECT="
+    external_build
     extra_warnings
     werror
     install_docs
@@ -291,6 +314,7 @@
     optimizations
     ccache
     runtime_cpu_detect
+    thumb
 
     libs
     examples
@@ -322,6 +346,8 @@
     unit_tests
     multi_res_encoding
     temporal_denoising
+    experimental
+    decrypt
 "
 
 process_cmdline() {
@@ -329,6 +355,18 @@
         optval="${opt#*=}"
         case "$opt" in
         --disable-codecs) for c in ${CODECS}; do disable $c; done ;;
+        --enable-?*|--disable-?*)
+        eval `echo "$opt" | sed 's/--/action=/;s/-/ option=/;s/-/_/g'`
+        if echo "${EXPERIMENT_LIST}" | grep "^ *$option\$" >/dev/null; then
+            if enabled experimental; then
+                $action $option
+            else
+                log_echo "Ignoring $opt -- not in experimental mode."
+            fi
+        else
+            process_common_cmdline $opt
+        fi
+        ;;
         *) process_common_cmdline "$opt"
         ;;
         esac
@@ -464,7 +502,7 @@
             fi
         fi
     fi
-    if [ -z "$CC" ]; then
+    if [ -z "$CC" ] || enabled external_build; then
         echo "Bypassing toolchain for environment detection."
         enable external_build
         check_header() {
@@ -473,6 +511,7 @@
             shift
             var=`echo $header | sed 's/[^A-Za-z0-9_]/_/g'`
             disable $var
+            # Headers common to all environments
             case $header in
                 stdio.h)
                     true;
@@ -484,6 +523,25 @@
                     done
                     ${result:-true}
             esac && enable $var
+
+            # Specialize windows and POSIX environments.
+            case $toolchain in
+                *-win*-*)
+                    case $header-$toolchain in
+                        stdint*-gcc) true;;
+                        *) false;;
+                    esac && enable $var
+                    ;;
+                *)
+                    case $header in
+                        stdint.h) true;;
+                        pthread.h) true;;
+                        sys/mman.h) true;;
+                        unistd.h) true;;
+                        *) false;;
+                    esac && enable $var
+            esac
+            enabled $var
         }
         check_ld() {
             true
@@ -497,6 +555,7 @@
     check_header stdint.h
     check_header pthread.h
     check_header sys/mman.h
+    check_header unistd.h # for sysconf(3) and friends.
 
     check_header vpx/vpx_integer.h -I${source_path} && enable vpx_ports
 }
@@ -537,10 +596,14 @@
         check_add_cflags -Wpointer-arith
         check_add_cflags -Wtype-limits
         check_add_cflags -Wcast-qual
+        check_add_cflags -Wvla
         check_add_cflags -Wimplicit-function-declaration
         check_add_cflags -Wuninitialized
         check_add_cflags -Wunused-variable
-        check_add_cflags -Wunused-but-set-variable
+        case ${CC} in
+          *clang*) ;;
+          *) check_add_cflags -Wunused-but-set-variable ;;
+        esac
         enabled extra_warnings || check_add_cflags -Wno-unused-function
     fi
 
@@ -585,7 +648,18 @@
         vs*) enable msvs
              enable solution
              vs_version=${tgt_cc##vs}
+             case $vs_version in
+             [789])
+                 VCPROJ_SFX=vcproj
+                 gen_vcproj_cmd=${source_path}/build/make/gen_msvs_proj.sh
+                 ;;
+             10|11)
+                 VCPROJ_SFX=vcxproj
+                 gen_vcproj_cmd=${source_path}/build/make/gen_msvs_vcxproj.sh
+                 ;;
+             esac
              all_targets="${all_targets} solution"
+             INLINE="__forceinline"
         ;;
     esac
 
@@ -596,7 +670,7 @@
         enabled postproc || die "postproc_visualizer requires postproc to be enabled"
     fi
 
-    # Enable unit tests if we have a working C++ compiler
+    # Enable unit tests by default if we have a working C++ compiler.
     case "$toolchain" in
         *-vs*)
             soft_enable unit_tests
@@ -604,8 +678,12 @@
         *-android-*)
             # GTestLog must be modified to use Android logging utilities.
         ;;
+        *-darwin-*)
+            # iOS/ARM builds do not work with gtest. This does not match
+            # x86 targets.
+        ;;
         *)
-            check_cxx "$@" <<EOF && soft_enable unit_tests
+            enabled pthread_h && check_cxx "$@" <<EOF && soft_enable unit_tests
 int z;
 EOF
         ;;
diff --git a/libvpx/example_xma.c b/libvpx/example_xma.c
index 72eb470..7aa8798 100644
--- a/libvpx/example_xma.c
+++ b/libvpx/example_xma.c
@@ -18,197 +18,174 @@
 #include "vpx_config.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx/vpx_integer.h"
-#if CONFIG_VP8_DECODER
+#if CONFIG_VP9_DECODER
 #include "vpx/vp8dx.h"
 #endif
 
 static char *exec_name;
 static int   verbose = 0;
 
-static const struct
-{
-    const char *name;
-    const vpx_codec_iface_t *iface;
-} ifaces[] =
-{
-#if CONFIG_VP8_DECODER
-    {"vp8",  &vpx_codec_vp8_dx_algo},
+static const struct {
+  const char *name;
+  const vpx_codec_iface_t *iface;
+} ifaces[] = {
+#if CONFIG_VP9_DECODER
+  {"vp9",  &vpx_codec_vp8_dx_algo},
 #endif
 };
 
-static void usage_exit(void)
-{
-    int i;
+static void usage_exit(void) {
+  int i;
 
-    printf("Usage: %s <options>\n\n"
-           "Options:\n"
-           "\t--codec <name>\tCodec to use (default=%s)\n"
-           "\t-h <height>\tHeight of the simulated video frame, in pixels\n"
-           "\t-w <width> \tWidth of the simulated video frame, in pixels\n"
-           "\t-v         \tVerbose mode (show individual segment sizes)\n"
-           "\t--help     \tShow this message\n"
-           "\n"
-           "Included decoders:\n"
-           "\n",
-           exec_name,
-           ifaces[0].name);
+  printf("Usage: %s <options>\n\n"
+         "Options:\n"
+         "\t--codec <name>\tCodec to use (default=%s)\n"
+         "\t-h <height>\tHeight of the simulated video frame, in pixels\n"
+         "\t-w <width> \tWidth of the simulated video frame, in pixels\n"
+         "\t-v         \tVerbose mode (show individual segment sizes)\n"
+         "\t--help     \tShow this message\n"
+         "\n"
+         "Included decoders:\n"
+         "\n",
+         exec_name,
+         ifaces[0].name);
 
-    for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
-        printf("    %-6s - %s\n",
-               ifaces[i].name,
-               vpx_codec_iface_name(ifaces[i].iface));
+  for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
+    printf("    %-6s - %s\n",
+           ifaces[i].name,
+           vpx_codec_iface_name(ifaces[i].iface));
 
-    exit(EXIT_FAILURE);
+  exit(EXIT_FAILURE);
 }
 
-static void usage_error(const char *fmt, ...)
-{
-    va_list ap;
-    va_start(ap, fmt);
-    vprintf(fmt, ap);
-    printf("\n");
-    usage_exit();
+static void usage_error(const char *fmt, ...) {
+  va_list ap;
+  va_start(ap, fmt);
+  vprintf(fmt, ap);
+  printf("\n");
+  usage_exit();
 }
 
-void my_mem_dtor(vpx_codec_mmap_t *mmap)
-{
-    if (verbose)
-        printf("freeing segment %d\n", mmap->id);
+void my_mem_dtor(vpx_codec_mmap_t *mmap) {
+  if (verbose)
+    printf("freeing segment %d\n", mmap->id);
 
-    free(mmap->priv);
+  free(mmap->priv);
 }
 
-int main(int argc, char **argv)
-{
-    vpx_codec_ctx_t           decoder;
-    vpx_codec_iface_t        *iface = ifaces[0].iface;
-    vpx_codec_iter_t          iter;
-    vpx_codec_dec_cfg_t       cfg;
-    vpx_codec_err_t           res = VPX_CODEC_OK;
-    unsigned int            alloc_sz = 0;
-    unsigned int            w = 352;
-    unsigned int            h = 288;
-    int                     i;
+int main(int argc, char **argv) {
+  vpx_codec_ctx_t           decoder;
+  vpx_codec_iface_t        *iface = ifaces[0].iface;
+  vpx_codec_iter_t          iter;
+  vpx_codec_dec_cfg_t       cfg;
+  vpx_codec_err_t           res = VPX_CODEC_OK;
+  unsigned int            alloc_sz = 0;
+  unsigned int            w = 352;
+  unsigned int            h = 288;
+  int                     i;
 
-    exec_name = argv[0];
+  exec_name = argv[0];
 
-    for (i = 1; i < argc; i++)
-    {
-        if (!strcmp(argv[i], "--codec"))
-        {
-            if (i + 1 < argc)
-            {
-                int j, k = -1;
+  for (i = 1; i < argc; i++) {
+    if (!strcmp(argv[i], "--codec")) {
+      if (i + 1 < argc) {
+        int j, k = -1;
 
-                i++;
+        i++;
 
-                for (j = 0; j < sizeof(ifaces) / sizeof(ifaces[0]); j++)
-                    if (!strcmp(ifaces[j].name, argv[i]))
-                        k = j;
+        for (j = 0; j < sizeof(ifaces) / sizeof(ifaces[0]); j++)
+          if (!strcmp(ifaces[j].name, argv[i]))
+            k = j;
 
-                if (k >= 0)
-                    iface = ifaces[k].iface;
-                else
-                    usage_error("Error: Unrecognized argument (%s) to --codec\n",
-                                argv[i]);
-            }
-            else
-                usage_error("Error: Option --codec requires argument.\n");
-        }
-        else if (!strcmp(argv[i], "-v"))
-            verbose = 1;
-        else if (!strcmp(argv[i], "-h"))
-            if (i + 1 < argc)
-            {
-                h = atoi(argv[++i]);
-            }
-            else
-                usage_error("Error: Option -h requires argument.\n");
-        else if (!strcmp(argv[i], "-w"))
-            if (i + 1 < argc)
-            {
-                w = atoi(argv[++i]);
-            }
-            else
-                usage_error("Error: Option -w requires argument.\n");
-        else if (!strcmp(argv[i], "--help"))
-            usage_exit();
+        if (k >= 0)
+          iface = ifaces[k].iface;
         else
-            usage_error("Error: Unrecognized option %s\n\n", argv[i]);
-    }
+          usage_error("Error: Unrecognized argument (%s) to --codec\n",
+                      argv[i]);
+      } else
+        usage_error("Error: Option --codec requires argument.\n");
+    } else if (!strcmp(argv[i], "-v"))
+      verbose = 1;
+    else if (!strcmp(argv[i], "-h"))
+      if (i + 1 < argc) {
+        h = atoi(argv[++i]);
+      } else
+        usage_error("Error: Option -h requires argument.\n");
+    else if (!strcmp(argv[i], "-w"))
+      if (i + 1 < argc) {
+        w = atoi(argv[++i]);
+      } else
+        usage_error("Error: Option -w requires argument.\n");
+    else if (!strcmp(argv[i], "--help"))
+      usage_exit();
+    else
+      usage_error("Error: Unrecognized option %s\n\n", argv[i]);
+  }
 
-    if (argc == 1)
-        printf("Using built-in defaults. For options, rerun with --help\n\n");
+  if (argc == 1)
+    printf("Using built-in defaults. For options, rerun with --help\n\n");
 
-    /* XMA mode is not supported on all decoders! */
-    if (!(vpx_codec_get_caps(iface) & VPX_CODEC_CAP_XMA))
-    {
-        printf("%s does not support XMA mode!\n", vpx_codec_iface_name(iface));
+  /* XMA mode is not supported on all decoders! */
+  if (!(vpx_codec_get_caps(iface) & VPX_CODEC_CAP_XMA)) {
+    printf("%s does not support XMA mode!\n", vpx_codec_iface_name(iface));
+    return EXIT_FAILURE;
+  }
+
+  /* The codec knows how much memory to allocate based on the size of the
+   * encoded frames. This data can be parsed from the bitstream with
+   * vpx_codec_peek_stream_info() if a bitstream is available. Otherwise,
+   * a fixed size can be used that will be the upper limit on the frame
+   * size the decoder can decode.
+   */
+  cfg.w = w;
+  cfg.h = h;
+
+  /* Initialize the decoder in XMA mode. */
+  if (vpx_codec_dec_init(&decoder, iface, &cfg, VPX_CODEC_USE_XMA)) {
+    printf("Failed to initialize decoder in XMA mode: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
+
+  /* Iterate through the list of memory maps, allocating them with the
+   * requested alignment.
+   */
+  iter = NULL;
+
+  do {
+    vpx_codec_mmap_t  mmap;
+    unsigned int    align;
+
+    res = vpx_codec_get_mem_map(&decoder, &mmap, &iter);
+    align = mmap.align ? mmap.align - 1 : 0;
+
+    if (!res) {
+      if (verbose)
+        printf("Allocating segment %u, size %lu, align %u %s\n",
+               mmap.id, mmap.sz, mmap.align,
+               mmap.flags & VPX_CODEC_MEM_ZERO ? "(ZEROED)" : "");
+
+      if (mmap.flags & VPX_CODEC_MEM_ZERO)
+        mmap.priv = calloc(1, mmap.sz + align);
+      else
+        mmap.priv = malloc(mmap.sz + align);
+
+      mmap.base = (void *)((((uintptr_t)mmap.priv) + align) & ~(uintptr_t)align);
+      mmap.dtor = my_mem_dtor;
+      alloc_sz += mmap.sz + align;
+
+      if (vpx_codec_set_mem_map(&decoder, &mmap, 1)) {
+        printf("Failed to set mmap: %s\n", vpx_codec_error(&decoder));
         return EXIT_FAILURE;
+      }
+    } else if (res != VPX_CODEC_LIST_END) {
+      printf("Failed to get mmap: %s\n", vpx_codec_error(&decoder));
+      return EXIT_FAILURE;
     }
+  } while (res != VPX_CODEC_LIST_END);
 
-    /* The codec knows how much memory to allocate based on the size of the
-     * encoded frames. This data can be parsed from the bitstream with
-     * vpx_codec_peek_stream_info() if a bitstream is available. Otherwise,
-     * a fixed size can be used that will be the upper limit on the frame
-     * size the decoder can decode.
-     */
-    cfg.w = w;
-    cfg.h = h;
-
-    /* Initialize the decoder in XMA mode. */
-    if (vpx_codec_dec_init(&decoder, iface, &cfg, VPX_CODEC_USE_XMA))
-    {
-        printf("Failed to initialize decoder in XMA mode: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
-
-    /* Iterate through the list of memory maps, allocating them with the
-     * requested alignment.
-     */
-    iter = NULL;
-
-    do
-    {
-        vpx_codec_mmap_t  mmap;
-        unsigned int    align;
-
-        res = vpx_codec_get_mem_map(&decoder, &mmap, &iter);
-        align = mmap.align ? mmap.align - 1 : 0;
-
-        if (!res)
-        {
-            if (verbose)
-                printf("Allocating segment %u, size %lu, align %u %s\n",
-                       mmap.id, mmap.sz, mmap.align,
-                       mmap.flags & VPX_CODEC_MEM_ZERO ? "(ZEROED)" : "");
-
-            if (mmap.flags & VPX_CODEC_MEM_ZERO)
-                mmap.priv = calloc(1, mmap.sz + align);
-            else
-                mmap.priv = malloc(mmap.sz + align);
-
-            mmap.base = (void *)((((uintptr_t)mmap.priv) + align) & ~(uintptr_t)align);
-            mmap.dtor = my_mem_dtor;
-            alloc_sz += mmap.sz + align;
-
-            if (vpx_codec_set_mem_map(&decoder, &mmap, 1))
-            {
-                printf("Failed to set mmap: %s\n", vpx_codec_error(&decoder));
-                return EXIT_FAILURE;
-            }
-        }
-        else if (res != VPX_CODEC_LIST_END)
-        {
-            printf("Failed to get mmap: %s\n", vpx_codec_error(&decoder));
-            return EXIT_FAILURE;
-        }
-    }
-    while (res != VPX_CODEC_LIST_END);
-
-    printf("%s\n    %d bytes external memory required for %dx%d.\n",
-           decoder.name, alloc_sz, cfg.w, cfg.h);
-    vpx_codec_destroy(&decoder);
-    return EXIT_SUCCESS;
+  printf("%s\n    %d bytes external memory required for %dx%d.\n",
+         decoder.name, alloc_sz, cfg.w, cfg.h);
+  vpx_codec_destroy(&decoder);
+  return EXIT_SUCCESS;
 
 }
diff --git a/libvpx/examples.mk b/libvpx/examples.mk
index 90913e6..5b5ca23 100644
--- a/libvpx/examples.mk
+++ b/libvpx/examples.mk
@@ -8,6 +8,12 @@
 ##  be found in the AUTHORS file in the root of the source tree.
 ##
 
+LIBYUV_SRCS +=  third_party/libyuv/include/libyuv/basic_types.h  \
+                third_party/libyuv/include/libyuv/cpu_id.h  \
+                third_party/libyuv/include/libyuv/scale.h  \
+                third_party/libyuv/source/row.h \
+                third_party/libyuv/source/scale.c  \
+                third_party/libyuv/source/cpu_id.c
 
 # List of examples to build. UTILS are files that are taken from the source
 # tree directly, and GEN_EXAMPLES are files that are created from the
@@ -25,6 +31,7 @@
 vpxdec.SRCS                 += nestegg/halloc/src/macros.h
 vpxdec.SRCS                 += nestegg/include/nestegg/nestegg.h
 vpxdec.SRCS                 += nestegg/src/nestegg.c
+vpxdec.SRCS                 += $(LIBYUV_SRCS)
 vpxdec.GUID                  = BA5FE66F-38DD-E034-F542-B1578C5FB950
 vpxdec.DESCRIPTION           = Full featured decoder
 UTILS-$(CONFIG_ENCODERS)    += vpxenc.c
@@ -36,9 +43,10 @@
 vpxenc.SRCS                 += libmkv/EbmlIDs.h
 vpxenc.SRCS                 += libmkv/EbmlWriter.c
 vpxenc.SRCS                 += libmkv/EbmlWriter.h
+vpxenc.SRCS                 += $(LIBYUV_SRCS)
 vpxenc.GUID                  = 548DEC74-7A15-4B2B-AFC3-AA102E7C25C1
 vpxenc.DESCRIPTION           = Full featured encoder
-UTILS-$(CONFIG_ENCODERS)    += vp8_scalable_patterns.c
+UTILS-$(CONFIG_VP8_ENCODER)    += vp8_scalable_patterns.c
 vp8_scalable_patterns.GUID   = 0D6A210B-F482-4D6F-8570-4A9C01ACC88C
 vp8_scalable_patterns.DESCRIPTION = Temporal Scalability Encoder
 
@@ -56,37 +64,37 @@
 #example_xma.GUID             = A955FC4A-73F1-44F7-135E-30D84D32F022
 #example_xma.DESCRIPTION      = External Memory Allocation mode usage
 
-GEN_EXAMPLES-$(CONFIG_DECODERS) += simple_decoder.c
+GEN_EXAMPLES-$(CONFIG_VP8_DECODER) += simple_decoder.c
 simple_decoder.GUID              = D3BBF1E9-2427-450D-BBFF-B2843C1D44CC
 simple_decoder.DESCRIPTION       = Simplified decoder loop
-GEN_EXAMPLES-$(CONFIG_DECODERS) += postproc.c
+GEN_EXAMPLES-$(CONFIG_VP8_DECODER) += postproc.c
 postproc.GUID                    = 65E33355-F35E-4088-884D-3FD4905881D7
 postproc.DESCRIPTION             = Decoder postprocessor control
-GEN_EXAMPLES-$(CONFIG_DECODERS) += decode_to_md5.c
+GEN_EXAMPLES-$(CONFIG_VP8_DECODER) += decode_to_md5.c
 decode_to_md5.SRCS              += md5_utils.h md5_utils.c
 decode_to_md5.GUID               = 59120B9B-2735-4BFE-B022-146CA340FE42
 decode_to_md5.DESCRIPTION        = Frame by frame MD5 checksum
 
-GEN_EXAMPLES-$(CONFIG_ENCODERS) += simple_encoder.c
+GEN_EXAMPLES-$(CONFIG_VP8_ENCODER) += simple_encoder.c
 simple_encoder.GUID              = 4607D299-8A71-4D2C-9B1D-071899B6FBFD
 simple_encoder.DESCRIPTION       = Simplified encoder loop
-GEN_EXAMPLES-$(CONFIG_ENCODERS) += twopass_encoder.c
+GEN_EXAMPLES-$(CONFIG_VP8_ENCODER) += twopass_encoder.c
 twopass_encoder.GUID             = 73494FA6-4AF9-4763-8FBB-265C92402FD8
 twopass_encoder.DESCRIPTION      = Two-pass encoder loop
-GEN_EXAMPLES-$(CONFIG_ENCODERS) += force_keyframe.c
+GEN_EXAMPLES-$(CONFIG_VP8_ENCODER) += force_keyframe.c
 force_keyframe.GUID              = 3C67CADF-029F-4C86-81F5-D6D4F51177F0
 force_keyframe.DESCRIPTION       = Force generation of keyframes
 ifeq ($(CONFIG_DECODERS),yes)
-GEN_EXAMPLES-$(CONFIG_ENCODERS) += decode_with_drops.c
+GEN_EXAMPLES-$(CONFIG_VP8_ENCODER) += decode_with_drops.c
 endif
 decode_with_drops.GUID           = CE5C53C4-8DDA-438A-86ED-0DDD3CDB8D26
 decode_with_drops.DESCRIPTION    = Drops frames while decoding
-ifeq ($(CONFIG_DECODERS),yes)
+ifeq ($(CONFIG_VP8_DECODER),yes)
 GEN_EXAMPLES-$(CONFIG_ERROR_CONCEALMENT) += decode_with_partial_drops.c
 endif
 decode_with_partial_drops.GUID           = 61C2D026-5754-46AC-916F-1343ECC5537E
 decode_with_partial_drops.DESCRIPTION    = Drops parts of frames while decoding
-GEN_EXAMPLES-$(CONFIG_ENCODERS) += error_resilient.c
+GEN_EXAMPLES-$(CONFIG_VP8_ENCODER) += error_resilient.c
 error_resilient.GUID             = DF5837B9-4145-4F92-A031-44E4F832E00C
 error_resilient.DESCRIPTION      = Error Resiliency Feature
 
@@ -99,13 +107,7 @@
 
 # C file is provided, not generated automatically.
 UTILS-$(CONFIG_MULTI_RES_ENCODING) += vp8_multi_resolution_encoder.c
-vp8_multi_resolution_encoder.SRCS  \
-                         += third_party/libyuv/include/libyuv/basic_types.h  \
-                            third_party/libyuv/include/libyuv/cpu_id.h  \
-                            third_party/libyuv/include/libyuv/scale.h  \
-                            third_party/libyuv/source/row.h \
-                            third_party/libyuv/source/scale.c  \
-                            third_party/libyuv/source/cpu_id.c
+vp8_multi_resolution_encoder.SRCS         += $(LIBYUV_SRCS)
 vp8_multi_resolution_encoder.GUID         = 04f8738e-63c8-423b-90fa-7c2703a374de
 vp8_multi_resolution_encoder.DESCRIPTION  = VP8 Multiple-resolution Encoding
 
@@ -115,9 +117,11 @@
 # when building for bare-metal targets
 ifeq ($(CONFIG_OS_SUPPORT), yes)
 CODEC_EXTRA_LIBS-$(CONFIG_VP8)         += m
+CODEC_EXTRA_LIBS-$(CONFIG_VP9)         += m
 else
     ifeq ($(CONFIG_GCC), yes)
     CODEC_EXTRA_LIBS-$(CONFIG_VP8)         += m
+    CODEC_EXTRA_LIBS-$(CONFIG_VP9)         += m
     endif
 endif
 #
@@ -136,6 +140,8 @@
     LIB_PATH-yes                     += $(if $(BUILD_PFX),$(BUILD_PFX),.)
     INC_PATH-$(CONFIG_VP8_DECODER)   += $(SRC_PATH_BARE)/vp8
     INC_PATH-$(CONFIG_VP8_ENCODER)   += $(SRC_PATH_BARE)/vp8
+    INC_PATH-$(CONFIG_VP9_DECODER)   += $(SRC_PATH_BARE)/vp9
+    INC_PATH-$(CONFIG_VP9_ENCODER)   += $(SRC_PATH_BARE)/vp9
     LIB_PATH := $(call enabled,LIB_PATH)
     INC_PATH := $(call enabled,INC_PATH)
 endif
@@ -179,7 +185,8 @@
 
 # Instantiate linker template for all examples.
 CODEC_LIB=$(if $(CONFIG_DEBUG_LIBS),vpx_g,vpx)
-CODEC_LIB_SUF=$(if $(CONFIG_SHARED),.so,.a)
+SHARED_LIB_SUF=$(if $(filter darwin%,$(TGT_OS)),.dylib,.so)
+CODEC_LIB_SUF=$(if $(CONFIG_SHARED),$(SHARED_LIB_SUF),.a)
 $(foreach bin,$(BINS-yes),\
     $(if $(BUILD_OBJS),$(eval $(bin):\
         $(LIB_PATH)/lib$(CODEC_LIB)$(CODEC_LIB_SUF)))\
@@ -209,7 +216,7 @@
 
 # Set up additional MSVS environment
 ifeq ($(CONFIG_MSVS),yes)
-CODEC_LIB=$(if $(CONFIG_STATIC_MSVCRT),vpxmt,vpxmd)
+CODEC_LIB=$(if $(CONFIG_SHARED),vpx,$(if $(CONFIG_STATIC_MSVCRT),vpxmt,vpxmd))
 # This variable uses deferred expansion intentionally, since the results of
 # $(wildcard) may change during the course of the Make.
 VS_PLATFORMS = $(foreach d,$(wildcard */Release/$(CODEC_LIB).lib),$(word 1,$(subst /, ,$(d))))
@@ -224,19 +231,19 @@
 # even though there is no real dependency there (the dependency is on
 # the makefiles). We may want to revisit this.
 define vcproj_template
-$(1): $($(1:.vcproj=).SRCS)
+$(1): $($(1:.$(VCPROJ_SFX)=).SRCS) vpx.$(VCPROJ_SFX)
 	@echo "    [vcproj] $$@"
-	$$(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh\
+	$$(GEN_VCPROJ)\
             --exe\
             --target=$$(TOOLCHAIN)\
-            --name=$$(@:.vcproj=)\
+            --name=$$(@:.$(VCPROJ_SFX)=)\
             --ver=$$(CONFIG_VS_VERSION)\
-            --proj-guid=$$($$(@:.vcproj=).GUID)\
+            --proj-guid=$$($$(@:.$(VCPROJ_SFX)=).GUID)\
             $$(if $$(CONFIG_STATIC_MSVCRT),--static-crt) \
             --out=$$@ $$(INTERNAL_CFLAGS) $$(CFLAGS) \
-            $$(INTERNAL_LDFLAGS) $$(LDFLAGS) -l$$(CODEC_LIB) -lwinmm $$^
+            $$(INTERNAL_LDFLAGS) $$(LDFLAGS) -l$$(CODEC_LIB) $$^
 endef
-PROJECTS-$(CONFIG_MSVS) += $(ALL_EXAMPLES:.c=.vcproj)
+PROJECTS-$(CONFIG_MSVS) += $(ALL_EXAMPLES:.c=.$(VCPROJ_SFX))
 INSTALL-BINS-$(CONFIG_MSVS) += $(foreach p,$(VS_PLATFORMS),\
                                $(addprefix bin/$(p)/,$(ALL_EXAMPLES:.c=.exe)))
 $(foreach proj,$(call enabled,PROJECTS),\
diff --git a/libvpx/examples/decoder_tmpl.c b/libvpx/examples/decoder_tmpl.c
index 8194f0a..597fea2 100644
--- a/libvpx/examples/decoder_tmpl.c
+++ b/libvpx/examples/decoder_tmpl.c
@@ -12,6 +12,7 @@
 /*
 @*INTRODUCTION
  */
+#include "vpx_config.h"
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdarg.h>
diff --git a/libvpx/examples/decoder_tmpl.txt b/libvpx/examples/decoder_tmpl.txt
index e652a63..3d230a5 100644
--- a/libvpx/examples/decoder_tmpl.txt
+++ b/libvpx/examples/decoder_tmpl.txt
@@ -1,7 +1,7 @@
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ DEC_INCLUDES
 #define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_decoder.h"
-#include "vpx/vp8dx.h"
+#include "vpx/vp9dx.h"
 #define interface (vpx_codec_vp8_dx())
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ DEC_INCLUDES
 
diff --git a/libvpx/examples/encoder_tmpl.txt b/libvpx/examples/encoder_tmpl.txt
index 1afbd8b..9f8f4af 100644
--- a/libvpx/examples/encoder_tmpl.txt
+++ b/libvpx/examples/encoder_tmpl.txt
@@ -1,7 +1,7 @@
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ENC_INCLUDES
 #define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_encoder.h"
-#include "vpx/vp8cx.h"
+#include "vpx/vp9cx.h"
 #define interface (vpx_codec_vp8_cx())
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ENC_INCLUDES
 
diff --git a/libvpx/examples/postproc.txt b/libvpx/examples/postproc.txt
index 51b251a..e00bf59 100644
--- a/libvpx/examples/postproc.txt
+++ b/libvpx/examples/postproc.txt
@@ -51,7 +51,7 @@
 postprocessors. VP8 is one example. The following sample code toggles
 postprocessing on and off every 15 frames.
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PRE_DECODE
-#if CONFIG_VP8_DECODER
+#if CONFIG_VP9_DECODER
 if(frame_cnt%30 == 1) {
     vp8_postproc_cfg_t  pp = {0, 0, 0};
 
diff --git a/libvpx/libmkv/EbmlBufferWriter.c b/libvpx/libmkv/EbmlBufferWriter.c
index d9b04a8..574e478 100644
--- a/libvpx/libmkv/EbmlBufferWriter.c
+++ b/libvpx/libmkv/EbmlBufferWriter.c
@@ -1,60 +1,54 @@
-//#include <strmif.h>
+// #include <strmif.h>
 #include "EbmlBufferWriter.h"
 #include "EbmlWriter.h"
-//#include <cassert>
-//#include <limits>
-//#include <malloc.h>  //_alloca
+// #include <cassert>
+// #include <limits>
+// #include <malloc.h>  //_alloca
 #include <stdlib.h>
 #include <wchar.h>
 #include <string.h>
 
-void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
-{
-    unsigned char *src = glob->buf;
-    src += glob->offset;
-    memcpy(src, buffer_in, len);
-    glob->offset += len;
+void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len) {
+  unsigned char *src = glob->buf;
+  src += glob->offset;
+  memcpy(src, buffer_in, len);
+  glob->offset += len;
 }
 
-static void _Serialize(EbmlGlobal *glob, const unsigned char *p, const unsigned char *q)
-{
-    while (q != p)
-    {
-        --q;
+static void _Serialize(EbmlGlobal *glob, const unsigned char *p, const unsigned char *q) {
+  while (q != p) {
+    --q;
 
-        unsigned long cbWritten;
-        memcpy(&(glob->buf[glob->offset]), q, 1);
-        glob->offset ++;
-    }
+    unsigned long cbWritten;
+    memcpy(&(glob->buf[glob->offset]), q, 1);
+    glob->offset++;
+  }
 }
 
-void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
-{
-    //assert(buf);
+void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, unsigned long len) {
+  // assert(buf);
 
-    const unsigned char *const p = (const unsigned char *)(buffer_in);
-    const unsigned char *const q = p + len;
+  const unsigned char *const p = (const unsigned char *)(buffer_in);
+  const unsigned char *const q = p + len;
 
-    _Serialize(glob, p, q);
+  _Serialize(glob, p, q);
 }
 
 
-void Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc, unsigned long class_id)
-{
-    Ebml_WriteID(glob, class_id);
-    ebmlLoc->offset = glob->offset;
-    //todo this is always taking 8 bytes, this may need later optimization
-    unsigned long long unknownLen =  0x01FFFFFFFFFFFFFFLLU;
-    Ebml_Serialize(glob, (void *)&unknownLen, 8); //this is a key that says lenght unknown
+void Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc, unsigned long class_id) {
+  Ebml_WriteID(glob, class_id);
+  ebmlLoc->offset = glob->offset;
+  // todo this is always taking 8 bytes, this may need later optimization
+  unsigned long long unknownLen =  0x01FFFFFFFFFFFFFFLLU;
+  Ebml_Serialize(glob, (void *)&unknownLen, 8); // this is a key that says lenght unknown
 }
 
-void Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
-{
-    unsigned long long size = glob->offset - ebmlLoc->offset - 8;
-    unsigned long long curOffset = glob->offset;
-    glob->offset = ebmlLoc->offset;
-    size |=  0x0100000000000000LLU;
-    Ebml_Serialize(glob, &size, 8);
-    glob->offset = curOffset;
+void Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc) {
+  unsigned long long size = glob->offset - ebmlLoc->offset - 8;
+  unsigned long long curOffset = glob->offset;
+  glob->offset = ebmlLoc->offset;
+  size |=  0x0100000000000000LLU;
+  Ebml_Serialize(glob, &size, 8);
+  glob->offset = curOffset;
 }
 
diff --git a/libvpx/libmkv/EbmlBufferWriter.h b/libvpx/libmkv/EbmlBufferWriter.h
index ba0a9b3..acd5c2a 100644
--- a/libvpx/libmkv/EbmlBufferWriter.h
+++ b/libvpx/libmkv/EbmlBufferWriter.h
@@ -1,16 +1,14 @@
 #ifndef EBMLBUFFERWRITER_HPP
 #define EBMLBUFFERWRITER_HPP
 
-typedef struct
-{
-    unsigned long long offset;
+typedef struct {
+  unsigned long long offset;
 } EbmlLoc;
 
-typedef struct
-{
-    unsigned char *buf;
-    unsigned int length;
-    unsigned int offset;
+typedef struct {
+  unsigned char *buf;
+  unsigned int length;
+  unsigned int offset;
 } EbmlGlobal;
 
 
diff --git a/libvpx/libmkv/EbmlIDs.h b/libvpx/libmkv/EbmlIDs.h
index e3ce585..44d4385 100644
--- a/libvpx/libmkv/EbmlIDs.h
+++ b/libvpx/libmkv/EbmlIDs.h
@@ -12,35 +12,34 @@
 
 /* Commenting out values not available in webm, but available in matroska */
 
-enum mkv
-{
-    EBML = 0x1A45DFA3,
-    EBMLVersion = 0x4286,
-    EBMLReadVersion = 0x42F7,
-    EBMLMaxIDLength = 0x42F2,
-    EBMLMaxSizeLength = 0x42F3,
-    DocType = 0x4282,
-    DocTypeVersion = 0x4287,
-    DocTypeReadVersion = 0x4285,
+enum mkv {
+  EBML = 0x1A45DFA3,
+  EBMLVersion = 0x4286,
+  EBMLReadVersion = 0x42F7,
+  EBMLMaxIDLength = 0x42F2,
+  EBMLMaxSizeLength = 0x42F3,
+  DocType = 0x4282,
+  DocTypeVersion = 0x4287,
+  DocTypeReadVersion = 0x4285,
 /* CRC_32 = 0xBF, */
-    Void = 0xEC,
-    SignatureSlot = 0x1B538667,
-    SignatureAlgo = 0x7E8A,
-    SignatureHash = 0x7E9A,
-    SignaturePublicKey = 0x7EA5,
-    Signature = 0x7EB5,
-    SignatureElements = 0x7E5B,
-    SignatureElementList = 0x7E7B,
-    SignedElement = 0x6532,
-    /* segment */
-    Segment = 0x18538067,
-    /* Meta Seek Information */
-    SeekHead = 0x114D9B74,
-    Seek = 0x4DBB,
-    SeekID = 0x53AB,
-    SeekPosition = 0x53AC,
-    /* Segment Information */
-    Info = 0x1549A966,
+  Void = 0xEC,
+  SignatureSlot = 0x1B538667,
+  SignatureAlgo = 0x7E8A,
+  SignatureHash = 0x7E9A,
+  SignaturePublicKey = 0x7EA5,
+  Signature = 0x7EB5,
+  SignatureElements = 0x7E5B,
+  SignatureElementList = 0x7E7B,
+  SignedElement = 0x6532,
+  /* segment */
+  Segment = 0x18538067,
+  /* Meta Seek Information */
+  SeekHead = 0x114D9B74,
+  Seek = 0x4DBB,
+  SeekID = 0x53AB,
+  SeekPosition = 0x53AC,
+  /* Segment Information */
+  Info = 0x1549A966,
 /* SegmentUID = 0x73A4, */
 /* SegmentFilename = 0x7384, */
 /* PrevUID = 0x3CB923, */
@@ -52,61 +51,61 @@
 /* ChapterTranslateEditionUID = 0x69FC, */
 /* ChapterTranslateCodec = 0x69BF, */
 /* ChapterTranslateID = 0x69A5, */
-    TimecodeScale = 0x2AD7B1,
-    Segment_Duration = 0x4489,
-    DateUTC = 0x4461,
+  TimecodeScale = 0x2AD7B1,
+  Segment_Duration = 0x4489,
+  DateUTC = 0x4461,
 /* Title = 0x7BA9, */
-    MuxingApp = 0x4D80,
-    WritingApp = 0x5741,
-    /* Cluster */
-    Cluster = 0x1F43B675,
-    Timecode = 0xE7,
+  MuxingApp = 0x4D80,
+  WritingApp = 0x5741,
+  /* Cluster */
+  Cluster = 0x1F43B675,
+  Timecode = 0xE7,
 /* SilentTracks = 0x5854, */
 /* SilentTrackNumber = 0x58D7, */
 /* Position = 0xA7, */
-    PrevSize = 0xAB,
-    BlockGroup = 0xA0,
-    Block = 0xA1,
+  PrevSize = 0xAB,
+  BlockGroup = 0xA0,
+  Block = 0xA1,
 /* BlockVirtual = 0xA2, */
-/* BlockAdditions = 0x75A1, */
-/* BlockMore = 0xA6, */
-/* BlockAddID = 0xEE, */
-/* BlockAdditional = 0xA5, */
-    BlockDuration = 0x9B,
+  BlockAdditions = 0x75A1,
+  BlockMore = 0xA6,
+  BlockAddID = 0xEE,
+  BlockAdditional = 0xA5,
+  BlockDuration = 0x9B,
 /* ReferencePriority = 0xFA, */
-    ReferenceBlock = 0xFB,
+  ReferenceBlock = 0xFB,
 /* ReferenceVirtual = 0xFD, */
 /* CodecState = 0xA4, */
 /* Slices = 0x8E, */
 /* TimeSlice = 0xE8, */
-    LaceNumber = 0xCC,
+  LaceNumber = 0xCC,
 /* FrameNumber = 0xCD, */
 /* BlockAdditionID = 0xCB, */
 /* MkvDelay = 0xCE, */
 /* Cluster_Duration = 0xCF, */
-    SimpleBlock = 0xA3,
+  SimpleBlock = 0xA3,
 /* EncryptedBlock = 0xAF, */
-    /* Track */
-    Tracks = 0x1654AE6B,
-    TrackEntry = 0xAE,
-    TrackNumber = 0xD7,
-    TrackUID = 0x73C5,
-    TrackType = 0x83,
-    FlagEnabled = 0xB9,
-    FlagDefault = 0x88,
-    FlagForced = 0x55AA,
-    FlagLacing = 0x9C,
+  /* Track */
+  Tracks = 0x1654AE6B,
+  TrackEntry = 0xAE,
+  TrackNumber = 0xD7,
+  TrackUID = 0x73C5,
+  TrackType = 0x83,
+  FlagEnabled = 0xB9,
+  FlagDefault = 0x88,
+  FlagForced = 0x55AA,
+  FlagLacing = 0x9C,
 /* MinCache = 0x6DE7, */
 /* MaxCache = 0x6DF8, */
-    DefaultDuration = 0x23E383,
+  DefaultDuration = 0x23E383,
 /* TrackTimecodeScale = 0x23314F, */
 /* TrackOffset = 0x537F, */
-/* MaxBlockAdditionID = 0x55EE, */
-    Name = 0x536E,
-    Language = 0x22B59C,
-    CodecID = 0x86,
-    CodecPrivate = 0x63A2,
-    CodecName = 0x258688,
+  MaxBlockAdditionID = 0x55EE,
+  Name = 0x536E,
+  Language = 0x22B59C,
+  CodecID = 0x86,
+  CodecPrivate = 0x63A2,
+  CodecName = 0x258688,
 /* AttachmentLink = 0x7446, */
 /* CodecSettings = 0x3A9697, */
 /* CodecInfoURL = 0x3B4040, */
@@ -117,33 +116,34 @@
 /* TrackTranslateEditionUID = 0x66FC, */
 /* TrackTranslateCodec = 0x66BF, */
 /* TrackTranslateTrackID = 0x66A5, */
-    /* video */
-    Video = 0xE0,
-    FlagInterlaced = 0x9A,
-    StereoMode = 0x53B8,
-    PixelWidth = 0xB0,
-    PixelHeight = 0xBA,
-    PixelCropBottom = 0x54AA,
-    PixelCropTop = 0x54BB,
-    PixelCropLeft = 0x54CC,
-    PixelCropRight = 0x54DD,
-    DisplayWidth = 0x54B0,
-    DisplayHeight = 0x54BA,
-    DisplayUnit = 0x54B2,
-    AspectRatioType = 0x54B3,
+  /* video */
+  Video = 0xE0,
+  FlagInterlaced = 0x9A,
+  StereoMode = 0x53B8,
+  AlphaMode = 0x53C0,
+  PixelWidth = 0xB0,
+  PixelHeight = 0xBA,
+  PixelCropBottom = 0x54AA,
+  PixelCropTop = 0x54BB,
+  PixelCropLeft = 0x54CC,
+  PixelCropRight = 0x54DD,
+  DisplayWidth = 0x54B0,
+  DisplayHeight = 0x54BA,
+  DisplayUnit = 0x54B2,
+  AspectRatioType = 0x54B3,
 /* ColourSpace = 0x2EB524, */
 /* GammaValue = 0x2FB523, */
-    FrameRate = 0x2383E3,
-    /* end video */
-    /* audio */
-    Audio = 0xE1,
-    SamplingFrequency = 0xB5,
-    OutputSamplingFrequency = 0x78B5,
-    Channels = 0x9F,
+  FrameRate = 0x2383E3,
+  /* end video */
+  /* audio */
+  Audio = 0xE1,
+  SamplingFrequency = 0xB5,
+  OutputSamplingFrequency = 0x78B5,
+  Channels = 0x9F,
 /* ChannelPositions = 0x7D7B, */
-    BitDepth = 0x6264,
-    /* end audio */
-    /* content encoding */
+  BitDepth = 0x6264,
+  /* end audio */
+  /* content encoding */
 /* ContentEncodings = 0x6d80, */
 /* ContentEncoding = 0x6240, */
 /* ContentEncodingOrder = 0x5031, */
@@ -159,22 +159,22 @@
 /* ContentSigKeyID = 0x47e4, */
 /* ContentSigAlgo = 0x47e5, */
 /* ContentSigHashAlgo = 0x47e6, */
-    /* end content encoding */
-    /* Cueing Data */
-    Cues = 0x1C53BB6B,
-    CuePoint = 0xBB,
-    CueTime = 0xB3,
-    CueTrackPositions = 0xB7,
-    CueTrack = 0xF7,
-    CueClusterPosition = 0xF1,
-    CueBlockNumber = 0x5378
+  /* end content encoding */
+  /* Cueing Data */
+  Cues = 0x1C53BB6B,
+  CuePoint = 0xBB,
+  CueTime = 0xB3,
+  CueTrackPositions = 0xB7,
+  CueTrack = 0xF7,
+  CueClusterPosition = 0xF1,
+  CueBlockNumber = 0x5378
 /* CueCodecState = 0xEA, */
 /* CueReference = 0xDB, */
 /* CueRefTime = 0x96, */
 /* CueRefCluster = 0x97, */
 /* CueRefNumber = 0x535F, */
 /* CueRefCodecState = 0xEB, */
-    /* Attachment */
+  /* Attachment */
 /* Attachments = 0x1941A469, */
 /* AttachedFile = 0x61A7, */
 /* FileDescription = 0x467E, */
@@ -183,7 +183,7 @@
 /* FileData = 0x465C, */
 /* FileUID = 0x46AE, */
 /* FileReferral = 0x4675, */
-    /* Chapters */
+  /* Chapters */
 /* Chapters = 0x1043A770, */
 /* EditionEntry = 0x45B9, */
 /* EditionUID = 0x45BC, */
@@ -211,7 +211,7 @@
 /* ChapProcessCommand = 0x6911, */
 /* ChapProcessTime = 0x6922, */
 /* ChapProcessData = 0x6933, */
-    /* Tagging */
+  /* Tagging */
 /* Tags = 0x1254C367, */
 /* Tag = 0x7373, */
 /* Targets = 0x63C0, */
diff --git a/libvpx/libmkv/EbmlWriter.c b/libvpx/libmkv/EbmlWriter.c
index d70f06e..5fc5ed2 100644
--- a/libvpx/libmkv/EbmlWriter.c
+++ b/libvpx/libmkv/EbmlWriter.c
@@ -18,158 +18,140 @@
 #define LITERALU64(n) n##LLU
 #endif
 
-void Ebml_WriteLen(EbmlGlobal *glob, int64_t val)
-{
-    /* TODO check and make sure we are not > than 0x0100000000000000LLU */
-    unsigned char size = 8; /* size in bytes to output */
+void Ebml_WriteLen(EbmlGlobal *glob, int64_t val) {
+  /* TODO check and make sure we are not > than 0x0100000000000000LLU */
+  unsigned char size = 8; /* size in bytes to output */
 
-    /* mask to compare for byte size */
-    int64_t minVal = 0xff;
+  /* mask to compare for byte size */
+  int64_t minVal = 0xff;
 
-    for (size = 1; size < 8; size ++)
-    {
-        if (val < minVal)
-            break;
+  for (size = 1; size < 8; size ++) {
+    if (val < minVal)
+      break;
 
-        minVal = (minVal << 7);
+    minVal = (minVal << 7);
+  }
+
+  val |= (((uint64_t)0x80) << ((size - 1) * 7));
+
+  Ebml_Serialize(glob, (void *) &val, sizeof(val), size);
+}
+
+void Ebml_WriteString(EbmlGlobal *glob, const char *str) {
+  const size_t size_ = strlen(str);
+  const uint64_t  size = size_;
+  Ebml_WriteLen(glob, size);
+  /* TODO: it's not clear from the spec whether the nul terminator
+   * should be serialized too.  For now we omit the null terminator.
+   */
+  Ebml_Write(glob, str, (unsigned long)size);
+}
+
+void Ebml_WriteUTF8(EbmlGlobal *glob, const wchar_t *wstr) {
+  const size_t strlen = wcslen(wstr);
+
+  /* TODO: it's not clear from the spec whether the nul terminator
+   * should be serialized too.  For now we include it.
+   */
+  const uint64_t  size = strlen;
+
+  Ebml_WriteLen(glob, size);
+  Ebml_Write(glob, wstr, (unsigned long)size);
+}
+
+void Ebml_WriteID(EbmlGlobal *glob, unsigned long class_id) {
+  int len;
+
+  if (class_id >= 0x01000000)
+    len = 4;
+  else if (class_id >= 0x00010000)
+    len = 3;
+  else if (class_id >= 0x00000100)
+    len = 2;
+  else
+    len = 1;
+
+  Ebml_Serialize(glob, (void *)&class_id, sizeof(class_id), len);
+}
+
+void Ebml_SerializeUnsigned64(EbmlGlobal *glob, unsigned long class_id, uint64_t ui) {
+  unsigned char sizeSerialized = 8 | 0x80;
+  Ebml_WriteID(glob, class_id);
+  Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
+  Ebml_Serialize(glob, &ui, sizeof(ui), 8);
+}
+
+void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned long ui) {
+  unsigned char size = 8; /* size in bytes to output */
+  unsigned char sizeSerialized = 0;
+  unsigned long minVal;
+
+  Ebml_WriteID(glob, class_id);
+  minVal = 0x7fLU; /* mask to compare for byte size */
+
+  for (size = 1; size < 4; size ++) {
+    if (ui < minVal) {
+      break;
     }
 
-    val |= (((uint64_t)0x80) << ((size - 1) * 7));
+    minVal <<= 7;
+  }
 
-    Ebml_Serialize(glob, (void *) &val, sizeof(val), size);
-}
-
-void Ebml_WriteString(EbmlGlobal *glob, const char *str)
-{
-    const size_t size_ = strlen(str);
-    const uint64_t  size = size_;
-    Ebml_WriteLen(glob, size);
-    /* TODO: it's not clear from the spec whether the nul terminator
-     * should be serialized too.  For now we omit the null terminator.
-     */
-    Ebml_Write(glob, str, (unsigned long)size);
-}
-
-void Ebml_WriteUTF8(EbmlGlobal *glob, const wchar_t *wstr)
-{
-    const size_t strlen = wcslen(wstr);
-
-    /* TODO: it's not clear from the spec whether the nul terminator
-     * should be serialized too.  For now we include it.
-     */
-    const uint64_t  size = strlen;
-
-    Ebml_WriteLen(glob, size);
-    Ebml_Write(glob, wstr, (unsigned long)size);
-}
-
-void Ebml_WriteID(EbmlGlobal *glob, unsigned long class_id)
-{
-    int len;
-
-    if (class_id >= 0x01000000)
-        len = 4;
-    else if (class_id >= 0x00010000)
-        len = 3;
-    else if (class_id >= 0x00000100)
-        len = 2;
-    else
-        len = 1;
-
-    Ebml_Serialize(glob, (void *)&class_id, sizeof(class_id), len);
-}
-
-void Ebml_SerializeUnsigned64(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
-{
-    unsigned char sizeSerialized = 8 | 0x80;
-    Ebml_WriteID(glob, class_id);
-    Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
-    Ebml_Serialize(glob, &ui, sizeof(ui), 8);
-}
-
-void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned long ui)
-{
-    unsigned char size = 8; /* size in bytes to output */
-    unsigned char sizeSerialized = 0;
-    unsigned long minVal;
-
-    Ebml_WriteID(glob, class_id);
-    minVal = 0x7fLU; /* mask to compare for byte size */
-
-    for (size = 1; size < 4; size ++)
-    {
-        if (ui < minVal)
-        {
-            break;
-        }
-
-        minVal <<= 7;
-    }
-
-    sizeSerialized = 0x80 | size;
-    Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
-    Ebml_Serialize(glob, &ui, sizeof(ui), size);
+  sizeSerialized = 0x80 | size;
+  Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
+  Ebml_Serialize(glob, &ui, sizeof(ui), size);
 }
 /* TODO: perhaps this is a poor name for this id serializer helper function */
-void Ebml_SerializeBinary(EbmlGlobal *glob, unsigned long class_id, unsigned long bin)
-{
-    int size;
-    for (size=4; size > 1; size--)
-    {
-        if (bin & 0x000000ff << ((size-1) * 8))
-            break;
-    }
-    Ebml_WriteID(glob, class_id);
-    Ebml_WriteLen(glob, size);
-    Ebml_WriteID(glob, bin);
+void Ebml_SerializeBinary(EbmlGlobal *glob, unsigned long class_id, unsigned long bin) {
+  int size;
+  for (size = 4; size > 1; size--) {
+    if (bin & 0x000000ff << ((size - 1) * 8))
+      break;
+  }
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteLen(glob, size);
+  Ebml_WriteID(glob, bin);
 }
 
-void Ebml_SerializeFloat(EbmlGlobal *glob, unsigned long class_id, double d)
-{
-    unsigned char len = 0x88;
+void Ebml_SerializeFloat(EbmlGlobal *glob, unsigned long class_id, double d) {
+  unsigned char len = 0x88;
 
-    Ebml_WriteID(glob, class_id);
-    Ebml_Serialize(glob, &len, sizeof(len), 1);
-    Ebml_Serialize(glob,  &d, sizeof(d), 8);
+  Ebml_WriteID(glob, class_id);
+  Ebml_Serialize(glob, &len, sizeof(len), 1);
+  Ebml_Serialize(glob,  &d, sizeof(d), 8);
 }
 
-void Ebml_WriteSigned16(EbmlGlobal *glob, short val)
-{
-    signed long out = ((val & 0x003FFFFF) | 0x00200000) << 8;
-    Ebml_Serialize(glob, &out, sizeof(out), 3);
+void Ebml_WriteSigned16(EbmlGlobal *glob, short val) {
+  signed long out = ((val & 0x003FFFFF) | 0x00200000) << 8;
+  Ebml_Serialize(glob, &out, sizeof(out), 3);
 }
 
-void Ebml_SerializeString(EbmlGlobal *glob, unsigned long class_id, const char *s)
-{
-    Ebml_WriteID(glob, class_id);
-    Ebml_WriteString(glob, s);
+void Ebml_SerializeString(EbmlGlobal *glob, unsigned long class_id, const char *s) {
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteString(glob, s);
 }
 
-void Ebml_SerializeUTF8(EbmlGlobal *glob, unsigned long class_id, wchar_t *s)
-{
-    Ebml_WriteID(glob,  class_id);
-    Ebml_WriteUTF8(glob,  s);
+void Ebml_SerializeUTF8(EbmlGlobal *glob, unsigned long class_id, wchar_t *s) {
+  Ebml_WriteID(glob,  class_id);
+  Ebml_WriteUTF8(glob,  s);
 }
 
-void Ebml_SerializeData(EbmlGlobal *glob, unsigned long class_id, unsigned char *data, unsigned long data_length)
-{
-    Ebml_WriteID(glob, class_id);
-    Ebml_WriteLen(glob, data_length);
-    Ebml_Write(glob,  data, data_length);
+void Ebml_SerializeData(EbmlGlobal *glob, unsigned long class_id, unsigned char *data, unsigned long data_length) {
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteLen(glob, data_length);
+  Ebml_Write(glob,  data, data_length);
 }
 
-void Ebml_WriteVoid(EbmlGlobal *glob, unsigned long vSize)
-{
-    unsigned char tmp = 0;
-    unsigned long i = 0;
+void Ebml_WriteVoid(EbmlGlobal *glob, unsigned long vSize) {
+  unsigned char tmp = 0;
+  unsigned long i = 0;
 
-    Ebml_WriteID(glob, 0xEC);
-    Ebml_WriteLen(glob, vSize);
+  Ebml_WriteID(glob, 0xEC);
+  Ebml_WriteLen(glob, vSize);
 
-    for (i = 0; i < vSize; i++)
-    {
-        Ebml_Write(glob, &tmp, 1);
-    }
+  for (i = 0; i < vSize; i++) {
+    Ebml_Write(glob, &tmp, 1);
+  }
 }
 
 /* TODO Serialize Date */
diff --git a/libvpx/libmkv/WebMElement.c b/libvpx/libmkv/WebMElement.c
index 0ef5100..2f79a3c 100644
--- a/libvpx/libmkv/WebMElement.c
+++ b/libvpx/libmkv/WebMElement.c
@@ -14,106 +14,100 @@
 
 #define kVorbisPrivateMaxSize  4000
 
-void writeHeader(EbmlGlobal *glob)
-{
-    EbmlLoc start;
-    Ebml_StartSubElement(glob, &start, EBML);
-    Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
-    Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
-    Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
-    Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
-    Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
-    Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
-    Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
-    Ebml_EndSubElement(glob, &start);
+void writeHeader(EbmlGlobal *glob) {
+  EbmlLoc start;
+  Ebml_StartSubElement(glob, &start, EBML);
+  Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
+  Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); // EBML Read Version
+  Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); // EBML Max ID Length
+  Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); // EBML Max Size Length
+  Ebml_SerializeString(glob, DocType, "webm"); // Doc Type
+  Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); // Doc Type Version
+  Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); // Doc Type Read Version
+  Ebml_EndSubElement(glob, &start);
 }
 
 void writeSimpleBlock(EbmlGlobal *glob, unsigned char trackNumber, short timeCode,
                       int isKeyframe, unsigned char lacingFlag, int discardable,
-                      unsigned char *data, unsigned long dataLength)
-{
-    Ebml_WriteID(glob, SimpleBlock);
-    unsigned long blockLength = 4 + dataLength;
-    blockLength |= 0x10000000; //TODO check length < 0x0FFFFFFFF
-    Ebml_Serialize(glob, &blockLength, sizeof(blockLength), 4);
-    trackNumber |= 0x80;  //TODO check track nubmer < 128
-    Ebml_Write(glob, &trackNumber, 1);
-    //Ebml_WriteSigned16(glob, timeCode,2); //this is 3 bytes
-    Ebml_Serialize(glob, &timeCode, sizeof(timeCode), 2);
-    unsigned char flags = 0x00 | (isKeyframe ? 0x80 : 0x00) | (lacingFlag << 1) | discardable;
-    Ebml_Write(glob, &flags, 1);
-    Ebml_Write(glob, data, dataLength);
+                      unsigned char *data, unsigned long dataLength) {
+  Ebml_WriteID(glob, SimpleBlock);
+  unsigned long blockLength = 4 + dataLength;
+  blockLength |= 0x10000000; // TODO check length < 0x0FFFFFFFF
+  Ebml_Serialize(glob, &blockLength, sizeof(blockLength), 4);
+  trackNumber |= 0x80;  // TODO check track nubmer < 128
+  Ebml_Write(glob, &trackNumber, 1);
+  // Ebml_WriteSigned16(glob, timeCode,2); //this is 3 bytes
+  Ebml_Serialize(glob, &timeCode, sizeof(timeCode), 2);
+  unsigned char flags = 0x00 | (isKeyframe ? 0x80 : 0x00) | (lacingFlag << 1) | discardable;
+  Ebml_Write(glob, &flags, 1);
+  Ebml_Write(glob, data, dataLength);
 }
 
-static UInt64 generateTrackID(unsigned int trackNumber)
-{
-    UInt64 t = time(NULL) * trackNumber;
-    UInt64 r = rand();
-    r = r << 32;
-    r +=  rand();
-    UInt64 rval = t ^ r;
-    return rval;
+static UInt64 generateTrackID(unsigned int trackNumber) {
+  UInt64 t = time(NULL) * trackNumber;
+  UInt64 r = rand();
+  r = r << 32;
+  r +=  rand();
+  UInt64 rval = t ^ r;
+  return rval;
 }
 
 void writeVideoTrack(EbmlGlobal *glob, unsigned int trackNumber, int flagLacing,
                      char *codecId, unsigned int pixelWidth, unsigned int pixelHeight,
-                     double frameRate)
-{
-    EbmlLoc start;
-    Ebml_StartSubElement(glob, &start, TrackEntry);
-    Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
-    UInt64 trackID = generateTrackID(trackNumber);
-    Ebml_SerializeUnsigned(glob, TrackUID, trackID);
-    Ebml_SerializeString(glob, CodecName, "VP8");  //TODO shouldn't be fixed
+                     double frameRate) {
+  EbmlLoc start;
+  Ebml_StartSubElement(glob, &start, TrackEntry);
+  Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
+  UInt64 trackID = generateTrackID(trackNumber);
+  Ebml_SerializeUnsigned(glob, TrackUID, trackID);
+  Ebml_SerializeString(glob, CodecName, "VP8");  // TODO shouldn't be fixed
 
-    Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
-    Ebml_SerializeString(glob, CodecID, codecId);
-    {
-        EbmlLoc videoStart;
-        Ebml_StartSubElement(glob, &videoStart, Video);
-        Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
-        Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
-        Ebml_SerializeFloat(glob, FrameRate, frameRate);
-        Ebml_EndSubElement(glob, &videoStart); //Video
-    }
-    Ebml_EndSubElement(glob, &start); //Track Entry
+  Ebml_SerializeUnsigned(glob, TrackType, 1); // video is always 1
+  Ebml_SerializeString(glob, CodecID, codecId);
+  {
+    EbmlLoc videoStart;
+    Ebml_StartSubElement(glob, &videoStart, Video);
+    Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
+    Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
+    Ebml_SerializeFloat(glob, FrameRate, frameRate);
+    Ebml_EndSubElement(glob, &videoStart); // Video
+  }
+  Ebml_EndSubElement(glob, &start); // Track Entry
 }
 void writeAudioTrack(EbmlGlobal *glob, unsigned int trackNumber, int flagLacing,
                      char *codecId, double samplingFrequency, unsigned int channels,
-                     unsigned char *private, unsigned long privateSize)
-{
-    EbmlLoc start;
-    Ebml_StartSubElement(glob, &start, TrackEntry);
-    Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
-    UInt64 trackID = generateTrackID(trackNumber);
-    Ebml_SerializeUnsigned(glob, TrackUID, trackID);
-    Ebml_SerializeUnsigned(glob, TrackType, 2); //audio is always 2
-    //I am using defaults for thesed required fields
-    /*  Ebml_SerializeUnsigned(glob, FlagEnabled, 1);
-        Ebml_SerializeUnsigned(glob, FlagDefault, 1);
-        Ebml_SerializeUnsigned(glob, FlagForced, 1);
-        Ebml_SerializeUnsigned(glob, FlagLacing, flagLacing);*/
-    Ebml_SerializeString(glob, CodecID, codecId);
-    Ebml_SerializeData(glob, CodecPrivate, private, privateSize);
+                     unsigned char *private, unsigned long privateSize) {
+  EbmlLoc start;
+  Ebml_StartSubElement(glob, &start, TrackEntry);
+  Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
+  UInt64 trackID = generateTrackID(trackNumber);
+  Ebml_SerializeUnsigned(glob, TrackUID, trackID);
+  Ebml_SerializeUnsigned(glob, TrackType, 2); // audio is always 2
+  // I am using defaults for thesed required fields
+  /*  Ebml_SerializeUnsigned(glob, FlagEnabled, 1);
+      Ebml_SerializeUnsigned(glob, FlagDefault, 1);
+      Ebml_SerializeUnsigned(glob, FlagForced, 1);
+      Ebml_SerializeUnsigned(glob, FlagLacing, flagLacing);*/
+  Ebml_SerializeString(glob, CodecID, codecId);
+  Ebml_SerializeData(glob, CodecPrivate, private, privateSize);
 
-    Ebml_SerializeString(glob, CodecName, "VORBIS");  //fixed for now
-    {
-        EbmlLoc AudioStart;
-        Ebml_StartSubElement(glob, &AudioStart, Audio);
-        Ebml_SerializeFloat(glob, SamplingFrequency, samplingFrequency);
-        Ebml_SerializeUnsigned(glob, Channels, channels);
-        Ebml_EndSubElement(glob, &AudioStart);
-    }
-    Ebml_EndSubElement(glob, &start);
+  Ebml_SerializeString(glob, CodecName, "VORBIS");  // fixed for now
+  {
+    EbmlLoc AudioStart;
+    Ebml_StartSubElement(glob, &AudioStart, Audio);
+    Ebml_SerializeFloat(glob, SamplingFrequency, samplingFrequency);
+    Ebml_SerializeUnsigned(glob, Channels, channels);
+    Ebml_EndSubElement(glob, &AudioStart);
+  }
+  Ebml_EndSubElement(glob, &start);
 }
-void writeSegmentInformation(EbmlGlobal *ebml, EbmlLoc* startInfo, unsigned long timeCodeScale, double duration)
-{
-    Ebml_StartSubElement(ebml, startInfo, Info);
-    Ebml_SerializeUnsigned(ebml, TimecodeScale, timeCodeScale);
-    Ebml_SerializeFloat(ebml, Segment_Duration, duration * 1000.0); //Currently fixed to using milliseconds
-    Ebml_SerializeString(ebml, 0x4D80, "QTmuxingAppLibWebM-0.0.1");
-    Ebml_SerializeString(ebml, 0x5741, "QTwritingAppLibWebM-0.0.1");
-    Ebml_EndSubElement(ebml, startInfo);
+void writeSegmentInformation(EbmlGlobal *ebml, EbmlLoc *startInfo, unsigned long timeCodeScale, double duration) {
+  Ebml_StartSubElement(ebml, startInfo, Info);
+  Ebml_SerializeUnsigned(ebml, TimecodeScale, timeCodeScale);
+  Ebml_SerializeFloat(ebml, Segment_Duration, duration * 1000.0); // Currently fixed to using milliseconds
+  Ebml_SerializeString(ebml, 0x4D80, "QTmuxingAppLibWebM-0.0.1");
+  Ebml_SerializeString(ebml, 0x5741, "QTwritingAppLibWebM-0.0.1");
+  Ebml_EndSubElement(ebml, startInfo);
 }
 
 /*
@@ -142,7 +136,7 @@
         Ebml_SerializeString(ebml_out, 0x7384, segmentInformation.filename);
     Ebml_SerializeUnsigned(ebml_out, 0x2AD7B1, segmentInformation.TimecodeScale);
     Ebml_SerializeUnsigned(ebml_out, 0x4489, segmentInformation.Duration);
-    //TODO date
+    // TODO date
     Ebml_SerializeWString(ebml_out, 0x4D80, L"MKVMUX");
     Ebml_SerializeWString(ebml_out, 0x5741, segmentInformation.WritingApp);
 }
@@ -173,9 +167,9 @@
 void Mkv_WriteVideoTrack(Ebml& ebml_out, TrackStruct & track, VideoTrackStruct& video)
 {
     EbmlLoc trackHeadLoc, videoHeadLoc;
-    Ebml_StartSubElement(ebml_out, trackHeadLoc, 0xAE);  //start Track
+    Ebml_StartSubElement(ebml_out, trackHeadLoc, 0xAE);  // start Track
     Mkv_WriteGenericTrackData(ebml_out, track);
-    Ebml_StartSubElement(ebml_out, videoHeadLoc, 0xE0);  //start Video
+    Ebml_StartSubElement(ebml_out, videoHeadLoc, 0xE0);  // start Video
     Ebml_SerializeUnsigned(ebml_out, 0x9A, video.FlagInterlaced ? 1 :0);
     Ebml_SerializeUnsigned(ebml_out, 0xB0, video.PixelWidth);
     Ebml_SerializeUnsigned(ebml_out, 0xBA, video.PixelHeight);
@@ -193,7 +187,7 @@
     EbmlLoc trackHeadLoc, audioHeadLoc;
     Ebml_StartSubElement(ebml_out, trackHeadLoc, 0xAE);
     Mkv_WriteGenericTrackData(ebml_out, track);
-    Ebml_StartSubElement(ebml_out, audioHeadLoc, 0xE0);  //start Audio
+    Ebml_StartSubElement(ebml_out, audioHeadLoc, 0xE0);  // start Audio
     Ebml_SerializeFloat(ebml_out, 0xB5, video.SamplingFrequency);
     Ebml_SerializeUnsigned(ebml_out, 0x9F, video.Channels);
     Ebml_SerializeUnsigned(ebml_out, 0x6264, video.BitDepth);
@@ -213,7 +207,7 @@
     Ebml_Write1UInt(ebml_out, block.TrackNumber);
     Ebml_WriteSigned16(ebml_out,block.TimeCode);
     unsigned char flags = 0x00 | (block.iskey ? 0x80:0x00) | (block.lacing << 1) | block.discardable;
-    Ebml_Write1UInt(ebml_out, flags);  //TODO this may be the wrong function
+    Ebml_Write1UInt(ebml_out, flags);  // TODO this may be the wrong function
     Ebml_Serialize(ebml_out, block.data, block.dataLength);
     Ebml_EndSubElement(ebml_out,ebmlLoc);
 }
diff --git a/libvpx/libmkv/WebMElement.h b/libvpx/libmkv/WebMElement.h
index b4208f2..d9ad0a0 100644
--- a/libvpx/libmkv/WebMElement.h
+++ b/libvpx/libmkv/WebMElement.h
@@ -17,8 +17,8 @@
 
 // these are helper functions
 void writeHeader(EbmlGlobal *ebml);
-void writeSegmentInformation(EbmlGlobal *ebml, EbmlLoc* startInfo , unsigned long timeCodeScale, double duration);
-//this function is a helper only, it assumes a lot of defaults
+void writeSegmentInformation(EbmlGlobal *ebml, EbmlLoc *startInfo, unsigned long timeCodeScale, double duration);
+// this function is a helper only, it assumes a lot of defaults
 void writeVideoTrack(EbmlGlobal *ebml, unsigned int trackNumber, int flagLacing,
                      char *codecId, unsigned int pixelWidth, unsigned int pixelHeight,
                      double frameRate);
diff --git a/libvpx/libmkv/testlibmkv.c b/libvpx/libmkv/testlibmkv.c
index 7edfc43..97bcf95 100644
--- a/libvpx/libmkv/testlibmkv.c
+++ b/libvpx/libmkv/testlibmkv.c
@@ -13,51 +13,50 @@
 #include "WebMElement.h"
 
 #include <stdio.h>
-int main(int argc, char *argv[])
-{
-    //init the datatype we're using for ebml output
-    unsigned char data[8192];
-    EbmlGlobal ebml;
-    ebml.buf = data;
-    ebml.offset = 0;
-    ebml.length = 8192;
+int main(int argc, char *argv[]) {
+  // init the datatype we're using for ebml output
+  unsigned char data[8192];
+  EbmlGlobal ebml;
+  ebml.buf = data;
+  ebml.offset = 0;
+  ebml.length = 8192;
 
-    writeHeader(&ebml);
+  writeHeader(&ebml);
+  {
+    EbmlLoc startSegment;
+    Ebml_StartSubElement(&ebml, &startSegment, Segment); // segment
     {
-        EbmlLoc startSegment;
-        Ebml_StartSubElement(&ebml, &startSegment, Segment); //segment
-        {
-            //segment info
-            EbmlLoc startInfo;
-            Ebml_StartSubElement(&ebml, &startInfo, Info);
-            Ebml_SerializeString(&ebml, 0x4D80, "muxingAppLibMkv");
-            Ebml_SerializeString(&ebml, 0x5741, "writingAppLibMkv");
-            Ebml_EndSubElement(&ebml, &startInfo);
-        }
-
-        {
-            EbmlLoc trackStart;
-            Ebml_StartSubElement(&ebml, &trackStart, Tracks);
-            writeVideoTrack(&ebml, 1, 1, "V_MS/VFW/FOURCC", 320, 240, 29.97);
-            //writeAudioTrack(&ebml,2,1, "A_VORBIS", 32000, 1, NULL, 0);
-            Ebml_EndSubElement(&ebml, &trackStart);
-        }
-
-        {
-            EbmlLoc clusterStart;
-            Ebml_StartSubElement(&ebml, &clusterStart, Cluster); //cluster
-            Ebml_SerializeUnsigned(&ebml, Timecode, 0);
-
-            unsigned char someData[4] = {1, 2, 3, 4};
-            writeSimpleBlock(&ebml, 1, 0, 1, 0, 0, someData, 4);
-            Ebml_EndSubElement(&ebml, &clusterStart);
-        }    //end cluster
-        Ebml_EndSubElement(&ebml, &startSegment);
+      // segment info
+      EbmlLoc startInfo;
+      Ebml_StartSubElement(&ebml, &startInfo, Info);
+      Ebml_SerializeString(&ebml, 0x4D80, "muxingAppLibMkv");
+      Ebml_SerializeString(&ebml, 0x5741, "writingAppLibMkv");
+      Ebml_EndSubElement(&ebml, &startInfo);
     }
 
-    //dump ebml stuff to the file
-    FILE *file_out = fopen("test.mkv", "wb");
-    size_t bytesWritten = fwrite(data, 1, ebml.offset, file_out);
-    fclose(file_out);
-    return 0;
+    {
+      EbmlLoc trackStart;
+      Ebml_StartSubElement(&ebml, &trackStart, Tracks);
+      writeVideoTrack(&ebml, 1, 1, "V_MS/VFW/FOURCC", 320, 240, 29.97);
+      // writeAudioTrack(&ebml,2,1, "A_VORBIS", 32000, 1, NULL, 0);
+      Ebml_EndSubElement(&ebml, &trackStart);
+    }
+
+    {
+      EbmlLoc clusterStart;
+      Ebml_StartSubElement(&ebml, &clusterStart, Cluster); // cluster
+      Ebml_SerializeUnsigned(&ebml, Timecode, 0);
+
+      unsigned char someData[4] = {1, 2, 3, 4};
+      writeSimpleBlock(&ebml, 1, 0, 1, 0, 0, someData, 4);
+      Ebml_EndSubElement(&ebml, &clusterStart);
+    }    // end cluster
+    Ebml_EndSubElement(&ebml, &startSegment);
+  }
+
+  // dump ebml stuff to the file
+  FILE *file_out = fopen("test.mkv", "wb");
+  size_t bytesWritten = fwrite(data, 1, ebml.offset, file_out);
+  fclose(file_out);
+  return 0;
 }
\ No newline at end of file
diff --git a/libvpx/libs.mk b/libvpx/libs.mk
index 4115dd8..f7ed95b 100644
--- a/libvpx/libs.mk
+++ b/libvpx/libs.mk
@@ -12,11 +12,51 @@
 # ARM assembly files are written in RVCT-style. We use some make magic to
 # filter those files to allow GCC compilation
 ifeq ($(ARCH_ARM),yes)
-  ASM:=$(if $(filter yes,$(CONFIG_GCC)),.asm.s,.asm)
+  ASM:=$(if $(filter yes,$(CONFIG_GCC)$(CONFIG_MSVS)),.asm.s,.asm)
 else
   ASM:=.asm
 endif
 
+#
+# Calculate platform- and compiler-specific offsets for hand coded assembly
+#
+ifeq ($(filter icc gcc,$(TGT_CC)), $(TGT_CC))
+OFFSET_PATTERN:='^[a-zA-Z0-9_]* EQU'
+define asm_offsets_template
+$$(BUILD_PFX)$(1): $$(BUILD_PFX)$(2).S
+	@echo "    [CREATE] $$@"
+	$$(qexec)LC_ALL=C grep $$(OFFSET_PATTERN) $$< | tr -d '$$$$\#' $$(ADS2GAS) > $$@
+$$(BUILD_PFX)$(2).S: $(2)
+CLEAN-OBJS += $$(BUILD_PFX)$(1) $(2).S
+endef
+else
+  ifeq ($(filter rvct,$(TGT_CC)), $(TGT_CC))
+define asm_offsets_template
+$$(BUILD_PFX)$(1): obj_int_extract
+$$(BUILD_PFX)$(1): $$(BUILD_PFX)$(2).o
+	@echo "    [CREATE] $$@"
+	$$(qexec)./obj_int_extract rvds $$< $$(ADS2GAS) > $$@
+OBJS-yes += $$(BUILD_PFX)$(2).o
+CLEAN-OBJS += $$(BUILD_PFX)$(1)
+$$(filter %$$(ASM).o,$$(OBJS-yes)): $$(BUILD_PFX)$(1)
+endef
+endif # rvct
+endif # !gcc
+
+#
+# Rule to generate runtime cpu detection files
+#
+define rtcd_h_template
+$$(BUILD_PFX)$(1).h: $$(SRC_PATH_BARE)/$(2)
+	@echo "    [CREATE] $$@"
+	$$(qexec)$$(SRC_PATH_BARE)/build/make/rtcd.sh --arch=$$(TGT_ISA) \
+          --sym=$(1) \
+          --config=$$(CONFIG_DIR)$$(target)$$(if $$(FAT_ARCHS),,-$$(TOOLCHAIN)).mk \
+          $$(RTCD_OPTIONS) $$^ > $$@
+CLEAN-OBJS += $$(BUILD_PFX)$(1).h
+RTCD += $$(BUILD_PFX)$(1).h
+endef
+
 CODEC_SRCS-yes += CHANGELOG
 CODEC_SRCS-yes += libs.mk
 
@@ -40,9 +80,12 @@
 include $(SRC_PATH_BARE)/vpx_ports/vpx_ports.mk
 CODEC_SRCS-yes += $(addprefix vpx_ports/,$(call enabled,PORTS_SRCS))
 
+ifneq ($(CONFIG_VP8_ENCODER)$(CONFIG_VP8_DECODER),)
+  VP8_PREFIX=vp8/
+  include $(SRC_PATH_BARE)/$(VP8_PREFIX)vp8_common.mk
+endif
 
 ifeq ($(CONFIG_VP8_ENCODER),yes)
-  VP8_PREFIX=vp8/
   include $(SRC_PATH_BARE)/$(VP8_PREFIX)vp8cx.mk
   CODEC_SRCS-yes += $(addprefix $(VP8_PREFIX),$(call enabled,VP8_CX_SRCS))
   CODEC_EXPORTS-yes += $(addprefix $(VP8_PREFIX),$(VP8_CX_EXPORTS))
@@ -52,7 +95,6 @@
 endif
 
 ifeq ($(CONFIG_VP8_DECODER),yes)
-  VP8_PREFIX=vp8/
   include $(SRC_PATH_BARE)/$(VP8_PREFIX)vp8dx.mk
   CODEC_SRCS-yes += $(addprefix $(VP8_PREFIX),$(call enabled,VP8_DX_SRCS))
   CODEC_EXPORTS-yes += $(addprefix $(VP8_PREFIX),$(VP8_DX_EXPORTS))
@@ -61,6 +103,35 @@
   CODEC_DOC_SECTIONS += vp8 vp8_decoder
 endif
 
+ifneq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_DECODER),)
+  VP9_PREFIX=vp9/
+  include $(SRC_PATH_BARE)/$(VP9_PREFIX)vp9_common.mk
+endif
+
+ifeq ($(CONFIG_VP9_ENCODER),yes)
+  VP9_PREFIX=vp9/
+  include $(SRC_PATH_BARE)/$(VP9_PREFIX)vp9cx.mk
+  CODEC_SRCS-yes += $(addprefix $(VP9_PREFIX),$(call enabled,VP9_CX_SRCS))
+  CODEC_EXPORTS-yes += $(addprefix $(VP9_PREFIX),$(VP9_CX_EXPORTS))
+  CODEC_SRCS-yes += $(VP9_PREFIX)vp9cx.mk vpx/vp8.h vpx/vp8cx.h
+  INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8cx.h
+  INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP9_PREFIX)/%
+  CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8cx.h
+  CODEC_DOC_SECTIONS += vp9 vp9_encoder
+endif
+
+ifeq ($(CONFIG_VP9_DECODER),yes)
+  VP9_PREFIX=vp9/
+  include $(SRC_PATH_BARE)/$(VP9_PREFIX)vp9dx.mk
+  CODEC_SRCS-yes += $(addprefix $(VP9_PREFIX),$(call enabled,VP9_DX_SRCS))
+  CODEC_EXPORTS-yes += $(addprefix $(VP9_PREFIX),$(VP9_DX_EXPORTS))
+  CODEC_SRCS-yes += $(VP9_PREFIX)vp9dx.mk vpx/vp8.h vpx/vp8dx.h
+  INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8dx.h
+  INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP9_PREFIX)/%
+  CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8dx.h
+  CODEC_DOC_SECTIONS += vp9 vp9_decoder
+endif
+
 
 ifeq ($(CONFIG_ENCODERS),yes)
   CODEC_DOC_SECTIONS += encoder
@@ -91,8 +162,13 @@
 
 CODEC_SRCS-$(BUILD_LIBVPX) += build/make/version.sh
 CODEC_SRCS-$(BUILD_LIBVPX) += build/make/rtcd.sh
+CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/emmintrin_compat.h
+CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/vpx_once.h
 CODEC_SRCS-$(BUILD_LIBVPX) += $(BUILD_PFX)vpx_config.c
 INSTALL-SRCS-no += $(BUILD_PFX)vpx_config.c
+ifeq ($(ARCH_X86)$(ARCH_X86_64),yes)
+CODEC_SRCS-$(BUILD_LIBVPX) += third_party/x86inc/x86inc.asm
+endif
 CODEC_EXPORTS-$(BUILD_LIBVPX) += vpx/exports_com
 CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_enc
 CODEC_EXPORTS-$(CONFIG_DECODERS) += vpx/exports_dec
@@ -116,7 +192,7 @@
 INSTALL-LIBS-$(CONFIG_DEBUG_LIBS) += $(LIBSUBDIR)/libvpx_g.a
 endif
 
-CODEC_SRCS=$(filter-out %_test.cc,$(call enabled,CODEC_SRCS))
+CODEC_SRCS=$(call enabled,CODEC_SRCS)
 INSTALL-SRCS-$(CONFIG_CODEC_SRCS) += $(CODEC_SRCS)
 INSTALL-SRCS-$(CONFIG_CODEC_SRCS) += $(call enabled,CODEC_EXPORTS)
 
@@ -131,10 +207,10 @@
 ifeq ($(CONFIG_EXTERNAL_BUILD),yes)
 ifeq ($(CONFIG_MSVS),yes)
 
-obj_int_extract.vcproj: $(SRC_PATH_BARE)/build/make/obj_int_extract.c
-	@cp $(SRC_PATH_BARE)/build/x86-msvs/obj_int_extract.bat .
+obj_int_extract.$(VCPROJ_SFX): $(SRC_PATH_BARE)/build/make/obj_int_extract.c
+	@cp $(SRC_PATH_BARE)/build/$(MSVS_ARCH_DIR)/obj_int_extract.bat .
 	@echo "    [CREATE] $@"
-	$(qexec)$(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
+	$(qexec)$(GEN_VCPROJ) \
     --exe \
     --target=$(TOOLCHAIN) \
     --name=obj_int_extract \
@@ -145,8 +221,7 @@
     -I. \
     -I"$(SRC_PATH_BARE)" \
 
-PROJECTS-$(BUILD_LIBVPX) += obj_int_extract.vcproj
-PROJECTS-$(BUILD_LIBVPX) += obj_int_extract.bat
+PROJECTS-$(BUILD_LIBVPX) += obj_int_extract.$(VCPROJ_SFX)
 
 vpx.def: $(call enabled,CODEC_EXPORTS)
 	@echo "    [CREATE] $@"
@@ -155,11 +230,11 @@
             --out=$@ $^
 CLEAN-OBJS += vpx.def
 
-vpx.vcproj: $(CODEC_SRCS) vpx.def
+vpx.$(VCPROJ_SFX): $(CODEC_SRCS) vpx.def obj_int_extract.$(VCPROJ_SFX)
 	@echo "    [CREATE] $@"
-	$(qexec)$(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
-			--lib \
-			--target=$(TOOLCHAIN) \
+	$(qexec)$(GEN_VCPROJ) \
+            $(if $(CONFIG_SHARED),--dll,--lib) \
+            --target=$(TOOLCHAIN) \
             $(if $(CONFIG_STATIC_MSVCRT),--static-crt) \
             --name=vpx \
             --proj-guid=DCE19DAF-69AC-46DB-B14A-39F0FAA5DB74 \
@@ -168,10 +243,10 @@
             --out=$@ $(CFLAGS) $^ \
             --src-path-bare="$(SRC_PATH_BARE)" \
 
-PROJECTS-$(BUILD_LIBVPX) += vpx.vcproj
+PROJECTS-$(BUILD_LIBVPX) += vpx.$(VCPROJ_SFX)
 
-vpx.vcproj: vpx_config.asm
-vpx.vcproj: vpx_rtcd.h
+vpx.$(VCPROJ_SFX): vpx_config.asm
+vpx.$(VCPROJ_SFX): $(RTCD)
 
 endif
 else
@@ -180,17 +255,29 @@
 LIBS-$(if $(BUILD_LIBVPX),$(CONFIG_STATIC)) += $(BUILD_PFX)libvpx.a $(BUILD_PFX)libvpx_g.a
 $(BUILD_PFX)libvpx_g.a: $(LIBVPX_OBJS)
 
+
 BUILD_LIBVPX_SO         := $(if $(BUILD_LIBVPX),$(CONFIG_SHARED))
+
+ifeq ($(filter darwin%,$(TGT_OS)),$(TGT_OS))
+LIBVPX_SO               := libvpx.$(VERSION_MAJOR).dylib
+EXPORT_FILE             := libvpx.syms
+LIBVPX_SO_SYMLINKS      := $(addprefix $(LIBSUBDIR)/, \
+                             libvpx.dylib  )
+else
 LIBVPX_SO               := libvpx.so.$(VERSION_MAJOR).$(VERSION_MINOR).$(VERSION_PATCH)
-LIBS-$(BUILD_LIBVPX_SO) += $(BUILD_PFX)$(LIBVPX_SO)\
-                           $(notdir $(LIBVPX_SO_SYMLINKS))
-$(BUILD_PFX)$(LIBVPX_SO): $(LIBVPX_OBJS) libvpx.ver
-$(BUILD_PFX)$(LIBVPX_SO): extralibs += -lm
-$(BUILD_PFX)$(LIBVPX_SO): SONAME = libvpx.so.$(VERSION_MAJOR)
-$(BUILD_PFX)$(LIBVPX_SO): SO_VERSION_SCRIPT = libvpx.ver
+EXPORT_FILE             := libvpx.ver
+SYM_LINK                := libvpx.so
 LIBVPX_SO_SYMLINKS      := $(addprefix $(LIBSUBDIR)/, \
                              libvpx.so libvpx.so.$(VERSION_MAJOR) \
                              libvpx.so.$(VERSION_MAJOR).$(VERSION_MINOR))
+endif
+
+LIBS-$(BUILD_LIBVPX_SO) += $(BUILD_PFX)$(LIBVPX_SO)\
+                           $(notdir $(LIBVPX_SO_SYMLINKS))
+$(BUILD_PFX)$(LIBVPX_SO): $(LIBVPX_OBJS) $(EXPORT_FILE)
+$(BUILD_PFX)$(LIBVPX_SO): extralibs += -lm
+$(BUILD_PFX)$(LIBVPX_SO): SONAME = libvpx.so.$(VERSION_MAJOR)
+$(BUILD_PFX)$(LIBVPX_SO): EXPORTS_FILE = $(EXPORT_FILE)
 
 libvpx.ver: $(call enabled,CODEC_EXPORTS)
 	@echo "    [CREATE] $@"
@@ -199,10 +286,16 @@
 	$(qexec)echo "local: *; };" >> $@
 CLEAN-OBJS += libvpx.ver
 
+libvpx.syms: $(call enabled,CODEC_EXPORTS)
+	@echo "    [CREATE] $@"
+	$(qexec)awk '{print "_"$$2}' $^ >$@
+CLEAN-OBJS += libvpx.syms
+
 define libvpx_symlink_template
 $(1): $(2)
-	@echo "    [LN]      $$@"
-	$(qexec)ln -sf $(LIBVPX_SO) $$@
+	@echo "    [LN]     $(2) $$@"
+	$(qexec)mkdir -p $$(dir $$@)
+	$(qexec)ln -sf $(2) $$@
 endef
 
 $(eval $(call libvpx_symlink_template,\
@@ -210,10 +303,12 @@
     $(BUILD_PFX)$(LIBVPX_SO)))
 $(eval $(call libvpx_symlink_template,\
     $(addprefix $(DIST_DIR)/,$(LIBVPX_SO_SYMLINKS)),\
-    $(DIST_DIR)/$(LIBSUBDIR)/$(LIBVPX_SO)))
+    $(LIBVPX_SO)))
 
-INSTALL-LIBS-$(CONFIG_SHARED) += $(LIBVPX_SO_SYMLINKS)
-INSTALL-LIBS-$(CONFIG_SHARED) += $(LIBSUBDIR)/$(LIBVPX_SO)
+
+INSTALL-LIBS-$(BUILD_LIBVPX_SO) += $(LIBVPX_SO_SYMLINKS)
+INSTALL-LIBS-$(BUILD_LIBVPX_SO) += $(LIBSUBDIR)/$(LIBVPX_SO)
+
 
 LIBS-$(BUILD_LIBVPX) += vpx.pc
 vpx.pc: config.mk libs.mk
@@ -229,8 +324,12 @@
 	$(qexec)echo 'Version: $(VERSION_MAJOR).$(VERSION_MINOR).$(VERSION_PATCH)' >> $@
 	$(qexec)echo 'Requires:' >> $@
 	$(qexec)echo 'Conflicts:' >> $@
-	$(qexec)echo 'Libs: -L$${libdir} -lvpx' >> $@
+	$(qexec)echo 'Libs: -L$${libdir} -lvpx -lm' >> $@
+ifeq ($(HAVE_PTHREAD_H),yes)
 	$(qexec)echo 'Libs.private: -lm -lpthread' >> $@
+else
+	$(qexec)echo 'Libs.private: -lm' >> $@
+endif
 	$(qexec)echo 'Cflags: -I$${includedir}' >> $@
 INSTALL-LIBS-yes += $(LIBSUBDIR)/pkgconfig/vpx.pc
 INSTALL_MAPS += $(LIBSUBDIR)/pkgconfig/%.pc %.pc
@@ -265,71 +364,10 @@
 $(filter %.s.o,$(OBJS-yes)):     $(BUILD_PFX)vpx_config.asm
 $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
 
-#
-# Calculate platform- and compiler-specific offsets for hand coded assembly
-#
-
-OFFSET_PATTERN:='^[a-zA-Z0-9_]* EQU'
-
-ifeq ($(filter icc gcc,$(TGT_CC)), $(TGT_CC))
-    $(BUILD_PFX)asm_com_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S
-	@echo "    [CREATE] $@"
-	$(qexec)LC_ALL=C grep $(OFFSET_PATTERN) $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S: $(VP8_PREFIX)common/asm_com_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_com_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S
-
-    $(BUILD_PFX)asm_enc_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S
-	@echo "    [CREATE] $@"
-	$(qexec)LC_ALL=C grep $(OFFSET_PATTERN) $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S: $(VP8_PREFIX)encoder/asm_enc_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_enc_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S
-
-    $(BUILD_PFX)asm_dec_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S
-	@echo "    [CREATE] $@"
-	$(qexec)LC_ALL=C grep $(OFFSET_PATTERN) $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S: $(VP8_PREFIX)decoder/asm_dec_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_dec_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S
-else
-  ifeq ($(filter rvct,$(TGT_CC)), $(TGT_CC))
-    asm_com_offsets.asm: obj_int_extract
-    asm_com_offsets.asm: $(VP8_PREFIX)common/asm_com_offsets.c.o
-	@echo "    [CREATE] $@"
-	$(qexec)./obj_int_extract rvds $< $(ADS2GAS) > $@
-    OBJS-yes += $(VP8_PREFIX)common/asm_com_offsets.c.o
-    CLEAN-OBJS += asm_com_offsets.asm
-    $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_com_offsets.asm
-
-    asm_enc_offsets.asm: obj_int_extract
-    asm_enc_offsets.asm: $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
-	@echo "    [CREATE] $@"
-	$(qexec)./obj_int_extract rvds $< $(ADS2GAS) > $@
-    OBJS-yes += $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
-    CLEAN-OBJS += asm_enc_offsets.asm
-    $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_enc_offsets.asm
-
-    asm_dec_offsets.asm: obj_int_extract
-    asm_dec_offsets.asm: $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
-	@echo "    [CREATE] $@"
-	$(qexec)./obj_int_extract rvds $< $(ADS2GAS) > $@
-    OBJS-yes += $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
-    CLEAN-OBJS += asm_dec_offsets.asm
-    $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_dec_offsets.asm
-  endif
-endif
 
 $(shell $(SRC_PATH_BARE)/build/make/version.sh "$(SRC_PATH_BARE)" $(BUILD_PFX)vpx_version.h)
 CLEAN-OBJS += $(BUILD_PFX)vpx_version.h
 
-#
-# Rule to generate runtime cpu detection files
-#
-$(BUILD_PFX)vpx_rtcd.h: $(SRC_PATH_BARE)/$(sort $(filter %rtcd_defs.sh,$(CODEC_SRCS)))
-	@echo "    [CREATE] $@"
-	$(qexec)$(SRC_PATH_BARE)/build/make/rtcd.sh --arch=$(TGT_ISA) \
-          --sym=vpx_rtcd \
-          --config=$(target)$(if $(FAT_ARCHS),,-$(TOOLCHAIN)).mk \
-          $(RTCD_OPTIONS) $^ > $@
-CLEAN-OBJS += $(BUILD_PFX)vpx_rtcd.h
 
 ##
 ## libvpx test directives
@@ -339,7 +377,7 @@
 
 include $(SRC_PATH_BARE)/test/test.mk
 LIBVPX_TEST_SRCS=$(addprefix test/,$(call enabled,LIBVPX_TEST_SRCS))
-LIBVPX_TEST_BINS=./test_libvpx
+LIBVPX_TEST_BINS=./test_libvpx$(EXE_SFX)
 LIBVPX_TEST_DATA=$(addprefix $(LIBVPX_TEST_DATA_PATH)/,\
                      $(call enabled,LIBVPX_TEST_DATA))
 libvpx_test_data_url=http://downloads.webmproject.org/test_data/libvpx/$(1)
@@ -365,9 +403,9 @@
 ifeq ($(CONFIG_EXTERNAL_BUILD),yes)
 ifeq ($(CONFIG_MSVS),yes)
 
-gtest.vcproj: $(SRC_PATH_BARE)/third_party/googletest/src/src/gtest-all.cc
+gtest.$(VCPROJ_SFX): $(SRC_PATH_BARE)/third_party/googletest/src/src/gtest-all.cc
 	@echo "    [CREATE] $@"
-	$(qexec)$(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
+	$(qexec)$(GEN_VCPROJ) \
             --lib \
             --target=$(TOOLCHAIN) \
             $(if $(CONFIG_STATIC_MSVCRT),--static-crt) \
@@ -375,28 +413,29 @@
             --proj-guid=EC00E1EC-AF68-4D92-A255-181690D1C9B1 \
             --ver=$(CONFIG_VS_VERSION) \
             --src-path-bare="$(SRC_PATH_BARE)" \
-            --out=gtest.vcproj $(SRC_PATH_BARE)/third_party/googletest/src/src/gtest-all.cc \
+            -D_VARIADIC_MAX=10 \
+            --out=gtest.$(VCPROJ_SFX) $(SRC_PATH_BARE)/third_party/googletest/src/src/gtest-all.cc \
             -I. -I"$(SRC_PATH_BARE)/third_party/googletest/src/include" -I"$(SRC_PATH_BARE)/third_party/googletest/src"
 
-PROJECTS-$(CONFIG_MSVS) += gtest.vcproj
+PROJECTS-$(CONFIG_MSVS) += gtest.$(VCPROJ_SFX)
 
-test_libvpx.vcproj: $(LIBVPX_TEST_SRCS)
+test_libvpx.$(VCPROJ_SFX): $(LIBVPX_TEST_SRCS) vpx.$(VCPROJ_SFX) gtest.$(VCPROJ_SFX)
 	@echo "    [CREATE] $@"
-	$(qexec)$(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
+	$(qexec)$(GEN_VCPROJ) \
             --exe \
             --target=$(TOOLCHAIN) \
             --name=test_libvpx \
+            -D_VARIADIC_MAX=10 \
             --proj-guid=CD837F5F-52D8-4314-A370-895D614166A7 \
             --ver=$(CONFIG_VS_VERSION) \
             $(if $(CONFIG_STATIC_MSVCRT),--static-crt) \
             --out=$@ $(INTERNAL_CFLAGS) $(CFLAGS) \
             -I. -I"$(SRC_PATH_BARE)/third_party/googletest/src/include" \
-            -L. -l$(CODEC_LIB) -lwinmm -l$(GTEST_LIB) $^
+            -L. -l$(CODEC_LIB) -l$(GTEST_LIB) $^
 
-PROJECTS-$(CONFIG_MSVS) += test_libvpx.vcproj
+PROJECTS-$(CONFIG_MSVS) += test_libvpx.$(VCPROJ_SFX)
 
-test:: testdata
-	@set -e; for t in $(addprefix Win32/Release/,$(notdir $(LIBVPX_TEST_BINS:.cc=.exe))); do $$t; done
+LIBVPX_TEST_BINS := $(addprefix $(TGT_OS:win64=x64)/Release/,$(notdir $(LIBVPX_TEST_BINS)))
 endif
 else
 
@@ -431,10 +470,24 @@
         )))\
     $(if $(LIPO_LIBS),$(eval $(call lipo_bin_template,$(bin))))\
 
-test:: $(LIBVPX_TEST_BINS) testdata
-	@set -e; for t in $(LIBVPX_TEST_BINS); do $$t; done
-
 endif
+
+define test_shard_template
+test:: test_shard.$(1)
+test_shard.$(1): $(LIBVPX_TEST_BINS) testdata
+	@set -e; \
+	 for t in $(LIBVPX_TEST_BINS); do \
+	   export GTEST_SHARD_INDEX=$(1); \
+	   export GTEST_TOTAL_SHARDS=$(2); \
+	   $$$$t; \
+	 done
+.PHONY: test_shard.$(1)
+endef
+
+NUM_SHARDS := 10
+SHARDS := 0 1 2 3 4 5 6 7 8 9
+$(foreach s,$(SHARDS),$(eval $(call test_shard_template,$(s),$(NUM_SHARDS))))
+
 endif
 
 ##
@@ -450,5 +503,8 @@
 	@echo "INCLUDE_PATH += ." >> $@;
 	@echo "ENABLED_SECTIONS += $(sort $(CODEC_DOC_SECTIONS))" >> $@
 
-## Generate vpx_rtcd.h for all objects
-$(OBJS-yes:.o=.d): $(BUILD_PFX)vpx_rtcd.h
+## Generate rtcd.h for all objects
+$(OBJS-yes:.o=.d): $(RTCD)
+
+## Update the global src list
+SRCS += $(CODEC_SRCS) $(LIBVPX_TEST_SRCS) $(GTEST_SRCS)
diff --git a/libvpx/md5_utils.c b/libvpx/md5_utils.c
index 9a584fa..8fb26e2 100644
--- a/libvpx/md5_utils.c
+++ b/libvpx/md5_utils.c
@@ -25,25 +25,22 @@
 #include "md5_utils.h"
 
 void
-byteSwap(UWORD32 *buf, unsigned words)
-{
-    md5byte *p;
+byteSwap(UWORD32 *buf, unsigned words) {
+  md5byte *p;
 
-    /* Only swap bytes for big endian machines */
-    int i = 1;
+  /* Only swap bytes for big endian machines */
+  int i = 1;
 
-    if (*(char *)&i == 1)
-        return;
+  if (*(char *)&i == 1)
+    return;
 
-    p = (md5byte *)buf;
+  p = (md5byte *)buf;
 
-    do
-    {
-        *buf++ = (UWORD32)((unsigned)p[3] << 8 | p[2]) << 16 |
-                 ((unsigned)p[1] << 8 | p[0]);
-        p += 4;
-    }
-    while (--words);
+  do {
+    *buf++ = (UWORD32)((unsigned)p[3] << 8 | p[2]) << 16 |
+             ((unsigned)p[1] << 8 | p[0]);
+    p += 4;
+  } while (--words);
 }
 
 /*
@@ -51,15 +48,14 @@
  * initialization constants.
  */
 void
-MD5Init(struct MD5Context *ctx)
-{
-    ctx->buf[0] = 0x67452301;
-    ctx->buf[1] = 0xefcdab89;
-    ctx->buf[2] = 0x98badcfe;
-    ctx->buf[3] = 0x10325476;
+MD5Init(struct MD5Context *ctx) {
+  ctx->buf[0] = 0x67452301;
+  ctx->buf[1] = 0xefcdab89;
+  ctx->buf[2] = 0x98badcfe;
+  ctx->buf[3] = 0x10325476;
 
-    ctx->bytes[0] = 0;
-    ctx->bytes[1] = 0;
+  ctx->bytes[0] = 0;
+  ctx->bytes[1] = 0;
 }
 
 /*
@@ -67,44 +63,41 @@
  * of bytes.
  */
 void
-MD5Update(struct MD5Context *ctx, md5byte const *buf, unsigned len)
-{
-    UWORD32 t;
+MD5Update(struct MD5Context *ctx, md5byte const *buf, unsigned len) {
+  UWORD32 t;
 
-    /* Update byte count */
+  /* Update byte count */
 
-    t = ctx->bytes[0];
+  t = ctx->bytes[0];
 
-    if ((ctx->bytes[0] = t + len) < t)
-        ctx->bytes[1]++;  /* Carry from low to high */
+  if ((ctx->bytes[0] = t + len) < t)
+    ctx->bytes[1]++;  /* Carry from low to high */
 
-    t = 64 - (t & 0x3f);  /* Space available in ctx->in (at least 1) */
+  t = 64 - (t & 0x3f);  /* Space available in ctx->in (at least 1) */
 
-    if (t > len)
-    {
-        memcpy((md5byte *)ctx->in + 64 - t, buf, len);
-        return;
-    }
+  if (t > len) {
+    memcpy((md5byte *)ctx->in + 64 - t, buf, len);
+    return;
+  }
 
-    /* First chunk is an odd size */
-    memcpy((md5byte *)ctx->in + 64 - t, buf, t);
+  /* First chunk is an odd size */
+  memcpy((md5byte *)ctx->in + 64 - t, buf, t);
+  byteSwap(ctx->in, 16);
+  MD5Transform(ctx->buf, ctx->in);
+  buf += t;
+  len -= t;
+
+  /* Process data in 64-byte chunks */
+  while (len >= 64) {
+    memcpy(ctx->in, buf, 64);
     byteSwap(ctx->in, 16);
     MD5Transform(ctx->buf, ctx->in);
-    buf += t;
-    len -= t;
+    buf += 64;
+    len -= 64;
+  }
 
-    /* Process data in 64-byte chunks */
-    while (len >= 64)
-    {
-        memcpy(ctx->in, buf, 64);
-        byteSwap(ctx->in, 16);
-        MD5Transform(ctx->buf, ctx->in);
-        buf += 64;
-        len -= 64;
-    }
-
-    /* Handle any remaining bytes of data. */
-    memcpy(ctx->in, buf, len);
+  /* Handle any remaining bytes of data. */
+  memcpy(ctx->in, buf, len);
 }
 
 /*
@@ -112,37 +105,35 @@
  * 1 0* (64-bit count of bits processed, MSB-first)
  */
 void
-MD5Final(md5byte digest[16], struct MD5Context *ctx)
-{
-    int count = ctx->bytes[0] & 0x3f; /* Number of bytes in ctx->in */
-    md5byte *p = (md5byte *)ctx->in + count;
+MD5Final(md5byte digest[16], struct MD5Context *ctx) {
+  int count = ctx->bytes[0] & 0x3f; /* Number of bytes in ctx->in */
+  md5byte *p = (md5byte *)ctx->in + count;
 
-    /* Set the first char of padding to 0x80.  There is always room. */
-    *p++ = 0x80;
+  /* Set the first char of padding to 0x80.  There is always room. */
+  *p++ = 0x80;
 
-    /* Bytes of padding needed to make 56 bytes (-8..55) */
-    count = 56 - 1 - count;
+  /* Bytes of padding needed to make 56 bytes (-8..55) */
+  count = 56 - 1 - count;
 
-    if (count < 0)    /* Padding forces an extra block */
-    {
-        memset(p, 0, count + 8);
-        byteSwap(ctx->in, 16);
-        MD5Transform(ctx->buf, ctx->in);
-        p = (md5byte *)ctx->in;
-        count = 56;
-    }
-
-    memset(p, 0, count);
-    byteSwap(ctx->in, 14);
-
-    /* Append length in bits and transform */
-    ctx->in[14] = ctx->bytes[0] << 3;
-    ctx->in[15] = ctx->bytes[1] << 3 | ctx->bytes[0] >> 29;
+  if (count < 0) {  /* Padding forces an extra block */
+    memset(p, 0, count + 8);
+    byteSwap(ctx->in, 16);
     MD5Transform(ctx->buf, ctx->in);
+    p = (md5byte *)ctx->in;
+    count = 56;
+  }
 
-    byteSwap(ctx->buf, 4);
-    memcpy(digest, ctx->buf, 16);
-    memset(ctx, 0, sizeof(*ctx)); /* In case it's sensitive */
+  memset(p, 0, count);
+  byteSwap(ctx->in, 14);
+
+  /* Append length in bits and transform */
+  ctx->in[14] = ctx->bytes[0] << 3;
+  ctx->in[15] = ctx->bytes[1] << 3 | ctx->bytes[0] >> 29;
+  MD5Transform(ctx->buf, ctx->in);
+
+  byteSwap(ctx->buf, 4);
+  memcpy(digest, ctx->buf, 16);
+  memset(ctx, 0, sizeof(*ctx)); /* In case it's sensitive */
 }
 
 #ifndef ASM_MD5
@@ -157,7 +148,7 @@
 
 /* This is the central step in the MD5 algorithm. */
 #define MD5STEP(f,w,x,y,z,in,s) \
-    (w += f(x,y,z) + in, w = (w<<s | w>>(32-s)) + x)
+  (w += f(x,y,z) + in, w = (w<<s | w>>(32-s)) + x)
 
 /*
  * The core of the MD5 algorithm, this alters an existing MD5 hash to
@@ -165,87 +156,86 @@
  * the data and converts bytes into longwords for this routine.
  */
 void
-MD5Transform(UWORD32 buf[4], UWORD32 const in[16])
-{
-    register UWORD32 a, b, c, d;
+MD5Transform(UWORD32 buf[4], UWORD32 const in[16]) {
+  register UWORD32 a, b, c, d;
 
-    a = buf[0];
-    b = buf[1];
-    c = buf[2];
-    d = buf[3];
+  a = buf[0];
+  b = buf[1];
+  c = buf[2];
+  d = buf[3];
 
-    MD5STEP(F1, a, b, c, d, in[0] + 0xd76aa478, 7);
-    MD5STEP(F1, d, a, b, c, in[1] + 0xe8c7b756, 12);
-    MD5STEP(F1, c, d, a, b, in[2] + 0x242070db, 17);
-    MD5STEP(F1, b, c, d, a, in[3] + 0xc1bdceee, 22);
-    MD5STEP(F1, a, b, c, d, in[4] + 0xf57c0faf, 7);
-    MD5STEP(F1, d, a, b, c, in[5] + 0x4787c62a, 12);
-    MD5STEP(F1, c, d, a, b, in[6] + 0xa8304613, 17);
-    MD5STEP(F1, b, c, d, a, in[7] + 0xfd469501, 22);
-    MD5STEP(F1, a, b, c, d, in[8] + 0x698098d8, 7);
-    MD5STEP(F1, d, a, b, c, in[9] + 0x8b44f7af, 12);
-    MD5STEP(F1, c, d, a, b, in[10] + 0xffff5bb1, 17);
-    MD5STEP(F1, b, c, d, a, in[11] + 0x895cd7be, 22);
-    MD5STEP(F1, a, b, c, d, in[12] + 0x6b901122, 7);
-    MD5STEP(F1, d, a, b, c, in[13] + 0xfd987193, 12);
-    MD5STEP(F1, c, d, a, b, in[14] + 0xa679438e, 17);
-    MD5STEP(F1, b, c, d, a, in[15] + 0x49b40821, 22);
+  MD5STEP(F1, a, b, c, d, in[0] + 0xd76aa478, 7);
+  MD5STEP(F1, d, a, b, c, in[1] + 0xe8c7b756, 12);
+  MD5STEP(F1, c, d, a, b, in[2] + 0x242070db, 17);
+  MD5STEP(F1, b, c, d, a, in[3] + 0xc1bdceee, 22);
+  MD5STEP(F1, a, b, c, d, in[4] + 0xf57c0faf, 7);
+  MD5STEP(F1, d, a, b, c, in[5] + 0x4787c62a, 12);
+  MD5STEP(F1, c, d, a, b, in[6] + 0xa8304613, 17);
+  MD5STEP(F1, b, c, d, a, in[7] + 0xfd469501, 22);
+  MD5STEP(F1, a, b, c, d, in[8] + 0x698098d8, 7);
+  MD5STEP(F1, d, a, b, c, in[9] + 0x8b44f7af, 12);
+  MD5STEP(F1, c, d, a, b, in[10] + 0xffff5bb1, 17);
+  MD5STEP(F1, b, c, d, a, in[11] + 0x895cd7be, 22);
+  MD5STEP(F1, a, b, c, d, in[12] + 0x6b901122, 7);
+  MD5STEP(F1, d, a, b, c, in[13] + 0xfd987193, 12);
+  MD5STEP(F1, c, d, a, b, in[14] + 0xa679438e, 17);
+  MD5STEP(F1, b, c, d, a, in[15] + 0x49b40821, 22);
 
-    MD5STEP(F2, a, b, c, d, in[1] + 0xf61e2562, 5);
-    MD5STEP(F2, d, a, b, c, in[6] + 0xc040b340, 9);
-    MD5STEP(F2, c, d, a, b, in[11] + 0x265e5a51, 14);
-    MD5STEP(F2, b, c, d, a, in[0] + 0xe9b6c7aa, 20);
-    MD5STEP(F2, a, b, c, d, in[5] + 0xd62f105d, 5);
-    MD5STEP(F2, d, a, b, c, in[10] + 0x02441453, 9);
-    MD5STEP(F2, c, d, a, b, in[15] + 0xd8a1e681, 14);
-    MD5STEP(F2, b, c, d, a, in[4] + 0xe7d3fbc8, 20);
-    MD5STEP(F2, a, b, c, d, in[9] + 0x21e1cde6, 5);
-    MD5STEP(F2, d, a, b, c, in[14] + 0xc33707d6, 9);
-    MD5STEP(F2, c, d, a, b, in[3] + 0xf4d50d87, 14);
-    MD5STEP(F2, b, c, d, a, in[8] + 0x455a14ed, 20);
-    MD5STEP(F2, a, b, c, d, in[13] + 0xa9e3e905, 5);
-    MD5STEP(F2, d, a, b, c, in[2] + 0xfcefa3f8, 9);
-    MD5STEP(F2, c, d, a, b, in[7] + 0x676f02d9, 14);
-    MD5STEP(F2, b, c, d, a, in[12] + 0x8d2a4c8a, 20);
+  MD5STEP(F2, a, b, c, d, in[1] + 0xf61e2562, 5);
+  MD5STEP(F2, d, a, b, c, in[6] + 0xc040b340, 9);
+  MD5STEP(F2, c, d, a, b, in[11] + 0x265e5a51, 14);
+  MD5STEP(F2, b, c, d, a, in[0] + 0xe9b6c7aa, 20);
+  MD5STEP(F2, a, b, c, d, in[5] + 0xd62f105d, 5);
+  MD5STEP(F2, d, a, b, c, in[10] + 0x02441453, 9);
+  MD5STEP(F2, c, d, a, b, in[15] + 0xd8a1e681, 14);
+  MD5STEP(F2, b, c, d, a, in[4] + 0xe7d3fbc8, 20);
+  MD5STEP(F2, a, b, c, d, in[9] + 0x21e1cde6, 5);
+  MD5STEP(F2, d, a, b, c, in[14] + 0xc33707d6, 9);
+  MD5STEP(F2, c, d, a, b, in[3] + 0xf4d50d87, 14);
+  MD5STEP(F2, b, c, d, a, in[8] + 0x455a14ed, 20);
+  MD5STEP(F2, a, b, c, d, in[13] + 0xa9e3e905, 5);
+  MD5STEP(F2, d, a, b, c, in[2] + 0xfcefa3f8, 9);
+  MD5STEP(F2, c, d, a, b, in[7] + 0x676f02d9, 14);
+  MD5STEP(F2, b, c, d, a, in[12] + 0x8d2a4c8a, 20);
 
-    MD5STEP(F3, a, b, c, d, in[5] + 0xfffa3942, 4);
-    MD5STEP(F3, d, a, b, c, in[8] + 0x8771f681, 11);
-    MD5STEP(F3, c, d, a, b, in[11] + 0x6d9d6122, 16);
-    MD5STEP(F3, b, c, d, a, in[14] + 0xfde5380c, 23);
-    MD5STEP(F3, a, b, c, d, in[1] + 0xa4beea44, 4);
-    MD5STEP(F3, d, a, b, c, in[4] + 0x4bdecfa9, 11);
-    MD5STEP(F3, c, d, a, b, in[7] + 0xf6bb4b60, 16);
-    MD5STEP(F3, b, c, d, a, in[10] + 0xbebfbc70, 23);
-    MD5STEP(F3, a, b, c, d, in[13] + 0x289b7ec6, 4);
-    MD5STEP(F3, d, a, b, c, in[0] + 0xeaa127fa, 11);
-    MD5STEP(F3, c, d, a, b, in[3] + 0xd4ef3085, 16);
-    MD5STEP(F3, b, c, d, a, in[6] + 0x04881d05, 23);
-    MD5STEP(F3, a, b, c, d, in[9] + 0xd9d4d039, 4);
-    MD5STEP(F3, d, a, b, c, in[12] + 0xe6db99e5, 11);
-    MD5STEP(F3, c, d, a, b, in[15] + 0x1fa27cf8, 16);
-    MD5STEP(F3, b, c, d, a, in[2] + 0xc4ac5665, 23);
+  MD5STEP(F3, a, b, c, d, in[5] + 0xfffa3942, 4);
+  MD5STEP(F3, d, a, b, c, in[8] + 0x8771f681, 11);
+  MD5STEP(F3, c, d, a, b, in[11] + 0x6d9d6122, 16);
+  MD5STEP(F3, b, c, d, a, in[14] + 0xfde5380c, 23);
+  MD5STEP(F3, a, b, c, d, in[1] + 0xa4beea44, 4);
+  MD5STEP(F3, d, a, b, c, in[4] + 0x4bdecfa9, 11);
+  MD5STEP(F3, c, d, a, b, in[7] + 0xf6bb4b60, 16);
+  MD5STEP(F3, b, c, d, a, in[10] + 0xbebfbc70, 23);
+  MD5STEP(F3, a, b, c, d, in[13] + 0x289b7ec6, 4);
+  MD5STEP(F3, d, a, b, c, in[0] + 0xeaa127fa, 11);
+  MD5STEP(F3, c, d, a, b, in[3] + 0xd4ef3085, 16);
+  MD5STEP(F3, b, c, d, a, in[6] + 0x04881d05, 23);
+  MD5STEP(F3, a, b, c, d, in[9] + 0xd9d4d039, 4);
+  MD5STEP(F3, d, a, b, c, in[12] + 0xe6db99e5, 11);
+  MD5STEP(F3, c, d, a, b, in[15] + 0x1fa27cf8, 16);
+  MD5STEP(F3, b, c, d, a, in[2] + 0xc4ac5665, 23);
 
-    MD5STEP(F4, a, b, c, d, in[0] + 0xf4292244, 6);
-    MD5STEP(F4, d, a, b, c, in[7] + 0x432aff97, 10);
-    MD5STEP(F4, c, d, a, b, in[14] + 0xab9423a7, 15);
-    MD5STEP(F4, b, c, d, a, in[5] + 0xfc93a039, 21);
-    MD5STEP(F4, a, b, c, d, in[12] + 0x655b59c3, 6);
-    MD5STEP(F4, d, a, b, c, in[3] + 0x8f0ccc92, 10);
-    MD5STEP(F4, c, d, a, b, in[10] + 0xffeff47d, 15);
-    MD5STEP(F4, b, c, d, a, in[1] + 0x85845dd1, 21);
-    MD5STEP(F4, a, b, c, d, in[8] + 0x6fa87e4f, 6);
-    MD5STEP(F4, d, a, b, c, in[15] + 0xfe2ce6e0, 10);
-    MD5STEP(F4, c, d, a, b, in[6] + 0xa3014314, 15);
-    MD5STEP(F4, b, c, d, a, in[13] + 0x4e0811a1, 21);
-    MD5STEP(F4, a, b, c, d, in[4] + 0xf7537e82, 6);
-    MD5STEP(F4, d, a, b, c, in[11] + 0xbd3af235, 10);
-    MD5STEP(F4, c, d, a, b, in[2] + 0x2ad7d2bb, 15);
-    MD5STEP(F4, b, c, d, a, in[9] + 0xeb86d391, 21);
+  MD5STEP(F4, a, b, c, d, in[0] + 0xf4292244, 6);
+  MD5STEP(F4, d, a, b, c, in[7] + 0x432aff97, 10);
+  MD5STEP(F4, c, d, a, b, in[14] + 0xab9423a7, 15);
+  MD5STEP(F4, b, c, d, a, in[5] + 0xfc93a039, 21);
+  MD5STEP(F4, a, b, c, d, in[12] + 0x655b59c3, 6);
+  MD5STEP(F4, d, a, b, c, in[3] + 0x8f0ccc92, 10);
+  MD5STEP(F4, c, d, a, b, in[10] + 0xffeff47d, 15);
+  MD5STEP(F4, b, c, d, a, in[1] + 0x85845dd1, 21);
+  MD5STEP(F4, a, b, c, d, in[8] + 0x6fa87e4f, 6);
+  MD5STEP(F4, d, a, b, c, in[15] + 0xfe2ce6e0, 10);
+  MD5STEP(F4, c, d, a, b, in[6] + 0xa3014314, 15);
+  MD5STEP(F4, b, c, d, a, in[13] + 0x4e0811a1, 21);
+  MD5STEP(F4, a, b, c, d, in[4] + 0xf7537e82, 6);
+  MD5STEP(F4, d, a, b, c, in[11] + 0xbd3af235, 10);
+  MD5STEP(F4, c, d, a, b, in[2] + 0x2ad7d2bb, 15);
+  MD5STEP(F4, b, c, d, a, in[9] + 0xeb86d391, 21);
 
-    buf[0] += a;
-    buf[1] += b;
-    buf[2] += c;
-    buf[3] += d;
+  buf[0] += a;
+  buf[1] += b;
+  buf[2] += c;
+  buf[3] += d;
 }
 
 #endif
diff --git a/libvpx/md5_utils.h b/libvpx/md5_utils.h
index 5ca1b5f..81792c4 100644
--- a/libvpx/md5_utils.h
+++ b/libvpx/md5_utils.h
@@ -27,11 +27,10 @@
 #define UWORD32 unsigned int
 
 typedef struct MD5Context MD5Context;
-struct MD5Context
-{
-    UWORD32 buf[4];
-    UWORD32 bytes[2];
-    UWORD32 in[16];
+struct MD5Context {
+  UWORD32 buf[4];
+  UWORD32 bytes[2];
+  UWORD32 in[16];
 };
 
 void MD5Init(struct MD5Context *context);
diff --git a/libvpx/nestegg/halloc/src/macros.h b/libvpx/nestegg/halloc/src/macros.h
index c36b516..1f84bc2 100644
--- a/libvpx/nestegg/halloc/src/macros.h
+++ b/libvpx/nestegg/halloc/src/macros.h
@@ -20,7 +20,7 @@
 /*
  	restore pointer to the structure by a pointer to its field
  */
-#define structof(p,t,f) ((t*)(- offsetof(t,f) + (char*)(p)))
+#define structof(p,t,f) ((t*)(- (ptrdiff_t) offsetof(t,f) + (char*)(p)))
 
 /*
  *	redefine for the target compiler
diff --git a/libvpx/nestegg/include/nestegg/nestegg.h b/libvpx/nestegg/include/nestegg/nestegg.h
index 7447d14..6510694 100644
--- a/libvpx/nestegg/include/nestegg/nestegg.h
+++ b/libvpx/nestegg/include/nestegg/nestegg.h
@@ -67,6 +67,7 @@
 
 #define NESTEGG_CODEC_VP8    0 /**< Track uses Google On2 VP8 codec. */
 #define NESTEGG_CODEC_VORBIS 1 /**< Track uses Xiph Vorbis codec. */
+#define NESTEGG_CODEC_VP9    2 /**< Track uses Google On2 VP9 codec. */
 
 #define NESTEGG_SEEK_SET 0 /**< Seek offset relative to beginning of stream. */
 #define NESTEGG_SEEK_CUR 1 /**< Seek offset relative to current position in stream. */
diff --git a/libvpx/nestegg/src/nestegg.c b/libvpx/nestegg/src/nestegg.c
index cc87788..ae87e8f 100644
--- a/libvpx/nestegg/src/nestegg.c
+++ b/libvpx/nestegg/src/nestegg.c
@@ -127,6 +127,7 @@
 
 /* Track IDs */
 #define TRACK_ID_VP8            "V_VP8"
+#define TRACK_ID_VP9            "V_VP9"
 #define TRACK_ID_VORBIS         "A_VORBIS"
 
 enum vint_mask {
@@ -1669,6 +1670,9 @@
   if (strcmp(codec_id, TRACK_ID_VP8) == 0)
     return NESTEGG_CODEC_VP8;
 
+  if (strcmp(codec_id, TRACK_ID_VP9) == 0)
+    return NESTEGG_CODEC_VP9;
+
   if (strcmp(codec_id, TRACK_ID_VORBIS) == 0)
     return NESTEGG_CODEC_VORBIS;
 
diff --git a/libvpx/solution.mk b/libvpx/solution.mk
index 948305f..2c8d29a 100644
--- a/libvpx/solution.mk
+++ b/libvpx/solution.mk
@@ -9,14 +9,14 @@
 ##
 
 # libvpx reverse dependencies (targets that depend on libvpx)
-VPX_NONDEPS=$(addsuffix .vcproj,vpx gtest obj_int_extract)
+VPX_NONDEPS=$(addsuffix .$(VCPROJ_SFX),vpx gtest obj_int_extract)
 VPX_RDEPS=$(foreach vcp,\
-              $(filter-out $(VPX_NONDEPS),$^), --dep=$(vcp:.vcproj=):vpx)
+              $(filter-out $(VPX_NONDEPS),$^), --dep=$(vcp:.$(VCPROJ_SFX)=):vpx)
 
-vpx.sln: $(wildcard *.vcproj)
+vpx.sln: $(wildcard *.$(VCPROJ_SFX))
 	@echo "    [CREATE] $@"
 	$(SRC_PATH_BARE)/build/make/gen_msvs_sln.sh \
-            $(if $(filter vpx.vcproj,$^),$(VPX_RDEPS)) \
+            $(if $(filter vpx.$(VCPROJ_SFX),$^),$(VPX_RDEPS)) \
             --dep=vpx:obj_int_extract \
             --dep=test_libvpx:gtest \
             --ver=$(CONFIG_VS_VERSION)\
diff --git a/libvpx/test/acm_random.h b/libvpx/test/acm_random.h
index 514894e..cd33d12 100644
--- a/libvpx/test/acm_random.h
+++ b/libvpx/test/acm_random.h
@@ -11,7 +11,7 @@
 #ifndef LIBVPX_TEST_ACM_RANDOM_H_
 #define LIBVPX_TEST_ACM_RANDOM_H_
 
-#include <stdlib.h>
+#include "third_party/googletest/src/include/gtest/gtest.h"
 
 #include "vpx/vpx_integer.h"
 
@@ -19,24 +19,30 @@
 
 class ACMRandom {
  public:
-  ACMRandom() {
-    Reset(DeterministicSeed());
-  }
+  ACMRandom() : random_(DeterministicSeed()) {}
 
-  explicit ACMRandom(int seed) {
-    Reset(seed);
-  }
+  explicit ACMRandom(int seed) : random_(seed) {}
 
   void Reset(int seed) {
-    srand(seed);
+    random_.Reseed(seed);
   }
 
   uint8_t Rand8(void) {
-    return (rand() >> 8) & 0xff;
+    const uint32_t value =
+        random_.Generate(testing::internal::Random::kMaxRange);
+    // There's a bit more entropy in the upper bits of this implementation.
+    return (value >> 24) & 0xff;
+  }
+
+  uint8_t Rand8Extremes(void) {
+    // Returns a random value near 0 or near 255, to better exercise
+    // saturation behavior.
+    const uint8_t r = Rand8();
+    return r < 128 ? r << 4 : r >> 4;
   }
 
   int PseudoUniform(int range) {
-    return (rand() >> 8) % range;
+    return random_.Generate(range);
   }
 
   int operator()(int n) {
@@ -46,6 +52,9 @@
   static int DeterministicSeed(void) {
     return 0xbaba;
   }
+
+ private:
+  testing::internal::Random random_;
 };
 
 }  // namespace libvpx_test
diff --git a/libvpx/test/altref_test.cc b/libvpx/test/altref_test.cc
index ca05577..14af265 100644
--- a/libvpx/test/altref_test.cc
+++ b/libvpx/test/altref_test.cc
@@ -8,19 +8,20 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
 #include "test/i420_video_source.h"
-
+#include "test/util.h"
 namespace {
 
 // lookahead range: [kLookAheadMin, kLookAheadMax).
 const int kLookAheadMin = 5;
 const int kLookAheadMax = 26;
 
-class AltRefTest : public libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<int> {
+class AltRefTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<int> {
  protected:
-  AltRefTest() : altref_count_(0) {}
+  AltRefTest() : EncoderTest(GET_PARAM(0)), altref_count_(0) {}
   virtual ~AltRefTest() {}
 
   virtual void SetUp() {
@@ -58,7 +59,7 @@
   const vpx_rational timebase = { 33333333, 1000000000 };
   cfg_.g_timebase = timebase;
   cfg_.rc_target_bitrate = 1000;
-  cfg_.g_lag_in_frames = GetParam();
+  cfg_.g_lag_in_frames = GET_PARAM(1);
 
   libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
                                      timebase.den, timebase.num, 0, 30);
@@ -66,6 +67,7 @@
   EXPECT_GE(altref_count(), 1);
 }
 
-INSTANTIATE_TEST_CASE_P(NonZeroLag, AltRefTest,
-                        ::testing::Range(kLookAheadMin, kLookAheadMax));
+
+VP8_INSTANTIATE_TEST_CASE(AltRefTest,
+                          ::testing::Range(kLookAheadMin, kLookAheadMax));
 }  // namespace
diff --git a/libvpx/test/borders_test.cc b/libvpx/test/borders_test.cc
new file mode 100644
index 0000000..49505ee
--- /dev/null
+++ b/libvpx/test/borders_test.cc
@@ -0,0 +1,86 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include <climits>
+#include <vector>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class BordersTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
+ protected:
+  BordersTest() : EncoderTest(GET_PARAM(0)) {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+  }
+
+  virtual bool Continue() const {
+    return !HasFatalFailure() && !abort_;
+  }
+
+  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource *video,
+                                  ::libvpx_test::Encoder *encoder) {
+    if ( video->frame() == 1) {
+      encoder->Control(VP8E_SET_CPUUSED, 0);
+      encoder->Control(VP8E_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(VP8E_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(VP8E_SET_ARNR_STRENGTH, 5);
+      encoder->Control(VP8E_SET_ARNR_TYPE, 3);
+    }
+  }
+
+  virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
+    if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
+    }
+  }
+};
+
+TEST_P(BordersTest, TestEncodeHighBitrate) {
+  // Validate that this non multiple of 64 wide clip encodes and decodes
+  // without a mismatch when passing in a very low max q.  This pushes
+  // the encoder to producing lots of big partitions which will likely
+  // extend into the border and test the border condition.
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_minsection_pct = 2000;
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.rc_max_quantizer = 10;
+
+  ::libvpx_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       40);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+TEST_P(BordersTest, TestLowBitrate) {
+  // Validate that this clip encodes and decodes without a mismatch
+  // when passing in a very high min q.  This pushes the encoder to producing
+  // lots of small partitions which might will test the other condition.
+
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_minsection_pct = 2000;
+  cfg_.rc_target_bitrate = 200;
+  cfg_.rc_min_quantizer = 40;
+
+  ::libvpx_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       40);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+VP9_INSTANTIATE_TEST_CASE(BordersTest, ::testing::Values(
+    ::libvpx_test::kTwoPassGood));
+}  // namespace
diff --git a/libvpx/test/clear_system_state.h b/libvpx/test/clear_system_state.h
new file mode 100644
index 0000000..e240981
--- /dev/null
+++ b/libvpx/test/clear_system_state.h
@@ -0,0 +1,31 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef TEST_CLEAR_SYSTEM_STATE_H_
+#define TEST_CLEAR_SYSTEM_STATE_H_
+
+#include "vpx_config.h"
+extern "C" {
+#if ARCH_X86 || ARCH_X86_64
+# include "vpx_ports/x86.h"
+#endif
+}
+
+namespace libvpx_test {
+
+// Reset system to a known state. This function should be used for all non-API
+// test cases.
+inline void ClearSystemState() {
+#if ARCH_X86 || ARCH_X86_64
+  vpx_reset_mmx_state();
+#endif
+}
+
+}  // namespace libvpx_test
+#endif  // TEST_CLEAR_SYSTEM_STATE_H_
diff --git a/libvpx/test/codec_factory.h b/libvpx/test/codec_factory.h
new file mode 100644
index 0000000..fdae572
--- /dev/null
+++ b/libvpx/test/codec_factory.h
@@ -0,0 +1,232 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef TEST_CODEC_FACTORY_H_
+#define TEST_CODEC_FACTORY_H_
+
+extern "C" {
+#include "./vpx_config.h"
+#include "vpx/vpx_decoder.h"
+#include "vpx/vpx_encoder.h"
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
+#include "vpx/vp8cx.h"
+#endif
+#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
+#include "vpx/vp8dx.h"
+#endif
+}
+
+#include "test/decode_test_driver.h"
+#include "test/encode_test_driver.h"
+namespace libvpx_test {
+
+class CodecFactory {
+ public:
+  CodecFactory() {}
+
+  virtual ~CodecFactory() {}
+
+  virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
+                                 unsigned long deadline) const = 0;
+
+  virtual Encoder* CreateEncoder(vpx_codec_enc_cfg_t cfg,
+                                 unsigned long deadline,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const = 0;
+
+  virtual vpx_codec_err_t DefaultEncoderConfig(vpx_codec_enc_cfg_t *cfg,
+                                               int usage) const = 0;
+};
+
+/* Provide CodecTestWith<n>Params classes for a variable number of parameters
+ * to avoid having to include a pointer to the CodecFactory in every test
+ * definition.
+ */
+template<class T1>
+class CodecTestWithParam : public ::testing::TestWithParam<
+    std::tr1::tuple< const libvpx_test::CodecFactory*, T1 > > {
+};
+
+template<class T1, class T2>
+class CodecTestWith2Params : public ::testing::TestWithParam<
+    std::tr1::tuple< const libvpx_test::CodecFactory*, T1, T2 > > {
+};
+
+template<class T1, class T2, class T3>
+class CodecTestWith3Params : public ::testing::TestWithParam<
+    std::tr1::tuple< const libvpx_test::CodecFactory*, T1, T2, T3 > > {
+};
+
+/*
+ * VP8 Codec Definitions
+ */
+#if CONFIG_VP8
+class VP8Decoder : public Decoder {
+ public:
+  VP8Decoder(vpx_codec_dec_cfg_t cfg, unsigned long deadline)
+      : Decoder(cfg, deadline) {}
+
+ protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP8_DECODER
+    return &vpx_codec_vp8_dx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP8Encoder : public Encoder {
+ public:
+  VP8Encoder(vpx_codec_enc_cfg_t cfg, unsigned long deadline,
+             const unsigned long init_flags, TwopassStatsStore *stats)
+      : Encoder(cfg, deadline, init_flags, stats) {}
+
+ protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP8_ENCODER
+    return &vpx_codec_vp8_cx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP8CodecFactory : public CodecFactory {
+ public:
+  VP8CodecFactory() : CodecFactory() {}
+
+  virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
+                                 unsigned long deadline) const {
+#if CONFIG_VP8_DECODER
+    return new VP8Decoder(cfg, deadline);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual Encoder* CreateEncoder(vpx_codec_enc_cfg_t cfg,
+                                 unsigned long deadline,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const {
+#if CONFIG_VP8_ENCODER
+    return new VP8Encoder(cfg, deadline, init_flags, stats);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual vpx_codec_err_t DefaultEncoderConfig(vpx_codec_enc_cfg_t *cfg,
+                                               int usage) const {
+#if CONFIG_VP8_ENCODER
+    return vpx_codec_enc_config_default(&vpx_codec_vp8_cx_algo, cfg, usage);
+#else
+    return VPX_CODEC_INCAPABLE;
+#endif
+  }
+};
+
+const libvpx_test::VP8CodecFactory kVP8;
+
+#define VP8_INSTANTIATE_TEST_CASE(test, params)\
+  INSTANTIATE_TEST_CASE_P(VP8, test, \
+      ::testing::Combine( \
+          ::testing::Values(static_cast<const libvpx_test::CodecFactory*>( \
+              &libvpx_test::kVP8)), \
+          params))
+#else
+#define VP8_INSTANTIATE_TEST_CASE(test, params)
+#endif  // CONFIG_VP8
+
+
+/*
+ * VP9 Codec Definitions
+ */
+#if CONFIG_VP9
+class VP9Decoder : public Decoder {
+ public:
+  VP9Decoder(vpx_codec_dec_cfg_t cfg, unsigned long deadline)
+      : Decoder(cfg, deadline) {}
+
+ protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP9_DECODER
+    return &vpx_codec_vp9_dx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP9Encoder : public Encoder {
+ public:
+  VP9Encoder(vpx_codec_enc_cfg_t cfg, unsigned long deadline,
+             const unsigned long init_flags, TwopassStatsStore *stats)
+      : Encoder(cfg, deadline, init_flags, stats) {}
+
+ protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP9_ENCODER
+    return &vpx_codec_vp9_cx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP9CodecFactory : public CodecFactory {
+ public:
+  VP9CodecFactory() : CodecFactory() {}
+
+  virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
+                                 unsigned long deadline) const {
+#if CONFIG_VP9_DECODER
+    return new VP9Decoder(cfg, deadline);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual Encoder* CreateEncoder(vpx_codec_enc_cfg_t cfg,
+                                 unsigned long deadline,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const {
+#if CONFIG_VP9_ENCODER
+    return new VP9Encoder(cfg, deadline, init_flags, stats);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual vpx_codec_err_t DefaultEncoderConfig(vpx_codec_enc_cfg_t *cfg,
+                                               int usage) const {
+#if CONFIG_VP9_ENCODER
+    return vpx_codec_enc_config_default(&vpx_codec_vp9_cx_algo, cfg, usage);
+#else
+    return VPX_CODEC_INCAPABLE;
+#endif
+  }
+};
+
+const libvpx_test::VP9CodecFactory kVP9;
+
+#define VP9_INSTANTIATE_TEST_CASE(test, params)\
+  INSTANTIATE_TEST_CASE_P(VP9, test, \
+      ::testing::Combine( \
+          ::testing::Values(static_cast<const libvpx_test::CodecFactory*>( \
+               &libvpx_test::kVP9)), \
+          params))
+#else
+#define VP9_INSTANTIATE_TEST_CASE(test, params)
+#endif  // CONFIG_VP9
+
+
+}  // namespace libvpx_test
+
+#endif  // TEST_CODEC_FACTORY_H_
diff --git a/libvpx/test/config_test.cc b/libvpx/test/config_test.cc
index c4da46e..9008728 100644
--- a/libvpx/test/config_test.cc
+++ b/libvpx/test/config_test.cc
@@ -8,20 +8,22 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
+#include "test/util.h"
 #include "test/video_source.h"
 
 namespace {
 
 class ConfigTest : public ::libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<enum libvpx_test::TestMode> {
- public:
-  ConfigTest() : frame_count_in_(0), frame_count_out_(0), frame_count_max_(0) {}
-
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
  protected:
+  ConfigTest() : EncoderTest(GET_PARAM(0)),
+                 frame_count_in_(0), frame_count_out_(0), frame_count_max_(0) {}
+
   virtual void SetUp() {
     InitializeConfig();
-    SetMode(GetParam());
+    SetMode(GET_PARAM(1));
   }
 
   virtual void BeginPassHook(unsigned int /*pass*/) {
@@ -57,5 +59,5 @@
   EXPECT_EQ(frame_count_in_, frame_count_out_);
 }
 
-INSTANTIATE_TEST_CASE_P(OnePassModes, ConfigTest, ONE_PASS_TEST_MODES);
+VP8_INSTANTIATE_TEST_CASE(ConfigTest, ONE_PASS_TEST_MODES);
 }  // namespace
diff --git a/libvpx/test/convolve_test.cc b/libvpx/test/convolve_test.cc
new file mode 100644
index 0000000..fd2bd36
--- /dev/null
+++ b/libvpx/test/convolve_test.cc
@@ -0,0 +1,549 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "test/acm_random.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_filter.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/mem.h"
+}
+
+namespace {
+typedef void (*convolve_fn_t)(const uint8_t *src, int src_stride,
+                              uint8_t *dst, int dst_stride,
+                              const int16_t *filter_x, int filter_x_stride,
+                              const int16_t *filter_y, int filter_y_stride,
+                              int w, int h);
+
+struct ConvolveFunctions {
+  ConvolveFunctions(convolve_fn_t h8, convolve_fn_t h8_avg,
+                    convolve_fn_t v8, convolve_fn_t v8_avg,
+                    convolve_fn_t hv8, convolve_fn_t hv8_avg)
+      : h8_(h8), v8_(v8), hv8_(hv8), h8_avg_(h8_avg), v8_avg_(v8_avg),
+        hv8_avg_(hv8_avg) {}
+
+  convolve_fn_t h8_;
+  convolve_fn_t v8_;
+  convolve_fn_t hv8_;
+  convolve_fn_t h8_avg_;
+  convolve_fn_t v8_avg_;
+  convolve_fn_t hv8_avg_;
+};
+
+// Reference 8-tap subpixel filter, slightly modified to fit into this test.
+#define VP9_FILTER_WEIGHT 128
+#define VP9_FILTER_SHIFT 7
+uint8_t clip_pixel(int x) {
+  return x < 0 ? 0 :
+         x > 255 ? 255 :
+         x;
+}
+
+void filter_block2d_8_c(const uint8_t *src_ptr,
+                        const unsigned int src_stride,
+                        const int16_t *HFilter,
+                        const int16_t *VFilter,
+                        uint8_t *dst_ptr,
+                        unsigned int dst_stride,
+                        unsigned int output_width,
+                        unsigned int output_height) {
+  // Between passes, we use an intermediate buffer whose height is extended to
+  // have enough horizontally filtered values as input for the vertical pass.
+  // This buffer is allocated to be big enough for the largest block type we
+  // support.
+  const int kInterp_Extend = 4;
+  const unsigned int intermediate_height =
+      (kInterp_Extend - 1) + output_height + kInterp_Extend;
+
+  /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
+   * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
+   *                                 + kInterp_Extend
+   *                               = 3 + 16 + 4
+   *                               = 23
+   * and filter_max_width = 16
+   */
+  uint8_t intermediate_buffer[71 * 64];
+  const int intermediate_next_stride = 1 - intermediate_height * output_width;
+
+  // Horizontal pass (src -> transposed intermediate).
+  {
+    uint8_t *output_ptr = intermediate_buffer;
+    const int src_next_row_stride = src_stride - output_width;
+    unsigned int i, j;
+    src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
+    for (i = 0; i < intermediate_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp = (src_ptr[0] * HFilter[0]) +
+                         (src_ptr[1] * HFilter[1]) +
+                         (src_ptr[2] * HFilter[2]) +
+                         (src_ptr[3] * HFilter[3]) +
+                         (src_ptr[4] * HFilter[4]) +
+                         (src_ptr[5] * HFilter[5]) +
+                         (src_ptr[6] * HFilter[6]) +
+                         (src_ptr[7] * HFilter[7]) +
+                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *output_ptr = clip_pixel(temp >> VP9_FILTER_SHIFT);
+        ++src_ptr;
+        output_ptr += intermediate_height;
+      }
+      src_ptr += src_next_row_stride;
+      output_ptr += intermediate_next_stride;
+    }
+  }
+
+  // Vertical pass (transposed intermediate -> dst).
+  {
+    uint8_t *src_ptr = intermediate_buffer;
+    const int dst_next_row_stride = dst_stride - output_width;
+    unsigned int i, j;
+    for (i = 0; i < output_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp = (src_ptr[0] * VFilter[0]) +
+                         (src_ptr[1] * VFilter[1]) +
+                         (src_ptr[2] * VFilter[2]) +
+                         (src_ptr[3] * VFilter[3]) +
+                         (src_ptr[4] * VFilter[4]) +
+                         (src_ptr[5] * VFilter[5]) +
+                         (src_ptr[6] * VFilter[6]) +
+                         (src_ptr[7] * VFilter[7]) +
+                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *dst_ptr++ = clip_pixel(temp >> VP9_FILTER_SHIFT);
+        src_ptr += intermediate_height;
+      }
+      src_ptr += intermediate_next_stride;
+      dst_ptr += dst_next_row_stride;
+    }
+  }
+}
+
+void block2d_average_c(uint8_t *src,
+                       unsigned int src_stride,
+                       uint8_t *output_ptr,
+                       unsigned int output_stride,
+                       unsigned int output_width,
+                       unsigned int output_height) {
+  unsigned int i, j;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
+    }
+    output_ptr += output_stride;
+  }
+}
+
+void filter_average_block2d_8_c(const uint8_t *src_ptr,
+                                const unsigned int src_stride,
+                                const int16_t *HFilter,
+                                const int16_t *VFilter,
+                                uint8_t *dst_ptr,
+                                unsigned int dst_stride,
+                                unsigned int output_width,
+                                unsigned int output_height) {
+  uint8_t tmp[64 * 64];
+
+  assert(output_width <= 64);
+  assert(output_height <= 64);
+  filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp, 64,
+                     output_width, output_height);
+  block2d_average_c(tmp, 64, dst_ptr, dst_stride,
+                    output_width, output_height);
+}
+
+class ConvolveTest : public PARAMS(int, int, const ConvolveFunctions*) {
+ public:
+  static void SetUpTestCase() {
+    // Force input_ to be unaligned, output to be 16 byte aligned.
+    input_ = reinterpret_cast<uint8_t*>(
+        vpx_memalign(kDataAlignment, kInputBufferSize + 1)) + 1;
+    output_ = reinterpret_cast<uint8_t*>(
+        vpx_memalign(kDataAlignment, kOutputBufferSize));
+  }
+
+  static void TearDownTestCase() {
+    vpx_free(input_ - 1);
+    input_ = NULL;
+    vpx_free(output_);
+    output_ = NULL;
+  }
+
+ protected:
+  static const int kDataAlignment = 16;
+  static const int kOuterBlockSize = 128;
+  static const int kInputStride = kOuterBlockSize;
+  static const int kOutputStride = kOuterBlockSize;
+  static const int kMaxDimension = 64;
+  static const int kInputBufferSize = kOuterBlockSize * kOuterBlockSize;
+  static const int kOutputBufferSize = kOuterBlockSize * kOuterBlockSize;
+
+  int Width() const { return GET_PARAM(0); }
+  int Height() const { return GET_PARAM(1); }
+  int BorderLeft() const {
+    const int center = (kOuterBlockSize - Width()) / 2;
+    return (center + (kDataAlignment - 1)) & ~(kDataAlignment - 1);
+  }
+  int BorderTop() const { return (kOuterBlockSize - Height()) / 2; }
+
+  bool IsIndexInBorder(int i) {
+    return (i < BorderTop() * kOuterBlockSize ||
+            i >= (BorderTop() + Height()) * kOuterBlockSize ||
+            i % kOuterBlockSize < BorderLeft() ||
+            i % kOuterBlockSize >= (BorderLeft() + Width()));
+  }
+
+  virtual void SetUp() {
+    UUT_ = GET_PARAM(2);
+    /* Set up guard blocks for an inner block cetered in the outer block */
+    for (int i = 0; i < kOutputBufferSize; ++i) {
+      if (IsIndexInBorder(i))
+        output_[i] = 255;
+      else
+        output_[i] = 0;
+    }
+
+    ::libvpx_test::ACMRandom prng;
+    for (int i = 0; i < kInputBufferSize; ++i)
+      input_[i] = prng.Rand8Extremes();
+  }
+
+  void CheckGuardBlocks() {
+    for (int i = 0; i < kOutputBufferSize; ++i) {
+      if (IsIndexInBorder(i))
+        EXPECT_EQ(255, output_[i]);
+    }
+  }
+
+  uint8_t* input() const {
+    return input_ + BorderTop() * kOuterBlockSize + BorderLeft();
+  }
+
+  uint8_t* output() const {
+    return output_ + BorderTop() * kOuterBlockSize + BorderLeft();
+  }
+
+  const ConvolveFunctions* UUT_;
+  static uint8_t* input_;
+  static uint8_t* output_;
+};
+uint8_t* ConvolveTest::input_ = NULL;
+uint8_t* ConvolveTest::output_ = NULL;
+
+TEST_P(ConvolveTest, GuardBlocks) {
+  CheckGuardBlocks();
+}
+
+TEST_P(ConvolveTest, CopyHoriz) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
+
+  REGISTER_STATE_CHECK(
+      UUT_->h8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
+                Width(), Height()));
+
+  CheckGuardBlocks();
+
+  for (int y = 0; y < Height(); ++y)
+    for (int x = 0; x < Width(); ++x)
+      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+          << "(" << x << "," << y << ")";
+}
+
+TEST_P(ConvolveTest, CopyVert) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
+
+  REGISTER_STATE_CHECK(
+      UUT_->v8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
+                Width(), Height()));
+
+  CheckGuardBlocks();
+
+  for (int y = 0; y < Height(); ++y)
+    for (int x = 0; x < Width(); ++x)
+      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+          << "(" << x << "," << y << ")";
+}
+
+TEST_P(ConvolveTest, Copy2D) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  DECLARE_ALIGNED(256, const int16_t, filter8[8]) = {0, 0, 0, 128, 0, 0, 0, 0};
+
+  REGISTER_STATE_CHECK(
+      UUT_->hv8_(in, kInputStride, out, kOutputStride, filter8, 16, filter8, 16,
+                 Width(), Height()));
+
+  CheckGuardBlocks();
+
+  for (int y = 0; y < Height(); ++y)
+    for (int x = 0; x < Width(); ++x)
+      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+          << "(" << x << "," << y << ")";
+}
+
+const int16_t (*kTestFilterList[])[8] = {
+  vp9_bilinear_filters,
+  vp9_sub_pel_filters_8,
+  vp9_sub_pel_filters_8s,
+  vp9_sub_pel_filters_8lp
+};
+const int kNumFilterBanks = sizeof(kTestFilterList) /
+                            sizeof(kTestFilterList[0]);
+const int kNumFilters = 16;
+
+TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
+  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+    const int16_t (*filters)[8] = kTestFilterList[filter_bank];
+    for (int i = 0; i < kNumFilters; i++) {
+      const int p0 = filters[i][0] + filters[i][1];
+      const int p1 = filters[i][2] + filters[i][3];
+      const int p2 = filters[i][4] + filters[i][5];
+      const int p3 = filters[i][6] + filters[i][7];
+      EXPECT_LE(p0, 128);
+      EXPECT_LE(p1, 128);
+      EXPECT_LE(p2, 128);
+      EXPECT_LE(p3, 128);
+      EXPECT_LE(p0 + p3, 128);
+      EXPECT_LE(p0 + p3 + p1, 128);
+      EXPECT_LE(p0 + p3 + p1 + p2, 128);
+      EXPECT_EQ(p0 + p1 + p2 + p3, 128);
+    }
+  }
+}
+
+const int16_t kInvalidFilter[8] = { 0 };
+
+TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  uint8_t ref[kOutputStride * kMaxDimension];
+
+
+  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+    const int16_t (*filters)[8] = kTestFilterList[filter_bank];
+
+    for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+      for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+        filter_block2d_8_c(in, kInputStride,
+                           filters[filter_x], filters[filter_y],
+                           ref, kOutputStride,
+                           Width(), Height());
+
+        if (filters == vp9_sub_pel_filters_8lp || (filter_x && filter_y))
+          REGISTER_STATE_CHECK(
+              UUT_->hv8_(in, kInputStride, out, kOutputStride,
+                         filters[filter_x], 16, filters[filter_y], 16,
+                         Width(), Height()));
+        else if (filter_y)
+          REGISTER_STATE_CHECK(
+              UUT_->v8_(in, kInputStride, out, kOutputStride,
+                        kInvalidFilter, 16, filters[filter_y], 16,
+                        Width(), Height()));
+        else
+          REGISTER_STATE_CHECK(
+              UUT_->h8_(in, kInputStride, out, kOutputStride,
+                        filters[filter_x], 16, kInvalidFilter, 16,
+                        Width(), Height()));
+
+        CheckGuardBlocks();
+
+        for (int y = 0; y < Height(); ++y)
+          for (int x = 0; x < Width(); ++x)
+            ASSERT_EQ(ref[y * kOutputStride + x], out[y * kOutputStride + x])
+                << "mismatch at (" << x << "," << y << "), "
+                << "filters (" << filter_bank << ","
+                << filter_x << "," << filter_y << ")";
+      }
+    }
+  }
+}
+
+TEST_P(ConvolveTest, MatchesReferenceAveragingSubpixelFilter) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  uint8_t ref[kOutputStride * kMaxDimension];
+
+  // Populate ref and out with some random data
+  ::libvpx_test::ACMRandom prng;
+  for (int y = 0; y < Height(); ++y) {
+    for (int x = 0; x < Width(); ++x) {
+      const uint8_t r = prng.Rand8Extremes();
+
+      out[y * kOutputStride + x] = r;
+      ref[y * kOutputStride + x] = r;
+    }
+  }
+
+  const int kNumFilterBanks = sizeof(kTestFilterList) /
+      sizeof(kTestFilterList[0]);
+
+  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+    const int16_t (*filters)[8] = kTestFilterList[filter_bank];
+    const int kNumFilters = 16;
+
+    for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+      for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+        filter_average_block2d_8_c(in, kInputStride,
+                                   filters[filter_x], filters[filter_y],
+                                   ref, kOutputStride,
+                                   Width(), Height());
+
+        if (filters == vp9_sub_pel_filters_8lp || (filter_x && filter_y))
+          REGISTER_STATE_CHECK(
+              UUT_->hv8_avg_(in, kInputStride, out, kOutputStride,
+                             filters[filter_x], 16, filters[filter_y], 16,
+                             Width(), Height()));
+        else if (filter_y)
+          REGISTER_STATE_CHECK(
+              UUT_->v8_avg_(in, kInputStride, out, kOutputStride,
+                            filters[filter_x], 16, filters[filter_y], 16,
+                            Width(), Height()));
+        else
+          REGISTER_STATE_CHECK(
+              UUT_->h8_avg_(in, kInputStride, out, kOutputStride,
+                            filters[filter_x], 16, filters[filter_y], 16,
+                            Width(), Height()));
+
+        CheckGuardBlocks();
+
+        for (int y = 0; y < Height(); ++y)
+          for (int x = 0; x < Width(); ++x)
+            ASSERT_EQ(ref[y * kOutputStride + x], out[y * kOutputStride + x])
+                << "mismatch at (" << x << "," << y << "), "
+                << "filters (" << filter_bank << ","
+                << filter_x << "," << filter_y << ")";
+      }
+    }
+  }
+}
+
+DECLARE_ALIGNED(256, const int16_t, kChangeFilters[16][8]) = {
+    { 0,   0,   0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0, 128},
+    { 0,   0,   0, 128},
+    { 0,   0, 128},
+    { 0, 128},
+    { 128},
+    { 0,   0,   0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0,   0, 128},
+    { 0,   0,   0,   0, 128},
+    { 0,   0,   0, 128},
+    { 0,   0, 128},
+    { 0, 128},
+    { 128}
+};
+
+TEST_P(ConvolveTest, ChangeFilterWorks) {
+  uint8_t* const in = input();
+  uint8_t* const out = output();
+  const int kPixelSelected = 4;
+
+  REGISTER_STATE_CHECK(UUT_->h8_(in, kInputStride, out, kOutputStride,
+                                 kChangeFilters[8], 17, kChangeFilters[4], 16,
+                                 Width(), Height()));
+
+  for (int x = 0; x < Width(); ++x) {
+    const int kQ4StepAdjust = x >> 4;
+    const int kFilterPeriodAdjust = (x >> 3) << 3;
+    const int ref_x = kQ4StepAdjust + kFilterPeriodAdjust + kPixelSelected;
+    ASSERT_EQ(in[ref_x], out[x]) << "x == " << x;
+  }
+
+  REGISTER_STATE_CHECK(UUT_->v8_(in, kInputStride, out, kOutputStride,
+                                 kChangeFilters[4], 16, kChangeFilters[8], 17,
+                                 Width(), Height()));
+
+  for (int y = 0; y < Height(); ++y) {
+    const int kQ4StepAdjust = y >> 4;
+    const int kFilterPeriodAdjust = (y >> 3) << 3;
+    const int ref_y = kQ4StepAdjust + kFilterPeriodAdjust + kPixelSelected;
+    ASSERT_EQ(in[ref_y * kInputStride], out[y * kInputStride]) << "y == " << y;
+  }
+
+  REGISTER_STATE_CHECK(UUT_->hv8_(in, kInputStride, out, kOutputStride,
+                                  kChangeFilters[8], 17, kChangeFilters[8], 17,
+                                  Width(), Height()));
+
+  for (int y = 0; y < Height(); ++y) {
+    const int kQ4StepAdjustY = y >> 4;
+    const int kFilterPeriodAdjustY = (y >> 3) << 3;
+    const int ref_y = kQ4StepAdjustY + kFilterPeriodAdjustY + kPixelSelected;
+    for (int x = 0; x < Width(); ++x) {
+      const int kQ4StepAdjustX = x >> 4;
+      const int kFilterPeriodAdjustX = (x >> 3) << 3;
+      const int ref_x = kQ4StepAdjustX + kFilterPeriodAdjustX + kPixelSelected;
+
+      ASSERT_EQ(in[ref_y * kInputStride + ref_x], out[y * kOutputStride + x])
+          << "x == " << x << ", y == " << y;
+    }
+  }
+}
+
+
+using std::tr1::make_tuple;
+
+const ConvolveFunctions convolve8_c(
+    vp9_convolve8_horiz_c, vp9_convolve8_avg_horiz_c,
+    vp9_convolve8_vert_c, vp9_convolve8_avg_vert_c,
+    vp9_convolve8_c, vp9_convolve8_avg_c);
+
+INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve8_c),
+    make_tuple(8, 4, &convolve8_c),
+    make_tuple(4, 8, &convolve8_c),
+    make_tuple(8, 8, &convolve8_c),
+    make_tuple(16, 8, &convolve8_c),
+    make_tuple(8, 16, &convolve8_c),
+    make_tuple(16, 16, &convolve8_c),
+    make_tuple(32, 16, &convolve8_c),
+    make_tuple(16, 32, &convolve8_c),
+    make_tuple(32, 32, &convolve8_c),
+    make_tuple(64, 32, &convolve8_c),
+    make_tuple(32, 64, &convolve8_c),
+    make_tuple(64, 64, &convolve8_c)));
+
+#if HAVE_SSSE3
+const ConvolveFunctions convolve8_ssse3(
+    vp9_convolve8_horiz_ssse3, vp9_convolve8_avg_horiz_c,
+    vp9_convolve8_vert_ssse3, vp9_convolve8_avg_vert_c,
+    vp9_convolve8_ssse3, vp9_convolve8_avg_c);
+
+INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve8_ssse3),
+    make_tuple(8, 4, &convolve8_ssse3),
+    make_tuple(4, 8, &convolve8_ssse3),
+    make_tuple(8, 8, &convolve8_ssse3),
+    make_tuple(16, 8, &convolve8_ssse3),
+    make_tuple(8, 16, &convolve8_ssse3),
+    make_tuple(16, 16, &convolve8_ssse3),
+    make_tuple(32, 16, &convolve8_ssse3),
+    make_tuple(16, 32, &convolve8_ssse3),
+    make_tuple(32, 32, &convolve8_ssse3),
+    make_tuple(64, 32, &convolve8_ssse3),
+    make_tuple(32, 64, &convolve8_ssse3),
+    make_tuple(64, 64, &convolve8_ssse3)));
+#endif
+}  // namespace
diff --git a/libvpx/test/cq_test.cc b/libvpx/test/cq_test.cc
index 42ee2a2..a6a4b8e 100644
--- a/libvpx/test/cq_test.cc
+++ b/libvpx/test/cq_test.cc
@@ -9,8 +9,12 @@
  */
 #include <cmath>
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
 #include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
 
 // CQ level range: [kCQLevelMin, kCQLevelMax).
 const int kCQLevelMin = 4;
@@ -18,12 +22,13 @@
 const int kCQLevelStep = 8;
 const int kCQTargetBitrate = 2000;
 
-namespace {
-
-class CQTest : public libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<int> {
+class CQTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<int> {
  protected:
-  CQTest() : cq_level_(GetParam()) { init_flags_ = VPX_CODEC_USE_PSNR; }
+  CQTest() : EncoderTest(GET_PARAM(0)), cq_level_(GET_PARAM(1)) {
+    init_flags_ = VPX_CODEC_USE_PSNR;
+  }
+
   virtual ~CQTest() {}
 
   virtual void SetUp() {
@@ -100,7 +105,7 @@
   EXPECT_GE(cq_psnr_lin, vbr_psnr_lin);
 }
 
-INSTANTIATE_TEST_CASE_P(CQLevelRange, CQTest,
-                        ::testing::Range(kCQLevelMin, kCQLevelMax,
-                                         kCQLevelStep));
+VP8_INSTANTIATE_TEST_CASE(CQTest,
+                          ::testing::Range(kCQLevelMin, kCQLevelMax,
+                                           kCQLevelStep));
 }  // namespace
diff --git a/libvpx/test/datarate_test.cc b/libvpx/test/datarate_test.cc
index 6fbcb64..85eeafb 100644
--- a/libvpx/test/datarate_test.cc
+++ b/libvpx/test/datarate_test.cc
@@ -7,17 +7,23 @@
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
 #include "test/i420_video_source.h"
-#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/util.h"
+
 namespace {
 
 class DatarateTest : public ::libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<enum libvpx_test::TestMode> {
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
+ public:
+  DatarateTest() : EncoderTest(GET_PARAM(0)) {}
+
  protected:
   virtual void SetUp() {
     InitializeConfig();
-    SetMode(GetParam());
+    SetMode(GET_PARAM(1));
     ResetModel();
   }
 
@@ -174,5 +180,6 @@
   }
 }
 
-INSTANTIATE_TEST_CASE_P(AllModes, DatarateTest, ALL_TEST_MODES);
+VP8_INSTANTIATE_TEST_CASE(DatarateTest, ALL_TEST_MODES);
+
 }  // namespace
diff --git a/libvpx/test/dct16x16_test.cc b/libvpx/test/dct16x16_test.cc
new file mode 100644
index 0000000..9fb45d6
--- /dev/null
+++ b/libvpx/test/dct16x16_test.cc
@@ -0,0 +1,366 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "vp9/common/vp9_entropy.h"
+#include "vp9_rtcd.h"
+void vp9_short_idct16x16_add_c(short *input, uint8_t *output, int pitch);
+}
+
+#include "acm_random.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+
+#ifdef _MSC_VER
+static int round(double x) {
+  if (x < 0)
+    return (int)ceil(x - 0.5);
+  else
+    return (int)floor(x + 0.5);
+}
+#endif
+
+const double PI = 3.1415926535898;
+void reference2_16x16_idct_2d(double *input, double *output) {
+  double x;
+  for (int l = 0; l < 16; ++l) {
+    for (int k = 0; k < 16; ++k) {
+      double s = 0;
+      for (int i = 0; i < 16; ++i) {
+        for (int j = 0; j < 16; ++j) {
+          x=cos(PI*j*(l+0.5)/16.0)*cos(PI*i*(k+0.5)/16.0)*input[i*16+j]/256;
+          if (i != 0)
+            x *= sqrt(2.0);
+          if (j != 0)
+            x *= sqrt(2.0);
+          s += x;
+        }
+      }
+      output[k*16+l] = s;
+    }
+  }
+}
+
+
+static const double C1 = 0.995184726672197;
+static const double C2 = 0.98078528040323;
+static const double C3 = 0.956940335732209;
+static const double C4 = 0.923879532511287;
+static const double C5 = 0.881921264348355;
+static const double C6 = 0.831469612302545;
+static const double C7 = 0.773010453362737;
+static const double C8 = 0.707106781186548;
+static const double C9 = 0.634393284163646;
+static const double C10 = 0.555570233019602;
+static const double C11 = 0.471396736825998;
+static const double C12 = 0.38268343236509;
+static const double C13 = 0.290284677254462;
+static const double C14 = 0.195090322016128;
+static const double C15 = 0.098017140329561;
+
+static void butterfly_16x16_dct_1d(double input[16], double output[16]) {
+  double step[16];
+  double intermediate[16];
+  double temp1, temp2;
+
+  // step 1
+  step[ 0] = input[0] + input[15];
+  step[ 1] = input[1] + input[14];
+  step[ 2] = input[2] + input[13];
+  step[ 3] = input[3] + input[12];
+  step[ 4] = input[4] + input[11];
+  step[ 5] = input[5] + input[10];
+  step[ 6] = input[6] + input[ 9];
+  step[ 7] = input[7] + input[ 8];
+  step[ 8] = input[7] - input[ 8];
+  step[ 9] = input[6] - input[ 9];
+  step[10] = input[5] - input[10];
+  step[11] = input[4] - input[11];
+  step[12] = input[3] - input[12];
+  step[13] = input[2] - input[13];
+  step[14] = input[1] - input[14];
+  step[15] = input[0] - input[15];
+
+  // step 2
+  output[0] = step[0] + step[7];
+  output[1] = step[1] + step[6];
+  output[2] = step[2] + step[5];
+  output[3] = step[3] + step[4];
+  output[4] = step[3] - step[4];
+  output[5] = step[2] - step[5];
+  output[6] = step[1] - step[6];
+  output[7] = step[0] - step[7];
+
+  temp1 = step[ 8]*C7;
+  temp2 = step[15]*C9;
+  output[ 8] = temp1 + temp2;
+
+  temp1 = step[ 9]*C11;
+  temp2 = step[14]*C5;
+  output[ 9] = temp1 - temp2;
+
+  temp1 = step[10]*C3;
+  temp2 = step[13]*C13;
+  output[10] = temp1 + temp2;
+
+  temp1 = step[11]*C15;
+  temp2 = step[12]*C1;
+  output[11] = temp1 - temp2;
+
+  temp1 = step[11]*C1;
+  temp2 = step[12]*C15;
+  output[12] = temp2 + temp1;
+
+  temp1 = step[10]*C13;
+  temp2 = step[13]*C3;
+  output[13] = temp2 - temp1;
+
+  temp1 = step[ 9]*C5;
+  temp2 = step[14]*C11;
+  output[14] = temp2 + temp1;
+
+  temp1 = step[ 8]*C9;
+  temp2 = step[15]*C7;
+  output[15] = temp2 - temp1;
+
+  // step 3
+  step[ 0] = output[0] + output[3];
+  step[ 1] = output[1] + output[2];
+  step[ 2] = output[1] - output[2];
+  step[ 3] = output[0] - output[3];
+
+  temp1 = output[4]*C14;
+  temp2 = output[7]*C2;
+  step[ 4] = temp1 + temp2;
+
+  temp1 = output[5]*C10;
+  temp2 = output[6]*C6;
+  step[ 5] = temp1 + temp2;
+
+  temp1 = output[5]*C6;
+  temp2 = output[6]*C10;
+  step[ 6] = temp2 - temp1;
+
+  temp1 = output[4]*C2;
+  temp2 = output[7]*C14;
+  step[ 7] = temp2 - temp1;
+
+  step[ 8] = output[ 8] + output[11];
+  step[ 9] = output[ 9] + output[10];
+  step[10] = output[ 9] - output[10];
+  step[11] = output[ 8] - output[11];
+
+  step[12] = output[12] + output[15];
+  step[13] = output[13] + output[14];
+  step[14] = output[13] - output[14];
+  step[15] = output[12] - output[15];
+
+  // step 4
+  output[ 0] = (step[ 0] + step[ 1]);
+  output[ 8] = (step[ 0] - step[ 1]);
+
+  temp1 = step[2]*C12;
+  temp2 = step[3]*C4;
+  temp1 = temp1 + temp2;
+  output[ 4] = 2*(temp1*C8);
+
+  temp1 = step[2]*C4;
+  temp2 = step[3]*C12;
+  temp1 = temp2 - temp1;
+  output[12] = 2*(temp1*C8);
+
+  output[ 2] = 2*((step[4] + step[ 5])*C8);
+  output[14] = 2*((step[7] - step[ 6])*C8);
+
+  temp1 = step[4] - step[5];
+  temp2 = step[6] + step[7];
+  output[ 6] = (temp1 + temp2);
+  output[10] = (temp1 - temp2);
+
+  intermediate[8] = step[8] + step[14];
+  intermediate[9] = step[9] + step[15];
+
+  temp1 = intermediate[8]*C12;
+  temp2 = intermediate[9]*C4;
+  temp1 = temp1 - temp2;
+  output[3] = 2*(temp1*C8);
+
+  temp1 = intermediate[8]*C4;
+  temp2 = intermediate[9]*C12;
+  temp1 = temp2 + temp1;
+  output[13] = 2*(temp1*C8);
+
+  output[ 9] = 2*((step[10] + step[11])*C8);
+
+  intermediate[11] = step[10] - step[11];
+  intermediate[12] = step[12] + step[13];
+  intermediate[13] = step[12] - step[13];
+  intermediate[14] = step[ 8] - step[14];
+  intermediate[15] = step[ 9] - step[15];
+
+  output[15] = (intermediate[11] + intermediate[12]);
+  output[ 1] = -(intermediate[11] - intermediate[12]);
+
+  output[ 7] = 2*(intermediate[13]*C8);
+
+  temp1 = intermediate[14]*C12;
+  temp2 = intermediate[15]*C4;
+  temp1 = temp1 - temp2;
+  output[11] = -2*(temp1*C8);
+
+  temp1 = intermediate[14]*C4;
+  temp2 = intermediate[15]*C12;
+  temp1 = temp2 + temp1;
+  output[ 5] = 2*(temp1*C8);
+}
+
+static void reference_16x16_dct_1d(double in[16], double out[16]) {
+  const double kPi = 3.141592653589793238462643383279502884;
+  const double kInvSqrt2 = 0.707106781186547524400844362104;
+  for (int k = 0; k < 16; k++) {
+    out[k] = 0.0;
+    for (int n = 0; n < 16; n++)
+      out[k] += in[n]*cos(kPi*(2*n+1)*k/32.0);
+    if (k == 0)
+      out[k] = out[k]*kInvSqrt2;
+  }
+}
+
+void reference_16x16_dct_2d(int16_t input[16*16], double output[16*16]) {
+  // First transform columns
+  for (int i = 0; i < 16; ++i) {
+    double temp_in[16], temp_out[16];
+    for (int j = 0; j < 16; ++j)
+      temp_in[j] = input[j*16 + i];
+    butterfly_16x16_dct_1d(temp_in, temp_out);
+    for (int j = 0; j < 16; ++j)
+      output[j*16 + i] = temp_out[j];
+  }
+  // Then transform rows
+  for (int i = 0; i < 16; ++i) {
+    double temp_in[16], temp_out[16];
+    for (int j = 0; j < 16; ++j)
+      temp_in[j] = output[j + i*16];
+    butterfly_16x16_dct_1d(temp_in, temp_out);
+    // Scale by some magic number
+    for (int j = 0; j < 16; ++j)
+      output[j + i*16] = temp_out[j]/2;
+  }
+}
+
+
+TEST(VP9Idct16x16Test, AccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t in[256], coeff[256];
+    uint8_t dst[256], src[256];
+    double out_r[256];
+
+    for (int j = 0; j < 256; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 256; ++j)
+      in[j] = src[j] - dst[j];
+
+    reference_16x16_dct_2d(in, out_r);
+    for (int j = 0; j < 256; j++)
+      coeff[j] = round(out_r[j]);
+    vp9_short_idct16x16_add_c(coeff, dst, 16);
+    for (int j = 0; j < 256; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      EXPECT_GE(1, error)
+          << "Error: 16x16 IDCT has error " << error
+          << " at index " << j;
+    }
+  }
+}
+
+// we need enable fdct test once we re-do the 16 point fdct.
+TEST(VP9Fdct16x16Test, AccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int max_error = 0;
+  double total_error = 0;
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t test_input_block[256];
+    int16_t test_temp_block[256];
+    uint8_t dst[256], src[256];
+
+    for (int j = 0; j < 256; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 256; ++j)
+      test_input_block[j] = src[j] - dst[j];
+
+    const int pitch = 32;
+    vp9_short_fdct16x16_c(test_input_block, test_temp_block, pitch);
+    vp9_short_idct16x16_add_c(test_temp_block, dst, 16);
+
+    for (int j = 0; j < 256; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+      total_error += error;
+    }
+  }
+
+  EXPECT_GE(1, max_error)
+      << "Error: 16x16 FDCT/IDCT has an individual round trip error > 1";
+
+  EXPECT_GE(count_test_block , total_error)
+      << "Error: 16x16 FDCT/IDCT has average round trip error > 1 per block";
+}
+
+TEST(VP9Fdct16x16Test, CoeffSizeCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t input_block[256], input_extreme_block[256];
+    int16_t output_block[256], output_extreme_block[256];
+
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 256; ++j) {
+      input_block[j] = rnd.Rand8() - rnd.Rand8();
+      input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+    }
+    if (i == 0)
+      for (int j = 0; j < 256; ++j)
+        input_extreme_block[j] = 255;
+
+    const int pitch = 32;
+    vp9_short_fdct16x16_c(input_block, output_block, pitch);
+    vp9_short_fdct16x16_c(input_extreme_block, output_extreme_block, pitch);
+
+    // The minimum quant value is 4.
+    for (int j = 0; j < 256; ++j) {
+      EXPECT_GE(4*DCT_MAX_VALUE, abs(output_block[j]))
+          << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
+      EXPECT_GE(4*DCT_MAX_VALUE, abs(output_extreme_block[j]))
+          << "Error: 16x16 FDCT extreme has coefficient larger than 4*DCT_MAX_VALUE";
+    }
+  }
+}
+}  // namespace
diff --git a/libvpx/test/dct32x32_test.cc b/libvpx/test/dct32x32_test.cc
new file mode 100644
index 0000000..e05d482
--- /dev/null
+++ b/libvpx/test/dct32x32_test.cc
@@ -0,0 +1,192 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "vp9/common/vp9_entropy.h"
+#include "./vp9_rtcd.h"
+  void vp9_short_fdct32x32_c(int16_t *input, int16_t *out, int pitch);
+  void vp9_short_idct32x32_add_c(short *input, uint8_t *output, int pitch);
+}
+
+#include "test/acm_random.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+#ifdef _MSC_VER
+static int round(double x) {
+  if (x < 0)
+    return (int)ceil(x - 0.5);
+  else
+    return (int)floor(x + 0.5);
+}
+#endif
+
+static const double kPi = 3.141592653589793238462643383279502884;
+static void reference2_32x32_idct_2d(double *input, double *output) {
+  double x;
+  for (int l = 0; l < 32; ++l) {
+    for (int k = 0; k < 32; ++k) {
+      double s = 0;
+      for (int i = 0; i < 32; ++i) {
+        for (int j = 0; j < 32; ++j) {
+          x = cos(kPi * j * (l + 0.5) / 32.0) *
+              cos(kPi * i * (k + 0.5) / 32.0) * input[i * 32 + j] / 1024;
+          if (i != 0)
+            x *= sqrt(2.0);
+          if (j != 0)
+            x *= sqrt(2.0);
+          s += x;
+        }
+      }
+      output[k * 32 + l] = s / 4;
+    }
+  }
+}
+
+static void reference_32x32_dct_1d(double in[32], double out[32], int stride) {
+  const double kInvSqrt2 = 0.707106781186547524400844362104;
+  for (int k = 0; k < 32; k++) {
+    out[k] = 0.0;
+    for (int n = 0; n < 32; n++)
+      out[k] += in[n] * cos(kPi * (2 * n + 1) * k / 64.0);
+    if (k == 0)
+      out[k] = out[k] * kInvSqrt2;
+  }
+}
+
+static void reference_32x32_dct_2d(int16_t input[32*32], double output[32*32]) {
+  // First transform columns
+  for (int i = 0; i < 32; ++i) {
+    double temp_in[32], temp_out[32];
+    for (int j = 0; j < 32; ++j)
+      temp_in[j] = input[j*32 + i];
+    reference_32x32_dct_1d(temp_in, temp_out, 1);
+    for (int j = 0; j < 32; ++j)
+      output[j * 32 + i] = temp_out[j];
+  }
+  // Then transform rows
+  for (int i = 0; i < 32; ++i) {
+    double temp_in[32], temp_out[32];
+    for (int j = 0; j < 32; ++j)
+      temp_in[j] = output[j + i*32];
+    reference_32x32_dct_1d(temp_in, temp_out, 1);
+    // Scale by some magic number
+    for (int j = 0; j < 32; ++j)
+      output[j + i * 32] = temp_out[j] / 4;
+  }
+}
+
+TEST(VP9Idct32x32Test, AccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t in[1024], coeff[1024];
+    uint8_t dst[1024], src[1024];
+    double out_r[1024];
+
+    for (int j = 0; j < 1024; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 1024; ++j)
+      in[j] = src[j] - dst[j];
+
+    reference_32x32_dct_2d(in, out_r);
+    for (int j = 0; j < 1024; j++)
+      coeff[j] = round(out_r[j]);
+    vp9_short_idct32x32_add_c(coeff, dst, 32);
+    for (int j = 0; j < 1024; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      EXPECT_GE(1, error)
+          << "Error: 32x32 IDCT has error " << error
+          << " at index " << j;
+    }
+  }
+}
+
+TEST(VP9Fdct32x32Test, AccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  unsigned int max_error = 0;
+  int64_t total_error = 0;
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t test_input_block[1024];
+    int16_t test_temp_block[1024];
+    uint8_t dst[1024], src[1024];
+
+    for (int j = 0; j < 1024; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 1024; ++j)
+      test_input_block[j] = src[j] - dst[j];
+
+    const int pitch = 64;
+    vp9_short_fdct32x32_c(test_input_block, test_temp_block, pitch);
+    vp9_short_idct32x32_add_c(test_temp_block, dst, 32);
+
+    for (int j = 0; j < 1024; ++j) {
+      const unsigned diff = dst[j] - src[j];
+      const unsigned error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+      total_error += error;
+    }
+  }
+
+  EXPECT_GE(1u, max_error)
+      << "Error: 32x32 FDCT/IDCT has an individual roundtrip error > 1";
+
+  EXPECT_GE(count_test_block, total_error)
+      << "Error: 32x32 FDCT/IDCT has average roundtrip error > 1 per block";
+}
+
+TEST(VP9Fdct32x32Test, CoeffSizeCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 1000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t input_block[1024], input_extreme_block[1024];
+    int16_t output_block[1024], output_extreme_block[1024];
+
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 1024; ++j) {
+      input_block[j] = rnd.Rand8() - rnd.Rand8();
+      input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+    }
+    if (i == 0)
+      for (int j = 0; j < 1024; ++j)
+        input_extreme_block[j] = 255;
+
+    const int pitch = 64;
+    vp9_short_fdct32x32_c(input_block, output_block, pitch);
+    vp9_short_fdct32x32_c(input_extreme_block, output_extreme_block, pitch);
+
+    // The minimum quant value is 4.
+    for (int j = 0; j < 1024; ++j) {
+      EXPECT_GE(4*DCT_MAX_VALUE, abs(output_block[j]))
+          << "Error: 32x32 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
+      EXPECT_GE(4*DCT_MAX_VALUE, abs(output_extreme_block[j]))
+          << "Error: 32x32 FDCT extreme has coefficient larger than "
+             "4*DCT_MAX_VALUE";
+    }
+  }
+}
+}  // namespace
diff --git a/libvpx/test/decode_test_driver.cc b/libvpx/test/decode_test_driver.cc
index 3610f02..1f6d540 100644
--- a/libvpx/test/decode_test_driver.cc
+++ b/libvpx/test/decode_test_driver.cc
@@ -7,40 +7,41 @@
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
+#include "test/codec_factory.h"
 #include "test/decode_test_driver.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/register_state_check.h"
 #include "test/video_source.h"
 
 namespace libvpx_test {
-#if CONFIG_VP8_DECODER
-void Decoder::DecodeFrame(const uint8_t *cxdata, int size) {
-  if (!decoder_.priv) {
-    const vpx_codec_err_t res_init = vpx_codec_dec_init(&decoder_,
-                                                        &vpx_codec_vp8_dx_algo,
-                                                        &cfg_, 0);
-    ASSERT_EQ(VPX_CODEC_OK, res_init) << DecodeError();
-  }
 
-  const vpx_codec_err_t res_dec = vpx_codec_decode(&decoder_,
-                                                   cxdata, size, NULL, 0);
-  ASSERT_EQ(VPX_CODEC_OK, res_dec) << DecodeError();
+vpx_codec_err_t Decoder::DecodeFrame(const uint8_t *cxdata, int size) {
+  vpx_codec_err_t res_dec;
+  InitOnce();
+  REGISTER_STATE_CHECK(res_dec = vpx_codec_decode(&decoder_,
+                                                  cxdata, size, NULL, 0));
+  return res_dec;
 }
 
 void DecoderTest::RunLoop(CompressedVideoSource *video) {
   vpx_codec_dec_cfg_t dec_cfg = {0};
-  Decoder decoder(dec_cfg, 0);
+  Decoder* const decoder = codec_->CreateDecoder(dec_cfg, 0);
+  ASSERT_TRUE(decoder != NULL);
 
   // Decode frames.
   for (video->Begin(); video->cxdata(); video->Next()) {
-    decoder.DecodeFrame(video->cxdata(), video->frame_size());
+    vpx_codec_err_t res_dec = decoder->DecodeFrame(video->cxdata(),
+                                                   video->frame_size());
+    ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder->DecodeError();
 
-    DxDataIterator dec_iter = decoder.GetDxData();
+    DxDataIterator dec_iter = decoder->GetDxData();
     const vpx_image_t *img = NULL;
 
     // Get decompressed data
     while ((img = dec_iter.Next()))
       DecompressedFrameHook(*img, video->frame_number());
   }
+
+  delete decoder;
 }
-#endif
 }  // namespace libvpx_test
diff --git a/libvpx/test/decode_test_driver.h b/libvpx/test/decode_test_driver.h
index 6408bee..49e7384 100644
--- a/libvpx/test/decode_test_driver.h
+++ b/libvpx/test/decode_test_driver.h
@@ -14,10 +14,10 @@
 #include "third_party/googletest/src/include/gtest/gtest.h"
 #include "vpx_config.h"
 #include "vpx/vpx_decoder.h"
-#include "vpx/vp8dx.h"
 
 namespace libvpx_test {
 
+class CodecFactory;
 class CompressedVideoSource;
 
 // Provides an object to handle decoding output
@@ -42,15 +42,15 @@
 class Decoder {
  public:
   Decoder(vpx_codec_dec_cfg_t cfg, unsigned long deadline)
-      : cfg_(cfg), deadline_(deadline) {
+      : cfg_(cfg), deadline_(deadline), init_done_(false) {
     memset(&decoder_, 0, sizeof(decoder_));
   }
 
-  ~Decoder() {
+  virtual ~Decoder() {
     vpx_codec_destroy(&decoder_);
   }
 
-  void DecodeFrame(const uint8_t *cxdata, int size);
+  vpx_codec_err_t DecodeFrame(const uint8_t *cxdata, int size);
 
   DxDataIterator GetDxData() {
     return DxDataIterator(&decoder_);
@@ -61,25 +61,45 @@
   }
 
   void Control(int ctrl_id, int arg) {
+    InitOnce();
     const vpx_codec_err_t res = vpx_codec_control_(&decoder_, ctrl_id, arg);
     ASSERT_EQ(VPX_CODEC_OK, res) << DecodeError();
   }
 
- protected:
-  const char *DecodeError() {
+  void Control(int ctrl_id, const void *arg) {
+    InitOnce();
+    const vpx_codec_err_t res = vpx_codec_control_(&decoder_, ctrl_id, arg);
+    ASSERT_EQ(VPX_CODEC_OK, res) << DecodeError();
+  }
+
+  const char* DecodeError() {
     const char *detail = vpx_codec_error_detail(&decoder_);
     return detail ? detail : vpx_codec_error(&decoder_);
   }
 
+ protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const = 0;
+
+  void InitOnce() {
+    if (!init_done_) {
+      const vpx_codec_err_t res = vpx_codec_dec_init(&decoder_,
+                                                     CodecInterface(),
+                                                     &cfg_, 0);
+      ASSERT_EQ(VPX_CODEC_OK, res) << DecodeError();
+      init_done_ = true;
+    }
+  }
+
   vpx_codec_ctx_t     decoder_;
   vpx_codec_dec_cfg_t cfg_;
   unsigned int        deadline_;
+  bool                init_done_;
 };
 
 // Common test functionality for all Decoder tests.
 class DecoderTest {
  public:
-  // Main loop.
+  // Main decoding loop
   virtual void RunLoop(CompressedVideoSource *video);
 
   // Hook to be called on every decompressed frame.
@@ -87,9 +107,11 @@
                                      const unsigned int frame_number) {}
 
  protected:
-  DecoderTest() {}
+  explicit DecoderTest(const CodecFactory *codec) : codec_(codec) {}
 
   virtual ~DecoderTest() {}
+
+  const CodecFactory *codec_;
 };
 
 }  // namespace libvpx_test
diff --git a/libvpx/test/encode_test_driver.cc b/libvpx/test/encode_test_driver.cc
index ebb3959..eed3e33 100644
--- a/libvpx/test/encode_test_driver.cc
+++ b/libvpx/test/encode_test_driver.cc
@@ -7,11 +7,12 @@
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
+
 #include "vpx_config.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
-#if CONFIG_VP8_DECODER
 #include "test/decode_test_driver.h"
-#endif
+#include "test/register_state_check.h"
 #include "test/video_source.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
@@ -44,7 +45,7 @@
     cfg_.g_h = img->d_h;
     cfg_.g_timebase = video.timebase();
     cfg_.rc_twopass_stats_in = stats_->buf();
-    res = vpx_codec_enc_init(&encoder_, &vpx_codec_vp8_cx_algo, &cfg_,
+    res = vpx_codec_enc_init(&encoder_, CodecInterface(), &cfg_,
                              init_flags_);
     ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
   }
@@ -58,9 +59,10 @@
   }
 
   // Encode the frame
-  res = vpx_codec_encode(&encoder_,
-                         video.img(), video.pts(), video.duration(),
-                         frame_flags, deadline_);
+  REGISTER_STATE_CHECK(
+      res = vpx_codec_encode(&encoder_,
+                             video.img(), video.pts(), video.duration(),
+                             frame_flags, deadline_));
   ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
 }
 
@@ -70,6 +72,11 @@
   ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
 }
 
+void EncoderTest::InitializeConfig() {
+  const vpx_codec_err_t res = codec_->DefaultEncoderConfig(&cfg_, 0);
+  ASSERT_EQ(VPX_CODEC_OK, res);
+}
+
 void EncoderTest::SetMode(TestMode mode) {
   switch (mode) {
     case kRealTime:
@@ -123,13 +130,17 @@
   return match;
 }
 
+void EncoderTest::MismatchHook(const vpx_image_t *img1,
+                               const vpx_image_t *img2) {
+  ASSERT_TRUE(0) << "Encode/Decode mismatch found";
+}
+
 void EncoderTest::RunLoop(VideoSource *video) {
-#if CONFIG_VP8_DECODER
   vpx_codec_dec_cfg_t dec_cfg = {0};
-#endif
 
   stats_.Reset();
 
+  ASSERT_TRUE(passes_ == 1 || passes_ == 2);
   for (unsigned int pass = 0; pass < passes_; pass++) {
     last_pts_ = 0;
 
@@ -141,31 +152,34 @@
       cfg_.g_pass = VPX_RC_LAST_PASS;
 
     BeginPassHook(pass);
-    Encoder encoder(cfg_, deadline_, init_flags_, &stats_);
-#if CONFIG_VP8_DECODER
-    Decoder decoder(dec_cfg, 0);
-    bool has_cxdata = false;
-#endif
+    Encoder* const encoder = codec_->CreateEncoder(cfg_, deadline_, init_flags_,
+                                                   &stats_);
+    ASSERT_TRUE(encoder != NULL);
+    Decoder* const decoder = codec_->CreateDecoder(dec_cfg, 0);
     bool again;
     for (again = true, video->Begin(); again; video->Next()) {
       again = video->img() != NULL;
 
       PreEncodeFrameHook(video);
-      PreEncodeFrameHook(video, &encoder);
-      encoder.EncodeFrame(video, frame_flags_);
+      PreEncodeFrameHook(video, encoder);
+      encoder->EncodeFrame(video, frame_flags_);
 
-      CxDataIterator iter = encoder.GetCxData();
+      CxDataIterator iter = encoder->GetCxData();
 
+      bool has_cxdata = false;
+      bool has_dxdata = false;
       while (const vpx_codec_cx_pkt_t *pkt = iter.Next()) {
+        pkt = MutateEncoderOutputHook(pkt);
         again = true;
-
         switch (pkt->kind) {
           case VPX_CODEC_CX_FRAME_PKT:
-#if CONFIG_VP8_DECODER
             has_cxdata = true;
-            decoder.DecodeFrame((const uint8_t*)pkt->data.frame.buf,
-                                pkt->data.frame.sz);
-#endif
+            if (decoder && DoDecode()) {
+              vpx_codec_err_t res_dec = decoder->DecodeFrame(
+                  (const uint8_t*)pkt->data.frame.buf, pkt->data.frame.sz);
+              ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder->DecodeError();
+              has_dxdata = true;
+            }
             ASSERT_GE(pkt->data.frame.pts, last_pts_);
             last_pts_ = pkt->data.frame.pts;
             FramePktHook(pkt);
@@ -180,25 +194,32 @@
         }
       }
 
-#if CONFIG_VP8_DECODER
-      if (has_cxdata) {
-        const vpx_image_t *img_enc = encoder.GetPreviewFrame();
-        DxDataIterator dec_iter = decoder.GetDxData();
+      if (has_dxdata && has_cxdata) {
+        const vpx_image_t *img_enc = encoder->GetPreviewFrame();
+        DxDataIterator dec_iter = decoder->GetDxData();
         const vpx_image_t *img_dec = dec_iter.Next();
-        if(img_enc && img_dec) {
+        if (img_enc && img_dec) {
           const bool res = compare_img(img_enc, img_dec);
-          ASSERT_TRUE(res)<< "Encoder/Decoder mismatch found.";
+          if (!res) {  // Mismatch
+            MismatchHook(img_enc, img_dec);
+          }
         }
+        if (img_dec)
+          DecompressedFrameHook(*img_dec, video->pts());
       }
-#endif
       if (!Continue())
         break;
     }
 
     EndPassHook();
 
+    if (decoder)
+      delete decoder;
+    delete encoder;
+
     if (!Continue())
       break;
   }
 }
+
 }  // namespace libvpx_test
diff --git a/libvpx/test/encode_test_driver.h b/libvpx/test/encode_test_driver.h
index 0141fa9..6aeb96b 100644
--- a/libvpx/test/encode_test_driver.h
+++ b/libvpx/test/encode_test_driver.h
@@ -9,14 +9,17 @@
  */
 #ifndef TEST_ENCODE_TEST_DRIVER_H_
 #define TEST_ENCODE_TEST_DRIVER_H_
+
 #include <string>
 #include <vector>
+
+#include "./vpx_config.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 #include "vpx/vpx_encoder.h"
-#include "vpx/vp8cx.h"
 
 namespace libvpx_test {
 
+class CodecFactory;
 class VideoSource;
 
 enum TestMode {
@@ -36,12 +39,15 @@
                                               ::libvpx_test::kOnePassGood, \
                                               ::libvpx_test::kOnePassBest)
 
+#define TWO_PASS_TEST_MODES ::testing::Values(::libvpx_test::kTwoPassGood, \
+                                              ::libvpx_test::kTwoPassBest)
+
 
 // Provides an object to handle the libvpx get_cx_data() iteration pattern
 class CxDataIterator {
  public:
   explicit CxDataIterator(vpx_codec_ctx_t *encoder)
-    : encoder_(encoder), iter_(NULL) {}
+      : encoder_(encoder), iter_(NULL) {}
 
   const vpx_codec_cx_pkt_t *Next() {
     return vpx_codec_get_cx_data(encoder_, &iter_);
@@ -83,11 +89,11 @@
  public:
   Encoder(vpx_codec_enc_cfg_t cfg, unsigned long deadline,
           const unsigned long init_flags, TwopassStatsStore *stats)
-    : cfg_(cfg), deadline_(deadline), init_flags_(init_flags), stats_(stats) {
+      : cfg_(cfg), deadline_(deadline), init_flags_(init_flags), stats_(stats) {
     memset(&encoder_, 0, sizeof(encoder_));
   }
 
-  ~Encoder() {
+  virtual ~Encoder() {
     vpx_codec_destroy(&encoder_);
   }
 
@@ -112,11 +118,18 @@
     ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
   }
 
+  void Control(int ctrl_id, struct vpx_scaling_mode *arg) {
+    const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
+  }
+
   void set_deadline(unsigned long deadline) {
     deadline_ = deadline;
   }
 
  protected:
+  virtual const vpx_codec_iface_t* CodecInterface() const = 0;
+
   const char *EncoderError() {
     const char *detail = vpx_codec_error_detail(&encoder_);
     return detail ? detail : vpx_codec_error(&encoder_);
@@ -145,22 +158,19 @@
 // classes directly, so that tests can be parameterized differently.
 class EncoderTest {
  protected:
-  EncoderTest() : abort_(false), init_flags_(0), frame_flags_(0),
-                  last_pts_(0) {}
+  explicit EncoderTest(const CodecFactory *codec)
+      : codec_(codec), abort_(false), init_flags_(0), frame_flags_(0),
+        last_pts_(0) {}
 
   virtual ~EncoderTest() {}
 
   // Initialize the cfg_ member with the default configuration.
-  void InitializeConfig() {
-    const vpx_codec_err_t res = vpx_codec_enc_config_default(
-                                    &vpx_codec_vp8_cx_algo, &cfg_, 0);
-    ASSERT_EQ(VPX_CODEC_OK, res);
-  }
+  void InitializeConfig();
 
   // Map the TestMode enum to the deadline_ and passes_ variables.
   void SetMode(TestMode mode);
 
-  // Main loop.
+  // Main loop
   virtual void RunLoop(VideoSource *video);
 
   // Hook to be called at the beginning of a pass.
@@ -182,6 +192,24 @@
   // Hook to determine whether the encode loop should continue.
   virtual bool Continue() const { return !abort_; }
 
+  const CodecFactory   *codec_;
+  // Hook to determine whether to decode frame after encoding
+  virtual bool DoDecode() const { return 1; }
+
+  // Hook to handle encode/decode mismatch
+  virtual void MismatchHook(const vpx_image_t *img1,
+                            const vpx_image_t *img2);
+
+  // Hook to be called on every decompressed frame.
+  virtual void DecompressedFrameHook(const vpx_image_t& img,
+                                     vpx_codec_pts_t pts) {}
+
+  // Hook that can modify the encoder's output data
+  virtual const vpx_codec_cx_pkt_t * MutateEncoderOutputHook(
+      const vpx_codec_cx_pkt_t *pkt) {
+    return pkt;
+  }
+
   bool                 abort_;
   vpx_codec_enc_cfg_t  cfg_;
   unsigned int         passes_;
diff --git a/libvpx/test/error_resilience_test.cc b/libvpx/test/error_resilience_test.cc
index 25c6731..ddfbd0f 100644
--- a/libvpx/test/error_resilience_test.cc
+++ b/libvpx/test/error_resilience_test.cc
@@ -7,22 +7,37 @@
   in the file PATENTS.  All contributing project authors may
   be found in the AUTHORS file in the root of the source tree.
 */
+
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
 #include "test/i420_video_source.h"
+#include "test/util.h"
 
 namespace {
 
-class ErrorResilienceTest : public libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<int> {
+const int kMaxErrorFrames = 8;
+const int kMaxDroppableFrames = 8;
+
+class ErrorResilienceTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
  protected:
-  ErrorResilienceTest() {
-    psnr_ = 0.0;
-    nframes_ = 0;
-    encoding_mode_ = static_cast<libvpx_test::TestMode>(GetParam());
+  ErrorResilienceTest() : EncoderTest(GET_PARAM(0)),
+                          psnr_(0.0),
+                          nframes_(0),
+                          mismatch_psnr_(0.0),
+                          mismatch_nframes_(0),
+                          encoding_mode_(GET_PARAM(1)) {
+    Reset();
   }
+
   virtual ~ErrorResilienceTest() {}
 
+  void Reset() {
+    error_nframes_ = 0;
+    droppable_nframes_ = 0;
+  }
+
   virtual void SetUp() {
     InitializeConfig();
     SetMode(encoding_mode_);
@@ -31,6 +46,8 @@
   virtual void BeginPassHook(unsigned int /*pass*/) {
     psnr_ = 0.0;
     nframes_ = 0;
+    mismatch_psnr_ = 0.0;
+    mismatch_nframes_ = 0;
   }
 
   virtual bool Continue() const {
@@ -42,15 +59,92 @@
     nframes_++;
   }
 
+  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video) {
+    frame_flags_ &= ~(VP8_EFLAG_NO_UPD_LAST |
+                      VP8_EFLAG_NO_UPD_GF |
+                      VP8_EFLAG_NO_UPD_ARF);
+    if (droppable_nframes_ > 0 &&
+        (cfg_.g_pass == VPX_RC_LAST_PASS || cfg_.g_pass == VPX_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < droppable_nframes_; ++i) {
+        if (droppable_frames_[i] == nframes_) {
+          std::cout << "             Encoding droppable frame: "
+                    << droppable_frames_[i] << "\n";
+          frame_flags_ |= (VP8_EFLAG_NO_UPD_LAST |
+                           VP8_EFLAG_NO_UPD_GF |
+                           VP8_EFLAG_NO_UPD_ARF);
+          return;
+        }
+      }
+    }
+  }
+
   double GetAveragePsnr() const {
     if (nframes_)
       return psnr_ / nframes_;
     return 0.0;
   }
 
+  double GetAverageMismatchPsnr() const {
+    if (mismatch_nframes_)
+      return mismatch_psnr_ / mismatch_nframes_;
+    return 0.0;
+  }
+
+  virtual bool DoDecode() const {
+    if (error_nframes_ > 0 &&
+        (cfg_.g_pass == VPX_RC_LAST_PASS || cfg_.g_pass == VPX_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < error_nframes_; ++i) {
+        if (error_frames_[i] == nframes_ - 1) {
+          std::cout << "             Skipping decoding frame: "
+                    << error_frames_[i] << "\n";
+          return 0;
+        }
+      }
+    }
+    return 1;
+  }
+
+  virtual void MismatchHook(const vpx_image_t *img1,
+                            const vpx_image_t *img2) {
+    double mismatch_psnr = compute_psnr(img1, img2);
+    mismatch_psnr_ += mismatch_psnr;
+    ++mismatch_nframes_;
+    // std::cout << "Mismatch frame psnr: " << mismatch_psnr << "\n";
+  }
+
+  void SetErrorFrames(int num, unsigned int *list) {
+    if (num > kMaxErrorFrames)
+      num = kMaxErrorFrames;
+    else if (num < 0)
+      num = 0;
+    error_nframes_ = num;
+    for (unsigned int i = 0; i < error_nframes_; ++i)
+      error_frames_[i] = list[i];
+  }
+
+  void SetDroppableFrames(int num, unsigned int *list) {
+    if (num > kMaxDroppableFrames)
+      num = kMaxDroppableFrames;
+    else if (num < 0)
+      num = 0;
+    droppable_nframes_ = num;
+    for (unsigned int i = 0; i < droppable_nframes_; ++i)
+      droppable_frames_[i] = list[i];
+  }
+
+  unsigned int GetMismatchFrames() {
+    return mismatch_nframes_;
+  }
+
  private:
   double psnr_;
   unsigned int nframes_;
+  unsigned int error_nframes_;
+  unsigned int droppable_nframes_;
+  double mismatch_psnr_;
+  unsigned int mismatch_nframes_;
+  unsigned int error_frames_[kMaxErrorFrames];
+  unsigned int droppable_frames_[kMaxDroppableFrames];
   libvpx_test::TestMode encoding_mode_;
 };
 
@@ -85,6 +179,56 @@
   }
 }
 
-INSTANTIATE_TEST_CASE_P(OnOffTest, ErrorResilienceTest,
-                        ONE_PASS_TEST_MODES);
+TEST_P(ErrorResilienceTest, DropFramesWithoutRecovery) {
+  const vpx_rational timebase = { 33333333, 1000000000 };
+  cfg_.g_timebase = timebase;
+  cfg_.rc_target_bitrate = 500;
+
+  init_flags_ = VPX_CODEC_USE_PSNR;
+
+  libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     timebase.den, timebase.num, 0, 30);
+
+  // Error resilient mode ON.
+  cfg_.g_error_resilient = 1;
+
+  // Set an arbitrary set of error frames same as droppable frames
+  unsigned int num_droppable_frames = 2;
+  unsigned int droppable_frame_list[] = {5, 16};
+  SetDroppableFrames(num_droppable_frames, droppable_frame_list);
+  SetErrorFrames(num_droppable_frames, droppable_frame_list);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  // Test that no mismatches have been found
+  std::cout << "             Mismatch frames: "
+            << GetMismatchFrames() << "\n";
+  EXPECT_EQ(GetMismatchFrames(), (unsigned int) 0);
+
+  // reset previously set error/droppable frames
+  Reset();
+
+#if 0
+  // TODO(jkoleszar): This test is disabled for the time being as too
+  // sensitive. It's not clear how to set a reasonable threshold for
+  // this behavior.
+
+  // Now set an arbitrary set of error frames that are non-droppable
+  unsigned int num_error_frames = 3;
+  unsigned int error_frame_list[] = {3, 10, 20};
+  SetErrorFrames(num_error_frames, error_frame_list);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Test that dropping an arbitrary set of inter frames does not hurt too much
+  // Note the Average Mismatch PSNR is the average of the PSNR between
+  // decoded frame and encoder's version of the same frame for all frames
+  // with mismatch.
+  const double psnr_resilience_mismatch = GetAverageMismatchPsnr();
+  std::cout << "             Mismatch PSNR: "
+            << psnr_resilience_mismatch << "\n";
+  EXPECT_GT(psnr_resilience_mismatch, 20.0);
+#endif
+}
+
+VP8_INSTANTIATE_TEST_CASE(ErrorResilienceTest, ONE_PASS_TEST_MODES);
+VP9_INSTANTIATE_TEST_CASE(ErrorResilienceTest, ONE_PASS_TEST_MODES);
+
 }  // namespace
diff --git a/libvpx/test/fdct4x4_test.cc b/libvpx/test/fdct4x4_test.cc
index 619b23d..1c887bb 100644
--- a/libvpx/test/fdct4x4_test.cc
+++ b/libvpx/test/fdct4x4_test.cc
@@ -1,79 +1,31 @@
 /*
-*  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
-*
-*  Use of this source code is governed by a BSD-style license
-*  that can be found in the LICENSE file in the root of the source
-*  tree. An additional intellectual property rights grant can be found
-*  in the file PATENTS.  All contributing project authors may
-*  be found in the AUTHORS file in the root of the source tree.
-*/
-
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
 
 #include <math.h>
-#include <stddef.h>
-#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/types.h>
 
+#include "third_party/googletest/src/include/gtest/gtest.h"
 
 extern "C" {
-#include "vpx_rtcd.h"
+#include "vp9_rtcd.h"
 }
 
-#include "test/acm_random.h"
-#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "acm_random.h"
 #include "vpx/vpx_integer.h"
 
-
-namespace {
-
-const int cospi8sqrt2minus1 = 20091;
-const int sinpi8sqrt2 = 35468;
-
-void reference_idct4x4(const int16_t *input, int16_t *output) {
-  const int16_t *ip = input;
-  int16_t *op = output;
-
-  for (int i = 0; i < 4; ++i) {
-    const int a1 = ip[0] + ip[8];
-    const int b1 = ip[0] - ip[8];
-    const int temp1 = (ip[4] * sinpi8sqrt2) >> 16;
-    const int temp2 = ip[12] + ((ip[12] * cospi8sqrt2minus1) >> 16);
-    const int c1 = temp1 - temp2;
-    const int temp3 = ip[4] + ((ip[4] * cospi8sqrt2minus1) >> 16);
-    const int temp4 = (ip[12] * sinpi8sqrt2) >> 16;
-    const int d1 = temp3 + temp4;
-    op[0] = a1 + d1;
-    op[12] = a1 - d1;
-    op[4] = b1 + c1;
-    op[8] = b1 - c1;
-    ++ip;
-    ++op;
-  }
-  ip = output;
-  op = output;
-  for (int i = 0; i < 4; ++i) {
-    const int a1 = ip[0] + ip[2];
-    const int b1 = ip[0] - ip[2];
-    const int temp1 = (ip[1] * sinpi8sqrt2) >> 16;
-    const int temp2 = ip[3] + ((ip[3] * cospi8sqrt2minus1) >> 16);
-    const int c1 = temp1 - temp2;
-    const int temp3 = ip[1] + ((ip[1] * cospi8sqrt2minus1) >> 16);
-    const int temp4 = (ip[3] * sinpi8sqrt2) >> 16;
-    const int d1 = temp3 + temp4;
-    op[0] = (a1 + d1 + 4) >> 3;
-    op[3] = (a1 - d1 + 4) >> 3;
-    op[1] = (b1 + c1 + 4) >> 3;
-    op[2] = (b1 - c1 + 4) >> 3;
-    ip += 4;
-    op += 4;
-  }
-}
-
 using libvpx_test::ACMRandom;
 
-TEST(Vp8FdctTest, SignBiasCheck) {
+namespace {
+
+TEST(Vp9Fdct4x4Test, SignBiasCheck) {
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   int16_t test_input_block[16];
   int16_t test_output_block[16];
@@ -88,7 +40,9 @@
     for (int j = 0; j < 16; ++j)
       test_input_block[j] = rnd.Rand8() - rnd.Rand8();
 
-    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+    // TODO(Yaowu): this should be converted to a parameterized test
+    // to test optimized versions of this function.
+    vp9_short_fdct4x4_c(test_input_block, test_output_block, pitch);
 
     for (int j = 0; j < 16; ++j) {
       if (test_output_block[j] < 0)
@@ -98,13 +52,13 @@
     }
   }
 
-  bool bias_acceptable = true;
-  for (int j = 0; j < 16; ++j)
-    bias_acceptable = bias_acceptable &&
-    (abs(count_sign_block[j][0] - count_sign_block[j][1]) < 10000);
-
-  EXPECT_EQ(true, bias_acceptable)
-    << "Error: 4x4 FDCT has a sign bias > 1% for input range [-255, 255]";
+  for (int j = 0; j < 16; ++j) {
+    const bool bias_acceptable = (abs(count_sign_block[j][0] -
+                                      count_sign_block[j][1]) < 10000);
+    EXPECT_TRUE(bias_acceptable)
+        << "Error: 4x4 FDCT has a sign bias > 1%"
+        << " for input range [-255, 255] at index " << j;
+  }
 
   memset(count_sign_block, 0, sizeof(count_sign_block));
 
@@ -113,7 +67,9 @@
     for (int j = 0; j < 16; ++j)
       test_input_block[j] = (rnd.Rand8() >> 4) - (rnd.Rand8() >> 4);
 
-    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+    // TODO(Yaowu): this should be converted to a parameterized test
+    // to test optimized versions of this function.
+    vp9_short_fdct4x4_c(test_input_block, test_output_block, pitch);
 
     for (int j = 0; j < 16; ++j) {
       if (test_output_block[j] < 0)
@@ -123,16 +79,16 @@
     }
   }
 
-  bias_acceptable = true;
-  for (int j = 0; j < 16; ++j)
-    bias_acceptable = bias_acceptable &&
-    (abs(count_sign_block[j][0] - count_sign_block[j][1]) < 100000);
-
-  EXPECT_EQ(true, bias_acceptable)
-    << "Error: 4x4 FDCT has a sign bias > 10% for input range [-15, 15]";
+  for (int j = 0; j < 16; ++j) {
+    const bool bias_acceptable = (abs(count_sign_block[j][0] -
+                                      count_sign_block[j][1]) < 100000);
+    EXPECT_TRUE(bias_acceptable)
+        << "Error: 4x4 FDCT has a sign bias > 10%"
+        << " for input range [-15, 15] at index " << j;
+  }
 };
 
-TEST(Vp8FdctTest, RoundTripErrorCheck) {
+TEST(Vp9Fdct4x4Test, RoundTripErrorCheck) {
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   int max_error = 0;
   double total_error = 0;
@@ -140,30 +96,49 @@
   for (int i = 0; i < count_test_block; ++i) {
     int16_t test_input_block[16];
     int16_t test_temp_block[16];
-    int16_t test_output_block[16];
-
-    // Initialize a test block with input range [-255, 255].
-    for (int j = 0; j < 16; ++j)
-      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
-
-    const int pitch = 8;
-    vp8_short_fdct4x4_c(test_input_block, test_temp_block, pitch);
-    reference_idct4x4(test_temp_block, test_output_block);
+    uint8_t dst[16], src[16];
 
     for (int j = 0; j < 16; ++j) {
-      const int diff = test_input_block[j] - test_output_block[j];
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 16; ++j)
+      test_input_block[j] = src[j] - dst[j];
+
+    // TODO(Yaowu): this should be converted to a parameterized test
+    // to test optimized versions of this function.
+    const int pitch = 8;
+    vp9_short_fdct4x4_c(test_input_block, test_temp_block, pitch);
+
+    for (int j = 0; j < 16; ++j) {
+        if(test_temp_block[j] > 0) {
+          test_temp_block[j] += 2;
+          test_temp_block[j] /= 4;
+          test_temp_block[j] *= 4;
+        } else {
+          test_temp_block[j] -= 2;
+          test_temp_block[j] /= 4;
+          test_temp_block[j] *= 4;
+        }
+    }
+
+    // Because the bitstream is not frozen yet, use the idct in the codebase.
+    vp9_short_idct4x4_add_c(test_temp_block, dst, 4);
+
+    for (int j = 0; j < 16; ++j) {
+      const int diff = dst[j] - src[j];
       const int error = diff * diff;
       if (max_error < error)
         max_error = error;
       total_error += error;
     }
   }
-
-  EXPECT_GE(1, max_error )
-    << "Error: FDCT/IDCT has an individual roundtrip error > 1";
+  EXPECT_GE(1, max_error)
+      << "Error: FDCT/IDCT has an individual roundtrip error > 1";
 
   EXPECT_GE(count_test_block, total_error)
-    << "Error: FDCT/IDCT has average roundtrip error > 1 per block";
+      << "Error: FDCT/IDCT has average roundtrip error > 1 per block";
 };
 
 }  // namespace
diff --git a/libvpx/test/fdct8x8_test.cc b/libvpx/test/fdct8x8_test.cc
new file mode 100644
index 0000000..90b4ecd
--- /dev/null
+++ b/libvpx/test/fdct8x8_test.cc
@@ -0,0 +1,185 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "vp9_rtcd.h"
+void vp9_short_idct8x8_add_c(short *input, uint8_t *output, int pitch);
+}
+
+#include "acm_random.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+
+TEST(VP9Fdct8x8Test, SignBiasCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int16_t test_input_block[64];
+  int16_t test_output_block[64];
+  const int pitch = 16;
+  int count_sign_block[64][2];
+  const int count_test_block = 100000;
+
+  memset(count_sign_block, 0, sizeof(count_sign_block));
+
+  for (int i = 0; i < count_test_block; ++i) {
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 64; ++j)
+      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+
+    vp9_short_fdct8x8_c(test_input_block, test_output_block, pitch);
+
+    for (int j = 0; j < 64; ++j) {
+      if (test_output_block[j] < 0)
+        ++count_sign_block[j][0];
+      else if (test_output_block[j] > 0)
+        ++count_sign_block[j][1];
+    }
+  }
+
+  for (int j = 0; j < 64; ++j) {
+    const int diff = abs(count_sign_block[j][0] - count_sign_block[j][1]);
+    const int max_diff = 1125;
+    EXPECT_LT(diff, max_diff)
+        << "Error: 8x8 FDCT has a sign bias > "
+        << 1. * max_diff / count_test_block * 100 << "%"
+        << " for input range [-255, 255] at index " << j
+        << " count0: " << count_sign_block[j][0]
+        << " count1: " << count_sign_block[j][1]
+        << " diff: " << diff;
+  }
+
+  memset(count_sign_block, 0, sizeof(count_sign_block));
+
+  for (int i = 0; i < count_test_block; ++i) {
+    // Initialize a test block with input range [-15, 15].
+    for (int j = 0; j < 64; ++j)
+      test_input_block[j] = (rnd.Rand8() >> 4) - (rnd.Rand8() >> 4);
+
+    vp9_short_fdct8x8_c(test_input_block, test_output_block, pitch);
+
+    for (int j = 0; j < 64; ++j) {
+      if (test_output_block[j] < 0)
+        ++count_sign_block[j][0];
+      else if (test_output_block[j] > 0)
+        ++count_sign_block[j][1];
+    }
+  }
+
+  for (int j = 0; j < 64; ++j) {
+    const int diff = abs(count_sign_block[j][0] - count_sign_block[j][1]);
+    const int max_diff = 10000;
+    EXPECT_LT(diff, max_diff)
+        << "Error: 4x4 FDCT has a sign bias > "
+        << 1. * max_diff / count_test_block * 100 << "%"
+        << " for input range [-15, 15] at index " << j
+        << " count0: " << count_sign_block[j][0]
+        << " count1: " << count_sign_block[j][1]
+        << " diff: " << diff;
+  }
+};
+
+TEST(VP9Fdct8x8Test, RoundTripErrorCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int max_error = 0;
+  double total_error = 0;
+  const int count_test_block = 100000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t test_input_block[64];
+    int16_t test_temp_block[64];
+    uint8_t dst[64], src[64];
+
+    for (int j = 0; j < 64; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 64; ++j)
+      test_input_block[j] = src[j] - dst[j];
+
+    const int pitch = 16;
+    vp9_short_fdct8x8_c(test_input_block, test_temp_block, pitch);
+    for (int j = 0; j < 64; ++j){
+        if(test_temp_block[j] > 0) {
+          test_temp_block[j] += 2;
+          test_temp_block[j] /= 4;
+          test_temp_block[j] *= 4;
+        } else {
+          test_temp_block[j] -= 2;
+          test_temp_block[j] /= 4;
+          test_temp_block[j] *= 4;
+        }
+    }
+    vp9_short_idct8x8_add_c(test_temp_block, dst, 8);
+
+    for (int j = 0; j < 64; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+      total_error += error;
+    }
+  }
+
+  EXPECT_GE(1, max_error)
+      << "Error: 8x8 FDCT/IDCT has an individual roundtrip error > 1";
+
+  EXPECT_GE(count_test_block/5, total_error)
+      << "Error: 8x8 FDCT/IDCT has average roundtrip error > 1/5 per block";
+};
+
+TEST(VP9Fdct8x8Test, ExtremalCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int max_error = 0;
+  double total_error = 0;
+  const int count_test_block = 100000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t test_input_block[64];
+    int16_t test_temp_block[64];
+    uint8_t dst[64], src[64];
+
+    for (int j = 0; j < 64; ++j) {
+      src[j] = rnd.Rand8() % 2 ? 255 : 0;
+      dst[j] = src[j] > 0 ? 0 : 255;
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 64; ++j)
+      test_input_block[j] = src[j] - dst[j];
+
+    const int pitch = 16;
+    vp9_short_fdct8x8_c(test_input_block, test_temp_block, pitch);
+    vp9_short_idct8x8_add_c(test_temp_block, dst, 8);
+
+    for (int j = 0; j < 64; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+      total_error += error;
+    }
+
+    EXPECT_GE(1, max_error)
+        << "Error: Extremal 8x8 FDCT/IDCT has an"
+        << " individual roundtrip error > 1";
+
+    EXPECT_GE(count_test_block/5, total_error)
+        << "Error: Extremal 8x8 FDCT/IDCT has average"
+        << " roundtrip error > 1/5 per block";
+  }
+};
+
+}  // namespace
diff --git a/libvpx/test/i420_video_source.h b/libvpx/test/i420_video_source.h
index 219bd33..12a6ab1 100644
--- a/libvpx/test/i420_video_source.h
+++ b/libvpx/test/i420_video_source.h
@@ -83,7 +83,7 @@
   void SetSize(unsigned int width, unsigned int height) {
     if (width != width_ || height != height_) {
       vpx_img_free(img_);
-      img_ = vpx_img_alloc(NULL, VPX_IMG_FMT_VPXI420, width, height, 1);
+      img_ = vpx_img_alloc(NULL, VPX_IMG_FMT_I420, width, height, 1);
       ASSERT_TRUE(img_ != NULL);
       width_ = width;
       height_ = height;
diff --git a/libvpx/test/idct8x8_test.cc b/libvpx/test/idct8x8_test.cc
new file mode 100644
index 0000000..67db78b
--- /dev/null
+++ b/libvpx/test/idct8x8_test.cc
@@ -0,0 +1,140 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "vp9_rtcd.h"
+}
+
+#include "acm_random.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+
+#ifdef _MSC_VER
+static int round(double x) {
+  if(x < 0)
+    return (int)ceil(x - 0.5);
+  else
+    return (int)floor(x + 0.5);
+}
+#endif
+
+void reference_dct_1d(double input[8], double output[8]) {
+  const double kPi = 3.141592653589793238462643383279502884;
+  const double kInvSqrt2 = 0.707106781186547524400844362104;
+  for (int k = 0; k < 8; k++) {
+    output[k] = 0.0;
+    for (int n = 0; n < 8; n++)
+      output[k] += input[n]*cos(kPi*(2*n+1)*k/16.0);
+    if (k == 0)
+      output[k] = output[k]*kInvSqrt2;
+  }
+}
+
+void reference_dct_2d(int16_t input[64], double output[64]) {
+  // First transform columns
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = input[j*8 + i];
+    reference_dct_1d(temp_in, temp_out);
+    for (int j = 0; j < 8; ++j)
+      output[j*8 + i] = temp_out[j];
+  }
+  // Then transform rows
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = output[j + i*8];
+    reference_dct_1d(temp_in, temp_out);
+    for (int j = 0; j < 8; ++j)
+      output[j + i*8] = temp_out[j];
+  }
+  // Scale by some magic number
+  for (int i = 0; i < 64; ++i)
+    output[i] *= 2;
+}
+
+void reference_idct_1d(double input[8], double output[8]) {
+  const double kPi = 3.141592653589793238462643383279502884;
+  const double kSqrt2 = 1.414213562373095048801688724209698;
+  for (int k = 0; k < 8; k++) {
+    output[k] = 0.0;
+    for (int n = 0; n < 8; n++) {
+      output[k] += input[n]*cos(kPi*(2*k+1)*n/16.0);
+      if (n == 0)
+        output[k] = output[k]/kSqrt2;
+    }
+  }
+}
+
+void reference_idct_2d(double input[64], int16_t output[64]) {
+  double out[64], out2[64];
+  // First transform rows
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = input[j + i*8];
+    reference_idct_1d(temp_in, temp_out);
+    for (int j = 0; j < 8; ++j)
+      out[j + i*8] = temp_out[j];
+  }
+  // Then transform columns
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = out[j*8 + i];
+    reference_idct_1d(temp_in, temp_out);
+    for (int j = 0; j < 8; ++j)
+      out2[j*8 + i] = temp_out[j];
+  }
+  for (int i = 0; i < 64; ++i)
+    output[i] = round(out2[i]/32);
+}
+
+TEST(VP9Idct8x8Test, AccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 10000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t input[64], coeff[64];
+    double output_r[64];
+    uint8_t dst[64], src[64];
+
+    for (int j = 0; j < 64; ++j) {
+      src[j] = rnd.Rand8();
+      dst[j] = rnd.Rand8();
+    }
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 64; ++j)
+      input[j] = src[j] - dst[j];
+
+    reference_dct_2d(input, output_r);
+    for (int j = 0; j < 64; ++j)
+      coeff[j] = round(output_r[j]);
+    vp9_short_idct8x8_add_c(coeff, dst, 8);
+    for (int j = 0; j < 64; ++j) {
+      const int diff = dst[j] - src[j];
+      const int error = diff * diff;
+      EXPECT_GE(1, error)
+          << "Error: 8x8 FDCT/IDCT has error " << error
+          << " at index " << j;
+    }
+  }
+}
+
+}  // namespace
diff --git a/libvpx/test/idctllm_test.cc b/libvpx/test/idct_test.cc
similarity index 61%
rename from libvpx/test/idctllm_test.cc
rename to libvpx/test/idct_test.cc
index dd42e22..659cce0 100644
--- a/libvpx/test/idctllm_test.cc
+++ b/libvpx/test/idct_test.cc
@@ -10,27 +10,31 @@
 
 
 extern "C" {
-#include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "./vpx_config.h"
+#include "./vp8_rtcd.h"
 }
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
 typedef void (*idct_fn_t)(short *input, unsigned char *pred_ptr,
                           int pred_stride, unsigned char *dst_ptr,
                           int dst_stride);
 namespace {
-class IDCTTest : public ::testing::TestWithParam<idct_fn_t>
-{
+class IDCTTest : public ::testing::TestWithParam<idct_fn_t> {
   protected:
-    virtual void SetUp()
-    {
+    virtual void SetUp() {
         int i;
 
         UUT = GetParam();
         memset(input, 0, sizeof(input));
         /* Set up guard blocks */
-        for(i=0; i<256; i++)
-            output[i] = ((i&0xF)<4&&(i<64))?0:-1;
+        for (i = 0; i < 256; i++)
+            output[i] = ((i & 0xF) < 4 && (i < 64)) ? 0 : -1;
+    }
+
+    virtual void TearDown() {
+      libvpx_test::ClearSystemState();
     }
 
     idct_fn_t UUT;
@@ -39,78 +43,72 @@
     unsigned char predict[256];
 };
 
-TEST_P(IDCTTest, TestGuardBlocks)
-{
+TEST_P(IDCTTest, TestGuardBlocks) {
     int i;
 
-    for(i=0; i<256; i++)
-        if((i&0xF) < 4 && i<64)
+    for (i = 0; i < 256; i++)
+        if ((i & 0xF) < 4 && i < 64)
             EXPECT_EQ(0, output[i]) << i;
         else
             EXPECT_EQ(255, output[i]);
 }
 
-TEST_P(IDCTTest, TestAllZeros)
-{
+TEST_P(IDCTTest, TestAllZeros) {
     int i;
 
-    UUT(input, output, 16, output, 16);
+    REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
 
-    for(i=0; i<256; i++)
-        if((i&0xF) < 4 && i<64)
+    for (i = 0; i < 256; i++)
+        if ((i & 0xF) < 4 && i < 64)
             EXPECT_EQ(0, output[i]) << "i==" << i;
         else
             EXPECT_EQ(255, output[i]) << "i==" << i;
 }
 
-TEST_P(IDCTTest, TestAllOnes)
-{
+TEST_P(IDCTTest, TestAllOnes) {
     int i;
 
     input[0] = 4;
-    UUT(input, output, 16, output, 16);
+    REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
 
-    for(i=0; i<256; i++)
-        if((i&0xF) < 4 && i<64)
+    for (i = 0; i < 256; i++)
+        if ((i & 0xF) < 4 && i < 64)
             EXPECT_EQ(1, output[i]) << "i==" << i;
         else
             EXPECT_EQ(255, output[i]) << "i==" << i;
 }
 
-TEST_P(IDCTTest, TestAddOne)
-{
+TEST_P(IDCTTest, TestAddOne) {
     int i;
 
-    for(i=0; i<256; i++)
+    for (i = 0; i < 256; i++)
         predict[i] = i;
-
     input[0] = 4;
-    UUT(input, predict, 16, output, 16);
+    REGISTER_STATE_CHECK(UUT(input, predict, 16, output, 16));
 
-    for(i=0; i<256; i++)
-        if((i&0xF) < 4 && i<64)
+    for (i = 0; i < 256; i++)
+        if ((i & 0xF) < 4 && i < 64)
             EXPECT_EQ(i+1, output[i]) << "i==" << i;
         else
             EXPECT_EQ(255, output[i]) << "i==" << i;
 }
 
-TEST_P(IDCTTest, TestWithData)
-{
+TEST_P(IDCTTest, TestWithData) {
     int i;
 
-    for(i=0; i<16; i++)
+    for (i = 0; i < 16; i++)
         input[i] = i;
 
-    UUT(input, output, 16, output, 16);
+    REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
 
-    for(i=0; i<256; i++)
-        if((i&0xF) > 3 || i>63)
+    for (i = 0; i < 256; i++)
+        if ((i & 0xF) > 3 || i > 63)
             EXPECT_EQ(255, output[i]) << "i==" << i;
-        else if(i == 0)
+        else if (i == 0)
             EXPECT_EQ(11, output[i]) << "i==" << i;
-        else if(i == 34)
+        else if (i == 34)
             EXPECT_EQ(1, output[i]) << "i==" << i;
-        else if(i == 2 || i == 17 || i == 32)
+        else if (i == 2 || i == 17 || i == 32)
             EXPECT_EQ(3, output[i]) << "i==" << i;
         else
             EXPECT_EQ(0, output[i]) << "i==" << i;
diff --git a/libvpx/test/intrapred_test.cc b/libvpx/test/intrapred_test.cc
index d2e0d61..39ec896 100644
--- a/libvpx/test/intrapred_test.cc
+++ b/libvpx/test/intrapred_test.cc
@@ -11,10 +11,12 @@
 
 #include <string.h>
 #include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 extern "C" {
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/blockd.h"
 #include "vpx_mem/vpx_mem.h"
 }
@@ -24,6 +26,11 @@
 using libvpx_test::ACMRandom;
 
 class IntraPredBase {
+ public:
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+  }
+
  protected:
   void SetupMacroblock(uint8_t *data, int block_size, int stride,
                        int num_planes) {
@@ -246,8 +253,10 @@
 
   virtual void Predict(MB_PREDICTION_MODE mode) {
     mb_.mode_info_context->mbmi.mode = mode;
-    pred_fn_(&mb_, data_ptr_[0] - kStride, data_ptr_[0] - 1, kStride,
-             data_ptr_[0], kStride);
+    REGISTER_STATE_CHECK(pred_fn_(&mb_,
+                                  data_ptr_[0] - kStride,
+                                  data_ptr_[0] - 1, kStride,
+                                  data_ptr_[0], kStride));
   }
 
   intra_pred_y_fn_t pred_fn_;
diff --git a/libvpx/test/keyframe_test.cc b/libvpx/test/keyframe_test.cc
index d0c81df..85ca0b9 100644
--- a/libvpx/test/keyframe_test.cc
+++ b/libvpx/test/keyframe_test.cc
@@ -9,18 +9,22 @@
  */
 #include <climits>
 #include <vector>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/encode_test_driver.h"
 #include "test/i420_video_source.h"
-#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/util.h"
 
 namespace {
 
 class KeyframeTest : public ::libvpx_test::EncoderTest,
-    public ::testing::TestWithParam<enum libvpx_test::TestMode> {
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
  protected:
+  KeyframeTest() : EncoderTest(GET_PARAM(0)) {}
+
   virtual void SetUp() {
     InitializeConfig();
-    SetMode(GetParam());
+    SetMode(GET_PARAM(1));
     kf_count_ = 0;
     kf_count_max_ = INT_MAX;
     kf_do_force_kf_ = false;
@@ -64,7 +68,7 @@
 
   // In realtime mode - auto placed keyframes are exceedingly rare,  don't
   // bother with this check   if(GetParam() > 0)
-  if(GetParam() > 0)
+  if (GET_PARAM(1) > 0)
     EXPECT_GT(kf_count_, 1);
 }
 
@@ -126,7 +130,7 @@
 
   // In realtime mode - auto placed keyframes are exceedingly rare,  don't
   // bother with this check
-  if(GetParam() > 0)
+  if (GET_PARAM(1) > 0)
     EXPECT_EQ(2u, kf_pts_list_.size()) << " Not the right number of keyframes ";
 
   // Verify that keyframes match the file keyframes in the file.
@@ -141,5 +145,5 @@
   }
 }
 
-INSTANTIATE_TEST_CASE_P(AllModes, KeyframeTest, ALL_TEST_MODES);
+VP8_INSTANTIATE_TEST_CASE(KeyframeTest, ALL_TEST_MODES);
 }  // namespace
diff --git a/libvpx/test/md5_helper.h b/libvpx/test/md5_helper.h
new file mode 100644
index 0000000..fc1a974
--- /dev/null
+++ b/libvpx/test/md5_helper.h
@@ -0,0 +1,64 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef LIBVPX_TEST_MD5_HELPER_H_
+#define LIBVPX_TEST_MD5_HELPER_H_
+
+extern "C" {
+#include "./md5_utils.h"
+#include "vpx/vpx_decoder.h"
+}
+
+namespace libvpx_test {
+class MD5 {
+ public:
+  MD5() {
+    MD5Init(&md5_);
+  }
+
+  void Add(const vpx_image_t *img) {
+    for (int plane = 0; plane < 3; ++plane) {
+      uint8_t *buf = img->planes[plane];
+      const int h = plane ? (img->d_h + 1) >> 1 : img->d_h;
+      const int w = plane ? (img->d_w + 1) >> 1 : img->d_w;
+
+      for (int y = 0; y < h; ++y) {
+        MD5Update(&md5_, buf, w);
+        buf += img->stride[plane];
+      }
+    }
+  }
+
+  const char *Get(void) {
+    static const char hex[16] = {
+      '0', '1', '2', '3', '4', '5', '6', '7',
+      '8', '9', 'a', 'b', 'c', 'd', 'e', 'f',
+    };
+    uint8_t tmp[16];
+    MD5Context ctx_tmp = md5_;
+
+    MD5Final(tmp, &ctx_tmp);
+    for (int i = 0; i < 16; i++) {
+      res_[i * 2 + 0]  = hex[tmp[i] >> 4];
+      res_[i * 2 + 1]  = hex[tmp[i] & 0xf];
+    }
+    res_[32] = 0;
+
+    return res_;
+  }
+
+ protected:
+  char res_[33];
+  MD5Context md5_;
+};
+
+}  // namespace libvpx_test
+
+#endif  // LIBVPX_TEST_MD5_HELPER_H_
diff --git a/libvpx/test/pp_filter_test.cc b/libvpx/test/pp_filter_test.cc
index af2f3bd..79896fe 100644
--- a/libvpx/test/pp_filter_test.cc
+++ b/libvpx/test/pp_filter_test.cc
@@ -7,10 +7,12 @@
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 extern "C" {
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/vpx_integer.h"
 #include "vpx_mem/vpx_mem.h"
 }
@@ -26,7 +28,12 @@
 namespace {
 
 class Vp8PostProcessingFilterTest
-    : public ::testing::TestWithParam<post_proc_func_t> {};
+    : public ::testing::TestWithParam<post_proc_func_t> {
+ public:
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+  }
+};
 
 // Test routine for the VP8 post-processing function
 // vp8_post_proc_down_and_across_mb_row_c.
@@ -74,8 +81,8 @@
   // Initialize pixels in the output to 99.
   (void)vpx_memset(dst_image, 99, output_size);
 
-  GetParam()(src_image_ptr, dst_image_ptr, input_stride,
-             output_stride, block_width, flimits, 16);
+  REGISTER_STATE_CHECK(GetParam()(src_image_ptr, dst_image_ptr, input_stride,
+                                  output_stride, block_width, flimits, 16));
 
   static const uint8_t expected_data[block_height] = {
     4, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 4
diff --git a/libvpx/test/register_state_check.h b/libvpx/test/register_state_check.h
new file mode 100644
index 0000000..fb3f53b
--- /dev/null
+++ b/libvpx/test/register_state_check.h
@@ -0,0 +1,95 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef LIBVPX_TEST_REGISTER_STATE_CHECK_H_
+#define LIBVPX_TEST_REGISTER_STATE_CHECK_H_
+
+#ifdef _WIN64
+
+#define _WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include <winnt.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+namespace testing {
+namespace internal {
+
+inline bool operator==(const M128A& lhs, const M128A& rhs) {
+  return (lhs.Low == rhs.Low && lhs.High == rhs.High);
+}
+
+}  // namespace internal
+}  // namespace testing
+
+namespace libvpx_test {
+
+// Compares the state of xmm[6-15] at construction with their state at
+// destruction. These registers should be preserved by the callee on
+// Windows x64.
+// Usage:
+// {
+//   RegisterStateCheck reg_check;
+//   FunctionToVerify();
+// }
+class RegisterStateCheck {
+ public:
+  RegisterStateCheck() { initialized_ = StoreRegisters(&pre_context_); }
+  ~RegisterStateCheck() { EXPECT_TRUE(Check()); }
+
+ private:
+  static bool StoreRegisters(CONTEXT* const context) {
+    const HANDLE this_thread = GetCurrentThread();
+    EXPECT_TRUE(this_thread != NULL);
+    context->ContextFlags = CONTEXT_FLOATING_POINT;
+    const bool context_saved = GetThreadContext(this_thread, context) == TRUE;
+    EXPECT_TRUE(context_saved) << "GetLastError: " << GetLastError();
+    return context_saved;
+  }
+
+  // Compares the register state. Returns true if the states match.
+  bool Check() const {
+    if (!initialized_) return false;
+    CONTEXT post_context;
+    if (!StoreRegisters(&post_context)) return false;
+
+    const M128A* xmm_pre = &pre_context_.Xmm6;
+    const M128A* xmm_post = &post_context.Xmm6;
+    for (int i = 6; i <= 15; ++i) {
+      EXPECT_EQ(*xmm_pre, *xmm_post) << "xmm" << i << " has been modified!";
+      ++xmm_pre;
+      ++xmm_post;
+    }
+    return !testing::Test::HasNonfatalFailure();
+  }
+
+  bool initialized_;
+  CONTEXT pre_context_;
+};
+
+#define REGISTER_STATE_CHECK(statement) do { \
+  libvpx_test::RegisterStateCheck reg_check; \
+  statement;                               \
+} while (false)
+
+}  // namespace libvpx_test
+
+#else  // !_WIN64
+
+namespace libvpx_test {
+
+class RegisterStateCheck {};
+#define REGISTER_STATE_CHECK(statement) statement
+
+}  // namespace libvpx_test
+
+#endif  // _WIN64
+
+#endif  // LIBVPX_TEST_REGISTER_STATE_CHECK_H_
diff --git a/libvpx/test/resize_test.cc b/libvpx/test/resize_test.cc
index c846157..0d591ad 100644
--- a/libvpx/test/resize_test.cc
+++ b/libvpx/test/resize_test.cc
@@ -9,9 +9,12 @@
  */
 #include <climits>
 #include <vector>
-#include "test/encode_test_driver.h"
-#include "test/video_source.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/video_source.h"
+#include "test/util.h"
 
 namespace {
 
@@ -49,8 +52,10 @@
 };
 
 class ResizeTest : public ::libvpx_test::EncoderTest,
-  public ::testing::TestWithParam<enum libvpx_test::TestMode> {
+  public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
  protected:
+  ResizeTest() : EncoderTest(GET_PARAM(0)) {}
+
   struct FrameInfo {
     FrameInfo(vpx_codec_pts_t _pts, unsigned int _w, unsigned int _h)
         : pts(_pts), w(_w), h(_h) {}
@@ -62,22 +67,16 @@
 
   virtual void SetUp() {
     InitializeConfig();
-    SetMode(GetParam());
+    SetMode(GET_PARAM(1));
   }
 
   virtual bool Continue() const {
     return !HasFatalFailure() && !abort_;
   }
 
-  virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
-    if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
-      const unsigned char *buf =
-          reinterpret_cast<const unsigned char *>(pkt->data.frame.buf);
-      const unsigned int w = (buf[6] | (buf[7] << 8)) & 0x3fff;
-      const unsigned int h = (buf[8] | (buf[9] << 8)) & 0x3fff;
-
-      frame_info_list_.push_back(FrameInfo(pkt->data.frame.pts, w, h));
-    }
+  virtual void DecompressedFrameHook(const vpx_image_t &img,
+                                     vpx_codec_pts_t pts) {
+    frame_info_list_.push_back(FrameInfo(pts, img.d_w, img.d_h));
   }
 
   std::vector< FrameInfo > frame_info_list_;
@@ -100,5 +99,53 @@
   }
 }
 
-INSTANTIATE_TEST_CASE_P(OnePass, ResizeTest, ONE_PASS_TEST_MODES);
+class ResizeInternalTest : public ResizeTest {
+ protected:
+  ResizeInternalTest() : ResizeTest(), frame0_psnr_(0.0) {}
+
+  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
+                                  libvpx_test::Encoder *encoder) {
+    if (video->frame() == 3) {
+      struct vpx_scaling_mode mode = {VP8E_FOURFIVE, VP8E_THREEFIVE};
+      encoder->Control(VP8E_SET_SCALEMODE, &mode);
+    }
+    if (video->frame() == 6) {
+      struct vpx_scaling_mode mode = {VP8E_NORMAL, VP8E_NORMAL};
+      encoder->Control(VP8E_SET_SCALEMODE, &mode);
+    }
+  }
+
+  virtual void PSNRPktHook(const vpx_codec_cx_pkt_t *pkt) {
+    if (!frame0_psnr_)
+      frame0_psnr_ = pkt->data.psnr.psnr[0];
+    EXPECT_NEAR(pkt->data.psnr.psnr[0], frame0_psnr_, 1.0);
+  }
+
+  double frame0_psnr_;
+};
+
+TEST_P(ResizeInternalTest, TestInternalResizeWorks) {
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 10);
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  // q picked such that initial keyframe on this clip is ~30dB PSNR
+  cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = 48;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  for (std::vector<FrameInfo>::iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    const vpx_codec_pts_t pts = info->pts;
+    if (pts >= 3 && pts < 6) {
+      ASSERT_EQ(282U, info->w) << "Frame " << pts << " had unexpected width";
+      ASSERT_EQ(173U, info->h) << "Frame " << pts << " had unexpected height";
+    } else {
+      EXPECT_EQ(352U, info->w) << "Frame " << pts << " had unexpected width";
+      EXPECT_EQ(288U, info->h) << "Frame " << pts << " had unexpected height";
+    }
+  }
+}
+
+VP8_INSTANTIATE_TEST_CASE(ResizeTest, ONE_PASS_TEST_MODES);
+VP9_INSTANTIATE_TEST_CASE(ResizeInternalTest,
+                          ::testing::Values(::libvpx_test::kOnePassBest));
 }  // namespace
diff --git a/libvpx/test/sad_test.cc b/libvpx/test/sad_test.cc
index 2b562e6..1f5435f 100644
--- a/libvpx/test/sad_test.cc
+++ b/libvpx/test/sad_test.cc
@@ -15,12 +15,19 @@
 
 extern "C" {
 #include "./vpx_config.h"
-#include "./vpx_rtcd.h"
-#include "vp8/common/blockd.h"
+#if CONFIG_VP8_ENCODER
+#include "./vp8_rtcd.h"
+//#include "vp8/common/blockd.h"
+#endif
+#if CONFIG_VP9_ENCODER
+#include "./vp9_rtcd.h"
+#endif
 #include "vpx_mem/vpx_mem.h"
 }
 
 #include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 #include "test/util.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
@@ -30,15 +37,26 @@
                                         const unsigned char *reference_ptr,
                                         int reference_stride,
                                         unsigned int max_sad);
+typedef std::tr1::tuple<int, int, sad_m_by_n_fn_t> sad_m_by_n_test_param_t;
+
+typedef void (*sad_n_by_n_by_4_fn_t)(const uint8_t *src_ptr,
+                                     int src_stride,
+                                     const unsigned char * const ref_ptr[],
+                                     int ref_stride,
+                                     unsigned int *sad_array);
+typedef std::tr1::tuple<int, int, sad_n_by_n_by_4_fn_t>
+        sad_n_by_n_by_4_test_param_t;
 
 using libvpx_test::ACMRandom;
 
 namespace {
-class SADTest : public PARAMS(int, int, sad_m_by_n_fn_t) {
+class SADTestBase : public ::testing::Test {
  public:
+  SADTestBase(int width, int height) : width_(width), height_(height) {}
+
   static void SetUpTestCase() {
     source_data_ = reinterpret_cast<uint8_t*>(
-        vpx_memalign(kDataAlignment, kDataBufferSize));
+        vpx_memalign(kDataAlignment, kDataBlockSize));
     reference_data_ = reinterpret_cast<uint8_t*>(
         vpx_memalign(kDataAlignment, kDataBufferSize));
   }
@@ -50,35 +68,36 @@
     reference_data_ = NULL;
   }
 
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+  }
+
  protected:
+  // Handle blocks up to 4 blocks 64x64 with stride up to 128
   static const int kDataAlignment = 16;
-  static const int kDataBufferSize = 16 * 32;
+  static const int kDataBlockSize = 64 * 128;
+  static const int kDataBufferSize = 4 * kDataBlockSize;
 
   virtual void SetUp() {
-    sad_fn_ = GET_PARAM(2);
-    height_ = GET_PARAM(1);
-    width_ = GET_PARAM(0);
-    source_stride_ = width_ * 2;
+    source_stride_ = (width_ + 31) & ~31;
     reference_stride_ = width_ * 2;
     rnd_.Reset(ACMRandom::DeterministicSeed());
   }
 
-  sad_m_by_n_fn_t sad_fn_;
-  virtual unsigned int SAD(unsigned int max_sad) {
-    return sad_fn_(source_data_, source_stride_,
-                   reference_data_, reference_stride_,
-                   max_sad);
+  virtual uint8_t* GetReference(int block_idx) {
+    return reference_data_ + block_idx * kDataBlockSize;
   }
 
   // Sum of Absolute Differences. Given two blocks, calculate the absolute
   // difference between two pixels in the same relative location; accumulate.
-  unsigned int ReferenceSAD(unsigned int max_sad) {
+  unsigned int ReferenceSAD(unsigned int max_sad, int block_idx = 0) {
     unsigned int sad = 0;
+    const uint8_t* const reference = GetReference(block_idx);
 
     for (int h = 0; h < height_; ++h) {
       for (int w = 0; w < width_; ++w) {
         sad += abs(source_data_[h * source_stride_ + w]
-               - reference_data_[h * reference_stride_ + w]);
+               - reference[h * reference_stride_ + w]);
       }
       if (sad > max_sad) {
         break;
@@ -103,6 +122,31 @@
     }
   }
 
+  int width_, height_;
+  static uint8_t* source_data_;
+  int source_stride_;
+  static uint8_t* reference_data_;
+  int reference_stride_;
+
+  ACMRandom rnd_;
+};
+
+class SADTest : public SADTestBase,
+    public ::testing::WithParamInterface<sad_m_by_n_test_param_t> {
+ public:
+  SADTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1)) {}
+
+ protected:
+  unsigned int SAD(unsigned int max_sad, int block_idx = 0) {
+    unsigned int ret;
+    const uint8_t* const reference = GetReference(block_idx);
+
+    REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+                                            reference, reference_stride_,
+                                            max_sad));
+    return ret;
+  }
+
   void CheckSad(unsigned int max_sad) {
     unsigned int reference_sad, exp_sad;
 
@@ -116,19 +160,37 @@
       ASSERT_GE(exp_sad, reference_sad);
     }
   }
-
-  // Handle blocks up to 16x16 with stride up to 32
-  int height_, width_;
-  static uint8_t* source_data_;
-  int source_stride_;
-  static uint8_t* reference_data_;
-  int reference_stride_;
-
-  ACMRandom rnd_;
 };
 
-uint8_t* SADTest::source_data_ = NULL;
-uint8_t* SADTest::reference_data_ = NULL;
+class SADx4Test : public SADTestBase,
+    public ::testing::WithParamInterface<sad_n_by_n_by_4_test_param_t> {
+ public:
+  SADx4Test() : SADTestBase(GET_PARAM(0), GET_PARAM(1)) {}
+
+ protected:
+  void SADs(unsigned int *results) {
+    const uint8_t* refs[] = {GetReference(0), GetReference(1),
+                             GetReference(2), GetReference(3)};
+
+    REGISTER_STATE_CHECK(GET_PARAM(2)(source_data_, source_stride_,
+                                      refs, reference_stride_,
+                                      results));
+  }
+
+  void CheckSADs() {
+    unsigned int reference_sad, exp_sad[4];
+
+    SADs(exp_sad);
+    for (int block = 0; block < 4; block++) {
+      reference_sad = ReferenceSAD(UINT_MAX, block);
+
+      EXPECT_EQ(exp_sad[block], reference_sad) << "block " << block;
+    }
+  }
+};
+
+uint8_t* SADTestBase::source_data_ = NULL;
+uint8_t* SADTestBase::reference_data_ = NULL;
 
 TEST_P(SADTest, MaxRef) {
   FillConstant(source_data_, source_stride_, 0);
@@ -136,12 +198,30 @@
   CheckSad(UINT_MAX);
 }
 
+TEST_P(SADx4Test, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(GetReference(0), reference_stride_, 255);
+  FillConstant(GetReference(1), reference_stride_, 255);
+  FillConstant(GetReference(2), reference_stride_, 255);
+  FillConstant(GetReference(3), reference_stride_, 255);
+  CheckSADs();
+}
+
 TEST_P(SADTest, MaxSrc) {
   FillConstant(source_data_, source_stride_, 255);
   FillConstant(reference_data_, reference_stride_, 0);
   CheckSad(UINT_MAX);
 }
 
+TEST_P(SADx4Test, MaxSrc) {
+  FillConstant(source_data_, source_stride_, 255);
+  FillConstant(GetReference(0), reference_stride_, 0);
+  FillConstant(GetReference(1), reference_stride_, 0);
+  FillConstant(GetReference(2), reference_stride_, 0);
+  FillConstant(GetReference(3), reference_stride_, 0);
+  CheckSADs();
+}
+
 TEST_P(SADTest, ShortRef) {
   int tmp_stride = reference_stride_;
   reference_stride_ >>= 1;
@@ -151,6 +231,18 @@
   reference_stride_ = tmp_stride;
 }
 
+TEST_P(SADx4Test, ShortRef) {
+  int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  reference_stride_ = tmp_stride;
+}
+
 TEST_P(SADTest, UnalignedRef) {
   // The reference frame, but not the source frame, may be unaligned for
   // certain types of searches.
@@ -162,6 +254,20 @@
   reference_stride_ = tmp_stride;
 }
 
+TEST_P(SADx4Test, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  reference_stride_ = tmp_stride;
+}
+
 TEST_P(SADTest, ShortSrc) {
   int tmp_stride = source_stride_;
   source_stride_ >>= 1;
@@ -171,6 +277,18 @@
   source_stride_ = tmp_stride;
 }
 
+TEST_P(SADx4Test, ShortSrc) {
+  int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  source_stride_ = tmp_stride;
+}
+
 TEST_P(SADTest, MaxSAD) {
   // Verify that, when max_sad is set, the implementation does not return a
   // value lower than the reference.
@@ -181,17 +299,75 @@
 
 using std::tr1::make_tuple;
 
+#if CONFIG_VP8_ENCODER
 const sad_m_by_n_fn_t sad_16x16_c = vp8_sad16x16_c;
 const sad_m_by_n_fn_t sad_8x16_c = vp8_sad8x16_c;
 const sad_m_by_n_fn_t sad_16x8_c = vp8_sad16x8_c;
 const sad_m_by_n_fn_t sad_8x8_c = vp8_sad8x8_c;
 const sad_m_by_n_fn_t sad_4x4_c = vp8_sad4x4_c;
-INSTANTIATE_TEST_CASE_P(C, SADTest, ::testing::Values(
-                        make_tuple(16, 16, sad_16x16_c),
-                        make_tuple(8, 16, sad_8x16_c),
-                        make_tuple(16, 8, sad_16x8_c),
-                        make_tuple(8, 8, sad_8x8_c),
-                        make_tuple(4, 4, sad_4x4_c)));
+#endif
+#if CONFIG_VP9_ENCODER
+const sad_m_by_n_fn_t sad_64x64_c_vp9 = vp9_sad64x64_c;
+const sad_m_by_n_fn_t sad_32x32_c_vp9 = vp9_sad32x32_c;
+const sad_m_by_n_fn_t sad_16x16_c_vp9 = vp9_sad16x16_c;
+const sad_m_by_n_fn_t sad_8x16_c_vp9 = vp9_sad8x16_c;
+const sad_m_by_n_fn_t sad_16x8_c_vp9 = vp9_sad16x8_c;
+const sad_m_by_n_fn_t sad_8x8_c_vp9 = vp9_sad8x8_c;
+const sad_m_by_n_fn_t sad_8x4_c_vp9 = vp9_sad8x4_c;
+const sad_m_by_n_fn_t sad_4x8_c_vp9 = vp9_sad4x8_c;
+const sad_m_by_n_fn_t sad_4x4_c_vp9 = vp9_sad4x4_c;
+#endif
+const sad_m_by_n_test_param_t c_tests[] = {
+#if CONFIG_VP8_ENCODER
+  make_tuple(16, 16, sad_16x16_c),
+  make_tuple(8, 16, sad_8x16_c),
+  make_tuple(16, 8, sad_16x8_c),
+  make_tuple(8, 8, sad_8x8_c),
+  make_tuple(4, 4, sad_4x4_c),
+#endif
+#if CONFIG_VP9_ENCODER
+  make_tuple(64, 64, sad_64x64_c_vp9),
+  make_tuple(32, 32, sad_32x32_c_vp9),
+  make_tuple(16, 16, sad_16x16_c_vp9),
+  make_tuple(8, 16, sad_8x16_c_vp9),
+  make_tuple(16, 8, sad_16x8_c_vp9),
+  make_tuple(8, 8, sad_8x8_c_vp9),
+  make_tuple(8, 4, sad_8x4_c_vp9),
+  make_tuple(4, 8, sad_4x8_c_vp9),
+  make_tuple(4, 4, sad_4x4_c_vp9),
+#endif
+};
+INSTANTIATE_TEST_CASE_P(C, SADTest, ::testing::ValuesIn(c_tests));
+
+#if CONFIG_VP9_ENCODER
+const sad_n_by_n_by_4_fn_t sad_64x64x4d_c = vp9_sad64x64x4d_c;
+const sad_n_by_n_by_4_fn_t sad_64x32x4d_c = vp9_sad64x32x4d_c;
+const sad_n_by_n_by_4_fn_t sad_32x64x4d_c = vp9_sad32x64x4d_c;
+const sad_n_by_n_by_4_fn_t sad_32x32x4d_c = vp9_sad32x32x4d_c;
+const sad_n_by_n_by_4_fn_t sad_32x16x4d_c = vp9_sad32x16x4d_c;
+const sad_n_by_n_by_4_fn_t sad_16x32x4d_c = vp9_sad16x32x4d_c;
+const sad_n_by_n_by_4_fn_t sad_16x16x4d_c = vp9_sad16x16x4d_c;
+const sad_n_by_n_by_4_fn_t sad_16x8x4d_c = vp9_sad16x8x4d_c;
+const sad_n_by_n_by_4_fn_t sad_8x16x4d_c = vp9_sad8x16x4d_c;
+const sad_n_by_n_by_4_fn_t sad_8x8x4d_c = vp9_sad8x8x4d_c;
+const sad_n_by_n_by_4_fn_t sad_8x4x4d_c = vp9_sad8x4x4d_c;
+const sad_n_by_n_by_4_fn_t sad_4x8x4d_c = vp9_sad4x8x4d_c;
+const sad_n_by_n_by_4_fn_t sad_4x4x4d_c = vp9_sad4x4x4d_c;
+INSTANTIATE_TEST_CASE_P(C, SADx4Test, ::testing::Values(
+                        make_tuple(64, 64, sad_64x64x4d_c),
+                        make_tuple(64, 32, sad_64x32x4d_c),
+                        make_tuple(32, 64, sad_32x64x4d_c),
+                        make_tuple(32, 32, sad_32x32x4d_c),
+                        make_tuple(32, 16, sad_32x16x4d_c),
+                        make_tuple(16, 32, sad_16x32x4d_c),
+                        make_tuple(16, 16, sad_16x16x4d_c),
+                        make_tuple(16, 8, sad_16x8x4d_c),
+                        make_tuple(8, 16, sad_8x16x4d_c),
+                        make_tuple(8, 8, sad_8x8x4d_c),
+                        make_tuple(8, 4, sad_8x4x4d_c),
+                        make_tuple(4, 8, sad_4x8x4d_c),
+                        make_tuple(4, 4, sad_4x4x4d_c)));
+#endif
 
 // ARM tests
 #if HAVE_MEDIA
@@ -216,31 +392,136 @@
 
 // X86 tests
 #if HAVE_MMX
+#if CONFIG_VP8_ENCODER
 const sad_m_by_n_fn_t sad_16x16_mmx = vp8_sad16x16_mmx;
 const sad_m_by_n_fn_t sad_8x16_mmx = vp8_sad8x16_mmx;
 const sad_m_by_n_fn_t sad_16x8_mmx = vp8_sad16x8_mmx;
 const sad_m_by_n_fn_t sad_8x8_mmx = vp8_sad8x8_mmx;
 const sad_m_by_n_fn_t sad_4x4_mmx = vp8_sad4x4_mmx;
-INSTANTIATE_TEST_CASE_P(MMX, SADTest, ::testing::Values(
-                        make_tuple(16, 16, sad_16x16_mmx),
-                        make_tuple(8, 16, sad_8x16_mmx),
-                        make_tuple(16, 8, sad_16x8_mmx),
-                        make_tuple(8, 8, sad_8x8_mmx),
-                        make_tuple(4, 4, sad_4x4_mmx)));
 #endif
+#if CONFIG_VP9_ENCODER
+const sad_m_by_n_fn_t sad_16x16_mmx_vp9 = vp9_sad16x16_mmx;
+const sad_m_by_n_fn_t sad_8x16_mmx_vp9 = vp9_sad8x16_mmx;
+const sad_m_by_n_fn_t sad_16x8_mmx_vp9 = vp9_sad16x8_mmx;
+const sad_m_by_n_fn_t sad_8x8_mmx_vp9 = vp9_sad8x8_mmx;
+const sad_m_by_n_fn_t sad_4x4_mmx_vp9 = vp9_sad4x4_mmx;
+#endif
+
+const sad_m_by_n_test_param_t mmx_tests[] = {
+#if CONFIG_VP8_ENCODER
+  make_tuple(16, 16, sad_16x16_mmx),
+  make_tuple(8, 16, sad_8x16_mmx),
+  make_tuple(16, 8, sad_16x8_mmx),
+  make_tuple(8, 8, sad_8x8_mmx),
+  make_tuple(4, 4, sad_4x4_mmx),
+#endif
+#if CONFIG_VP9_ENCODER
+  make_tuple(16, 16, sad_16x16_mmx_vp9),
+  make_tuple(8, 16, sad_8x16_mmx_vp9),
+  make_tuple(16, 8, sad_16x8_mmx_vp9),
+  make_tuple(8, 8, sad_8x8_mmx_vp9),
+  make_tuple(4, 4, sad_4x4_mmx_vp9),
+#endif
+};
+INSTANTIATE_TEST_CASE_P(MMX, SADTest, ::testing::ValuesIn(mmx_tests));
+#endif
+
+#if HAVE_SSE
+#if CONFIG_VP9_ENCODER
+const sad_m_by_n_fn_t sad_4x4_sse_vp9 = vp9_sad4x4_sse;
+const sad_m_by_n_fn_t sad_4x8_sse_vp9 = vp9_sad4x8_sse;
+INSTANTIATE_TEST_CASE_P(SSE, SADTest, ::testing::Values(
+                        make_tuple(4, 4, sad_4x4_sse_vp9),
+                        make_tuple(4, 8, sad_4x8_sse_vp9)));
+
+const sad_n_by_n_by_4_fn_t sad_4x8x4d_sse = vp9_sad4x8x4d_sse;
+const sad_n_by_n_by_4_fn_t sad_4x4x4d_sse = vp9_sad4x4x4d_sse;
+INSTANTIATE_TEST_CASE_P(SSE, SADx4Test, ::testing::Values(
+                        make_tuple(4, 8, sad_4x8x4d_sse),
+                        make_tuple(4, 4, sad_4x4x4d_sse)));
+#endif
+#endif
+
 #if HAVE_SSE2
+#if CONFIG_VP8_ENCODER
 const sad_m_by_n_fn_t sad_16x16_wmt = vp8_sad16x16_wmt;
 const sad_m_by_n_fn_t sad_8x16_wmt = vp8_sad8x16_wmt;
 const sad_m_by_n_fn_t sad_16x8_wmt = vp8_sad16x8_wmt;
 const sad_m_by_n_fn_t sad_8x8_wmt = vp8_sad8x8_wmt;
 const sad_m_by_n_fn_t sad_4x4_wmt = vp8_sad4x4_wmt;
-INSTANTIATE_TEST_CASE_P(SSE2, SADTest, ::testing::Values(
-                        make_tuple(16, 16, sad_16x16_wmt),
-                        make_tuple(8, 16, sad_8x16_wmt),
-                        make_tuple(16, 8, sad_16x8_wmt),
-                        make_tuple(8, 8, sad_8x8_wmt),
-                        make_tuple(4, 4, sad_4x4_wmt)));
 #endif
+#if CONFIG_VP9_ENCODER
+const sad_m_by_n_fn_t sad_64x64_sse2_vp9 = vp9_sad64x64_sse2;
+const sad_m_by_n_fn_t sad_32x32_sse2_vp9 = vp9_sad32x32_sse2;
+const sad_m_by_n_fn_t sad_16x16_sse2_vp9 = vp9_sad16x16_sse2;
+const sad_m_by_n_fn_t sad_8x16_sse2_vp9 = vp9_sad8x16_sse2;
+const sad_m_by_n_fn_t sad_16x8_sse2_vp9 = vp9_sad16x8_sse2;
+const sad_m_by_n_fn_t sad_8x8_sse2_vp9 = vp9_sad8x8_sse2;
+const sad_m_by_n_fn_t sad_8x4_sse2_vp9 = vp9_sad8x4_sse2;
+#endif
+const sad_m_by_n_test_param_t sse2_tests[] = {
+#if CONFIG_VP8_ENCODER
+  make_tuple(16, 16, sad_16x16_wmt),
+  make_tuple(8, 16, sad_8x16_wmt),
+  make_tuple(16, 8, sad_16x8_wmt),
+  make_tuple(8, 8, sad_8x8_wmt),
+  make_tuple(4, 4, sad_4x4_wmt),
+#endif
+#if CONFIG_VP9_ENCODER
+  make_tuple(64, 64, sad_64x64_sse2_vp9),
+  make_tuple(32, 32, sad_32x32_sse2_vp9),
+  make_tuple(16, 16, sad_16x16_sse2_vp9),
+  make_tuple(8, 16, sad_8x16_sse2_vp9),
+  make_tuple(16, 8, sad_16x8_sse2_vp9),
+  make_tuple(8, 8, sad_8x8_sse2_vp9),
+  make_tuple(8, 4, sad_8x4_sse2_vp9),
+#endif
+};
+INSTANTIATE_TEST_CASE_P(SSE2, SADTest, ::testing::ValuesIn(sse2_tests));
+
+#if CONFIG_VP9_ENCODER
+const sad_n_by_n_by_4_fn_t sad_64x64x4d_sse2 = vp9_sad64x64x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_64x32x4d_sse2 = vp9_sad64x32x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_32x64x4d_sse2 = vp9_sad32x64x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_32x32x4d_sse2 = vp9_sad32x32x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_32x16x4d_sse2 = vp9_sad32x16x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_16x32x4d_sse2 = vp9_sad16x32x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_16x16x4d_sse2 = vp9_sad16x16x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_16x8x4d_sse2 = vp9_sad16x8x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_8x16x4d_sse2 = vp9_sad8x16x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_8x8x4d_sse2 = vp9_sad8x8x4d_sse2;
+const sad_n_by_n_by_4_fn_t sad_8x4x4d_sse2 = vp9_sad8x4x4d_sse2;
+INSTANTIATE_TEST_CASE_P(SSE2, SADx4Test, ::testing::Values(
+                        make_tuple(64, 64, sad_64x64x4d_sse2),
+                        make_tuple(64, 32, sad_64x32x4d_sse2),
+                        make_tuple(32, 64, sad_32x64x4d_sse2),
+                        make_tuple(32, 32, sad_32x32x4d_sse2),
+                        make_tuple(32, 16, sad_32x16x4d_sse2),
+                        make_tuple(16, 32, sad_16x32x4d_sse2),
+                        make_tuple(16, 16, sad_16x16x4d_sse2),
+                        make_tuple(16, 8, sad_16x8x4d_sse2),
+                        make_tuple(8, 16, sad_8x16x4d_sse2),
+                        make_tuple(8, 8, sad_8x8x4d_sse2),
+                        make_tuple(8, 4, sad_8x4x4d_sse2)));
+#endif
+#endif
+
+#if HAVE_SSE3
+#if CONFIG_VP8_ENCODER
+const sad_n_by_n_by_4_fn_t sad_16x16x4d_sse3 = vp8_sad16x16x4d_sse3;
+const sad_n_by_n_by_4_fn_t sad_16x8x4d_sse3 = vp8_sad16x8x4d_sse3;
+const sad_n_by_n_by_4_fn_t sad_8x16x4d_sse3 = vp8_sad8x16x4d_sse3;
+const sad_n_by_n_by_4_fn_t sad_8x8x4d_sse3 = vp8_sad8x8x4d_sse3;
+const sad_n_by_n_by_4_fn_t sad_4x4x4d_sse3 = vp8_sad4x4x4d_sse3;
+INSTANTIATE_TEST_CASE_P(SSE3, SADx4Test, ::testing::Values(
+                        make_tuple(16, 16, sad_16x16x4d_sse3),
+                        make_tuple(16, 8, sad_16x8x4d_sse3),
+                        make_tuple(8, 16, sad_8x16x4d_sse3),
+                        make_tuple(8, 8, sad_8x8x4d_sse3),
+                        make_tuple(4, 4, sad_4x4x4d_sse3)));
+#endif
+#endif
+
 #if HAVE_SSSE3
 const sad_m_by_n_fn_t sad_16x16_sse3 = vp8_sad16x16_sse3;
 INSTANTIATE_TEST_CASE_P(SSE3, SADTest, ::testing::Values(
diff --git a/libvpx/test/sixtap_predict_test.cc b/libvpx/test/sixtap_predict_test.cc
index 06f14a1..ee4faac 100644
--- a/libvpx/test/sixtap_predict_test.cc
+++ b/libvpx/test/sixtap_predict_test.cc
@@ -12,11 +12,13 @@
 #include <stdlib.h>
 #include <string.h>
 #include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 #include "test/util.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
 extern "C" {
 #include "./vpx_config.h"
-#include "./vpx_rtcd.h"
+#include "./vp8_rtcd.h"
 #include "vpx/vpx_integer.h"
 #include "vpx_mem/vpx_mem.h"
 }
@@ -47,6 +49,10 @@
     dst_c_ = NULL;
   }
 
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+  }
+
  protected:
   // Make test arrays big enough for 16x16 functions. Six-tap filters
   // need 5 extra pixels outside of the macroblock.
@@ -60,9 +66,9 @@
     width_ = GET_PARAM(0);
     height_ = GET_PARAM(1);
     sixtap_predict_ = GET_PARAM(2);
-    memset(src_, 0, sizeof(src_));
-    memset(dst_, 0, sizeof(dst_));
-    memset(dst_c_, 0, sizeof(dst_c_));
+    memset(src_, 0, kSrcSize);
+    memset(dst_, 0, kDstSize);
+    memset(dst_c_, 0, kDstSize);
   }
 
   int width_;
@@ -136,8 +142,8 @@
 
   uint8_t *src = const_cast<uint8_t*>(test_data);
 
-  sixtap_predict_(&src[kSrcStride * 2 + 2 + 1], kSrcStride,
-                  2, 2, dst_, kDstStride);
+  REGISTER_STATE_CHECK(sixtap_predict_(&src[kSrcStride * 2 + 2 + 1], kSrcStride,
+                                       2, 2, dst_, kDstStride));
 
   for (int i = 0; i < height_; ++i)
     for (int j = 0; j < width_; ++j)
@@ -162,8 +168,9 @@
                                 xoffset, yoffset, dst_c_, kDstStride);
 
       // Run test.
-      sixtap_predict_(&src_[kSrcStride * 2 + 2 + 1], kSrcStride,
-                      xoffset, yoffset, dst_, kDstStride);
+      REGISTER_STATE_CHECK(
+          sixtap_predict_(&src_[kSrcStride * 2 + 2 + 1], kSrcStride,
+                          xoffset, yoffset, dst_, kDstStride));
 
       for (int i = 0; i < height_; ++i)
         for (int j = 0; j < width_; ++j)
diff --git a/libvpx/test/subtract_test.cc b/libvpx/test/subtract_test.cc
index 99363de..81bfb66 100644
--- a/libvpx/test/subtract_test.cc
+++ b/libvpx/test/subtract_test.cc
@@ -10,9 +10,11 @@
 
 #include "third_party/googletest/src/include/gtest/gtest.h"
 #include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
 extern "C" {
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/blockd.h"
 #include "vp8/encoder/block.h"
 #include "vpx_mem/vpx_mem.h"
@@ -22,7 +24,12 @@
 
 namespace {
 
-class SubtractBlockTest : public ::testing::TestWithParam<subtract_b_fn_t> {};
+class SubtractBlockTest : public ::testing::TestWithParam<subtract_b_fn_t> {
+ public:
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+  }
+};
 
 using libvpx_test::ACMRandom;
 
@@ -77,7 +84,7 @@
       predictor += kDiffPredStride;
     }
 
-    GetParam()(&be, &bd, kDiffPredStride);
+    REGISTER_STATE_CHECK(GetParam()(&be, &bd, kDiffPredStride));
 
     base_src = *be.base_src;
     src_diff = be.src_diff;
diff --git a/libvpx/test/superframe_test.cc b/libvpx/test/superframe_test.cc
new file mode 100644
index 0000000..062ec6c
--- /dev/null
+++ b/libvpx/test/superframe_test.cc
@@ -0,0 +1,100 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include <climits>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class SuperframeTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
+ protected:
+  SuperframeTest() : EncoderTest(GET_PARAM(0)), modified_buf_(NULL),
+      last_sf_pts_(0) {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    sf_count_ = 0;
+    sf_count_max_ = INT_MAX;
+  }
+
+  virtual void TearDown() {
+    delete[] modified_buf_;
+  }
+
+  virtual bool Continue() const {
+    return !HasFatalFailure() && !abort_;
+  }
+
+  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
+                                  libvpx_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(VP8E_SET_ENABLEAUTOALTREF, 1);
+    }
+  }
+
+  virtual const vpx_codec_cx_pkt_t * MutateEncoderOutputHook(
+      const vpx_codec_cx_pkt_t *pkt) {
+    if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
+      return pkt;
+
+    const uint8_t *buffer = reinterpret_cast<uint8_t*>(pkt->data.frame.buf);
+    const uint8_t marker = buffer[pkt->data.frame.sz - 1];
+    const int frames = (marker & 0x7) + 1;
+    const int mag = ((marker >> 3) & 3) + 1;
+    const unsigned int index_sz = 2 + mag  * frames;
+    if ((marker & 0xe0) == 0xc0 &&
+        pkt->data.frame.sz >= index_sz &&
+        buffer[pkt->data.frame.sz - index_sz] == marker) {
+      // frame is a superframe. strip off the index.
+      if (modified_buf_)
+        delete[] modified_buf_;
+      modified_buf_ = new uint8_t[pkt->data.frame.sz - index_sz];
+      memcpy(modified_buf_, pkt->data.frame.buf,
+             pkt->data.frame.sz - index_sz);
+      modified_pkt_ = *pkt;
+      modified_pkt_.data.frame.buf = modified_buf_;
+      modified_pkt_.data.frame.sz -= index_sz;
+
+      sf_count_++;
+      last_sf_pts_ = pkt->data.frame.pts;
+      return &modified_pkt_;
+    }
+
+    // Make sure we do a few frames after the last SF
+    abort_ |= sf_count_ > sf_count_max_ &&
+              pkt->data.frame.pts - last_sf_pts_ >= 5;
+    return pkt;
+  }
+
+  int sf_count_;
+  int sf_count_max_;
+  vpx_codec_cx_pkt_t modified_pkt_;
+  uint8_t *modified_buf_;
+  vpx_codec_pts_t last_sf_pts_;
+};
+
+TEST_P(SuperframeTest, TestSuperframeIndexIsOptional) {
+  sf_count_max_ = 0;  // early exit on successful test.
+  cfg_.g_lag_in_frames = 25;
+
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 40);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  EXPECT_EQ(sf_count_, 1);
+}
+
+VP9_INSTANTIATE_TEST_CASE(SuperframeTest, ::testing::Values(
+    ::libvpx_test::kTwoPassGood));
+}  // namespace
diff --git a/libvpx/test/test-data.sha1 b/libvpx/test/test-data.sha1
index c1b6a83..1036d7c 100644
--- a/libvpx/test/test-data.sha1
+++ b/libvpx/test/test-data.sha1
@@ -1,4 +1,5 @@
 d5dfb0151c9051f8c85999255645d7a23916d3c0  hantro_collage_w352h288.yuv
+b87815bf86020c592ccc7a846ba2e28ec8043902  hantro_odd.yuv
 5184c46ddca8b1fadd16742e8500115bc8f749da  vp80-00-comprehensive-001.ivf
 65bf1bbbced81b97bd030f376d1b7f61a224793f  vp80-00-comprehensive-002.ivf
 906b4c1e99eb734504c504b3f1ad8052137ce672  vp80-00-comprehensive-003.ivf
@@ -120,4 +121,224 @@
 41d70bb5fa45bc88da1604a0af466930b8dd77b5  vp80-05-sharpness-1438.ivf.md5
 086c56378df81b6cee264d7540a7b8f2b405c7a4  vp80-05-sharpness-1439.ivf.md5
 d32dc2c4165eb266ea4c23c14a45459b363def32  vp80-05-sharpness-1440.ivf.md5
-8c69dc3d8e563f56ffab5ad1e400d9e689dd23df  vp80-05-sharpness-1443.ivf.md5
\ No newline at end of file
+8c69dc3d8e563f56ffab5ad1e400d9e689dd23df  vp80-05-sharpness-1443.ivf.md5
+c5b6fc822d7b4ed97b5a0d69e3a71d9de6cab815  vp90-00-akiyo-100.webm
+1cd8ee73b53f4ecc2511effd233f9af6ecdfac7e  vp90-00-akiyo-100.webm.md5
+a854b0f2313efde7767a4465afbcbe35005ffb07  vp90-00-akiyo-200.webm
+b0f53ad309611246821174b642f6808cc1e670de  vp90-00-akiyo-200.webm.md5
+38a5c0e5465f884474b1a5a9184685f17f961ba1  vp90-00-akiyo-300.webm
+756a34417fc10dc2a49464eccaa6b7f987227b57  vp90-00-akiyo-300.webm.md5
+1047e6f19dd137ae7bbd5b93d407fc7186f8a98e  vp90-00-akiyo-50.webm
+0fa08a76901a6a5b2d4b58a6b20bfa5239409b9d  vp90-00-akiyo-50.webm.md5
+767511b25dde2c5926f5284782a9f1e04fe7afda  vp90-00-bowing-150.webm
+b259c3c6afb30fd1ae7d3a563c1fe9fe6a4644cd  vp90-00-bowing-150.webm.md5
+2ef831c75c021a03176536fb652196e9afc37888  vp90-00-bowing-25.webm
+37d3522cd76b7bab3b5e973e2b2c51edea49ef3f  vp90-00-bowing-25.webm.md5
+c1e4639f14914516ca704f38c875d01f4c06be14  vp90-00-bowing-400.webm
+ca35c574512185d5f20f3b81517d6ac3333a1377  vp90-00-bowing-400.webm.md5
+e20fc293db095e52f29b891bc09458e7568e8603  vp90-00-bus-100.webm
+a754ea588cc409546936c09fb1ad06b3014b94f9  vp90-00-bus-100.webm.md5
+da5eb45fa42f55ff70ec7b71999e6fd8489d12f9  vp90-00-bus-2000.webm
+2a7356328eb991175cbddebd51a30018e48632f2  vp90-00-bus-2000.webm.md5
+607169c774664176aca7c7d46dabf04b9c3634e4  vp90-00-bus-300.webm
+c84daa3a0290d73226b243dd630820ac97bf4fbd  vp90-00-bus-300.webm.md5
+655902b54b9a8a882c11bc8bce1447f3b2085035  vp90-00-bus-4400.webm
+f719ecd7b53c8e35fae735396629d1915ffc1ff9  vp90-00-bus-4400.webm.md5
+afcdca9763d233dd63fd67165a7b92ea679822af  vp90-00-bus-800.webm
+66e2a55560e570cae09520060f1ae315c7ea0a07  vp90-00-bus-800.webm.md5
+390b91c8566d94c3a869af77531585c38f9f78da  vp90-00-cheer-1600.webm
+3d47da26375a75afef0cf2123f5c808d0862e25d  vp90-00-cheer-1600.webm.md5
+23419784db17a50e129e3bd030c20256cf0d6eb0  vp90-00-cheer-2800.webm
+0df4676171f19e7807d719a9b8a6fadcefc8f1fc  vp90-00-cheer-2800.webm.md5
+45ed3c42874d5ec88852798691cf54bfb0cf652a  vp90-00-cheer-400.webm
+374fd67ac9ae0e8146051b77963459c54b9eaaa2  vp90-00-cheer-400.webm.md5
+1c9459d824116a297ff0e90bed9be783005f9ac1  vp90-00-cheer-600.webm
+9dc0d43f72c8eb49d51a9748fb9948495529a6b5  vp90-00-cheer-600.webm.md5
+a86c5af1929d2f929a5caf6ef847d0066086223b  vp90-00-city-1200.webm
+231c7f0f406e3a8d2328daee4c4466e1b4d47354  vp90-00-city-1200.webm.md5
+be9cf927e6ab517d7876925d21b3193b1373d03d  vp90-00-city-2000.webm
+487d60226a3a3039528a049e9c6e8243b07404e6  vp90-00-city-2000.webm.md5
+1f3cd649d5829d52c08da3323baa86b1dcf2d2de  vp90-00-city-300.webm
+8e3b38cfa2be757e46ea12cff11762cb50134615  vp90-00-city-300.webm.md5
+286f6ea64c33ce735b5b7806aac4ca5ee331af66  vp90-00-city-600.webm
+7c51ead147ef4029094a2b455239090c1999d8fe  vp90-00-city-600.webm.md5
+f7ecbd63bed06ed15afe0ba2a192f2cf7943714c  vp90-00-coastguard-1200.webm
+8c8fed2c64cc8fb330e9200e1e0f58a79b953b79  vp90-00-coastguard-1200.webm.md5
+2e63178e5b2c2cc84226df2b514c4dde46c32d70  vp90-00-coastguard-200.webm
+128f2b22fdcfd02bc50e63b1cd6d40c0cc4998d6  vp90-00-coastguard-200.webm.md5
+97b779617d3c1ca8f50beda7126be5df913d071d  vp90-00-coastguard-3600.webm
+0da0ab4794439e6b8ab9ced41239e1307686be69  vp90-00-coastguard-3600.webm.md5
+5e060d66573a40f7f0a46ae9b6acb51b0afb2e3c  vp90-00-coastguard-5200.webm
+4ba526d4bb895c4794dc20edeb38b102a9b1bd92  vp90-00-coastguard-5200.webm.md5
+17810fa737f29d5b032836e38243bbb666f06636  vp90-00-container-1000.webm
+7e0fd7e93c5a16394818f844aa5f2d5fa7a73ee2  vp90-00-container-1000.webm.md5
+38deb4f59cec9e62715dec2f3670ffe7b1cf493e  vp90-00-container-200.webm
+aa3229017f920750bd5d919e19ea6127ea05adc0  vp90-00-container-200.webm.md5
+8b1a67ef35d3f00981d23c41b56a0a2e09976312  vp90-00-container-50.webm
+0a6f1a793b936ff1287326882f1165065a2dcea0  vp90-00-container-50.webm.md5
+4c724db691b7202b60b56107ec7b0abc6cc52bdc  vp90-00-deadline-1000.webm
+5903bd89be457be681a6c6c8fd8c19f4570173db  vp90-00-deadline-1000.webm.md5
+ee5e19a8fe14d3e72b1314a012b49a3bc0586375  vp90-00-deadline-200.webm
+77095f98406fa27a2da8661f21664c00292dcefc  vp90-00-deadline-200.webm.md5
+8230b07aa0ee7adf3caabae4e3bef997929001eb  vp90-00-deadline-50.webm
+fc47a159b2d2b0bed93d4e2c35408243e70b6d24  vp90-00-deadline-50.webm.md5
+244d12cda51235dcc421fedbe12422b326f539e7  vp90-00-flower-100.webm
+dfeca236450b5ff19c1558ad33fba7ab7ff75f27  vp90-00-flower-100.webm.md5
+d5b7057564f670f7bf82017e2abc3aed5656b810  vp90-00-flower-2000.webm
+65118811f4d46ef1e911d520296731536d3a507e  vp90-00-flower-2000.webm.md5
+a9c226643365f0c8ae03e780d55aa6c6fa9cc0e7  vp90-00-flower-300.webm
+fa5193d1a6e6b9e8bb91f75e91a3a377f00fa42e  vp90-00-flower-300.webm.md5
+b206284b51dec6219c46e9b03def38a94d91bf89  vp90-00-flower-4400.webm
+c8a73acd8234b287e86465d03fbf4f886d1fefb2  vp90-00-flower-4400.webm.md5
+faff83d7b6aa89f5d9518ffc5d4b145eb02b6800  vp90-00-flower-800.webm
+328dd1969804afc094d010f54f350bd05390d6a9  vp90-00-flower-800.webm.md5
+42caa40d3b76b8ae5e7573b95e09bc4e57bea835  vp90-00-football-1600.webm
+167b8f58a85d83050d4c56391d6b2d9a9a205b9a  vp90-00-football-1600.webm.md5
+4c4f93f594a8ef89a9ba903bbcff914022a5ad9d  vp90-00-football-2800.webm
+7995f7f91b13d4ab5badcd3f9282bd1fceba38f3  vp90-00-football-2800.webm.md5
+c3ff724e79b4ae0202929f3ed1a1a5b67d10901f  vp90-00-football-400.webm
+19164a0e58ca5d407282a867866e8ec4a0a08fea  vp90-00-football-400.webm.md5
+95de1c4abceab3706f0225e3b9c5dc719901a6cf  vp90-00-football-600.webm
+4a4454ae4d65748a45eaa3decb783bbe0ba190dc  vp90-00-football-600.webm.md5
+80eebcdae76459c00d14b6c50f7529377e53a1c2  vp90-00-foreman-1200.webm
+8228cc5a7cc83970b3a65f9b49bc74733255b09c  vp90-00-foreman-1200.webm.md5
+601d0ff4f058a3da3af4409e4117795f7c231fda  vp90-00-foreman-2000.webm
+e0c0b0aa6f9597984a2d78e799a00e0052710b2c  vp90-00-foreman-2000.webm.md5
+30ebc327645d68bcc83eab72610bba22f877fb4c  vp90-00-foreman-300.webm
+080fc2adf29a84f02a3e4b5508fc2f8dc32f1440  vp90-00-foreman-300.webm.md5
+6b1a6be0f7bd7605b565750b3080be397d4c48a0  vp90-00-foreman-600.webm
+f7713d3eba8d34d511ba1c9585a5a3f34e133ba5  vp90-00-foreman-600.webm.md5
+b080d9786abc89b4be59bffc5baba7b42fbc286a  vp90-00-hallmonitor-1200.webm
+77be47800b58001eb7a854d4d4a9b9823bbbe158  vp90-00-hallmonitor-1200.webm.md5
+05cd8e8d58ab8311ad528c27b4c89cdf268e749b  vp90-00-hallmonitor-2000.webm
+de1aa35c7172e78e07d6b197280214bbd362cc4e  vp90-00-hallmonitor-2000.webm.md5
+908676b32b190e956518bb742d1415efceeb8c75  vp90-00-hallmonitor-300.webm
+f9d39866db341d18256339e9fd2c0ec296f47702  vp90-00-hallmonitor-300.webm.md5
+1307c7f7558de34a6230912e684ff9571a05db5f  vp90-00-hallmonitor-600.webm
+954b292dd56be5c1bf153df440b132e1b1fbcb68  vp90-00-hallmonitor-600.webm.md5
+05f556288c5c4211420f7c332daded816f9b31b7  vp90-00-harbour-1200.webm
+399481f93cc252f20ad5141dd402cf5363673578  vp90-00-harbour-1200.webm.md5
+fa62e449485c544c281030c5ccff32c60d4dd169  vp90-00-harbour-200.webm
+3d0e1885befb2493c477384917797164d4fe58e4  vp90-00-harbour-200.webm.md5
+fa3a5e563c3d2215703c1a68f71fbe2168a42468  vp90-00-harbour-3600.webm
+9af392f6b2cb5ec5c9446b7262206773df535319  vp90-00-harbour-3600.webm.md5
+476db4b15989a5a078f1d2fc5f9734d1d24f1da1  vp90-00-harbour-5200.webm
+352a05b179dc1f86cf6ce27494a4a8fb42379d72  vp90-00-harbour-5200.webm.md5
+0ea17a4892383a2fd0be9f88f213f5f48f2a61f4  vp90-00-highway-100.webm
+a2fe942955bafa83295d1381c9a25264764924c5  vp90-00-highway-100.webm.md5
+7ab80485670a5343a74c4a2454761ed3bed7ceef  vp90-00-highway-1600.webm
+fda9c82cb5d28a5ff5f7dae7c537e9187dfbd4cc  vp90-00-highway-1600.webm.md5
+162d42e033dad04fd7ae3bf9d39e9e204c022edc  vp90-00-highway-2800.webm
+b882c93a2dc89feb6090b0f72e67ac8a59fc0986  vp90-00-highway-2800.webm.md5
+79b9a0e6fa6cdd2367228e9ac8d6a369a8d647e6  vp90-00-highway-50.webm
+80ecf926372dbe8c1b4bcd68ea2101f78a93b02e  vp90-00-highway-50.webm.md5
+a67fd02cbb75c1a757b5ea56b9eee46069bfadbf  vp90-00-husky-100.webm
+12cd583e791c8e5b40b5dffe4a9dbcc1929dc645  vp90-00-husky-100.webm.md5
+1a8b4302eb6f88b14a9acd4a6cbe62d0b380f2e4  vp90-00-husky-2000.webm
+a9c2532e5d867d7627bb6767008b43b653cce904  vp90-00-husky-2000.webm.md5
+f56f66afd4d4512a49904275a1c942ba7379fec4  vp90-00-husky-300.webm
+196dc386f104b7b9ed2ec6c6a1f104ce0319c2eb  vp90-00-husky-300.webm.md5
+6ba3c16fd98d37a8de7023419682a3595778b9bc  vp90-00-husky-4400.webm
+2f4815ba97e352fcd0089d1a5883a0aff1e5394a  vp90-00-husky-4400.webm.md5
+db04a296c377693dd6e974bea36256f4b14cddef  vp90-00-husky-800.webm
+7658473ad17ee689a37fda558c5a23816131cfc3  vp90-00-husky-800.webm.md5
+50cf9e34b61e1cf32c9dde2ebcc5f5703c379a41  vp90-00-ice-150.webm
+806ceba91dc40c45eafc4d7ee61df9346c6fe5f9  vp90-00-ice-150.webm.md5
+4cfca1bea7aae6e4405abfca603cfbded13ded1a  vp90-00-ice-400.webm
+e4298abf05419973da89c0bfcdf0006b1606ebcd  vp90-00-ice-400.webm.md5
+12e3ccfdf96c3f4eebeed8106c5daef6c2b28d83  vp90-00-ice-800.webm
+6fb2aacb4d8131dcabaa61a9cd2497cd09854377  vp90-00-ice-800.webm.md5
+124977938c47ba739e918533bc5d6d73e41ce2ec  vp90-00-mobile-1600.webm
+603b2b523c8ed5922121d285567a845bb6693d35  vp90-00-mobile-1600.webm.md5
+93f204b90250791b884479be5da534a5bc6304ff  vp90-00-mobile-2800.webm
+21ec8735b774c66e192f7270c12075f598f700d5  vp90-00-mobile-2800.webm.md5
+fe9cdbfdeee2b7554efb532f646703cff55c2d2c  vp90-00-mobile-400.webm
+4def63c78ee09e90e6385d3122ada95343246102  vp90-00-mobile-400.webm.md5
+2a042aa8a06c45770dcb52c56a7f5cea6d51b8dd  vp90-00-mobile-600.webm
+03169f031dece0db3d89ce16cc3e0ee3eca21065  vp90-00-mobile-600.webm.md5
+7fc5b0b0c684d63e161c9c5932e1374327e15dd4  vp90-00-motherdaughter-100.webm
+290ac7722caf4b15136b307a239c9b903113b9c4  vp90-00-motherdaughter-100.webm.md5
+67ddfce82bff083a1ceb108a7dcfb801791102f1  vp90-00-motherdaughter-300.webm
+7696698d38e32f0afeb3a3e9a45b7fe3f237aaba  vp90-00-motherdaughter-300.webm.md5
+ff65a1bee2fe384728017c5148df61379043d5b6  vp90-00-motherdaughter-600.webm
+f0b167000bf40877d1ba7ba52a08b4310011c032  vp90-00-motherdaughter-600.webm.md5
+d73c54e676bd63424fc9ad8d0cef64e929081cf4  vp90-00-news-100.webm
+71821b71a97823e9ba58563efc841dc6beefe9df  vp90-00-news-100.webm.md5
+2937238d094863951eb8f218438b966d2b7b5430  vp90-00-news-300.webm
+2587d0859a330cf6d8e0a135d1f586bb2a5033fc  vp90-00-news-300.webm.md5
+65afdd4fc411951115b48435b8b65155594b5c99  vp90-00-news-600.webm
+5815bb341db976f44dab97bb9cfba8ea0ca55502  vp90-00-news-600.webm.md5
+de5dd99ac04d3a937fc0951d06fb8f533fdc393a  vp90-00-pamphlet-150.webm
+0381d705fa490f35c772e3048b423b382088d546  vp90-00-pamphlet-150.webm.md5
+46f283284cb64b79243b2ea6aad709a526c26393  vp90-00-pamphlet-25.webm
+f100fbebcad96f27ed8f340414b939bc738d49d0  vp90-00-pamphlet-25.webm.md5
+8df04ece12455c5c40f14cb089348260798c5f2b  vp90-00-pamphlet-400.webm
+66a2c87cd4194368d3477e9a334880b76c87e991  vp90-00-pamphlet-400.webm.md5
+a00e97e4a71f5e24f194c59cde7d41bc2c3af325  vp90-00-paris-1000.webm
+53ef896e16d1b83aa5166945d149c7133401b3f0  vp90-00-paris-1000.webm.md5
+6b03388e0236f6171e20c73834858e3c87b441b2  vp90-00-paris-200.webm
+55a324b0153c5d54cd0c0492fed8755c441fa18c  vp90-00-paris-200.webm.md5
+429ec362a9600c8822652cf7e122e22bca033d69  vp90-00-paris-50.webm
+4406226b7bddb11ede8ee0c442d52e5d3bbbde78  vp90-00-paris-50.webm.md5
+a7996d4e757ea484aa72e14f623d6c9e72537888  vp90-00-signirene-1000.webm
+f65a1ac6e1ce77102e63fb363dbca361b8108c02  vp90-00-signirene-1000.webm.md5
+8c2f686179bc3e87a18b48bcb5058f3cd61e1b4c  vp90-00-signirene-200.webm
+b8ab16cba9392e49169c374eb1e0c1b763ccaefb  vp90-00-signirene-200.webm.md5
+5f8f99c386dce64931bbd4fc42a59a78dc6fdba1  vp90-00-signirene-50.webm
+fdb8c4bc302884d413a256634d3e2fbd92867c90  vp90-00-signirene-50.webm.md5
+d5074f0a5bcefe9fd651afbbebf0e0f3fedb965b  vp90-00-silent-1000.webm
+9c075894fbfb84791fcc7dbd3fcab15b0a9bd64e  vp90-00-silent-1000.webm.md5
+32101f334f675715a8f411638dfda80afacc37a6  vp90-00-silent-200.webm
+fb0dac37f31ca711443832046a6aaf868e69b357  vp90-00-silent-200.webm.md5
+0aaef50d7f94873e99ec7e39f59a6b74e92ad946  vp90-00-silent-50.webm
+be9fc41965b5b63f7c7bbd6c91191e940903e012  vp90-00-silent-50.webm.md5
+5e22ad14c562733d4d4a3ce163b580ed4a64e6fe  vp90-00-soccer-100.webm
+1ca9a0016910cfca26def9944568749a168131d8  vp90-00-soccer-100.webm.md5
+2d9b2a0fa5ac210f8d7c646578698e045733ad4a  vp90-00-soccer-2000.webm
+f979078650057606ca770b3f03be4c509efb40a9  vp90-00-soccer-2000.webm.md5
+7b789360ffc1eb5a3735f8a1f8d248a24ca4267c  vp90-00-soccer-300.webm
+195d33b23ca8304519bd6e38e9657e53a04779d8  vp90-00-soccer-300.webm.md5
+3907318ef35573e4efc5c150d3aff271c7157501  vp90-00-soccer-4400.webm
+4b43ceecae9a9a7d39a47347f9e20af3613827d1  vp90-00-soccer-4400.webm.md5
+c89920aa89194cb6a36f77dff8722573f0df7241  vp90-00-soccer-800.webm
+1da71751009afa483a03e274a538df24c9f5e513  vp90-00-soccer-800.webm.md5
+efca14e8e0515a8f8ed3ded11fdbff24b09a7f9d  vp90-00-stefan-1600.webm
+6f103270ce03cc85b28dd1c86d0447922d810671  vp90-00-stefan-1600.webm.md5
+b99ab6a983d48c15aa3a9160d06286fca0074193  vp90-00-stefan-2800.webm
+986a72dd9988c6bf4246cd5bd966ce991ba55319  vp90-00-stefan-2800.webm.md5
+eb962244ca51a101ad8f585df6be8f5f96691f18  vp90-00-stefan-400.webm
+2747cfd8f74aedc370767f08129b35ace70e1fe7  vp90-00-stefan-400.webm.md5
+b507b8cedd0147c5316db8f84f35ace768c25069  vp90-00-stefan-600.webm
+daeb369046c2dc27ecfde978b87fd8b49d83789f  vp90-00-stefan-600.webm.md5
+c5c2dd891c2b5fe4a70845858ccb859df3455ee7  vp90-00-students-100.webm
+d1be06dc636ece0c34ab8c17399888aaf19e0c19  vp90-00-students-100.webm.md5
+c9e4da3a8b455aa690d89338f32f9d76773cdd18  vp90-00-students-300.webm
+a9aa72e1ee27063f8e9f13b4647cec01c8efb2d6  vp90-00-students-300.webm.md5
+e9e5072cd944a8994e50fce367975e3ce526bd67  vp90-00-students-600.webm
+86525ce188a98a51f86fad27341729bb61d1ca8b  vp90-00-students-600.webm.md5
+58deb053aeafefdfdf13741accf9fcbe4584ea94  vp90-00-tempete-1200.webm
+ec395a2ec76b4c1e64e243366a8840da22ee3a65  vp90-00-tempete-1200.webm.md5
+5d35232eaa8ee149a917ff94536968fb37dad50e  vp90-00-tempete-200.webm
+7f8c7529f40d6b6d6de8e89dbf9697623d27c234  vp90-00-tempete-200.webm.md5
+c44eb147bc3f8682b96096fccef8beb4380c40db  vp90-00-tempete-3600.webm
+01fd23e412530fa2d5319a22886161957a747ee0  vp90-00-tempete-3600.webm.md5
+56ab322b34a750e16dcc8ccfb735a5b9270cedc4  vp90-00-tempete-5200.webm
+1cf803409ae53b991bff10079af4ab07aaa2853d  vp90-00-tempete-5200.webm.md5
+ffe48d52019c228e919f4b123028664b8d0c2f4b  vp90-00-tennis-100.webm
+406fda3367899995d4e37170063495832e2be372  vp90-00-tennis-100.webm.md5
+6c030f8142b1932fbe8eb5c2b39b3452a5eea3aa  vp90-00-tennis-2000.webm
+dcf20921e2a8ab0dcd09f7f6bdcdd35f979205ae  vp90-00-tennis-2000.webm.md5
+3fe0df7b74f301b39e1b21e6926c69a8418b9b70  vp90-00-tennis-300.webm
+80c8301d3a37b33ca50318ba000066a6ae9929dc  vp90-00-tennis-300.webm.md5
+82a2497083b8dce6b1c73bcdf16323ea69d1cca9  vp90-00-tennis-4400.webm
+83ce97bc09a7e1b2f2c3437195a8931d7608a62b  vp90-00-tennis-4400.webm.md5
+2c8bd3a29bbd1085169bfcba9fdf65a37f4a16bb  vp90-00-tennis-800.webm
+9920a65e06d2e7025f13f3d8bf35670503875aed  vp90-00-tennis-800.webm.md5
+26469062c5724c2cc4914436ef032bb55373f843  vp90-00-waterfall-150.webm
+9b86373ce15302a9b22cef8f808ce0e37e6d2b65  vp90-00-waterfall-150.webm.md5
+410ba6af2ddca5110fa7a4c383dc8b28f38cf565  vp90-00-waterfall-200.webm
+251892d3fdcbc9d7a20c22ba202ed4935222e5b8  vp90-00-waterfall-200.webm.md5
+40b643aff88aed3764c5b58c446a8fbbc5fb36d7  vp90-00-waterfall-400.webm
+51f31a6b6408f8af4d107e0f2a3c1a274d4da6bb  vp90-00-waterfall-400.webm.md5
+bd421141e01f53dc15ced790f9a96ab70a613260  vp90-00-waterfall-800.webm
+1366efe772fccaa2b8a6ac3ce45255b312a2ef6c  vp90-00-waterfall-800.webm.md5
diff --git a/libvpx/test/test.mk b/libvpx/test/test.mk
index 7a11a27..806901d 100644
--- a/libvpx/test/test.mk
+++ b/libvpx/test/test.mk
@@ -1,5 +1,9 @@
-LIBVPX_TEST_SRCS-yes += acm_random.h
+LIBVPX_TEST_SRCS-yes += clear_system_state.h
+LIBVPX_TEST_SRCS-yes += register_state_check.h
 LIBVPX_TEST_SRCS-yes += test.mk
+LIBVPX_TEST_SRCS-yes += acm_random.h
+LIBVPX_TEST_SRCS-yes += md5_helper.h
+LIBVPX_TEST_SRCS-yes += codec_factory.h
 LIBVPX_TEST_SRCS-yes += test_libvpx.cc
 LIBVPX_TEST_SRCS-yes += util.h
 LIBVPX_TEST_SRCS-yes += video_source.h
@@ -13,18 +17,32 @@
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += config_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += cq_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += datarate_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += encode_test_driver.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += encode_test_driver.h
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += error_resilience_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += i420_video_source.h
+
+LIBVPX_TEST_SRCS-yes                   += encode_test_driver.cc
+LIBVPX_TEST_SRCS-yes                   += encode_test_driver.h
+LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS)    += error_resilience_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS)    += i420_video_source.h
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += keyframe_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += borders_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += resize_test.cc
 
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += ../md5_utils.h ../md5_utils.c
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += decode_test_driver.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += decode_test_driver.h
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += ivf_video_source.h
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += test_vector_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += ../md5_utils.h ../md5_utils.c
+LIBVPX_TEST_SRCS-yes                   += decode_test_driver.cc
+LIBVPX_TEST_SRCS-yes                   += decode_test_driver.h
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += ivf_video_source.h
+
+## WebM Parsing
+NESTEGG_SRCS                           += ../nestegg/halloc/halloc.h
+NESTEGG_SRCS                           += ../nestegg/halloc/src/align.h
+NESTEGG_SRCS                           += ../nestegg/halloc/src/halloc.c
+NESTEGG_SRCS                           += ../nestegg/halloc/src/hlist.h
+NESTEGG_SRCS                           += ../nestegg/include/nestegg/nestegg.h
+NESTEGG_SRCS                           += ../nestegg/src/nestegg.c
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += $(NESTEGG_SRCS)
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += webm_video_source.h
+
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += test_vector_test.cc
+
 ##
 ## WHITE BOX TESTS
 ##
@@ -33,19 +51,50 @@
 ##
 ifeq ($(CONFIG_SHARED),)
 
+## VP8
+ifneq ($(CONFIG_VP8_ENCODER)$(CONFIG_VP8_DECODER),)
+
 # These tests require both the encoder and decoder to be built.
 ifeq ($(CONFIG_VP8_ENCODER)$(CONFIG_VP8_DECODER),yesyes)
-LIBVPX_TEST_SRCS-yes                   += boolcoder_test.cc
+LIBVPX_TEST_SRCS-yes                   += vp8_boolcoder_test.cc
 endif
 
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += fdct4x4_test.cc
-LIBVPX_TEST_SRCS-yes                   += idctllm_test.cc
+LIBVPX_TEST_SRCS-yes                   += idct_test.cc
 LIBVPX_TEST_SRCS-yes                   += intrapred_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_POSTPROC)    += pp_filter_test.cc
-LIBVPX_TEST_SRCS-yes                   += sad_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS)    += sad_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += set_roi.cc
 LIBVPX_TEST_SRCS-yes                   += sixtap_predict_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += subtract_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += variance_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += vp8_decrypt_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += vp8_fdct4x4_test.cc
+
+endif # VP8
+
+## VP9
+ifneq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_DECODER),)
+
+# These tests require both the encoder and decoder to be built.
+ifeq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_DECODER),yesyes)
+LIBVPX_TEST_SRCS-yes                   += vp9_boolcoder_test.cc
+
+# IDCT test currently depends on FDCT function
+LIBVPX_TEST_SRCS-yes                   += idct8x8_test.cc
+LIBVPX_TEST_SRCS-yes                   += superframe_test.cc
+LIBVPX_TEST_SRCS-yes                   += tile_independence_test.cc
+endif
+
+LIBVPX_TEST_SRCS-$(CONFIG_VP9)         += convolve_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
+
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct16x16_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct32x32_test.cc
+
+endif # VP9
+
 
 endif
 
@@ -53,7 +102,9 @@
 ##
 ## TEST DATA
 ##
-LIBVPX_TEST_DATA-$(CONFIG_VP8_ENCODER) += hantro_collage_w352h288.yuv
+LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += hantro_collage_w352h288.yuv
+LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += hantro_odd.yuv
+
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-001.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-002.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-003.ivf
@@ -176,3 +227,223 @@
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1439.ivf.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1440.ivf.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1443.ivf.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-akiyo-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-150.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-150.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-25.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-25.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bowing-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-4400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-4400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-bus-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-1600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-1600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-2800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-2800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-cheer-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-city-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-3600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-3600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-5200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-coastguard-5200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-1000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-1000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-container-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-1000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-1000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-deadline-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-4400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-4400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-flower-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-1600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-1600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-2800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-2800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-football-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-foreman-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-hallmonitor-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-3600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-3600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-5200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-harbour-5200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-1600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-1600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-2800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-2800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-highway-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-4400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-4400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-husky-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-150.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-150.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-ice-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-1600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-1600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-2800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-2800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-mobile-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-motherdaughter-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-news-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-150.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-150.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-25.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-25.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-pamphlet-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-1000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-1000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-paris-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-1000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-1000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-signirene-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-1000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-1000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-50.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-silent-50.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-4400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-4400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-soccer-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-1600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-1600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-2800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-2800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-stefan-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-students-600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-1200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-1200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-3600.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-3600.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-5200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tempete-5200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-100.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-100.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-2000.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-2000.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-300.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-300.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-4400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-4400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-tennis-800.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-150.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-150.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-200.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-200.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-400.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-400.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-800.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-00-waterfall-800.webm.md5
diff --git a/libvpx/test/test_libvpx.cc b/libvpx/test/test_libvpx.cc
index cfd5d28..5610c26 100644
--- a/libvpx/test/test_libvpx.cc
+++ b/libvpx/test/test_libvpx.cc
@@ -9,11 +9,17 @@
  */
 #include <string>
 #include "vpx_config.h"
-#if ARCH_X86 || ARCH_X86_64
 extern "C" {
+#if ARCH_X86 || ARCH_X86_64
 #include "vpx_ports/x86.h"
-}
 #endif
+#if CONFIG_VP8
+extern void vp8_rtcd();
+#endif
+#if CONFIG_VP9
+extern void vp9_rtcd();
+#endif
+}
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
 static void append_gtest_filter(const char *str) {
@@ -27,19 +33,29 @@
 
 #if ARCH_X86 || ARCH_X86_64
   const int simd_caps = x86_simd_caps();
-  if(!(simd_caps & HAS_MMX))
+  if (!(simd_caps & HAS_MMX))
     append_gtest_filter(":-MMX/*");
-  if(!(simd_caps & HAS_SSE))
+  if (!(simd_caps & HAS_SSE))
     append_gtest_filter(":-SSE/*");
-  if(!(simd_caps & HAS_SSE2))
+  if (!(simd_caps & HAS_SSE2))
     append_gtest_filter(":-SSE2/*");
-  if(!(simd_caps & HAS_SSE3))
+  if (!(simd_caps & HAS_SSE3))
     append_gtest_filter(":-SSE3/*");
-  if(!(simd_caps & HAS_SSSE3))
+  if (!(simd_caps & HAS_SSSE3))
     append_gtest_filter(":-SSSE3/*");
-  if(!(simd_caps & HAS_SSE4_1))
+  if (!(simd_caps & HAS_SSE4_1))
     append_gtest_filter(":-SSE4_1/*");
 #endif
 
+#if !CONFIG_SHARED
+  /* Shared library builds don't support whitebox tests that exercise internal symbols. */
+#if CONFIG_VP8
+  vp8_rtcd();
+#endif
+#if CONFIG_VP9
+  vp9_rtcd();
+#endif
+#endif
+
   return RUN_ALL_TESTS();
 }
diff --git a/libvpx/test/test_vector_test.cc b/libvpx/test/test_vector_test.cc
index 938457b..d7bd184 100644
--- a/libvpx/test/test_vector_test.cc
+++ b/libvpx/test/test_vector_test.cc
@@ -12,20 +12,19 @@
 #include <cstdlib>
 #include <string>
 #include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
 #include "test/decode_test_driver.h"
 #include "test/ivf_video_source.h"
+#include "test/webm_video_source.h"
+#include "test/util.h"
+#include "test/md5_helper.h"
 extern "C" {
-#include "./md5_utils.h"
 #include "vpx_mem/vpx_mem.h"
 }
 
-#if defined(_MSC_VER)
-#define snprintf sprintf_s
-#endif
-
 namespace {
-// There are 61 test vectors in total.
-const char *kTestVectors[] = {
+#if CONFIG_VP8_DECODER
+const char *kVP8TestVectors[] = {
   "vp80-00-comprehensive-001.ivf",
   "vp80-00-comprehensive-002.ivf", "vp80-00-comprehensive-003.ivf",
   "vp80-00-comprehensive-004.ivf", "vp80-00-comprehensive-005.ivf",
@@ -58,11 +57,71 @@
   "vp80-05-sharpness-1438.ivf", "vp80-05-sharpness-1439.ivf",
   "vp80-05-sharpness-1440.ivf", "vp80-05-sharpness-1443.ivf"
 };
+#endif
+#if CONFIG_VP9_DECODER
+const char *kVP9TestVectors[] = {
+  "vp90-00-akiyo-200.webm", "vp90-00-akiyo-300.webm",
+  "vp90-00-akiyo-50.webm", "vp90-00-bowing-150.webm",
+  "vp90-00-bowing-25.webm", "vp90-00-bowing-400.webm",
+  "vp90-00-bus-100.webm", "vp90-00-bus-2000.webm",
+  "vp90-00-bus-300.webm", "vp90-00-bus-4400.webm",
+  "vp90-00-bus-800.webm", "vp90-00-cheer-1600.webm",
+  "vp90-00-cheer-2800.webm", "vp90-00-cheer-400.webm",
+  "vp90-00-cheer-600.webm", "vp90-00-city-1200.webm",
+  "vp90-00-city-2000.webm", "vp90-00-city-300.webm",
+  "vp90-00-city-600.webm", "vp90-00-coastguard-1200.webm",
+  "vp90-00-coastguard-200.webm", "vp90-00-coastguard-3600.webm",
+  "vp90-00-coastguard-5200.webm", "vp90-00-container-1000.webm",
+  "vp90-00-container-200.webm", "vp90-00-container-50.webm",
+  "vp90-00-deadline-1000.webm", "vp90-00-deadline-200.webm",
+  "vp90-00-deadline-50.webm", "vp90-00-flower-100.webm",
+  "vp90-00-flower-2000.webm", "vp90-00-flower-300.webm",
+  "vp90-00-flower-4400.webm", "vp90-00-flower-800.webm",
+  "vp90-00-football-1600.webm", "vp90-00-football-2800.webm",
+  "vp90-00-football-400.webm", "vp90-00-football-600.webm",
+  "vp90-00-foreman-1200.webm", "vp90-00-foreman-2000.webm",
+  "vp90-00-foreman-300.webm", "vp90-00-foreman-600.webm",
+  "vp90-00-hallmonitor-1200.webm", "vp90-00-hallmonitor-2000.webm",
+  "vp90-00-hallmonitor-300.webm", "vp90-00-hallmonitor-600.webm",
+  "vp90-00-harbour-1200.webm", "vp90-00-harbour-200.webm",
+  "vp90-00-harbour-3600.webm", "vp90-00-harbour-5200.webm",
+  "vp90-00-highway-100.webm", "vp90-00-highway-1600.webm",
+  "vp90-00-highway-2800.webm", "vp90-00-highway-50.webm",
+  "vp90-00-husky-100.webm", "vp90-00-husky-2000.webm",
+  "vp90-00-husky-300.webm", "vp90-00-husky-4400.webm",
+  "vp90-00-husky-800.webm", "vp90-00-ice-150.webm",
+  "vp90-00-ice-400.webm", "vp90-00-ice-800.webm",
+  "vp90-00-mobile-1600.webm", "vp90-00-mobile-2800.webm",
+  "vp90-00-mobile-400.webm", "vp90-00-mobile-600.webm",
+  "vp90-00-motherdaughter-100.webm", "vp90-00-motherdaughter-300.webm",
+  "vp90-00-motherdaughter-600.webm", "vp90-00-news-100.webm",
+  "vp90-00-news-300.webm", "vp90-00-news-600.webm",
+  "vp90-00-pamphlet-150.webm", "vp90-00-pamphlet-25.webm",
+  "vp90-00-pamphlet-400.webm", "vp90-00-paris-1000.webm",
+  "vp90-00-paris-200.webm", "vp90-00-paris-50.webm",
+  "vp90-00-signirene-1000.webm", "vp90-00-signirene-200.webm",
+  "vp90-00-signirene-50.webm", "vp90-00-silent-1000.webm",
+  "vp90-00-silent-200.webm", "vp90-00-silent-50.webm",
+  "vp90-00-soccer-100.webm", "vp90-00-soccer-2000.webm",
+  "vp90-00-soccer-300.webm", "vp90-00-soccer-4400.webm",
+  "vp90-00-soccer-800.webm", "vp90-00-stefan-1600.webm",
+  "vp90-00-stefan-2800.webm", "vp90-00-stefan-400.webm",
+  "vp90-00-stefan-600.webm", "vp90-00-students-100.webm",
+  "vp90-00-students-300.webm", "vp90-00-students-600.webm",
+  "vp90-00-tempete-1200.webm", "vp90-00-tempete-200.webm",
+  "vp90-00-tempete-3600.webm", "vp90-00-tempete-5200.webm",
+  "vp90-00-tennis-100.webm", "vp90-00-tennis-2000.webm",
+  "vp90-00-tennis-300.webm", "vp90-00-tennis-4400.webm",
+  "vp90-00-tennis-800.webm", "vp90-00-waterfall-150.webm",
+  "vp90-00-waterfall-200.webm", "vp90-00-waterfall-400.webm",
+  "vp90-00-waterfall-800.webm",
+};
+#endif
 
-class TestVectorTest : public libvpx_test::DecoderTest,
-    public ::testing::TestWithParam<const char*> {
+class TestVectorTest : public ::libvpx_test::DecoderTest,
+    public ::libvpx_test::CodecTestWithParam<const char*> {
  protected:
-  TestVectorTest() : md5_file_(NULL) {}
+  TestVectorTest() : DecoderTest(GET_PARAM(0)), md5_file_(NULL) {}
 
   virtual ~TestVectorTest() {
     if (md5_file_)
@@ -85,30 +144,9 @@
     ASSERT_NE(res, EOF) << "Read md5 data failed";
     expected_md5[32] = '\0';
 
-    MD5Context md5;
-    MD5Init(&md5);
-
-    // Compute and update md5 for each raw in decompressed data.
-    for (int plane = 0; plane < 3; ++plane) {
-      uint8_t *buf = img.planes[plane];
-
-      for (unsigned int y = 0; y < (plane ? (img.d_h + 1) >> 1 : img.d_h);
-           ++y) {
-        MD5Update(&md5, buf, (plane ? (img.d_w + 1) >> 1 : img.d_w));
-        buf += img.stride[plane];
-      }
-    }
-
-    uint8_t md5_sum[16];
-    MD5Final(md5_sum, &md5);
-
-    char actual_md5[33];
-    // Convert to get the actual md5.
-    for (int i = 0; i < 16; i++) {
-      snprintf(&actual_md5[i * 2], sizeof(actual_md5) - i * 2, "%02x",
-               md5_sum[i]);
-    }
-    actual_md5[32] = '\0';
+    ::libvpx_test::MD5 md5_res;
+    md5_res.Add(&img);
+    const char *actual_md5 = md5_res.Get();
 
     // Check md5 match.
     ASSERT_STREQ(expected_md5, actual_md5)
@@ -124,21 +162,29 @@
 // checksums match the correct md5 data, then the test is passed. Otherwise,
 // the test failed.
 TEST_P(TestVectorTest, MD5Match) {
-  const std::string filename = GetParam();
-  // Open compressed video file.
-  libvpx_test::IVFVideoSource video(filename);
+  const std::string filename = GET_PARAM(1);
+  libvpx_test::CompressedVideoSource *video = NULL;
 
-  video.Init();
+  // Open compressed video file.
+  if (filename.substr(filename.length() - 3, 3) == "ivf") {
+    video = new libvpx_test::IVFVideoSource(filename);
+  } else if (filename.substr(filename.length() - 4, 4) == "webm") {
+    video = new libvpx_test::WebMVideoSource(filename);
+  }
+  video->Init();
 
   // Construct md5 file name.
   const std::string md5_filename = filename + ".md5";
   OpenMD5File(md5_filename);
 
   // Decode frame, and check the md5 matching.
-  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video));
+  delete video;
 }
 
-INSTANTIATE_TEST_CASE_P(TestVectorSequence, TestVectorTest,
-                        ::testing::ValuesIn(kTestVectors));
+VP8_INSTANTIATE_TEST_CASE(TestVectorTest,
+                          ::testing::ValuesIn(kVP8TestVectors));
+VP9_INSTANTIATE_TEST_CASE(TestVectorTest,
+                          ::testing::ValuesIn(kVP9TestVectors));
 
 }  // namespace
diff --git a/libvpx/test/tile_independence_test.cc b/libvpx/test/tile_independence_test.cc
new file mode 100644
index 0000000..9633ed7
--- /dev/null
+++ b/libvpx/test/tile_independence_test.cc
@@ -0,0 +1,108 @@
+/*
+ Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+
+ Use of this source code is governed by a BSD-style license
+ that can be found in the LICENSE file in the root of the source
+ tree. An additional intellectual property rights grant can be found
+ in the file PATENTS.  All contributing project authors may
+ be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/md5_helper.h"
+extern "C" {
+#include "vpx_mem/vpx_mem.h"
+}
+
+namespace {
+class TileIndependenceTest : public ::libvpx_test::EncoderTest,
+    public ::libvpx_test::CodecTestWithParam<int> {
+ protected:
+  TileIndependenceTest() : EncoderTest(GET_PARAM(0)), n_tiles_(GET_PARAM(1)),
+      md5_fw_order_(), md5_inv_order_() {
+    init_flags_ = VPX_CODEC_USE_PSNR;
+    vpx_codec_dec_cfg_t cfg;
+    cfg.w = 704;
+    cfg.h = 144;
+    cfg.threads = 1;
+    fw_dec_ = codec_->CreateDecoder(cfg, 0);
+    inv_dec_ = codec_->CreateDecoder(cfg, 0);
+    inv_dec_->Control(VP9_INVERT_TILE_DECODE_ORDER, 1);
+  }
+
+  virtual ~TileIndependenceTest() {
+    delete fw_dec_;
+    delete inv_dec_;
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(libvpx_test::kTwoPassGood);
+  }
+
+  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
+                                  libvpx_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(VP9E_SET_TILE_COLUMNS, n_tiles_);
+    }
+  }
+
+  void UpdateMD5(::libvpx_test::Decoder *dec, const vpx_codec_cx_pkt_t *pkt,
+                 ::libvpx_test::MD5 *md5) {
+    const vpx_codec_err_t res =
+        dec->DecodeFrame(reinterpret_cast<uint8_t*>(pkt->data.frame.buf),
+                         pkt->data.frame.sz);
+    if (res != VPX_CODEC_OK) {
+      abort_ = true;
+      ASSERT_EQ(VPX_CODEC_OK, res);
+    }
+    const vpx_image_t *img = dec->GetDxData().Next();
+    md5->Add(img);
+  }
+
+  virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
+    UpdateMD5(fw_dec_, pkt, &md5_fw_order_);
+    UpdateMD5(inv_dec_, pkt, &md5_inv_order_);
+  }
+
+ private:
+  int n_tiles_;
+ protected:
+  ::libvpx_test::MD5 md5_fw_order_, md5_inv_order_;
+  ::libvpx_test::Decoder *fw_dec_, *inv_dec_;
+};
+
+// run an encode with 2 or 4 tiles, and do the decode both in normal and
+// inverted tile ordering. Ensure that the MD5 of the output in both cases
+// is identical. If so, tiles are considered independent and the test passes.
+TEST_P(TileIndependenceTest, MD5Match) {
+  const vpx_rational timebase = { 33333333, 1000000000 };
+  cfg_.g_timebase = timebase;
+  cfg_.rc_target_bitrate = 500;
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_end_usage = VPX_VBR;
+
+  libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 704, 144,
+                                     timebase.den, timebase.num, 0, 30);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  const char *md5_fw_str  = md5_fw_order_.Get();
+  const char *md5_inv_str = md5_inv_order_.Get();
+
+  // could use ASSERT_EQ(!memcmp(.., .., 16) here, but this gives nicer
+  // output if it fails. Not sure if it's helpful since it's really just
+  // a MD5...
+  ASSERT_STREQ(md5_fw_str, md5_inv_str);
+}
+
+VP9_INSTANTIATE_TEST_CASE(TileIndependenceTest,
+                          ::testing::Range(0, 2, 1));
+
+}  // namespace
diff --git a/libvpx/test/util.h b/libvpx/test/util.h
index 06a70cc..533a1db 100644
--- a/libvpx/test/util.h
+++ b/libvpx/test/util.h
@@ -11,8 +11,38 @@
 #ifndef TEST_UTIL_H_
 #define TEST_UTIL_H_
 
+#include <stdio.h>
+#include <math.h>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "vpx/vpx_image.h"
+
 // Macros
 #define PARAMS(...) ::testing::TestWithParam< std::tr1::tuple< __VA_ARGS__ > >
 #define GET_PARAM(k) std::tr1::get< k >(GetParam())
 
+static double compute_psnr(const vpx_image_t *img1,
+                           const vpx_image_t *img2) {
+  assert((img1->fmt == img2->fmt) &&
+         (img1->d_w == img2->d_w) &&
+         (img1->d_h == img2->d_h));
+
+  const unsigned int width_y  = img1->d_w;
+  const unsigned int height_y = img1->d_h;
+  unsigned int i, j;
+
+  int64_t sqrerr = 0;
+  for (i = 0; i < height_y; ++i)
+    for (j = 0; j < width_y; ++j) {
+      int64_t d = img1->planes[VPX_PLANE_Y][i * img1->stride[VPX_PLANE_Y] + j] -
+                  img2->planes[VPX_PLANE_Y][i * img2->stride[VPX_PLANE_Y] + j];
+      sqrerr += d * d;
+    }
+  double mse = sqrerr / (width_y * height_y);
+  double psnr = 100.0;
+  if (mse > 0.0) {
+    psnr = 10 * log10(255.0 * 255.0 / mse);
+  }
+  return psnr;
+}
+
 #endif  // TEST_UTIL_H_
diff --git a/libvpx/test/variance_test.cc b/libvpx/test/variance_test.cc
new file mode 100644
index 0000000..dfa1a07
--- /dev/null
+++ b/libvpx/test/variance_test.cc
@@ -0,0 +1,211 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include <stdlib.h>
+#include <new>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "test/clear_system_state.h"
+
+#include "vpx/vpx_integer.h"
+#include "vpx_config.h"
+extern "C" {
+#if CONFIG_VP8_ENCODER
+# include "vp8/common/variance.h"
+# include "vp8_rtcd.h"
+#endif
+#if CONFIG_VP9_ENCODER
+# include "vp9/encoder/vp9_variance.h"
+# include "vp9_rtcd.h"
+#endif
+}
+
+namespace {
+
+using ::std::tr1::get;
+using ::std::tr1::make_tuple;
+using ::std::tr1::tuple;
+
+template<typename VarianceFunctionType>
+class VarianceTest :
+    public ::testing::TestWithParam<tuple<int, int, VarianceFunctionType> > {
+ public:
+  virtual void SetUp() {
+    const tuple<int, int, VarianceFunctionType>& params = this->GetParam();
+    width_  = get<0>(params);
+    height_ = get<1>(params);
+    variance_ = get<2>(params);
+
+    block_size_ = width_ * height_;
+    src_ = new uint8_t[block_size_];
+    ref_ = new uint8_t[block_size_];
+    ASSERT_TRUE(src_ != NULL);
+    ASSERT_TRUE(ref_ != NULL);
+  }
+
+  virtual void TearDown() {
+    delete[] src_;
+    delete[] ref_;
+    libvpx_test::ClearSystemState();
+  }
+
+ protected:
+  void ZeroTest();
+  void OneQuarterTest();
+
+  uint8_t* src_;
+  uint8_t* ref_;
+  int width_;
+  int height_;
+  int block_size_;
+  VarianceFunctionType variance_;
+
+};
+
+template<typename VarianceFunctionType>
+void VarianceTest<VarianceFunctionType>::ZeroTest() {
+  for (int i = 0; i <= 255; ++i) {
+    memset(src_, i, block_size_);
+    for (int j = 0; j <= 255; ++j) {
+      memset(ref_, j, block_size_);
+      unsigned int sse;
+      const unsigned int var = variance_(src_, width_, ref_, width_, &sse);
+      EXPECT_EQ(0u, var) << "src values: " << i << "ref values: " << j;
+    }
+  }
+}
+
+template<typename VarianceFunctionType>
+void VarianceTest<VarianceFunctionType>::OneQuarterTest() {
+  memset(src_, 255, block_size_);
+  const int half = block_size_ / 2;
+  memset(ref_, 255, half);
+  memset(ref_ + half, 0, half);
+  unsigned int sse;
+  const unsigned int var = variance_(src_, width_, ref_, width_, &sse);
+  const unsigned int expected = block_size_ * 255 * 255 / 4;
+  EXPECT_EQ(expected, var);
+}
+
+// -----------------------------------------------------------------------------
+// VP8 test cases.
+
+namespace vp8 {
+
+#if CONFIG_VP8_ENCODER
+typedef VarianceTest<vp8_variance_fn_t> VP8VarianceTest;
+
+TEST_P(VP8VarianceTest, Zero) { ZeroTest(); }
+TEST_P(VP8VarianceTest, OneQuarter) { OneQuarterTest(); }
+
+const vp8_variance_fn_t variance4x4_c = vp8_variance4x4_c;
+const vp8_variance_fn_t variance8x8_c = vp8_variance8x8_c;
+const vp8_variance_fn_t variance8x16_c = vp8_variance8x16_c;
+const vp8_variance_fn_t variance16x8_c = vp8_variance16x8_c;
+const vp8_variance_fn_t variance16x16_c = vp8_variance16x16_c;
+INSTANTIATE_TEST_CASE_P(
+    C, VP8VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_c),
+                      make_tuple(8, 8, variance8x8_c),
+                      make_tuple(8, 16, variance8x16_c),
+                      make_tuple(16, 8, variance16x8_c),
+                      make_tuple(16, 16, variance16x16_c)));
+
+#if HAVE_MMX
+const vp8_variance_fn_t variance4x4_mmx = vp8_variance4x4_mmx;
+const vp8_variance_fn_t variance8x8_mmx = vp8_variance8x8_mmx;
+const vp8_variance_fn_t variance8x16_mmx = vp8_variance8x16_mmx;
+const vp8_variance_fn_t variance16x8_mmx = vp8_variance16x8_mmx;
+const vp8_variance_fn_t variance16x16_mmx = vp8_variance16x16_mmx;
+INSTANTIATE_TEST_CASE_P(
+    MMX, VP8VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_mmx),
+                      make_tuple(8, 8, variance8x8_mmx),
+                      make_tuple(8, 16, variance8x16_mmx),
+                      make_tuple(16, 8, variance16x8_mmx),
+                      make_tuple(16, 16, variance16x16_mmx)));
+#endif
+
+#if HAVE_SSE2
+const vp8_variance_fn_t variance4x4_wmt = vp8_variance4x4_wmt;
+const vp8_variance_fn_t variance8x8_wmt = vp8_variance8x8_wmt;
+const vp8_variance_fn_t variance8x16_wmt = vp8_variance8x16_wmt;
+const vp8_variance_fn_t variance16x8_wmt = vp8_variance16x8_wmt;
+const vp8_variance_fn_t variance16x16_wmt = vp8_variance16x16_wmt;
+INSTANTIATE_TEST_CASE_P(
+    SSE2, VP8VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_wmt),
+                      make_tuple(8, 8, variance8x8_wmt),
+                      make_tuple(8, 16, variance8x16_wmt),
+                      make_tuple(16, 8, variance16x8_wmt),
+                      make_tuple(16, 16, variance16x16_wmt)));
+#endif
+#endif  // CONFIG_VP8_ENCODER
+
+}  // namespace vp8
+
+// -----------------------------------------------------------------------------
+// VP9 test cases.
+
+namespace vp9 {
+
+#if CONFIG_VP9_ENCODER
+typedef VarianceTest<vp9_variance_fn_t> VP9VarianceTest;
+
+TEST_P(VP9VarianceTest, Zero) { ZeroTest(); }
+TEST_P(VP9VarianceTest, OneQuarter) { OneQuarterTest(); }
+
+const vp9_variance_fn_t variance4x4_c = vp9_variance4x4_c;
+const vp9_variance_fn_t variance8x8_c = vp9_variance8x8_c;
+const vp9_variance_fn_t variance8x16_c = vp9_variance8x16_c;
+const vp9_variance_fn_t variance16x8_c = vp9_variance16x8_c;
+const vp9_variance_fn_t variance16x16_c = vp9_variance16x16_c;
+INSTANTIATE_TEST_CASE_P(
+    C, VP9VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_c),
+                      make_tuple(8, 8, variance8x8_c),
+                      make_tuple(8, 16, variance8x16_c),
+                      make_tuple(16, 8, variance16x8_c),
+                      make_tuple(16, 16, variance16x16_c)));
+
+#if HAVE_MMX
+const vp9_variance_fn_t variance4x4_mmx = vp9_variance4x4_mmx;
+const vp9_variance_fn_t variance8x8_mmx = vp9_variance8x8_mmx;
+const vp9_variance_fn_t variance8x16_mmx = vp9_variance8x16_mmx;
+const vp9_variance_fn_t variance16x8_mmx = vp9_variance16x8_mmx;
+const vp9_variance_fn_t variance16x16_mmx = vp9_variance16x16_mmx;
+INSTANTIATE_TEST_CASE_P(
+    MMX, VP9VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_mmx),
+                      make_tuple(8, 8, variance8x8_mmx),
+                      make_tuple(8, 16, variance8x16_mmx),
+                      make_tuple(16, 8, variance16x8_mmx),
+                      make_tuple(16, 16, variance16x16_mmx)));
+#endif
+
+#if HAVE_SSE2
+const vp9_variance_fn_t variance4x4_wmt = vp9_variance4x4_sse2;
+const vp9_variance_fn_t variance8x8_wmt = vp9_variance8x8_sse2;
+const vp9_variance_fn_t variance8x16_wmt = vp9_variance8x16_sse2;
+const vp9_variance_fn_t variance16x8_wmt = vp9_variance16x8_sse2;
+const vp9_variance_fn_t variance16x16_wmt = vp9_variance16x16_sse2;
+INSTANTIATE_TEST_CASE_P(
+    SSE2, VP9VarianceTest,
+    ::testing::Values(make_tuple(4, 4, variance4x4_wmt),
+                      make_tuple(8, 8, variance8x8_wmt),
+                      make_tuple(8, 16, variance8x16_wmt),
+                      make_tuple(16, 8, variance16x8_wmt),
+                      make_tuple(16, 16, variance16x16_wmt)));
+#endif
+#endif  // CONFIG_VP9_ENCODER
+
+}  // namespace vp9
+
+}  // namespace
diff --git a/libvpx/test/video_source.h b/libvpx/test/video_source.h
index 9772657..26d5328 100644
--- a/libvpx/test/video_source.h
+++ b/libvpx/test/video_source.h
@@ -103,7 +103,7 @@
     if (width != width_ || height != height_) {
       vpx_img_free(img_);
       raw_sz_ = ((width + 31)&~31) * height * 3 / 2;
-      img_ = vpx_img_alloc(NULL, VPX_IMG_FMT_VPXI420, width, height, 32);
+      img_ = vpx_img_alloc(NULL, VPX_IMG_FMT_I420, width, height, 32);
       width_ = width;
       height_ = height;
     }
diff --git a/libvpx/test/boolcoder_test.cc b/libvpx/test/vp8_boolcoder_test.cc
similarity index 72%
rename from libvpx/test/boolcoder_test.cc
rename to libvpx/test/vp8_boolcoder_test.cc
index 4e21be8..c3a8d12 100644
--- a/libvpx/test/boolcoder_test.cc
+++ b/libvpx/test/vp8_boolcoder_test.cc
@@ -26,6 +26,30 @@
 
 namespace {
 const int num_tests = 10;
+
+// In a real use the 'decrypt_state' parameter will be a pointer to a struct
+// with whatever internal state the decryptor uses. For testing we'll just
+// xor with a constant key, and decrypt_state will point to the start of
+// the original buffer.
+const uint8_t secret_key[16] = {
+  0x01, 0x12, 0x23, 0x34, 0x45, 0x56, 0x67, 0x78,
+  0x89, 0x9a, 0xab, 0xbc, 0xcd, 0xde, 0xef, 0xf0
+};
+
+void encrypt_buffer(uint8_t *buffer, int size) {
+  for (int i = 0; i < size; ++i) {
+    buffer[i] ^= secret_key[i & 15];
+  }
+}
+
+void test_decrypt_cb(void *decrypt_state, const uint8_t *input,
+                           uint8_t *output, int count) {
+  int offset = input - (uint8_t *)decrypt_state;
+  for (int i = 0; i < count; i++) {
+    output[i] = input[i] ^ secret_key[(offset + i) & 15];
+  }
+}
+
 }  // namespace
 
 using libvpx_test::ACMRandom;
@@ -71,7 +95,13 @@
         vp8_stop_encode(&bw);
 
         BOOL_DECODER br;
-        vp8dx_start_decode(&br, bw_buffer, buffer_size);
+#if CONFIG_DECRYPT
+        encrypt_buffer(bw_buffer, buffer_size);
+        vp8dx_start_decode(&br, bw_buffer, buffer_size,
+                           test_decrypt_cb, (void *)bw_buffer);
+#else
+        vp8dx_start_decode(&br, bw_buffer, buffer_size, NULL, NULL);
+#endif
         bit_rnd.Reset(random_seed);
         for (int i = 0; i < bits_to_test; ++i) {
           if (bit_method == 2) {
diff --git a/libvpx/test/vp8_decrypt_test.cc b/libvpx/test/vp8_decrypt_test.cc
new file mode 100644
index 0000000..d850f00
--- /dev/null
+++ b/libvpx/test/vp8_decrypt_test.cc
@@ -0,0 +1,71 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include <vector>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/ivf_video_source.h"
+
+namespace {
+// In a real use the 'decrypt_state' parameter will be a pointer to a struct
+// with whatever internal state the decryptor uses. For testing we'll just
+// xor with a constant key, and decrypt_state will point to the start of
+// the original buffer.
+const uint8_t test_key[16] = {
+  0x01, 0x12, 0x23, 0x34, 0x45, 0x56, 0x67, 0x78,
+  0x89, 0x9a, 0xab, 0xbc, 0xcd, 0xde, 0xef, 0xf0
+};
+
+void encrypt_buffer(const uint8_t *src, uint8_t *dst, int size, int offset = 0) {
+  for (int i = 0; i < size; ++i) {
+    dst[i] = src[i] ^ test_key[(offset + i) & 15];
+  }
+}
+
+void test_decrypt_cb(void *decrypt_state, const uint8_t *input,
+                     uint8_t *output, int count) {
+  encrypt_buffer(input, output, count, input - (uint8_t *)decrypt_state);
+}
+
+} // namespace
+
+namespace libvpx_test {
+
+TEST(TestDecrypt, DecryptWorks) {
+  libvpx_test::IVFVideoSource video("vp80-00-comprehensive-001.ivf");
+  video.Init();
+
+  vpx_codec_dec_cfg_t dec_cfg = {0};
+  VP8Decoder decoder(dec_cfg, 0);
+
+  video.Begin();
+
+  // no decryption
+  vpx_codec_err_t res = decoder.DecodeFrame(video.cxdata(), video.frame_size());
+  ASSERT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
+
+  // decrypt frame
+  video.Next();
+
+#if CONFIG_DECRYPT
+  std::vector<uint8_t> encrypted(video.frame_size());
+  encrypt_buffer(video.cxdata(), &encrypted[0], video.frame_size());
+  vp8_decrypt_init di = { test_decrypt_cb, &encrypted[0] };
+  decoder.Control(VP8D_SET_DECRYPTOR, &di);
+#endif  // CONFIG_DECRYPT
+
+  res = decoder.DecodeFrame(video.cxdata(), video.frame_size());
+  ASSERT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
+}
+
+}  // namespace libvpx_test
diff --git a/libvpx/test/vp8_fdct4x4_test.cc b/libvpx/test/vp8_fdct4x4_test.cc
new file mode 100644
index 0000000..3c60011
--- /dev/null
+++ b/libvpx/test/vp8_fdct4x4_test.cc
@@ -0,0 +1,169 @@
+/*
+*  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+*
+*  Use of this source code is governed by a BSD-style license
+*  that can be found in the LICENSE file in the root of the source
+*  tree. An additional intellectual property rights grant can be found
+*  in the file PATENTS.  All contributing project authors may
+*  be found in the AUTHORS file in the root of the source tree.
+*/
+
+
+#include <math.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/types.h>
+
+
+extern "C" {
+#include "vp8_rtcd.h"
+}
+
+#include "test/acm_random.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "vpx/vpx_integer.h"
+
+
+namespace {
+
+const int cospi8sqrt2minus1 = 20091;
+const int sinpi8sqrt2 = 35468;
+
+void reference_idct4x4(const int16_t *input, int16_t *output) {
+  const int16_t *ip = input;
+  int16_t *op = output;
+
+  for (int i = 0; i < 4; ++i) {
+    const int a1 = ip[0] + ip[8];
+    const int b1 = ip[0] - ip[8];
+    const int temp1 = (ip[4] * sinpi8sqrt2) >> 16;
+    const int temp2 = ip[12] + ((ip[12] * cospi8sqrt2minus1) >> 16);
+    const int c1 = temp1 - temp2;
+    const int temp3 = ip[4] + ((ip[4] * cospi8sqrt2minus1) >> 16);
+    const int temp4 = (ip[12] * sinpi8sqrt2) >> 16;
+    const int d1 = temp3 + temp4;
+    op[0] = a1 + d1;
+    op[12] = a1 - d1;
+    op[4] = b1 + c1;
+    op[8] = b1 - c1;
+    ++ip;
+    ++op;
+  }
+  ip = output;
+  op = output;
+  for (int i = 0; i < 4; ++i) {
+    const int a1 = ip[0] + ip[2];
+    const int b1 = ip[0] - ip[2];
+    const int temp1 = (ip[1] * sinpi8sqrt2) >> 16;
+    const int temp2 = ip[3] + ((ip[3] * cospi8sqrt2minus1) >> 16);
+    const int c1 = temp1 - temp2;
+    const int temp3 = ip[1] + ((ip[1] * cospi8sqrt2minus1) >> 16);
+    const int temp4 = (ip[3] * sinpi8sqrt2) >> 16;
+    const int d1 = temp3 + temp4;
+    op[0] = (a1 + d1 + 4) >> 3;
+    op[3] = (a1 - d1 + 4) >> 3;
+    op[1] = (b1 + c1 + 4) >> 3;
+    op[2] = (b1 - c1 + 4) >> 3;
+    ip += 4;
+    op += 4;
+  }
+}
+
+using libvpx_test::ACMRandom;
+
+TEST(Vp8FdctTest, SignBiasCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int16_t test_input_block[16];
+  int16_t test_output_block[16];
+  const int pitch = 8;
+  int count_sign_block[16][2];
+  const int count_test_block = 1000000;
+
+  memset(count_sign_block, 0, sizeof(count_sign_block));
+
+  for (int i = 0; i < count_test_block; ++i) {
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 16; ++j)
+      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+
+    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+
+    for (int j = 0; j < 16; ++j) {
+      if (test_output_block[j] < 0)
+        ++count_sign_block[j][0];
+      else if (test_output_block[j] > 0)
+        ++count_sign_block[j][1];
+    }
+  }
+
+  bool bias_acceptable = true;
+  for (int j = 0; j < 16; ++j)
+    bias_acceptable = bias_acceptable &&
+    (abs(count_sign_block[j][0] - count_sign_block[j][1]) < 10000);
+
+  EXPECT_EQ(true, bias_acceptable)
+    << "Error: 4x4 FDCT has a sign bias > 1% for input range [-255, 255]";
+
+  memset(count_sign_block, 0, sizeof(count_sign_block));
+
+  for (int i = 0; i < count_test_block; ++i) {
+    // Initialize a test block with input range [-15, 15].
+    for (int j = 0; j < 16; ++j)
+      test_input_block[j] = (rnd.Rand8() >> 4) - (rnd.Rand8() >> 4);
+
+    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+
+    for (int j = 0; j < 16; ++j) {
+      if (test_output_block[j] < 0)
+        ++count_sign_block[j][0];
+      else if (test_output_block[j] > 0)
+        ++count_sign_block[j][1];
+    }
+  }
+
+  bias_acceptable = true;
+  for (int j = 0; j < 16; ++j)
+    bias_acceptable = bias_acceptable &&
+    (abs(count_sign_block[j][0] - count_sign_block[j][1]) < 100000);
+
+  EXPECT_EQ(true, bias_acceptable)
+    << "Error: 4x4 FDCT has a sign bias > 10% for input range [-15, 15]";
+};
+
+TEST(Vp8FdctTest, RoundTripErrorCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int max_error = 0;
+  double total_error = 0;
+  const int count_test_block = 1000000;
+  for (int i = 0; i < count_test_block; ++i) {
+    int16_t test_input_block[16];
+    int16_t test_temp_block[16];
+    int16_t test_output_block[16];
+
+    // Initialize a test block with input range [-255, 255].
+    for (int j = 0; j < 16; ++j)
+      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+
+    const int pitch = 8;
+    vp8_short_fdct4x4_c(test_input_block, test_temp_block, pitch);
+    reference_idct4x4(test_temp_block, test_output_block);
+
+    for (int j = 0; j < 16; ++j) {
+      const int diff = test_input_block[j] - test_output_block[j];
+      const int error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+      total_error += error;
+    }
+  }
+
+  EXPECT_GE(1, max_error )
+    << "Error: FDCT/IDCT has an individual roundtrip error > 1";
+
+  EXPECT_GE(count_test_block, total_error)
+    << "Error: FDCT/IDCT has average roundtrip error > 1 per block";
+};
+
+}  // namespace
diff --git a/libvpx/test/boolcoder_test.cc b/libvpx/test/vp9_boolcoder_test.cc
similarity index 70%
copy from libvpx/test/boolcoder_test.cc
copy to libvpx/test/vp9_boolcoder_test.cc
index 4e21be8..42b2229 100644
--- a/libvpx/test/boolcoder_test.cc
+++ b/libvpx/test/vp9_boolcoder_test.cc
@@ -8,29 +8,27 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-extern "C" {
-#include "vp8/encoder/boolhuff.h"
-#include "vp8/decoder/dboolhuff.h"
-}
-
 #include <math.h>
-#include <stddef.h>
-#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/types.h>
 
-#include "test/acm_random.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
+
+extern "C" {
+#include "vp9/encoder/vp9_boolhuff.h"
+#include "vp9/decoder/vp9_dboolhuff.h"
+}
+
+#include "acm_random.h"
 #include "vpx/vpx_integer.h"
 
+using libvpx_test::ACMRandom;
+
 namespace {
 const int num_tests = 10;
 }  // namespace
 
-using libvpx_test::ACMRandom;
-
-TEST(VP8, TestBitIO) {
+TEST(VP9, TestBitIO) {
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   for (int n = 0; n < num_tests; ++n) {
     for (int method = 0; method <= 7; ++method) {   // we generate various proba
@@ -40,23 +38,23 @@
       for (int i = 0; i < bits_to_test; ++i) {
         const int parity = i & 1;
         probas[i] =
-            (method == 0) ? 0 : (method == 1) ? 255 :
-            (method == 2) ? 128 :
-            (method == 3) ? rnd.Rand8() :
-            (method == 4) ? (parity ? 0 : 255) :
+          (method == 0) ? 0 : (method == 1) ? 255 :
+          (method == 2) ? 128 :
+          (method == 3) ? rnd.Rand8() :
+          (method == 4) ? (parity ? 0 : 255) :
             // alternate between low and high proba:
             (method == 5) ? (parity ? rnd(128) : 255 - rnd(128)) :
             (method == 6) ?
-                (parity ? rnd(64) : 255 - rnd(64)) :
-                (parity ? rnd(32) : 255 - rnd(32));
+            (parity ? rnd(64) : 255 - rnd(64)) :
+            (parity ? rnd(32) : 255 - rnd(32));
       }
       for (int bit_method = 0; bit_method <= 3; ++bit_method) {
         const int random_seed = 6432;
         const int buffer_size = 10000;
         ACMRandom bit_rnd(random_seed);
-        BOOL_CODER bw;
+        vp9_writer bw;
         uint8_t bw_buffer[buffer_size];
-        vp8_start_encode(&bw, bw_buffer, bw_buffer + buffer_size);
+        vp9_start_encode(&bw, bw_buffer);
 
         int bit = (bit_method == 0) ? 0 : (bit_method == 1) ? 1 : 0;
         for (int i = 0; i < bits_to_test; ++i) {
@@ -65,13 +63,16 @@
           } else if (bit_method == 3) {
             bit = bit_rnd(2);
           }
-          vp8_encode_bool(&bw, bit, static_cast<int>(probas[i]));
+          vp9_write(&bw, bit, static_cast<int>(probas[i]));
         }
 
-        vp8_stop_encode(&bw);
+        vp9_stop_encode(&bw);
 
-        BOOL_DECODER br;
-        vp8dx_start_decode(&br, bw_buffer, buffer_size);
+        // First bit should be zero
+        GTEST_ASSERT_EQ(bw_buffer[0] & 0x80, 0);
+
+        vp9_reader br;
+        vp9_reader_init(&br, bw_buffer, buffer_size);
         bit_rnd.Reset(random_seed);
         for (int i = 0; i < bits_to_test; ++i) {
           if (bit_method == 2) {
@@ -79,8 +80,8 @@
           } else if (bit_method == 3) {
             bit = bit_rnd(2);
           }
-          GTEST_ASSERT_EQ(vp8dx_decode_bool(&br, probas[i]), bit)
-              << "pos: "<< i << " / " << bits_to_test
+          GTEST_ASSERT_EQ(vp9_read(&br, probas[i]), bit)
+              << "pos: " << i << " / " << bits_to_test
               << " bit_method: " << bit_method
               << " method: " << method;
         }
diff --git a/libvpx/test/webm_video_source.h b/libvpx/test/webm_video_source.h
new file mode 100644
index 0000000..c7919a9
--- /dev/null
+++ b/libvpx/test/webm_video_source.h
@@ -0,0 +1,184 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef TEST_WEBM_VIDEO_SOURCE_H_
+#define TEST_WEBM_VIDEO_SOURCE_H_
+#include <cstdarg>
+#include <cstdio>
+#include <cstdlib>
+#include <new>
+#include <string>
+#include "nestegg/include/nestegg/nestegg.h"
+#include "test/video_source.h"
+
+namespace libvpx_test {
+
+static int
+nestegg_read_cb(void *buffer, size_t length, void *userdata) {
+  FILE *f = reinterpret_cast<FILE *>(userdata);
+
+  if (fread(buffer, 1, length, f) < length) {
+    if (ferror(f))
+      return -1;
+    if (feof(f))
+      return 0;
+  }
+  return 1;
+}
+
+
+static int
+nestegg_seek_cb(int64_t offset, int whence, void *userdata) {
+  FILE *f = reinterpret_cast<FILE *>(userdata);
+  switch (whence) {
+    case NESTEGG_SEEK_SET:
+      whence = SEEK_SET;
+      break;
+    case NESTEGG_SEEK_CUR:
+      whence = SEEK_CUR;
+      break;
+    case NESTEGG_SEEK_END:
+      whence = SEEK_END;
+      break;
+  };
+  return fseek(f, (long)offset, whence) ? -1 : 0;
+}
+
+
+static int64_t
+nestegg_tell_cb(void *userdata) {
+  FILE *f = reinterpret_cast<FILE *>(userdata);
+  return ftell(f);
+}
+
+
+static void
+nestegg_log_cb(nestegg *context, unsigned int severity, char const *format,
+               ...) {
+  va_list ap;
+
+  va_start(ap, format);
+  vfprintf(stderr, format, ap);
+  fprintf(stderr, "\n");
+  va_end(ap);
+}
+
+// This class extends VideoSource to allow parsing of WebM files,
+// so that we can do actual file decodes.
+class WebMVideoSource : public CompressedVideoSource {
+ public:
+  explicit WebMVideoSource(const std::string &file_name)
+      : file_name_(file_name),
+        input_file_(NULL),
+        nestegg_ctx_(NULL),
+        pkt_(NULL),
+        video_track_(0),
+        chunk_(0),
+        chunks_(0),
+        buf_(NULL),
+        buf_sz_(0),
+        frame_(0),
+        end_of_file_(false) {
+  }
+
+  virtual ~WebMVideoSource() {
+    if (input_file_)
+      fclose(input_file_);
+    if (nestegg_ctx_)
+      nestegg_destroy(nestegg_ctx_);
+  }
+
+  virtual void Init() {
+  }
+
+  virtual void Begin() {
+    input_file_ = OpenTestDataFile(file_name_);
+    ASSERT_TRUE(input_file_) << "Input file open failed. Filename: "
+        << file_name_;
+
+    nestegg_io io = {nestegg_read_cb, nestegg_seek_cb, nestegg_tell_cb,
+                     input_file_};
+    ASSERT_FALSE(nestegg_init(&nestegg_ctx_, io, NULL))
+        << "nestegg_init failed";
+
+    unsigned int n;
+    ASSERT_FALSE(nestegg_track_count(nestegg_ctx_, &n))
+        << "failed to get track count";
+
+    for (unsigned int i = 0; i < n; i++) {
+      int track_type = nestegg_track_type(nestegg_ctx_, i);
+      ASSERT_GE(track_type, 0) << "failed to get track type";
+
+      if (track_type == NESTEGG_TRACK_VIDEO) {
+        video_track_ = i;
+        break;
+      }
+    }
+
+    FillFrame();
+  }
+
+  virtual void Next() {
+    ++frame_;
+    FillFrame();
+  }
+
+  void FillFrame() {
+    if (chunk_ >= chunks_) {
+      unsigned int track;
+
+      do {
+        /* End of this packet, get another. */
+        if (pkt_)
+          nestegg_free_packet(pkt_);
+
+        int again = nestegg_read_packet(nestegg_ctx_, &pkt_);
+        ASSERT_GE(again, 0) << "nestegg_read_packet failed";
+        if (!again) {
+          end_of_file_ = true;
+          return;
+        }
+
+        ASSERT_FALSE(nestegg_packet_track(pkt_, &track))
+            << "nestegg_packet_track failed";
+      } while (track != video_track_);
+
+      ASSERT_FALSE(nestegg_packet_count(pkt_, &chunks_))
+          << "nestegg_packet_count failed";
+      chunk_ = 0;
+    }
+
+    ASSERT_FALSE(nestegg_packet_data(pkt_, chunk_, &buf_, &buf_sz_))
+        << "nestegg_packet_data failed";
+    chunk_++;
+  }
+
+  virtual const uint8_t *cxdata() const {
+    return end_of_file_ ? NULL : buf_;
+  }
+  virtual const unsigned int frame_size() const { return buf_sz_; }
+  virtual const unsigned int frame_number() const { return frame_; }
+
+ protected:
+  std::string file_name_;
+  FILE *input_file_;
+  nestegg *nestegg_ctx_;
+  nestegg_packet *pkt_;
+  unsigned int video_track_;
+  unsigned int chunk_;
+  unsigned int chunks_;
+  uint8_t *buf_;
+  size_t buf_sz_;
+  unsigned int frame_;
+  bool end_of_file_;
+};
+
+}  // namespace libvpx_test
+
+#endif  // TEST_WEBM_VIDEO_SOURCE_H_
diff --git a/libvpx/third_party/libyuv/source/scale.c b/libvpx/third_party/libyuv/source/scale.c
index c142a17..72a817d 100644
--- a/libvpx/third_party/libyuv/source/scale.c
+++ b/libvpx/third_party/libyuv/source/scale.c
@@ -632,7 +632,7 @@
   { 65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3, 65536 / 3, 65536 / 2, 0, 0 };
 #endif
 
-#if defined(_M_IX86) && !defined(YUV_DISABLE_ASM)
+#if defined(_M_IX86) && !defined(YUV_DISABLE_ASM) && defined(_MSC_VER)
 
 #define HAS_SCALEROWDOWN2_SSE2
 // Reads 32 pixels, throws half away and writes 16 pixels.
diff --git a/libvpx/third_party/x86inc/LICENSE b/libvpx/third_party/x86inc/LICENSE
new file mode 100644
index 0000000..7d07645
--- /dev/null
+++ b/libvpx/third_party/x86inc/LICENSE
@@ -0,0 +1,18 @@
+Copyright (C) 2005-2012 x264 project
+
+Authors: Loren Merritt <lorenm@u.washington.edu>
+         Anton Mitrofanov <BugMaster@narod.ru>
+         Jason Garrett-Glaser <darkshikari@gmail.com>
+         Henrik Gramner <hengar-6@student.ltu.se>
+
+Permission to use, copy, modify, and/or distribute this software for any
+purpose with or without fee is hereby granted, provided that the above
+copyright notice and this permission notice appear in all copies.
+
+THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
diff --git a/libvpx/third_party/x86inc/README.webm b/libvpx/third_party/x86inc/README.webm
new file mode 100644
index 0000000..02cd9ab
--- /dev/null
+++ b/libvpx/third_party/x86inc/README.webm
@@ -0,0 +1,11 @@
+URL: http://git.videolan.org/?p=x264.git
+Version: 999b753ff0f4dc872077f4fa90d465e948cbe656
+License: ISC
+License File: LICENSE
+
+Description:
+x264/libav's framework for x86 assembly. Contains a variety of macros and
+defines that help automatically allow assembly to work cross-platform.
+
+Local Modifications:
+Some modifications to allow PIC to work with x86inc.
diff --git a/libvpx/third_party/x86inc/x86inc.asm b/libvpx/third_party/x86inc/x86inc.asm
new file mode 100644
index 0000000..a66a96b
--- /dev/null
+++ b/libvpx/third_party/x86inc/x86inc.asm
@@ -0,0 +1,1125 @@
+;*****************************************************************************
+;* x86inc.asm: x264asm abstraction layer
+;*****************************************************************************
+;* Copyright (C) 2005-2012 x264 project
+;*
+;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;*          Anton Mitrofanov <BugMaster@narod.ru>
+;*          Jason Garrett-Glaser <darkshikari@gmail.com>
+;*          Henrik Gramner <hengar-6@student.ltu.se>
+;*
+;* Permission to use, copy, modify, and/or distribute this software for any
+;* purpose with or without fee is hereby granted, provided that the above
+;* copyright notice and this permission notice appear in all copies.
+;*
+;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+;*****************************************************************************
+
+; This is a header file for the x264ASM assembly language, which uses
+; NASM/YASM syntax combined with a large number of macros to provide easy
+; abstraction between different calling conventions (x86_32, win64, linux64).
+; It also has various other useful features to simplify writing the kind of
+; DSP functions that are most often used in x264.
+
+; Unlike the rest of x264, this file is available under an ISC license, as it
+; has significant usefulness outside of x264 and we want it to be available
+; to the largest audience possible.  Of course, if you modify it for your own
+; purposes to add a new feature, we strongly encourage contributing a patch
+; as this feature might be useful for others as well.  Send patches or ideas
+; to x264-devel@videolan.org .
+
+%include "vpx_config.asm"
+
+%define program_name vp9
+
+
+%define UNIX64 0
+%define WIN64  0
+%if ARCH_X86_64
+    %ifidn __OUTPUT_FORMAT__,win32
+        %define WIN64  1
+    %elifidn __OUTPUT_FORMAT__,win64
+        %define WIN64  1
+    %elifidn __OUTPUT_FORMAT__,x64
+        %define WIN64  1
+    %else
+        %define UNIX64 1
+    %endif
+%endif
+
+%ifidn   __OUTPUT_FORMAT__,elf32
+    %define mangle(x) x
+%elifidn __OUTPUT_FORMAT__,elf64
+    %define mangle(x) x
+%elifidn __OUTPUT_FORMAT__,elf
+    %define mangle(x) x
+%elifidn __OUTPUT_FORMAT__,x64
+    %define mangle(x) x
+%elifidn __OUTPUT_FORMAT__,win64
+    %define mangle(x) x
+%else
+    %define mangle(x) _ %+ x
+%endif
+
+; FIXME: All of the 64bit asm functions that take a stride as an argument
+; via register, assume that the high dword of that register is filled with 0.
+; This is true in practice (since we never do any 64bit arithmetic on strides,
+; and x264's strides are all positive), but is not guaranteed by the ABI.
+
+; Name of the .rodata section.
+; Kludge: Something on OS X fails to align .rodata even given an align attribute,
+; so use a different read-only section.
+%macro SECTION_RODATA 0-1 16
+    %ifidn __OUTPUT_FORMAT__,macho64
+        SECTION .text align=%1
+    %elifidn __OUTPUT_FORMAT__,macho
+        SECTION .text align=%1
+        fakegot:
+    %elifidn __OUTPUT_FORMAT__,aout
+        section .text
+    %else
+        SECTION .rodata align=%1
+    %endif
+%endmacro
+
+; aout does not support align=
+%macro SECTION_TEXT 0-1 16
+    %ifidn __OUTPUT_FORMAT__,aout
+        SECTION .text
+    %else
+        SECTION .text align=%1
+    %endif
+%endmacro
+
+%if WIN64
+    %define PIC
+%elifidn __OUTPUT_FORMAT__,macho64
+    %define PIC
+%elif ARCH_X86_64 == 0
+; x86_32 doesn't require PIC.
+; Some distros prefer shared objects to be PIC, but nothing breaks if
+; the code contains a few textrels, so we'll skip that complexity.
+    %undef PIC
+%elif CONFIG_PIC
+    %define PIC
+%endif
+%ifdef PIC
+    default rel
+%endif
+
+; Always use long nops (reduces 0x90 spam in disassembly on x86_32)
+%ifndef __NASM_VER__
+CPU amdnop
+%else
+%use smartalign
+ALIGNMODE k7
+%endif
+
+; Macros to eliminate most code duplication between x86_32 and x86_64:
+; Currently this works only for leaf functions which load all their arguments
+; into registers at the start, and make no other use of the stack. Luckily that
+; covers most of x264's asm.
+
+; PROLOGUE:
+; %1 = number of arguments. loads them from stack if needed.
+; %2 = number of registers used. pushes callee-saved regs if needed.
+; %3 = number of xmm registers used. pushes callee-saved xmm regs if needed.
+; %4 = list of names to define to registers
+; PROLOGUE can also be invoked by adding the same options to cglobal
+
+; e.g.
+; cglobal foo, 2,3,0, dst, src, tmp
+; declares a function (foo), taking two args (dst and src) and one local variable (tmp)
+
+; TODO Some functions can use some args directly from the stack. If they're the
+; last args then you can just not declare them, but if they're in the middle
+; we need more flexible macro.
+
+; RET:
+; Pops anything that was pushed by PROLOGUE, and returns.
+
+; REP_RET:
+; Same, but if it doesn't pop anything it becomes a 2-byte ret, for athlons
+; which are slow when a normal ret follows a branch.
+
+; registers:
+; rN and rNq are the native-size register holding function argument N
+; rNd, rNw, rNb are dword, word, and byte size
+; rNm is the original location of arg N (a register or on the stack), dword
+; rNmp is native size
+
+%macro DECLARE_REG 5-6
+    %define r%1q %2
+    %define r%1d %3
+    %define r%1w %4
+    %define r%1b %5
+    %if %0 == 5
+        %define r%1m  %3
+        %define r%1mp %2
+    %elif ARCH_X86_64 ; memory
+        %define r%1m [rsp + stack_offset + %6]
+        %define r%1mp qword r %+ %1m
+    %else
+        %define r%1m [esp + stack_offset + %6]
+        %define r%1mp dword r %+ %1m
+    %endif
+    %define r%1  %2
+%endmacro
+
+%macro DECLARE_REG_SIZE 2
+    %define r%1q r%1
+    %define e%1q r%1
+    %define r%1d e%1
+    %define e%1d e%1
+    %define r%1w %1
+    %define e%1w %1
+    %define r%1b %2
+    %define e%1b %2
+%if ARCH_X86_64 == 0
+    %define r%1  e%1
+%endif
+%endmacro
+
+DECLARE_REG_SIZE ax, al
+DECLARE_REG_SIZE bx, bl
+DECLARE_REG_SIZE cx, cl
+DECLARE_REG_SIZE dx, dl
+DECLARE_REG_SIZE si, sil
+DECLARE_REG_SIZE di, dil
+DECLARE_REG_SIZE bp, bpl
+
+; t# defines for when per-arch register allocation is more complex than just function arguments
+
+%macro DECLARE_REG_TMP 1-*
+    %assign %%i 0
+    %rep %0
+        CAT_XDEFINE t, %%i, r%1
+        %assign %%i %%i+1
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro DECLARE_REG_TMP_SIZE 0-*
+    %rep %0
+        %define t%1q t%1 %+ q
+        %define t%1d t%1 %+ d
+        %define t%1w t%1 %+ w
+        %define t%1b t%1 %+ b
+        %rotate 1
+    %endrep
+%endmacro
+
+DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14
+
+%if ARCH_X86_64
+    %define gprsize 8
+%else
+    %define gprsize 4
+%endif
+
+%macro PUSH 1
+    push %1
+    %assign stack_offset stack_offset+gprsize
+%endmacro
+
+%macro POP 1
+    pop %1
+    %assign stack_offset stack_offset-gprsize
+%endmacro
+
+%macro PUSH_IF_USED 1-*
+    %rep %0
+        %if %1 < regs_used
+            PUSH r%1
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro POP_IF_USED 1-*
+    %rep %0
+        %if %1 < regs_used
+            pop r%1
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro LOAD_IF_USED 1-*
+    %rep %0
+        %if %1 < num_args
+            mov r%1, r %+ %1 %+ mp
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro SUB 2
+    sub %1, %2
+    %ifidn %1, rsp
+        %assign stack_offset stack_offset+(%2)
+    %endif
+%endmacro
+
+%macro ADD 2
+    add %1, %2
+    %ifidn %1, rsp
+        %assign stack_offset stack_offset-(%2)
+    %endif
+%endmacro
+
+%macro movifnidn 2
+    %ifnidn %1, %2
+        mov %1, %2
+    %endif
+%endmacro
+
+%macro movsxdifnidn 2
+    %ifnidn %1, %2
+        movsxd %1, %2
+    %endif
+%endmacro
+
+%macro ASSERT 1
+    %if (%1) == 0
+        %error assert failed
+    %endif
+%endmacro
+
+%macro DEFINE_ARGS 0-*
+    %ifdef n_arg_names
+        %assign %%i 0
+        %rep n_arg_names
+            CAT_UNDEF arg_name %+ %%i, q
+            CAT_UNDEF arg_name %+ %%i, d
+            CAT_UNDEF arg_name %+ %%i, w
+            CAT_UNDEF arg_name %+ %%i, b
+            CAT_UNDEF arg_name %+ %%i, m
+            CAT_UNDEF arg_name %+ %%i, mp
+            CAT_UNDEF arg_name, %%i
+            %assign %%i %%i+1
+        %endrep
+    %endif
+
+    %xdefine %%stack_offset stack_offset
+    %undef stack_offset ; so that the current value of stack_offset doesn't get baked in by xdefine
+    %assign %%i 0
+    %rep %0
+        %xdefine %1q r %+ %%i %+ q
+        %xdefine %1d r %+ %%i %+ d
+        %xdefine %1w r %+ %%i %+ w
+        %xdefine %1b r %+ %%i %+ b
+        %xdefine %1m r %+ %%i %+ m
+        %xdefine %1mp r %+ %%i %+ mp
+        CAT_XDEFINE arg_name, %%i, %1
+        %assign %%i %%i+1
+        %rotate 1
+    %endrep
+    %xdefine stack_offset %%stack_offset
+    %assign n_arg_names %0
+%endmacro
+
+%if WIN64 ; Windows x64 ;=================================================
+
+DECLARE_REG 0,  rcx, ecx,  cx,   cl
+DECLARE_REG 1,  rdx, edx,  dx,   dl
+DECLARE_REG 2,  R8,  R8D,  R8W,  R8B
+DECLARE_REG 3,  R9,  R9D,  R9W,  R9B
+DECLARE_REG 4,  R10, R10D, R10W, R10B, 40
+DECLARE_REG 5,  R11, R11D, R11W, R11B, 48
+DECLARE_REG 6,  rax, eax,  ax,   al,   56
+DECLARE_REG 7,  rdi, edi,  di,   dil,  64
+DECLARE_REG 8,  rsi, esi,  si,   sil,  72
+DECLARE_REG 9,  rbx, ebx,  bx,   bl,   80
+DECLARE_REG 10, rbp, ebp,  bp,   bpl,  88
+DECLARE_REG 11, R12, R12D, R12W, R12B, 96
+DECLARE_REG 12, R13, R13D, R13W, R13B, 104
+DECLARE_REG 13, R14, R14D, R14W, R14B, 112
+DECLARE_REG 14, R15, R15D, R15W, R15B, 120
+
+%macro PROLOGUE 2-4+ 0 ; #args, #regs, #xmm_regs, arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    ASSERT regs_used >= num_args
+    ASSERT regs_used <= 15
+    PUSH_IF_USED 7, 8, 9, 10, 11, 12, 13, 14
+    %if mmsize == 8
+        %assign xmm_regs_used 0
+    %else
+        WIN64_SPILL_XMM %3
+    %endif
+    LOAD_IF_USED 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14
+    DEFINE_ARGS %4
+%endmacro
+
+%macro WIN64_SPILL_XMM 1
+    %assign xmm_regs_used %1
+    ASSERT xmm_regs_used <= 16
+    %if xmm_regs_used > 6
+        SUB rsp, (xmm_regs_used-6)*16+16
+        %assign %%i xmm_regs_used
+        %rep (xmm_regs_used-6)
+            %assign %%i %%i-1
+            movdqa [rsp + (%%i-6)*16+(~stack_offset&8)], xmm %+ %%i
+        %endrep
+    %endif
+%endmacro
+
+%macro WIN64_RESTORE_XMM_INTERNAL 1
+    %if xmm_regs_used > 6
+        %assign %%i xmm_regs_used
+        %rep (xmm_regs_used-6)
+            %assign %%i %%i-1
+            movdqa xmm %+ %%i, [%1 + (%%i-6)*16+(~stack_offset&8)]
+        %endrep
+        add %1, (xmm_regs_used-6)*16+16
+    %endif
+%endmacro
+
+%macro WIN64_RESTORE_XMM 1
+    WIN64_RESTORE_XMM_INTERNAL %1
+    %assign stack_offset stack_offset-(xmm_regs_used-6)*16+16
+    %assign xmm_regs_used 0
+%endmacro
+
+%macro RET 0
+    WIN64_RESTORE_XMM_INTERNAL rsp
+    POP_IF_USED 14, 13, 12, 11, 10, 9, 8, 7
+    ret
+%endmacro
+
+%macro REP_RET 0
+    %if regs_used > 7 || xmm_regs_used > 6
+        RET
+    %else
+        rep ret
+    %endif
+%endmacro
+
+%elif ARCH_X86_64 ; *nix x64 ;=============================================
+
+DECLARE_REG 0,  rdi, edi,  di,   dil
+DECLARE_REG 1,  rsi, esi,  si,   sil
+DECLARE_REG 2,  rdx, edx,  dx,   dl
+DECLARE_REG 3,  rcx, ecx,  cx,   cl
+DECLARE_REG 4,  R8,  R8D,  R8W,  R8B
+DECLARE_REG 5,  R9,  R9D,  R9W,  R9B
+DECLARE_REG 6,  rax, eax,  ax,   al,   8
+DECLARE_REG 7,  R10, R10D, R10W, R10B, 16
+DECLARE_REG 8,  R11, R11D, R11W, R11B, 24
+DECLARE_REG 9,  rbx, ebx,  bx,   bl,   32
+DECLARE_REG 10, rbp, ebp,  bp,   bpl,  40
+DECLARE_REG 11, R12, R12D, R12W, R12B, 48
+DECLARE_REG 12, R13, R13D, R13W, R13B, 56
+DECLARE_REG 13, R14, R14D, R14W, R14B, 64
+DECLARE_REG 14, R15, R15D, R15W, R15B, 72
+
+%macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    ASSERT regs_used >= num_args
+    ASSERT regs_used <= 15
+    PUSH_IF_USED 9, 10, 11, 12, 13, 14
+    LOAD_IF_USED 6, 7, 8, 9, 10, 11, 12, 13, 14
+    DEFINE_ARGS %4
+%endmacro
+
+%macro RET 0
+    POP_IF_USED 14, 13, 12, 11, 10, 9
+    ret
+%endmacro
+
+%macro REP_RET 0
+    %if regs_used > 9
+        RET
+    %else
+        rep ret
+    %endif
+%endmacro
+
+%else ; X86_32 ;==============================================================
+
+DECLARE_REG 0, eax, eax, ax, al,   4
+DECLARE_REG 1, ecx, ecx, cx, cl,   8
+DECLARE_REG 2, edx, edx, dx, dl,   12
+DECLARE_REG 3, ebx, ebx, bx, bl,   16
+DECLARE_REG 4, esi, esi, si, null, 20
+DECLARE_REG 5, edi, edi, di, null, 24
+DECLARE_REG 6, ebp, ebp, bp, null, 28
+%define rsp esp
+
+%macro DECLARE_ARG 1-*
+    %rep %0
+        %define r%1m [esp + stack_offset + 4*%1 + 4]
+        %define r%1mp dword r%1m
+        %rotate 1
+    %endrep
+%endmacro
+
+DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
+
+%macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    %if regs_used > 7
+        %assign regs_used 7
+    %endif
+    ASSERT regs_used >= num_args
+    PUSH_IF_USED 3, 4, 5, 6
+    LOAD_IF_USED 0, 1, 2, 3, 4, 5, 6
+    DEFINE_ARGS %4
+%endmacro
+
+%macro RET 0
+    POP_IF_USED 6, 5, 4, 3
+    ret
+%endmacro
+
+%macro REP_RET 0
+    %if regs_used > 3
+        RET
+    %else
+        rep ret
+    %endif
+%endmacro
+
+%endif ;======================================================================
+
+%if WIN64 == 0
+%macro WIN64_SPILL_XMM 1
+%endmacro
+%macro WIN64_RESTORE_XMM 1
+%endmacro
+%endif
+
+;=============================================================================
+; arch-independent part
+;=============================================================================
+
+%assign function_align 16
+
+; Begin a function.
+; Applies any symbol mangling needed for C linkage, and sets up a define such that
+; subsequent uses of the function name automatically refer to the mangled version.
+; Appends cpuflags to the function name if cpuflags has been specified.
+%macro cglobal 1-2+ ; name, [PROLOGUE args]
+%if %0 == 1
+    cglobal_internal %1 %+ SUFFIX
+%else
+    cglobal_internal %1 %+ SUFFIX, %2
+%endif
+%endmacro
+%macro cglobal_internal 1-2+
+    %ifndef cglobaled_%1
+        %xdefine %1 mangle(program_name %+ _ %+ %1)
+        %xdefine %1.skip_prologue %1 %+ .skip_prologue
+        CAT_XDEFINE cglobaled_, %1, 1
+    %endif
+    %xdefine current_function %1
+    %ifidn __OUTPUT_FORMAT__,elf
+        global %1:function hidden
+    %elifidn __OUTPUT_FORMAT__,elf32
+        global %1:function hidden
+    %elifidn __OUTPUT_FORMAT__,elf64
+        global %1:function hidden
+    %else
+        global %1
+    %endif
+    align function_align
+    %1:
+    RESET_MM_PERMUTATION ; not really needed, but makes disassembly somewhat nicer
+    %assign stack_offset 0
+    %if %0 > 1
+        PROLOGUE %2
+    %endif
+%endmacro
+
+%macro cextern 1
+    %xdefine %1 mangle(program_name %+ _ %+ %1)
+    CAT_XDEFINE cglobaled_, %1, 1
+    extern %1
+%endmacro
+
+; like cextern, but without the prefix
+%macro cextern_naked 1
+    %xdefine %1 mangle(%1)
+    CAT_XDEFINE cglobaled_, %1, 1
+    extern %1
+%endmacro
+
+%macro const 2+
+    %xdefine %1 mangle(program_name %+ _ %+ %1)
+    global %1
+    %1: %2
+%endmacro
+
+; This is needed for ELF, otherwise the GNU linker assumes the stack is
+; executable by default.
+%ifidn __OUTPUT_FORMAT__,elf
+SECTION .note.GNU-stack noalloc noexec nowrite progbits
+%elifidn __OUTPUT_FORMAT__,elf32
+SECTION .note.GNU-stack noalloc noexec nowrite progbits
+%elifidn __OUTPUT_FORMAT__,elf64
+SECTION .note.GNU-stack noalloc noexec nowrite progbits
+%endif
+
+; cpuflags
+
+%assign cpuflags_mmx      (1<<0)
+%assign cpuflags_mmx2     (1<<1) | cpuflags_mmx
+%assign cpuflags_3dnow    (1<<2) | cpuflags_mmx
+%assign cpuflags_3dnow2   (1<<3) | cpuflags_3dnow
+%assign cpuflags_sse      (1<<4) | cpuflags_mmx2
+%assign cpuflags_sse2     (1<<5) | cpuflags_sse
+%assign cpuflags_sse2slow (1<<6) | cpuflags_sse2
+%assign cpuflags_sse3     (1<<7) | cpuflags_sse2
+%assign cpuflags_ssse3    (1<<8) | cpuflags_sse3
+%assign cpuflags_sse4     (1<<9) | cpuflags_ssse3
+%assign cpuflags_sse42    (1<<10)| cpuflags_sse4
+%assign cpuflags_avx      (1<<11)| cpuflags_sse42
+%assign cpuflags_xop      (1<<12)| cpuflags_avx
+%assign cpuflags_fma4     (1<<13)| cpuflags_avx
+
+%assign cpuflags_cache32  (1<<16)
+%assign cpuflags_cache64  (1<<17)
+%assign cpuflags_slowctz  (1<<18)
+%assign cpuflags_lzcnt    (1<<19)
+%assign cpuflags_misalign (1<<20)
+%assign cpuflags_aligned  (1<<21) ; not a cpu feature, but a function variant
+%assign cpuflags_atom     (1<<22)
+
+%define    cpuflag(x) ((cpuflags & (cpuflags_ %+ x)) == (cpuflags_ %+ x))
+%define notcpuflag(x) ((cpuflags & (cpuflags_ %+ x)) != (cpuflags_ %+ x))
+
+; Takes up to 2 cpuflags from the above list.
+; All subsequent functions (up to the next INIT_CPUFLAGS) is built for the specified cpu.
+; You shouldn't need to invoke this macro directly, it's a subroutine for INIT_MMX &co.
+%macro INIT_CPUFLAGS 0-2
+    %if %0 >= 1
+        %xdefine cpuname %1
+        %assign cpuflags cpuflags_%1
+        %if %0 >= 2
+            %xdefine cpuname %1_%2
+            %assign cpuflags cpuflags | cpuflags_%2
+        %endif
+        %xdefine SUFFIX _ %+ cpuname
+        %if cpuflag(avx)
+            %assign avx_enabled 1
+        %endif
+        %if mmsize == 16 && notcpuflag(sse2)
+            %define mova movaps
+            %define movu movups
+            %define movnta movntps
+        %endif
+        %if cpuflag(aligned)
+            %define movu mova
+        %elifidn %1, sse3
+            %define movu lddqu
+        %endif
+    %else
+        %xdefine SUFFIX
+        %undef cpuname
+        %undef cpuflags
+    %endif
+%endmacro
+
+; merge mmx and sse*
+
+%macro CAT_XDEFINE 3
+    %xdefine %1%2 %3
+%endmacro
+
+%macro CAT_UNDEF 2
+    %undef %1%2
+%endmacro
+
+%macro INIT_MMX 0-1+
+    %assign avx_enabled 0
+    %define RESET_MM_PERMUTATION INIT_MMX %1
+    %define mmsize 8
+    %define num_mmregs 8
+    %define mova movq
+    %define movu movq
+    %define movh movd
+    %define movnta movntq
+    %assign %%i 0
+    %rep 8
+    CAT_XDEFINE m, %%i, mm %+ %%i
+    CAT_XDEFINE nmm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    %rep 8
+    CAT_UNDEF m, %%i
+    CAT_UNDEF nmm, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+%macro INIT_XMM 0-1+
+    %assign avx_enabled 0
+    %define RESET_MM_PERMUTATION INIT_XMM %1
+    %define mmsize 16
+    %define num_mmregs 8
+    %if ARCH_X86_64
+    %define num_mmregs 16
+    %endif
+    %define mova movdqa
+    %define movu movdqu
+    %define movh movq
+    %define movnta movntdq
+    %assign %%i 0
+    %rep num_mmregs
+    CAT_XDEFINE m, %%i, xmm %+ %%i
+    CAT_XDEFINE nxmm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+; FIXME: INIT_AVX can be replaced by INIT_XMM avx
+%macro INIT_AVX 0
+    INIT_XMM
+    %assign avx_enabled 1
+    %define PALIGNR PALIGNR_SSSE3
+    %define RESET_MM_PERMUTATION INIT_AVX
+%endmacro
+
+%macro INIT_YMM 0-1+
+    %assign avx_enabled 1
+    %define RESET_MM_PERMUTATION INIT_YMM %1
+    %define mmsize 32
+    %define num_mmregs 8
+    %if ARCH_X86_64
+    %define num_mmregs 16
+    %endif
+    %define mova vmovaps
+    %define movu vmovups
+    %undef movh
+    %define movnta vmovntps
+    %assign %%i 0
+    %rep num_mmregs
+    CAT_XDEFINE m, %%i, ymm %+ %%i
+    CAT_XDEFINE nymm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+INIT_XMM
+
+; I often want to use macros that permute their arguments. e.g. there's no
+; efficient way to implement butterfly or transpose or dct without swapping some
+; arguments.
+;
+; I would like to not have to manually keep track of the permutations:
+; If I insert a permutation in the middle of a function, it should automatically
+; change everything that follows. For more complex macros I may also have multiple
+; implementations, e.g. the SSE2 and SSSE3 versions may have different permutations.
+;
+; Hence these macros. Insert a PERMUTE or some SWAPs at the end of a macro that
+; permutes its arguments. It's equivalent to exchanging the contents of the
+; registers, except that this way you exchange the register names instead, so it
+; doesn't cost any cycles.
+
+%macro PERMUTE 2-* ; takes a list of pairs to swap
+%rep %0/2
+    %xdefine tmp%2 m%2
+    %xdefine ntmp%2 nm%2
+    %rotate 2
+%endrep
+%rep %0/2
+    %xdefine m%1 tmp%2
+    %xdefine nm%1 ntmp%2
+    %undef tmp%2
+    %undef ntmp%2
+    %rotate 2
+%endrep
+%endmacro
+
+%macro SWAP 2-* ; swaps a single chain (sometimes more concise than pairs)
+%rep %0-1
+%ifdef m%1
+    %xdefine tmp m%1
+    %xdefine m%1 m%2
+    %xdefine m%2 tmp
+    CAT_XDEFINE n, m%1, %1
+    CAT_XDEFINE n, m%2, %2
+%else
+    ; If we were called as "SWAP m0,m1" rather than "SWAP 0,1" infer the original numbers here.
+    ; Be careful using this mode in nested macros though, as in some cases there may be
+    ; other copies of m# that have already been dereferenced and don't get updated correctly.
+    %xdefine %%n1 n %+ %1
+    %xdefine %%n2 n %+ %2
+    %xdefine tmp m %+ %%n1
+    CAT_XDEFINE m, %%n1, m %+ %%n2
+    CAT_XDEFINE m, %%n2, tmp
+    CAT_XDEFINE n, m %+ %%n1, %%n1
+    CAT_XDEFINE n, m %+ %%n2, %%n2
+%endif
+    %undef tmp
+    %rotate 1
+%endrep
+%endmacro
+
+; If SAVE_MM_PERMUTATION is placed at the end of a function, then any later
+; calls to that function will automatically load the permutation, so values can
+; be returned in mmregs.
+%macro SAVE_MM_PERMUTATION 0-1
+    %if %0
+        %xdefine %%f %1_m
+    %else
+        %xdefine %%f current_function %+ _m
+    %endif
+    %assign %%i 0
+    %rep num_mmregs
+        CAT_XDEFINE %%f, %%i, m %+ %%i
+    %assign %%i %%i+1
+    %endrep
+%endmacro
+
+%macro LOAD_MM_PERMUTATION 1 ; name to load from
+    %ifdef %1_m0
+        %assign %%i 0
+        %rep num_mmregs
+            CAT_XDEFINE m, %%i, %1_m %+ %%i
+            CAT_XDEFINE n, m %+ %%i, %%i
+        %assign %%i %%i+1
+        %endrep
+    %endif
+%endmacro
+
+; Append cpuflags to the callee's name iff the appended name is known and the plain name isn't
+%macro call 1
+    call_internal %1, %1 %+ SUFFIX
+%endmacro
+%macro call_internal 2
+    %xdefine %%i %1
+    %ifndef cglobaled_%1
+        %ifdef cglobaled_%2
+            %xdefine %%i %2
+        %endif
+    %endif
+    call %%i
+    LOAD_MM_PERMUTATION %%i
+%endmacro
+
+; Substitutions that reduce instruction size but are functionally equivalent
+%macro add 2
+    %ifnum %2
+        %if %2==128
+            sub %1, -128
+        %else
+            add %1, %2
+        %endif
+    %else
+        add %1, %2
+    %endif
+%endmacro
+
+%macro sub 2
+    %ifnum %2
+        %if %2==128
+            add %1, -128
+        %else
+            sub %1, %2
+        %endif
+    %else
+        sub %1, %2
+    %endif
+%endmacro
+
+;=============================================================================
+; AVX abstraction layer
+;=============================================================================
+
+%assign i 0
+%rep 16
+    %if i < 8
+        CAT_XDEFINE sizeofmm, i, 8
+    %endif
+    CAT_XDEFINE sizeofxmm, i, 16
+    CAT_XDEFINE sizeofymm, i, 32
+%assign i i+1
+%endrep
+%undef i
+
+;%1 == instruction
+;%2 == 1 if float, 0 if int
+;%3 == 1 if 4-operand (xmm, xmm, xmm, imm), 0 if 2- or 3-operand (xmm, xmm, xmm)
+;%4 == number of operands given
+;%5+: operands
+%macro RUN_AVX_INSTR 6-7+
+    %ifid %5
+        %define %%size sizeof%5
+    %else
+        %define %%size mmsize
+    %endif
+    %if %%size==32
+        %if %0 >= 7
+            v%1 %5, %6, %7
+        %else
+            v%1 %5, %6
+        %endif
+    %else
+        %if %%size==8
+            %define %%regmov movq
+        %elif %2
+            %define %%regmov movaps
+        %else
+            %define %%regmov movdqa
+        %endif
+
+        %if %4>=3+%3
+            %ifnidn %5, %6
+                %if avx_enabled && sizeof%5==16
+                    v%1 %5, %6, %7
+                %else
+                    %%regmov %5, %6
+                    %1 %5, %7
+                %endif
+            %else
+                %1 %5, %7
+            %endif
+        %elif %3
+            %1 %5, %6, %7
+        %else
+            %1 %5, %6
+        %endif
+    %endif
+%endmacro
+
+; 3arg AVX ops with a memory arg can only have it in src2,
+; whereas SSE emulation of 3arg prefers to have it in src1 (i.e. the mov).
+; So, if the op is symmetric and the wrong one is memory, swap them.
+%macro RUN_AVX_INSTR1 8
+    %assign %%swap 0
+    %if avx_enabled
+        %ifnid %6
+            %assign %%swap 1
+        %endif
+    %elifnidn %5, %6
+        %ifnid %7
+            %assign %%swap 1
+        %endif
+    %endif
+    %if %%swap && %3 == 0 && %8 == 1
+        RUN_AVX_INSTR %1, %2, %3, %4, %5, %7, %6
+    %else
+        RUN_AVX_INSTR %1, %2, %3, %4, %5, %6, %7
+    %endif
+%endmacro
+
+;%1 == instruction
+;%2 == 1 if float, 0 if int
+;%3 == 1 if 4-operand (xmm, xmm, xmm, imm), 0 if 3-operand (xmm, xmm, xmm)
+;%4 == 1 if symmetric (i.e. doesn't matter which src arg is which), 0 if not
+%macro AVX_INSTR 4
+    %macro %1 2-9 fnord, fnord, fnord, %1, %2, %3, %4
+        %ifidn %3, fnord
+            RUN_AVX_INSTR %6, %7, %8, 2, %1, %2
+        %elifidn %4, fnord
+            RUN_AVX_INSTR1 %6, %7, %8, 3, %1, %2, %3, %9
+        %elifidn %5, fnord
+            RUN_AVX_INSTR %6, %7, %8, 4, %1, %2, %3, %4
+        %else
+            RUN_AVX_INSTR %6, %7, %8, 5, %1, %2, %3, %4, %5
+        %endif
+    %endmacro
+%endmacro
+
+AVX_INSTR addpd, 1, 0, 1
+AVX_INSTR addps, 1, 0, 1
+AVX_INSTR addsd, 1, 0, 1
+AVX_INSTR addss, 1, 0, 1
+AVX_INSTR addsubpd, 1, 0, 0
+AVX_INSTR addsubps, 1, 0, 0
+AVX_INSTR andpd, 1, 0, 1
+AVX_INSTR andps, 1, 0, 1
+AVX_INSTR andnpd, 1, 0, 0
+AVX_INSTR andnps, 1, 0, 0
+AVX_INSTR blendpd, 1, 0, 0
+AVX_INSTR blendps, 1, 0, 0
+AVX_INSTR blendvpd, 1, 0, 0
+AVX_INSTR blendvps, 1, 0, 0
+AVX_INSTR cmppd, 1, 0, 0
+AVX_INSTR cmpps, 1, 0, 0
+AVX_INSTR cmpsd, 1, 0, 0
+AVX_INSTR cmpss, 1, 0, 0
+AVX_INSTR cvtdq2ps, 1, 0, 0
+AVX_INSTR cvtps2dq, 1, 0, 0
+AVX_INSTR divpd, 1, 0, 0
+AVX_INSTR divps, 1, 0, 0
+AVX_INSTR divsd, 1, 0, 0
+AVX_INSTR divss, 1, 0, 0
+AVX_INSTR dppd, 1, 1, 0
+AVX_INSTR dpps, 1, 1, 0
+AVX_INSTR haddpd, 1, 0, 0
+AVX_INSTR haddps, 1, 0, 0
+AVX_INSTR hsubpd, 1, 0, 0
+AVX_INSTR hsubps, 1, 0, 0
+AVX_INSTR maxpd, 1, 0, 1
+AVX_INSTR maxps, 1, 0, 1
+AVX_INSTR maxsd, 1, 0, 1
+AVX_INSTR maxss, 1, 0, 1
+AVX_INSTR minpd, 1, 0, 1
+AVX_INSTR minps, 1, 0, 1
+AVX_INSTR minsd, 1, 0, 1
+AVX_INSTR minss, 1, 0, 1
+AVX_INSTR movhlps, 1, 0, 0
+AVX_INSTR movlhps, 1, 0, 0
+AVX_INSTR movsd, 1, 0, 0
+AVX_INSTR movss, 1, 0, 0
+AVX_INSTR mpsadbw, 0, 1, 0
+AVX_INSTR mulpd, 1, 0, 1
+AVX_INSTR mulps, 1, 0, 1
+AVX_INSTR mulsd, 1, 0, 1
+AVX_INSTR mulss, 1, 0, 1
+AVX_INSTR orpd, 1, 0, 1
+AVX_INSTR orps, 1, 0, 1
+AVX_INSTR packsswb, 0, 0, 0
+AVX_INSTR packssdw, 0, 0, 0
+AVX_INSTR packuswb, 0, 0, 0
+AVX_INSTR packusdw, 0, 0, 0
+AVX_INSTR paddb, 0, 0, 1
+AVX_INSTR paddw, 0, 0, 1
+AVX_INSTR paddd, 0, 0, 1
+AVX_INSTR paddq, 0, 0, 1
+AVX_INSTR paddsb, 0, 0, 1
+AVX_INSTR paddsw, 0, 0, 1
+AVX_INSTR paddusb, 0, 0, 1
+AVX_INSTR paddusw, 0, 0, 1
+AVX_INSTR palignr, 0, 1, 0
+AVX_INSTR pand, 0, 0, 1
+AVX_INSTR pandn, 0, 0, 0
+AVX_INSTR pavgb, 0, 0, 1
+AVX_INSTR pavgw, 0, 0, 1
+AVX_INSTR pblendvb, 0, 0, 0
+AVX_INSTR pblendw, 0, 1, 0
+AVX_INSTR pcmpestri, 0, 0, 0
+AVX_INSTR pcmpestrm, 0, 0, 0
+AVX_INSTR pcmpistri, 0, 0, 0
+AVX_INSTR pcmpistrm, 0, 0, 0
+AVX_INSTR pcmpeqb, 0, 0, 1
+AVX_INSTR pcmpeqw, 0, 0, 1
+AVX_INSTR pcmpeqd, 0, 0, 1
+AVX_INSTR pcmpeqq, 0, 0, 1
+AVX_INSTR pcmpgtb, 0, 0, 0
+AVX_INSTR pcmpgtw, 0, 0, 0
+AVX_INSTR pcmpgtd, 0, 0, 0
+AVX_INSTR pcmpgtq, 0, 0, 0
+AVX_INSTR phaddw, 0, 0, 0
+AVX_INSTR phaddd, 0, 0, 0
+AVX_INSTR phaddsw, 0, 0, 0
+AVX_INSTR phsubw, 0, 0, 0
+AVX_INSTR phsubd, 0, 0, 0
+AVX_INSTR phsubsw, 0, 0, 0
+AVX_INSTR pmaddwd, 0, 0, 1
+AVX_INSTR pmaddubsw, 0, 0, 0
+AVX_INSTR pmaxsb, 0, 0, 1
+AVX_INSTR pmaxsw, 0, 0, 1
+AVX_INSTR pmaxsd, 0, 0, 1
+AVX_INSTR pmaxub, 0, 0, 1
+AVX_INSTR pmaxuw, 0, 0, 1
+AVX_INSTR pmaxud, 0, 0, 1
+AVX_INSTR pminsb, 0, 0, 1
+AVX_INSTR pminsw, 0, 0, 1
+AVX_INSTR pminsd, 0, 0, 1
+AVX_INSTR pminub, 0, 0, 1
+AVX_INSTR pminuw, 0, 0, 1
+AVX_INSTR pminud, 0, 0, 1
+AVX_INSTR pmulhuw, 0, 0, 1
+AVX_INSTR pmulhrsw, 0, 0, 1
+AVX_INSTR pmulhw, 0, 0, 1
+AVX_INSTR pmullw, 0, 0, 1
+AVX_INSTR pmulld, 0, 0, 1
+AVX_INSTR pmuludq, 0, 0, 1
+AVX_INSTR pmuldq, 0, 0, 1
+AVX_INSTR por, 0, 0, 1
+AVX_INSTR psadbw, 0, 0, 1
+AVX_INSTR pshufb, 0, 0, 0
+AVX_INSTR psignb, 0, 0, 0
+AVX_INSTR psignw, 0, 0, 0
+AVX_INSTR psignd, 0, 0, 0
+AVX_INSTR psllw, 0, 0, 0
+AVX_INSTR pslld, 0, 0, 0
+AVX_INSTR psllq, 0, 0, 0
+AVX_INSTR pslldq, 0, 0, 0
+AVX_INSTR psraw, 0, 0, 0
+AVX_INSTR psrad, 0, 0, 0
+AVX_INSTR psrlw, 0, 0, 0
+AVX_INSTR psrld, 0, 0, 0
+AVX_INSTR psrlq, 0, 0, 0
+AVX_INSTR psrldq, 0, 0, 0
+AVX_INSTR psubb, 0, 0, 0
+AVX_INSTR psubw, 0, 0, 0
+AVX_INSTR psubd, 0, 0, 0
+AVX_INSTR psubq, 0, 0, 0
+AVX_INSTR psubsb, 0, 0, 0
+AVX_INSTR psubsw, 0, 0, 0
+AVX_INSTR psubusb, 0, 0, 0
+AVX_INSTR psubusw, 0, 0, 0
+AVX_INSTR punpckhbw, 0, 0, 0
+AVX_INSTR punpckhwd, 0, 0, 0
+AVX_INSTR punpckhdq, 0, 0, 0
+AVX_INSTR punpckhqdq, 0, 0, 0
+AVX_INSTR punpcklbw, 0, 0, 0
+AVX_INSTR punpcklwd, 0, 0, 0
+AVX_INSTR punpckldq, 0, 0, 0
+AVX_INSTR punpcklqdq, 0, 0, 0
+AVX_INSTR pxor, 0, 0, 1
+AVX_INSTR shufps, 1, 1, 0
+AVX_INSTR subpd, 1, 0, 0
+AVX_INSTR subps, 1, 0, 0
+AVX_INSTR subsd, 1, 0, 0
+AVX_INSTR subss, 1, 0, 0
+AVX_INSTR unpckhpd, 1, 0, 0
+AVX_INSTR unpckhps, 1, 0, 0
+AVX_INSTR unpcklpd, 1, 0, 0
+AVX_INSTR unpcklps, 1, 0, 0
+AVX_INSTR xorpd, 1, 0, 1
+AVX_INSTR xorps, 1, 0, 1
+
+; 3DNow instructions, for sharing code between AVX, SSE and 3DN
+AVX_INSTR pfadd, 1, 0, 1
+AVX_INSTR pfsub, 1, 0, 0
+AVX_INSTR pfmul, 1, 0, 1
+
+; base-4 constants for shuffles
+%assign i 0
+%rep 256
+    %assign j ((i>>6)&3)*1000 + ((i>>4)&3)*100 + ((i>>2)&3)*10 + (i&3)
+    %if j < 10
+        CAT_XDEFINE q000, j, i
+    %elif j < 100
+        CAT_XDEFINE q00, j, i
+    %elif j < 1000
+        CAT_XDEFINE q0, j, i
+    %else
+        CAT_XDEFINE q, j, i
+    %endif
+%assign i i+1
+%endrep
+%undef i
+%undef j
+
+%macro FMA_INSTR 3
+    %macro %1 4-7 %1, %2, %3
+        %if cpuflag(xop)
+            v%5 %1, %2, %3, %4
+        %else
+            %6 %1, %2, %3
+            %7 %1, %4
+        %endif
+    %endmacro
+%endmacro
+
+FMA_INSTR  pmacsdd,  pmulld, paddd
+FMA_INSTR  pmacsww,  pmullw, paddw
+FMA_INSTR pmadcswd, pmaddwd, paddd
diff --git a/libvpx/tools/all_builds.py b/libvpx/tools/all_builds.py
new file mode 100755
index 0000000..d1f0c80
--- /dev/null
+++ b/libvpx/tools/all_builds.py
@@ -0,0 +1,72 @@
+#!/usr/bin/python
+
+import getopt
+import subprocess
+import sys
+
+LONG_OPTIONS = ["shard=", "shards="]
+BASE_COMMAND = "./configure --enable-internal-stats --enable-experimental"
+
+def RunCommand(command):
+  run = subprocess.Popen(command, shell=True)
+  output = run.communicate()
+  if run.returncode:
+    print "Non-zero return code: " + str(run.returncode) + " => exiting!"
+    sys.exit(1)
+
+def list_of_experiments():
+  experiments = []
+  configure_file = open("configure")
+  list_start = False
+  for line in configure_file.read().split("\n"):
+    if line == 'EXPERIMENT_LIST="':
+      list_start = True
+    elif line == '"':
+      list_start = False
+    elif list_start:
+      currently_broken = ["csm"]
+      experiment = line[4:]
+      if experiment not in currently_broken:
+        experiments.append(experiment)
+  return experiments
+
+def main(argv):
+  # Parse arguments
+  options = {"--shard": 0, "--shards": 1}
+  if "--" in argv:
+    opt_end_index = argv.index("--")
+  else:
+    opt_end_index = len(argv)
+  try:
+    o, _ = getopt.getopt(argv[1:opt_end_index], None, LONG_OPTIONS)
+  except getopt.GetoptError, err:
+    print str(err)
+    print "Usage: %s [--shard=<n> --shards=<n>] -- [configure flag ...]"%argv[0]
+    sys.exit(2)
+
+  options.update(o)
+  extra_args = argv[opt_end_index + 1:]
+
+  # Shard experiment list
+  shard = int(options["--shard"])
+  shards = int(options["--shards"])
+  experiments = list_of_experiments()
+  base_command = " ".join([BASE_COMMAND] + extra_args)
+  configs = [base_command]
+  configs += ["%s --enable-%s" % (base_command, e) for e in experiments]
+  my_configs = zip(configs, range(len(configs)))
+  my_configs = filter(lambda x: x[1] % shards == shard, my_configs)
+  my_configs = [e[0] for e in my_configs]
+
+  # Run configs for this shard
+  for config in my_configs:
+    test_build(config)
+
+def test_build(configure_command):
+  print "\033[34m\033[47mTesting %s\033[0m" % (configure_command)
+  RunCommand(configure_command)
+  RunCommand("make clean")
+  RunCommand("make")
+
+if __name__ == "__main__":
+  main(sys.argv)
diff --git a/libvpx/tools/cpplint.py b/libvpx/tools/cpplint.py
new file mode 100755
index 0000000..159dbbb
--- /dev/null
+++ b/libvpx/tools/cpplint.py
@@ -0,0 +1,4020 @@
+#!/usr/bin/python
+#
+# Copyright (c) 2009 Google Inc. All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#    * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#    * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#    * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# Here are some issues that I've had people identify in my code during reviews,
+# that I think are possible to flag automatically in a lint tool.  If these were
+# caught by lint, it would save time both for myself and that of my reviewers.
+# Most likely, some of these are beyond the scope of the current lint framework,
+# but I think it is valuable to retain these wish-list items even if they cannot
+# be immediately implemented.
+#
+#  Suggestions
+#  -----------
+#  - Check for no 'explicit' for multi-arg ctor
+#  - Check for boolean assign RHS in parens
+#  - Check for ctor initializer-list colon position and spacing
+#  - Check that if there's a ctor, there should be a dtor
+#  - Check accessors that return non-pointer member variables are
+#    declared const
+#  - Check accessors that return non-const pointer member vars are
+#    *not* declared const
+#  - Check for using public includes for testing
+#  - Check for spaces between brackets in one-line inline method
+#  - Check for no assert()
+#  - Check for spaces surrounding operators
+#  - Check for 0 in pointer context (should be NULL)
+#  - Check for 0 in char context (should be '\0')
+#  - Check for camel-case method name conventions for methods
+#    that are not simple inline getters and setters
+#  - Do not indent namespace contents
+#  - Avoid inlining non-trivial constructors in header files
+#  - Check for old-school (void) cast for call-sites of functions
+#    ignored return value
+#  - Check gUnit usage of anonymous namespace
+#  - Check for class declaration order (typedefs, consts, enums,
+#    ctor(s?), dtor, friend declarations, methods, member vars)
+#
+
+"""Does google-lint on c++ files.
+
+The goal of this script is to identify places in the code that *may*
+be in non-compliance with google style.  It does not attempt to fix
+up these problems -- the point is to educate.  It does also not
+attempt to find all problems, or to ensure that everything it does
+find is legitimately a problem.
+
+In particular, we can get very confused by /* and // inside strings!
+We do a small hack, which is to ignore //'s with "'s after them on the
+same line, but it is far from perfect (in either direction).
+"""
+
+import codecs
+import copy
+import getopt
+import math  # for log
+import os
+import re
+import sre_compile
+import string
+import sys
+import unicodedata
+
+
+_USAGE = """
+Syntax: cpplint.py [--verbose=#] [--output=vs7] [--filter=-x,+y,...]
+                   [--counting=total|toplevel|detailed]
+        <file> [file] ...
+
+  The style guidelines this tries to follow are those in
+    http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml
+
+  Every problem is given a confidence score from 1-5, with 5 meaning we are
+  certain of the problem, and 1 meaning it could be a legitimate construct.
+  This will miss some errors, and is not a substitute for a code review.
+
+  To suppress false-positive errors of a certain category, add a
+  'NOLINT(category)' comment to the line.  NOLINT or NOLINT(*)
+  suppresses errors of all categories on that line.
+
+  The files passed in will be linted; at least one file must be provided.
+  Linted extensions are .cc, .cpp, and .h.  Other file types will be ignored.
+
+  Flags:
+
+    output=vs7
+      By default, the output is formatted to ease emacs parsing.  Visual Studio
+      compatible output (vs7) may also be used.  Other formats are unsupported.
+
+    verbose=#
+      Specify a number 0-5 to restrict errors to certain verbosity levels.
+
+    filter=-x,+y,...
+      Specify a comma-separated list of category-filters to apply: only
+      error messages whose category names pass the filters will be printed.
+      (Category names are printed with the message and look like
+      "[whitespace/indent]".)  Filters are evaluated left to right.
+      "-FOO" and "FOO" means "do not print categories that start with FOO".
+      "+FOO" means "do print categories that start with FOO".
+
+      Examples: --filter=-whitespace,+whitespace/braces
+                --filter=whitespace,runtime/printf,+runtime/printf_format
+                --filter=-,+build/include_what_you_use
+
+      To see a list of all the categories used in cpplint, pass no arg:
+         --filter=
+
+    counting=total|toplevel|detailed
+      The total number of errors found is always printed. If
+      'toplevel' is provided, then the count of errors in each of
+      the top-level categories like 'build' and 'whitespace' will
+      also be printed. If 'detailed' is provided, then a count
+      is provided for each category like 'build/class'.
+
+    root=subdir
+      The root directory used for deriving header guard CPP variable.
+      By default, the header guard CPP variable is calculated as the relative
+      path to the directory that contains .git, .hg, or .svn.  When this flag
+      is specified, the relative path is calculated from the specified
+      directory. If the specified directory does not exist, this flag is
+      ignored.
+
+      Examples:
+        Assuing that src/.git exists, the header guard CPP variables for
+        src/chrome/browser/ui/browser.h are:
+
+        No flag => CHROME_BROWSER_UI_BROWSER_H_
+        --root=chrome => BROWSER_UI_BROWSER_H_
+        --root=chrome/browser => UI_BROWSER_H_
+"""
+
+# We categorize each error message we print.  Here are the categories.
+# We want an explicit list so we can list them all in cpplint --filter=.
+# If you add a new error message with a new category, add it to the list
+# here!  cpplint_unittest.py should tell you if you forget to do this.
+# \ used for clearer layout -- pylint: disable-msg=C6013
+_ERROR_CATEGORIES = [
+  'build/class',
+  'build/deprecated',
+  'build/endif_comment',
+  'build/explicit_make_pair',
+  'build/forward_decl',
+  'build/header_guard',
+  'build/include',
+  'build/include_alpha',
+  'build/include_order',
+  'build/include_what_you_use',
+  'build/namespaces',
+  'build/printf_format',
+  'build/storage_class',
+  'legal/copyright',
+  'readability/alt_tokens',
+  'readability/braces',
+  'readability/casting',
+  'readability/check',
+  'readability/constructors',
+  'readability/fn_size',
+  'readability/function',
+  'readability/multiline_comment',
+  'readability/multiline_string',
+  'readability/namespace',
+  'readability/nolint',
+  'readability/streams',
+  'readability/todo',
+  'readability/utf8',
+  'runtime/arrays',
+  'runtime/casting',
+  'runtime/explicit',
+  'runtime/int',
+  'runtime/init',
+  'runtime/invalid_increment',
+  'runtime/member_string_references',
+  'runtime/memset',
+  'runtime/operator',
+  'runtime/printf',
+  'runtime/printf_format',
+  'runtime/references',
+  'runtime/rtti',
+  'runtime/sizeof',
+  'runtime/string',
+  'runtime/threadsafe_fn',
+  'whitespace/blank_line',
+  'whitespace/braces',
+  'whitespace/comma',
+  'whitespace/comments',
+  'whitespace/empty_loop_body',
+  'whitespace/end_of_line',
+  'whitespace/ending_newline',
+  'whitespace/forcolon',
+  'whitespace/indent',
+  'whitespace/labels',
+  'whitespace/line_length',
+  'whitespace/newline',
+  'whitespace/operators',
+  'whitespace/parens',
+  'whitespace/semicolon',
+  'whitespace/tab',
+  'whitespace/todo'
+  ]
+
+# The default state of the category filter. This is overrided by the --filter=
+# flag. By default all errors are on, so only add here categories that should be
+# off by default (i.e., categories that must be enabled by the --filter= flags).
+# All entries here should start with a '-' or '+', as in the --filter= flag.
+_DEFAULT_FILTERS = ['-build/include_alpha']
+
+# We used to check for high-bit characters, but after much discussion we
+# decided those were OK, as long as they were in UTF-8 and didn't represent
+# hard-coded international strings, which belong in a separate i18n file.
+
+# Headers that we consider STL headers.
+_STL_HEADERS = frozenset([
+    'algobase.h', 'algorithm', 'alloc.h', 'bitset', 'deque', 'exception',
+    'function.h', 'functional', 'hash_map', 'hash_map.h', 'hash_set',
+    'hash_set.h', 'iterator', 'list', 'list.h', 'map', 'memory', 'new',
+    'pair.h', 'pthread_alloc', 'queue', 'set', 'set.h', 'sstream', 'stack',
+    'stl_alloc.h', 'stl_relops.h', 'type_traits.h',
+    'utility', 'vector', 'vector.h',
+    ])
+
+
+# Non-STL C++ system headers.
+_CPP_HEADERS = frozenset([
+    'algo.h', 'builtinbuf.h', 'bvector.h', 'cassert', 'cctype',
+    'cerrno', 'cfloat', 'ciso646', 'climits', 'clocale', 'cmath',
+    'complex', 'complex.h', 'csetjmp', 'csignal', 'cstdarg', 'cstddef',
+    'cstdio', 'cstdlib', 'cstring', 'ctime', 'cwchar', 'cwctype',
+    'defalloc.h', 'deque.h', 'editbuf.h', 'exception', 'fstream',
+    'fstream.h', 'hashtable.h', 'heap.h', 'indstream.h', 'iomanip',
+    'iomanip.h', 'ios', 'iosfwd', 'iostream', 'iostream.h', 'istream',
+    'istream.h', 'iterator.h', 'limits', 'map.h', 'multimap.h', 'multiset.h',
+    'numeric', 'ostream', 'ostream.h', 'parsestream.h', 'pfstream.h',
+    'PlotFile.h', 'procbuf.h', 'pthread_alloc.h', 'rope', 'rope.h',
+    'ropeimpl.h', 'SFile.h', 'slist', 'slist.h', 'stack.h', 'stdexcept',
+    'stdiostream.h', 'streambuf.h', 'stream.h', 'strfile.h', 'string',
+    'strstream', 'strstream.h', 'tempbuf.h', 'tree.h', 'typeinfo', 'valarray',
+    ])
+
+
+# Assertion macros.  These are defined in base/logging.h and
+# testing/base/gunit.h.  Note that the _M versions need to come first
+# for substring matching to work.
+_CHECK_MACROS = [
+    'DCHECK', 'CHECK',
+    'EXPECT_TRUE_M', 'EXPECT_TRUE',
+    'ASSERT_TRUE_M', 'ASSERT_TRUE',
+    'EXPECT_FALSE_M', 'EXPECT_FALSE',
+    'ASSERT_FALSE_M', 'ASSERT_FALSE',
+    ]
+
+# Replacement macros for CHECK/DCHECK/EXPECT_TRUE/EXPECT_FALSE
+_CHECK_REPLACEMENT = dict([(m, {}) for m in _CHECK_MACROS])
+
+for op, replacement in [('==', 'EQ'), ('!=', 'NE'),
+                        ('>=', 'GE'), ('>', 'GT'),
+                        ('<=', 'LE'), ('<', 'LT')]:
+  _CHECK_REPLACEMENT['DCHECK'][op] = 'DCHECK_%s' % replacement
+  _CHECK_REPLACEMENT['CHECK'][op] = 'CHECK_%s' % replacement
+  _CHECK_REPLACEMENT['EXPECT_TRUE'][op] = 'EXPECT_%s' % replacement
+  _CHECK_REPLACEMENT['ASSERT_TRUE'][op] = 'ASSERT_%s' % replacement
+  _CHECK_REPLACEMENT['EXPECT_TRUE_M'][op] = 'EXPECT_%s_M' % replacement
+  _CHECK_REPLACEMENT['ASSERT_TRUE_M'][op] = 'ASSERT_%s_M' % replacement
+
+for op, inv_replacement in [('==', 'NE'), ('!=', 'EQ'),
+                            ('>=', 'LT'), ('>', 'LE'),
+                            ('<=', 'GT'), ('<', 'GE')]:
+  _CHECK_REPLACEMENT['EXPECT_FALSE'][op] = 'EXPECT_%s' % inv_replacement
+  _CHECK_REPLACEMENT['ASSERT_FALSE'][op] = 'ASSERT_%s' % inv_replacement
+  _CHECK_REPLACEMENT['EXPECT_FALSE_M'][op] = 'EXPECT_%s_M' % inv_replacement
+  _CHECK_REPLACEMENT['ASSERT_FALSE_M'][op] = 'ASSERT_%s_M' % inv_replacement
+
+# Alternative tokens and their replacements.  For full list, see section 2.5
+# Alternative tokens [lex.digraph] in the C++ standard.
+#
+# Digraphs (such as '%:') are not included here since it's a mess to
+# match those on a word boundary.
+_ALT_TOKEN_REPLACEMENT = {
+    'and': '&&',
+    'bitor': '|',
+    'or': '||',
+    'xor': '^',
+    'compl': '~',
+    'bitand': '&',
+    'and_eq': '&=',
+    'or_eq': '|=',
+    'xor_eq': '^=',
+    'not': '!',
+    'not_eq': '!='
+    }
+
+# Compile regular expression that matches all the above keywords.  The "[ =()]"
+# bit is meant to avoid matching these keywords outside of boolean expressions.
+#
+# False positives include C-style multi-line comments (http://go/nsiut )
+# and multi-line strings (http://go/beujw ), but those have always been
+# troublesome for cpplint.
+_ALT_TOKEN_REPLACEMENT_PATTERN = re.compile(
+    r'[ =()](' + ('|'.join(_ALT_TOKEN_REPLACEMENT.keys())) + r')(?=[ (]|$)')
+
+
+# These constants define types of headers for use with
+# _IncludeState.CheckNextIncludeOrder().
+_C_SYS_HEADER = 1
+_CPP_SYS_HEADER = 2
+_LIKELY_MY_HEADER = 3
+_POSSIBLE_MY_HEADER = 4
+_OTHER_HEADER = 5
+
+# These constants define the current inline assembly state
+_NO_ASM = 0       # Outside of inline assembly block
+_INSIDE_ASM = 1   # Inside inline assembly block
+_END_ASM = 2      # Last line of inline assembly block
+_BLOCK_ASM = 3    # The whole block is an inline assembly block
+
+# Match start of assembly blocks
+_MATCH_ASM = re.compile(r'^\s*(?:asm|_asm|__asm|__asm__)'
+                        r'(?:\s+(volatile|__volatile__))?'
+                        r'\s*[{(]')
+
+
+_regexp_compile_cache = {}
+
+# Finds occurrences of NOLINT or NOLINT(...).
+_RE_SUPPRESSION = re.compile(r'\bNOLINT\b(\([^)]*\))?')
+
+# {str, set(int)}: a map from error categories to sets of linenumbers
+# on which those errors are expected and should be suppressed.
+_error_suppressions = {}
+
+# The root directory used for deriving header guard CPP variable.
+# This is set by --root flag.
+_root = None
+
+def ParseNolintSuppressions(filename, raw_line, linenum, error):
+  """Updates the global list of error-suppressions.
+
+  Parses any NOLINT comments on the current line, updating the global
+  error_suppressions store.  Reports an error if the NOLINT comment
+  was malformed.
+
+  Args:
+    filename: str, the name of the input file.
+    raw_line: str, the line of input text, with comments.
+    linenum: int, the number of the current line.
+    error: function, an error handler.
+  """
+  # FIXME(adonovan): "NOLINT(" is misparsed as NOLINT(*).
+  matched = _RE_SUPPRESSION.search(raw_line)
+  if matched:
+    category = matched.group(1)
+    if category in (None, '(*)'):  # => "suppress all"
+      _error_suppressions.setdefault(None, set()).add(linenum)
+    else:
+      if category.startswith('(') and category.endswith(')'):
+        category = category[1:-1]
+        if category in _ERROR_CATEGORIES:
+          _error_suppressions.setdefault(category, set()).add(linenum)
+        else:
+          error(filename, linenum, 'readability/nolint', 5,
+                'Unknown NOLINT error category: %s' % category)
+
+
+def ResetNolintSuppressions():
+  "Resets the set of NOLINT suppressions to empty."
+  _error_suppressions.clear()
+
+
+def IsErrorSuppressedByNolint(category, linenum):
+  """Returns true if the specified error category is suppressed on this line.
+
+  Consults the global error_suppressions map populated by
+  ParseNolintSuppressions/ResetNolintSuppressions.
+
+  Args:
+    category: str, the category of the error.
+    linenum: int, the current line number.
+  Returns:
+    bool, True iff the error should be suppressed due to a NOLINT comment.
+  """
+  return (linenum in _error_suppressions.get(category, set()) or
+          linenum in _error_suppressions.get(None, set()))
+
+def Match(pattern, s):
+  """Matches the string with the pattern, caching the compiled regexp."""
+  # The regexp compilation caching is inlined in both Match and Search for
+  # performance reasons; factoring it out into a separate function turns out
+  # to be noticeably expensive.
+  if not pattern in _regexp_compile_cache:
+    _regexp_compile_cache[pattern] = sre_compile.compile(pattern)
+  return _regexp_compile_cache[pattern].match(s)
+
+
+def Search(pattern, s):
+  """Searches the string for the pattern, caching the compiled regexp."""
+  if not pattern in _regexp_compile_cache:
+    _regexp_compile_cache[pattern] = sre_compile.compile(pattern)
+  return _regexp_compile_cache[pattern].search(s)
+
+
+class _IncludeState(dict):
+  """Tracks line numbers for includes, and the order in which includes appear.
+
+  As a dict, an _IncludeState object serves as a mapping between include
+  filename and line number on which that file was included.
+
+  Call CheckNextIncludeOrder() once for each header in the file, passing
+  in the type constants defined above. Calls in an illegal order will
+  raise an _IncludeError with an appropriate error message.
+
+  """
+  # self._section will move monotonically through this set. If it ever
+  # needs to move backwards, CheckNextIncludeOrder will raise an error.
+  _INITIAL_SECTION = 0
+  _MY_H_SECTION = 1
+  _C_SECTION = 2
+  _CPP_SECTION = 3
+  _OTHER_H_SECTION = 4
+
+  _TYPE_NAMES = {
+      _C_SYS_HEADER: 'C system header',
+      _CPP_SYS_HEADER: 'C++ system header',
+      _LIKELY_MY_HEADER: 'header this file implements',
+      _POSSIBLE_MY_HEADER: 'header this file may implement',
+      _OTHER_HEADER: 'other header',
+      }
+  _SECTION_NAMES = {
+      _INITIAL_SECTION: "... nothing. (This can't be an error.)",
+      _MY_H_SECTION: 'a header this file implements',
+      _C_SECTION: 'C system header',
+      _CPP_SECTION: 'C++ system header',
+      _OTHER_H_SECTION: 'other header',
+      }
+
+  def __init__(self):
+    dict.__init__(self)
+    # The name of the current section.
+    self._section = self._INITIAL_SECTION
+    # The path of last found header.
+    self._last_header = ''
+
+  def CanonicalizeAlphabeticalOrder(self, header_path):
+    """Returns a path canonicalized for alphabetical comparison.
+
+    - replaces "-" with "_" so they both cmp the same.
+    - removes '-inl' since we don't require them to be after the main header.
+    - lowercase everything, just in case.
+
+    Args:
+      header_path: Path to be canonicalized.
+
+    Returns:
+      Canonicalized path.
+    """
+    return header_path.replace('-inl.h', '.h').replace('-', '_').lower()
+
+  def IsInAlphabeticalOrder(self, header_path):
+    """Check if a header is in alphabetical order with the previous header.
+
+    Args:
+      header_path: Header to be checked.
+
+    Returns:
+      Returns true if the header is in alphabetical order.
+    """
+    canonical_header = self.CanonicalizeAlphabeticalOrder(header_path)
+    if self._last_header > canonical_header:
+      return False
+    self._last_header = canonical_header
+    return True
+
+  def CheckNextIncludeOrder(self, header_type):
+    """Returns a non-empty error message if the next header is out of order.
+
+    This function also updates the internal state to be ready to check
+    the next include.
+
+    Args:
+      header_type: One of the _XXX_HEADER constants defined above.
+
+    Returns:
+      The empty string if the header is in the right order, or an
+      error message describing what's wrong.
+
+    """
+    error_message = ('Found %s after %s' %
+                     (self._TYPE_NAMES[header_type],
+                      self._SECTION_NAMES[self._section]))
+
+    last_section = self._section
+
+    if header_type == _C_SYS_HEADER:
+      if self._section <= self._C_SECTION:
+        self._section = self._C_SECTION
+      else:
+        self._last_header = ''
+        return error_message
+    elif header_type == _CPP_SYS_HEADER:
+      if self._section <= self._CPP_SECTION:
+        self._section = self._CPP_SECTION
+      else:
+        self._last_header = ''
+        return error_message
+    elif header_type == _LIKELY_MY_HEADER:
+      if self._section <= self._MY_H_SECTION:
+        self._section = self._MY_H_SECTION
+      else:
+        self._section = self._OTHER_H_SECTION
+    elif header_type == _POSSIBLE_MY_HEADER:
+      if self._section <= self._MY_H_SECTION:
+        self._section = self._MY_H_SECTION
+      else:
+        # This will always be the fallback because we're not sure
+        # enough that the header is associated with this file.
+        self._section = self._OTHER_H_SECTION
+    else:
+      assert header_type == _OTHER_HEADER
+      self._section = self._OTHER_H_SECTION
+
+    if last_section != self._section:
+      self._last_header = ''
+
+    return ''
+
+
+class _CppLintState(object):
+  """Maintains module-wide state.."""
+
+  def __init__(self):
+    self.verbose_level = 1  # global setting.
+    self.error_count = 0    # global count of reported errors
+    # filters to apply when emitting error messages
+    self.filters = _DEFAULT_FILTERS[:]
+    self.counting = 'total'  # In what way are we counting errors?
+    self.errors_by_category = {}  # string to int dict storing error counts
+
+    # output format:
+    # "emacs" - format that emacs can parse (default)
+    # "vs7" - format that Microsoft Visual Studio 7 can parse
+    self.output_format = 'emacs'
+
+  def SetOutputFormat(self, output_format):
+    """Sets the output format for errors."""
+    self.output_format = output_format
+
+  def SetVerboseLevel(self, level):
+    """Sets the module's verbosity, and returns the previous setting."""
+    last_verbose_level = self.verbose_level
+    self.verbose_level = level
+    return last_verbose_level
+
+  def SetCountingStyle(self, counting_style):
+    """Sets the module's counting options."""
+    self.counting = counting_style
+
+  def SetFilters(self, filters):
+    """Sets the error-message filters.
+
+    These filters are applied when deciding whether to emit a given
+    error message.
+
+    Args:
+      filters: A string of comma-separated filters (eg "+whitespace/indent").
+               Each filter should start with + or -; else we die.
+
+    Raises:
+      ValueError: The comma-separated filters did not all start with '+' or '-'.
+                  E.g. "-,+whitespace,-whitespace/indent,whitespace/badfilter"
+    """
+    # Default filters always have less priority than the flag ones.
+    self.filters = _DEFAULT_FILTERS[:]
+    for filt in filters.split(','):
+      clean_filt = filt.strip()
+      if clean_filt:
+        self.filters.append(clean_filt)
+    for filt in self.filters:
+      if not (filt.startswith('+') or filt.startswith('-')):
+        raise ValueError('Every filter in --filters must start with + or -'
+                         ' (%s does not)' % filt)
+
+  def ResetErrorCounts(self):
+    """Sets the module's error statistic back to zero."""
+    self.error_count = 0
+    self.errors_by_category = {}
+
+  def IncrementErrorCount(self, category):
+    """Bumps the module's error statistic."""
+    self.error_count += 1
+    if self.counting in ('toplevel', 'detailed'):
+      if self.counting != 'detailed':
+        category = category.split('/')[0]
+      if category not in self.errors_by_category:
+        self.errors_by_category[category] = 0
+      self.errors_by_category[category] += 1
+
+  def PrintErrorCounts(self):
+    """Print a summary of errors by category, and the total."""
+    for category, count in self.errors_by_category.iteritems():
+      sys.stderr.write('Category \'%s\' errors found: %d\n' %
+                       (category, count))
+    sys.stderr.write('Total errors found: %d\n' % self.error_count)
+
+_cpplint_state = _CppLintState()
+
+
+def _OutputFormat():
+  """Gets the module's output format."""
+  return _cpplint_state.output_format
+
+
+def _SetOutputFormat(output_format):
+  """Sets the module's output format."""
+  _cpplint_state.SetOutputFormat(output_format)
+
+
+def _VerboseLevel():
+  """Returns the module's verbosity setting."""
+  return _cpplint_state.verbose_level
+
+
+def _SetVerboseLevel(level):
+  """Sets the module's verbosity, and returns the previous setting."""
+  return _cpplint_state.SetVerboseLevel(level)
+
+
+def _SetCountingStyle(level):
+  """Sets the module's counting options."""
+  _cpplint_state.SetCountingStyle(level)
+
+
+def _Filters():
+  """Returns the module's list of output filters, as a list."""
+  return _cpplint_state.filters
+
+
+def _SetFilters(filters):
+  """Sets the module's error-message filters.
+
+  These filters are applied when deciding whether to emit a given
+  error message.
+
+  Args:
+    filters: A string of comma-separated filters (eg "whitespace/indent").
+             Each filter should start with + or -; else we die.
+  """
+  _cpplint_state.SetFilters(filters)
+
+
+class _FunctionState(object):
+  """Tracks current function name and the number of lines in its body."""
+
+  _NORMAL_TRIGGER = 250  # for --v=0, 500 for --v=1, etc.
+  _TEST_TRIGGER = 400    # about 50% more than _NORMAL_TRIGGER.
+
+  def __init__(self):
+    self.in_a_function = False
+    self.lines_in_function = 0
+    self.current_function = ''
+
+  def Begin(self, function_name):
+    """Start analyzing function body.
+
+    Args:
+      function_name: The name of the function being tracked.
+    """
+    self.in_a_function = True
+    self.lines_in_function = 0
+    self.current_function = function_name
+
+  def Count(self):
+    """Count line in current function body."""
+    if self.in_a_function:
+      self.lines_in_function += 1
+
+  def Check(self, error, filename, linenum):
+    """Report if too many lines in function body.
+
+    Args:
+      error: The function to call with any errors found.
+      filename: The name of the current file.
+      linenum: The number of the line to check.
+    """
+    if Match(r'T(EST|est)', self.current_function):
+      base_trigger = self._TEST_TRIGGER
+    else:
+      base_trigger = self._NORMAL_TRIGGER
+    trigger = base_trigger * 2**_VerboseLevel()
+
+    if self.lines_in_function > trigger:
+      error_level = int(math.log(self.lines_in_function / base_trigger, 2))
+      # 50 => 0, 100 => 1, 200 => 2, 400 => 3, 800 => 4, 1600 => 5, ...
+      if error_level > 5:
+        error_level = 5
+      error(filename, linenum, 'readability/fn_size', error_level,
+            'Small and focused functions are preferred:'
+            ' %s has %d non-comment lines'
+            ' (error triggered by exceeding %d lines).'  % (
+                self.current_function, self.lines_in_function, trigger))
+
+  def End(self):
+    """Stop analyzing function body."""
+    self.in_a_function = False
+
+
+class _IncludeError(Exception):
+  """Indicates a problem with the include order in a file."""
+  pass
+
+
+class FileInfo:
+  """Provides utility functions for filenames.
+
+  FileInfo provides easy access to the components of a file's path
+  relative to the project root.
+  """
+
+  def __init__(self, filename):
+    self._filename = filename
+
+  def FullName(self):
+    """Make Windows paths like Unix."""
+    return os.path.abspath(self._filename).replace('\\', '/')
+
+  def RepositoryName(self):
+    """FullName after removing the local path to the repository.
+
+    If we have a real absolute path name here we can try to do something smart:
+    detecting the root of the checkout and truncating /path/to/checkout from
+    the name so that we get header guards that don't include things like
+    "C:\Documents and Settings\..." or "/home/username/..." in them and thus
+    people on different computers who have checked the source out to different
+    locations won't see bogus errors.
+    """
+    fullname = self.FullName()
+
+    if os.path.exists(fullname):
+      project_dir = os.path.dirname(fullname)
+
+      if os.path.exists(os.path.join(project_dir, ".svn")):
+        # If there's a .svn file in the current directory, we recursively look
+        # up the directory tree for the top of the SVN checkout
+        root_dir = project_dir
+        one_up_dir = os.path.dirname(root_dir)
+        while os.path.exists(os.path.join(one_up_dir, ".svn")):
+          root_dir = os.path.dirname(root_dir)
+          one_up_dir = os.path.dirname(one_up_dir)
+
+        prefix = os.path.commonprefix([root_dir, project_dir])
+        return fullname[len(prefix) + 1:]
+
+      # Not SVN <= 1.6? Try to find a git, hg, or svn top level directory by
+      # searching up from the current path.
+      root_dir = os.path.dirname(fullname)
+      while (root_dir != os.path.dirname(root_dir) and
+             not os.path.exists(os.path.join(root_dir, ".git")) and
+             not os.path.exists(os.path.join(root_dir, ".hg")) and
+             not os.path.exists(os.path.join(root_dir, ".svn"))):
+        root_dir = os.path.dirname(root_dir)
+
+      if (os.path.exists(os.path.join(root_dir, ".git")) or
+          os.path.exists(os.path.join(root_dir, ".hg")) or
+          os.path.exists(os.path.join(root_dir, ".svn"))):
+        prefix = os.path.commonprefix([root_dir, project_dir])
+        return fullname[len(prefix) + 1:]
+
+    # Don't know what to do; header guard warnings may be wrong...
+    return fullname
+
+  def Split(self):
+    """Splits the file into the directory, basename, and extension.
+
+    For 'chrome/browser/browser.cc', Split() would
+    return ('chrome/browser', 'browser', '.cc')
+
+    Returns:
+      A tuple of (directory, basename, extension).
+    """
+
+    googlename = self.RepositoryName()
+    project, rest = os.path.split(googlename)
+    return (project,) + os.path.splitext(rest)
+
+  def BaseName(self):
+    """File base name - text after the final slash, before the final period."""
+    return self.Split()[1]
+
+  def Extension(self):
+    """File extension - text following the final period."""
+    return self.Split()[2]
+
+  def NoExtension(self):
+    """File has no source file extension."""
+    return '/'.join(self.Split()[0:2])
+
+  def IsSource(self):
+    """File has a source file extension."""
+    return self.Extension()[1:] in ('c', 'cc', 'cpp', 'cxx')
+
+
+def _ShouldPrintError(category, confidence, linenum):
+  """If confidence >= verbose, category passes filter and is not suppressed."""
+
+  # There are three ways we might decide not to print an error message:
+  # a "NOLINT(category)" comment appears in the source,
+  # the verbosity level isn't high enough, or the filters filter it out.
+  if IsErrorSuppressedByNolint(category, linenum):
+    return False
+  if confidence < _cpplint_state.verbose_level:
+    return False
+
+  is_filtered = False
+  for one_filter in _Filters():
+    if one_filter.startswith('-'):
+      if category.startswith(one_filter[1:]):
+        is_filtered = True
+    elif one_filter.startswith('+'):
+      if category.startswith(one_filter[1:]):
+        is_filtered = False
+    else:
+      assert False  # should have been checked for in SetFilter.
+  if is_filtered:
+    return False
+
+  return True
+
+
+def Error(filename, linenum, category, confidence, message):
+  """Logs the fact we've found a lint error.
+
+  We log where the error was found, and also our confidence in the error,
+  that is, how certain we are this is a legitimate style regression, and
+  not a misidentification or a use that's sometimes justified.
+
+  False positives can be suppressed by the use of
+  "cpplint(category)"  comments on the offending line.  These are
+  parsed into _error_suppressions.
+
+  Args:
+    filename: The name of the file containing the error.
+    linenum: The number of the line containing the error.
+    category: A string used to describe the "category" this bug
+      falls under: "whitespace", say, or "runtime".  Categories
+      may have a hierarchy separated by slashes: "whitespace/indent".
+    confidence: A number from 1-5 representing a confidence score for
+      the error, with 5 meaning that we are certain of the problem,
+      and 1 meaning that it could be a legitimate construct.
+    message: The error message.
+  """
+  if _ShouldPrintError(category, confidence, linenum):
+    _cpplint_state.IncrementErrorCount(category)
+    if _cpplint_state.output_format == 'vs7':
+      sys.stderr.write('%s(%s):  %s  [%s] [%d]\n' % (
+          filename, linenum, message, category, confidence))
+    else:
+      sys.stderr.write('%s:%s:  %s  [%s] [%d]\n' % (
+          filename, linenum, message, category, confidence))
+
+
+# Matches standard C++ escape esequences per 2.13.2.3 of the C++ standard.
+_RE_PATTERN_CLEANSE_LINE_ESCAPES = re.compile(
+    r'\\([abfnrtv?"\\\']|\d+|x[0-9a-fA-F]+)')
+# Matches strings.  Escape codes should already be removed by ESCAPES.
+_RE_PATTERN_CLEANSE_LINE_DOUBLE_QUOTES = re.compile(r'"[^"]*"')
+# Matches characters.  Escape codes should already be removed by ESCAPES.
+_RE_PATTERN_CLEANSE_LINE_SINGLE_QUOTES = re.compile(r"'.'")
+# Matches multi-line C++ comments.
+# This RE is a little bit more complicated than one might expect, because we
+# have to take care of space removals tools so we can handle comments inside
+# statements better.
+# The current rule is: We only clear spaces from both sides when we're at the
+# end of the line. Otherwise, we try to remove spaces from the right side,
+# if this doesn't work we try on left side but only if there's a non-character
+# on the right.
+_RE_PATTERN_CLEANSE_LINE_C_COMMENTS = re.compile(
+    r"""(\s*/\*.*\*/\s*$|
+            /\*.*\*/\s+|
+         \s+/\*.*\*/(?=\W)|
+            /\*.*\*/)""", re.VERBOSE)
+
+
+def IsCppString(line):
+  """Does line terminate so, that the next symbol is in string constant.
+
+  This function does not consider single-line nor multi-line comments.
+
+  Args:
+    line: is a partial line of code starting from the 0..n.
+
+  Returns:
+    True, if next character appended to 'line' is inside a
+    string constant.
+  """
+
+  line = line.replace(r'\\', 'XX')  # after this, \\" does not match to \"
+  return ((line.count('"') - line.count(r'\"') - line.count("'\"'")) & 1) == 1
+
+
+def FindNextMultiLineCommentStart(lines, lineix):
+  """Find the beginning marker for a multiline comment."""
+  while lineix < len(lines):
+    if lines[lineix].strip().startswith('/*'):
+      # Only return this marker if the comment goes beyond this line
+      if lines[lineix].strip().find('*/', 2) < 0:
+        return lineix
+    lineix += 1
+  return len(lines)
+
+
+def FindNextMultiLineCommentEnd(lines, lineix):
+  """We are inside a comment, find the end marker."""
+  while lineix < len(lines):
+    if lines[lineix].strip().endswith('*/'):
+      return lineix
+    lineix += 1
+  return len(lines)
+
+
+def RemoveMultiLineCommentsFromRange(lines, begin, end):
+  """Clears a range of lines for multi-line comments."""
+  # Having // dummy comments makes the lines non-empty, so we will not get
+  # unnecessary blank line warnings later in the code.
+  for i in range(begin, end):
+    lines[i] = '// dummy'
+
+
+def RemoveMultiLineComments(filename, lines, error):
+  """Removes multiline (c-style) comments from lines."""
+  lineix = 0
+  while lineix < len(lines):
+    lineix_begin = FindNextMultiLineCommentStart(lines, lineix)
+    if lineix_begin >= len(lines):
+      return
+    lineix_end = FindNextMultiLineCommentEnd(lines, lineix_begin)
+    if lineix_end >= len(lines):
+      error(filename, lineix_begin + 1, 'readability/multiline_comment', 5,
+            'Could not find end of multi-line comment')
+      return
+    RemoveMultiLineCommentsFromRange(lines, lineix_begin, lineix_end + 1)
+    lineix = lineix_end + 1
+
+
+def CleanseComments(line):
+  """Removes //-comments and single-line C-style /* */ comments.
+
+  Args:
+    line: A line of C++ source.
+
+  Returns:
+    The line with single-line comments removed.
+  """
+  commentpos = line.find('//')
+  if commentpos != -1 and not IsCppString(line[:commentpos]):
+    line = line[:commentpos].rstrip()
+  # get rid of /* ... */
+  return _RE_PATTERN_CLEANSE_LINE_C_COMMENTS.sub('', line)
+
+
+class CleansedLines(object):
+  """Holds 3 copies of all lines with different preprocessing applied to them.
+
+  1) elided member contains lines without strings and comments,
+  2) lines member contains lines without comments, and
+  3) raw_lines member contains all the lines without processing.
+  All these three members are of <type 'list'>, and of the same length.
+  """
+
+  def __init__(self, lines):
+    self.elided = []
+    self.lines = []
+    self.raw_lines = lines
+    self.num_lines = len(lines)
+    for linenum in range(len(lines)):
+      self.lines.append(CleanseComments(lines[linenum]))
+      elided = self._CollapseStrings(lines[linenum])
+      self.elided.append(CleanseComments(elided))
+
+  def NumLines(self):
+    """Returns the number of lines represented."""
+    return self.num_lines
+
+  @staticmethod
+  def _CollapseStrings(elided):
+    """Collapses strings and chars on a line to simple "" or '' blocks.
+
+    We nix strings first so we're not fooled by text like '"http://"'
+
+    Args:
+      elided: The line being processed.
+
+    Returns:
+      The line with collapsed strings.
+    """
+    if not _RE_PATTERN_INCLUDE.match(elided):
+      # Remove escaped characters first to make quote/single quote collapsing
+      # basic.  Things that look like escaped characters shouldn't occur
+      # outside of strings and chars.
+      elided = _RE_PATTERN_CLEANSE_LINE_ESCAPES.sub('', elided)
+      elided = _RE_PATTERN_CLEANSE_LINE_SINGLE_QUOTES.sub("''", elided)
+      elided = _RE_PATTERN_CLEANSE_LINE_DOUBLE_QUOTES.sub('""', elided)
+    return elided
+
+
+def FindEndOfExpressionInLine(line, startpos, depth, startchar, endchar):
+  """Find the position just after the matching endchar.
+
+  Args:
+    line: a CleansedLines line.
+    startpos: start searching at this position.
+    depth: nesting level at startpos.
+    startchar: expression opening character.
+    endchar: expression closing character.
+
+  Returns:
+    Index just after endchar.
+  """
+  for i in xrange(startpos, len(line)):
+    if line[i] == startchar:
+      depth += 1
+    elif line[i] == endchar:
+      depth -= 1
+      if depth == 0:
+        return i + 1
+  return -1
+
+
+def CloseExpression(clean_lines, linenum, pos):
+  """If input points to ( or { or [, finds the position that closes it.
+
+  If lines[linenum][pos] points to a '(' or '{' or '[', finds the
+  linenum/pos that correspond to the closing of the expression.
+
+  Args:
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    pos: A position on the line.
+
+  Returns:
+    A tuple (line, linenum, pos) pointer *past* the closing brace, or
+    (line, len(lines), -1) if we never find a close.  Note we ignore
+    strings and comments when matching; and the line we return is the
+    'cleansed' line at linenum.
+  """
+
+  line = clean_lines.elided[linenum]
+  startchar = line[pos]
+  if startchar not in '({[':
+    return (line, clean_lines.NumLines(), -1)
+  if startchar == '(': endchar = ')'
+  if startchar == '[': endchar = ']'
+  if startchar == '{': endchar = '}'
+
+  # Check first line
+  end_pos = FindEndOfExpressionInLine(line, pos, 0, startchar, endchar)
+  if end_pos > -1:
+    return (line, linenum, end_pos)
+  tail = line[pos:]
+  num_open = tail.count(startchar) - tail.count(endchar)
+  while linenum < clean_lines.NumLines() - 1:
+    linenum += 1
+    line = clean_lines.elided[linenum]
+    delta = line.count(startchar) - line.count(endchar)
+    if num_open + delta <= 0:
+      return (line, linenum,
+              FindEndOfExpressionInLine(line, 0, num_open, startchar, endchar))
+    num_open += delta
+
+  # Did not find endchar before end of file, give up
+  return (line, clean_lines.NumLines(), -1)
+
+def CheckForCopyright(filename, lines, error):
+  """Logs an error if no Copyright message appears at the top of the file."""
+
+  # We'll say it should occur by line 10. Don't forget there's a
+  # dummy line at the front.
+  for line in xrange(1, min(len(lines), 11)):
+    if re.search(r'Copyright', lines[line], re.I): break
+  else:                       # means no copyright line was found
+    error(filename, 0, 'legal/copyright', 5,
+          'No copyright message found.  '
+          'You should have a line: "Copyright [year] <Copyright Owner>"')
+
+
+def GetHeaderGuardCPPVariable(filename):
+  """Returns the CPP variable that should be used as a header guard.
+
+  Args:
+    filename: The name of a C++ header file.
+
+  Returns:
+    The CPP variable that should be used as a header guard in the
+    named file.
+
+  """
+
+  # Restores original filename in case that cpplint is invoked from Emacs's
+  # flymake.
+  filename = re.sub(r'_flymake\.h$', '.h', filename)
+  filename = re.sub(r'/\.flymake/([^/]*)$', r'/\1', filename)
+
+  fileinfo = FileInfo(filename)
+  file_path_from_root = fileinfo.RepositoryName()
+  if _root:
+    file_path_from_root = re.sub('^' + _root + os.sep, '', file_path_from_root)
+  return re.sub(r'[-./\s]', '_', file_path_from_root).upper() + '_'
+
+
+def CheckForHeaderGuard(filename, lines, error):
+  """Checks that the file contains a header guard.
+
+  Logs an error if no #ifndef header guard is present.  For other
+  headers, checks that the full pathname is used.
+
+  Args:
+    filename: The name of the C++ header file.
+    lines: An array of strings, each representing a line of the file.
+    error: The function to call with any errors found.
+  """
+
+  cppvar = GetHeaderGuardCPPVariable(filename)
+
+  ifndef = None
+  ifndef_linenum = 0
+  define = None
+  endif = None
+  endif_linenum = 0
+  for linenum, line in enumerate(lines):
+    linesplit = line.split()
+    if len(linesplit) >= 2:
+      # find the first occurrence of #ifndef and #define, save arg
+      if not ifndef and linesplit[0] == '#ifndef':
+        # set ifndef to the header guard presented on the #ifndef line.
+        ifndef = linesplit[1]
+        ifndef_linenum = linenum
+      if not define and linesplit[0] == '#define':
+        define = linesplit[1]
+    # find the last occurrence of #endif, save entire line
+    if line.startswith('#endif'):
+      endif = line
+      endif_linenum = linenum
+
+  if not ifndef:
+    error(filename, 0, 'build/header_guard', 5,
+          'No #ifndef header guard found, suggested CPP variable is: %s' %
+          cppvar)
+    return
+
+  if not define:
+    error(filename, 0, 'build/header_guard', 5,
+          'No #define header guard found, suggested CPP variable is: %s' %
+          cppvar)
+    return
+
+  # The guard should be PATH_FILE_H_, but we also allow PATH_FILE_H__
+  # for backward compatibility.
+  if ifndef != cppvar:
+    error_level = 0
+    if ifndef != cppvar + '_':
+      error_level = 5
+
+    ParseNolintSuppressions(filename, lines[ifndef_linenum], ifndef_linenum,
+                            error)
+    error(filename, ifndef_linenum, 'build/header_guard', error_level,
+          '#ifndef header guard has wrong style, please use: %s' % cppvar)
+
+  if define != ifndef:
+    error(filename, 0, 'build/header_guard', 5,
+          '#ifndef and #define don\'t match, suggested CPP variable is: %s' %
+          cppvar)
+    return
+
+  if endif != ('#endif  // %s' % cppvar):
+    error_level = 0
+    if endif != ('#endif  // %s' % (cppvar + '_')):
+      error_level = 5
+
+    ParseNolintSuppressions(filename, lines[endif_linenum], endif_linenum,
+                            error)
+    error(filename, endif_linenum, 'build/header_guard', error_level,
+          '#endif line should be "#endif  // %s"' % cppvar)
+
+
+def CheckForUnicodeReplacementCharacters(filename, lines, error):
+  """Logs an error for each line containing Unicode replacement characters.
+
+  These indicate that either the file contained invalid UTF-8 (likely)
+  or Unicode replacement characters (which it shouldn't).  Note that
+  it's possible for this to throw off line numbering if the invalid
+  UTF-8 occurred adjacent to a newline.
+
+  Args:
+    filename: The name of the current file.
+    lines: An array of strings, each representing a line of the file.
+    error: The function to call with any errors found.
+  """
+  for linenum, line in enumerate(lines):
+    if u'\ufffd' in line:
+      error(filename, linenum, 'readability/utf8', 5,
+            'Line contains invalid UTF-8 (or Unicode replacement character).')
+
+
+def CheckForNewlineAtEOF(filename, lines, error):
+  """Logs an error if there is no newline char at the end of the file.
+
+  Args:
+    filename: The name of the current file.
+    lines: An array of strings, each representing a line of the file.
+    error: The function to call with any errors found.
+  """
+
+  # The array lines() was created by adding two newlines to the
+  # original file (go figure), then splitting on \n.
+  # To verify that the file ends in \n, we just have to make sure the
+  # last-but-two element of lines() exists and is empty.
+  if len(lines) < 3 or lines[-2]:
+    error(filename, len(lines) - 2, 'whitespace/ending_newline', 5,
+          'Could not find a newline character at the end of the file.')
+
+
+def CheckForMultilineCommentsAndStrings(filename, clean_lines, linenum, error):
+  """Logs an error if we see /* ... */ or "..." that extend past one line.
+
+  /* ... */ comments are legit inside macros, for one line.
+  Otherwise, we prefer // comments, so it's ok to warn about the
+  other.  Likewise, it's ok for strings to extend across multiple
+  lines, as long as a line continuation character (backslash)
+  terminates each line. Although not currently prohibited by the C++
+  style guide, it's ugly and unnecessary. We don't do well with either
+  in this lint program, so we warn about both.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  line = clean_lines.elided[linenum]
+
+  # Remove all \\ (escaped backslashes) from the line. They are OK, and the
+  # second (escaped) slash may trigger later \" detection erroneously.
+  line = line.replace('\\\\', '')
+
+  if line.count('/*') > line.count('*/'):
+    error(filename, linenum, 'readability/multiline_comment', 5,
+          'Complex multi-line /*...*/-style comment found. '
+          'Lint may give bogus warnings.  '
+          'Consider replacing these with //-style comments, '
+          'with #if 0...#endif, '
+          'or with more clearly structured multi-line comments.')
+
+  if (line.count('"') - line.count('\\"')) % 2:
+    error(filename, linenum, 'readability/multiline_string', 5,
+          'Multi-line string ("...") found.  This lint script doesn\'t '
+          'do well with such strings, and may give bogus warnings.  They\'re '
+          'ugly and unnecessary, and you should use concatenation instead".')
+
+
+threading_list = (
+    ('asctime(', 'asctime_r('),
+    ('ctime(', 'ctime_r('),
+    ('getgrgid(', 'getgrgid_r('),
+    ('getgrnam(', 'getgrnam_r('),
+    ('getlogin(', 'getlogin_r('),
+    ('getpwnam(', 'getpwnam_r('),
+    ('getpwuid(', 'getpwuid_r('),
+    ('gmtime(', 'gmtime_r('),
+    ('localtime(', 'localtime_r('),
+    ('rand(', 'rand_r('),
+    ('readdir(', 'readdir_r('),
+    ('strtok(', 'strtok_r('),
+    ('ttyname(', 'ttyname_r('),
+    )
+
+
+def CheckPosixThreading(filename, clean_lines, linenum, error):
+  """Checks for calls to thread-unsafe functions.
+
+  Much code has been originally written without consideration of
+  multi-threading. Also, engineers are relying on their old experience;
+  they have learned posix before threading extensions were added. These
+  tests guide the engineers to use thread-safe functions (when using
+  posix directly).
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  line = clean_lines.elided[linenum]
+  for single_thread_function, multithread_safe_function in threading_list:
+    ix = line.find(single_thread_function)
+    # Comparisons made explicit for clarity -- pylint: disable-msg=C6403
+    if ix >= 0 and (ix == 0 or (not line[ix - 1].isalnum() and
+                                line[ix - 1] not in ('_', '.', '>'))):
+      error(filename, linenum, 'runtime/threadsafe_fn', 2,
+            'Consider using ' + multithread_safe_function +
+            '...) instead of ' + single_thread_function +
+            '...) for improved thread safety.')
+
+
+# Matches invalid increment: *count++, which moves pointer instead of
+# incrementing a value.
+_RE_PATTERN_INVALID_INCREMENT = re.compile(
+    r'^\s*\*\w+(\+\+|--);')
+
+
+def CheckInvalidIncrement(filename, clean_lines, linenum, error):
+  """Checks for invalid increment *count++.
+
+  For example following function:
+  void increment_counter(int* count) {
+    *count++;
+  }
+  is invalid, because it effectively does count++, moving pointer, and should
+  be replaced with ++*count, (*count)++ or *count += 1.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  line = clean_lines.elided[linenum]
+  if _RE_PATTERN_INVALID_INCREMENT.match(line):
+    error(filename, linenum, 'runtime/invalid_increment', 5,
+          'Changing pointer instead of value (or unused value of operator*).')
+
+
+class _BlockInfo(object):
+  """Stores information about a generic block of code."""
+
+  def __init__(self, seen_open_brace):
+    self.seen_open_brace = seen_open_brace
+    self.open_parentheses = 0
+    self.inline_asm = _NO_ASM
+
+  def CheckBegin(self, filename, clean_lines, linenum, error):
+    """Run checks that applies to text up to the opening brace.
+
+    This is mostly for checking the text after the class identifier
+    and the "{", usually where the base class is specified.  For other
+    blocks, there isn't much to check, so we always pass.
+
+    Args:
+      filename: The name of the current file.
+      clean_lines: A CleansedLines instance containing the file.
+      linenum: The number of the line to check.
+      error: The function to call with any errors found.
+    """
+    pass
+
+  def CheckEnd(self, filename, clean_lines, linenum, error):
+    """Run checks that applies to text after the closing brace.
+
+    This is mostly used for checking end of namespace comments.
+
+    Args:
+      filename: The name of the current file.
+      clean_lines: A CleansedLines instance containing the file.
+      linenum: The number of the line to check.
+      error: The function to call with any errors found.
+    """
+    pass
+
+
+class _ClassInfo(_BlockInfo):
+  """Stores information about a class."""
+
+  def __init__(self, name, class_or_struct, clean_lines, linenum):
+    _BlockInfo.__init__(self, False)
+    self.name = name
+    self.starting_linenum = linenum
+    self.is_derived = False
+    if class_or_struct == 'struct':
+      self.access = 'public'
+    else:
+      self.access = 'private'
+
+    # Try to find the end of the class.  This will be confused by things like:
+    #   class A {
+    #   } *x = { ...
+    #
+    # But it's still good enough for CheckSectionSpacing.
+    self.last_line = 0
+    depth = 0
+    for i in range(linenum, clean_lines.NumLines()):
+      line = clean_lines.elided[i]
+      depth += line.count('{') - line.count('}')
+      if not depth:
+        self.last_line = i
+        break
+
+  def CheckBegin(self, filename, clean_lines, linenum, error):
+    # Look for a bare ':'
+    if Search('(^|[^:]):($|[^:])', clean_lines.elided[linenum]):
+      self.is_derived = True
+
+
+class _NamespaceInfo(_BlockInfo):
+  """Stores information about a namespace."""
+
+  def __init__(self, name, linenum):
+    _BlockInfo.__init__(self, False)
+    self.name = name or ''
+    self.starting_linenum = linenum
+
+  def CheckEnd(self, filename, clean_lines, linenum, error):
+    """Check end of namespace comments."""
+    line = clean_lines.raw_lines[linenum]
+
+    # Check how many lines is enclosed in this namespace.  Don't issue
+    # warning for missing namespace comments if there aren't enough
+    # lines.  However, do apply checks if there is already an end of
+    # namespace comment and it's incorrect.
+    #
+    # TODO(unknown): We always want to check end of namespace comments
+    # if a namespace is large, but sometimes we also want to apply the
+    # check if a short namespace contained nontrivial things (something
+    # other than forward declarations).  There is currently no logic on
+    # deciding what these nontrivial things are, so this check is
+    # triggered by namespace size only, which works most of the time.
+    if (linenum - self.starting_linenum < 10
+        and not Match(r'};*\s*(//|/\*).*\bnamespace\b', line)):
+      return
+
+    # Look for matching comment at end of namespace.
+    #
+    # Note that we accept C style "/* */" comments for terminating
+    # namespaces, so that code that terminate namespaces inside
+    # preprocessor macros can be cpplint clean.  Example: http://go/nxpiz
+    #
+    # We also accept stuff like "// end of namespace <name>." with the
+    # period at the end.
+    #
+    # Besides these, we don't accept anything else, otherwise we might
+    # get false negatives when existing comment is a substring of the
+    # expected namespace.  Example: http://go/ldkdc, http://cl/23548205
+    if self.name:
+      # Named namespace
+      if not Match((r'};*\s*(//|/\*).*\bnamespace\s+' + re.escape(self.name) +
+                    r'[\*/\.\\\s]*$'),
+                   line):
+        error(filename, linenum, 'readability/namespace', 5,
+              'Namespace should be terminated with "// namespace %s"' %
+              self.name)
+    else:
+      # Anonymous namespace
+      if not Match(r'};*\s*(//|/\*).*\bnamespace[\*/\.\\\s]*$', line):
+        error(filename, linenum, 'readability/namespace', 5,
+              'Namespace should be terminated with "// namespace"')
+
+
+class _PreprocessorInfo(object):
+  """Stores checkpoints of nesting stacks when #if/#else is seen."""
+
+  def __init__(self, stack_before_if):
+    # The entire nesting stack before #if
+    self.stack_before_if = stack_before_if
+
+    # The entire nesting stack up to #else
+    self.stack_before_else = []
+
+    # Whether we have already seen #else or #elif
+    self.seen_else = False
+
+
+class _NestingState(object):
+  """Holds states related to parsing braces."""
+
+  def __init__(self):
+    # Stack for tracking all braces.  An object is pushed whenever we
+    # see a "{", and popped when we see a "}".  Only 3 types of
+    # objects are possible:
+    # - _ClassInfo: a class or struct.
+    # - _NamespaceInfo: a namespace.
+    # - _BlockInfo: some other type of block.
+    self.stack = []
+
+    # Stack of _PreprocessorInfo objects.
+    self.pp_stack = []
+
+  def SeenOpenBrace(self):
+    """Check if we have seen the opening brace for the innermost block.
+
+    Returns:
+      True if we have seen the opening brace, False if the innermost
+      block is still expecting an opening brace.
+    """
+    return (not self.stack) or self.stack[-1].seen_open_brace
+
+  def InNamespaceBody(self):
+    """Check if we are currently one level inside a namespace body.
+
+    Returns:
+      True if top of the stack is a namespace block, False otherwise.
+    """
+    return self.stack and isinstance(self.stack[-1], _NamespaceInfo)
+
+  def UpdatePreprocessor(self, line):
+    """Update preprocessor stack.
+
+    We need to handle preprocessors due to classes like this:
+      #ifdef SWIG
+      struct ResultDetailsPageElementExtensionPoint {
+      #else
+      struct ResultDetailsPageElementExtensionPoint : public Extension {
+      #endif
+    (see http://go/qwddn for original example)
+
+    We make the following assumptions (good enough for most files):
+    - Preprocessor condition evaluates to true from #if up to first
+      #else/#elif/#endif.
+
+    - Preprocessor condition evaluates to false from #else/#elif up
+      to #endif.  We still perform lint checks on these lines, but
+      these do not affect nesting stack.
+
+    Args:
+      line: current line to check.
+    """
+    if Match(r'^\s*#\s*(if|ifdef|ifndef)\b', line):
+      # Beginning of #if block, save the nesting stack here.  The saved
+      # stack will allow us to restore the parsing state in the #else case.
+      self.pp_stack.append(_PreprocessorInfo(copy.deepcopy(self.stack)))
+    elif Match(r'^\s*#\s*(else|elif)\b', line):
+      # Beginning of #else block
+      if self.pp_stack:
+        if not self.pp_stack[-1].seen_else:
+          # This is the first #else or #elif block.  Remember the
+          # whole nesting stack up to this point.  This is what we
+          # keep after the #endif.
+          self.pp_stack[-1].seen_else = True
+          self.pp_stack[-1].stack_before_else = copy.deepcopy(self.stack)
+
+        # Restore the stack to how it was before the #if
+        self.stack = copy.deepcopy(self.pp_stack[-1].stack_before_if)
+      else:
+        # TODO(unknown): unexpected #else, issue warning?
+        pass
+    elif Match(r'^\s*#\s*endif\b', line):
+      # End of #if or #else blocks.
+      if self.pp_stack:
+        # If we saw an #else, we will need to restore the nesting
+        # stack to its former state before the #else, otherwise we
+        # will just continue from where we left off.
+        if self.pp_stack[-1].seen_else:
+          # Here we can just use a shallow copy since we are the last
+          # reference to it.
+          self.stack = self.pp_stack[-1].stack_before_else
+        # Drop the corresponding #if
+        self.pp_stack.pop()
+      else:
+        # TODO(unknown): unexpected #endif, issue warning?
+        pass
+
+  def Update(self, filename, clean_lines, linenum, error):
+    """Update nesting state with current line.
+
+    Args:
+      filename: The name of the current file.
+      clean_lines: A CleansedLines instance containing the file.
+      linenum: The number of the line to check.
+      error: The function to call with any errors found.
+    """
+    line = clean_lines.elided[linenum]
+
+    # Update pp_stack first
+    self.UpdatePreprocessor(line)
+
+    # Count parentheses.  This is to avoid adding struct arguments to
+    # the nesting stack.
+    if self.stack:
+      inner_block = self.stack[-1]
+      depth_change = line.count('(') - line.count(')')
+      inner_block.open_parentheses += depth_change
+
+      # Also check if we are starting or ending an inline assembly block.
+      if inner_block.inline_asm in (_NO_ASM, _END_ASM):
+        if (depth_change != 0 and
+            inner_block.open_parentheses == 1 and
+            _MATCH_ASM.match(line)):
+          # Enter assembly block
+          inner_block.inline_asm = _INSIDE_ASM
+        else:
+          # Not entering assembly block.  If previous line was _END_ASM,
+          # we will now shift to _NO_ASM state.
+          inner_block.inline_asm = _NO_ASM
+      elif (inner_block.inline_asm == _INSIDE_ASM and
+            inner_block.open_parentheses == 0):
+        # Exit assembly block
+        inner_block.inline_asm = _END_ASM
+
+    # Consume namespace declaration at the beginning of the line.  Do
+    # this in a loop so that we catch same line declarations like this:
+    #   namespace proto2 { namespace bridge { class MessageSet; } }
+    while True:
+      # Match start of namespace.  The "\b\s*" below catches namespace
+      # declarations even if it weren't followed by a whitespace, this
+      # is so that we don't confuse our namespace checker.  The
+      # missing spaces will be flagged by CheckSpacing.
+      namespace_decl_match = Match(r'^\s*namespace\b\s*([:\w]+)?(.*)$', line)
+      if not namespace_decl_match:
+        break
+
+      new_namespace = _NamespaceInfo(namespace_decl_match.group(1), linenum)
+      self.stack.append(new_namespace)
+
+      line = namespace_decl_match.group(2)
+      if line.find('{') != -1:
+        new_namespace.seen_open_brace = True
+        line = line[line.find('{') + 1:]
+
+    # Look for a class declaration in whatever is left of the line
+    # after parsing namespaces.  The regexp accounts for decorated classes
+    # such as in:
+    #   class LOCKABLE API Object {
+    #   };
+    #
+    # Templates with class arguments may confuse the parser, for example:
+    #   template <class T
+    #             class Comparator = less<T>,
+    #             class Vector = vector<T> >
+    #   class HeapQueue {
+    #
+    # Because this parser has no nesting state about templates, by the
+    # time it saw "class Comparator", it may think that it's a new class.
+    # Nested templates have a similar problem:
+    #   template <
+    #       typename ExportedType,
+    #       typename TupleType,
+    #       template <typename, typename> class ImplTemplate>
+    #
+    # To avoid these cases, we ignore classes that are followed by '=' or '>'
+    class_decl_match = Match(
+        r'\s*(template\s*<[\w\s<>,:]*>\s*)?'
+        '(class|struct)\s+([A-Z_]+\s+)*(\w+(?:::\w+)*)'
+        '(([^=>]|<[^<>]*>)*)$', line)
+    if (class_decl_match and
+        (not self.stack or self.stack[-1].open_parentheses == 0)):
+      self.stack.append(_ClassInfo(
+          class_decl_match.group(4), class_decl_match.group(2),
+          clean_lines, linenum))
+      line = class_decl_match.group(5)
+
+    # If we have not yet seen the opening brace for the innermost block,
+    # run checks here.
+    if not self.SeenOpenBrace():
+      self.stack[-1].CheckBegin(filename, clean_lines, linenum, error)
+
+    # Update access control if we are inside a class/struct
+    if self.stack and isinstance(self.stack[-1], _ClassInfo):
+      access_match = Match(r'\s*(public|private|protected)\s*:', line)
+      if access_match:
+        self.stack[-1].access = access_match.group(1)
+
+    # Consume braces or semicolons from what's left of the line
+    while True:
+      # Match first brace, semicolon, or closed parenthesis.
+      matched = Match(r'^[^{;)}]*([{;)}])(.*)$', line)
+      if not matched:
+        break
+
+      token = matched.group(1)
+      if token == '{':
+        # If namespace or class hasn't seen a opening brace yet, mark
+        # namespace/class head as complete.  Push a new block onto the
+        # stack otherwise.
+        if not self.SeenOpenBrace():
+          self.stack[-1].seen_open_brace = True
+        else:
+          self.stack.append(_BlockInfo(True))
+          if _MATCH_ASM.match(line):
+            self.stack[-1].inline_asm = _BLOCK_ASM
+      elif token == ';' or token == ')':
+        # If we haven't seen an opening brace yet, but we already saw
+        # a semicolon, this is probably a forward declaration.  Pop
+        # the stack for these.
+        #
+        # Similarly, if we haven't seen an opening brace yet, but we
+        # already saw a closing parenthesis, then these are probably
+        # function arguments with extra "class" or "struct" keywords.
+        # Also pop these stack for these.
+        if not self.SeenOpenBrace():
+          self.stack.pop()
+      else:  # token == '}'
+        # Perform end of block checks and pop the stack.
+        if self.stack:
+          self.stack[-1].CheckEnd(filename, clean_lines, linenum, error)
+          self.stack.pop()
+      line = matched.group(2)
+
+  def InnermostClass(self):
+    """Get class info on the top of the stack.
+
+    Returns:
+      A _ClassInfo object if we are inside a class, or None otherwise.
+    """
+    for i in range(len(self.stack), 0, -1):
+      classinfo = self.stack[i - 1]
+      if isinstance(classinfo, _ClassInfo):
+        return classinfo
+    return None
+
+  def CheckClassFinished(self, filename, error):
+    """Checks that all classes have been completely parsed.
+
+    Call this when all lines in a file have been processed.
+    Args:
+      filename: The name of the current file.
+      error: The function to call with any errors found.
+    """
+    # Note: This test can result in false positives if #ifdef constructs
+    # get in the way of brace matching. See the testBuildClass test in
+    # cpplint_unittest.py for an example of this.
+    for obj in self.stack:
+      if isinstance(obj, _ClassInfo):
+        error(filename, obj.starting_linenum, 'build/class', 5,
+              'Failed to find complete declaration of class %s' %
+              obj.name)
+
+
+def CheckForNonStandardConstructs(filename, clean_lines, linenum,
+                                  nesting_state, error):
+  """Logs an error if we see certain non-ANSI constructs ignored by gcc-2.
+
+  Complain about several constructs which gcc-2 accepts, but which are
+  not standard C++.  Warning about these in lint is one way to ease the
+  transition to new compilers.
+  - put storage class first (e.g. "static const" instead of "const static").
+  - "%lld" instead of %qd" in printf-type functions.
+  - "%1$d" is non-standard in printf-type functions.
+  - "\%" is an undefined character escape sequence.
+  - text after #endif is not allowed.
+  - invalid inner-style forward declaration.
+  - >? and <? operators, and their >?= and <?= cousins.
+
+  Additionally, check for constructor/destructor style violations and reference
+  members, as it is very convenient to do so while checking for
+  gcc-2 compliance.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    nesting_state: A _NestingState instance which maintains information about
+                   the current stack of nested blocks being parsed.
+    error: A callable to which errors are reported, which takes 4 arguments:
+           filename, line number, error level, and message
+  """
+
+  # Remove comments from the line, but leave in strings for now.
+  line = clean_lines.lines[linenum]
+
+  if Search(r'printf\s*\(.*".*%[-+ ]?\d*q', line):
+    error(filename, linenum, 'runtime/printf_format', 3,
+          '%q in format strings is deprecated.  Use %ll instead.')
+
+  if Search(r'printf\s*\(.*".*%\d+\$', line):
+    error(filename, linenum, 'runtime/printf_format', 2,
+          '%N$ formats are unconventional.  Try rewriting to avoid them.')
+
+  # Remove escaped backslashes before looking for undefined escapes.
+  line = line.replace('\\\\', '')
+
+  if Search(r'("|\').*\\(%|\[|\(|{)', line):
+    error(filename, linenum, 'build/printf_format', 3,
+          '%, [, (, and { are undefined character escapes.  Unescape them.')
+
+  # For the rest, work with both comments and strings removed.
+  line = clean_lines.elided[linenum]
+
+  if Search(r'\b(const|volatile|void|char|short|int|long'
+            r'|float|double|signed|unsigned'
+            r'|schar|u?int8|u?int16|u?int32|u?int64)'
+            r'\s+(register|static|extern|typedef)\b',
+            line):
+    error(filename, linenum, 'build/storage_class', 5,
+          'Storage class (static, extern, typedef, etc) should be first.')
+
+  if Match(r'\s*#\s*endif\s*[^/\s]+', line):
+    error(filename, linenum, 'build/endif_comment', 5,
+          'Uncommented text after #endif is non-standard.  Use a comment.')
+
+  if Match(r'\s*class\s+(\w+\s*::\s*)+\w+\s*;', line):
+    error(filename, linenum, 'build/forward_decl', 5,
+          'Inner-style forward declarations are invalid.  Remove this line.')
+
+  if Search(r'(\w+|[+-]?\d+(\.\d*)?)\s*(<|>)\?=?\s*(\w+|[+-]?\d+)(\.\d*)?',
+            line):
+    error(filename, linenum, 'build/deprecated', 3,
+          '>? and <? (max and min) operators are non-standard and deprecated.')
+
+  if Search(r'^\s*const\s*string\s*&\s*\w+\s*;', line):
+    # TODO(unknown): Could it be expanded safely to arbitrary references,
+    # without triggering too many false positives? The first
+    # attempt triggered 5 warnings for mostly benign code in the regtest, hence
+    # the restriction.
+    # Here's the original regexp, for the reference:
+    # type_name = r'\w+((\s*::\s*\w+)|(\s*<\s*\w+?\s*>))?'
+    # r'\s*const\s*' + type_name + '\s*&\s*\w+\s*;'
+    error(filename, linenum, 'runtime/member_string_references', 2,
+          'const string& members are dangerous. It is much better to use '
+          'alternatives, such as pointers or simple constants.')
+
+  # Everything else in this function operates on class declarations.
+  # Return early if the top of the nesting stack is not a class, or if
+  # the class head is not completed yet.
+  classinfo = nesting_state.InnermostClass()
+  if not classinfo or not classinfo.seen_open_brace:
+    return
+
+  # The class may have been declared with namespace or classname qualifiers.
+  # The constructor and destructor will not have those qualifiers.
+  base_classname = classinfo.name.split('::')[-1]
+
+  # Look for single-argument constructors that aren't marked explicit.
+  # Technically a valid construct, but against style.
+  args = Match(r'\s+(?:inline\s+)?%s\s*\(([^,()]+)\)'
+               % re.escape(base_classname),
+               line)
+  if (args and
+      args.group(1) != 'void' and
+      not Match(r'(const\s+)?%s\s*(?:<\w+>\s*)?&' % re.escape(base_classname),
+                args.group(1).strip())):
+    error(filename, linenum, 'runtime/explicit', 5,
+          'Single-argument constructors should be marked explicit.')
+
+
+def CheckSpacingForFunctionCall(filename, line, linenum, error):
+  """Checks for the correctness of various spacing around function calls.
+
+  Args:
+    filename: The name of the current file.
+    line: The text of the line to check.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+
+  # Since function calls often occur inside if/for/while/switch
+  # expressions - which have their own, more liberal conventions - we
+  # first see if we should be looking inside such an expression for a
+  # function call, to which we can apply more strict standards.
+  fncall = line    # if there's no control flow construct, look at whole line
+  for pattern in (r'\bif\s*\((.*)\)\s*{',
+                  r'\bfor\s*\((.*)\)\s*{',
+                  r'\bwhile\s*\((.*)\)\s*[{;]',
+                  r'\bswitch\s*\((.*)\)\s*{'):
+    match = Search(pattern, line)
+    if match:
+      fncall = match.group(1)    # look inside the parens for function calls
+      break
+
+  # Except in if/for/while/switch, there should never be space
+  # immediately inside parens (eg "f( 3, 4 )").  We make an exception
+  # for nested parens ( (a+b) + c ).  Likewise, there should never be
+  # a space before a ( when it's a function argument.  I assume it's a
+  # function argument when the char before the whitespace is legal in
+  # a function name (alnum + _) and we're not starting a macro. Also ignore
+  # pointers and references to arrays and functions coz they're too tricky:
+  # we use a very simple way to recognize these:
+  # " (something)(maybe-something)" or
+  # " (something)(maybe-something," or
+  # " (something)[something]"
+  # Note that we assume the contents of [] to be short enough that
+  # they'll never need to wrap.
+  if (  # Ignore control structures.
+      not Search(r'\b(if|for|while|switch|return|delete)\b', fncall) and
+      # Ignore pointers/references to functions.
+      not Search(r' \([^)]+\)\([^)]*(\)|,$)', fncall) and
+      # Ignore pointers/references to arrays.
+      not Search(r' \([^)]+\)\[[^\]]+\]', fncall)):
+    if Search(r'\w\s*\(\s(?!\s*\\$)', fncall):      # a ( used for a fn call
+      error(filename, linenum, 'whitespace/parens', 4,
+            'Extra space after ( in function call')
+    elif Search(r'\(\s+(?!(\s*\\)|\()', fncall):
+      error(filename, linenum, 'whitespace/parens', 2,
+            'Extra space after (')
+    if (Search(r'\w\s+\(', fncall) and
+        not Search(r'#\s*define|typedef', fncall) and
+        not Search(r'\w\s+\((\w+::)?\*\w+\)\(', fncall)):
+      error(filename, linenum, 'whitespace/parens', 4,
+            'Extra space before ( in function call')
+    # If the ) is followed only by a newline or a { + newline, assume it's
+    # part of a control statement (if/while/etc), and don't complain
+    if Search(r'[^)]\s+\)\s*[^{\s]', fncall):
+      # If the closing parenthesis is preceded by only whitespaces,
+      # try to give a more descriptive error message.
+      if Search(r'^\s+\)', fncall):
+        error(filename, linenum, 'whitespace/parens', 2,
+              'Closing ) should be moved to the previous line')
+      else:
+        error(filename, linenum, 'whitespace/parens', 2,
+              'Extra space before )')
+
+
+def IsBlankLine(line):
+  """Returns true if the given line is blank.
+
+  We consider a line to be blank if the line is empty or consists of
+  only white spaces.
+
+  Args:
+    line: A line of a string.
+
+  Returns:
+    True, if the given line is blank.
+  """
+  return not line or line.isspace()
+
+
+def CheckForFunctionLengths(filename, clean_lines, linenum,
+                            function_state, error):
+  """Reports for long function bodies.
+
+  For an overview why this is done, see:
+  http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml#Write_Short_Functions
+
+  Uses a simplistic algorithm assuming other style guidelines
+  (especially spacing) are followed.
+  Only checks unindented functions, so class members are unchecked.
+  Trivial bodies are unchecked, so constructors with huge initializer lists
+  may be missed.
+  Blank/comment lines are not counted so as to avoid encouraging the removal
+  of vertical space and comments just to get through a lint check.
+  NOLINT *on the last line of a function* disables this check.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    function_state: Current function name and lines in body so far.
+    error: The function to call with any errors found.
+  """
+  lines = clean_lines.lines
+  line = lines[linenum]
+  raw = clean_lines.raw_lines
+  raw_line = raw[linenum]
+  joined_line = ''
+
+  starting_func = False
+  regexp = r'(\w(\w|::|\*|\&|\s)*)\('  # decls * & space::name( ...
+  match_result = Match(regexp, line)
+  if match_result:
+    # If the name is all caps and underscores, figure it's a macro and
+    # ignore it, unless it's TEST or TEST_F.
+    function_name = match_result.group(1).split()[-1]
+    if function_name == 'TEST' or function_name == 'TEST_F' or (
+        not Match(r'[A-Z_]+$', function_name)):
+      starting_func = True
+
+  if starting_func:
+    body_found = False
+    for start_linenum in xrange(linenum, clean_lines.NumLines()):
+      start_line = lines[start_linenum]
+      joined_line += ' ' + start_line.lstrip()
+      if Search(r'(;|})', start_line):  # Declarations and trivial functions
+        body_found = True
+        break                              # ... ignore
+      elif Search(r'{', start_line):
+        body_found = True
+        function = Search(r'((\w|:)*)\(', line).group(1)
+        if Match(r'TEST', function):    # Handle TEST... macros
+          parameter_regexp = Search(r'(\(.*\))', joined_line)
+          if parameter_regexp:             # Ignore bad syntax
+            function += parameter_regexp.group(1)
+        else:
+          function += '()'
+        function_state.Begin(function)
+        break
+    if not body_found:
+      # No body for the function (or evidence of a non-function) was found.
+      error(filename, linenum, 'readability/fn_size', 5,
+            'Lint failed to find start of function body.')
+  elif Match(r'^\}\s*$', line):  # function end
+    function_state.Check(error, filename, linenum)
+    function_state.End()
+  elif not Match(r'^\s*$', line):
+    function_state.Count()  # Count non-blank/non-comment lines.
+
+
+_RE_PATTERN_TODO = re.compile(r'^//(\s*)TODO(\(.+?\))?:?(\s|$)?')
+
+
+def CheckComment(comment, filename, linenum, error):
+  """Checks for common mistakes in TODO comments.
+
+  Args:
+    comment: The text of the comment from the line in question.
+    filename: The name of the current file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  match = _RE_PATTERN_TODO.match(comment)
+  if match:
+    # One whitespace is correct; zero whitespace is handled elsewhere.
+    leading_whitespace = match.group(1)
+    if len(leading_whitespace) > 1:
+      error(filename, linenum, 'whitespace/todo', 2,
+            'Too many spaces before TODO')
+
+    username = match.group(2)
+    if not username:
+      error(filename, linenum, 'readability/todo', 2,
+            'Missing username in TODO; it should look like '
+            '"// TODO(my_username): Stuff."')
+
+    middle_whitespace = match.group(3)
+    # Comparisons made explicit for correctness -- pylint: disable-msg=C6403
+    if middle_whitespace != ' ' and middle_whitespace != '':
+      error(filename, linenum, 'whitespace/todo', 2,
+            'TODO(my_username) should be followed by a space')
+
+def CheckAccess(filename, clean_lines, linenum, nesting_state, error):
+  """Checks for improper use of DISALLOW* macros.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    nesting_state: A _NestingState instance which maintains information about
+                   the current stack of nested blocks being parsed.
+    error: The function to call with any errors found.
+  """
+  line = clean_lines.elided[linenum]  # get rid of comments and strings
+
+  matched = Match((r'\s*(DISALLOW_COPY_AND_ASSIGN|'
+                   r'DISALLOW_EVIL_CONSTRUCTORS|'
+                   r'DISALLOW_IMPLICIT_CONSTRUCTORS)'), line)
+  if not matched:
+    return
+  if nesting_state.stack and isinstance(nesting_state.stack[-1], _ClassInfo):
+    if nesting_state.stack[-1].access != 'private':
+      error(filename, linenum, 'readability/constructors', 3,
+            '%s must be in the private: section' % matched.group(1))
+
+  else:
+    # Found DISALLOW* macro outside a class declaration, or perhaps it
+    # was used inside a function when it should have been part of the
+    # class declaration.  We could issue a warning here, but it
+    # probably resulted in a compiler error already.
+    pass
+
+
+def FindNextMatchingAngleBracket(clean_lines, linenum, init_suffix):
+  """Find the corresponding > to close a template.
+
+  Args:
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: Current line number.
+    init_suffix: Remainder of the current line after the initial <.
+
+  Returns:
+    True if a matching bracket exists.
+  """
+  line = init_suffix
+  nesting_stack = ['<']
+  while True:
+    # Find the next operator that can tell us whether < is used as an
+    # opening bracket or as a less-than operator.  We only want to
+    # warn on the latter case.
+    #
+    # We could also check all other operators and terminate the search
+    # early, e.g. if we got something like this "a<b+c", the "<" is
+    # most likely a less-than operator, but then we will get false
+    # positives for default arguments (e.g. http://go/prccd) and
+    # other template expressions (e.g. http://go/oxcjq).
+    match = Search(r'^[^<>(),;\[\]]*([<>(),;\[\]])(.*)$', line)
+    if match:
+      # Found an operator, update nesting stack
+      operator = match.group(1)
+      line = match.group(2)
+
+      if nesting_stack[-1] == '<':
+        # Expecting closing angle bracket
+        if operator in ('<', '(', '['):
+          nesting_stack.append(operator)
+        elif operator == '>':
+          nesting_stack.pop()
+          if not nesting_stack:
+            # Found matching angle bracket
+            return True
+        elif operator == ',':
+          # Got a comma after a bracket, this is most likely a template
+          # argument.  We have not seen a closing angle bracket yet, but
+          # it's probably a few lines later if we look for it, so just
+          # return early here.
+          return True
+        else:
+          # Got some other operator.
+          return False
+
+      else:
+        # Expecting closing parenthesis or closing bracket
+        if operator in ('<', '(', '['):
+          nesting_stack.append(operator)
+        elif operator in (')', ']'):
+          # We don't bother checking for matching () or [].  If we got
+          # something like (] or [), it would have been a syntax error.
+          nesting_stack.pop()
+
+    else:
+      # Scan the next line
+      linenum += 1
+      if linenum >= len(clean_lines.elided):
+        break
+      line = clean_lines.elided[linenum]
+
+  # Exhausted all remaining lines and still no matching angle bracket.
+  # Most likely the input was incomplete, otherwise we should have
+  # seen a semicolon and returned early.
+  return True
+
+
+def FindPreviousMatchingAngleBracket(clean_lines, linenum, init_prefix):
+  """Find the corresponding < that started a template.
+
+  Args:
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: Current line number.
+    init_prefix: Part of the current line before the initial >.
+
+  Returns:
+    True if a matching bracket exists.
+  """
+  line = init_prefix
+  nesting_stack = ['>']
+  while True:
+    # Find the previous operator
+    match = Search(r'^(.*)([<>(),;\[\]])[^<>(),;\[\]]*$', line)
+    if match:
+      # Found an operator, update nesting stack
+      operator = match.group(2)
+      line = match.group(1)
+
+      if nesting_stack[-1] == '>':
+        # Expecting opening angle bracket
+        if operator in ('>', ')', ']'):
+          nesting_stack.append(operator)
+        elif operator == '<':
+          nesting_stack.pop()
+          if not nesting_stack:
+            # Found matching angle bracket
+            return True
+        elif operator == ',':
+          # Got a comma before a bracket, this is most likely a
+          # template argument.  The opening angle bracket is probably
+          # there if we look for it, so just return early here.
+          return True
+        else:
+          # Got some other operator.
+          return False
+
+      else:
+        # Expecting opening parenthesis or opening bracket
+        if operator in ('>', ')', ']'):
+          nesting_stack.append(operator)
+        elif operator in ('(', '['):
+          nesting_stack.pop()
+
+    else:
+      # Scan the previous line
+      linenum -= 1
+      if linenum < 0:
+        break
+      line = clean_lines.elided[linenum]
+
+  # Exhausted all earlier lines and still no matching angle bracket.
+  return False
+
+
+def CheckSpacing(filename, clean_lines, linenum, nesting_state, error):
+  """Checks for the correctness of various spacing issues in the code.
+
+  Things we check for: spaces around operators, spaces after
+  if/for/while/switch, no spaces around parens in function calls, two
+  spaces between code and comment, don't start a block with a blank
+  line, don't end a function with a blank line, don't add a blank line
+  after public/protected/private, don't have too many blank lines in a row.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    nesting_state: A _NestingState instance which maintains information about
+                   the current stack of nested blocks being parsed.
+    error: The function to call with any errors found.
+  """
+
+  raw = clean_lines.raw_lines
+  line = raw[linenum]
+
+  # Before nixing comments, check if the line is blank for no good
+  # reason.  This includes the first line after a block is opened, and
+  # blank lines at the end of a function (ie, right before a line like '}'
+  #
+  # Skip all the blank line checks if we are immediately inside a
+  # namespace body.  In other words, don't issue blank line warnings
+  # for this block:
+  #   namespace {
+  #
+  #   }
+  #
+  # A warning about missing end of namespace comments will be issued instead.
+  if IsBlankLine(line) and not nesting_state.InNamespaceBody():
+    elided = clean_lines.elided
+    prev_line = elided[linenum - 1]
+    prevbrace = prev_line.rfind('{')
+    # TODO(unknown): Don't complain if line before blank line, and line after,
+    #                both start with alnums and are indented the same amount.
+    #                This ignores whitespace at the start of a namespace block
+    #                because those are not usually indented.
+    if prevbrace != -1 and prev_line[prevbrace:].find('}') == -1:
+      # OK, we have a blank line at the start of a code block.  Before we
+      # complain, we check if it is an exception to the rule: The previous
+      # non-empty line has the parameters of a function header that are indented
+      # 4 spaces (because they did not fit in a 80 column line when placed on
+      # the same line as the function name).  We also check for the case where
+      # the previous line is indented 6 spaces, which may happen when the
+      # initializers of a constructor do not fit into a 80 column line.
+      exception = False
+      if Match(r' {6}\w', prev_line):  # Initializer list?
+        # We are looking for the opening column of initializer list, which
+        # should be indented 4 spaces to cause 6 space indentation afterwards.
+        search_position = linenum-2
+        while (search_position >= 0
+               and Match(r' {6}\w', elided[search_position])):
+          search_position -= 1
+        exception = (search_position >= 0
+                     and elided[search_position][:5] == '    :')
+      else:
+        # Search for the function arguments or an initializer list.  We use a
+        # simple heuristic here: If the line is indented 4 spaces; and we have a
+        # closing paren, without the opening paren, followed by an opening brace
+        # or colon (for initializer lists) we assume that it is the last line of
+        # a function header.  If we have a colon indented 4 spaces, it is an
+        # initializer list.
+        exception = (Match(r' {4}\w[^\(]*\)\s*(const\s*)?(\{\s*$|:)',
+                           prev_line)
+                     or Match(r' {4}:', prev_line))
+
+      if not exception:
+        error(filename, linenum, 'whitespace/blank_line', 2,
+              'Blank line at the start of a code block.  Is this needed?')
+    # Ignore blank lines at the end of a block in a long if-else
+    # chain, like this:
+    #   if (condition1) {
+    #     // Something followed by a blank line
+    #
+    #   } else if (condition2) {
+    #     // Something else
+    #   }
+    if linenum + 1 < clean_lines.NumLines():
+      next_line = raw[linenum + 1]
+      if (next_line
+          and Match(r'\s*}', next_line)
+          and next_line.find('} else ') == -1):
+        error(filename, linenum, 'whitespace/blank_line', 3,
+              'Blank line at the end of a code block.  Is this needed?')
+
+    matched = Match(r'\s*(public|protected|private):', prev_line)
+    if matched:
+      error(filename, linenum, 'whitespace/blank_line', 3,
+            'Do not leave a blank line after "%s:"' % matched.group(1))
+
+  # Next, we complain if there's a comment too near the text
+  commentpos = line.find('//')
+  if commentpos != -1:
+    # Check if the // may be in quotes.  If so, ignore it
+    # Comparisons made explicit for clarity -- pylint: disable-msg=C6403
+    if (line.count('"', 0, commentpos) -
+        line.count('\\"', 0, commentpos)) % 2 == 0:   # not in quotes
+      # Allow one space for new scopes, two spaces otherwise:
+      if (not Match(r'^\s*{ //', line) and
+          ((commentpos >= 1 and
+            line[commentpos-1] not in string.whitespace) or
+           (commentpos >= 2 and
+            line[commentpos-2] not in string.whitespace))):
+        error(filename, linenum, 'whitespace/comments', 2,
+              'At least two spaces is best between code and comments')
+      # There should always be a space between the // and the comment
+      commentend = commentpos + 2
+      if commentend < len(line) and not line[commentend] == ' ':
+        # but some lines are exceptions -- e.g. if they're big
+        # comment delimiters like:
+        # //----------------------------------------------------------
+        # or are an empty C++ style Doxygen comment, like:
+        # ///
+        # or they begin with multiple slashes followed by a space:
+        # //////// Header comment
+        match = (Search(r'[=/-]{4,}\s*$', line[commentend:]) or
+                 Search(r'^/$', line[commentend:]) or
+                 Search(r'^/+ ', line[commentend:]))
+        if not match:
+          error(filename, linenum, 'whitespace/comments', 4,
+                'Should have a space between // and comment')
+      CheckComment(line[commentpos:], filename, linenum, error)
+
+  line = clean_lines.elided[linenum]  # get rid of comments and strings
+
+  # Don't try to do spacing checks for operator methods
+  line = re.sub(r'operator(==|!=|<|<<|<=|>=|>>|>)\(', 'operator\(', line)
+
+  # We allow no-spaces around = within an if: "if ( (a=Foo()) == 0 )".
+  # Otherwise not.  Note we only check for non-spaces on *both* sides;
+  # sometimes people put non-spaces on one side when aligning ='s among
+  # many lines (not that this is behavior that I approve of...)
+  if Search(r'[\w.]=[\w.]', line) and not Search(r'\b(if|while) ', line):
+    error(filename, linenum, 'whitespace/operators', 4,
+          'Missing spaces around =')
+
+  # It's ok not to have spaces around binary operators like + - * /, but if
+  # there's too little whitespace, we get concerned.  It's hard to tell,
+  # though, so we punt on this one for now.  TODO.
+
+  # You should always have whitespace around binary operators.
+  #
+  # Check <= and >= first to avoid false positives with < and >, then
+  # check non-include lines for spacing around < and >.
+  match = Search(r'[^<>=!\s](==|!=|<=|>=)[^<>=!\s]', line)
+  if match:
+    error(filename, linenum, 'whitespace/operators', 3,
+          'Missing spaces around %s' % match.group(1))
+  # We allow no-spaces around << when used like this: 10<<20, but
+  # not otherwise (particularly, not when used as streams)
+  match = Search(r'(\S)(?:L|UL|ULL|l|ul|ull)?<<(\S)', line)
+  if match and not (match.group(1).isdigit() and match.group(2).isdigit()):
+    error(filename, linenum, 'whitespace/operators', 3,
+          'Missing spaces around <<')
+  elif not Match(r'#.*include', line):
+    # Avoid false positives on ->
+    reduced_line = line.replace('->', '')
+
+    # Look for < that is not surrounded by spaces.  This is only
+    # triggered if both sides are missing spaces, even though
+    # technically should should flag if at least one side is missing a
+    # space.  This is done to avoid some false positives with shifts.
+    match = Search(r'[^\s<]<([^\s=<].*)', reduced_line)
+    if (match and
+        not FindNextMatchingAngleBracket(clean_lines, linenum, match.group(1))):
+      error(filename, linenum, 'whitespace/operators', 3,
+            'Missing spaces around <')
+
+    # Look for > that is not surrounded by spaces.  Similar to the
+    # above, we only trigger if both sides are missing spaces to avoid
+    # false positives with shifts.
+    match = Search(r'^(.*[^\s>])>[^\s=>]', reduced_line)
+    if (match and
+        not FindPreviousMatchingAngleBracket(clean_lines, linenum,
+                                             match.group(1))):
+      error(filename, linenum, 'whitespace/operators', 3,
+            'Missing spaces around >')
+
+  # We allow no-spaces around >> for almost anything.  This is because
+  # C++11 allows ">>" to close nested templates, which accounts for
+  # most cases when ">>" is not followed by a space.
+  #
+  # We still warn on ">>" followed by alpha character, because that is
+  # likely due to ">>" being used for right shifts, e.g.:
+  #   value >> alpha
+  #
+  # When ">>" is used to close templates, the alphanumeric letter that
+  # follows would be part of an identifier, and there should still be
+  # a space separating the template type and the identifier.
+  #   type<type<type>> alpha
+  match = Search(r'>>[a-zA-Z_]', line)
+  if match:
+    error(filename, linenum, 'whitespace/operators', 3,
+          'Missing spaces around >>')
+
+  # There shouldn't be space around unary operators
+  match = Search(r'(!\s|~\s|[\s]--[\s;]|[\s]\+\+[\s;])', line)
+  if match:
+    error(filename, linenum, 'whitespace/operators', 4,
+          'Extra space for operator %s' % match.group(1))
+
+  # A pet peeve of mine: no spaces after an if, while, switch, or for
+  match = Search(r' (if\(|for\(|while\(|switch\()', line)
+  if match:
+    error(filename, linenum, 'whitespace/parens', 5,
+          'Missing space before ( in %s' % match.group(1))
+
+  # For if/for/while/switch, the left and right parens should be
+  # consistent about how many spaces are inside the parens, and
+  # there should either be zero or one spaces inside the parens.
+  # We don't want: "if ( foo)" or "if ( foo   )".
+  # Exception: "for ( ; foo; bar)" and "for (foo; bar; )" are allowed.
+  match = Search(r'\b(if|for|while|switch)\s*'
+                 r'\(([ ]*)(.).*[^ ]+([ ]*)\)\s*{\s*$',
+                 line)
+  if match:
+    if len(match.group(2)) != len(match.group(4)):
+      if not (match.group(3) == ';' and
+              len(match.group(2)) == 1 + len(match.group(4)) or
+              not match.group(2) and Search(r'\bfor\s*\(.*; \)', line)):
+        error(filename, linenum, 'whitespace/parens', 5,
+              'Mismatching spaces inside () in %s' % match.group(1))
+    if not len(match.group(2)) in [0, 1]:
+      error(filename, linenum, 'whitespace/parens', 5,
+            'Should have zero or one spaces inside ( and ) in %s' %
+            match.group(1))
+
+  # You should always have a space after a comma (either as fn arg or operator)
+  if Search(r',[^\s]', line):
+    error(filename, linenum, 'whitespace/comma', 3,
+          'Missing space after ,')
+
+  # You should always have a space after a semicolon
+  # except for few corner cases
+  # TODO(unknown): clarify if 'if (1) { return 1;}' is requires one more
+  # space after ;
+  if Search(r';[^\s};\\)/]', line):
+    error(filename, linenum, 'whitespace/semicolon', 3,
+          'Missing space after ;')
+
+  # Next we will look for issues with function calls.
+  CheckSpacingForFunctionCall(filename, line, linenum, error)
+
+  # Except after an opening paren, or after another opening brace (in case of
+  # an initializer list, for instance), you should have spaces before your
+  # braces. And since you should never have braces at the beginning of a line,
+  # this is an easy test.
+  if Search(r'[^ ({]{', line):
+    error(filename, linenum, 'whitespace/braces', 5,
+          'Missing space before {')
+
+  # Make sure '} else {' has spaces.
+  if Search(r'}else', line):
+    error(filename, linenum, 'whitespace/braces', 5,
+          'Missing space before else')
+
+  # You shouldn't have spaces before your brackets, except maybe after
+  # 'delete []' or 'new char * []'.
+  if Search(r'\w\s+\[', line) and not Search(r'delete\s+\[', line):
+    error(filename, linenum, 'whitespace/braces', 5,
+          'Extra space before [')
+
+  # You shouldn't have a space before a semicolon at the end of the line.
+  # There's a special case for "for" since the style guide allows space before
+  # the semicolon there.
+  if Search(r':\s*;\s*$', line):
+    error(filename, linenum, 'whitespace/semicolon', 5,
+          'Semicolon defining empty statement. Use {} instead.')
+  elif Search(r'^\s*;\s*$', line):
+    error(filename, linenum, 'whitespace/semicolon', 5,
+          'Line contains only semicolon. If this should be an empty statement, '
+          'use {} instead.')
+  elif (Search(r'\s+;\s*$', line) and
+        not Search(r'\bfor\b', line)):
+    error(filename, linenum, 'whitespace/semicolon', 5,
+          'Extra space before last semicolon. If this should be an empty '
+          'statement, use {} instead.')
+
+  # In range-based for, we wanted spaces before and after the colon, but
+  # not around "::" tokens that might appear.
+  if (Search('for *\(.*[^:]:[^: ]', line) or
+      Search('for *\(.*[^: ]:[^:]', line)):
+    error(filename, linenum, 'whitespace/forcolon', 2,
+          'Missing space around colon in range-based for loop')
+
+
+def CheckSectionSpacing(filename, clean_lines, class_info, linenum, error):
+  """Checks for additional blank line issues related to sections.
+
+  Currently the only thing checked here is blank line before protected/private.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    class_info: A _ClassInfo objects.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  # Skip checks if the class is small, where small means 25 lines or less.
+  # 25 lines seems like a good cutoff since that's the usual height of
+  # terminals, and any class that can't fit in one screen can't really
+  # be considered "small".
+  #
+  # Also skip checks if we are on the first line.  This accounts for
+  # classes that look like
+  #   class Foo { public: ... };
+  #
+  # If we didn't find the end of the class, last_line would be zero,
+  # and the check will be skipped by the first condition.
+  if (class_info.last_line - class_info.starting_linenum <= 24 or
+      linenum <= class_info.starting_linenum):
+    return
+
+  matched = Match(r'\s*(public|protected|private):', clean_lines.lines[linenum])
+  if matched:
+    # Issue warning if the line before public/protected/private was
+    # not a blank line, but don't do this if the previous line contains
+    # "class" or "struct".  This can happen two ways:
+    #  - We are at the beginning of the class.
+    #  - We are forward-declaring an inner class that is semantically
+    #    private, but needed to be public for implementation reasons.
+    # Also ignores cases where the previous line ends with a backslash as can be
+    # common when defining classes in C macros.
+    prev_line = clean_lines.lines[linenum - 1]
+    if (not IsBlankLine(prev_line) and
+        not Search(r'\b(class|struct)\b', prev_line) and
+        not Search(r'\\$', prev_line)):
+      # Try a bit harder to find the beginning of the class.  This is to
+      # account for multi-line base-specifier lists, e.g.:
+      #   class Derived
+      #       : public Base {
+      end_class_head = class_info.starting_linenum
+      for i in range(class_info.starting_linenum, linenum):
+        if Search(r'\{\s*$', clean_lines.lines[i]):
+          end_class_head = i
+          break
+      if end_class_head < linenum - 1:
+        error(filename, linenum, 'whitespace/blank_line', 3,
+              '"%s:" should be preceded by a blank line' % matched.group(1))
+
+
+def GetPreviousNonBlankLine(clean_lines, linenum):
+  """Return the most recent non-blank line and its line number.
+
+  Args:
+    clean_lines: A CleansedLines instance containing the file contents.
+    linenum: The number of the line to check.
+
+  Returns:
+    A tuple with two elements.  The first element is the contents of the last
+    non-blank line before the current line, or the empty string if this is the
+    first non-blank line.  The second is the line number of that line, or -1
+    if this is the first non-blank line.
+  """
+
+  prevlinenum = linenum - 1
+  while prevlinenum >= 0:
+    prevline = clean_lines.elided[prevlinenum]
+    if not IsBlankLine(prevline):     # if not a blank line...
+      return (prevline, prevlinenum)
+    prevlinenum -= 1
+  return ('', -1)
+
+
+def CheckBraces(filename, clean_lines, linenum, error):
+  """Looks for misplaced braces (e.g. at the end of line).
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+
+  line = clean_lines.elided[linenum]        # get rid of comments and strings
+
+  if Match(r'\s*{\s*$', line):
+    # We allow an open brace to start a line in the case where someone
+    # is using braces in a block to explicitly create a new scope,
+    # which is commonly used to control the lifetime of
+    # stack-allocated variables.  We don't detect this perfectly: we
+    # just don't complain if the last non-whitespace character on the
+    # previous non-blank line is ';', ':', '{', or '}', or if the previous
+    # line starts a preprocessor block.
+    prevline = GetPreviousNonBlankLine(clean_lines, linenum)[0]
+    if (not Search(r'[;:}{]\s*$', prevline) and
+        not Match(r'\s*#', prevline)):
+      error(filename, linenum, 'whitespace/braces', 4,
+            '{ should almost always be at the end of the previous line')
+
+  # An else clause should be on the same line as the preceding closing brace.
+  if Match(r'\s*else\s*', line):
+    prevline = GetPreviousNonBlankLine(clean_lines, linenum)[0]
+    if Match(r'\s*}\s*$', prevline):
+      error(filename, linenum, 'whitespace/newline', 4,
+            'An else should appear on the same line as the preceding }')
+
+  # If braces come on one side of an else, they should be on both.
+  # However, we have to worry about "else if" that spans multiple lines!
+  if Search(r'}\s*else[^{]*$', line) or Match(r'[^}]*else\s*{', line):
+    if Search(r'}\s*else if([^{]*)$', line):       # could be multi-line if
+      # find the ( after the if
+      pos = line.find('else if')
+      pos = line.find('(', pos)
+      if pos > 0:
+        (endline, _, endpos) = CloseExpression(clean_lines, linenum, pos)
+        if endline[endpos:].find('{') == -1:    # must be brace after if
+          error(filename, linenum, 'readability/braces', 5,
+                'If an else has a brace on one side, it should have it on both')
+    else:            # common case: else not followed by a multi-line if
+      error(filename, linenum, 'readability/braces', 5,
+            'If an else has a brace on one side, it should have it on both')
+
+  # Likewise, an else should never have the else clause on the same line
+  if Search(r'\belse [^\s{]', line) and not Search(r'\belse if\b', line):
+    error(filename, linenum, 'whitespace/newline', 4,
+          'Else clause should never be on same line as else (use 2 lines)')
+
+  # In the same way, a do/while should never be on one line
+  if Match(r'\s*do [^\s{]', line):
+    error(filename, linenum, 'whitespace/newline', 4,
+          'do/while clauses should not be on a single line')
+
+  # Braces shouldn't be followed by a ; unless they're defining a struct
+  # or initializing an array.
+  # We can't tell in general, but we can for some common cases.
+  prevlinenum = linenum
+  while True:
+    (prevline, prevlinenum) = GetPreviousNonBlankLine(clean_lines, prevlinenum)
+    if Match(r'\s+{.*}\s*;', line) and not prevline.count(';'):
+      line = prevline + line
+    else:
+      break
+  if (Search(r'{.*}\s*;', line) and
+      line.count('{') == line.count('}') and
+      not Search(r'struct|class|enum|\s*=\s*{', line)):
+    error(filename, linenum, 'readability/braces', 4,
+          "You don't need a ; after a }")
+
+
+def CheckEmptyLoopBody(filename, clean_lines, linenum, error):
+  """Loop for empty loop body with only a single semicolon.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+
+  # Search for loop keywords at the beginning of the line.  Because only
+  # whitespaces are allowed before the keywords, this will also ignore most
+  # do-while-loops, since those lines should start with closing brace.
+  line = clean_lines.elided[linenum]
+  if Match(r'\s*(for|while)\s*\(', line):
+    # Find the end of the conditional expression
+    (end_line, end_linenum, end_pos) = CloseExpression(
+        clean_lines, linenum, line.find('('))
+
+    # Output warning if what follows the condition expression is a semicolon.
+    # No warning for all other cases, including whitespace or newline, since we
+    # have a separate check for semicolons preceded by whitespace.
+    if end_pos >= 0 and Match(r';', end_line[end_pos:]):
+      error(filename, end_linenum, 'whitespace/empty_loop_body', 5,
+            'Empty loop bodies should use {} or continue')
+
+
+def ReplaceableCheck(operator, macro, line):
+  """Determine whether a basic CHECK can be replaced with a more specific one.
+
+  For example suggest using CHECK_EQ instead of CHECK(a == b) and
+  similarly for CHECK_GE, CHECK_GT, CHECK_LE, CHECK_LT, CHECK_NE.
+
+  Args:
+    operator: The C++ operator used in the CHECK.
+    macro: The CHECK or EXPECT macro being called.
+    line: The current source line.
+
+  Returns:
+    True if the CHECK can be replaced with a more specific one.
+  """
+
+  # This matches decimal and hex integers, strings, and chars (in that order).
+  match_constant = r'([-+]?(\d+|0[xX][0-9a-fA-F]+)[lLuU]{0,3}|".*"|\'.*\')'
+
+  # Expression to match two sides of the operator with something that
+  # looks like a literal, since CHECK(x == iterator) won't compile.
+  # This means we can't catch all the cases where a more specific
+  # CHECK is possible, but it's less annoying than dealing with
+  # extraneous warnings.
+  match_this = (r'\s*' + macro + r'\((\s*' +
+                match_constant + r'\s*' + operator + r'[^<>].*|'
+                r'.*[^<>]' + operator + r'\s*' + match_constant +
+                r'\s*\))')
+
+  # Don't complain about CHECK(x == NULL) or similar because
+  # CHECK_EQ(x, NULL) won't compile (requires a cast).
+  # Also, don't complain about more complex boolean expressions
+  # involving && or || such as CHECK(a == b || c == d).
+  return Match(match_this, line) and not Search(r'NULL|&&|\|\|', line)
+
+
+def CheckCheck(filename, clean_lines, linenum, error):
+  """Checks the use of CHECK and EXPECT macros.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+
+  # Decide the set of replacement macros that should be suggested
+  raw_lines = clean_lines.raw_lines
+  current_macro = ''
+  for macro in _CHECK_MACROS:
+    if raw_lines[linenum].find(macro) >= 0:
+      current_macro = macro
+      break
+  if not current_macro:
+    # Don't waste time here if line doesn't contain 'CHECK' or 'EXPECT'
+    return
+
+  line = clean_lines.elided[linenum]        # get rid of comments and strings
+
+  # Encourage replacing plain CHECKs with CHECK_EQ/CHECK_NE/etc.
+  for operator in ['==', '!=', '>=', '>', '<=', '<']:
+    if ReplaceableCheck(operator, current_macro, line):
+      error(filename, linenum, 'readability/check', 2,
+            'Consider using %s instead of %s(a %s b)' % (
+                _CHECK_REPLACEMENT[current_macro][operator],
+                current_macro, operator))
+      break
+
+
+def CheckAltTokens(filename, clean_lines, linenum, error):
+  """Check alternative keywords being used in boolean expressions.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  line = clean_lines.elided[linenum]
+
+  # Avoid preprocessor lines
+  if Match(r'^\s*#', line):
+    return
+
+  # Last ditch effort to avoid multi-line comments.  This will not help
+  # if the comment started before the current line or ended after the
+  # current line, but it catches most of the false positives.  At least,
+  # it provides a way to workaround this warning for people who use
+  # multi-line comments in preprocessor macros.
+  #
+  # TODO(unknown): remove this once cpplint has better support for
+  # multi-line comments.
+  if line.find('/*') >= 0 or line.find('*/') >= 0:
+    return
+
+  for match in _ALT_TOKEN_REPLACEMENT_PATTERN.finditer(line):
+    error(filename, linenum, 'readability/alt_tokens', 2,
+          'Use operator %s instead of %s' % (
+              _ALT_TOKEN_REPLACEMENT[match.group(1)], match.group(1)))
+
+
+def GetLineWidth(line):
+  """Determines the width of the line in column positions.
+
+  Args:
+    line: A string, which may be a Unicode string.
+
+  Returns:
+    The width of the line in column positions, accounting for Unicode
+    combining characters and wide characters.
+  """
+  if isinstance(line, unicode):
+    width = 0
+    for uc in unicodedata.normalize('NFC', line):
+      if unicodedata.east_asian_width(uc) in ('W', 'F'):
+        width += 2
+      elif not unicodedata.combining(uc):
+        width += 1
+    return width
+  else:
+    return len(line)
+
+
+def CheckStyle(filename, clean_lines, linenum, file_extension, nesting_state,
+               error):
+  """Checks rules from the 'C++ style rules' section of cppguide.html.
+
+  Most of these rules are hard to test (naming, comment style), but we
+  do what we can.  In particular we check for 2-space indents, line lengths,
+  tab usage, spaces inside code, etc.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    file_extension: The extension (without the dot) of the filename.
+    nesting_state: A _NestingState instance which maintains information about
+                   the current stack of nested blocks being parsed.
+    error: The function to call with any errors found.
+  """
+
+  raw_lines = clean_lines.raw_lines
+  line = raw_lines[linenum]
+
+  if line.find('\t') != -1:
+    error(filename, linenum, 'whitespace/tab', 1,
+          'Tab found; better to use spaces')
+
+  # One or three blank spaces at the beginning of the line is weird; it's
+  # hard to reconcile that with 2-space indents.
+  # NOTE: here are the conditions rob pike used for his tests.  Mine aren't
+  # as sophisticated, but it may be worth becoming so:  RLENGTH==initial_spaces
+  # if(RLENGTH > 20) complain = 0;
+  # if(match($0, " +(error|private|public|protected):")) complain = 0;
+  # if(match(prev, "&& *$")) complain = 0;
+  # if(match(prev, "\\|\\| *$")) complain = 0;
+  # if(match(prev, "[\",=><] *$")) complain = 0;
+  # if(match($0, " <<")) complain = 0;
+  # if(match(prev, " +for \\(")) complain = 0;
+  # if(prevodd && match(prevprev, " +for \\(")) complain = 0;
+  initial_spaces = 0
+  cleansed_line = clean_lines.elided[linenum]
+  while initial_spaces < len(line) and line[initial_spaces] == ' ':
+    initial_spaces += 1
+  if line and line[-1].isspace():
+    error(filename, linenum, 'whitespace/end_of_line', 4,
+          'Line ends in whitespace.  Consider deleting these extra spaces.')
+  # There are certain situations we allow one space, notably for labels
+  elif ((initial_spaces == 1 or initial_spaces == 3) and
+        not Match(r'\s*\w+\s*:\s*$', cleansed_line)):
+    error(filename, linenum, 'whitespace/indent', 3,
+          'Weird number of spaces at line-start.  '
+          'Are you using a 2-space indent?')
+  # Labels should always be indented at least one space.
+  elif not initial_spaces and line[:2] != '//' and Search(r'[^:]:\s*$',
+                                                          line):
+    error(filename, linenum, 'whitespace/labels', 4,
+          'Labels should always be indented at least one space.  '
+          'If this is a member-initializer list in a constructor or '
+          'the base class list in a class definition, the colon should '
+          'be on the following line.')
+
+
+  # Check if the line is a header guard.
+  is_header_guard = False
+  if file_extension == 'h':
+    cppvar = GetHeaderGuardCPPVariable(filename)
+    if (line.startswith('#ifndef %s' % cppvar) or
+        line.startswith('#define %s' % cppvar) or
+        line.startswith('#endif  // %s' % cppvar)):
+      is_header_guard = True
+  # #include lines and header guards can be long, since there's no clean way to
+  # split them.
+  #
+  # URLs can be long too.  It's possible to split these, but it makes them
+  # harder to cut&paste.
+  #
+  # The "$Id:...$" comment may also get very long without it being the
+  # developers fault.
+  if (not line.startswith('#include') and not is_header_guard and
+      not Match(r'^\s*//.*http(s?)://\S*$', line) and
+      not Match(r'^// \$Id:.*#[0-9]+ \$$', line)):
+    line_width = GetLineWidth(line)
+    if line_width > 100:
+      error(filename, linenum, 'whitespace/line_length', 4,
+            'Lines should very rarely be longer than 100 characters')
+    elif line_width > 80:
+      error(filename, linenum, 'whitespace/line_length', 2,
+            'Lines should be <= 80 characters long')
+
+  if (cleansed_line.count(';') > 1 and
+      # for loops are allowed two ;'s (and may run over two lines).
+      cleansed_line.find('for') == -1 and
+      (GetPreviousNonBlankLine(clean_lines, linenum)[0].find('for') == -1 or
+       GetPreviousNonBlankLine(clean_lines, linenum)[0].find(';') != -1) and
+      # It's ok to have many commands in a switch case that fits in 1 line
+      not ((cleansed_line.find('case ') != -1 or
+            cleansed_line.find('default:') != -1) and
+           cleansed_line.find('break;') != -1)):
+    error(filename, linenum, 'whitespace/newline', 0,
+          'More than one command on the same line')
+
+  # Some more style checks
+  CheckBraces(filename, clean_lines, linenum, error)
+  CheckEmptyLoopBody(filename, clean_lines, linenum, error)
+  CheckAccess(filename, clean_lines, linenum, nesting_state, error)
+  CheckSpacing(filename, clean_lines, linenum, nesting_state, error)
+  CheckCheck(filename, clean_lines, linenum, error)
+  CheckAltTokens(filename, clean_lines, linenum, error)
+  classinfo = nesting_state.InnermostClass()
+  if classinfo:
+    CheckSectionSpacing(filename, clean_lines, classinfo, linenum, error)
+
+
+_RE_PATTERN_INCLUDE_NEW_STYLE = re.compile(r'#include +"[^/]+\.h"')
+_RE_PATTERN_INCLUDE = re.compile(r'^\s*#\s*include\s*([<"])([^>"]*)[>"].*$')
+# Matches the first component of a filename delimited by -s and _s. That is:
+#  _RE_FIRST_COMPONENT.match('foo').group(0) == 'foo'
+#  _RE_FIRST_COMPONENT.match('foo.cc').group(0) == 'foo'
+#  _RE_FIRST_COMPONENT.match('foo-bar_baz.cc').group(0) == 'foo'
+#  _RE_FIRST_COMPONENT.match('foo_bar-baz.cc').group(0) == 'foo'
+_RE_FIRST_COMPONENT = re.compile(r'^[^-_.]+')
+
+
+def _DropCommonSuffixes(filename):
+  """Drops common suffixes like _test.cc or -inl.h from filename.
+
+  For example:
+    >>> _DropCommonSuffixes('foo/foo-inl.h')
+    'foo/foo'
+    >>> _DropCommonSuffixes('foo/bar/foo.cc')
+    'foo/bar/foo'
+    >>> _DropCommonSuffixes('foo/foo_internal.h')
+    'foo/foo'
+    >>> _DropCommonSuffixes('foo/foo_unusualinternal.h')
+    'foo/foo_unusualinternal'
+
+  Args:
+    filename: The input filename.
+
+  Returns:
+    The filename with the common suffix removed.
+  """
+  for suffix in ('test.cc', 'regtest.cc', 'unittest.cc',
+                 'inl.h', 'impl.h', 'internal.h'):
+    if (filename.endswith(suffix) and len(filename) > len(suffix) and
+        filename[-len(suffix) - 1] in ('-', '_')):
+      return filename[:-len(suffix) - 1]
+  return os.path.splitext(filename)[0]
+
+
+def _IsTestFilename(filename):
+  """Determines if the given filename has a suffix that identifies it as a test.
+
+  Args:
+    filename: The input filename.
+
+  Returns:
+    True if 'filename' looks like a test, False otherwise.
+  """
+  if (filename.endswith('_test.cc') or
+      filename.endswith('_unittest.cc') or
+      filename.endswith('_regtest.cc')):
+    return True
+  else:
+    return False
+
+
+def _ClassifyInclude(fileinfo, include, is_system):
+  """Figures out what kind of header 'include' is.
+
+  Args:
+    fileinfo: The current file cpplint is running over. A FileInfo instance.
+    include: The path to a #included file.
+    is_system: True if the #include used <> rather than "".
+
+  Returns:
+    One of the _XXX_HEADER constants.
+
+  For example:
+    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'stdio.h', True)
+    _C_SYS_HEADER
+    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'string', True)
+    _CPP_SYS_HEADER
+    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'foo/foo.h', False)
+    _LIKELY_MY_HEADER
+    >>> _ClassifyInclude(FileInfo('foo/foo_unknown_extension.cc'),
+    ...                  'bar/foo_other_ext.h', False)
+    _POSSIBLE_MY_HEADER
+    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'foo/bar.h', False)
+    _OTHER_HEADER
+  """
+  # This is a list of all standard c++ header files, except
+  # those already checked for above.
+  is_stl_h = include in _STL_HEADERS
+  is_cpp_h = is_stl_h or include in _CPP_HEADERS
+
+  if is_system:
+    if is_cpp_h:
+      return _CPP_SYS_HEADER
+    else:
+      return _C_SYS_HEADER
+
+  # If the target file and the include we're checking share a
+  # basename when we drop common extensions, and the include
+  # lives in . , then it's likely to be owned by the target file.
+  target_dir, target_base = (
+      os.path.split(_DropCommonSuffixes(fileinfo.RepositoryName())))
+  include_dir, include_base = os.path.split(_DropCommonSuffixes(include))
+  if target_base == include_base and (
+      include_dir == target_dir or
+      include_dir == os.path.normpath(target_dir + '/../public')):
+    return _LIKELY_MY_HEADER
+
+  # If the target and include share some initial basename
+  # component, it's possible the target is implementing the
+  # include, so it's allowed to be first, but we'll never
+  # complain if it's not there.
+  target_first_component = _RE_FIRST_COMPONENT.match(target_base)
+  include_first_component = _RE_FIRST_COMPONENT.match(include_base)
+  if (target_first_component and include_first_component and
+      target_first_component.group(0) ==
+      include_first_component.group(0)):
+    return _POSSIBLE_MY_HEADER
+
+  return _OTHER_HEADER
+
+
+
+def CheckIncludeLine(filename, clean_lines, linenum, include_state, error):
+  """Check rules that are applicable to #include lines.
+
+  Strings on #include lines are NOT removed from elided line, to make
+  certain tasks easier. However, to prevent false positives, checks
+  applicable to #include lines in CheckLanguage must be put here.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    include_state: An _IncludeState instance in which the headers are inserted.
+    error: The function to call with any errors found.
+  """
+  fileinfo = FileInfo(filename)
+
+  line = clean_lines.lines[linenum]
+
+  # "include" should use the new style "foo/bar.h" instead of just "bar.h"
+  if _RE_PATTERN_INCLUDE_NEW_STYLE.search(line):
+    error(filename, linenum, 'build/include', 4,
+          'Include the directory when naming .h files')
+
+  # we shouldn't include a file more than once. actually, there are a
+  # handful of instances where doing so is okay, but in general it's
+  # not.
+  match = _RE_PATTERN_INCLUDE.search(line)
+  if match:
+    include = match.group(2)
+    is_system = (match.group(1) == '<')
+    if include in include_state:
+      error(filename, linenum, 'build/include', 4,
+            '"%s" already included at %s:%s' %
+            (include, filename, include_state[include]))
+    else:
+      include_state[include] = linenum
+
+      # We want to ensure that headers appear in the right order:
+      # 1) for foo.cc, foo.h  (preferred location)
+      # 2) c system files
+      # 3) cpp system files
+      # 4) for foo.cc, foo.h  (deprecated location)
+      # 5) other google headers
+      #
+      # We classify each include statement as one of those 5 types
+      # using a number of techniques. The include_state object keeps
+      # track of the highest type seen, and complains if we see a
+      # lower type after that.
+      error_message = include_state.CheckNextIncludeOrder(
+          _ClassifyInclude(fileinfo, include, is_system))
+      if error_message:
+        error(filename, linenum, 'build/include_order', 4,
+              '%s. Should be: %s.h, c system, c++ system, other.' %
+              (error_message, fileinfo.BaseName()))
+      if not include_state.IsInAlphabeticalOrder(include):
+        error(filename, linenum, 'build/include_alpha', 4,
+              'Include "%s" not in alphabetical order' % include)
+
+  # Look for any of the stream classes that are part of standard C++.
+  match = _RE_PATTERN_INCLUDE.match(line)
+  if match:
+    include = match.group(2)
+    if Match(r'(f|ind|io|i|o|parse|pf|stdio|str|)?stream$', include):
+      # Many unit tests use cout, so we exempt them.
+      if not _IsTestFilename(filename):
+        error(filename, linenum, 'readability/streams', 3,
+              'Streams are highly discouraged.')
+
+
+def _GetTextInside(text, start_pattern):
+  """Retrieves all the text between matching open and close parentheses.
+
+  Given a string of lines and a regular expression string, retrieve all the text
+  following the expression and between opening punctuation symbols like
+  (, [, or {, and the matching close-punctuation symbol. This properly nested
+  occurrences of the punctuations, so for the text like
+    printf(a(), b(c()));
+  a call to _GetTextInside(text, r'printf\(') will return 'a(), b(c())'.
+  start_pattern must match string having an open punctuation symbol at the end.
+
+  Args:
+    text: The lines to extract text. Its comments and strings must be elided.
+           It can be single line and can span multiple lines.
+    start_pattern: The regexp string indicating where to start extracting
+                   the text.
+  Returns:
+    The extracted text.
+    None if either the opening string or ending punctuation could not be found.
+  """
+  # TODO(sugawarayu): Audit cpplint.py to see what places could be profitably
+  # rewritten to use _GetTextInside (and use inferior regexp matching today).
+
+  # Give opening punctuations to get the matching close-punctuations.
+  matching_punctuation = {'(': ')', '{': '}', '[': ']'}
+  closing_punctuation = set(matching_punctuation.itervalues())
+
+  # Find the position to start extracting text.
+  match = re.search(start_pattern, text, re.M)
+  if not match:  # start_pattern not found in text.
+    return None
+  start_position = match.end(0)
+
+  assert start_position > 0, (
+      'start_pattern must ends with an opening punctuation.')
+  assert text[start_position - 1] in matching_punctuation, (
+      'start_pattern must ends with an opening punctuation.')
+  # Stack of closing punctuations we expect to have in text after position.
+  punctuation_stack = [matching_punctuation[text[start_position - 1]]]
+  position = start_position
+  while punctuation_stack and position < len(text):
+    if text[position] == punctuation_stack[-1]:
+      punctuation_stack.pop()
+    elif text[position] in closing_punctuation:
+      # A closing punctuation without matching opening punctuations.
+      return None
+    elif text[position] in matching_punctuation:
+      punctuation_stack.append(matching_punctuation[text[position]])
+    position += 1
+  if punctuation_stack:
+    # Opening punctuations left without matching close-punctuations.
+    return None
+  # punctuations match.
+  return text[start_position:position - 1]
+
+
+def CheckLanguage(filename, clean_lines, linenum, file_extension, include_state,
+                  error):
+  """Checks rules from the 'C++ language rules' section of cppguide.html.
+
+  Some of these rules are hard to test (function overloading, using
+  uint32 inappropriately), but we do the best we can.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    file_extension: The extension (without the dot) of the filename.
+    include_state: An _IncludeState instance in which the headers are inserted.
+    error: The function to call with any errors found.
+  """
+  # If the line is empty or consists of entirely a comment, no need to
+  # check it.
+  line = clean_lines.elided[linenum]
+  if not line:
+    return
+
+  match = _RE_PATTERN_INCLUDE.search(line)
+  if match:
+    CheckIncludeLine(filename, clean_lines, linenum, include_state, error)
+    return
+
+  # Create an extended_line, which is the concatenation of the current and
+  # next lines, for more effective checking of code that may span more than one
+  # line.
+  if linenum + 1 < clean_lines.NumLines():
+    extended_line = line + clean_lines.elided[linenum + 1]
+  else:
+    extended_line = line
+
+  # Make Windows paths like Unix.
+  fullname = os.path.abspath(filename).replace('\\', '/')
+
+  # TODO(unknown): figure out if they're using default arguments in fn proto.
+
+  # Check for non-const references in functions.  This is tricky because &
+  # is also used to take the address of something.  We allow <> for templates,
+  # (ignoring whatever is between the braces) and : for classes.
+  # These are complicated re's.  They try to capture the following:
+  # paren (for fn-prototype start), typename, &, varname.  For the const
+  # version, we're willing for const to be before typename or after
+  # Don't check the implementation on same line.
+  fnline = line.split('{', 1)[0]
+  if (len(re.findall(r'\([^()]*\b(?:[\w:]|<[^()]*>)+(\s?&|&\s?)\w+', fnline)) >
+      len(re.findall(r'\([^()]*\bconst\s+(?:typename\s+)?(?:struct\s+)?'
+                     r'(?:[\w:]|<[^()]*>)+(\s?&|&\s?)\w+', fnline)) +
+      len(re.findall(r'\([^()]*\b(?:[\w:]|<[^()]*>)+\s+const(\s?&|&\s?)[\w]+',
+                     fnline))):
+
+    # We allow non-const references in a few standard places, like functions
+    # called "swap()" or iostream operators like "<<" or ">>". We also filter
+    # out for loops, which lint otherwise mistakenly thinks are functions.
+    if not Search(
+        r'(for|swap|Swap|operator[<>][<>])\s*\(\s*'
+        r'(?:(?:typename\s*)?[\w:]|<.*>)+\s*&',
+        fnline):
+      error(filename, linenum, 'runtime/references', 2,
+            'Is this a non-const reference? '
+            'If so, make const or use a pointer.')
+
+  # Check to see if they're using an conversion function cast.
+  # I just try to capture the most common basic types, though there are more.
+  # Parameterless conversion functions, such as bool(), are allowed as they are
+  # probably a member operator declaration or default constructor.
+  match = Search(
+      r'(\bnew\s+)?\b'  # Grab 'new' operator, if it's there
+      r'(int|float|double|bool|char|int32|uint32|int64|uint64)\([^)]', line)
+  if match:
+    # gMock methods are defined using some variant of MOCK_METHODx(name, type)
+    # where type may be float(), int(string), etc.  Without context they are
+    # virtually indistinguishable from int(x) casts. Likewise, gMock's
+    # MockCallback takes a template parameter of the form return_type(arg_type),
+    # which looks much like the cast we're trying to detect.
+    if (match.group(1) is None and  # If new operator, then this isn't a cast
+        not (Match(r'^\s*MOCK_(CONST_)?METHOD\d+(_T)?\(', line) or
+             Match(r'^\s*MockCallback<.*>', line))):
+      # Try a bit harder to catch gmock lines: the only place where
+      # something looks like an old-style cast is where we declare the
+      # return type of the mocked method, and the only time when we
+      # are missing context is if MOCK_METHOD was split across
+      # multiple lines (for example http://go/hrfhr ), so we only need
+      # to check the previous line for MOCK_METHOD.
+      if (linenum == 0 or
+          not Match(r'^\s*MOCK_(CONST_)?METHOD\d+(_T)?\(\S+,\s*$',
+                    clean_lines.elided[linenum - 1])):
+        error(filename, linenum, 'readability/casting', 4,
+              'Using deprecated casting style.  '
+              'Use static_cast<%s>(...) instead' %
+              match.group(2))
+
+  CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
+                  'static_cast',
+                  r'\((int|float|double|bool|char|u?int(16|32|64))\)', error)
+
+  # This doesn't catch all cases. Consider (const char * const)"hello".
+  #
+  # (char *) "foo" should always be a const_cast (reinterpret_cast won't
+  # compile).
+  if CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
+                     'const_cast', r'\((char\s?\*+\s?)\)\s*"', error):
+    pass
+  else:
+    # Check pointer casts for other than string constants
+    CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
+                    'reinterpret_cast', r'\((\w+\s?\*+\s?)\)', error)
+
+  # In addition, we look for people taking the address of a cast.  This
+  # is dangerous -- casts can assign to temporaries, so the pointer doesn't
+  # point where you think.
+  if Search(
+      r'(&\([^)]+\)[\w(])|(&(static|dynamic|reinterpret)_cast\b)', line):
+    error(filename, linenum, 'runtime/casting', 4,
+          ('Are you taking an address of a cast?  '
+           'This is dangerous: could be a temp var.  '
+           'Take the address before doing the cast, rather than after'))
+
+  # Check for people declaring static/global STL strings at the top level.
+  # This is dangerous because the C++ language does not guarantee that
+  # globals with constructors are initialized before the first access.
+  match = Match(
+      r'((?:|static +)(?:|const +))string +([a-zA-Z0-9_:]+)\b(.*)',
+      line)
+  # Make sure it's not a function.
+  # Function template specialization looks like: "string foo<Type>(...".
+  # Class template definitions look like: "string Foo<Type>::Method(...".
+  if match and not Match(r'\s*(<.*>)?(::[a-zA-Z0-9_]+)?\s*\(([^"]|$)',
+                         match.group(3)):
+    error(filename, linenum, 'runtime/string', 4,
+          'For a static/global string constant, use a C style string instead: '
+          '"%schar %s[]".' %
+          (match.group(1), match.group(2)))
+
+  # Check that we're not using RTTI outside of testing code.
+  if Search(r'\bdynamic_cast<', line) and not _IsTestFilename(filename):
+    error(filename, linenum, 'runtime/rtti', 5,
+          'Do not use dynamic_cast<>.  If you need to cast within a class '
+          "hierarchy, use static_cast<> to upcast.  Google doesn't support "
+          'RTTI.')
+
+  if Search(r'\b([A-Za-z0-9_]*_)\(\1\)', line):
+    error(filename, linenum, 'runtime/init', 4,
+          'You seem to be initializing a member variable with itself.')
+
+  if file_extension == 'h':
+    # TODO(unknown): check that 1-arg constructors are explicit.
+    #                How to tell it's a constructor?
+    #                (handled in CheckForNonStandardConstructs for now)
+    # TODO(unknown): check that classes have DISALLOW_EVIL_CONSTRUCTORS
+    #                (level 1 error)
+    pass
+
+  # Check if people are using the verboten C basic types.  The only exception
+  # we regularly allow is "unsigned short port" for port.
+  if Search(r'\bshort port\b', line):
+    if not Search(r'\bunsigned short port\b', line):
+      error(filename, linenum, 'runtime/int', 4,
+            'Use "unsigned short" for ports, not "short"')
+  else:
+    match = Search(r'\b(short|long(?! +double)|long long)\b', line)
+    if match:
+      error(filename, linenum, 'runtime/int', 4,
+            'Use int16/int64/etc, rather than the C type %s' % match.group(1))
+
+  # When snprintf is used, the second argument shouldn't be a literal.
+  match = Search(r'snprintf\s*\(([^,]*),\s*([0-9]*)\s*,', line)
+  if match and match.group(2) != '0':
+    # If 2nd arg is zero, snprintf is used to calculate size.
+    error(filename, linenum, 'runtime/printf', 3,
+          'If you can, use sizeof(%s) instead of %s as the 2nd arg '
+          'to snprintf.' % (match.group(1), match.group(2)))
+
+  # Check if some verboten C functions are being used.
+  if Search(r'\bsprintf\b', line):
+    error(filename, linenum, 'runtime/printf', 5,
+          'Never use sprintf.  Use snprintf instead.')
+  match = Search(r'\b(strcpy|strcat)\b', line)
+  if match:
+    error(filename, linenum, 'runtime/printf', 4,
+          'Almost always, snprintf is better than %s' % match.group(1))
+
+  if Search(r'\bsscanf\b', line):
+    error(filename, linenum, 'runtime/printf', 1,
+          'sscanf can be ok, but is slow and can overflow buffers.')
+
+  # Check if some verboten operator overloading is going on
+  # TODO(unknown): catch out-of-line unary operator&:
+  #   class X {};
+  #   int operator&(const X& x) { return 42; }  // unary operator&
+  # The trick is it's hard to tell apart from binary operator&:
+  #   class Y { int operator&(const Y& x) { return 23; } }; // binary operator&
+  if Search(r'\boperator\s*&\s*\(\s*\)', line):
+    error(filename, linenum, 'runtime/operator', 4,
+          'Unary operator& is dangerous.  Do not use it.')
+
+  # Check for suspicious usage of "if" like
+  # } if (a == b) {
+  if Search(r'\}\s*if\s*\(', line):
+    error(filename, linenum, 'readability/braces', 4,
+          'Did you mean "else if"? If not, start a new line for "if".')
+
+  # Check for potential format string bugs like printf(foo).
+  # We constrain the pattern not to pick things like DocidForPrintf(foo).
+  # Not perfect but it can catch printf(foo.c_str()) and printf(foo->c_str())
+  # TODO(sugawarayu): Catch the following case. Need to change the calling
+  # convention of the whole function to process multiple line to handle it.
+  #   printf(
+  #       boy_this_is_a_really_long_variable_that_cannot_fit_on_the_prev_line);
+  printf_args = _GetTextInside(line, r'(?i)\b(string)?printf\s*\(')
+  if printf_args:
+    match = Match(r'([\w.\->()]+)$', printf_args)
+    if match and match.group(1) != '__VA_ARGS__':
+      function_name = re.search(r'\b((?:string)?printf)\s*\(',
+                                line, re.I).group(1)
+      error(filename, linenum, 'runtime/printf', 4,
+            'Potential format string bug. Do %s("%%s", %s) instead.'
+            % (function_name, match.group(1)))
+
+  # Check for potential memset bugs like memset(buf, sizeof(buf), 0).
+  match = Search(r'memset\s*\(([^,]*),\s*([^,]*),\s*0\s*\)', line)
+  if match and not Match(r"^''|-?[0-9]+|0x[0-9A-Fa-f]$", match.group(2)):
+    error(filename, linenum, 'runtime/memset', 4,
+          'Did you mean "memset(%s, 0, %s)"?'
+          % (match.group(1), match.group(2)))
+
+  if Search(r'\busing namespace\b', line):
+    error(filename, linenum, 'build/namespaces', 5,
+          'Do not use namespace using-directives.  '
+          'Use using-declarations instead.')
+
+  # Detect variable-length arrays.
+  match = Match(r'\s*(.+::)?(\w+) [a-z]\w*\[(.+)];', line)
+  if (match and match.group(2) != 'return' and match.group(2) != 'delete' and
+      match.group(3).find(']') == -1):
+    # Split the size using space and arithmetic operators as delimiters.
+    # If any of the resulting tokens are not compile time constants then
+    # report the error.
+    tokens = re.split(r'\s|\+|\-|\*|\/|<<|>>]', match.group(3))
+    is_const = True
+    skip_next = False
+    for tok in tokens:
+      if skip_next:
+        skip_next = False
+        continue
+
+      if Search(r'sizeof\(.+\)', tok): continue
+      if Search(r'arraysize\(\w+\)', tok): continue
+
+      tok = tok.lstrip('(')
+      tok = tok.rstrip(')')
+      if not tok: continue
+      if Match(r'\d+', tok): continue
+      if Match(r'0[xX][0-9a-fA-F]+', tok): continue
+      if Match(r'k[A-Z0-9]\w*', tok): continue
+      if Match(r'(.+::)?k[A-Z0-9]\w*', tok): continue
+      if Match(r'(.+::)?[A-Z][A-Z0-9_]*', tok): continue
+      # A catch all for tricky sizeof cases, including 'sizeof expression',
+      # 'sizeof(*type)', 'sizeof(const type)', 'sizeof(struct StructName)'
+      # requires skipping the next token because we split on ' ' and '*'.
+      if tok.startswith('sizeof'):
+        skip_next = True
+        continue
+      is_const = False
+      break
+    if not is_const:
+      error(filename, linenum, 'runtime/arrays', 1,
+            'Do not use variable-length arrays.  Use an appropriately named '
+            "('k' followed by CamelCase) compile-time constant for the size.")
+
+  # If DISALLOW_EVIL_CONSTRUCTORS, DISALLOW_COPY_AND_ASSIGN, or
+  # DISALLOW_IMPLICIT_CONSTRUCTORS is present, then it should be the last thing
+  # in the class declaration.
+  match = Match(
+      (r'\s*'
+       r'(DISALLOW_(EVIL_CONSTRUCTORS|COPY_AND_ASSIGN|IMPLICIT_CONSTRUCTORS))'
+       r'\(.*\);$'),
+      line)
+  if match and linenum + 1 < clean_lines.NumLines():
+    next_line = clean_lines.elided[linenum + 1]
+    # We allow some, but not all, declarations of variables to be present
+    # in the statement that defines the class.  The [\w\*,\s]* fragment of
+    # the regular expression below allows users to declare instances of
+    # the class or pointers to instances, but not less common types such
+    # as function pointers or arrays.  It's a tradeoff between allowing
+    # reasonable code and avoiding trying to parse more C++ using regexps.
+    if not Search(r'^\s*}[\w\*,\s]*;', next_line):
+      error(filename, linenum, 'readability/constructors', 3,
+            match.group(1) + ' should be the last thing in the class')
+
+  # Check for use of unnamed namespaces in header files.  Registration
+  # macros are typically OK, so we allow use of "namespace {" on lines
+  # that end with backslashes.
+  if (file_extension == 'h'
+      and Search(r'\bnamespace\s*{', line)
+      and line[-1] != '\\'):
+    error(filename, linenum, 'build/namespaces', 4,
+          'Do not use unnamed namespaces in header files.  See '
+          'http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml#Namespaces'
+          ' for more information.')
+
+
+def CheckCStyleCast(filename, linenum, line, raw_line, cast_type, pattern,
+                    error):
+  """Checks for a C-style cast by looking for the pattern.
+
+  This also handles sizeof(type) warnings, due to similarity of content.
+
+  Args:
+    filename: The name of the current file.
+    linenum: The number of the line to check.
+    line: The line of code to check.
+    raw_line: The raw line of code to check, with comments.
+    cast_type: The string for the C++ cast to recommend.  This is either
+      reinterpret_cast, static_cast, or const_cast, depending.
+    pattern: The regular expression used to find C-style casts.
+    error: The function to call with any errors found.
+
+  Returns:
+    True if an error was emitted.
+    False otherwise.
+  """
+  match = Search(pattern, line)
+  if not match:
+    return False
+
+  # e.g., sizeof(int)
+  sizeof_match = Match(r'.*sizeof\s*$', line[0:match.start(1) - 1])
+  if sizeof_match:
+    error(filename, linenum, 'runtime/sizeof', 1,
+          'Using sizeof(type).  Use sizeof(varname) instead if possible')
+    return True
+
+  # operator++(int) and operator--(int)
+  if (line[0:match.start(1) - 1].endswith(' operator++') or
+      line[0:match.start(1) - 1].endswith(' operator--')):
+    return False
+
+  remainder = line[match.end(0):]
+
+  # The close paren is for function pointers as arguments to a function.
+  # eg, void foo(void (*bar)(int));
+  # The semicolon check is a more basic function check; also possibly a
+  # function pointer typedef.
+  # eg, void foo(int); or void foo(int) const;
+  # The equals check is for function pointer assignment.
+  # eg, void *(*foo)(int) = ...
+  # The > is for MockCallback<...> ...
+  #
+  # Right now, this will only catch cases where there's a single argument, and
+  # it's unnamed.  It should probably be expanded to check for multiple
+  # arguments with some unnamed.
+  function_match = Match(r'\s*(\)|=|(const)?\s*(;|\{|throw\(\)|>))', remainder)
+  if function_match:
+    if (not function_match.group(3) or
+        function_match.group(3) == ';' or
+        ('MockCallback<' not in raw_line and
+         '/*' not in raw_line)):
+      error(filename, linenum, 'readability/function', 3,
+            'All parameters should be named in a function')
+    return True
+
+  # At this point, all that should be left is actual casts.
+  error(filename, linenum, 'readability/casting', 4,
+        'Using C-style cast.  Use %s<%s>(...) instead' %
+        (cast_type, match.group(1)))
+
+  return True
+
+
+_HEADERS_CONTAINING_TEMPLATES = (
+    ('<deque>', ('deque',)),
+    ('<functional>', ('unary_function', 'binary_function',
+                      'plus', 'minus', 'multiplies', 'divides', 'modulus',
+                      'negate',
+                      'equal_to', 'not_equal_to', 'greater', 'less',
+                      'greater_equal', 'less_equal',
+                      'logical_and', 'logical_or', 'logical_not',
+                      'unary_negate', 'not1', 'binary_negate', 'not2',
+                      'bind1st', 'bind2nd',
+                      'pointer_to_unary_function',
+                      'pointer_to_binary_function',
+                      'ptr_fun',
+                      'mem_fun_t', 'mem_fun', 'mem_fun1_t', 'mem_fun1_ref_t',
+                      'mem_fun_ref_t',
+                      'const_mem_fun_t', 'const_mem_fun1_t',
+                      'const_mem_fun_ref_t', 'const_mem_fun1_ref_t',
+                      'mem_fun_ref',
+                     )),
+    ('<limits>', ('numeric_limits',)),
+    ('<list>', ('list',)),
+    ('<map>', ('map', 'multimap',)),
+    ('<memory>', ('allocator',)),
+    ('<queue>', ('queue', 'priority_queue',)),
+    ('<set>', ('set', 'multiset',)),
+    ('<stack>', ('stack',)),
+    ('<string>', ('char_traits', 'basic_string',)),
+    ('<utility>', ('pair',)),
+    ('<vector>', ('vector',)),
+
+    # gcc extensions.
+    # Note: std::hash is their hash, ::hash is our hash
+    ('<hash_map>', ('hash_map', 'hash_multimap',)),
+    ('<hash_set>', ('hash_set', 'hash_multiset',)),
+    ('<slist>', ('slist',)),
+    )
+
+_RE_PATTERN_STRING = re.compile(r'\bstring\b')
+
+_re_pattern_algorithm_header = []
+for _template in ('copy', 'max', 'min', 'min_element', 'sort', 'swap',
+                  'transform'):
+  # Match max<type>(..., ...), max(..., ...), but not foo->max, foo.max or
+  # type::max().
+  _re_pattern_algorithm_header.append(
+      (re.compile(r'[^>.]\b' + _template + r'(<.*?>)?\([^\)]'),
+       _template,
+       '<algorithm>'))
+
+_re_pattern_templates = []
+for _header, _templates in _HEADERS_CONTAINING_TEMPLATES:
+  for _template in _templates:
+    _re_pattern_templates.append(
+        (re.compile(r'(\<|\b)' + _template + r'\s*\<'),
+         _template + '<>',
+         _header))
+
+
+def FilesBelongToSameModule(filename_cc, filename_h):
+  """Check if these two filenames belong to the same module.
+
+  The concept of a 'module' here is a as follows:
+  foo.h, foo-inl.h, foo.cc, foo_test.cc and foo_unittest.cc belong to the
+  same 'module' if they are in the same directory.
+  some/path/public/xyzzy and some/path/internal/xyzzy are also considered
+  to belong to the same module here.
+
+  If the filename_cc contains a longer path than the filename_h, for example,
+  '/absolute/path/to/base/sysinfo.cc', and this file would include
+  'base/sysinfo.h', this function also produces the prefix needed to open the
+  header. This is used by the caller of this function to more robustly open the
+  header file. We don't have access to the real include paths in this context,
+  so we need this guesswork here.
+
+  Known bugs: tools/base/bar.cc and base/bar.h belong to the same module
+  according to this implementation. Because of this, this function gives
+  some false positives. This should be sufficiently rare in practice.
+
+  Args:
+    filename_cc: is the path for the .cc file
+    filename_h: is the path for the header path
+
+  Returns:
+    Tuple with a bool and a string:
+    bool: True if filename_cc and filename_h belong to the same module.
+    string: the additional prefix needed to open the header file.
+  """
+
+  if not filename_cc.endswith('.cc'):
+    return (False, '')
+  filename_cc = filename_cc[:-len('.cc')]
+  if filename_cc.endswith('_unittest'):
+    filename_cc = filename_cc[:-len('_unittest')]
+  elif filename_cc.endswith('_test'):
+    filename_cc = filename_cc[:-len('_test')]
+  filename_cc = filename_cc.replace('/public/', '/')
+  filename_cc = filename_cc.replace('/internal/', '/')
+
+  if not filename_h.endswith('.h'):
+    return (False, '')
+  filename_h = filename_h[:-len('.h')]
+  if filename_h.endswith('-inl'):
+    filename_h = filename_h[:-len('-inl')]
+  filename_h = filename_h.replace('/public/', '/')
+  filename_h = filename_h.replace('/internal/', '/')
+
+  files_belong_to_same_module = filename_cc.endswith(filename_h)
+  common_path = ''
+  if files_belong_to_same_module:
+    common_path = filename_cc[:-len(filename_h)]
+  return files_belong_to_same_module, common_path
+
+
+def UpdateIncludeState(filename, include_state, io=codecs):
+  """Fill up the include_state with new includes found from the file.
+
+  Args:
+    filename: the name of the header to read.
+    include_state: an _IncludeState instance in which the headers are inserted.
+    io: The io factory to use to read the file. Provided for testability.
+
+  Returns:
+    True if a header was succesfully added. False otherwise.
+  """
+  headerfile = None
+  try:
+    headerfile = io.open(filename, 'r', 'utf8', 'replace')
+  except IOError:
+    return False
+  linenum = 0
+  for line in headerfile:
+    linenum += 1
+    clean_line = CleanseComments(line)
+    match = _RE_PATTERN_INCLUDE.search(clean_line)
+    if match:
+      include = match.group(2)
+      # The value formatting is cute, but not really used right now.
+      # What matters here is that the key is in include_state.
+      include_state.setdefault(include, '%s:%d' % (filename, linenum))
+  return True
+
+
+def CheckForIncludeWhatYouUse(filename, clean_lines, include_state, error,
+                              io=codecs):
+  """Reports for missing stl includes.
+
+  This function will output warnings to make sure you are including the headers
+  necessary for the stl containers and functions that you use. We only give one
+  reason to include a header. For example, if you use both equal_to<> and
+  less<> in a .h file, only one (the latter in the file) of these will be
+  reported as a reason to include the <functional>.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    include_state: An _IncludeState instance.
+    error: The function to call with any errors found.
+    io: The IO factory to use to read the header file. Provided for unittest
+        injection.
+  """
+  required = {}  # A map of header name to linenumber and the template entity.
+                 # Example of required: { '<functional>': (1219, 'less<>') }
+
+  for linenum in xrange(clean_lines.NumLines()):
+    line = clean_lines.elided[linenum]
+    if not line or line[0] == '#':
+      continue
+
+    # String is special -- it is a non-templatized type in STL.
+    matched = _RE_PATTERN_STRING.search(line)
+    if matched:
+      # Don't warn about strings in non-STL namespaces:
+      # (We check only the first match per line; good enough.)
+      prefix = line[:matched.start()]
+      if prefix.endswith('std::') or not prefix.endswith('::'):
+        required['<string>'] = (linenum, 'string')
+
+    for pattern, template, header in _re_pattern_algorithm_header:
+      if pattern.search(line):
+        required[header] = (linenum, template)
+
+    # The following function is just a speed up, no semantics are changed.
+    if not '<' in line:  # Reduces the cpu time usage by skipping lines.
+      continue
+
+    for pattern, template, header in _re_pattern_templates:
+      if pattern.search(line):
+        required[header] = (linenum, template)
+
+  # The policy is that if you #include something in foo.h you don't need to
+  # include it again in foo.cc. Here, we will look at possible includes.
+  # Let's copy the include_state so it is only messed up within this function.
+  include_state = include_state.copy()
+
+  # Did we find the header for this file (if any) and succesfully load it?
+  header_found = False
+
+  # Use the absolute path so that matching works properly.
+  abs_filename = FileInfo(filename).FullName()
+
+  # For Emacs's flymake.
+  # If cpplint is invoked from Emacs's flymake, a temporary file is generated
+  # by flymake and that file name might end with '_flymake.cc'. In that case,
+  # restore original file name here so that the corresponding header file can be
+  # found.
+  # e.g. If the file name is 'foo_flymake.cc', we should search for 'foo.h'
+  # instead of 'foo_flymake.h'
+  abs_filename = re.sub(r'_flymake\.cc$', '.cc', abs_filename)
+
+  # include_state is modified during iteration, so we iterate over a copy of
+  # the keys.
+  header_keys = include_state.keys()
+  for header in header_keys:
+    (same_module, common_path) = FilesBelongToSameModule(abs_filename, header)
+    fullpath = common_path + header
+    if same_module and UpdateIncludeState(fullpath, include_state, io):
+      header_found = True
+
+  # If we can't find the header file for a .cc, assume it's because we don't
+  # know where to look. In that case we'll give up as we're not sure they
+  # didn't include it in the .h file.
+  # TODO(unknown): Do a better job of finding .h files so we are confident that
+  # not having the .h file means there isn't one.
+  if filename.endswith('.cc') and not header_found:
+    return
+
+  # All the lines have been processed, report the errors found.
+  for required_header_unstripped in required:
+    template = required[required_header_unstripped][1]
+    if required_header_unstripped.strip('<>"') not in include_state:
+      error(filename, required[required_header_unstripped][0],
+            'build/include_what_you_use', 4,
+            'Add #include ' + required_header_unstripped + ' for ' + template)
+
+
+_RE_PATTERN_EXPLICIT_MAKEPAIR = re.compile(r'\bmake_pair\s*<')
+
+
+def CheckMakePairUsesDeduction(filename, clean_lines, linenum, error):
+  """Check that make_pair's template arguments are deduced.
+
+  G++ 4.6 in C++0x mode fails badly if make_pair's template arguments are
+  specified explicitly, and such use isn't intended in any case.
+
+  Args:
+    filename: The name of the current file.
+    clean_lines: A CleansedLines instance containing the file.
+    linenum: The number of the line to check.
+    error: The function to call with any errors found.
+  """
+  raw = clean_lines.raw_lines
+  line = raw[linenum]
+  match = _RE_PATTERN_EXPLICIT_MAKEPAIR.search(line)
+  if match:
+    error(filename, linenum, 'build/explicit_make_pair',
+          4,  # 4 = high confidence
+          'For C++11-compatibility, omit template arguments from make_pair'
+          ' OR use pair directly OR if appropriate, construct a pair directly')
+
+
+def ProcessLine(filename, file_extension, clean_lines, line,
+                include_state, function_state, nesting_state, error,
+                extra_check_functions=[]):
+  """Processes a single line in the file.
+
+  Args:
+    filename: Filename of the file that is being processed.
+    file_extension: The extension (dot not included) of the file.
+    clean_lines: An array of strings, each representing a line of the file,
+                 with comments stripped.
+    line: Number of line being processed.
+    include_state: An _IncludeState instance in which the headers are inserted.
+    function_state: A _FunctionState instance which counts function lines, etc.
+    nesting_state: A _NestingState instance which maintains information about
+                   the current stack of nested blocks being parsed.
+    error: A callable to which errors are reported, which takes 4 arguments:
+           filename, line number, error level, and message
+    extra_check_functions: An array of additional check functions that will be
+                           run on each source line. Each function takes 4
+                           arguments: filename, clean_lines, line, error
+  """
+  raw_lines = clean_lines.raw_lines
+  ParseNolintSuppressions(filename, raw_lines[line], line, error)
+  nesting_state.Update(filename, clean_lines, line, error)
+  if nesting_state.stack and nesting_state.stack[-1].inline_asm != _NO_ASM:
+    return
+  CheckForFunctionLengths(filename, clean_lines, line, function_state, error)
+  CheckForMultilineCommentsAndStrings(filename, clean_lines, line, error)
+  CheckStyle(filename, clean_lines, line, file_extension, nesting_state, error)
+  CheckLanguage(filename, clean_lines, line, file_extension, include_state,
+                error)
+  CheckForNonStandardConstructs(filename, clean_lines, line,
+                                nesting_state, error)
+  CheckPosixThreading(filename, clean_lines, line, error)
+  CheckInvalidIncrement(filename, clean_lines, line, error)
+  CheckMakePairUsesDeduction(filename, clean_lines, line, error)
+  for check_fn in extra_check_functions:
+    check_fn(filename, clean_lines, line, error)
+
+def ProcessFileData(filename, file_extension, lines, error,
+                    extra_check_functions=[]):
+  """Performs lint checks and reports any errors to the given error function.
+
+  Args:
+    filename: Filename of the file that is being processed.
+    file_extension: The extension (dot not included) of the file.
+    lines: An array of strings, each representing a line of the file, with the
+           last element being empty if the file is terminated with a newline.
+    error: A callable to which errors are reported, which takes 4 arguments:
+           filename, line number, error level, and message
+    extra_check_functions: An array of additional check functions that will be
+                           run on each source line. Each function takes 4
+                           arguments: filename, clean_lines, line, error
+  """
+  lines = (['// marker so line numbers and indices both start at 1'] + lines +
+           ['// marker so line numbers end in a known way'])
+
+  include_state = _IncludeState()
+  function_state = _FunctionState()
+  nesting_state = _NestingState()
+
+  ResetNolintSuppressions()
+
+  CheckForCopyright(filename, lines, error)
+
+  if file_extension == 'h':
+    CheckForHeaderGuard(filename, lines, error)
+
+  RemoveMultiLineComments(filename, lines, error)
+  clean_lines = CleansedLines(lines)
+  for line in xrange(clean_lines.NumLines()):
+    ProcessLine(filename, file_extension, clean_lines, line,
+                include_state, function_state, nesting_state, error,
+                extra_check_functions)
+  nesting_state.CheckClassFinished(filename, error)
+
+  CheckForIncludeWhatYouUse(filename, clean_lines, include_state, error)
+
+  # We check here rather than inside ProcessLine so that we see raw
+  # lines rather than "cleaned" lines.
+  CheckForUnicodeReplacementCharacters(filename, lines, error)
+
+  CheckForNewlineAtEOF(filename, lines, error)
+
+def ProcessFile(filename, vlevel, extra_check_functions=[]):
+  """Does google-lint on a single file.
+
+  Args:
+    filename: The name of the file to parse.
+
+    vlevel: The level of errors to report.  Every error of confidence
+    >= verbose_level will be reported.  0 is a good default.
+
+    extra_check_functions: An array of additional check functions that will be
+                           run on each source line. Each function takes 4
+                           arguments: filename, clean_lines, line, error
+  """
+
+  _SetVerboseLevel(vlevel)
+
+  try:
+    # Support the UNIX convention of using "-" for stdin.  Note that
+    # we are not opening the file with universal newline support
+    # (which codecs doesn't support anyway), so the resulting lines do
+    # contain trailing '\r' characters if we are reading a file that
+    # has CRLF endings.
+    # If after the split a trailing '\r' is present, it is removed
+    # below. If it is not expected to be present (i.e. os.linesep !=
+    # '\r\n' as in Windows), a warning is issued below if this file
+    # is processed.
+
+    if filename == '-':
+      lines = codecs.StreamReaderWriter(sys.stdin,
+                                        codecs.getreader('utf8'),
+                                        codecs.getwriter('utf8'),
+                                        'replace').read().split('\n')
+    else:
+      lines = codecs.open(filename, 'r', 'utf8', 'replace').read().split('\n')
+
+    carriage_return_found = False
+    # Remove trailing '\r'.
+    for linenum in range(len(lines)):
+      if lines[linenum].endswith('\r'):
+        lines[linenum] = lines[linenum].rstrip('\r')
+        carriage_return_found = True
+
+  except IOError:
+    sys.stderr.write(
+        "Skipping input '%s': Can't open for reading\n" % filename)
+    return
+
+  # Note, if no dot is found, this will give the entire filename as the ext.
+  file_extension = filename[filename.rfind('.') + 1:]
+
+  # When reading from stdin, the extension is unknown, so no cpplint tests
+  # should rely on the extension.
+  if (filename != '-' and file_extension != 'cc' and file_extension != 'h'
+      and file_extension != 'cpp'):
+    sys.stderr.write('Ignoring %s; not a .cc or .h file\n' % filename)
+  else:
+    ProcessFileData(filename, file_extension, lines, Error,
+                    extra_check_functions)
+    if carriage_return_found and os.linesep != '\r\n':
+      # Use 0 for linenum since outputting only one error for potentially
+      # several lines.
+      Error(filename, 0, 'whitespace/newline', 1,
+            'One or more unexpected \\r (^M) found;'
+            'better to use only a \\n')
+
+  sys.stderr.write('Done processing %s\n' % filename)
+
+
+def PrintUsage(message):
+  """Prints a brief usage string and exits, optionally with an error message.
+
+  Args:
+    message: The optional error message.
+  """
+  sys.stderr.write(_USAGE)
+  if message:
+    sys.exit('\nFATAL ERROR: ' + message)
+  else:
+    sys.exit(1)
+
+
+def PrintCategories():
+  """Prints a list of all the error-categories used by error messages.
+
+  These are the categories used to filter messages via --filter.
+  """
+  sys.stderr.write(''.join('  %s\n' % cat for cat in _ERROR_CATEGORIES))
+  sys.exit(0)
+
+
+def ParseArguments(args):
+  """Parses the command line arguments.
+
+  This may set the output format and verbosity level as side-effects.
+
+  Args:
+    args: The command line arguments:
+
+  Returns:
+    The list of filenames to lint.
+  """
+  try:
+    (opts, filenames) = getopt.getopt(args, '', ['help', 'output=', 'verbose=',
+                                                 'counting=',
+                                                 'filter=',
+                                                 'root='])
+  except getopt.GetoptError:
+    PrintUsage('Invalid arguments.')
+
+  verbosity = _VerboseLevel()
+  output_format = _OutputFormat()
+  filters = ''
+  counting_style = ''
+
+  for (opt, val) in opts:
+    if opt == '--help':
+      PrintUsage(None)
+    elif opt == '--output':
+      if not val in ('emacs', 'vs7'):
+        PrintUsage('The only allowed output formats are emacs and vs7.')
+      output_format = val
+    elif opt == '--verbose':
+      verbosity = int(val)
+    elif opt == '--filter':
+      filters = val
+      if not filters:
+        PrintCategories()
+    elif opt == '--counting':
+      if val not in ('total', 'toplevel', 'detailed'):
+        PrintUsage('Valid counting options are total, toplevel, and detailed')
+      counting_style = val
+    elif opt == '--root':
+      global _root
+      _root = val
+
+  if not filenames:
+    PrintUsage('No files were specified.')
+
+  _SetOutputFormat(output_format)
+  _SetVerboseLevel(verbosity)
+  _SetFilters(filters)
+  _SetCountingStyle(counting_style)
+
+  return filenames
+
+
+def main():
+  filenames = ParseArguments(sys.argv[1:])
+
+  # Change stderr to write with replacement characters so we don't die
+  # if we try to print something containing non-ASCII characters.
+  sys.stderr = codecs.StreamReaderWriter(sys.stderr,
+                                         codecs.getreader('utf8'),
+                                         codecs.getwriter('utf8'),
+                                         'replace')
+
+  _cpplint_state.ResetErrorCounts()
+  for filename in filenames:
+    ProcessFile(filename, _cpplint_state.verbose_level)
+  _cpplint_state.PrintErrorCounts()
+
+  sys.exit(_cpplint_state.error_count > 0)
+
+
+if __name__ == '__main__':
+  main()
diff --git a/libvpx/tools/diff.py b/libvpx/tools/diff.py
new file mode 100644
index 0000000..a42a4dc
--- /dev/null
+++ b/libvpx/tools/diff.py
@@ -0,0 +1,127 @@
+#!/usr/bin/env python
+##  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+"""Classes for representing diff pieces."""
+
+__author__ = "jkoleszar@google.com"
+
+import re
+
+
+class DiffLines(object):
+    """A container for one half of a diff."""
+
+    def __init__(self, filename, offset, length):
+        self.filename = filename
+        self.offset = offset
+        self.length = length
+        self.lines = []
+        self.delta_line_nums = []
+
+    def Append(self, line):
+        l = len(self.lines)
+        if line[0] != " ":
+            self.delta_line_nums.append(self.offset + l)
+        self.lines.append(line[1:])
+        assert l+1 <= self.length
+
+    def Complete(self):
+        return len(self.lines) == self.length
+
+    def __contains__(self, item):
+        return item >= self.offset and item <= self.offset + self.length - 1
+
+
+class DiffHunk(object):
+    """A container for one diff hunk, consisting of two DiffLines."""
+
+    def __init__(self, header, file_a, file_b, start_a, len_a, start_b, len_b):
+        self.header = header
+        self.left = DiffLines(file_a, start_a, len_a)
+        self.right = DiffLines(file_b, start_b, len_b)
+        self.lines = []
+
+    def Append(self, line):
+        """Adds a line to the DiffHunk and its DiffLines children."""
+        if line[0] == "-":
+            self.left.Append(line)
+        elif line[0] == "+":
+            self.right.Append(line)
+        elif line[0] == " ":
+            self.left.Append(line)
+            self.right.Append(line)
+        else:
+            assert False, ("Unrecognized character at start of diff line "
+                           "%r" % line[0])
+        self.lines.append(line)
+
+    def Complete(self):
+        return self.left.Complete() and self.right.Complete()
+
+    def __repr__(self):
+        return "DiffHunk(%s, %s, len %d)" % (
+            self.left.filename, self.right.filename,
+            max(self.left.length, self.right.length))
+
+
+def ParseDiffHunks(stream):
+    """Walk a file-like object, yielding DiffHunks as they're parsed."""
+
+    file_regex = re.compile(r"(\+\+\+|---) (\S+)")
+    range_regex = re.compile(r"@@ -(\d+)(,(\d+))? \+(\d+)(,(\d+))?")
+    hunk = None
+    while True:
+        line = stream.readline()
+        if not line:
+            break
+
+        if hunk is None:
+            # Parse file names
+            diff_file = file_regex.match(line)
+            if diff_file:
+              if line.startswith("---"):
+                  a_line = line
+                  a = diff_file.group(2)
+                  continue
+              if line.startswith("+++"):
+                  b_line = line
+                  b = diff_file.group(2)
+                  continue
+
+            # Parse offset/lengths
+            diffrange = range_regex.match(line)
+            if diffrange:
+                if diffrange.group(2):
+                    start_a = int(diffrange.group(1))
+                    len_a = int(diffrange.group(3))
+                else:
+                    start_a = 1
+                    len_a = int(diffrange.group(1))
+
+                if diffrange.group(5):
+                    start_b = int(diffrange.group(4))
+                    len_b = int(diffrange.group(6))
+                else:
+                    start_b = 1
+                    len_b = int(diffrange.group(4))
+
+                header = [a_line, b_line, line]
+                hunk = DiffHunk(header, a, b, start_a, len_a, start_b, len_b)
+        else:
+            # Add the current line to the hunk
+            hunk.Append(line)
+
+            # See if the whole hunk has been parsed. If so, yield it and prepare
+            # for the next hunk.
+            if hunk.Complete():
+                yield hunk
+                hunk = None
+
+    # Partial hunks are a parse error
+    assert hunk is None
diff --git a/libvpx/tools/ftfy.sh b/libvpx/tools/ftfy.sh
index c5cfdea..92059f5 100755
--- a/libvpx/tools/ftfy.sh
+++ b/libvpx/tools/ftfy.sh
@@ -29,12 +29,13 @@
 
 
 vpx_style() {
-  astyle --style=bsd --min-conditional-indent=0 --break-blocks \
-         --pad-oper --pad-header --unpad-paren \
-         --align-pointer=name \
-         --indent-preprocessor --convert-tabs --indent-labels \
-         --suffix=none --quiet "$@"
-  sed -i "" 's/[[:space:]]\{1,\},/,/g' "$@"
+  for f; do
+    case "$f" in
+      *.h|*.c|*.cc)
+        "${dirname_self}"/vpx-astyle.sh "$f"
+        ;;
+    esac
+  done
 }
 
 
@@ -119,8 +120,7 @@
 git show > "${ORIG_DIFF}"
 
 # Apply the style guide on new and modified files and collect its diff
-for f in $(git diff HEAD^ --name-only -M90 --diff-filter=AM \
-           | grep '\.[ch]$'); do
+for f in $(git diff HEAD^ --name-only -M90 --diff-filter=AM); do
   case "$f" in
     third_party/*) continue;;
     nestegg/*) continue;;
diff --git a/libvpx/tools/intersect-diffs.py b/libvpx/tools/intersect-diffs.py
index be9dea5..4dbafa9 100755
--- a/libvpx/tools/intersect-diffs.py
+++ b/libvpx/tools/intersect-diffs.py
@@ -16,121 +16,9 @@
 
 __author__ = "jkoleszar@google.com"
 
-import re
 import sys
 
-
-class DiffLines(object):
-    """A container for one half of a diff."""
-
-    def __init__(self, filename, offset, length):
-        self.filename = filename
-        self.offset = offset
-        self.length = length
-        self.lines = []
-        self.delta_line_nums = []
-
-    def Append(self, line):
-        l = len(self.lines)
-        if line[0] != " ":
-            self.delta_line_nums.append(self.offset + l)
-        self.lines.append(line[1:])
-        assert l+1 <= self.length
-
-    def Complete(self):
-        return len(self.lines) == self.length
-
-    def __contains__(self, item):
-        return item >= self.offset and item <= self.offset + self.length - 1
-
-
-class DiffHunk(object):
-    """A container for one diff hunk, consisting of two DiffLines."""
-
-    def __init__(self, header, file_a, file_b, start_a, len_a, start_b, len_b):
-        self.header = header
-        self.left = DiffLines(file_a, start_a, len_a)
-        self.right = DiffLines(file_b, start_b, len_b)
-        self.lines = []
-
-    def Append(self, line):
-        """Adds a line to the DiffHunk and its DiffLines children."""
-        if line[0] == "-":
-            self.left.Append(line)
-        elif line[0] == "+":
-            self.right.Append(line)
-        elif line[0] == " ":
-            self.left.Append(line)
-            self.right.Append(line)
-        else:
-            assert False, ("Unrecognized character at start of diff line "
-                           "%r" % line[0])
-        self.lines.append(line)
-
-    def Complete(self):
-        return self.left.Complete() and self.right.Complete()
-
-    def __repr__(self):
-        return "DiffHunk(%s, %s, len %d)" % (
-            self.left.filename, self.right.filename,
-            max(self.left.length, self.right.length))
-
-
-def ParseDiffHunks(stream):
-    """Walk a file-like object, yielding DiffHunks as they're parsed."""
-
-    file_regex = re.compile(r"(\+\+\+|---) (\S+)")
-    range_regex = re.compile(r"@@ -(\d+)(,(\d+))? \+(\d+)(,(\d+))?")
-    hunk = None
-    while True:
-        line = stream.readline()
-        if not line:
-            break
-
-        if hunk is None:
-            # Parse file names
-            diff_file = file_regex.match(line)
-            if diff_file:
-              if line.startswith("---"):
-                  a_line = line
-                  a = diff_file.group(2)
-                  continue
-              if line.startswith("+++"):
-                  b_line = line
-                  b = diff_file.group(2)
-                  continue
-
-            # Parse offset/lengths
-            diffrange = range_regex.match(line)
-            if diffrange:
-                if diffrange.group(2):
-                    start_a = int(diffrange.group(1))
-                    len_a = int(diffrange.group(3))
-                else:
-                    start_a = 1
-                    len_a = int(diffrange.group(1))
-
-                if diffrange.group(5):
-                    start_b = int(diffrange.group(4))
-                    len_b = int(diffrange.group(6))
-                else:
-                    start_b = 1
-                    len_b = int(diffrange.group(4))
-
-                header = [a_line, b_line, line]
-                hunk = DiffHunk(header, a, b, start_a, len_a, start_b, len_b)
-        else:
-            # Add the current line to the hunk
-            hunk.Append(line)
-
-            # See if the whole hunk has been parsed. If so, yield it and prepare
-            # for the next hunk.
-            if hunk.Complete():
-                yield hunk
-                hunk = None
-
-    # Partial hunks are a parse error
-    assert hunk is None
+import diff
 
 
 def FormatDiffHunks(hunks):
@@ -162,8 +50,8 @@
 
 
 def main():
-    old_hunks = [x for x in ParseDiffHunks(open(sys.argv[1], "r"))]
-    new_hunks = [x for x in ParseDiffHunks(open(sys.argv[2], "r"))]
+    old_hunks = [x for x in diff.ParseDiffHunks(open(sys.argv[1], "r"))]
+    new_hunks = [x for x in diff.ParseDiffHunks(open(sys.argv[2], "r"))]
     out_hunks = []
 
     # Join the right hand side of the older diff with the left hand side of the
diff --git a/libvpx/tools/lint-hunks.py b/libvpx/tools/lint-hunks.py
new file mode 100755
index 0000000..b15a691
--- /dev/null
+++ b/libvpx/tools/lint-hunks.py
@@ -0,0 +1,144 @@
+#!/usr/bin/python
+##  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+"""Performs style checking on each diff hunk."""
+import getopt
+import os
+import StringIO
+import subprocess
+import sys
+
+import diff
+
+
+SHORT_OPTIONS = "h"
+LONG_OPTIONS = ["help"]
+
+TOPLEVEL_CMD = ["git", "rev-parse", "--show-toplevel"]
+DIFF_CMD = ["git", "diff"]
+DIFF_INDEX_CMD = ["git", "diff-index", "-u", "HEAD", "--"]
+SHOW_CMD = ["git", "show"]
+CPPLINT_FILTERS = ["-readability/casting", "-runtime/int"]
+
+
+class Usage(Exception):
+    pass
+
+
+class SubprocessException(Exception):
+    def __init__(self, args):
+        msg = "Failed to execute '%s'"%(" ".join(args))
+        super(SubprocessException, self).__init__(msg)
+
+
+class Subprocess(subprocess.Popen):
+    """Adds the notion of an expected returncode to Popen."""
+
+    def __init__(self, args, expected_returncode=0, **kwargs):
+        self._args = args
+        self._expected_returncode = expected_returncode
+        super(Subprocess, self).__init__(args, **kwargs)
+
+    def communicate(self, *args, **kwargs):
+        result = super(Subprocess, self).communicate(*args, **kwargs)
+        if self._expected_returncode is not None:
+            try:
+                ok = self.returncode in self._expected_returncode
+            except TypeError:
+                ok = self.returncode == self._expected_returncode
+            if not ok:
+                raise SubprocessException(self._args)
+        return result
+
+
+def main(argv=None):
+    if argv is None:
+        argv = sys.argv
+    try:
+        try:
+            opts, args = getopt.getopt(argv[1:], SHORT_OPTIONS, LONG_OPTIONS)
+        except getopt.error, msg:
+            raise Usage(msg)
+
+        # process options
+        for o, _ in opts:
+            if o in ("-h", "--help"):
+                print __doc__
+                sys.exit(0)
+
+        if args and len(args) > 1:
+            print __doc__
+            sys.exit(0)
+
+        # Find the fully qualified path to the root of the tree
+        tl = Subprocess(TOPLEVEL_CMD, stdout=subprocess.PIPE)
+        tl = tl.communicate()[0].strip()
+
+        # See if we're working on the index or not.
+        if args:
+            diff_cmd = DIFF_CMD + [args[0] + "^!"]
+        else:
+            diff_cmd = DIFF_INDEX_CMD
+
+        # Build the command line to execute cpplint
+        cpplint_cmd = [os.path.join(tl, "tools", "cpplint.py"),
+                       "--filter=" + ",".join(CPPLINT_FILTERS),
+                       "-"]
+
+        # Get a list of all affected lines
+        file_affected_line_map = {}
+        p = Subprocess(diff_cmd, stdout=subprocess.PIPE)
+        stdout = p.communicate()[0]
+        for hunk in diff.ParseDiffHunks(StringIO.StringIO(stdout)):
+            filename = hunk.right.filename[2:]
+            if filename not in file_affected_line_map:
+                file_affected_line_map[filename] = set()
+            file_affected_line_map[filename].update(hunk.right.delta_line_nums)
+
+        # Run each affected file through cpplint
+        lint_failed = False
+        for filename, affected_lines in file_affected_line_map.iteritems():
+            if filename.split(".")[-1] not in ("c", "h", "cc"):
+                continue
+
+            if args:
+                # File contents come from git
+                show_cmd = SHOW_CMD + [args[0] + ":" + filename]
+                show = Subprocess(show_cmd, stdout=subprocess.PIPE)
+                lint = Subprocess(cpplint_cmd, expected_returncode=(0, 1),
+                                  stdin=show.stdout, stderr=subprocess.PIPE)
+                lint_out = lint.communicate()[1]
+            else:
+                # File contents come from the working tree
+                lint = Subprocess(cpplint_cmd, expected_returncode=(0, 1),
+                                  stdin=subprocess.PIPE, stderr=subprocess.PIPE)
+                stdin = open(os.path.join(tl, filename)).read()
+                lint_out = lint.communicate(stdin)[1]
+
+            for line in lint_out.split("\n"):
+                fields = line.split(":")
+                if fields[0] != "-":
+                    continue
+                warning_line_num = int(fields[1])
+                if warning_line_num in affected_lines:
+                    print "%s:%d:%s"%(filename, warning_line_num,
+                                      ":".join(fields[2:]))
+                    lint_failed = True
+
+        # Set exit code if any relevant lint errors seen
+        if lint_failed:
+            return 1
+
+    except Usage, err:
+        print >>sys.stderr, err
+        print >>sys.stderr, "for help use --help"
+        return 2
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/libvpx/tools/vpx-astyle.sh b/libvpx/tools/vpx-astyle.sh
new file mode 100755
index 0000000..6340426
--- /dev/null
+++ b/libvpx/tools/vpx-astyle.sh
@@ -0,0 +1,27 @@
+#!/bin/sh
+set -e
+astyle --style=java --indent=spaces=2 --indent-switches\
+       --min-conditional-indent=0 \
+       --pad-oper --pad-header --unpad-paren \
+       --align-pointer=name \
+       --indent-preprocessor --convert-tabs --indent-labels \
+       --suffix=none --quiet --max-instatement-indent=80 "$@"
+# Disabled, too greedy?
+#sed -i 's;[[:space:]]\{1,\}\[;[;g' "$@"
+
+sed_i() {
+  # Incompatible sed parameter parsing.
+  if sed -i 2>&1 | grep -q 'requires an argument'; then
+    sed -i '' "$@"
+  else
+    sed -i "$@"
+  fi
+}
+
+sed_i -e 's/[[:space:]]\{1,\}\([,;]\)/\1/g' \
+      -e 's/[[:space:]]\{1,\}\([+-]\{2\};\)/\1/g' \
+      -e 's/,[[:space:]]*}/}/g' \
+      -e 's;//\([^/[:space:]].*$\);// \1;g' \
+      -e 's/^\(public\|private\|protected\):$/ \1:/g' \
+      -e 's/[[:space:]]\{1,\}$//g' \
+      "$@"
diff --git a/libvpx/tools_common.c b/libvpx/tools_common.c
index 6f95028..92de794 100644
--- a/libvpx/tools_common.c
+++ b/libvpx/tools_common.c
@@ -20,11 +20,10 @@
 #endif
 #endif
 
-FILE* set_binary_mode(FILE *stream)
-{
-    (void)stream;
+FILE *set_binary_mode(FILE *stream) {
+  (void)stream;
 #if defined(_WIN32) || defined(__OS2__)
-    _setmode(_fileno(stream), _O_BINARY);
+  _setmode(_fileno(stream), _O_BINARY);
 #endif
-    return stream;
+  return stream;
 }
diff --git a/libvpx/tools_common.h b/libvpx/tools_common.h
index 80c9747..9e56149 100644
--- a/libvpx/tools_common.h
+++ b/libvpx/tools_common.h
@@ -11,6 +11,6 @@
 #define TOOLS_COMMON_H
 
 /* Sets a stdio stream into binary mode */
-FILE* set_binary_mode(FILE *stream);
+FILE *set_binary_mode(FILE *stream);
 
 #endif
diff --git a/libvpx/vp8/common/arm/armv6/filter_v6.asm b/libvpx/vp8/common/arm/armv6/filter_v6.asm
index 1ba91dd..eb4b75b 100644
--- a/libvpx/vp8/common/arm/armv6/filter_v6.asm
+++ b/libvpx/vp8/common/arm/armv6/filter_v6.asm
@@ -394,7 +394,7 @@
     mov     r4, #0x40                       ; rounding factor (for smlad{x})
 
 |height_loop_2nd_4|
-    ldrd    r8, [r0, #-4]                   ; load the data
+    ldrd    r8, r9, [r0, #-4]               ; load the data
     orr     r7, r7, r3, lsr #1              ; loop counter
 
 |width_loop_2nd_4|
diff --git a/libvpx/vp8/common/arm/armv6/idct_blk_v6.c b/libvpx/vp8/common/arm/armv6/idct_blk_v6.c
index 6002c0f..c94f84a 100644
--- a/libvpx/vp8/common/arm/armv6/idct_blk_v6.c
+++ b/libvpx/vp8/common/arm/armv6/idct_blk_v6.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 
 void vp8_dequant_idct_add_y_block_v6(short *q, short *dq,
diff --git a/libvpx/vp8/common/arm/bilinearfilter_arm.c b/libvpx/vp8/common/arm/bilinearfilter_arm.c
index c63073c..799c8bd 100644
--- a/libvpx/vp8/common/arm/bilinearfilter_arm.c
+++ b/libvpx/vp8/common/arm/bilinearfilter_arm.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include <math.h>
 #include "vp8/common/filter.h"
 #include "bilinearfilter_arm.h"
diff --git a/libvpx/vp8/common/arm/filter_arm.c b/libvpx/vp8/common/arm/filter_arm.c
index 148951a..7fe3967 100644
--- a/libvpx/vp8/common/arm/filter_arm.c
+++ b/libvpx/vp8/common/arm/filter_arm.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include <math.h>
 #include "vp8/common/filter.h"
 #include "vpx_ports/mem.h"
diff --git a/libvpx/vp8/common/arm/loopfilter_arm.c b/libvpx/vp8/common/arm/loopfilter_arm.c
index b8f9bd9..3bdc967 100644
--- a/libvpx/vp8/common/arm/loopfilter_arm.c
+++ b/libvpx/vp8/common/arm/loopfilter_arm.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/loopfilter.h"
 #include "vp8/common/onyxc_int.h"
 
diff --git a/libvpx/vp8/common/arm/neon/idct_blk_neon.c b/libvpx/vp8/common/arm/neon/idct_blk_neon.c
index ee7f223..fb327a7 100644
--- a/libvpx/vp8/common/arm/neon/idct_blk_neon.c
+++ b/libvpx/vp8/common/arm/neon/idct_blk_neon.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 /* place these declarations here because we don't want to maintain them
  * outside of this scope
diff --git a/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm b/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm
index e7a3ed1..9d22c52 100644
--- a/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm
+++ b/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm
@@ -9,9 +9,6 @@
 ;
 
 
-bilinear_taps_coeff
-    DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
 ;-----------------
 
     EXPORT  |vp8_sub_pixel_variance16x16_neon_func|
@@ -29,6 +26,9 @@
 ; stack(r6) unsigned int *sse
 ;note: most of the code is copied from bilinear_predict16x16_neon and vp8_variance16x16_neon.
 
+bilinear_taps_coeff
+    DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
+
 |vp8_sub_pixel_variance16x16_neon_func| PROC
     push            {r4-r6, lr}
 
diff --git a/libvpx/vp8/common/arm/reconintra_arm.c b/libvpx/vp8/common/arm/reconintra_arm.c
index 121e090..2874896 100644
--- a/libvpx/vp8/common/arm/reconintra_arm.c
+++ b/libvpx/vp8/common/arm/reconintra_arm.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/blockd.h"
 #include "vpx_mem/vpx_mem.h"
 
diff --git a/libvpx/vp8/common/arm/variance_arm.c b/libvpx/vp8/common/arm/variance_arm.c
index 891d767..467a509 100644
--- a/libvpx/vp8/common/arm/variance_arm.c
+++ b/libvpx/vp8/common/arm/variance_arm.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/variance.h"
 #include "vp8/common/filter.h"
 
diff --git a/libvpx/vp8/common/dequantize.c b/libvpx/vp8/common/dequantize.c
index 8eda486..6e2f69a 100644
--- a/libvpx/vp8/common/dequantize.c
+++ b/libvpx/vp8/common/dequantize.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/blockd.h"
 #include "vpx_mem/vpx_mem.h"
 
diff --git a/libvpx/vp8/common/generic/systemdependent.c b/libvpx/vp8/common/generic/systemdependent.c
index 5a6ac7b..d84df33 100644
--- a/libvpx/vp8/common/generic/systemdependent.c
+++ b/libvpx/vp8/common/generic/systemdependent.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #if ARCH_ARM
 #include "vpx_ports/arm.h"
 #elif ARCH_X86 || ARCH_X86_64
@@ -82,6 +82,7 @@
 }
 #endif
 
+void vp8_clear_system_state_c() {};
 
 void vp8_machine_specific_config(VP8_COMMON *ctx)
 {
diff --git a/libvpx/vp8/common/idct_blk.c b/libvpx/vp8/common/idct_blk.c
index 0b058c7..8edfffb 100644
--- a/libvpx/vp8/common/idct_blk.c
+++ b/libvpx/vp8/common/idct_blk.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 void vp8_dequant_idct_add_c(short *input, short *dq,
                             unsigned char *dest, int stride);
diff --git a/libvpx/vp8/common/invtrans.h b/libvpx/vp8/common/invtrans.h
index d048665..9262640 100644
--- a/libvpx/vp8/common/invtrans.h
+++ b/libvpx/vp8/common/invtrans.h
@@ -13,7 +13,7 @@
 #define __INC_INVTRANS_H
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "blockd.h"
 #include "onyxc_int.h"
 
diff --git a/libvpx/vp8/common/loopfilter.c b/libvpx/vp8/common/loopfilter.c
index 41b4f12..19857a7 100644
--- a/libvpx/vp8/common/loopfilter.c
+++ b/libvpx/vp8/common/loopfilter.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "loopfilter.h"
 #include "onyxc_int.h"
 #include "vpx_mem/vpx_mem.h"
@@ -156,39 +156,38 @@
             continue;
         }
 
-        lvl_ref = lvl_seg;
-
         /* INTRA_FRAME */
         ref = INTRA_FRAME;
 
         /* Apply delta for reference frame */
-        lvl_ref += mbd->ref_lf_deltas[ref];
+        lvl_ref = lvl_seg + mbd->ref_lf_deltas[ref];
 
         /* Apply delta for Intra modes */
         mode = 0; /* B_PRED */
         /* Only the split mode BPRED has a further special case */
-        lvl_mode = lvl_ref +  mbd->mode_lf_deltas[mode];
-        lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; /* clamp */
+        lvl_mode = lvl_ref + mbd->mode_lf_deltas[mode];
+        /* clamp */
+        lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0;
 
         lfi->lvl[seg][ref][mode] = lvl_mode;
 
         mode = 1; /* all the rest of Intra modes */
-        lvl_mode = (lvl_ref > 0) ? (lvl_ref > 63 ? 63 : lvl_ref)  : 0; /* clamp */
+        /* clamp */
+        lvl_mode = (lvl_ref > 0) ? (lvl_ref > 63 ? 63 : lvl_ref) : 0;
         lfi->lvl[seg][ref][mode] = lvl_mode;
 
         /* LAST, GOLDEN, ALT */
         for(ref = 1; ref < MAX_REF_FRAMES; ref++)
         {
-            int lvl_ref = lvl_seg;
-
             /* Apply delta for reference frame */
-            lvl_ref += mbd->ref_lf_deltas[ref];
+            lvl_ref = lvl_seg + mbd->ref_lf_deltas[ref];
 
             /* Apply delta for Inter modes */
             for (mode = 1; mode < 4; mode++)
             {
                 lvl_mode = lvl_ref + mbd->mode_lf_deltas[mode];
-                lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; /* clamp */
+                /* clamp */
+                lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0;
 
                 lfi->lvl[seg][ref][mode] = lvl_mode;
             }
@@ -567,46 +566,28 @@
     int mb_cols = post->y_width >> 4;
     int mb_rows = post->y_height >> 4;
 
-    int linestocopy, i;
+    int linestocopy;
 
     loop_filter_info_n *lfi_n = &cm->lf_info;
     loop_filter_info lfi;
 
     int filter_level;
-    int alt_flt_enabled = mbd->segmentation_enabled;
     FRAME_TYPE frame_type = cm->frame_type;
 
     const MODE_INFO *mode_info_context;
 
-    int lvl_seg[MAX_MB_SEGMENTS];
+#if 0
+    if(default_filt_lvl == 0) /* no filter applied */
+        return;
+#endif
+
+    /* Initialize the loop filter for this frame. */
+    vp8_loop_filter_frame_init( cm, mbd, default_filt_lvl);
 
     /* number of MB rows to use in partial filtering */
     linestocopy = mb_rows / PARTIAL_FRAME_FRACTION;
     linestocopy = linestocopy ? linestocopy << 4 : 16;     /* 16 lines per MB */
 
-    /* Note the baseline filter values for each segment */
-    /* See vp8_loop_filter_frame_init. Rather than call that for each change
-     * to default_filt_lvl, copy the relevant calculation here.
-     */
-    if (alt_flt_enabled)
-    {
-        for (i = 0; i < MAX_MB_SEGMENTS; i++)
-        {    /* Abs value */
-            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
-            {
-                lvl_seg[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
-            }
-            /* Delta Value */
-            else
-            {
-                lvl_seg[i] = default_filt_lvl
-                        + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
-                lvl_seg[i] = (lvl_seg[i] > 0) ?
-                        ((lvl_seg[i] > 63) ? 63: lvl_seg[i]) : 0;
-            }
-        }
-    }
-
     /* Set up the buffer pointers; partial image starts at ~middle of frame */
     y_ptr = post->y_buffer + ((post->y_height >> 5) * 16) * post->y_stride;
     mode_info_context = cm->mi + (post->y_height >> 5) * (mb_cols + 1);
@@ -620,10 +601,12 @@
                            mode_info_context->mbmi.mode != SPLITMV &&
                            mode_info_context->mbmi.mb_skip_coeff);
 
-            if (alt_flt_enabled)
-                filter_level = lvl_seg[mode_info_context->mbmi.segment_id];
-            else
-                filter_level = default_filt_lvl;
+            const int mode_index =
+                lfi_n->mode_lf_lut[mode_info_context->mbmi.mode];
+            const int seg = mode_info_context->mbmi.segment_id;
+            const int ref_frame = mode_info_context->mbmi.ref_frame;
+
+            filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
 
             if (filter_level)
             {
diff --git a/libvpx/vp8/common/loopfilter.h b/libvpx/vp8/common/loopfilter.h
index b3af2d6..1e47f34 100644
--- a/libvpx/vp8/common/loopfilter.h
+++ b/libvpx/vp8/common/loopfilter.h
@@ -14,7 +14,7 @@
 
 #include "vpx_ports/mem.h"
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 #define MAX_LOOP_FILTER             63
 /* fraction of total macroblock rows to be used in fast filter level picking */
diff --git a/libvpx/vp8/common/loopfilter_filters.c b/libvpx/vp8/common/loopfilter_filters.c
index 8235f6e..1d51696 100644
--- a/libvpx/vp8/common/loopfilter_filters.c
+++ b/libvpx/vp8/common/loopfilter_filters.c
@@ -54,7 +54,7 @@
 {
     signed char ps0, qs0;
     signed char ps1, qs1;
-    signed char vp8_filter, Filter1, Filter2;
+    signed char filter_value, Filter1, Filter2;
     signed char u;
 
     ps1 = (signed char) * op1 ^ 0x80;
@@ -63,35 +63,35 @@
     qs1 = (signed char) * oq1 ^ 0x80;
 
     /* add outer taps if we have high edge variance */
-    vp8_filter = vp8_signed_char_clamp(ps1 - qs1);
-    vp8_filter &= hev;
+    filter_value = vp8_signed_char_clamp(ps1 - qs1);
+    filter_value &= hev;
 
     /* inner taps */
-    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (qs0 - ps0));
-    vp8_filter &= mask;
+    filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0));
+    filter_value &= mask;
 
     /* save bottom 3 bits so that we round one side +4 and the other +3
      * if it equals 4 we'll set to adjust by -1 to account for the fact
      * we'd round 3 the other way
      */
-    Filter1 = vp8_signed_char_clamp(vp8_filter + 4);
-    Filter2 = vp8_signed_char_clamp(vp8_filter + 3);
+    Filter1 = vp8_signed_char_clamp(filter_value + 4);
+    Filter2 = vp8_signed_char_clamp(filter_value + 3);
     Filter1 >>= 3;
     Filter2 >>= 3;
     u = vp8_signed_char_clamp(qs0 - Filter1);
     *oq0 = u ^ 0x80;
     u = vp8_signed_char_clamp(ps0 + Filter2);
     *op0 = u ^ 0x80;
-    vp8_filter = Filter1;
+    filter_value = Filter1;
 
     /* outer tap adjustments */
-    vp8_filter += 1;
-    vp8_filter >>= 1;
-    vp8_filter &= ~hev;
+    filter_value += 1;
+    filter_value >>= 1;
+    filter_value &= ~hev;
 
-    u = vp8_signed_char_clamp(qs1 - vp8_filter);
+    u = vp8_signed_char_clamp(qs1 - filter_value);
     *oq1 = u ^ 0x80;
-    u = vp8_signed_char_clamp(ps1 + vp8_filter);
+    u = vp8_signed_char_clamp(ps1 + filter_value);
     *op1 = u ^ 0x80;
 
 }
@@ -162,7 +162,7 @@
                            uc *op2, uc *op1, uc *op0, uc *oq0, uc *oq1, uc *oq2)
 {
     signed char s, u;
-    signed char vp8_filter, Filter1, Filter2;
+    signed char filter_value, Filter1, Filter2;
     signed char ps2 = (signed char) * op2 ^ 0x80;
     signed char ps1 = (signed char) * op1 ^ 0x80;
     signed char ps0 = (signed char) * op0 ^ 0x80;
@@ -171,11 +171,11 @@
     signed char qs2 = (signed char) * oq2 ^ 0x80;
 
     /* add outer taps if we have high edge variance */
-    vp8_filter = vp8_signed_char_clamp(ps1 - qs1);
-    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (qs0 - ps0));
-    vp8_filter &= mask;
+    filter_value = vp8_signed_char_clamp(ps1 - qs1);
+    filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0));
+    filter_value &= mask;
 
-    Filter2 = vp8_filter;
+    Filter2 = filter_value;
     Filter2 &= hev;
 
     /* save bottom 3 bits so that we round one side +4 and the other +3 */
@@ -188,8 +188,8 @@
 
 
     /* only apply wider filter if not high edge variance */
-    vp8_filter &= ~hev;
-    Filter2 = vp8_filter;
+    filter_value &= ~hev;
+    Filter2 = filter_value;
 
     /* roughly 3/7th difference across boundary */
     u = vp8_signed_char_clamp((63 + Filter2 * 27) >> 7);
@@ -291,24 +291,24 @@
 
 static void vp8_simple_filter(signed char mask, uc *op1, uc *op0, uc *oq0, uc *oq1)
 {
-    signed char vp8_filter, Filter1, Filter2;
+    signed char filter_value, Filter1, Filter2;
     signed char p1 = (signed char) * op1 ^ 0x80;
     signed char p0 = (signed char) * op0 ^ 0x80;
     signed char q0 = (signed char) * oq0 ^ 0x80;
     signed char q1 = (signed char) * oq1 ^ 0x80;
     signed char u;
 
-    vp8_filter = vp8_signed_char_clamp(p1 - q1);
-    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (q0 - p0));
-    vp8_filter &= mask;
+    filter_value = vp8_signed_char_clamp(p1 - q1);
+    filter_value = vp8_signed_char_clamp(filter_value + 3 * (q0 - p0));
+    filter_value &= mask;
 
     /* save bottom 3 bits so that we round one side +4 and the other +3 */
-    Filter1 = vp8_signed_char_clamp(vp8_filter + 4);
+    Filter1 = vp8_signed_char_clamp(filter_value + 4);
     Filter1 >>= 3;
     u = vp8_signed_char_clamp(q0 - Filter1);
     *oq0  = u ^ 0x80;
 
-    Filter2 = vp8_signed_char_clamp(vp8_filter + 3);
+    Filter2 = vp8_signed_char_clamp(filter_value + 3);
     Filter2 >>= 3;
     u = vp8_signed_char_clamp(p0 + Filter2);
     *op0 = u ^ 0x80;
diff --git a/libvpx/vp8/common/mfqe.c b/libvpx/vp8/common/mfqe.c
index 3dff150..0693326 100644
--- a/libvpx/vp8/common/mfqe.c
+++ b/libvpx/vp8/common/mfqe.c
@@ -20,7 +20,7 @@
 #include "postproc.h"
 #include "variance.h"
 #include "vpx_mem/vpx_mem.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_scale/yv12config.h"
 
 #include <limits.h>
@@ -280,7 +280,7 @@
 
     FRAME_TYPE frame_type = cm->frame_type;
     /* Point at base of Mb MODE_INFO list has motion vectors etc */
-    const MODE_INFO *mode_info_context = cm->mi;
+    const MODE_INFO *mode_info_context = cm->show_frame_mi;
     int mb_row;
     int mb_col;
     int totmap, map[4];
diff --git a/libvpx/vp8/common/mips/dspr2/dequantize_dspr2.c b/libvpx/vp8/common/mips/dspr2/dequantize_dspr2.c
index 6823325..619ee80 100644
--- a/libvpx/vp8/common/mips/dspr2/dequantize_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/dequantize_dspr2.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_mem/vpx_mem.h"
 
 #if HAVE_DSPR2
diff --git a/libvpx/vp8/common/mips/dspr2/filter_dspr2.c b/libvpx/vp8/common/mips/dspr2/filter_dspr2.c
index 71fdcd7..ace5d40 100644
--- a/libvpx/vp8/common/mips/dspr2/filter_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/filter_dspr2.c
@@ -10,7 +10,7 @@
 
 
 #include <stdlib.h>
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_ports/mem.h"
 
 #if HAVE_DSPR2
diff --git a/libvpx/vp8/common/mips/dspr2/idct_blk_dspr2.c b/libvpx/vp8/common/mips/dspr2/idct_blk_dspr2.c
index 1e0ebd1..ab938cd 100644
--- a/libvpx/vp8/common/mips/dspr2/idct_blk_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/idct_blk_dspr2.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 #if HAVE_DSPR2
 
diff --git a/libvpx/vp8/common/mips/dspr2/idctllm_dspr2.c b/libvpx/vp8/common/mips/dspr2/idctllm_dspr2.c
index 25b7936..2eff710 100644
--- a/libvpx/vp8/common/mips/dspr2/idctllm_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/idctllm_dspr2.c
@@ -8,7 +8,7 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 #if HAVE_DSPR2
 #define CROP_WIDTH 256
diff --git a/libvpx/vp8/common/mips/dspr2/loopfilter_filters_dspr2.c b/libvpx/vp8/common/mips/dspr2/loopfilter_filters_dspr2.c
index b8e5e4d..9ae6bc8 100644
--- a/libvpx/vp8/common/mips/dspr2/loopfilter_filters_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/loopfilter_filters_dspr2.c
@@ -10,7 +10,7 @@
 
 
 #include <stdlib.h>
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/onyxc_int.h"
 
 #if HAVE_DSPR2
diff --git a/libvpx/vp8/common/mips/dspr2/reconinter_dspr2.c b/libvpx/vp8/common/mips/dspr2/reconinter_dspr2.c
index a5239a3..a14b397 100644
--- a/libvpx/vp8/common/mips/dspr2/reconinter_dspr2.c
+++ b/libvpx/vp8/common/mips/dspr2/reconinter_dspr2.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/vpx_integer.h"
 
 #if HAVE_DSPR2
diff --git a/libvpx/vp8/common/onyxc_int.h b/libvpx/vp8/common/onyxc_int.h
index 5325bac..276dd72 100644
--- a/libvpx/vp8/common/onyxc_int.h
+++ b/libvpx/vp8/common/onyxc_int.h
@@ -13,7 +13,7 @@
 #define __INC_VP8C_INT_H
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/internal/vpx_codec_internal.h"
 #include "loopfilter.h"
 #include "entropymv.h"
@@ -127,7 +127,8 @@
     MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
     MODE_INFO *prev_mi;  /* 'mi' from last frame (points into prev_mip) */
 #endif
-
+    MODE_INFO *show_frame_mi;  /* MODE_INFO for the last decoded frame
+                                  to show */
     LOOPFILTERTYPE filter_type;
 
     loop_filter_info_n lf_info;
diff --git a/libvpx/vp8/common/onyxd.h b/libvpx/vp8/common/onyxd.h
index fd7e051..97c81c1 100644
--- a/libvpx/vp8/common/onyxd.h
+++ b/libvpx/vp8/common/onyxd.h
@@ -34,7 +34,6 @@
         int     postprocess;
         int     max_threads;
         int     error_concealment;
-        int     input_fragments;
     } VP8D_CONFIG;
 
     typedef enum
@@ -56,10 +55,6 @@
     vpx_codec_err_t vp8dx_get_reference(struct VP8D_COMP* comp, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd);
     vpx_codec_err_t vp8dx_set_reference(struct VP8D_COMP* comp, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd);
 
-    struct VP8D_COMP* vp8dx_create_decompressor(VP8D_CONFIG *oxcf);
-
-    void vp8dx_remove_decompressor(struct VP8D_COMP* comp);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/libvpx/vp8/common/postproc.c b/libvpx/vp8/common/postproc.c
index 80fa530..0266f4c 100644
--- a/libvpx/vp8/common/postproc.c
+++ b/libvpx/vp8/common/postproc.c
@@ -10,11 +10,12 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
+#include "vpx_scale_rtcd.h"
 #include "vpx_scale/yv12config.h"
 #include "postproc.h"
 #include "common.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "systemdependent.h"
 
 #include <limits.h>
@@ -333,7 +334,7 @@
     double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
     int ppl = (int)(level + .5);
 
-    const MODE_INFO *mode_info_context = cm->mi;
+    const MODE_INFO *mode_info_context = cm->show_frame_mi;
     int mbr, mbc;
 
     /* The pixel thresholds are adjusted according to if or not the macroblock
@@ -438,29 +439,28 @@
     char char_dist[300];
 
     double sigma;
-    int ai = a, qi = q, i;
+    int i;
 
     vp8_clear_system_state();
 
 
-    sigma = ai + .5 + .6 * (63 - qi) / 63.0;
+    sigma = a + .5 + .6 * (63 - q) / 63.0;
 
     /* set up a lookup table of 256 entries that matches
      * a gaussian distribution with sigma determined by q.
      */
     {
-        double i;
         int next, j;
 
         next = 0;
 
         for (i = -32; i < 32; i++)
         {
-            int a = (int)(.5 + 256 * vp8_gaussian(sigma, 0, i));
+            const int v = (int)(.5 + 256 * vp8_gaussian(sigma, 0, i));
 
-            if (a)
+            if (v)
             {
-                for (j = 0; j < a; j++)
+                for (j = 0; j < v; j++)
                 {
                     char_dist[next+j] = (char) i;
                 }
@@ -543,12 +543,12 @@
  * filled with the same color block.
  */
 void vp8_blend_mb_inner_c (unsigned char *y, unsigned char *u, unsigned char *v,
-                        int y1, int u1, int v1, int alpha, int stride)
+                        int y_1, int u_1, int v_1, int alpha, int stride)
 {
     int i, j;
-    int y1_const = y1*((1<<16)-alpha);
-    int u1_const = u1*((1<<16)-alpha);
-    int v1_const = v1*((1<<16)-alpha);
+    int y1_const = y_1*((1<<16)-alpha);
+    int u1_const = u_1*((1<<16)-alpha);
+    int v1_const = v_1*((1<<16)-alpha);
 
     y += 2*stride + 2;
     for (i = 0; i < 12; i++)
@@ -581,12 +581,12 @@
  * unblended to allow for other visualizations to be layered.
  */
 void vp8_blend_mb_outer_c (unsigned char *y, unsigned char *u, unsigned char *v,
-                        int y1, int u1, int v1, int alpha, int stride)
+                        int y_1, int u_1, int v_1, int alpha, int stride)
 {
     int i, j;
-    int y1_const = y1*((1<<16)-alpha);
-    int u1_const = u1*((1<<16)-alpha);
-    int v1_const = v1*((1<<16)-alpha);
+    int y1_const = y_1*((1<<16)-alpha);
+    int u1_const = u_1*((1<<16)-alpha);
+    int v1_const = v_1*((1<<16)-alpha);
 
     for (i = 0; i < 2; i++)
     {
@@ -645,12 +645,12 @@
 }
 
 void vp8_blend_b_c (unsigned char *y, unsigned char *u, unsigned char *v,
-                        int y1, int u1, int v1, int alpha, int stride)
+                        int y_1, int u_1, int v_1, int alpha, int stride)
 {
     int i, j;
-    int y1_const = y1*((1<<16)-alpha);
-    int u1_const = u1*((1<<16)-alpha);
-    int v1_const = v1*((1<<16)-alpha);
+    int y1_const = y_1*((1<<16)-alpha);
+    int u1_const = u_1*((1<<16)-alpha);
+    int v1_const = v_1*((1<<16)-alpha);
 
     for (i = 0; i < 4; i++)
     {
@@ -675,46 +675,46 @@
     }
 }
 
-static void constrain_line (int x0, int *x1, int y0, int *y1, int width, int height)
+static void constrain_line (int x_0, int *x_1, int y_0, int *y_1, int width, int height)
 {
     int dx;
     int dy;
 
-    if (*x1 > width)
+    if (*x_1 > width)
     {
-        dx = *x1 - x0;
-        dy = *y1 - y0;
+        dx = *x_1 - x_0;
+        dy = *y_1 - y_0;
 
-        *x1 = width;
+        *x_1 = width;
         if (dx)
-            *y1 = ((width-x0)*dy)/dx + y0;
+            *y_1 = ((width-x_0)*dy)/dx + y_0;
     }
-    if (*x1 < 0)
+    if (*x_1 < 0)
     {
-        dx = *x1 - x0;
-        dy = *y1 - y0;
+        dx = *x_1 - x_0;
+        dy = *y_1 - y_0;
 
-        *x1 = 0;
+        *x_1 = 0;
         if (dx)
-            *y1 = ((0-x0)*dy)/dx + y0;
+            *y_1 = ((0-x_0)*dy)/dx + y_0;
     }
-    if (*y1 > height)
+    if (*y_1 > height)
     {
-        dx = *x1 - x0;
-        dy = *y1 - y0;
+        dx = *x_1 - x_0;
+        dy = *y_1 - y_0;
 
-        *y1 = height;
+        *y_1 = height;
         if (dy)
-            *x1 = ((height-y0)*dx)/dy + x0;
+            *x_1 = ((height-y_0)*dx)/dy + x_0;
     }
-    if (*y1 < 0)
+    if (*y_1 < 0)
     {
-        dx = *x1 - x0;
-        dy = *y1 - y0;
+        dx = *x_1 - x_0;
+        dy = *y_1 - y_0;
 
-        *y1 = 0;
+        *y_1 = 0;
         if (dy)
-            *x1 = ((0-y0)*dx)/dy + x0;
+            *x_1 = ((0-y_0)*dx)/dy + x_0;
     }
 }
 
diff --git a/libvpx/vp8/common/ppc/systemdependent.c b/libvpx/vp8/common/ppc/systemdependent.c
index 87f4cac..6899c0e 100644
--- a/libvpx/vp8/common/ppc/systemdependent.c
+++ b/libvpx/vp8/common/ppc/systemdependent.c
@@ -12,13 +12,8 @@
 #include "subpixel.h"
 #include "loopfilter.h"
 #include "recon.h"
-#include "idct.h"
 #include "onyxc_int.h"
 
-void (*vp8_short_idct4x4)(short *input, short *output, int pitch);
-void (*vp8_short_idct4x4_1)(short *input, short *output, int pitch);
-void (*vp8_dc_only_idct)(short input_dc, short *output, int pitch);
-
 extern void (*vp8_post_proc_down_and_across_mb_row)(
     unsigned char *src_ptr,
     unsigned char *dst_ptr,
diff --git a/libvpx/vp8/common/reconinter.c b/libvpx/vp8/common/reconinter.c
index 3da3bc7..43f84d0 100644
--- a/libvpx/vp8/common/reconinter.c
+++ b/libvpx/vp8/common/reconinter.c
@@ -11,7 +11,7 @@
 
 #include <limits.h>
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/vpx_integer.h"
 #include "blockd.h"
 #include "reconinter.h"
diff --git a/libvpx/vp8/common/reconintra.c b/libvpx/vp8/common/reconintra.c
index 4067a68..ec51ffe 100644
--- a/libvpx/vp8/common/reconintra.c
+++ b/libvpx/vp8/common/reconintra.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_mem/vpx_mem.h"
 #include "blockd.h"
 
@@ -36,7 +36,6 @@
     case DC_PRED:
     {
         int expected_dc;
-        int i;
         int shift;
         int average = 0;
 
@@ -168,7 +167,6 @@
     {
         int expected_udc;
         int expected_vdc;
-        int i;
         int shift;
         int Uaverage = 0;
         int Vaverage = 0;
@@ -217,8 +215,6 @@
     break;
     case V_PRED:
     {
-        int i;
-
         for (i = 0; i < 8; i++)
         {
             vpx_memcpy(upred_ptr, uabove_row, 8);
@@ -231,8 +227,6 @@
     break;
     case H_PRED:
     {
-        int i;
-
         for (i = 0; i < 8; i++)
         {
             vpx_memset(upred_ptr, uleft_col[i], 8);
@@ -245,8 +239,6 @@
     break;
     case TM_PRED:
     {
-        int i;
-
         for (i = 0; i < 8; i++)
         {
             for (j = 0; j < 8; j++)
diff --git a/libvpx/vp8/common/reconintra4x4.c b/libvpx/vp8/common/reconintra4x4.c
index 7bb8d0a..3d4f2c4 100644
--- a/libvpx/vp8/common/reconintra4x4.c
+++ b/libvpx/vp8/common/reconintra4x4.c
@@ -10,17 +10,17 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "blockd.h"
 
 void vp8_intra4x4_predict_c(unsigned char *Above,
                             unsigned char *yleft, int left_stride,
-                            B_PREDICTION_MODE b_mode,
+                            int           _b_mode,
                             unsigned char *dst, int dst_stride,
                             unsigned char top_left)
 {
     int i, r, c;
-
+    B_PREDICTION_MODE b_mode = (B_PREDICTION_MODE)_b_mode;
     unsigned char Left[4];
     Left[0] = yleft[0];
     Left[1] = yleft[left_stride];
diff --git a/libvpx/vp8/common/rtcd.c b/libvpx/vp8/common/rtcd.c
index 01dad46..0b371b0 100644
--- a/libvpx/vp8/common/rtcd.c
+++ b/libvpx/vp8/common/rtcd.c
@@ -9,97 +9,13 @@
  */
 #include "vpx_config.h"
 #define RTCD_C
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
+#include "vpx_ports/vpx_once.h"
 
-#if CONFIG_MULTITHREAD && defined(_WIN32)
-#include <windows.h>
-#include <stdlib.h>
-static void once(void (*func)(void))
+extern void vpx_scale_rtcd(void);
+
+void vp8_rtcd()
 {
-    static CRITICAL_SECTION *lock;
-    static LONG waiters;
-    static int done;
-    void *lock_ptr = &lock;
-
-    /* If the initialization is complete, return early. This isn't just an
-     * optimization, it prevents races on the destruction of the global
-     * lock.
-     */
-    if(done)
-        return;
-
-    InterlockedIncrement(&waiters);
-
-    /* Get a lock. We create one and try to make it the one-true-lock,
-     * throwing it away if we lost the race.
-     */
-
-    {
-        /* Scope to protect access to new_lock */
-        CRITICAL_SECTION *new_lock = malloc(sizeof(CRITICAL_SECTION));
-        InitializeCriticalSection(new_lock);
-        if (InterlockedCompareExchangePointer(lock_ptr, new_lock, NULL) != NULL)
-        {
-            DeleteCriticalSection(new_lock);
-            free(new_lock);
-        }
-    }
-
-    /* At this point, we have a lock that can be synchronized on. We don't
-     * care which thread actually performed the allocation.
-     */
-
-    EnterCriticalSection(lock);
-
-    if (!done)
-    {
-        func();
-        done = 1;
-    }
-
-    LeaveCriticalSection(lock);
-
-    /* Last one out should free resources. The destructed objects are
-     * protected by checking if(done) above.
-     */
-    if(!InterlockedDecrement(&waiters))
-    {
-        DeleteCriticalSection(lock);
-        free(lock);
-        lock = NULL;
-    }
-}
-
-
-#elif CONFIG_MULTITHREAD && HAVE_PTHREAD_H
-#include <pthread.h>
-static void once(void (*func)(void))
-{
-    static pthread_once_t lock = PTHREAD_ONCE_INIT;
-    pthread_once(&lock, func);
-}
-
-
-#else
-/* No-op version that performs no synchronization. vpx_rtcd() is idempotent,
- * so as long as your platform provides atomic loads/stores of pointers
- * no synchronization is strictly necessary.
- */
-
-static void once(void (*func)(void))
-{
-    static int done;
-
-    if(!done)
-    {
-        func();
-        done = 1;
-    }
-}
-#endif
-
-
-void vpx_rtcd()
-{
+    vpx_scale_rtcd();
     once(setup_rtcd_internal);
 }
diff --git a/libvpx/vp8/common/rtcd_defs.sh b/libvpx/vp8/common/rtcd_defs.sh
index 0f950f8..9ebf389 100644
--- a/libvpx/vp8/common/rtcd_defs.sh
+++ b/libvpx/vp8/common/rtcd_defs.sh
@@ -1,6 +1,8 @@
-common_forward_decls() {
+vp8_common_forward_decls() {
 cat <<EOF
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -14,7 +16,14 @@
 struct yv12_buffer_config;
 EOF
 }
-forward_decls common_forward_decls
+forward_decls vp8_common_forward_decls
+
+#
+# system state
+#
+prototype void vp8_clear_system_state ""
+specialize vp8_clear_system_state mmx
+vp8_clear_system_state_mmx=vpx_reset_mmx_state
 
 #
 # Dequant
@@ -146,7 +155,7 @@
 prototype void vp8_build_intra_predictors_mbuv_s "struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride"
 specialize vp8_build_intra_predictors_mbuv_s sse2 ssse3
 
-prototype void vp8_intra4x4_predict "unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left"
+prototype void vp8_intra4x4_predict "unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left"
 specialize vp8_intra4x4_predict media
 vp8_intra4x4_predict_media=vp8_intra4x4_predict_armv6
 
@@ -442,8 +451,9 @@
 # Quantizer
 #
 prototype void vp8_regular_quantize_b "struct block *, struct blockd *"
-specialize vp8_regular_quantize_b sse2 sse4_1
-vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4
+specialize vp8_regular_quantize_b sse2 #sse4_1
+# TODO(johann) Update sse4 implementation and re-enable
+#vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4
 
 prototype void vp8_fast_quantize_b "struct block *, struct blockd *"
 specialize vp8_fast_quantize_b sse2 ssse3 media neon
@@ -530,39 +540,3 @@
 
 # End of encoder only functions
 fi
-
-# Scaler functions
-if [ "CONFIG_SPATIAL_RESAMPLING" != "yes" ]; then
-    prototype void vp8_horizontal_line_4_5_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_4_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_last_vertical_band_4_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_2_3_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_2_3_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_last_vertical_band_2_3_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_3_5_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_3_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_last_vertical_band_3_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_3_4_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_3_4_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_last_vertical_band_3_4_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_1_2_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_1_2_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_last_vertical_band_1_2_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_5_4_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_5_4_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_5_3_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_5_3_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_horizontal_line_2_1_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
-    prototype void vp8_vertical_band_2_1_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-    prototype void vp8_vertical_band_2_1_scale_i "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
-fi
-
-prototype void vp8_yv12_extend_frame_borders "struct yv12_buffer_config *ybf"
-specialize vp8_yv12_extend_frame_borders neon
-
-prototype void vp8_yv12_copy_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
-specialize vp8_yv12_copy_frame neon
-
-prototype void vp8_yv12_copy_y "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
-specialize vp8_yv12_copy_y neon
-
diff --git a/libvpx/vp8/common/systemdependent.h b/libvpx/vp8/common/systemdependent.h
index f99c4bb..e6b0456 100644
--- a/libvpx/vp8/common/systemdependent.h
+++ b/libvpx/vp8/common/systemdependent.h
@@ -10,12 +10,6 @@
 
 
 #include "vpx_config.h"
-#if ARCH_X86 || ARCH_X86_64
-void vpx_reset_mmx_state(void);
-#define vp8_clear_system_state() vpx_reset_mmx_state()
-#else
-#define vp8_clear_system_state()
-#endif
 
 struct VP8Common;
 void vp8_machine_specific_config(struct VP8Common *);
diff --git a/libvpx/vp8/common/variance_c.c b/libvpx/vp8/common/variance_c.c
index da08aff..773b655 100644
--- a/libvpx/vp8/common/variance_c.c
+++ b/libvpx/vp8/common/variance_c.c
@@ -75,7 +75,7 @@
 
     variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 8));
+    return (var - (((unsigned int)avg * avg) >> 8));
 }
 
 unsigned int vp8_variance8x16_c(
@@ -91,7 +91,7 @@
 
     variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 }
 
 unsigned int vp8_variance16x8_c(
@@ -107,7 +107,7 @@
 
     variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 }
 
 
@@ -124,7 +124,7 @@
 
     variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 6));
+    return (var - (((unsigned int)avg * avg) >> 6));
 }
 
 unsigned int vp8_variance4x4_c(
@@ -140,7 +140,7 @@
 
     variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4, &var, &avg);
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 4));
+    return (var - (((unsigned int)avg * avg) >> 4));
 }
 
 
diff --git a/libvpx/vp8/common/vp8_asm_com_offsets.c b/libvpx/vp8/common/vp8_asm_com_offsets.c
new file mode 100644
index 0000000..7bab90f
--- /dev/null
+++ b/libvpx/vp8/common/vp8_asm_com_offsets.c
@@ -0,0 +1,52 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_config.h"
+#include "vpx/vpx_codec.h"
+#include "vpx_ports/asm_offsets.h"
+#include "vp8/common/blockd.h"
+
+#if CONFIG_POSTPROC
+#include "postproc.h"
+#endif /* CONFIG_POSTPROC */
+
+BEGIN
+
+#if CONFIG_POSTPROC
+/* mfqe.c / filter_by_weight */
+DEFINE(MFQE_PRECISION_VAL,                      MFQE_PRECISION);
+#endif /* CONFIG_POSTPROC */
+
+END
+
+/* add asserts for any offset that is not supported by assembly code */
+/* add asserts for any size that is not supported by assembly code */
+
+#if HAVE_MEDIA
+/* switch case in vp8_intra4x4_predict_armv6 is based on these enumerated values */
+ct_assert(B_DC_PRED, B_DC_PRED == 0);
+ct_assert(B_TM_PRED, B_TM_PRED == 1);
+ct_assert(B_VE_PRED, B_VE_PRED == 2);
+ct_assert(B_HE_PRED, B_HE_PRED == 3);
+ct_assert(B_LD_PRED, B_LD_PRED == 4);
+ct_assert(B_RD_PRED, B_RD_PRED == 5);
+ct_assert(B_VR_PRED, B_VR_PRED == 6);
+ct_assert(B_VL_PRED, B_VL_PRED == 7);
+ct_assert(B_HD_PRED, B_HD_PRED == 8);
+ct_assert(B_HU_PRED, B_HU_PRED == 9);
+#endif
+
+#if HAVE_SSE2
+#if CONFIG_POSTPROC
+/* vp8_filter_by_weight16x16 and 8x8 */
+ct_assert(MFQE_PRECISION_VAL, MFQE_PRECISION == 4)
+#endif /* CONFIG_POSTPROC */
+#endif /* HAVE_SSE2 */
diff --git a/libvpx/vp8/common/x86/idct_blk_mmx.c b/libvpx/vp8/common/x86/idct_blk_mmx.c
index 4adf3f5..49b2013 100644
--- a/libvpx/vp8/common/x86/idct_blk_mmx.c
+++ b/libvpx/vp8/common/x86/idct_blk_mmx.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/blockd.h"
 
 extern void vp8_dequantize_b_impl_mmx(short *sq, short *dq, short *q);
diff --git a/libvpx/vp8/common/x86/idct_blk_sse2.c b/libvpx/vp8/common/x86/idct_blk_sse2.c
index 056e052..ae96ec8 100644
--- a/libvpx/vp8/common/x86/idct_blk_sse2.c
+++ b/libvpx/vp8/common/x86/idct_blk_sse2.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 void vp8_idct_dequant_0_2x_sse2
             (short *q, short *dq ,
diff --git a/libvpx/vp8/common/x86/iwalsh_mmx.asm b/libvpx/vp8/common/x86/iwalsh_mmx.asm
index 4aac094..158c3b7 100644
--- a/libvpx/vp8/common/x86/iwalsh_mmx.asm
+++ b/libvpx/vp8/common/x86/iwalsh_mmx.asm
@@ -24,7 +24,7 @@
 
     movq        mm0, [rdx + 0]    ;ip[0]
     movq        mm1, [rdx + 8]    ;ip[4]
-    movd        mm7, rax
+    movq        mm7, rax
 
     movq        mm2, [rdx + 16]   ;ip[8]
     movq        mm3, [rdx + 24]   ;ip[12]
diff --git a/libvpx/vp8/common/x86/loopfilter_block_sse2.asm b/libvpx/vp8/common/x86/loopfilter_block_sse2.asm
index 1c445ef..6d5aaa1 100644
--- a/libvpx/vp8/common/x86/loopfilter_block_sse2.asm
+++ b/libvpx/vp8/common/x86/loopfilter_block_sse2.asm
@@ -136,7 +136,7 @@
 global sym(vp8_loop_filter_bh_y_sse2) PRIVATE
 sym(vp8_loop_filter_bh_y_sse2):
 
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
     %define src      rcx ; src_ptr
     %define stride   rdx ; src_pixel_step
     %define blimit   r8
@@ -150,6 +150,7 @@
 
     push    rbp
     mov     rbp, rsp
+    SAVE_XMM 11
     push    r12
     push    r13
     mov     thresh, arg(4)
@@ -255,9 +256,10 @@
         movdqa       i12, xmm3
         movdqa       i13, xmm8
 
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
     pop    r13
     pop    r12
+    RESTORE_XMM
     pop    rbp
 %endif
 
@@ -276,7 +278,7 @@
 global sym(vp8_loop_filter_bv_y_sse2) PRIVATE
 sym(vp8_loop_filter_bv_y_sse2):
 
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
     %define src      rcx ; src_ptr
     %define stride   rdx ; src_pixel_step
     %define blimit   r8
@@ -777,7 +779,7 @@
     ; un-ALIGN_STACK
     pop          rsp
 
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
     pop    r13
     pop    r12
     RESTORE_XMM
diff --git a/libvpx/vp8/common/x86/mfqe_sse2.asm b/libvpx/vp8/common/x86/mfqe_sse2.asm
index c1d2174..a8a7f56 100644
--- a/libvpx/vp8/common/x86/mfqe_sse2.asm
+++ b/libvpx/vp8/common/x86/mfqe_sse2.asm
@@ -271,7 +271,13 @@
 SECTION_RODATA
 align 16
 t128:
+%ifndef __NASM_VER__
     ddq 128
+%elif CONFIG_BIG_ENDIAN
+    dq  0, 128
+%else
+    dq  128, 0
+%endif
 align 16
 tMFQE: ; 1 << MFQE_PRECISION
     times 8 dw 0x10
diff --git a/libvpx/vp8/common/x86/postproc_mmx.asm b/libvpx/vp8/common/x86/postproc_mmx.asm
index 966c586..5cf110b 100644
--- a/libvpx/vp8/common/x86/postproc_mmx.asm
+++ b/libvpx/vp8/common/x86/postproc_mmx.asm
@@ -61,7 +61,7 @@
             mov         rcx,        8
 .init_borderd                                                    ; initialize borders
             lea         rdi,        [rdi + rax]
-            movq        [rdi],      xmm1
+            movq        [rdi],      mm1
 
             dec         rcx
             jne         .init_borderd
@@ -193,7 +193,6 @@
             movq        mm4,        [sym(vp8_rv) + rcx*2]
 %endif
             paddw       mm1,        mm4
-            ;paddw     xmm1,       eight8s
             psraw       mm1,        4
 
             packuswb    mm1,        mm0
diff --git a/libvpx/vp8/common/x86/recon_sse2.asm b/libvpx/vp8/common/x86/recon_sse2.asm
index fe77450..1434bcd 100644
--- a/libvpx/vp8/common/x86/recon_sse2.asm
+++ b/libvpx/vp8/common/x86/recon_sse2.asm
@@ -890,6 +890,7 @@
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 5
+    SAVE_XMM 7
     push        rsi
     push        rdi
     GET_GOT     rbx
@@ -957,6 +958,7 @@
     RESTORE_GOT
     pop         rdi
     pop         rsi
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
diff --git a/libvpx/vp8/common/x86/recon_wrapper_sse2.c b/libvpx/vp8/common/x86/recon_wrapper_sse2.c
index b482faa..65f4251 100644
--- a/libvpx/vp8/common/x86/recon_wrapper_sse2.c
+++ b/libvpx/vp8/common/x86/recon_wrapper_sse2.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vp8/common/blockd.h"
 
diff --git a/libvpx/vp8/common/x86/sad_sse3.asm b/libvpx/vp8/common/x86/sad_sse3.asm
index f90a589..69c8d37 100644
--- a/libvpx/vp8/common/x86/sad_sse3.asm
+++ b/libvpx/vp8/common/x86/sad_sse3.asm
@@ -33,7 +33,7 @@
     movsxd      rax,        dword ptr arg(1)    ; src_stride
     movsxd      rdx,        dword ptr arg(3)    ; ref_stride
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     SAVE_XMM 7, u
     %define     src_ptr     rcx
     %define     src_stride  rdx
@@ -76,7 +76,7 @@
     pop         rsi
     pop         rbp
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     RESTORE_XMM
   %endif
 %endif
@@ -111,7 +111,7 @@
 
     xchg        rbx,        rax
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     SAVE_XMM 7, u
     %define     src_ptr     rcx
     %define     src_stride  rdx
@@ -156,7 +156,7 @@
     pop         rsi
     pop         rbp
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     pop         rsi
     RESTORE_XMM
   %endif
diff --git a/libvpx/vp8/common/x86/subpixel_ssse3.asm b/libvpx/vp8/common/x86/subpixel_ssse3.asm
index 13bcaf6..c06f245 100644
--- a/libvpx/vp8/common/x86/subpixel_ssse3.asm
+++ b/libvpx/vp8/common/x86/subpixel_ssse3.asm
@@ -352,6 +352,7 @@
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
diff --git a/libvpx/vp8/common/x86/variance_mmx.c b/libvpx/vp8/common/x86/variance_mmx.c
index 0c4dd4a..36995db 100644
--- a/libvpx/vp8/common/x86/variance_mmx.c
+++ b/libvpx/vp8/common/x86/variance_mmx.c
@@ -91,7 +91,7 @@
 
     vp8_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 4));
+    return (var - (((unsigned int)avg * avg) >> 4));
 
 }
 
@@ -108,7 +108,7 @@
     vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
     *sse = var;
 
-    return (var - ((unsigned int)(avg * avg) >> 6));
+    return (var - (((unsigned int)avg * avg) >> 6));
 
 }
 
@@ -153,7 +153,7 @@
     var = sse0 + sse1 + sse2 + sse3;
     avg = sum0 + sum1 + sum2 + sum3;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 8));
+    return (var - (((unsigned int)avg * avg) >> 8));
 }
 
 unsigned int vp8_variance16x8_mmx(
@@ -172,7 +172,7 @@
     var = sse0 + sse1;
     avg = sum0 + sum1;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 
 }
 
@@ -194,7 +194,7 @@
     avg = sum0 + sum1;
     *sse = var;
 
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 
 }
 
@@ -219,7 +219,7 @@
         &xsum, &xxsum
     );
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 4));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 4));
 }
 
 
@@ -244,7 +244,7 @@
         &xsum, &xxsum
     );
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 6));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 6));
 }
 
 unsigned int vp8_sub_pixel_variance16x16_mmx
@@ -282,7 +282,7 @@
     xxsum0 += xxsum1;
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 
 
 }
@@ -335,7 +335,7 @@
     xxsum0 += xxsum1;
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 7));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
 }
 
 unsigned int vp8_sub_pixel_variance8x16_mmx
@@ -358,7 +358,7 @@
         &xsum, &xxsum
     );
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 7));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 7));
 }
 
 
diff --git a/libvpx/vp8/common/x86/variance_sse2.c b/libvpx/vp8/common/x86/variance_sse2.c
index afd6429..7fa5f53 100644
--- a/libvpx/vp8/common/x86/variance_sse2.c
+++ b/libvpx/vp8/common/x86/variance_sse2.c
@@ -148,7 +148,7 @@
 
     vp8_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 4));
+    return (var - (((unsigned int)avg * avg) >> 4));
 
 }
 
@@ -165,7 +165,7 @@
 
     vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg) ;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 6));
+    return (var - (((unsigned int)avg * avg) >> 6));
 
 }
 
@@ -184,7 +184,7 @@
 
     vp8_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ;
     *sse = sse0;
-    return (sse0 - ((unsigned int)(sum0 * sum0) >> 8));
+    return (sse0 - (((unsigned int)sum0 * sum0) >> 8));
 }
 unsigned int vp8_mse16x16_wmt(
     const unsigned char *src_ptr,
@@ -220,7 +220,7 @@
     var = sse0 + sse1;
     avg = sum0 + sum1;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 
 }
 
@@ -241,7 +241,7 @@
     var = sse0 + sse1;
     avg = sum0 + sum1;
     *sse = var;
-    return (var - ((unsigned int)(avg * avg) >> 7));
+    return (var - (((unsigned int)avg * avg) >> 7));
 
 }
 
@@ -265,7 +265,7 @@
         &xsum, &xxsum
     );
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 4));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 4));
 }
 
 
@@ -314,7 +314,7 @@
     }
 
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 6));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 6));
 }
 
 unsigned int vp8_sub_pixel_variance16x16_wmt
@@ -376,7 +376,7 @@
     }
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 }
 
 unsigned int vp8_sub_pixel_mse16x16_wmt(
@@ -447,7 +447,7 @@
     }
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 7));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
 }
 
 unsigned int vp8_sub_pixel_variance8x16_wmt
@@ -495,7 +495,7 @@
     }
 
     *sse = xxsum;
-    return (xxsum - ((unsigned int)(xsum * xsum) >> 7));
+    return (xxsum - (((unsigned int)xsum * xsum) >> 7));
 }
 
 
@@ -515,7 +515,7 @@
         &xsum0, &xxsum0);
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 }
 
 
@@ -534,7 +534,7 @@
         &xsum0, &xxsum0);
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 }
 
 
@@ -554,5 +554,5 @@
         &xsum0, &xxsum0);
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 }
diff --git a/libvpx/vp8/common/x86/variance_ssse3.c b/libvpx/vp8/common/x86/variance_ssse3.c
index ba2055c..f90f811 100644
--- a/libvpx/vp8/common/x86/variance_ssse3.c
+++ b/libvpx/vp8/common/x86/variance_ssse3.c
@@ -113,7 +113,7 @@
     }
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 8));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
 }
 
 unsigned int vp8_sub_pixel_variance16x8_ssse3
@@ -162,5 +162,5 @@
     }
 
     *sse = xxsum0;
-    return (xxsum0 - ((unsigned int)(xsum0 * xsum0) >> 7));
+    return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
 }
diff --git a/libvpx/vp8/common/x86/vp8_asm_stubs.c b/libvpx/vp8/common/x86/vp8_asm_stubs.c
index 3437a23..c0416b7 100644
--- a/libvpx/vp8/common/x86/vp8_asm_stubs.c
+++ b/libvpx/vp8/common/x86/vp8_asm_stubs.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_ports/mem.h"
 #include "filter_x86.h"
 
diff --git a/libvpx/vp8/decoder/dboolhuff.c b/libvpx/vp8/decoder/dboolhuff.c
index 7e7b05a..546fb2d 100644
--- a/libvpx/vp8/decoder/dboolhuff.c
+++ b/libvpx/vp8/decoder/dboolhuff.c
@@ -10,18 +10,20 @@
 
 
 #include "dboolhuff.h"
-#include "vpx_ports/mem.h"
-#include "vpx_mem/vpx_mem.h"
 
 int vp8dx_start_decode(BOOL_DECODER *br,
                        const unsigned char *source,
-                       unsigned int source_sz)
+                       unsigned int source_sz,
+                       vp8_decrypt_cb *decrypt_cb,
+                       void *decrypt_state)
 {
     br->user_buffer_end = source+source_sz;
     br->user_buffer     = source;
     br->value    = 0;
     br->count    = -8;
     br->range    = 255;
+    br->decrypt_cb = decrypt_cb;
+    br->decrypt_state = decrypt_state;
 
     if (source_sz && !source)
         return 1;
@@ -32,21 +34,42 @@
     return 0;
 }
 
-
 void vp8dx_bool_decoder_fill(BOOL_DECODER *br)
 {
-    const unsigned char *bufptr;
-    const unsigned char *bufend;
-    VP8_BD_VALUE         value;
-    int                  count;
-    bufend = br->user_buffer_end;
-    bufptr = br->user_buffer;
-    value = br->value;
-    count = br->count;
+    const unsigned char *bufptr = br->user_buffer;
+    VP8_BD_VALUE value = br->value;
+    int count = br->count;
+    int shift = VP8_BD_VALUE_SIZE - 8 - (count + 8);
+    size_t bytes_left = br->user_buffer_end - bufptr;
+    size_t bits_left = bytes_left * CHAR_BIT;
+    int x = (int)(shift + CHAR_BIT - bits_left);
+    int loop_end = 0;
+    unsigned char decrypted[sizeof(VP8_BD_VALUE) + 1];
 
-    VP8DX_BOOL_DECODER_FILL(count, value, bufptr, bufend);
+    if (br->decrypt_cb) {
+        int n = bytes_left > sizeof(decrypted) ? sizeof(decrypted) : bytes_left;
+        br->decrypt_cb(br->decrypt_state, bufptr, decrypted, n);
+        bufptr = decrypted;
+    }
 
-    br->user_buffer = bufptr;
+    if(x >= 0)
+    {
+        count += VP8_LOTS_OF_BITS;
+        loop_end = x;
+    }
+
+    if (x < 0 || bits_left)
+    {
+        while(shift >= loop_end)
+        {
+            count += CHAR_BIT;
+            value |= (VP8_BD_VALUE)*bufptr << shift;
+            ++bufptr;
+            ++br->user_buffer;
+            shift -= CHAR_BIT;
+        }
+    }
+
     br->value = value;
     br->count = count;
 }
diff --git a/libvpx/vp8/decoder/dboolhuff.h b/libvpx/vp8/decoder/dboolhuff.h
index 1a08c05..4c0ca1c 100644
--- a/libvpx/vp8/decoder/dboolhuff.h
+++ b/libvpx/vp8/decoder/dboolhuff.h
@@ -9,21 +9,30 @@
  */
 
 
-#ifndef DBOOLHUFF_H
-#define DBOOLHUFF_H
+#ifndef DBOOLHUFF_H_
+#define DBOOLHUFF_H_
+
 #include <stddef.h>
 #include <limits.h>
+
 #include "vpx_config.h"
 #include "vpx_ports/mem.h"
 #include "vpx/vpx_integer.h"
 
 typedef size_t VP8_BD_VALUE;
 
-# define VP8_BD_VALUE_SIZE ((int)sizeof(VP8_BD_VALUE)*CHAR_BIT)
+#define VP8_BD_VALUE_SIZE ((int)sizeof(VP8_BD_VALUE)*CHAR_BIT)
+
 /*This is meant to be a large, positive constant that can still be efficiently
    loaded as an immediate (on platforms like ARM, for example).
   Even relatively modest values like 100 would work fine.*/
-# define VP8_LOTS_OF_BITS (0x40000000)
+#define VP8_LOTS_OF_BITS (0x40000000)
+
+/*Decrypt n bytes of data from input -> output, using the decrypt_state
+   passed in VP8D_SET_DECRYPTOR.
+*/
+typedef void (vp8_decrypt_cb)(void *decrypt_state, const unsigned char *input,
+                              unsigned char *output, int count);
 
 typedef struct
 {
@@ -32,46 +41,20 @@
     VP8_BD_VALUE         value;
     int                  count;
     unsigned int         range;
+    vp8_decrypt_cb      *decrypt_cb;
+    void                *decrypt_state;
 } BOOL_DECODER;
 
 DECLARE_ALIGNED(16, extern const unsigned char, vp8_norm[256]);
 
 int vp8dx_start_decode(BOOL_DECODER *br,
                        const unsigned char *source,
-                       unsigned int source_sz);
+                       unsigned int source_sz,
+                       vp8_decrypt_cb *decrypt_cb,
+                       void *decrypt_state);
 
 void vp8dx_bool_decoder_fill(BOOL_DECODER *br);
 
-/*The refill loop is used in several places, so define it in a macro to make
-   sure they're all consistent.
-  An inline function would be cleaner, but has a significant penalty, because
-   multiple BOOL_DECODER fields must be modified, and the compiler is not smart
-   enough to eliminate the stores to those fields and the subsequent reloads
-   from them when inlining the function.*/
-#define VP8DX_BOOL_DECODER_FILL(_count,_value,_bufptr,_bufend) \
-    do \
-    { \
-        int shift = VP8_BD_VALUE_SIZE - 8 - ((_count) + 8); \
-        int loop_end, x; \
-        size_t bits_left = ((_bufend)-(_bufptr))*CHAR_BIT; \
-        \
-        x = (int)(shift + CHAR_BIT - bits_left); \
-        loop_end = 0; \
-        if(x >= 0) \
-        { \
-            (_count) += VP8_LOTS_OF_BITS; \
-            loop_end = x; \
-            if(!bits_left) break; \
-        } \
-        while(shift >= loop_end) \
-        { \
-            (_count) += CHAR_BIT; \
-            (_value) |= (VP8_BD_VALUE)*(_bufptr)++ << shift; \
-            shift -= CHAR_BIT; \
-        } \
-    } \
-    while(0) \
-
 
 static int vp8dx_decode_bool(BOOL_DECODER *br, int probability) {
     unsigned int bit = 0;
@@ -151,4 +134,5 @@
     /* No error. */
     return 0;
 }
-#endif
+
+#endif  // DBOOLHUFF_H_
diff --git a/libvpx/vp8/decoder/decodemv.h b/libvpx/vp8/decoder/decodemv.h
index 9403424..05a33d2 100644
--- a/libvpx/vp8/decoder/decodemv.h
+++ b/libvpx/vp8/decoder/decodemv.h
@@ -8,7 +8,11 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#ifndef DECODEMV_H_
+#define DECODEMV_H_
 
 #include "onyxd_int.h"
 
 void vp8_decode_mode_mvs(VP8D_COMP *);
+
+#endif  // DECODEMV_H_
diff --git a/libvpx/vp8/decoder/decoderthreading.h b/libvpx/vp8/decoder/decoderthreading.h
index 60c39d1..bc716e4 100644
--- a/libvpx/vp8/decoder/decoderthreading.h
+++ b/libvpx/vp8/decoder/decoderthreading.h
@@ -8,19 +8,15 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
-
-
-
-#ifndef _DECODER_THREADING_H
-#define _DECODER_THREADING_H
+#ifndef DECODERTHREADING_H_
+#define DECODERTHREADING_H_
 
 #if CONFIG_MULTITHREAD
-extern void vp8mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd);
-extern void vp8_decoder_remove_threads(VP8D_COMP *pbi);
-extern void vp8_decoder_create_threads(VP8D_COMP *pbi);
-extern void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows);
-extern void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows);
+void vp8mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd);
+void vp8_decoder_remove_threads(VP8D_COMP *pbi);
+void vp8_decoder_create_threads(VP8D_COMP *pbi);
+void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows);
+void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows);
 #endif
 
-#endif
+#endif  // DECODERTHREADING_H_
diff --git a/libvpx/vp8/decoder/decodframe.c b/libvpx/vp8/decoder/decodframe.c
index a4a00f6..44c35ef 100644
--- a/libvpx/vp8/decoder/decodframe.c
+++ b/libvpx/vp8/decoder/decodframe.c
@@ -10,7 +10,8 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
+#include "./vpx_scale_rtcd.h"
 #include "onyxd_int.h"
 #include "vp8/common/header.h"
 #include "vp8/common/reconintra4x4.h"
@@ -20,7 +21,7 @@
 #include "vp8/common/alloccommon.h"
 #include "vp8/common/entropymode.h"
 #include "vp8/common/quant_common.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vp8/common/setupintrarecon.h"
 
 #include "decodemv.h"
@@ -758,11 +759,16 @@
 
 }
 
-static unsigned int read_partition_size(const unsigned char *cx_size)
+static unsigned int read_partition_size(VP8D_COMP *pbi,
+                                        const unsigned char *cx_size)
 {
-    const unsigned int size =
-        cx_size[0] + (cx_size[1] << 8) + (cx_size[2] << 16);
-    return size;
+    unsigned char temp[3];
+    if (pbi->decrypt_cb)
+    {
+        pbi->decrypt_cb(pbi->decrypt_state, cx_size, temp, 3);
+        cx_size = temp;
+    }
+    return cx_size[0] + (cx_size[1] << 8) + (cx_size[2] << 16);
 }
 
 static int read_is_valid(const unsigned char *start,
@@ -793,7 +799,7 @@
     if (i < num_part - 1)
     {
         if (read_is_valid(partition_size_ptr, 3, first_fragment_end))
-            partition_size = read_partition_size(partition_size_ptr);
+            partition_size = read_partition_size(pbi, partition_size_ptr);
         else if (pbi->ec_active)
             partition_size = (unsigned int)bytes_left;
         else
@@ -827,8 +833,8 @@
     unsigned int partition_idx;
     unsigned int fragment_idx;
     unsigned int num_token_partitions;
-    const unsigned char *first_fragment_end = pbi->fragments[0] +
-                                          pbi->fragment_sizes[0];
+    const unsigned char *first_fragment_end = pbi->fragments.ptrs[0] +
+                                          pbi->fragments.sizes[0];
 
     TOKEN_PARTITION multi_token_partition =
             (TOKEN_PARTITION)vp8_read_literal(&pbi->mbc[8], 2);
@@ -838,10 +844,10 @@
 
     /* Check for partitions within the fragments and unpack the fragments
      * so that each fragment pointer points to its corresponding partition. */
-    for (fragment_idx = 0; fragment_idx < pbi->num_fragments; ++fragment_idx)
+    for (fragment_idx = 0; fragment_idx < pbi->fragments.count; ++fragment_idx)
     {
-        unsigned int fragment_size = pbi->fragment_sizes[fragment_idx];
-        const unsigned char *fragment_end = pbi->fragments[fragment_idx] +
+        unsigned int fragment_size = pbi->fragments.sizes[fragment_idx];
+        const unsigned char *fragment_end = pbi->fragments.ptrs[fragment_idx] +
                                             fragment_size;
         /* Special case for handling the first partition since we have already
          * read its size. */
@@ -849,16 +855,16 @@
         {
             /* Size of first partition + token partition sizes element */
             ptrdiff_t ext_first_part_size = token_part_sizes -
-                pbi->fragments[0] + 3 * (num_token_partitions - 1);
+                pbi->fragments.ptrs[0] + 3 * (num_token_partitions - 1);
             fragment_size -= (unsigned int)ext_first_part_size;
             if (fragment_size > 0)
             {
-                pbi->fragment_sizes[0] = (unsigned int)ext_first_part_size;
+                pbi->fragments.sizes[0] = (unsigned int)ext_first_part_size;
                 /* The fragment contains an additional partition. Move to
                  * next. */
                 fragment_idx++;
-                pbi->fragments[fragment_idx] = pbi->fragments[0] +
-                  pbi->fragment_sizes[0];
+                pbi->fragments.ptrs[fragment_idx] = pbi->fragments.ptrs[0] +
+                  pbi->fragments.sizes[0];
             }
         }
         /* Split the chunk into partitions read from the bitstream */
@@ -867,12 +873,12 @@
             ptrdiff_t partition_size = read_available_partition_size(
                                                  pbi,
                                                  token_part_sizes,
-                                                 pbi->fragments[fragment_idx],
+                                                 pbi->fragments.ptrs[fragment_idx],
                                                  first_fragment_end,
                                                  fragment_end,
                                                  fragment_idx - 1,
                                                  num_token_partitions);
-            pbi->fragment_sizes[fragment_idx] = (unsigned int)partition_size;
+            pbi->fragments.sizes[fragment_idx] = (unsigned int)partition_size;
             fragment_size -= (unsigned int)partition_size;
             assert(fragment_idx <= num_token_partitions);
             if (fragment_size > 0)
@@ -880,19 +886,20 @@
                 /* The fragment contains an additional partition.
                  * Move to next. */
                 fragment_idx++;
-                pbi->fragments[fragment_idx] =
-                    pbi->fragments[fragment_idx - 1] + partition_size;
+                pbi->fragments.ptrs[fragment_idx] =
+                    pbi->fragments.ptrs[fragment_idx - 1] + partition_size;
             }
         }
     }
 
-    pbi->num_fragments = num_token_partitions + 1;
+    pbi->fragments.count = num_token_partitions + 1;
 
-    for (partition_idx = 1; partition_idx < pbi->num_fragments; ++partition_idx)
+    for (partition_idx = 1; partition_idx < pbi->fragments.count; ++partition_idx)
     {
         if (vp8dx_start_decode(bool_decoder,
-                               pbi->fragments[partition_idx],
-                               pbi->fragment_sizes[partition_idx]))
+                               pbi->fragments.ptrs[partition_idx],
+                               pbi->fragments.sizes[partition_idx],
+                               pbi->decrypt_cb, pbi->decrypt_state))
             vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,
                                "Failed to allocate bool decoder %d",
                                partition_idx);
@@ -979,11 +986,11 @@
 
 int vp8_decode_frame(VP8D_COMP *pbi)
 {
-    vp8_reader *const bc = & pbi->mbc[8];
-    VP8_COMMON *const pc = & pbi->common;
-    MACROBLOCKD *const xd  = & pbi->mb;
-    const unsigned char *data = pbi->fragments[0];
-    const unsigned char *data_end =  data + pbi->fragment_sizes[0];
+    vp8_reader *const bc = &pbi->mbc[8];
+    VP8_COMMON *const pc = &pbi->common;
+    MACROBLOCKD *const xd  = &pbi->mb;
+    const unsigned char *data = pbi->fragments.ptrs[0];
+    const unsigned char *data_end =  data + pbi->fragments.sizes[0];
     ptrdiff_t first_partition_length_in_bytes;
 
     int i, j, k, l;
@@ -1015,18 +1022,30 @@
     }
     else
     {
-        pc->frame_type = (FRAME_TYPE)(data[0] & 1);
-        pc->version = (data[0] >> 1) & 7;
-        pc->show_frame = (data[0] >> 4) & 1;
-        first_partition_length_in_bytes =
-            (data[0] | (data[1] << 8) | (data[2] << 16)) >> 5;
+        unsigned char clear_buffer[10];
+        const unsigned char *clear = data;
+        if (pbi->decrypt_cb)
+        {
+            int n = data_end - data;
+            if (n > 10) n = 10;
+            pbi->decrypt_cb(pbi->decrypt_state, data, clear_buffer, n);
+            clear = clear_buffer;
+        }
 
-        if (!pbi->ec_active && (data + first_partition_length_in_bytes > data_end
+        pc->frame_type = (FRAME_TYPE)(clear[0] & 1);
+        pc->version = (clear[0] >> 1) & 7;
+        pc->show_frame = (clear[0] >> 4) & 1;
+        first_partition_length_in_bytes =
+            (clear[0] | (clear[1] << 8) | (clear[2] << 16)) >> 5;
+
+        if (!pbi->ec_active &&
+            (data + first_partition_length_in_bytes > data_end
             || data + first_partition_length_in_bytes < data))
             vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
                                "Truncated packet or corrupt partition 0 length");
 
         data += 3;
+        clear += 3;
 
         vp8_setup_version(pc);
 
@@ -1039,7 +1058,7 @@
              */
             if (!pbi->ec_active || data + 3 < data_end)
             {
-                if (data[0] != 0x9d || data[1] != 0x01 || data[2] != 0x2a)
+                if (clear[0] != 0x9d || clear[1] != 0x01 || clear[2] != 0x2a)
                     vpx_internal_error(&pc->error, VPX_CODEC_UNSUP_BITSTREAM,
                                    "Invalid frame sync code");
             }
@@ -1050,13 +1069,13 @@
              */
             if (!pbi->ec_active || data + 6 < data_end)
             {
-                pc->Width = (data[3] | (data[4] << 8)) & 0x3fff;
-                pc->horiz_scale = data[4] >> 6;
-                pc->Height = (data[5] | (data[6] << 8)) & 0x3fff;
-                pc->vert_scale = data[6] >> 6;
+                pc->Width = (clear[3] | (clear[4] << 8)) & 0x3fff;
+                pc->horiz_scale = clear[4] >> 6;
+                pc->Height = (clear[5] | (clear[6] << 8)) & 0x3fff;
+                pc->vert_scale = clear[6] >> 6;
             }
             data += 7;
-
+            clear += 7;
         }
         else
         {
@@ -1071,7 +1090,8 @@
 
     init_frame(pbi);
 
-    if (vp8dx_start_decode(bc, data, (unsigned int)(data_end - data)))
+    if (vp8dx_start_decode(bc, data, (unsigned int)(data_end - data),
+                           pbi->decrypt_cb, pbi->decrypt_state))
         vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
                            "Failed to allocate bool decoder 0");
     if (pc->frame_type == KEY_FRAME) {
@@ -1334,11 +1354,11 @@
 #if CONFIG_MULTITHREAD
     if (pbi->b_multithreaded_rd && pc->multi_token_partition != ONE_PARTITION)
     {
-        unsigned int i;
+        unsigned int thread;
         vp8mt_decode_mb_rows(pbi, xd);
         vp8_yv12_extend_frame_borders(yv12_fb_new);
-        for (i = 0; i < pbi->decoding_thread_count; ++i)
-            corrupt_tokens |= pbi->mb_row_di[i].mbd.corrupted;
+        for (thread = 0; thread < pbi->decoding_thread_count; ++thread)
+            corrupt_tokens |= pbi->mb_row_di[thread].mbd.corrupted;
     }
     else
 #endif
diff --git a/libvpx/vp8/decoder/detokenize.h b/libvpx/vp8/decoder/detokenize.h
index 8640bda..f2130b3 100644
--- a/libvpx/vp8/decoder/detokenize.h
+++ b/libvpx/vp8/decoder/detokenize.h
@@ -8,13 +8,12 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
-#ifndef DETOKENIZE_H
-#define DETOKENIZE_H
+#ifndef DETOKENIZE_H_
+#define DETOKENIZE_H_
 
 #include "onyxd_int.h"
 
 void vp8_reset_mb_tokens_context(MACROBLOCKD *x);
 int vp8_decode_mb_tokens(VP8D_COMP *, MACROBLOCKD *);
 
-#endif /* DETOKENIZE_H */
+#endif  // DETOKENIZE_H
diff --git a/libvpx/vp8/decoder/ec_types.h b/libvpx/vp8/decoder/ec_types.h
index ccb5ddb..b24bfd9 100644
--- a/libvpx/vp8/decoder/ec_types.h
+++ b/libvpx/vp8/decoder/ec_types.h
@@ -14,7 +14,6 @@
 #define MAX_OVERLAPS 16
 
 
-
 /* The area (pixel area in Q6) the block pointed to by bmi overlaps
  * another block with.
  */
@@ -48,4 +47,4 @@
     MV_REFERENCE_FRAME ref_frame;
 } EC_BLOCK;
 
-#endif /* VP8_DEC_EC_TYPES_H */
+#endif  // VP8_DEC_EC_TYPES_H
diff --git a/libvpx/vp8/decoder/error_concealment.c b/libvpx/vp8/decoder/error_concealment.c
index 8b2e32b..0b58c98 100644
--- a/libvpx/vp8/decoder/error_concealment.c
+++ b/libvpx/vp8/decoder/error_concealment.c
@@ -8,14 +8,14 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include <assert.h>
+
 #include "error_concealment.h"
 #include "onyxd_int.h"
 #include "decodemv.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vp8/common/findnearmv.h"
 
-#include <assert.h>
-
 #define MIN(x,y) (((x)<(y))?(x):(y))
 #define MAX(x,y) (((x)>(y))?(x):(y))
 
diff --git a/libvpx/vp8/decoder/error_concealment.h b/libvpx/vp8/decoder/error_concealment.h
index 65ae9d9..fb96b36 100644
--- a/libvpx/vp8/decoder/error_concealment.h
+++ b/libvpx/vp8/decoder/error_concealment.h
@@ -9,8 +9,8 @@
  */
 
 
-#ifndef ERROR_CONCEALMENT_H
-#define ERROR_CONCEALMENT_H
+#ifndef ERROR_CONCEALMENT_H_
+#define ERROR_CONCEALMENT_H_
 
 #include "onyxd_int.h"
 #include "ec_types.h"
@@ -38,4 +38,4 @@
  */
 void vp8_conceal_corrupt_mb(MACROBLOCKD *xd);
 
-#endif
+#endif  // ERROR_CONCEALMENT_H_
diff --git a/libvpx/vp8/decoder/onyxd_if.c b/libvpx/vp8/decoder/onyxd_if.c
index 8d6871b..2db3096 100644
--- a/libvpx/vp8/decoder/onyxd_if.c
+++ b/libvpx/vp8/decoder/onyxd_if.c
@@ -25,7 +25,8 @@
 #include <assert.h>
 
 #include "vp8/common/quant_common.h"
-#include "vpx_scale/vpxscale.h"
+#include "./vpx_scale_rtcd.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vp8/common/systemdependent.h"
 #include "vpx_ports/vpx_timer.h"
 #include "detokenize.h"
@@ -41,7 +42,16 @@
 static int get_free_fb (VP8_COMMON *cm);
 static void ref_cnt_fb (int *buf, int *idx, int new_idx);
 
-struct VP8D_COMP * vp8dx_create_decompressor(VP8D_CONFIG *oxcf)
+static void remove_decompressor(VP8D_COMP *pbi)
+{
+#if CONFIG_ERROR_CONCEALMENT
+    vp8_de_alloc_overlap_lists(pbi);
+#endif
+    vp8_remove_common(&pbi->common);
+    vpx_free(pbi);
+}
+
+static struct VP8D_COMP * create_decompressor(VP8D_CONFIG *oxcf)
 {
     VP8D_COMP *pbi = vpx_memalign(32, sizeof(VP8D_COMP));
 
@@ -53,7 +63,7 @@
     if (setjmp(pbi->common.error.jmp))
     {
         pbi->common.error.setjmp = 0;
-        vp8dx_remove_decompressor(pbi);
+        remove_decompressor(pbi);
         return 0;
     }
 
@@ -64,11 +74,6 @@
     pbi->common.current_video_frame = 0;
     pbi->ready_for_new_data = 1;
 
-#if CONFIG_MULTITHREAD
-    pbi->max_threads = oxcf->max_threads;
-    vp8_decoder_create_threads(pbi);
-#endif
-
     /* vp8cx_init_de_quantizer() is first called here. Add check in frame_init_dequantizer() to avoid
      *  unnecessary calling of vp8cx_init_de_quantizer() for every frame.
      */
@@ -91,9 +96,6 @@
 
     pbi->decoded_key_frame = 0;
 
-    pbi->input_fragments = oxcf->input_fragments;
-    pbi->num_fragments = 0;
-
     /* Independent partitions is activated when a frame updates the
      * token probability table to have equal probabilities over the
      * PREV_COEF context.
@@ -105,25 +107,6 @@
     return pbi;
 }
 
-
-void vp8dx_remove_decompressor(VP8D_COMP *pbi)
-{
-    if (!pbi)
-        return;
-
-#if CONFIG_MULTITHREAD
-    if (pbi->b_multithreaded_rd)
-        vp8mt_de_alloc_temp_buffers(pbi, pbi->common.mb_rows);
-    vp8_decoder_remove_threads(pbi);
-#endif
-#if CONFIG_ERROR_CONCEALMENT
-    vp8_de_alloc_overlap_lists(pbi);
-#endif
-    vp8_remove_common(&pbi->common);
-    vpx_free(pbi);
-}
-
-
 vpx_codec_err_t vp8dx_get_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd)
 {
     VP8_COMMON *cm = &pbi->common;
@@ -281,60 +264,13 @@
     return err;
 }
 
-int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size,
-                                  const uint8_t *source,
-                                  int64_t time_stamp)
+int check_fragments_for_errors(VP8D_COMP *pbi)
 {
-#if HAVE_NEON
-    int64_t dx_store_reg[8];
-#endif
-    VP8_COMMON *cm = &pbi->common;
-    int retcode = -1;
-
-    pbi->common.error.error_code = VPX_CODEC_OK;
-
-    if (pbi->num_fragments == 0)
-    {
-        /* New frame, reset fragment pointers and sizes */
-        vpx_memset((void*)pbi->fragments, 0, sizeof(pbi->fragments));
-        vpx_memset(pbi->fragment_sizes, 0, sizeof(pbi->fragment_sizes));
-    }
-    if (pbi->input_fragments && !(source == NULL && size == 0))
-    {
-        /* Store a pointer to this fragment and return. We haven't
-         * received the complete frame yet, so we will wait with decoding.
-         */
-        assert(pbi->num_fragments < MAX_PARTITIONS);
-        pbi->fragments[pbi->num_fragments] = source;
-        pbi->fragment_sizes[pbi->num_fragments] = size;
-        pbi->num_fragments++;
-        if (pbi->num_fragments > (1 << EIGHT_PARTITION) + 1)
-        {
-            pbi->common.error.error_code = VPX_CODEC_UNSUP_BITSTREAM;
-            pbi->common.error.setjmp = 0;
-            pbi->num_fragments = 0;
-            return -1;
-        }
-        return 0;
-    }
-
-    if (!pbi->input_fragments)
-    {
-        pbi->fragments[0] = source;
-        pbi->fragment_sizes[0] = size;
-        pbi->num_fragments = 1;
-    }
-    assert(pbi->common.multi_token_partition <= EIGHT_PARTITION);
-    if (pbi->num_fragments == 0)
-    {
-        pbi->num_fragments = 1;
-        pbi->fragments[0] = NULL;
-        pbi->fragment_sizes[0] = 0;
-    }
-
     if (!pbi->ec_active &&
-        pbi->num_fragments <= 1 && pbi->fragment_sizes[0] == 0)
+        pbi->fragments.count <= 1 && pbi->fragments.sizes[0] == 0)
     {
+        VP8_COMMON *cm = &pbi->common;
+
         /* If error concealment is disabled we won't signal missing frames
          * to the decoder.
          */
@@ -360,12 +296,29 @@
         /* Signal that we have no frame to show. */
         cm->show_frame = 0;
 
-        pbi->num_fragments = 0;
-
         /* Nothing more to do. */
         return 0;
     }
 
+    return 1;
+}
+
+int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size,
+                                  const uint8_t *source,
+                                  int64_t time_stamp)
+{
+#if HAVE_NEON
+    int64_t dx_store_reg[8];
+#endif
+    VP8_COMMON *cm = &pbi->common;
+    int retcode = -1;
+
+    pbi->common.error.error_code = VPX_CODEC_OK;
+
+    retcode = check_fragments_for_errors(pbi);
+    if(retcode <= 0)
+        return retcode;
+
 #if HAVE_NEON
 #if CONFIG_RUNTIME_CPU_DETECT
     if (cm->cpu_caps & HAS_NEON)
@@ -418,7 +371,13 @@
 
     vp8_clear_system_state();
 
-#if CONFIG_ERROR_CONCEALMENT
+    if (cm->show_frame)
+    {
+        cm->current_video_frame++;
+        cm->show_frame_mi = cm->mi;
+    }
+
+    #if CONFIG_ERROR_CONCEALMENT
     /* swap the mode infos to storage for future error concealment */
     if (pbi->ec_enabled && pbi->common.prev_mi)
     {
@@ -440,9 +399,6 @@
     }
 #endif
 
-    if (cm->show_frame)
-        cm->current_video_frame++;
-
     pbi->ready_for_new_data = 0;
     pbi->last_time_stamp = time_stamp;
 
@@ -457,7 +413,6 @@
 #endif
 
     pbi->common.error.setjmp = 0;
-    pbi->num_fragments = 0;
     return retcode;
 }
 int vp8dx_get_raw_frame(VP8D_COMP *pbi, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags)
@@ -520,3 +475,54 @@
     return 0;
 
 }
+
+int vp8_create_decoder_instances(struct frame_buffers *fb, VP8D_CONFIG *oxcf)
+{
+    if(!fb->use_frame_threads)
+    {
+        /* decoder instance for single thread mode */
+        fb->pbi[0] = create_decompressor(oxcf);
+        if(!fb->pbi[0])
+            return VPX_CODEC_ERROR;
+
+#if CONFIG_MULTITHREAD
+        /* enable row-based threading only when use_frame_threads
+         * is disabled */
+        fb->pbi[0]->max_threads = oxcf->max_threads;
+        vp8_decoder_create_threads(fb->pbi[0]);
+#endif
+    }
+    else
+    {
+        /* TODO : create frame threads and decoder instances for each
+         * thread here */
+    }
+
+    return VPX_CODEC_OK;
+}
+
+int vp8_remove_decoder_instances(struct frame_buffers *fb)
+{
+    if(!fb->use_frame_threads)
+    {
+        VP8D_COMP *pbi = fb->pbi[0];
+
+        if (!pbi)
+            return VPX_CODEC_ERROR;
+#if CONFIG_MULTITHREAD
+        if (pbi->b_multithreaded_rd)
+            vp8mt_de_alloc_temp_buffers(pbi, pbi->common.mb_rows);
+        vp8_decoder_remove_threads(pbi);
+#endif
+
+        /* decoder instance for single thread mode */
+        remove_decompressor(pbi);
+    }
+    else
+    {
+        /* TODO : remove frame threads and decoder instances for each
+         * thread here */
+    }
+
+    return VPX_CODEC_OK;
+}
diff --git a/libvpx/vp8/decoder/onyxd_int.h b/libvpx/vp8/decoder/onyxd_int.h
index 0063beb..54a98f7 100644
--- a/libvpx/vp8/decoder/onyxd_int.h
+++ b/libvpx/vp8/decoder/onyxd_int.h
@@ -9,8 +9,9 @@
  */
 
 
-#ifndef __INC_VP8D_INT_H
-#define __INC_VP8D_INT_H
+#ifndef ONYXD_INT_H_
+#define ONYXD_INT_H_
+
 #include "vpx_config.h"
 #include "vp8/common/onyxd.h"
 #include "treereader.h"
@@ -33,6 +34,31 @@
     MACROBLOCKD  mbd;
 } MB_ROW_DEC;
 
+
+typedef struct
+{
+    int enabled;
+    unsigned int count;
+    const unsigned char *ptrs[MAX_PARTITIONS];
+    unsigned int sizes[MAX_PARTITIONS];
+} FRAGMENT_DATA;
+
+#define MAX_FB_MT_DEC 32
+
+struct frame_buffers
+{
+    /*
+     * this struct will be populated with frame buffer management
+     * info in future commits. */
+
+    /* enable/disable frame-based threading */
+    int     use_frame_threads;
+
+    /* decoder instances */
+    struct VP8D_COMP *pbi[MAX_FB_MT_DEC];
+
+};
+
 typedef struct VP8D_COMP
 {
     DECLARE_ALIGNED(16, MACROBLOCKD, mb);
@@ -46,10 +72,7 @@
 
     VP8D_CONFIG oxcf;
 
-
-    const unsigned char *fragments[MAX_PARTITIONS];
-    unsigned int   fragment_sizes[MAX_PARTITIONS];
-    unsigned int   num_fragments;
+    FRAGMENT_DATA fragments;
 
 #if CONFIG_MULTITHREAD
     /* variable for threading */
@@ -95,15 +118,19 @@
 #endif
     int ec_enabled;
     int ec_active;
-    int input_fragments;
     int decoded_key_frame;
     int independent_partitions;
     int frame_corrupt_residual;
 
+    vp8_decrypt_cb *decrypt_cb;
+    void *decrypt_state;
 } VP8D_COMP;
 
 int vp8_decode_frame(VP8D_COMP *cpi);
 
+int vp8_create_decoder_instances(struct frame_buffers *fb, VP8D_CONFIG *oxcf);
+int vp8_remove_decoder_instances(struct frame_buffers *fb);
+
 #if CONFIG_DEBUG
 #define CHECK_MEM_ERROR(lval,expr) do {\
         lval = (expr); \
@@ -121,4 +148,4 @@
     } while(0)
 #endif
 
-#endif
+#endif  // ONYXD_INT_H_
diff --git a/libvpx/vp8/decoder/threading.c b/libvpx/vp8/decoder/threading.c
index 88c06be..7303189 100644
--- a/libvpx/vp8/decoder/threading.c
+++ b/libvpx/vp8/decoder/threading.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #if !defined(WIN32) && CONFIG_OS_SUPPORT == 1
 # include <unistd.h>
 #endif
@@ -36,7 +36,7 @@
 } while (0)
 
 
-extern void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
+void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
 
 static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count)
 {
@@ -343,7 +343,6 @@
 
     for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1))
     {
-       int i;
        int recon_yoffset, recon_uvoffset;
        int mb_col;
        int filter_level;
diff --git a/libvpx/vp8/decoder/treereader.h b/libvpx/vp8/decoder/treereader.h
index 238ff85..9393bb4 100644
--- a/libvpx/vp8/decoder/treereader.h
+++ b/libvpx/vp8/decoder/treereader.h
@@ -9,18 +9,17 @@
  */
 
 
-#ifndef tree_reader_h
-#define tree_reader_h 1
+#ifndef TREEREADER_H_
+#define TREEREADER_H_
 
 #include "vp8/common/treecoder.h"
-
 #include "dboolhuff.h"
 
 typedef BOOL_DECODER vp8_reader;
 
 #define vp8_read vp8dx_decode_bool
 #define vp8_read_literal vp8_decode_value
-#define vp8_read_bit( R) vp8_read( R, vp8_prob_half)
+#define vp8_read_bit(R) vp8_read(R, vp8_prob_half)
 
 
 /* Intent of tree data structure is to make decoding trivial. */
@@ -38,4 +37,4 @@
     return -i;
 }
 
-#endif /* tree_reader_h */
+#endif  // TREEREADER_H_
diff --git a/libvpx/vp8/decoder/asm_dec_offsets.c b/libvpx/vp8/decoder/vp8_asm_dec_offsets.c
similarity index 100%
rename from libvpx/vp8/decoder/asm_dec_offsets.c
rename to libvpx/vp8/decoder/vp8_asm_dec_offsets.c
diff --git a/libvpx/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm b/libvpx/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm
index a644a00..4abe818 100644
--- a/libvpx/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm
+++ b/libvpx/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm
@@ -15,7 +15,7 @@
     EXPORT |vp8_encode_value|
     IMPORT |vp8_validate_buffer_arm|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm
index a1cd467..90a141c 100644
--- a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm
+++ b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_armv5.asm
@@ -12,7 +12,7 @@
     EXPORT |vp8cx_pack_tokens_armv5|
     IMPORT |vp8_validate_buffer_arm|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm
index 1fa5e6c..3a8d17a 100644
--- a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm
+++ b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_mbrow_armv5.asm
@@ -12,7 +12,7 @@
     EXPORT |vp8cx_pack_mb_row_tokens_armv5|
     IMPORT |vp8_validate_buffer_arm|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm
index 90a98fe..e9aa495 100644
--- a/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm
+++ b/libvpx/vp8/encoder/arm/armv5te/vp8_packtokens_partitions_armv5.asm
@@ -12,7 +12,7 @@
     EXPORT |vp8cx_pack_tokens_into_partitions_armv5|
     IMPORT |vp8_validate_buffer_arm|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm b/libvpx/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
index d61f5d9..de35a1e 100644
--- a/libvpx/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
+++ b/libvpx/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
@@ -11,7 +11,7 @@
 
     EXPORT  |vp8_fast_quantize_b_armv6|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/armv6/vp8_subtract_armv6.asm b/libvpx/vp8/encoder/arm/armv6/vp8_subtract_armv6.asm
index f329f8f..05746cf 100644
--- a/libvpx/vp8/encoder/arm/armv6/vp8_subtract_armv6.asm
+++ b/libvpx/vp8/encoder/arm/armv6/vp8_subtract_armv6.asm
@@ -13,7 +13,7 @@
     EXPORT  |vp8_subtract_mbuv_armv6|
     EXPORT  |vp8_subtract_b_armv6|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/dct_arm.c b/libvpx/vp8/encoder/arm/dct_arm.c
index af0fb27..f71300d 100644
--- a/libvpx/vp8/encoder/arm/dct_arm.c
+++ b/libvpx/vp8/encoder/arm/dct_arm.c
@@ -9,7 +9,7 @@
  */
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 #if HAVE_MEDIA
 
diff --git a/libvpx/vp8/encoder/arm/neon/fastquantizeb_neon.asm b/libvpx/vp8/encoder/arm/neon/fastquantizeb_neon.asm
index 1430588..9374310 100644
--- a/libvpx/vp8/encoder/arm/neon/fastquantizeb_neon.asm
+++ b/libvpx/vp8/encoder/arm/neon/fastquantizeb_neon.asm
@@ -12,7 +12,7 @@
     EXPORT  |vp8_fast_quantize_b_neon|
     EXPORT  |vp8_fast_quantize_b_pair_neon|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/neon/shortfdct_neon.asm b/libvpx/vp8/encoder/arm/neon/shortfdct_neon.asm
index 09dd011..5ea8dd8 100644
--- a/libvpx/vp8/encoder/arm/neon/shortfdct_neon.asm
+++ b/libvpx/vp8/encoder/arm/neon/shortfdct_neon.asm
@@ -97,7 +97,7 @@
     vmlal.s16       q11, d6, d17    ; c1*2217 + d1*5352 + 12000
     vmlsl.s16       q12, d6, d16    ; d1*2217 - c1*5352 + 51000
 
-    vmvn.s16        d4, d4
+    vmvn            d4, d4
     vshrn.s32       d1, q11, #16    ; op[4] = (c1*2217 + d1*5352 + 12000)>>16
     vsub.s16        d1, d1, d4      ; op[4] += (d1!=0)
     vshrn.s32       d3, q12, #16    ; op[12]= (d1*2217 - c1*5352 + 51000)>>16
@@ -200,7 +200,7 @@
     vmlal.s16       q11, d27, d17   ; B[4]  = c1*2217 + d1*5352 + 12000
     vmlsl.s16       q12, d27, d16   ; B[12] = d1*2217 - c1*5352 + 51000
 
-    vmvn.s16        q14, q14
+    vmvn            q14, q14
 
     vshrn.s32       d1, q9, #16     ; A[4] = (c1*2217 + d1*5352 + 12000)>>16
     vshrn.s32       d3, q10, #16    ; A[12]= (d1*2217 - c1*5352 + 51000)>>16
diff --git a/libvpx/vp8/encoder/arm/neon/subtract_neon.asm b/libvpx/vp8/encoder/arm/neon/subtract_neon.asm
index 91a328c..5bda786 100644
--- a/libvpx/vp8/encoder/arm/neon/subtract_neon.asm
+++ b/libvpx/vp8/encoder/arm/neon/subtract_neon.asm
@@ -12,7 +12,7 @@
     EXPORT |vp8_subtract_mby_neon|
     EXPORT |vp8_subtract_mbuv_neon|
 
-    INCLUDE asm_enc_offsets.asm
+    INCLUDE vp8_asm_enc_offsets.asm
 
     ARM
     REQUIRE8
diff --git a/libvpx/vp8/encoder/arm/quantize_arm.c b/libvpx/vp8/encoder/arm/quantize_arm.c
index 8999e34..80d9ad0 100644
--- a/libvpx/vp8/encoder/arm/quantize_arm.c
+++ b/libvpx/vp8/encoder/arm/quantize_arm.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/encoder/block.h"
 #include <math.h>
 #include "vpx_mem/vpx_mem.h"
diff --git a/libvpx/vp8/encoder/bitstream.c b/libvpx/vp8/encoder/bitstream.c
index f84ae68..4707ae5 100644
--- a/libvpx/vp8/encoder/bitstream.c
+++ b/libvpx/vp8/encoder/bitstream.c
@@ -50,7 +50,7 @@
 unsigned __int64 Sectionbits[500];
 #endif
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 int intra_mode_stats[10][10][10];
 static unsigned int tree_update_hist [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] [2];
 extern unsigned int active_section;
@@ -90,17 +90,17 @@
 
     if (new_b + (n << 8) < old_b)
     {
-        int i = 0;
+        int j = 0;
 
         vp8_write_bit(w, 1);
 
         do
         {
-            const vp8_prob p = Pnew[i];
+            const vp8_prob p = Pnew[j];
 
-            vp8_write_literal(w, Pcur[i] = p ? p : 1, 8);
+            vp8_write_literal(w, Pcur[j] = p ? p : 1, 8);
         }
-        while (++i < n);
+        while (++j < n);
     }
     else
         vp8_write_bit(w, 0);
@@ -245,15 +245,15 @@
 
             if (L)
             {
-                const unsigned char *pp = b->prob;
-                int v = e >> 1;
-                int n = L;              /* number of bits in v, assumed nonzero */
-                int i = 0;
+                const unsigned char *proba = b->prob;
+                const int v2 = e >> 1;
+                int n2 = L;              /* number of bits in v2, assumed nonzero */
+                i = 0;
 
                 do
                 {
-                    const int bb = (v >> --n) & 1;
-                    split = 1 + (((range - 1) * pp[i>>1]) >> 8);
+                    const int bb = (v2 >> --n2) & 1;
+                    split = 1 + (((range - 1) * proba[i>>1]) >> 8);
                     i = b->tree[i+bb];
 
                     if (bb)
@@ -301,7 +301,7 @@
 
                     lowvalue <<= shift;
                 }
-                while (n);
+                while (n2);
             }
 
 
@@ -493,7 +493,7 @@
 }
 void vp8_convert_rfct_to_prob(VP8_COMP *const cpi)
 {
-    const int *const rfct = cpi->count_mb_ref_frame_usage;
+    const int *const rfct = cpi->mb.count_mb_ref_frame_usage;
     const int rf_intra = rfct[INTRA_FRAME];
     const int rf_inter = rfct[LAST_FRAME] + rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME];
 
@@ -531,7 +531,7 @@
 
     vp8_convert_rfct_to_prob(cpi);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     active_section = 1;
 #endif
 
@@ -580,7 +580,7 @@
             xd->mb_to_top_edge = -((mb_row * 16)) << 3;
             xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
             active_section = 9;
 #endif
 
@@ -593,7 +593,7 @@
             if (rf == INTRA_FRAME)
             {
                 vp8_write(w, 0, cpi->prob_intra_coded);
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                 active_section = 6;
 #endif
                 write_ymode(w, mode, pc->fc.ymode_prob);
@@ -633,13 +633,13 @@
 
                     vp8_mv_ref_probs(mv_ref_p, ct);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                     accum_mv_refs(mode, ct);
 #endif
 
                 }
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                 active_section = 3;
 #endif
 
@@ -649,7 +649,7 @@
                 {
                 case NEWMV:
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                     active_section = 5;
 #endif
 
@@ -692,7 +692,7 @@
 
                         if (blockmode == NEW4X4)
                         {
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                             active_section = 11;
 #endif
                             write_mv(w, &blockmv.as_mv, &best_mv, (const MV_CONTEXT *) mvc);
@@ -769,7 +769,7 @@
                     const B_PREDICTION_MODE L = left_block_mode(m, i);
                     const int bm = m->bmi[i].as_mode;
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                     ++intra_mode_stats [A] [L] [bm];
 #endif
 
@@ -980,6 +980,12 @@
                               int prob_garf
                              )
 {
+    assert(prob_intra >= 0);
+    assert(prob_intra <= 255);
+    assert(prob_last >= 0);
+    assert(prob_last <= 255);
+    assert(prob_garf >= 0);
+    assert(prob_garf <= 255);
     ref_frame_cost[INTRA_FRAME]   = vp8_cost_zero(prob_intra);
     ref_frame_cost[LAST_FRAME]    = vp8_cost_one(prob_intra)
                                     + vp8_cost_zero(prob_last);
@@ -996,7 +1002,7 @@
 {
     int savings = 0;
 
-    const int *const rfct = cpi->count_mb_ref_frame_usage;
+    const int *const rfct = cpi->mb.count_mb_ref_frame_usage;
     const int rf_intra = rfct[INTRA_FRAME];
     const int rf_inter = rfct[LAST_FRAME] + rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME];
     int new_intra, new_last, new_garf, oldtotal, newtotal;
@@ -1154,7 +1160,7 @@
 #endif
 
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                     ++ tree_update_hist [i][j][k][t] [u];
 #endif
 
@@ -1175,7 +1181,7 @@
                 while (++t < ENTROPY_NODES);
 
                 /* Accum token counts for generation of default statistics */
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
                 t = 0;
 
                 do
@@ -1521,7 +1527,7 @@
     if (pc->frame_type != KEY_FRAME)
         vp8_write_bit(bc, pc->refresh_last_frame);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 
     if (pc->frame_type == INTER_FRAME)
         active_section = 0;
@@ -1544,7 +1550,7 @@
     vp8_update_coef_probs(cpi);
 #endif
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     active_section = 2;
 #endif
 
@@ -1555,7 +1561,7 @@
     {
         write_kfmodes(cpi);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
         active_section = 8;
 #endif
     }
@@ -1563,7 +1569,7 @@
     {
         pack_inter_mode_mvs(cpi);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
         active_section = 1;
 #endif
     }
@@ -1681,7 +1687,7 @@
 #endif
 }
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 void print_tree_update_probs()
 {
     int i, j, k, l;
diff --git a/libvpx/vp8/encoder/block.h b/libvpx/vp8/encoder/block.h
index f9d63eb..cf74c7a 100644
--- a/libvpx/vp8/encoder/block.h
+++ b/libvpx/vp8/encoder/block.h
@@ -18,6 +18,9 @@
 #include "vp8/common/entropy.h"
 #include "vpx_ports/mem.h"
 
+#define MAX_MODES 20
+#define MAX_ERROR_BINS 1024
+
 /* motion search site */
 typedef struct
 {
@@ -34,7 +37,7 @@
     /* 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries */
     short *quant;
     short *quant_fast;
-    unsigned char *quant_shift;
+    short *quant_shift;
     short *zbin;
     short *zrun_zbin_boost;
     short *round;
@@ -134,7 +137,19 @@
     int uv_mode_count[VP8_UV_MODES];     /* intra MB type cts this frame */
     int64_t prediction_error;
     int64_t intra_error;
+    int count_mb_ref_frame_usage[MAX_REF_FRAMES];
 
+    int rd_thresh_mult[MAX_MODES];
+    int rd_threshes[MAX_MODES];
+    unsigned int mbs_tested_so_far;
+    unsigned int mode_test_hit_counts[MAX_MODES];
+    int zbin_mode_boost_enabled;
+    int zbin_mode_boost;
+    int last_zbin_mode_boost;
+
+    int last_zbin_over_quant;
+    int zbin_over_quant;
+    int error_bins[MAX_ERROR_BINS];
 
     void (*short_fdct4x4)(short *input, short *output, int pitch);
     void (*short_fdct8x4)(short *input, short *output, int pitch);
diff --git a/libvpx/vp8/encoder/boolhuff.c b/libvpx/vp8/encoder/boolhuff.c
index 74770a2..3b0c03a 100644
--- a/libvpx/vp8/encoder/boolhuff.c
+++ b/libvpx/vp8/encoder/boolhuff.c
@@ -16,7 +16,7 @@
 
 #endif
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 unsigned int active_section = 0;
 #endif
 
diff --git a/libvpx/vp8/encoder/boolhuff.h b/libvpx/vp8/encoder/boolhuff.h
index 8309063..39ab586 100644
--- a/libvpx/vp8/encoder/boolhuff.h
+++ b/libvpx/vp8/encoder/boolhuff.h
@@ -67,7 +67,7 @@
     unsigned int lowvalue = br->lowvalue;
     register unsigned int shift;
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 #if defined(SECTIONBITS_OUTPUT)
 
     if (bit)
diff --git a/libvpx/vp8/encoder/denoising.c b/libvpx/vp8/encoder/denoising.c
index c0dd7c1..7819265 100644
--- a/libvpx/vp8/encoder/denoising.c
+++ b/libvpx/vp8/encoder/denoising.c
@@ -13,7 +13,7 @@
 #include "vp8/common/reconinter.h"
 #include "vpx/vpx_integer.h"
 #include "vpx_mem/vpx_mem.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 static const unsigned int NOISE_MOTION_THRESHOLD = 25 * 25;
 /* SSE_DIFF_THRESHOLD is selected as ~95% confidence assuming
@@ -140,8 +140,7 @@
     int i;
     assert(denoiser);
 
-    /* don't need one for intra start at 1 */
-    for (i = 1; i < MAX_REF_FRAMES; i++)
+    for (i = 0; i < MAX_REF_FRAMES; i++)
     {
         denoiser->yv12_running_avg[i].flags = 0;
 
@@ -175,8 +174,7 @@
     int i;
     assert(denoiser);
 
-    /* we don't have one for intra ref frame */
-    for (i = 1; i < MAX_REF_FRAMES ; i++)
+    for (i = 0; i < MAX_REF_FRAMES ; i++)
     {
         vp8_yv12_de_alloc_frame_buffer(&denoiser->yv12_running_avg[i]);
     }
@@ -208,8 +206,6 @@
         MB_MODE_INFO saved_mbmi;
         MACROBLOCKD *filter_xd = &x->e_mbd;
         MB_MODE_INFO *mbmi = &filter_xd->mode_info_context->mbmi;
-        int mv_col;
-        int mv_row;
         int sse_diff = zero_mv_sse - best_sse;
 
         saved_mbmi = *mbmi;
@@ -291,7 +287,7 @@
     {
         /* Filter. */
         decision = vp8_denoiser_filter(&denoiser->yv12_mc_running_avg,
-                                       &denoiser->yv12_running_avg[LAST_FRAME],
+                                       &denoiser->yv12_running_avg[INTRA_FRAME],
                                        x,
                                        motion_magnitude2,
                                        recon_yoffset, recon_uvoffset);
@@ -303,7 +299,7 @@
          */
         vp8_copy_mem16x16(
                 x->thismb, 16,
-                denoiser->yv12_running_avg[LAST_FRAME].y_buffer + recon_yoffset,
-                denoiser->yv12_running_avg[LAST_FRAME].y_stride);
+                denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset,
+                denoiser->yv12_running_avg[INTRA_FRAME].y_stride);
     }
 }
diff --git a/libvpx/vp8/encoder/encodeframe.c b/libvpx/vp8/encoder/encodeframe.c
index 8828dd9..b550f6b 100644
--- a/libvpx/vp8/encoder/encodeframe.c
+++ b/libvpx/vp8/encoder/encodeframe.c
@@ -10,6 +10,7 @@
 
 
 #include "vpx_config.h"
+#include "vp8_rtcd.h"
 #include "encodemb.h"
 #include "encodemv.h"
 #include "vp8/common/common.h"
@@ -45,7 +46,6 @@
 extern void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
                                       MACROBLOCK *x,
                                       MB_ROW_COMP *mbr_ei,
-                                      int mb_row,
                                       int count);
 static void adjust_act_zbin( VP8_COMP *cpi, MACROBLOCK *x );
 
@@ -530,7 +530,8 @@
              * segmentation map
              */
             if ((cpi->current_layer == 0) &&
-                (cpi->cyclic_refresh_mode_enabled && xd->segmentation_enabled))
+                (cpi->cyclic_refresh_mode_enabled &&
+                 xd->segmentation_enabled))
             {
                 cpi->segmentation_map[map_index+mb_col] = xd->mode_info_context->mbmi.segment_id;
 
@@ -642,8 +643,6 @@
 
     xd->left_context = &cm->left_context;
 
-    vp8_zero(cpi->count_mb_ref_frame_usage)
-
     x->mvc = cm->fc.mvc;
 
     vpx_memset(cm->above_context, 0,
@@ -678,6 +677,7 @@
     vp8_zero(x->uv_mode_count)
     x->prediction_error = 0;
     x->intra_error = 0;
+    vp8_zero(x->count_mb_ref_frame_usage);
 }
 
 static void sum_coef_counts(MACROBLOCK *x, MACROBLOCK *x_thread)
@@ -766,7 +766,7 @@
 
     vp8cx_frame_init_quantizer(cpi);
 
-    vp8_initialize_rd_consts(cpi,
+    vp8_initialize_rd_consts(cpi, x,
                              vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q));
 
     vp8cx_initialize_me_consts(cpi, cm->base_qindex);
@@ -805,7 +805,8 @@
         {
             int i;
 
-            vp8cx_init_mbrthread_data(cpi, x, cpi->mb_row_ei, 1,  cpi->encoding_thread_count);
+            vp8cx_init_mbrthread_data(cpi, x, cpi->mb_row_ei,
+                                      cpi->encoding_thread_count);
 
             for (i = 0; i < cm->mb_rows; i++)
                 cpi->mt_current_mb_col[i] = -1;
@@ -852,11 +853,10 @@
 
             if (xd->segmentation_enabled)
             {
-                int i, j;
+                int j;
 
                 if (xd->segmentation_enabled)
                 {
-
                     for (i = 0; i < cpi->encoding_thread_count; i++)
                     {
                         for (j = 0; j < 4; j++)
@@ -868,7 +868,7 @@
             for (i = 0; i < cpi->encoding_thread_count; i++)
             {
                 int mode_count;
-                int mv_vals;
+                int c_idx;
                 totalrate += cpi->mb_row_ei[i].totalrate;
 
                 cpi->mb.skip_true_count += cpi->mb_row_ei[i].mb.skip_true_count;
@@ -881,18 +881,26 @@
                     cpi->mb.uv_mode_count[mode_count] +=
                         cpi->mb_row_ei[i].mb.uv_mode_count[mode_count];
 
-                for(mv_vals = 0; mv_vals < MVvals; mv_vals++)
+                for(c_idx = 0; c_idx < MVvals; c_idx++)
                 {
-                    cpi->mb.MVcount[0][mv_vals] +=
-                        cpi->mb_row_ei[i].mb.MVcount[0][mv_vals];
-                    cpi->mb.MVcount[1][mv_vals] +=
-                        cpi->mb_row_ei[i].mb.MVcount[1][mv_vals];
+                    cpi->mb.MVcount[0][c_idx] +=
+                        cpi->mb_row_ei[i].mb.MVcount[0][c_idx];
+                    cpi->mb.MVcount[1][c_idx] +=
+                        cpi->mb_row_ei[i].mb.MVcount[1][c_idx];
                 }
 
                 cpi->mb.prediction_error +=
                     cpi->mb_row_ei[i].mb.prediction_error;
                 cpi->mb.intra_error += cpi->mb_row_ei[i].mb.intra_error;
 
+                for(c_idx = 0; c_idx < MAX_REF_FRAMES; c_idx++)
+                    cpi->mb.count_mb_ref_frame_usage[c_idx] +=
+                        cpi->mb_row_ei[i].mb.count_mb_ref_frame_usage[c_idx];
+
+                for(c_idx = 0; c_idx < MAX_ERROR_BINS; c_idx++)
+                    cpi->mb.error_bins[c_idx] +=
+                        cpi->mb_row_ei[i].mb.error_bins[c_idx];
+
                 /* add up counts for each thread */
                 sum_coef_counts(x, &cpi->mb_row_ei[i].mb);
             }
@@ -987,13 +995,14 @@
     {
         int tot_modes;
 
-        tot_modes = cpi->count_mb_ref_frame_usage[INTRA_FRAME]
-                    + cpi->count_mb_ref_frame_usage[LAST_FRAME]
-                    + cpi->count_mb_ref_frame_usage[GOLDEN_FRAME]
-                    + cpi->count_mb_ref_frame_usage[ALTREF_FRAME];
+        tot_modes = cpi->mb.count_mb_ref_frame_usage[INTRA_FRAME]
+                    + cpi->mb.count_mb_ref_frame_usage[LAST_FRAME]
+                    + cpi->mb.count_mb_ref_frame_usage[GOLDEN_FRAME]
+                    + cpi->mb.count_mb_ref_frame_usage[ALTREF_FRAME];
 
         if (tot_modes)
-            cpi->this_frame_percent_intra = cpi->count_mb_ref_frame_usage[INTRA_FRAME] * 100 / tot_modes;
+            cpi->this_frame_percent_intra =
+                cpi->mb.count_mb_ref_frame_usage[INTRA_FRAME] * 100 / tot_modes;
 
     }
 
@@ -1224,17 +1233,17 @@
 
     if (cpi->sf.RD)
     {
-        int zbin_mode_boost_enabled = cpi->zbin_mode_boost_enabled;
+        int zbin_mode_boost_enabled = x->zbin_mode_boost_enabled;
 
         /* Are we using the fast quantizer for the mode selection? */
         if(cpi->sf.use_fastquant_for_pick)
         {
-            cpi->mb.quantize_b      = vp8_fast_quantize_b;
-            cpi->mb.quantize_b_pair = vp8_fast_quantize_b_pair;
+            x->quantize_b      = vp8_fast_quantize_b;
+            x->quantize_b_pair = vp8_fast_quantize_b_pair;
 
             /* the fast quantizer does not use zbin_extra, so
              * do not recalculate */
-            cpi->zbin_mode_boost_enabled = 0;
+            x->zbin_mode_boost_enabled = 0;
         }
         vp8_rd_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate,
                                &distortion, &intra_error);
@@ -1242,12 +1251,12 @@
         /* switch back to the regular quantizer for the encode */
         if (cpi->sf.improved_quant)
         {
-            cpi->mb.quantize_b      = vp8_regular_quantize_b;
-            cpi->mb.quantize_b_pair = vp8_regular_quantize_b_pair;
+            x->quantize_b      = vp8_regular_quantize_b;
+            x->quantize_b_pair = vp8_regular_quantize_b_pair;
         }
 
         /* restore cpi->zbin_mode_boost_enabled */
-        cpi->zbin_mode_boost_enabled = zbin_mode_boost_enabled;
+        x->zbin_mode_boost_enabled = zbin_mode_boost_enabled;
 
     }
     else
@@ -1290,25 +1299,27 @@
     }
 
     {
-        /* Experimental code. Special case for gf and arf zeromv modes.
-         * Increase zbin size to supress noise
+        /* Experimental code.
+         * Special case for gf and arf zeromv modes, for 1 temporal layer.
+         * Increase zbin size to supress noise.
          */
-        cpi->zbin_mode_boost = 0;
-        if (cpi->zbin_mode_boost_enabled)
+        x->zbin_mode_boost = 0;
+        if (x->zbin_mode_boost_enabled)
         {
             if ( xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME )
             {
                 if (xd->mode_info_context->mbmi.mode == ZEROMV)
                 {
-                    if (xd->mode_info_context->mbmi.ref_frame != LAST_FRAME)
-                        cpi->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
+                    if (xd->mode_info_context->mbmi.ref_frame != LAST_FRAME &&
+                        cpi->oxcf.number_of_layers == 1)
+                        x->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
                     else
-                        cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
+                        x->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
                 }
                 else if (xd->mode_info_context->mbmi.mode == SPLITMV)
-                    cpi->zbin_mode_boost = 0;
+                    x->zbin_mode_boost = 0;
                 else
-                    cpi->zbin_mode_boost = MV_ZBIN_BOOST;
+                    x->zbin_mode_boost = MV_ZBIN_BOOST;
             }
         }
 
@@ -1318,7 +1329,7 @@
             vp8_update_zbin_extra(cpi, x);
     }
 
-    cpi->count_mb_ref_frame_usage[xd->mode_info_context->mbmi.ref_frame] ++;
+    x->count_mb_ref_frame_usage[xd->mode_info_context->mbmi.ref_frame] ++;
 
     if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME)
     {
diff --git a/libvpx/vp8/encoder/encodeintra.c b/libvpx/vp8/encoder/encodeintra.c
index 340dd63..cfa4cb9 100644
--- a/libvpx/vp8/encoder/encodeintra.c
+++ b/libvpx/vp8/encoder/encodeintra.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "quantize.h"
 #include "vp8/common/reconintra4x4.h"
 #include "encodemb.h"
diff --git a/libvpx/vp8/encoder/encodemb.c b/libvpx/vp8/encoder/encodemb.c
index 7d494f2..7ed2fe1 100644
--- a/libvpx/vp8/encoder/encodemb.c
+++ b/libvpx/vp8/encoder/encodemb.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "encodemb.h"
 #include "vp8/common/reconinter.h"
 #include "quantize.h"
diff --git a/libvpx/vp8/encoder/encodemv.c b/libvpx/vp8/encoder/encodemv.c
index 0c43d06..2a74ff4 100644
--- a/libvpx/vp8/encoder/encodemv.c
+++ b/libvpx/vp8/encoder/encodemv.c
@@ -16,7 +16,7 @@
 
 #include <math.h>
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 extern unsigned int active_section;
 #endif
 
@@ -359,7 +359,7 @@
     vp8_writer *const w  = cpi->bc;
     MV_CONTEXT *mvc = cpi->common.fc.mvc;
     int flags[2] = {0, 0};
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     active_section = 4;
 #endif
     write_component_probs(
@@ -374,7 +374,7 @@
     if (flags[0] || flags[1])
         vp8_build_component_cost_table(cpi->mb.mvcost, (const MV_CONTEXT *) cpi->common.fc.mvc, flags);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     active_section = 5;
 #endif
 }
diff --git a/libvpx/vp8/encoder/ethreading.c b/libvpx/vp8/encoder/ethreading.c
index 39340f2..d4b17ce 100644
--- a/libvpx/vp8/encoder/ethreading.c
+++ b/libvpx/vp8/encoder/ethreading.c
@@ -214,7 +214,9 @@
                          * vp8cx_encode_inter_macroblock()) back into the
                          * global segmentation map
                          */
-                        if (cpi->cyclic_refresh_mode_enabled && xd->segmentation_enabled)
+                        if ((cpi->current_layer == 0) &&
+                            (cpi->cyclic_refresh_mode_enabled &&
+                             xd->segmentation_enabled))
                         {
                             const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
                             cpi->segmentation_map[map_index + mb_col] = mbmi->segment_id;
@@ -416,13 +418,23 @@
             zd->block[i].dequant = zd->dequant_uv;
         zd->block[24].dequant = zd->dequant_y2;
 #endif
+
+
+        vpx_memcpy(z->rd_threshes, x->rd_threshes, sizeof(x->rd_threshes));
+        vpx_memcpy(z->rd_thresh_mult, x->rd_thresh_mult,
+                   sizeof(x->rd_thresh_mult));
+
+        z->zbin_over_quant = x->zbin_over_quant;
+        z->zbin_mode_boost_enabled = x->zbin_mode_boost_enabled;
+        z->zbin_mode_boost = x->zbin_mode_boost;
+
+        vpx_memset(z->error_bins, 0, sizeof(z->error_bins));
     }
 }
 
 void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
                                MACROBLOCK *x,
                                MB_ROW_COMP *mbr_ei,
-                               int mb_row,
                                int count
                               )
 {
@@ -430,7 +442,6 @@
     VP8_COMMON *const cm = & cpi->common;
     MACROBLOCKD *const xd = & x->e_mbd;
     int i;
-    (void) mb_row;
 
     for (i = 0; i < count; i++)
     {
@@ -478,6 +489,8 @@
         vp8_zero(mb->MVcount);
         mb->prediction_error = 0;
         mb->intra_error = 0;
+        vp8_zero(mb->count_mb_ref_frame_usage);
+        mb->mbs_tested_so_far = 0;
     }
 }
 
diff --git a/libvpx/vp8/encoder/firstpass.c b/libvpx/vp8/encoder/firstpass.c
index b668c8f..433726d 100644
--- a/libvpx/vp8/encoder/firstpass.c
+++ b/libvpx/vp8/encoder/firstpass.c
@@ -12,6 +12,7 @@
 #include <limits.h>
 #include <stdio.h>
 
+#include "./vpx_scale_rtcd.h"
 #include "block.h"
 #include "onyx_int.h"
 #include "vp8/common/variance.h"
@@ -20,7 +21,7 @@
 #include "vp8/common/systemdependent.h"
 #include "mcomp.h"
 #include "firstpass.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "encodemb.h"
 #include "vp8/common/extend.h"
 #include "vpx_mem/vpx_mem.h"
@@ -569,7 +570,7 @@
     /* Initialise the MV cost table to the defaults */
     {
         int flag[2] = {1, 1};
-        vp8_initialize_rd_consts(cpi, vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q));
+        vp8_initialize_rd_consts(cpi, x, vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q));
         vpx_memcpy(cm->fc.mvc, vp8_default_mv_context, sizeof(vp8_default_mv_context));
         vp8_build_component_cost_table(cpi->mb.mvcost, (const MV_CONTEXT *) cm->fc.mvc, flag);
     }
@@ -857,7 +858,9 @@
      */
     if ((cm->current_video_frame > 0) &&
         (cpi->twopass.this_frame_stats.pcnt_inter > 0.20) &&
-        ((cpi->twopass.this_frame_stats.intra_error / cpi->twopass.this_frame_stats.coded_error) > 2.0))
+        ((cpi->twopass.this_frame_stats.intra_error /
+          DOUBLE_DIVIDE_CHECK(cpi->twopass.this_frame_stats.coded_error)) >
+         2.0))
     {
         vp8_yv12_copy_frame(lst_yv12, gld_yv12);
     }
@@ -2115,23 +2118,25 @@
         (cpi->twopass.kf_group_error_left > 0))
     {
         cpi->twopass.gf_group_bits =
-            (int)((double)cpi->twopass.kf_group_bits *
-                  (gf_group_err / (double)cpi->twopass.kf_group_error_left));
+            (int64_t)(cpi->twopass.kf_group_bits *
+                      (gf_group_err / cpi->twopass.kf_group_error_left));
     }
     else
         cpi->twopass.gf_group_bits = 0;
 
-    cpi->twopass.gf_group_bits = (int)(
+    cpi->twopass.gf_group_bits =
         (cpi->twopass.gf_group_bits < 0)
             ? 0
             : (cpi->twopass.gf_group_bits > cpi->twopass.kf_group_bits)
-                ? cpi->twopass.kf_group_bits : cpi->twopass.gf_group_bits);
+                ? cpi->twopass.kf_group_bits : cpi->twopass.gf_group_bits;
 
     /* Clip cpi->twopass.gf_group_bits based on user supplied data rate
      * variability limit (cpi->oxcf.two_pass_vbrmax_section)
      */
-    if (cpi->twopass.gf_group_bits > max_bits * cpi->baseline_gf_interval)
-        cpi->twopass.gf_group_bits = max_bits * cpi->baseline_gf_interval;
+    if (cpi->twopass.gf_group_bits >
+        (int64_t)max_bits * cpi->baseline_gf_interval)
+        cpi->twopass.gf_group_bits =
+            (int64_t)max_bits * cpi->baseline_gf_interval;
 
     /* Reset the file position */
     reset_fpf_position(cpi, start_pos);
@@ -2445,7 +2450,7 @@
          */
         if (cpi->oxcf.error_resilient_mode)
         {
-            cpi->twopass.gf_group_bits = (int)cpi->twopass.kf_group_bits;
+            cpi->twopass.gf_group_bits = cpi->twopass.kf_group_bits;
             cpi->twopass.gf_group_error_left =
                                   (int)cpi->twopass.kf_group_error_left;
             cpi->baseline_gf_interval = cpi->twopass.frames_to_key;
diff --git a/libvpx/vp8/encoder/mcomp.c b/libvpx/vp8/encoder/mcomp.c
index b08c7a5..83c3989 100644
--- a/libvpx/vp8/encoder/mcomp.c
+++ b/libvpx/vp8/encoder/mcomp.c
@@ -18,7 +18,7 @@
 #include <math.h>
 #include "vp8/common/findnearmv.h"
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 static int mv_ref_ct [31] [4] [2];
 static int mv_mode_cts [4] [2];
 #endif
@@ -233,19 +233,18 @@
 
 #if ARCH_X86 || ARCH_X86_64
     MACROBLOCKD *xd = &x->e_mbd;
-    unsigned char *y0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
+    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
     unsigned char *y;
-    int buf_r1, buf_r2, buf_c1, buf_c2;
+    int buf_r1, buf_r2, buf_c1;
 
     /* Clamping to avoid out-of-range data access */
     buf_r1 = ((bestmv->as_mv.row - 3) < x->mv_row_min)?(bestmv->as_mv.row - x->mv_row_min):3;
     buf_r2 = ((bestmv->as_mv.row + 3) > x->mv_row_max)?(x->mv_row_max - bestmv->as_mv.row):3;
     buf_c1 = ((bestmv->as_mv.col - 3) < x->mv_col_min)?(bestmv->as_mv.col - x->mv_col_min):3;
-    buf_c2 = ((bestmv->as_mv.col + 3) > x->mv_col_max)?(x->mv_col_max - bestmv->as_mv.col):3;
     y_stride = 32;
 
     /* Copy to intermediate buffer before searching. */
-    vfp->copymem(y0 - buf_c1 - pre_stride*buf_r1, pre_stride, xd->y_buf, y_stride, 16+buf_r1+buf_r2);
+    vfp->copymem(y_0 - buf_c1 - pre_stride*buf_r1, pre_stride, xd->y_buf, y_stride, 16+buf_r1+buf_r2);
     y = xd->y_buf + y_stride*buf_r1 +buf_c1;
 #else
     unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
@@ -376,12 +375,12 @@
 
 #if ARCH_X86 || ARCH_X86_64
     MACROBLOCKD *xd = &x->e_mbd;
-    unsigned char *y0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
+    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
     unsigned char *y;
 
     y_stride = 32;
     /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
-     vfp->copymem(y0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
+     vfp->copymem(y_0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
      y = xd->y_buf + y_stride + 1;
 #else
      unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
@@ -687,12 +686,12 @@
 
 #if ARCH_X86 || ARCH_X86_64
     MACROBLOCKD *xd = &x->e_mbd;
-    unsigned char *y0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
+    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
     unsigned char *y;
 
     y_stride = 32;
     /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
-    vfp->copymem(y0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
+    vfp->copymem(y_0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
     y = xd->y_buf + y_stride + 1;
 #else
     unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
@@ -1913,7 +1912,7 @@
            + mv_err_cost(&this_mv, center_mv, mvcost, x->errorperbit);
 }
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 void print_mode_context(void)
 {
     FILE *f = fopen("modecont.c", "w");
@@ -1966,8 +1965,8 @@
     fclose(f);
 }
 
-/* MV ref count ENTROPY_STATS stats code */
-#ifdef ENTROPY_STATS
+/* MV ref count VP8_ENTROPY_STATS stats code */
+#ifdef VP8_ENTROPY_STATS
 void init_mv_ref_counts()
 {
     vpx_memset(mv_ref_ct, 0, sizeof(mv_ref_ct));
@@ -2021,6 +2020,6 @@
     }
 }
 
-#endif/* END MV ref count ENTROPY_STATS stats code */
+#endif/* END MV ref count VP8_ENTROPY_STATS stats code */
 
 #endif
diff --git a/libvpx/vp8/encoder/mcomp.h b/libvpx/vp8/encoder/mcomp.h
index 890113f..e36c515 100644
--- a/libvpx/vp8/encoder/mcomp.h
+++ b/libvpx/vp8/encoder/mcomp.h
@@ -15,7 +15,7 @@
 #include "block.h"
 #include "vp8/common/variance.h"
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 extern void init_mv_ref_counts();
 extern void accum_mv_refs(MB_PREDICTION_MODE, const int near_mv_ref_cts[4]);
 #endif
diff --git a/libvpx/vp8/encoder/onyx_if.c b/libvpx/vp8/encoder/onyx_if.c
index c7d81b1..73f6583 100644
--- a/libvpx/vp8/encoder/onyx_if.c
+++ b/libvpx/vp8/encoder/onyx_if.c
@@ -10,6 +10,7 @@
 
 
 #include "vpx_config.h"
+#include "./vpx_scale_rtcd.h"
 #include "vp8/common/onyxc_int.h"
 #include "vp8/common/blockd.h"
 #include "onyx_int.h"
@@ -19,7 +20,7 @@
 #include "mcomp.h"
 #include "firstpass.h"
 #include "psnr.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vp8/common/extend.h"
 #include "ratectrl.h"
 #include "vp8/common/quant_common.h"
@@ -110,7 +111,7 @@
 #endif
 
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 extern int intra_mode_stats[10][10][10];
 #endif
 
@@ -238,7 +239,7 @@
     lc->rate_correction_factor           = cpi->rate_correction_factor;
     lc->key_frame_rate_correction_factor = cpi->key_frame_rate_correction_factor;
     lc->gf_rate_correction_factor        = cpi->gf_rate_correction_factor;
-    lc->zbin_over_quant                  = cpi->zbin_over_quant;
+    lc->zbin_over_quant                  = cpi->mb.zbin_over_quant;
     lc->inter_frame_target               = cpi->inter_frame_target;
     lc->total_byte_count                 = cpi->total_byte_count;
     lc->filter_level                     = cpi->common.filter_level;
@@ -246,8 +247,8 @@
     lc->last_frame_percent_intra         = cpi->last_frame_percent_intra;
 
     memcpy (lc->count_mb_ref_frame_usage,
-            cpi->count_mb_ref_frame_usage,
-            sizeof(cpi->count_mb_ref_frame_usage));
+            cpi->mb.count_mb_ref_frame_usage,
+            sizeof(cpi->mb.count_mb_ref_frame_usage));
 }
 
 static void restore_layer_context(VP8_COMP *cpi, const int layer)
@@ -276,16 +277,135 @@
     cpi->rate_correction_factor           = lc->rate_correction_factor;
     cpi->key_frame_rate_correction_factor = lc->key_frame_rate_correction_factor;
     cpi->gf_rate_correction_factor        = lc->gf_rate_correction_factor;
-    cpi->zbin_over_quant                  = lc->zbin_over_quant;
+    cpi->mb.zbin_over_quant                  = lc->zbin_over_quant;
     cpi->inter_frame_target               = lc->inter_frame_target;
     cpi->total_byte_count                 = lc->total_byte_count;
     cpi->common.filter_level              = lc->filter_level;
 
     cpi->last_frame_percent_intra         = lc->last_frame_percent_intra;
 
-    memcpy (cpi->count_mb_ref_frame_usage,
+    memcpy (cpi->mb.count_mb_ref_frame_usage,
             lc->count_mb_ref_frame_usage,
-            sizeof(cpi->count_mb_ref_frame_usage));
+            sizeof(cpi->mb.count_mb_ref_frame_usage));
+}
+
+static int rescale(int val, int num, int denom)
+{
+    int64_t llnum = num;
+    int64_t llden = denom;
+    int64_t llval = val;
+
+    return (int)(llval * llnum / llden);
+}
+
+static void init_temporal_layer_context(VP8_COMP *cpi,
+                                        VP8_CONFIG *oxcf,
+                                        const int layer,
+                                        double prev_layer_frame_rate)
+{
+    LAYER_CONTEXT *lc = &cpi->layer_context[layer];
+
+    lc->frame_rate = cpi->output_frame_rate / cpi->oxcf.rate_decimator[layer];
+    lc->target_bandwidth = cpi->oxcf.target_bitrate[layer] * 1000;
+
+    lc->starting_buffer_level_in_ms = oxcf->starting_buffer_level;
+    lc->optimal_buffer_level_in_ms  = oxcf->optimal_buffer_level;
+    lc->maximum_buffer_size_in_ms   = oxcf->maximum_buffer_size;
+
+    lc->starting_buffer_level =
+        rescale((int)(oxcf->starting_buffer_level),
+                lc->target_bandwidth, 1000);
+
+    if (oxcf->optimal_buffer_level == 0)
+      lc->optimal_buffer_level = lc->target_bandwidth / 8;
+    else
+      lc->optimal_buffer_level =
+          rescale((int)(oxcf->optimal_buffer_level),
+                  lc->target_bandwidth, 1000);
+
+    if (oxcf->maximum_buffer_size == 0)
+      lc->maximum_buffer_size = lc->target_bandwidth / 8;
+    else
+      lc->maximum_buffer_size =
+          rescale((int)(oxcf->maximum_buffer_size),
+                  lc->target_bandwidth, 1000);
+
+    /* Work out the average size of a frame within this layer */
+    if (layer > 0)
+      lc->avg_frame_size_for_layer =
+          (int)((cpi->oxcf.target_bitrate[layer] -
+                cpi->oxcf.target_bitrate[layer-1]) * 1000 /
+                (lc->frame_rate - prev_layer_frame_rate));
+
+     lc->active_worst_quality         = cpi->oxcf.worst_allowed_q;
+     lc->active_best_quality          = cpi->oxcf.best_allowed_q;
+     lc->avg_frame_qindex             = cpi->oxcf.worst_allowed_q;
+
+     lc->buffer_level                 = lc->starting_buffer_level;
+     lc->bits_off_target              = lc->starting_buffer_level;
+
+     lc->total_actual_bits                 = 0;
+     lc->ni_av_qi                          = 0;
+     lc->ni_tot_qi                         = 0;
+     lc->ni_frames                         = 0;
+     lc->rate_correction_factor            = 1.0;
+     lc->key_frame_rate_correction_factor  = 1.0;
+     lc->gf_rate_correction_factor         = 1.0;
+     lc->inter_frame_target                = 0;
+}
+
+// Upon a run-time change in temporal layers, reset the layer context parameters
+// for any "new" layers. For "existing" layers, let them inherit the parameters
+// from the previous layer state (at the same layer #). In future we may want
+// to better map the previous layer state(s) to the "new" ones.
+static void reset_temporal_layer_change(VP8_COMP *cpi,
+                                        VP8_CONFIG *oxcf,
+                                        const int prev_num_layers)
+{
+    int i;
+    double prev_layer_frame_rate = 0;
+    const int curr_num_layers = cpi->oxcf.number_of_layers;
+    // If the previous state was 1 layer, get current layer context from cpi.
+    // We need this to set the layer context for the new layers below.
+    if (prev_num_layers == 1)
+    {
+        cpi->current_layer = 0;
+        save_layer_context(cpi);
+    }
+    for (i = 0; i < curr_num_layers; i++)
+    {
+        LAYER_CONTEXT *lc = &cpi->layer_context[i];
+        if (i >= prev_num_layers)
+        {
+           init_temporal_layer_context(cpi, oxcf, i, prev_layer_frame_rate);
+        }
+        // The initial buffer levels are set based on their starting levels.
+        // We could set the buffer levels based on the previous state (normalized
+        // properly by the layer bandwidths) but we would need to keep track of
+        // the previous set of layer bandwidths (i.e., target_bitrate[i])
+        // before the layer change. For now, reset to the starting levels.
+        lc->buffer_level = cpi->oxcf.starting_buffer_level_in_ms *
+                           cpi->oxcf.target_bitrate[i];
+        lc->bits_off_target = lc->buffer_level;
+        // TDOD(marpan): Should we set the rate_correction_factor and
+        // active_worst/best_quality to values derived from the previous layer
+        // state (to smooth-out quality dips/rate fluctuation at transition)?
+
+        // We need to treat the 1 layer case separately: oxcf.target_bitrate[i]
+        // is not set for 1 layer, and the restore_layer_context/save_context()
+        // are not called in the encoding loop, so we need to call it here to
+        // pass the layer context state to |cpi|.
+        if (curr_num_layers == 1)
+        {
+            lc->target_bandwidth = cpi->oxcf.target_bandwidth;
+            lc->buffer_level = cpi->oxcf.starting_buffer_level_in_ms *
+                               lc->target_bandwidth  / 1000;
+            lc->bits_off_target = lc->buffer_level;
+            restore_layer_context(cpi, 0);
+        }
+        prev_layer_frame_rate =  cpi->output_frame_rate /
+                                 cpi->oxcf.rate_decimator[i];
+    }
 }
 
 static void setup_features(VP8_COMP *cpi)
@@ -640,11 +760,9 @@
     for (i = 0; i < MAX_MODES; i ++)
     {
         cpi->mode_check_freq[i] = 0;
-        cpi->mode_test_hit_counts[i] = 0;
-        cpi->mode_chosen_counts[i] = 0;
     }
 
-    cpi->mbs_tested_so_far = 0;
+    cpi->mb.mbs_tested_so_far = 0;
 
     /* best quality defaults */
     sf->RD = 1;
@@ -826,7 +944,7 @@
         {
             unsigned int sum = 0;
             unsigned int total_mbs = cm->MBs;
-            int i, thresh;
+            int thresh;
             unsigned int total_skip;
 
             int min = 2000;
@@ -838,7 +956,7 @@
 
             for (i = 0; i < min; i++)
             {
-                sum += cpi->error_bins[i];
+                sum += cpi->mb.error_bins[i];
             }
 
             total_skip = sum;
@@ -847,7 +965,7 @@
             /* i starts from 2 to make sure thresh started from 2048 */
             for (; i < 1024; i++)
             {
-                sum += cpi->error_bins[i];
+                sum += cpi->mb.error_bins[i];
 
                 if (10 * sum >= (unsigned int)(cpi->Speed - 6)*(total_mbs - total_skip))
                     break;
@@ -902,7 +1020,7 @@
         if (Speed >= 15)
             sf->half_pixel_search = 0;
 
-        vpx_memset(cpi->error_bins, 0, sizeof(cpi->error_bins));
+        vpx_memset(cpi->mb.error_bins, 0, sizeof(cpi->mb.error_bins));
 
     }; /* switch */
 
@@ -1077,10 +1195,7 @@
     }
 
     /* Data used for real time vc mode to see if gf needs refreshing */
-    cpi->inter_zz_count = 0;
     cpi->zeromv_count = 0;
-    cpi->gf_bad_count = 0;
-    cpi->gf_update_recommended = 0;
 
 
     /* Structures used to monitor GF usage */
@@ -1204,17 +1319,6 @@
 }
 
 
-static int
-rescale(int val, int num, int denom)
-{
-    int64_t llnum = num;
-    int64_t llden = denom;
-    int64_t llval = val;
-
-    return (int)(llval * llnum / llden);
-}
-
-
 static void init_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
 {
     VP8_COMMON *cm = &cpi->common;
@@ -1269,59 +1373,9 @@
 
         for (i=0; i<cpi->oxcf.number_of_layers; i++)
         {
-            LAYER_CONTEXT *lc = &cpi->layer_context[i];
-
-            /* Layer configuration */
-            lc->frame_rate =
-                        cpi->output_frame_rate / cpi->oxcf.rate_decimator[i];
-            lc->target_bandwidth = cpi->oxcf.target_bitrate[i] * 1000;
-
-            lc->starting_buffer_level_in_ms = oxcf->starting_buffer_level;
-            lc->optimal_buffer_level_in_ms  = oxcf->optimal_buffer_level;
-            lc->maximum_buffer_size_in_ms   = oxcf->maximum_buffer_size;
-
-            lc->starting_buffer_level =
-              rescale((int)(oxcf->starting_buffer_level),
-                          lc->target_bandwidth, 1000);
-
-            if (oxcf->optimal_buffer_level == 0)
-                lc->optimal_buffer_level = lc->target_bandwidth / 8;
-            else
-                lc->optimal_buffer_level =
-                  rescale((int)(oxcf->optimal_buffer_level),
-                          lc->target_bandwidth, 1000);
-
-            if (oxcf->maximum_buffer_size == 0)
-                lc->maximum_buffer_size = lc->target_bandwidth / 8;
-            else
-                lc->maximum_buffer_size =
-                  rescale((int)oxcf->maximum_buffer_size,
-                          lc->target_bandwidth, 1000);
-
-            /* Work out the average size of a frame within this layer */
-            if (i > 0)
-                lc->avg_frame_size_for_layer =
-                  (int)((cpi->oxcf.target_bitrate[i] -
-                         cpi->oxcf.target_bitrate[i-1]) * 1000 /
-                        (lc->frame_rate - prev_layer_frame_rate));
-
-            lc->active_worst_quality         = cpi->oxcf.worst_allowed_q;
-            lc->active_best_quality          = cpi->oxcf.best_allowed_q;
-            lc->avg_frame_qindex             = cpi->oxcf.worst_allowed_q;
-
-            lc->buffer_level                 = lc->starting_buffer_level;
-            lc->bits_off_target              = lc->starting_buffer_level;
-
-            lc->total_actual_bits                 = 0;
-            lc->ni_av_qi                          = 0;
-            lc->ni_tot_qi                         = 0;
-            lc->ni_frames                         = 0;
-            lc->rate_correction_factor            = 1.0;
-            lc->key_frame_rate_correction_factor  = 1.0;
-            lc->gf_rate_correction_factor         = 1.0;
-            lc->inter_frame_target                = 0;
-
-            prev_layer_frame_rate = lc->frame_rate;
+            init_temporal_layer_context(cpi, oxcf, i, prev_layer_frame_rate);
+            prev_layer_frame_rate = cpi->output_frame_rate /
+                                    cpi->oxcf.rate_decimator[i];
         }
     }
 
@@ -1388,7 +1442,7 @@
 void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
 {
     VP8_COMMON *cm = &cpi->common;
-    int last_w, last_h;
+    int last_w, last_h, prev_number_of_layers;
 
     if (!cpi)
         return;
@@ -1413,6 +1467,7 @@
 
     last_w = cpi->oxcf.Width;
     last_h = cpi->oxcf.Height;
+    prev_number_of_layers = cpi->oxcf.number_of_layers;
 
     cpi->oxcf = *oxcf;
 
@@ -1605,6 +1660,16 @@
 
     cpi->target_bandwidth = cpi->oxcf.target_bandwidth;
 
+    // Check if the number of temporal layers has changed, and if so reset the
+    // pattern counter and set/initialize the temporal layer context for the
+    // new layer configuration.
+    if (cpi->oxcf.number_of_layers != prev_number_of_layers)
+    {
+        // If the number of temporal layers are changed we must start at the
+        // base of the pattern cycle, so reset temporal_pattern_counter.
+        cpi->temporal_pattern_counter = 0;
+        reset_temporal_layer_change(cpi, oxcf, prev_number_of_layers);
+    }
 
     cm->Width       = cpi->oxcf.Width;
     cm->Height      = cpi->oxcf.Height;
@@ -1742,6 +1807,7 @@
 
     memcpy(cpi->base_skip_false_prob, vp8cx_base_skip_false_prob, sizeof(vp8cx_base_skip_false_prob));
     cpi->common.current_video_frame   = 0;
+    cpi->temporal_pattern_counter     = 0;
     cpi->kf_overspend_bits            = 0;
     cpi->kf_bitrate_adjustment        = 0;
     cpi->frames_till_gf_update_due      = 0;
@@ -1809,7 +1875,7 @@
     else
         cpi->cyclic_refresh_map = (signed char *) NULL;
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     init_context_counters();
 #endif
 
@@ -1924,10 +1990,10 @@
     /* Set starting values of RD threshold multipliers (128 = *1) */
     for (i = 0; i < MAX_MODES; i++)
     {
-        cpi->rd_thresh_mult[i] = 128;
+        cpi->mb.rd_thresh_mult[i] = 128;
     }
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
     init_mv_ref_counts();
 #endif
 
@@ -2002,7 +2068,7 @@
     cpi->refining_search_sad = vp8_refining_search_sad;
 
     /* make sure frame 1 is okay */
-    cpi->error_bins[0] = cpi->common.MBs;
+    cpi->mb.error_bins[0] = cpi->common.MBs;
 
     /* vp8cx_init_quantizer() is first called here. Add check in
      * vp8cx_frame_init_quantizer() so that vp8cx_init_quantizer is only
@@ -2064,7 +2130,7 @@
 
 #endif
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
         print_context_counters();
         print_tree_update_probs();
         print_mode_context();
@@ -2246,7 +2312,7 @@
         }
 #endif
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
         {
             int i, j, k;
             FILE *fmode = fopen("modecontext.c", "w");
@@ -2591,7 +2657,7 @@
         Scale2Ratio(cm->horiz_scale, &hr, &hs);
         Scale2Ratio(cm->vert_scale, &vr, &vs);
 
-        vp8_scale_frame(sd, &cpi->scaled_source, cm->temp_scale_frame.y_buffer,
+        vpx_scale_frame(sd, &cpi->scaled_source, cm->temp_scale_frame.y_buffer,
                         tmp_height, hs, hr, vs, vr, 0);
 
         vp8_yv12_extend_frame_borders(&cpi->scaled_source);
@@ -2689,7 +2755,7 @@
     /* Clear the alternate reference update pending flag. */
     cpi->source_alt_ref_pending = 0;
 
-    /* Set the alternate refernce frame active flag */
+    /* Set the alternate reference frame active flag */
     cpi->source_alt_ref_active = 1;
 
 
@@ -2775,10 +2841,14 @@
 
         if (cpi->common.frames_since_golden > 1)
         {
-            cpi->recent_ref_frame_usage[INTRA_FRAME] += cpi->count_mb_ref_frame_usage[INTRA_FRAME];
-            cpi->recent_ref_frame_usage[LAST_FRAME] += cpi->count_mb_ref_frame_usage[LAST_FRAME];
-            cpi->recent_ref_frame_usage[GOLDEN_FRAME] += cpi->count_mb_ref_frame_usage[GOLDEN_FRAME];
-            cpi->recent_ref_frame_usage[ALTREF_FRAME] += cpi->count_mb_ref_frame_usage[ALTREF_FRAME];
+            cpi->recent_ref_frame_usage[INTRA_FRAME] +=
+                cpi->mb.count_mb_ref_frame_usage[INTRA_FRAME];
+            cpi->recent_ref_frame_usage[LAST_FRAME] +=
+                cpi->mb.count_mb_ref_frame_usage[LAST_FRAME];
+            cpi->recent_ref_frame_usage[GOLDEN_FRAME] +=
+                cpi->mb.count_mb_ref_frame_usage[GOLDEN_FRAME];
+            cpi->recent_ref_frame_usage[ALTREF_FRAME] +=
+                cpi->mb.count_mb_ref_frame_usage[ALTREF_FRAME];
         }
     }
 }
@@ -2790,7 +2860,7 @@
 {
     VP8_COMMON *cm = &cpi->common;
 
-    const int *const rfct = cpi->count_mb_ref_frame_usage;
+    const int *const rfct = cpi->mb.count_mb_ref_frame_usage;
     const int rf_intra = rfct[INTRA_FRAME];
     const int rf_inter = rfct[LAST_FRAME] + rfct[GOLDEN_FRAME] + rfct[ALTREF_FRAME];
 
@@ -2815,6 +2885,8 @@
         if (cpi->common.refresh_alt_ref_frame)
         {
             cpi->prob_intra_coded += 40;
+            if (cpi->prob_intra_coded > 255)
+                cpi->prob_intra_coded = 255;
             cpi->prob_last_coded = 200;
             cpi->prob_gf_coded = 1;
         }
@@ -3131,6 +3203,57 @@
         cpi->current_ref_frames[LAST_FRAME] = cm->current_video_frame;
 #endif
     }
+
+#if CONFIG_TEMPORAL_DENOISING
+    if (cpi->oxcf.noise_sensitivity)
+    {
+        /* we shouldn't have to keep multiple copies as we know in advance which
+         * buffer we should start - for now to get something up and running
+         * I've chosen to copy the buffers
+         */
+        if (cm->frame_type == KEY_FRAME)
+        {
+            int i;
+            vp8_yv12_copy_frame(
+                    cpi->Source,
+                    &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
+
+            vp8_yv12_extend_frame_borders(
+                    &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
+
+            for (i = 2; i < MAX_REF_FRAMES - 1; i++)
+                vp8_yv12_copy_frame(
+                        &cpi->denoiser.yv12_running_avg[LAST_FRAME],
+                        &cpi->denoiser.yv12_running_avg[i]);
+        }
+        else /* For non key frames */
+        {
+            vp8_yv12_extend_frame_borders(
+                    &cpi->denoiser.yv12_running_avg[INTRA_FRAME]);
+
+            if (cm->refresh_alt_ref_frame || cm->copy_buffer_to_arf)
+            {
+                vp8_yv12_copy_frame(
+                        &cpi->denoiser.yv12_running_avg[INTRA_FRAME],
+                        &cpi->denoiser.yv12_running_avg[ALTREF_FRAME]);
+            }
+            if (cm->refresh_golden_frame || cm->copy_buffer_to_gf)
+            {
+                vp8_yv12_copy_frame(
+                        &cpi->denoiser.yv12_running_avg[INTRA_FRAME],
+                        &cpi->denoiser.yv12_running_avg[GOLDEN_FRAME]);
+            }
+            if(cm->refresh_last_frame)
+            {
+                vp8_yv12_copy_frame(
+                        &cpi->denoiser.yv12_running_avg[INTRA_FRAME],
+                        &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
+            }
+        }
+
+    }
+#endif
+
 }
 
 void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
@@ -3174,51 +3297,6 @@
     }
 
     vp8_yv12_extend_frame_borders(cm->frame_to_show);
-#if CONFIG_TEMPORAL_DENOISING
-    if (cpi->oxcf.noise_sensitivity)
-    {
-
-
-        /* we shouldn't have to keep multiple copies as we know in advance which
-         * buffer we should start - for now to get something up and running
-         * I've chosen to copy the buffers
-         */
-        if (cm->frame_type == KEY_FRAME)
-        {
-            int i;
-            vp8_yv12_copy_frame(
-                    cpi->Source,
-                    &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
-
-            vp8_yv12_extend_frame_borders(
-                    &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
-
-            for (i = 2; i < MAX_REF_FRAMES - 1; i++)
-                vp8_yv12_copy_frame(
-                        cpi->Source,
-                        &cpi->denoiser.yv12_running_avg[i]);
-        }
-        else /* For non key frames */
-        {
-            vp8_yv12_extend_frame_borders(
-                    &cpi->denoiser.yv12_running_avg[LAST_FRAME]);
-
-            if (cm->refresh_alt_ref_frame || cm->copy_buffer_to_arf)
-            {
-                vp8_yv12_copy_frame(
-                        &cpi->denoiser.yv12_running_avg[LAST_FRAME],
-                        &cpi->denoiser.yv12_running_avg[ALTREF_FRAME]);
-            }
-            if (cm->refresh_golden_frame || cm->copy_buffer_to_gf)
-            {
-                vp8_yv12_copy_frame(
-                        &cpi->denoiser.yv12_running_avg[LAST_FRAME],
-                        &cpi->denoiser.yv12_running_avg[GOLDEN_FRAME]);
-            }
-        }
-
-    }
-#endif
 
 }
 
@@ -3302,19 +3380,19 @@
     cm->copy_buffer_to_arf = 0;
 
     /* Clear zbin over-quant value and mode boost values. */
-    cpi->zbin_over_quant = 0;
-    cpi->zbin_mode_boost = 0;
+    cpi->mb.zbin_over_quant = 0;
+    cpi->mb.zbin_mode_boost = 0;
 
     /* Enable or disable mode based tweaking of the zbin
      * For 2 Pass Only used where GF/ARF prediction quality
      * is above a threshold
      */
-    cpi->zbin_mode_boost_enabled = 1;
+    cpi->mb.zbin_mode_boost_enabled = 1;
     if (cpi->pass == 2)
     {
         if ( cpi->gfu_boost <= 400 )
         {
-            cpi->zbin_mode_boost_enabled = 0;
+            cpi->mb.zbin_mode_boost_enabled = 0;
         }
     }
 
@@ -3324,7 +3402,7 @@
     else
         cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = 0;
 
-    /* Check to see if a key frame is signalled
+    /* Check to see if a key frame is signaled
      * For two pass with auto key frame enabled cm->frame_type may already
      * be set, but not for one pass.
      */
@@ -3381,7 +3459,7 @@
         /* Reset the RD threshold multipliers to default of * 1 (128) */
         for (i = 0; i < MAX_MODES; i++)
         {
-            cpi->rd_thresh_mult[i] = 128;
+            cpi->mb.rd_thresh_mult[i] = 128;
         }
     }
 
@@ -3459,7 +3537,7 @@
         /* Note that we should not throw out a key frame (especially when
          * spatial resampling is enabled).
          */
-        if ((cm->frame_type == KEY_FRAME))
+        if (cm->frame_type == KEY_FRAME)
         {
             cpi->decimation_count = cpi->decimation_factor;
         }
@@ -3477,6 +3555,8 @@
 
             cm->current_video_frame++;
             cpi->frames_since_key++;
+            // We advance the temporal pattern for dropped frames.
+            cpi->temporal_pattern_counter++;
 
 #if CONFIG_INTERNAL_STATS
             cpi->count ++;
@@ -3518,6 +3598,8 @@
 #endif
         cm->current_video_frame++;
         cpi->frames_since_key++;
+        // We advance the temporal pattern for dropped frames.
+        cpi->temporal_pattern_counter++;
         return;
     }
 
@@ -4070,8 +4152,9 @@
                 q_low = (Q < q_high) ? (Q + 1) : q_high;
 
                 /* If we are using over quant do the same for zbin_oq_low */
-                if (cpi->zbin_over_quant > 0)
-                    zbin_oq_low = (cpi->zbin_over_quant < zbin_oq_high) ? (cpi->zbin_over_quant + 1) : zbin_oq_high;
+                if (cpi->mb.zbin_over_quant > 0)
+                    zbin_oq_low = (cpi->mb.zbin_over_quant < zbin_oq_high) ?
+                        (cpi->mb.zbin_over_quant + 1) : zbin_oq_high;
 
                 if (undershoot_seen)
                 {
@@ -4087,11 +4170,13 @@
                      * is max)
                      */
                     if (Q < MAXQ)
-                        cpi->zbin_over_quant = 0;
+                        cpi->mb.zbin_over_quant = 0;
                     else
                     {
-                        zbin_oq_low = (cpi->zbin_over_quant < zbin_oq_high) ? (cpi->zbin_over_quant + 1) : zbin_oq_high;
-                        cpi->zbin_over_quant = (zbin_oq_high + zbin_oq_low) / 2;
+                        zbin_oq_low = (cpi->mb.zbin_over_quant < zbin_oq_high) ?
+                            (cpi->mb.zbin_over_quant + 1) : zbin_oq_high;
+                        cpi->mb.zbin_over_quant =
+                            (zbin_oq_high + zbin_oq_low) / 2;
                     }
                 }
                 else
@@ -4104,7 +4189,9 @@
 
                     Q = vp8_regulate_q(cpi, cpi->this_frame_target);
 
-                    while (((Q < q_low) || (cpi->zbin_over_quant < zbin_oq_low)) && (Retries < 10))
+                    while (((Q < q_low) ||
+                        (cpi->mb.zbin_over_quant < zbin_oq_low)) &&
+                        (Retries < 10))
                     {
                         vp8_update_rate_correction_factors(cpi, 0);
                         Q = vp8_regulate_q(cpi, cpi->this_frame_target);
@@ -4117,12 +4204,13 @@
             /* Frame is too small */
             else
             {
-                if (cpi->zbin_over_quant == 0)
+                if (cpi->mb.zbin_over_quant == 0)
                     /* Lower q_high if not using over quant */
                     q_high = (Q > q_low) ? (Q - 1) : q_low;
                 else
                     /* else lower zbin_oq_high */
-                    zbin_oq_high = (cpi->zbin_over_quant > zbin_oq_low) ? (cpi->zbin_over_quant - 1) : zbin_oq_low;
+                    zbin_oq_high = (cpi->mb.zbin_over_quant > zbin_oq_low) ?
+                        (cpi->mb.zbin_over_quant - 1) : zbin_oq_low;
 
                 if (overshoot_seen)
                 {
@@ -4138,9 +4226,10 @@
                      * is max)
                      */
                     if (Q < MAXQ)
-                        cpi->zbin_over_quant = 0;
+                        cpi->mb.zbin_over_quant = 0;
                     else
-                        cpi->zbin_over_quant = (zbin_oq_high + zbin_oq_low) / 2;
+                        cpi->mb.zbin_over_quant =
+                            (zbin_oq_high + zbin_oq_low) / 2;
                 }
                 else
                 {
@@ -4163,7 +4252,9 @@
                         q_low = Q;
                     }
 
-                    while (((Q > q_high) || (cpi->zbin_over_quant > zbin_oq_high)) && (Retries < 10))
+                    while (((Q > q_high) ||
+                        (cpi->mb.zbin_over_quant > zbin_oq_high)) &&
+                        (Retries < 10))
                     {
                         vp8_update_rate_correction_factors(cpi, 0);
                         Q = vp8_regulate_q(cpi, cpi->this_frame_target);
@@ -4181,7 +4272,9 @@
                 Q = q_low;
 
             /* Clamp cpi->zbin_over_quant */
-            cpi->zbin_over_quant = (cpi->zbin_over_quant < zbin_oq_low) ? zbin_oq_low : (cpi->zbin_over_quant > zbin_oq_high) ? zbin_oq_high : cpi->zbin_over_quant;
+            cpi->mb.zbin_over_quant = (cpi->mb.zbin_over_quant < zbin_oq_low) ?
+                zbin_oq_low : (cpi->mb.zbin_over_quant > zbin_oq_high) ?
+                    zbin_oq_high : cpi->mb.zbin_over_quant;
 
             Loop = Q != last_q;
         }
@@ -4263,7 +4356,6 @@
         /* Point to beginning of MODE_INFO arrays. */
         MODE_INFO *tmp = cm->mi;
 
-        cpi->inter_zz_count = 0;
         cpi->zeromv_count = 0;
 
         if(cm->frame_type != KEY_FRAME)
@@ -4272,8 +4364,6 @@
             {
                 for (mb_col = 0; mb_col < cm->mb_cols; mb_col ++)
                 {
-                    if(tmp->mbmi.mode == ZEROMV && tmp->mbmi.ref_frame == LAST_FRAME)
-                        cpi->inter_zz_count++;
                     if(tmp->mbmi.mode == ZEROMV)
                         cpi->zeromv_count++;
                     tmp++;
@@ -4583,9 +4673,6 @@
                         cm->frame_type, cm->refresh_golden_frame,
                         cm->refresh_alt_ref_frame);
 
-            for (i = 0; i < MAX_MODES; i++)
-                fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
-
             fprintf(fmodes, "\n");
 
             fclose(fmodes);
@@ -4680,6 +4767,7 @@
     {
         cm->current_video_frame++;
         cpi->frames_since_key++;
+        cpi->temporal_pattern_counter++;
     }
 
     /* reset to normal state now that we are done. */
@@ -4703,67 +4791,6 @@
 
 
 }
-
-
-static void check_gf_quality(VP8_COMP *cpi)
-{
-    VP8_COMMON *cm = &cpi->common;
-    int gf_active_pct = (100 * cpi->gf_active_count) / (cm->mb_rows * cm->mb_cols);
-    int gf_ref_usage_pct = (cpi->count_mb_ref_frame_usage[GOLDEN_FRAME] * 100) / (cm->mb_rows * cm->mb_cols);
-    int last_ref_zz_useage = (cpi->inter_zz_count * 100) / (cm->mb_rows * cm->mb_cols);
-
-    /* Gf refresh is not currently being signalled */
-    if (cpi->gf_update_recommended == 0)
-    {
-        if (cpi->common.frames_since_golden > 7)
-        {
-            /* Low use of gf */
-            if ((gf_active_pct < 10) || ((gf_active_pct + gf_ref_usage_pct) < 15))
-            {
-                /* ...but last frame zero zero usage is reasonbable so a
-                 * new gf might be appropriate
-                 */
-                if (last_ref_zz_useage >= 25)
-                {
-                    cpi->gf_bad_count ++;
-
-                    /* Check that the condition is stable */
-                    if (cpi->gf_bad_count >= 8)
-                    {
-                        cpi->gf_update_recommended = 1;
-                        cpi->gf_bad_count = 0;
-                    }
-                }
-                else
-                    /* Restart count as the background is not stable enough */
-                    cpi->gf_bad_count = 0;
-            }
-            else
-                /* Gf useage has picked up so reset count */
-                cpi->gf_bad_count = 0;
-        }
-    }
-    /* If the signal is set but has not been read should we cancel it. */
-    else if (last_ref_zz_useage < 15)
-    {
-        cpi->gf_update_recommended = 0;
-        cpi->gf_bad_count = 0;
-    }
-
-#if 0
-    {
-        FILE *f = fopen("gfneeded.stt", "a");
-        fprintf(f, "%10d %10d %10d %10d %10ld \n",
-                cm->current_video_frame,
-                cpi->common.frames_since_golden,
-                gf_active_pct, gf_ref_usage_pct,
-                cpi->gf_update_recommended);
-        fclose(f);
-    }
-
-#endif
-}
-
 #if !(CONFIG_REALTIME_ONLY)
 static void Pass2Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest, unsigned char * dest_end, unsigned int *frame_flags)
 {
@@ -5060,15 +5087,13 @@
 
         /* Restore layer specific context & set frame rate */
         layer = cpi->oxcf.layer_id[
-                            cm->current_video_frame % cpi->oxcf.periodicity];
+                cpi->temporal_pattern_counter % cpi->oxcf.periodicity];
         restore_layer_context (cpi, layer);
         vp8_new_frame_rate (cpi, cpi->layer_context[layer].frame_rate);
     }
 
     if (cpi->compressor_speed == 2)
     {
-        if (cpi->oxcf.number_of_layers == 1)
-            check_gf_quality(cpi);
         vpx_usec_timer_start(&tsctimer);
         vpx_usec_timer_start(&ticktimer);
     }
@@ -5229,7 +5254,7 @@
 
         if (cm->show_frame)
         {
-
+            cpi->common.show_frame_mi = cpi->common.mi;
             cpi->count ++;
 
             if (cpi->b_calculate_psnr)
@@ -5410,6 +5435,7 @@
 #endif
 
 #if CONFIG_POSTPROC
+        cpi->common.show_frame_mi = cpi->common.mi;
         ret = vp8_post_proc_frame(&cpi->common, dest, flags);
 #else
 
diff --git a/libvpx/vp8/encoder/onyx_int.h b/libvpx/vp8/encoder/onyx_int.h
index ed9c762..5120fcc 100644
--- a/libvpx/vp8/encoder/onyx_int.h
+++ b/libvpx/vp8/encoder/onyx_int.h
@@ -43,7 +43,7 @@
 #define AF_THRESH   25
 #define AF_THRESH2  100
 #define ARF_DECAY_THRESH 12
-#define MAX_MODES 20
+
 
 #define MIN_THRESHMULT  32
 #define MAX_THRESHMULT  512
@@ -282,17 +282,17 @@
 {
 
     DECLARE_ALIGNED(16, short, Y1quant[QINDEX_RANGE][16]);
-    DECLARE_ALIGNED(16, unsigned char, Y1quant_shift[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, Y1quant_shift[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, Y1zbin[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, Y1round[QINDEX_RANGE][16]);
 
     DECLARE_ALIGNED(16, short, Y2quant[QINDEX_RANGE][16]);
-    DECLARE_ALIGNED(16, unsigned char, Y2quant_shift[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, Y2quant_shift[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, Y2zbin[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, Y2round[QINDEX_RANGE][16]);
 
     DECLARE_ALIGNED(16, short, UVquant[QINDEX_RANGE][16]);
-    DECLARE_ALIGNED(16, unsigned char, UVquant_shift[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, UVquant_shift[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, UVzbin[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, UVround[QINDEX_RANGE][16]);
 
@@ -349,13 +349,8 @@
     int ambient_err;
 
     unsigned int mode_check_freq[MAX_MODES];
-    unsigned int mode_test_hit_counts[MAX_MODES];
-    unsigned int mode_chosen_counts[MAX_MODES];
-    unsigned int mbs_tested_so_far;
 
-    int rd_thresh_mult[MAX_MODES];
     int rd_baseline_thresh[MAX_MODES];
-    int rd_threshes[MAX_MODES];
 
     int RDMULT;
     int RDDIV ;
@@ -416,12 +411,6 @@
     int ni_frames;
     int avg_frame_qindex;
 
-    int zbin_over_quant;
-    int zbin_mode_boost;
-    int zbin_mode_boost_enabled;
-    int last_zbin_over_quant;
-    int last_zbin_mode_boost;
-
     int64_t total_byte_count;
 
     int buffered_mode;
@@ -477,7 +466,6 @@
     int Speed;
     int compressor_speed;
 
-    int interquantizer;
     int auto_gold;
     int auto_adjust_gold_quantizer;
     int auto_worst_q;
@@ -493,24 +481,16 @@
     int last_skip_probs_q[3];
     int recent_ref_frame_usage[MAX_REF_FRAMES];
 
-    int count_mb_ref_frame_usage[MAX_REF_FRAMES];
     int this_frame_percent_intra;
     int last_frame_percent_intra;
 
     int ref_frame_flags;
 
     SPEED_FEATURES sf;
-    int error_bins[1024];
 
-    /* Data used for real time conferencing mode to help determine if it
-     * would be good to update the gf
-     */
-    int inter_zz_count;
     /* Count ZEROMV on all reference frames. */
     int zeromv_count;
     int lf_zeromv_pct;
-    int gf_bad_count;
-    int gf_update_recommended;
 
     unsigned char *segmentation_map;
     signed char segment_feature_data[MB_LVL_MAX][MAX_MB_SEGMENTS];
@@ -529,6 +509,10 @@
     int cyclic_refresh_q;
     signed char *cyclic_refresh_map;
 
+    // Frame counter for the temporal pattern. Counter is rest when the temporal
+    // layers are changed dynamically (run-time change).
+    unsigned int temporal_pattern_counter;
+
 #if CONFIG_MULTITHREAD
     /* multithread data */
     int * mt_current_mb_col;
@@ -606,7 +590,7 @@
         /* Error score of frames still to be coded in kf group */
         int64_t kf_group_error_left;
         /* Projected Bits available for a group including 1 GF or ARF */
-        int gf_group_bits;
+        int64_t gf_group_bits;
         /* Bits for the golden frame or ARF */
         int gf_bits;
         int alt_extra_bits;
@@ -712,11 +696,8 @@
     } rd_costs;
 } VP8_COMP;
 
-void control_data_rate(VP8_COMP *cpi);
-
-void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned char *dest_end, unsigned long *size);
-
-int rd_cost_intra_mb(MACROBLOCKD *x);
+void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest,
+                        unsigned char *dest_end, unsigned long *size);
 
 void vp8_tokenize_mb(VP8_COMP *, MACROBLOCK *, TOKENEXTRA **);
 
diff --git a/libvpx/vp8/encoder/pickinter.c b/libvpx/vp8/encoder/pickinter.c
index 3f09a9f..c5279fe 100644
--- a/libvpx/vp8/encoder/pickinter.c
+++ b/libvpx/vp8/encoder/pickinter.c
@@ -389,7 +389,7 @@
 
 }
 
-static void update_mvcount(VP8_COMP *cpi, MACROBLOCK *x, int_mv *best_ref_mv)
+static void update_mvcount(MACROBLOCK *x, int_mv *best_ref_mv)
 {
     MACROBLOCKD *xd = &x->e_mbd;
     /* Split MV modes currently not supported when RD is nopt enabled,
@@ -594,6 +594,7 @@
     unsigned int zero_mv_sse = INT_MAX, best_sse = INT_MAX;
 #endif
 
+    int sf_improved_mv_pred = cpi->sf.improved_mv_pred;
     int_mv mvp;
 
     int near_sadidx[8] = {0, 1, 2, 3, 4, 5, 6, 7};
@@ -680,7 +681,7 @@
     get_predictor_pointers(cpi, plane, recon_yoffset, recon_uvoffset);
 
     /* Count of the number of MBs tested so far this frame */
-    cpi->mbs_tested_so_far++;
+    x->mbs_tested_so_far++;
 
     *returnintra = INT_MAX;
     x->skip = 0;
@@ -701,7 +702,7 @@
         int this_rd = INT_MAX;
         int this_ref_frame = ref_frame_map[vp8_ref_frame_order[mode_index]];
 
-        if (best_rd <= cpi->rd_threshes[mode_index])
+        if (best_rd <= x->rd_threshes[mode_index])
             continue;
 
         if (this_ref_frame < 0)
@@ -746,22 +747,22 @@
         /* Check to see if the testing frequency for this mode is at its max
          * If so then prevent it from being tested and increase the threshold
          * for its testing */
-        if (cpi->mode_test_hit_counts[mode_index] &&
+        if (x->mode_test_hit_counts[mode_index] &&
                                          (cpi->mode_check_freq[mode_index] > 1))
         {
-            if (cpi->mbs_tested_so_far <= (cpi->mode_check_freq[mode_index] *
-                                         cpi->mode_test_hit_counts[mode_index]))
+            if (x->mbs_tested_so_far <= (cpi->mode_check_freq[mode_index] *
+                                         x->mode_test_hit_counts[mode_index]))
             {
                 /* Increase the threshold for coding this mode to make it less
                  * likely to be chosen */
-                cpi->rd_thresh_mult[mode_index] += 4;
+                x->rd_thresh_mult[mode_index] += 4;
 
-                if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
-                    cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
+                if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
+                    x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
 
-                cpi->rd_threshes[mode_index] =
+                x->rd_threshes[mode_index] =
                                  (cpi->rd_baseline_thresh[mode_index] >> 7) *
-                                 cpi->rd_thresh_mult[mode_index];
+                                 x->rd_thresh_mult[mode_index];
                 continue;
             }
         }
@@ -769,7 +770,7 @@
         /* We have now reached the point where we are going to test the current
          * mode so increment the counter for the number of times it has been
          * tested */
-        cpi->mode_test_hit_counts[mode_index] ++;
+        x->mode_test_hit_counts[mode_index] ++;
 
         rate2 = 0;
         distortion2 = 0;
@@ -882,7 +883,7 @@
                last frame motion info is not stored, then we can not
                use improved_mv_pred. */
             if (cpi->oxcf.mr_encoder_id && !parent_ref_valid)
-                cpi->sf.improved_mv_pred = 0;
+                sf_improved_mv_pred = 0;
 
             if (parent_ref_valid && parent_ref_frame)
             {
@@ -899,7 +900,7 @@
             }else
 #endif
             {
-                if(cpi->sf.improved_mv_pred)
+                if(sf_improved_mv_pred)
                 {
                     if(!saddone)
                     {
@@ -1109,12 +1110,12 @@
             /* Testing this mode gave rise to an improvement in best error
              * score. Lower threshold a bit for next time
              */
-            cpi->rd_thresh_mult[mode_index] =
-                     (cpi->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
-                     cpi->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
-            cpi->rd_threshes[mode_index] =
+            x->rd_thresh_mult[mode_index] =
+                     (x->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
+                     x->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
+            x->rd_threshes[mode_index] =
                                    (cpi->rd_baseline_thresh[mode_index] >> 7) *
-                                   cpi->rd_thresh_mult[mode_index];
+                                   x->rd_thresh_mult[mode_index];
         }
 
         /* If the mode did not help improve the best error case then raise the
@@ -1122,14 +1123,14 @@
          */
         else
         {
-            cpi->rd_thresh_mult[mode_index] += 4;
+            x->rd_thresh_mult[mode_index] += 4;
 
-            if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
-                cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
+            if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
+                x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
 
-            cpi->rd_threshes[mode_index] =
+            x->rd_threshes[mode_index] =
                          (cpi->rd_baseline_thresh[mode_index] >> 7) *
-                         cpi->rd_thresh_mult[mode_index];
+                         x->rd_thresh_mult[mode_index];
         }
 
         if (x->skip)
@@ -1139,16 +1140,16 @@
     /* Reduce the activation RD thresholds for the best choice mode */
     if ((cpi->rd_baseline_thresh[best_mode_index] > 0) && (cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2)))
     {
-        int best_adjustment = (cpi->rd_thresh_mult[best_mode_index] >> 3);
+        int best_adjustment = (x->rd_thresh_mult[best_mode_index] >> 3);
 
-        cpi->rd_thresh_mult[best_mode_index] =
-                        (cpi->rd_thresh_mult[best_mode_index]
+        x->rd_thresh_mult[best_mode_index] =
+                        (x->rd_thresh_mult[best_mode_index]
                         >= (MIN_THRESHMULT + best_adjustment)) ?
-                        cpi->rd_thresh_mult[best_mode_index] - best_adjustment :
+                        x->rd_thresh_mult[best_mode_index] - best_adjustment :
                         MIN_THRESHMULT;
-        cpi->rd_threshes[best_mode_index] =
+        x->rd_threshes[best_mode_index] =
                         (cpi->rd_baseline_thresh[best_mode_index] >> 7) *
-                        cpi->rd_thresh_mult[best_mode_index];
+                        x->rd_thresh_mult[best_mode_index];
     }
 
 
@@ -1160,7 +1161,7 @@
             this_rdbin = 1023;
         }
 
-        cpi->error_bins[this_rdbin] ++;
+        x->error_bins[this_rdbin] ++;
     }
 
 #if CONFIG_TEMPORAL_DENOISING
@@ -1241,7 +1242,7 @@
       != cpi->common.ref_frame_sign_bias[xd->mode_info_context->mbmi.ref_frame])
         best_ref_mv.as_int = best_ref_mv_sb[!sign_bias].as_int;
 
-    update_mvcount(cpi, x, &best_ref_mv);
+    update_mvcount(x, &best_ref_mv);
 }
 
 
diff --git a/libvpx/vp8/encoder/picklpf.c b/libvpx/vp8/encoder/picklpf.c
index 4121349..841e1e4 100644
--- a/libvpx/vp8/encoder/picklpf.c
+++ b/libvpx/vp8/encoder/picklpf.c
@@ -9,11 +9,12 @@
  */
 
 
+#include "./vpx_scale_rtcd.h"
 #include "vp8/common/onyxc_int.h"
 #include "onyx_int.h"
 #include "quantize.h"
 #include "vpx_mem/vpx_mem.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vp8/common/alloccommon.h"
 #include "vp8/common/loopfilter.h"
 #if ARCH_ARM
diff --git a/libvpx/vp8/encoder/psnr.c b/libvpx/vp8/encoder/psnr.c
index 5bb49ad..b3a3d95 100644
--- a/libvpx/vp8/encoder/psnr.c
+++ b/libvpx/vp8/encoder/psnr.c
@@ -13,7 +13,7 @@
 #include "math.h"
 #include "vp8/common/systemdependent.h" /* for vp8_clear_system_state() */
 
-#define MAX_PSNR 60
+#define MAX_PSNR 100
 
 double vp8_mse2psnr(double Samples, double Peak, double Mse)
 {
diff --git a/libvpx/vp8/encoder/quantize.c b/libvpx/vp8/encoder/quantize.c
index 88fea11..fda997f 100644
--- a/libvpx/vp8/encoder/quantize.c
+++ b/libvpx/vp8/encoder/quantize.c
@@ -50,8 +50,8 @@
         if (x >= zbin)
         {
             x += round_ptr[rc];
-            y  = (((x * quant_ptr[rc]) >> 16) + x)
-                 >> quant_shift_ptr[rc];             /* quantize (x) */
+            y  = ((((x * quant_ptr[rc]) >> 16) + x)
+                 * quant_shift_ptr[rc]) >> 16;       /* quantize (x) */
             x  = (y ^ sz) - sz;                      /* get the sign back */
             qcoeff_ptr[rc] = x;                      /* write to destination */
             dqcoeff_ptr[rc] = x * dequant_ptr[rc];   /* dequantized value */
@@ -113,7 +113,7 @@
     short *zbin_ptr        = b->zbin;
     short *round_ptr       = b->round;
     short *quant_ptr       = b->quant;
-    unsigned char *quant_shift_ptr = b->quant_shift;
+    short *quant_shift_ptr = b->quant_shift;
     short *qcoeff_ptr      = d->qcoeff;
     short *dqcoeff_ptr     = d->dqcoeff;
     short *dequant_ptr     = d->dequant;
@@ -138,8 +138,8 @@
         if (x >= zbin)
         {
             x += round_ptr[rc];
-            y  = (((x * quant_ptr[rc]) >> 16) + x)
-                 >> quant_shift_ptr[rc];             /* quantize (x) */
+            y  = ((((x * quant_ptr[rc]) >> 16) + x)
+                 * quant_shift_ptr[rc]) >> 16;       /* quantize (x) */
             x  = (y ^ sz) - sz;                      /* get the sign back */
             qcoeff_ptr[rc]  = x;                     /* write to destination */
             dqcoeff_ptr[rc] = x * dequant_ptr[rc];   /* dequantized value */
@@ -167,7 +167,7 @@
     int sz;
     short *coeff_ptr;
     short *quant_ptr;
-    unsigned char *quant_shift_ptr;
+    short *quant_shift_ptr;
     short *qcoeff_ptr;
     short *dqcoeff_ptr;
     short *dequant_ptr;
@@ -184,21 +184,21 @@
     for (i = 0; i < 16; i++)
     {
         int dq;
-        int round;
+        int rounding;
 
         /*TODO: These arrays should be stored in zig-zag order.*/
         rc = vp8_default_zig_zag1d[i];
         z = coeff_ptr[rc];
         dq = dequant_ptr[rc];
-        round = dq >> 1;
+        rounding = dq >> 1;
         /* Sign of z. */
         sz = -(z < 0);
         x = (z + sz) ^ sz;
-        x += round;
+        x += rounding;
         if (x >= dq)
         {
             /* Quantize x. */
-            y  = (((x * quant_ptr[rc]) >> 16) + x) >> quant_shift_ptr[rc];
+            y  = ((((x * quant_ptr[rc]) >> 16) + x) * quant_shift_ptr[rc]) >> 16;
             /* Put the sign back. */
             x = (y + sz) ^ sz;
             /* Save the coefficient and its dequantized value. */
@@ -406,7 +406,7 @@
 #define EXACT_QUANT
 #ifdef EXACT_QUANT
 static void invert_quant(int improved_quant, short *quant,
-                               unsigned char *shift, short d)
+                         short *shift, short d)
 {
     if(improved_quant)
     {
@@ -418,11 +418,15 @@
         t = 1 + (1<<(16+l))/d;
         *quant = (short)(t - (1<<16));
         *shift = l;
+        /* use multiplication and constant shift by 16 */
+        *shift = 1 << (16 - *shift);
     }
     else
     {
         *quant = (1 << 16) / d;
         *shift = 0;
+        /* use multiplication and constant shift by 16 */
+        *shift = 1 << (16 - *shift);
     }
 }
 
@@ -587,20 +591,20 @@
 
 #define ZBIN_EXTRA_Y \
     (( cpi->common.Y1dequant[QIndex][1] *  \
-    ( cpi->zbin_over_quant +  \
-      cpi->zbin_mode_boost +  \
+    ( x->zbin_over_quant +  \
+      x->zbin_mode_boost +  \
       x->act_zbin_adj ) ) >> 7)
 
 #define ZBIN_EXTRA_UV \
     (( cpi->common.UVdequant[QIndex][1] *  \
-    ( cpi->zbin_over_quant +  \
-      cpi->zbin_mode_boost +  \
+    ( x->zbin_over_quant +  \
+      x->zbin_mode_boost +  \
       x->act_zbin_adj ) ) >> 7)
 
 #define ZBIN_EXTRA_Y2 \
     (( cpi->common.Y2dequant[QIndex][1] *  \
-    ( (cpi->zbin_over_quant / 2) +  \
-       cpi->zbin_mode_boost +  \
+    ( (x->zbin_over_quant / 2) +  \
+       x->zbin_mode_boost +  \
        x->act_zbin_adj ) ) >> 7)
 
 void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
@@ -702,15 +706,15 @@
         /* save this macroblock QIndex for vp8_update_zbin_extra() */
         x->q_index = QIndex;
 
-        cpi->last_zbin_over_quant = cpi->zbin_over_quant;
-        cpi->last_zbin_mode_boost = cpi->zbin_mode_boost;
+        x->last_zbin_over_quant = x->zbin_over_quant;
+        x->last_zbin_mode_boost = x->zbin_mode_boost;
         x->last_act_zbin_adj = x->act_zbin_adj;
 
 
 
     }
-    else if(cpi->last_zbin_over_quant != cpi->zbin_over_quant
-            || cpi->last_zbin_mode_boost != cpi->zbin_mode_boost
+    else if(x->last_zbin_over_quant != x->zbin_over_quant
+            || x->last_zbin_mode_boost != x->zbin_mode_boost
             || x->last_act_zbin_adj != x->act_zbin_adj)
     {
         /* Y */
@@ -729,8 +733,8 @@
         zbin_extra = ZBIN_EXTRA_Y2;
         x->block[24].zbin_extra = (short)zbin_extra;
 
-        cpi->last_zbin_over_quant = cpi->zbin_over_quant;
-        cpi->last_zbin_mode_boost = cpi->zbin_mode_boost;
+        x->last_zbin_over_quant = x->zbin_over_quant;
+        x->last_zbin_mode_boost = x->zbin_mode_boost;
         x->last_act_zbin_adj = x->act_zbin_adj;
     }
 }
@@ -764,7 +768,7 @@
 void vp8cx_frame_init_quantizer(VP8_COMP *cpi)
 {
     /* Clear Zbin mode boost for default case */
-    cpi->zbin_mode_boost = 0;
+    cpi->mb.zbin_mode_boost = 0;
 
     /* MB level quantizer setup */
     vp8cx_mb_init_quantizer(cpi, &cpi->mb, 0);
diff --git a/libvpx/vp8/encoder/ratectrl.c b/libvpx/vp8/encoder/ratectrl.c
index 77c1c5a..8e3c01d 100644
--- a/libvpx/vp8/encoder/ratectrl.c
+++ b/libvpx/vp8/encoder/ratectrl.c
@@ -614,7 +614,6 @@
 static void calc_pframe_target_size(VP8_COMP *cpi)
 {
     int min_frame_target;
-    int Adjustment;
     int old_per_frame_bandwidth = cpi->per_frame_bandwidth;
 
     if ( cpi->current_layer > 0)
@@ -658,6 +657,7 @@
         /* 1 pass */
         else
         {
+            int Adjustment;
             /* Make rate adjustment to recover bits spent in key frame
              * Test to see if the key frame inter data rate correction
              * should still be in force
@@ -688,7 +688,7 @@
              */
             if ((cpi->gf_overspend_bits > 0) && (cpi->this_frame_target > min_frame_target))
             {
-                int Adjustment = (cpi->non_gf_bitrate_adjustment <= cpi->gf_overspend_bits) ? cpi->non_gf_bitrate_adjustment : cpi->gf_overspend_bits;
+                Adjustment = (cpi->non_gf_bitrate_adjustment <= cpi->gf_overspend_bits) ? cpi->non_gf_bitrate_adjustment : cpi->gf_overspend_bits;
 
                 if (Adjustment > (cpi->this_frame_target - min_frame_target))
                     Adjustment = (cpi->this_frame_target - min_frame_target);
@@ -1109,7 +1109,9 @@
     }
     else
     {
-        if (cpi->common.refresh_alt_ref_frame || cpi->common.refresh_golden_frame)
+        if (cpi->oxcf.number_of_layers == 1 &&
+           (cpi->common.refresh_alt_ref_frame ||
+            cpi->common.refresh_golden_frame))
             rate_correction_factor = cpi->gf_rate_correction_factor;
         else
             rate_correction_factor = cpi->rate_correction_factor;
@@ -1122,9 +1124,9 @@
     projected_size_based_on_q = (int)(((.5 + rate_correction_factor * vp8_bits_per_mb[cpi->common.frame_type][Q]) * cpi->common.MBs) / (1 << BPER_MB_NORMBITS));
 
     /* Make some allowance for cpi->zbin_over_quant */
-    if (cpi->zbin_over_quant > 0)
+    if (cpi->mb.zbin_over_quant > 0)
     {
-        int Z = cpi->zbin_over_quant;
+        int Z = cpi->mb.zbin_over_quant;
         double Factor = 0.99;
         double factor_adjustment = 0.01 / 256.0;
 
@@ -1186,7 +1188,9 @@
         cpi->key_frame_rate_correction_factor = rate_correction_factor;
     else
     {
-        if (cpi->common.refresh_alt_ref_frame || cpi->common.refresh_golden_frame)
+        if (cpi->oxcf.number_of_layers == 1 &&
+           (cpi->common.refresh_alt_ref_frame ||
+            cpi->common.refresh_golden_frame))
             cpi->gf_rate_correction_factor = rate_correction_factor;
         else
             cpi->rate_correction_factor = rate_correction_factor;
@@ -1199,7 +1203,7 @@
     int Q = cpi->active_worst_quality;
 
     /* Reset Zbin OQ value */
-    cpi->zbin_over_quant = 0;
+    cpi->mb.zbin_over_quant = 0;
 
     if (cpi->oxcf.fixed_q >= 0)
     {
@@ -1209,11 +1213,13 @@
         {
             Q = cpi->oxcf.key_q;
         }
-        else if (cpi->common.refresh_alt_ref_frame)
+        else if (cpi->oxcf.number_of_layers == 1 &&
+            cpi->common.refresh_alt_ref_frame)
         {
             Q = cpi->oxcf.alt_q;
         }
-        else if (cpi->common.refresh_golden_frame)
+        else if (cpi->oxcf.number_of_layers == 1  &&
+            cpi->common.refresh_golden_frame)
         {
             Q = cpi->oxcf.gold_q;
         }
@@ -1232,7 +1238,9 @@
             correction_factor = cpi->key_frame_rate_correction_factor;
         else
         {
-            if (cpi->common.refresh_alt_ref_frame || cpi->common.refresh_golden_frame)
+            if (cpi->oxcf.number_of_layers == 1 &&
+               (cpi->common.refresh_alt_ref_frame ||
+                cpi->common.refresh_golden_frame))
                 correction_factor = cpi->gf_rate_correction_factor;
             else
                 correction_factor = cpi->rate_correction_factor;
@@ -1281,7 +1289,10 @@
 
             if (cpi->common.frame_type == KEY_FRAME)
                 zbin_oqmax = 0;
-            else if (cpi->common.refresh_alt_ref_frame || (cpi->common.refresh_golden_frame && !cpi->source_alt_ref_active))
+            else if (cpi->oxcf.number_of_layers == 1 &&
+                (cpi->common.refresh_alt_ref_frame ||
+                (cpi->common.refresh_golden_frame &&
+                 !cpi->source_alt_ref_active)))
                 zbin_oqmax = 16;
             else
                 zbin_oqmax = ZBIN_OQ_MAX;
@@ -1307,12 +1318,12 @@
              * normal maximum by expanding the zero bin and hence
              * decreasing the number of low magnitude non zero coefficients.
              */
-            while (cpi->zbin_over_quant < zbin_oqmax)
+            while (cpi->mb.zbin_over_quant < zbin_oqmax)
             {
-                cpi->zbin_over_quant ++;
+                cpi->mb.zbin_over_quant ++;
 
-                if (cpi->zbin_over_quant > zbin_oqmax)
-                    cpi->zbin_over_quant = zbin_oqmax;
+                if (cpi->mb.zbin_over_quant > zbin_oqmax)
+                    cpi->mb.zbin_over_quant = zbin_oqmax;
 
                 /* Adjust bits_per_mb_at_this_q estimate */
                 bits_per_mb_at_this_q = (int)(Factor * bits_per_mb_at_this_q);
@@ -1349,10 +1360,10 @@
          * whichever is smaller.
          */
         int key_freq = cpi->oxcf.key_freq>0 ? cpi->oxcf.key_freq : 1;
-        av_key_frame_frequency = (int)cpi->output_frame_rate * 2;
+        av_key_frame_frequency = 1 + (int)cpi->output_frame_rate * 2;
 
         if (cpi->oxcf.auto_key && av_key_frame_frequency > key_freq)
-            av_key_frame_frequency = cpi->oxcf.key_freq;
+            av_key_frame_frequency = key_freq;
 
         cpi->prior_key_frame_distance[KEY_FRAME_CONTEXT - 1]
             = av_key_frame_frequency;
@@ -1382,6 +1393,10 @@
         av_key_frame_frequency  /= total_weight;
 
     }
+    // TODO (marpan): Given the checks above, |av_key_frame_frequency|
+    // should always be above 0. But for now we keep the sanity check in.
+    if (av_key_frame_frequency == 0)
+        av_key_frame_frequency = 1;
     return av_key_frame_frequency;
 }
 
diff --git a/libvpx/vp8/encoder/rdopt.c b/libvpx/vp8/encoder/rdopt.c
index 7d80606..8579614 100644
--- a/libvpx/vp8/encoder/rdopt.c
+++ b/libvpx/vp8/encoder/rdopt.c
@@ -14,7 +14,7 @@
 #include <limits.h>
 #include <assert.h>
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vp8/common/pragmas.h"
 #include "tokenize.h"
 #include "treewriter.h"
@@ -223,7 +223,7 @@
     cpi->mb.sadperbit4  =  sad_per_bit4lut[QIndex];
 }
 
-void vp8_initialize_rd_consts(VP8_COMP *cpi, int Qvalue)
+void vp8_initialize_rd_consts(VP8_COMP *cpi, MACROBLOCK *x, int Qvalue)
 {
     int q;
     int i;
@@ -238,15 +238,15 @@
     cpi->RDMULT = (int)(rdconst * (capped_q * capped_q));
 
     /* Extend rate multiplier along side quantizer zbin increases */
-    if (cpi->zbin_over_quant  > 0)
+    if (cpi->mb.zbin_over_quant  > 0)
     {
         double oq_factor;
         double modq;
 
         /* Experimental code using the same basic equation as used for Q above
-         * The units of cpi->zbin_over_quant are 1/128 of Q bin size
+         * The units of cpi->mb.zbin_over_quant are 1/128 of Q bin size
          */
-        oq_factor = 1.0 + ((double)0.0015625 * cpi->zbin_over_quant);
+        oq_factor = 1.0 + ((double)0.0015625 * cpi->mb.zbin_over_quant);
         modq = (int)((double)capped_q * oq_factor);
         cpi->RDMULT = (int)(rdconst * (modq * modq));
     }
@@ -265,6 +265,11 @@
 
     vp8_set_speed_features(cpi);
 
+    for (i = 0; i < MAX_MODES; i++)
+    {
+        x->mode_test_hit_counts[i] = 0;
+    }
+
     q = (int)pow(Qvalue, 1.25);
 
     if (q < 8)
@@ -279,14 +284,14 @@
         {
             if (cpi->sf.thresh_mult[i] < INT_MAX)
             {
-                cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q / 100;
+                x->rd_threshes[i] = cpi->sf.thresh_mult[i] * q / 100;
             }
             else
             {
-                cpi->rd_threshes[i] = INT_MAX;
+                x->rd_threshes[i] = INT_MAX;
             }
 
-            cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
+            cpi->rd_baseline_thresh[i] = x->rd_threshes[i];
         }
     }
     else
@@ -297,14 +302,14 @@
         {
             if (cpi->sf.thresh_mult[i] < (INT_MAX / q))
             {
-                cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q;
+                x->rd_threshes[i] = cpi->sf.thresh_mult[i] * q;
             }
             else
             {
-                cpi->rd_threshes[i] = INT_MAX;
+                x->rd_threshes[i] = INT_MAX;
             }
 
-            cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
+            cpi->rd_baseline_thresh[i] = x->rd_threshes[i];
         }
     }
 
@@ -879,8 +884,8 @@
 
     for (mode = DC_PRED; mode <= TM_PRED; mode++)
     {
-        int rate;
-        int distortion;
+        int this_rate;
+        int this_distortion;
         int this_rd;
 
         xd->mode_info_context->mbmi.uv_mode = mode;
@@ -902,17 +907,17 @@
         vp8_quantize_mbuv(x);
 
         rate_to = rd_cost_mbuv(x);
-        rate = rate_to + x->intra_uv_mode_cost[xd->frame_type][xd->mode_info_context->mbmi.uv_mode];
+        this_rate = rate_to + x->intra_uv_mode_cost[xd->frame_type][xd->mode_info_context->mbmi.uv_mode];
 
-        distortion = vp8_mbuverror(x) / 4;
+        this_distortion = vp8_mbuverror(x) / 4;
 
-        this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
+        this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
 
         if (this_rd < best_rd)
         {
             best_rd = this_rd;
-            d = distortion;
-            r = rate;
+            d = this_distortion;
+            r = this_rate;
             *rate_tokenonly = rate_to;
             mode_selected = mode;
         }
@@ -1289,12 +1294,11 @@
 
                 if (bestsme < INT_MAX)
                 {
-                    int distortion;
+                    int disto;
                     unsigned int sse;
                     cpi->find_fractional_mv_step(x, c, e, &mode_mv[NEW4X4],
                         bsi->ref_mv, x->errorperbit, v_fn_ptr, x->mvcost,
-                        &distortion, &sse);
-
+                        &disto, &sse);
                 }
             } /* NEW4X4 */
 
@@ -1728,7 +1732,7 @@
     }
 }
 
-static void rd_update_mvcount(VP8_COMP *cpi, MACROBLOCK *x, int_mv *best_ref_mv)
+static void rd_update_mvcount(MACROBLOCK *x, int_mv *best_ref_mv)
 {
     if (x->e_mbd.mode_info_context->mbmi.mode == SPLITMV)
     {
@@ -2010,7 +2014,7 @@
 
     *returnintra = INT_MAX;
     /* Count of the number of MBs tested so far this frame */
-    cpi->mbs_tested_so_far++;
+    x->mbs_tested_so_far++;
 
     x->skip = 0;
 
@@ -2022,7 +2026,7 @@
         int this_ref_frame = ref_frame_map[vp8_ref_frame_order[mode_index]];
 
         /* Test best rd so far against threshold for trying this mode. */
-        if (best_mode.rd <= cpi->rd_threshes[mode_index])
+        if (best_mode.rd <= x->rd_threshes[mode_index])
             continue;
 
         if (this_ref_frame < 0)
@@ -2068,19 +2072,21 @@
          * max If so then prevent it from being tested and increase the
          * threshold for its testing
          */
-        if (cpi->mode_test_hit_counts[mode_index] && (cpi->mode_check_freq[mode_index] > 1))
+        if (x->mode_test_hit_counts[mode_index] && (cpi->mode_check_freq[mode_index] > 1))
         {
-            if (cpi->mbs_tested_so_far  <= cpi->mode_check_freq[mode_index] * cpi->mode_test_hit_counts[mode_index])
+            if (x->mbs_tested_so_far  <= cpi->mode_check_freq[mode_index] * x->mode_test_hit_counts[mode_index])
             {
                 /* Increase the threshold for coding this mode to make it
                  * less likely to be chosen
                  */
-                cpi->rd_thresh_mult[mode_index] += 4;
+                x->rd_thresh_mult[mode_index] += 4;
 
-                if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
-                    cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
+                if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
+                    x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
 
-                cpi->rd_threshes[mode_index] = (cpi->rd_baseline_thresh[mode_index] >> 7) * cpi->rd_thresh_mult[mode_index];
+                x->rd_threshes[mode_index] =
+                    (cpi->rd_baseline_thresh[mode_index] >> 7) *
+                    x->rd_thresh_mult[mode_index];
 
                 continue;
             }
@@ -2090,28 +2096,28 @@
          * current mode so increment the counter for the number of times
          * it has been tested
          */
-        cpi->mode_test_hit_counts[mode_index] ++;
+        x->mode_test_hit_counts[mode_index] ++;
 
         /* Experimental code. Special case for gf and arf zeromv modes.
          * Increase zbin size to supress noise
          */
-        if (cpi->zbin_mode_boost_enabled)
+        if (x->zbin_mode_boost_enabled)
         {
             if ( this_ref_frame == INTRA_FRAME )
-                cpi->zbin_mode_boost = 0;
+                x->zbin_mode_boost = 0;
             else
             {
                 if (vp8_mode_order[mode_index] == ZEROMV)
                 {
                     if (this_ref_frame != LAST_FRAME)
-                        cpi->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
+                        x->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
                     else
-                        cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
+                        x->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
                 }
                 else if (vp8_mode_order[mode_index] == SPLITMV)
-                    cpi->zbin_mode_boost = 0;
+                    x->zbin_mode_boost = 0;
                 else
-                    cpi->zbin_mode_boost = MV_ZBIN_BOOST;
+                    x->zbin_mode_boost = MV_ZBIN_BOOST;
             }
 
             vp8_update_zbin_extra(cpi, x);
@@ -2170,8 +2176,10 @@
             int this_rd_thresh;
             int distortion;
 
-            this_rd_thresh = (vp8_ref_frame_order[mode_index] == 1) ? cpi->rd_threshes[THR_NEW1] : cpi->rd_threshes[THR_NEW3];
-            this_rd_thresh = (vp8_ref_frame_order[mode_index] == 2) ? cpi->rd_threshes[THR_NEW2] : this_rd_thresh;
+            this_rd_thresh = (vp8_ref_frame_order[mode_index] == 1) ?
+                x->rd_threshes[THR_NEW1] : x->rd_threshes[THR_NEW3];
+            this_rd_thresh = (vp8_ref_frame_order[mode_index] == 2) ?
+                x->rd_threshes[THR_NEW2] : this_rd_thresh;
 
             tmp_rd = vp8_rd_pick_best_mbsegmentation(cpi, x, &best_ref_mv,
                                                      best_mode.yrd, mdcounts,
@@ -2464,8 +2472,9 @@
             /* Testing this mode gave rise to an improvement in best error
              * score. Lower threshold a bit for next time
              */
-            cpi->rd_thresh_mult[mode_index] = (cpi->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ? cpi->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
-            cpi->rd_threshes[mode_index] = (cpi->rd_baseline_thresh[mode_index] >> 7) * cpi->rd_thresh_mult[mode_index];
+            x->rd_thresh_mult[mode_index] =
+                (x->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
+                    x->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
         }
 
         /* If the mode did not help improve the best error case then raise
@@ -2473,13 +2482,14 @@
          */
         else
         {
-            cpi->rd_thresh_mult[mode_index] += 4;
+            x->rd_thresh_mult[mode_index] += 4;
 
-            if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
-                cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
-
-            cpi->rd_threshes[mode_index] = (cpi->rd_baseline_thresh[mode_index] >> 7) * cpi->rd_thresh_mult[mode_index];
+            if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
+                x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
         }
+        x->rd_threshes[mode_index] =
+            (cpi->rd_baseline_thresh[mode_index] >> 7) *
+                x->rd_thresh_mult[mode_index];
 
         if (x->skip)
             break;
@@ -2489,15 +2499,18 @@
     /* Reduce the activation RD thresholds for the best choice mode */
     if ((cpi->rd_baseline_thresh[best_mode_index] > 0) && (cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2)))
     {
-        int best_adjustment = (cpi->rd_thresh_mult[best_mode_index] >> 2);
+        int best_adjustment = (x->rd_thresh_mult[best_mode_index] >> 2);
 
-        cpi->rd_thresh_mult[best_mode_index] = (cpi->rd_thresh_mult[best_mode_index] >= (MIN_THRESHMULT + best_adjustment)) ? cpi->rd_thresh_mult[best_mode_index] - best_adjustment : MIN_THRESHMULT;
-        cpi->rd_threshes[best_mode_index] = (cpi->rd_baseline_thresh[best_mode_index] >> 7) * cpi->rd_thresh_mult[best_mode_index];
+        x->rd_thresh_mult[best_mode_index] =
+            (x->rd_thresh_mult[best_mode_index] >=
+                (MIN_THRESHMULT + best_adjustment)) ?
+                    x->rd_thresh_mult[best_mode_index] - best_adjustment :
+                    MIN_THRESHMULT;
+        x->rd_threshes[best_mode_index] =
+            (cpi->rd_baseline_thresh[best_mode_index] >> 7) *
+                x->rd_thresh_mult[best_mode_index];
     }
 
-    /* Note how often each mode chosen as best */
-    cpi->mode_chosen_counts[best_mode_index] ++;
-
 #if CONFIG_TEMPORAL_DENOISING
     if (cpi->oxcf.noise_sensitivity)
     {
@@ -2591,7 +2604,7 @@
         != cpi->common.ref_frame_sign_bias[xd->mode_info_context->mbmi.ref_frame])
         best_ref_mv.as_int = best_ref_mv_sb[!sign_bias].as_int;
 
-    rd_update_mvcount(cpi, x, &best_ref_mv);
+    rd_update_mvcount(x, &best_ref_mv);
 }
 
 void vp8_rd_pick_intra_mode(MACROBLOCK *x, int *rate_)
diff --git a/libvpx/vp8/encoder/rdopt.h b/libvpx/vp8/encoder/rdopt.h
index d7b0442..1e11fa7 100644
--- a/libvpx/vp8/encoder/rdopt.h
+++ b/libvpx/vp8/encoder/rdopt.h
@@ -65,7 +65,7 @@
     }
 }
 
-extern void vp8_initialize_rd_consts(VP8_COMP *cpi, int Qvalue);
+extern void vp8_initialize_rd_consts(VP8_COMP *cpi, MACROBLOCK *x, int Qvalue);
 extern void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int recon_uvoffset, int *returnrate, int *returndistortion, int *returnintra);
 extern void vp8_rd_pick_intra_mode(MACROBLOCK *x, int *rate);
 
diff --git a/libvpx/vp8/encoder/temporal_filter.c b/libvpx/vp8/encoder/temporal_filter.c
index b83ae89..7e3af71 100644
--- a/libvpx/vp8/encoder/temporal_filter.c
+++ b/libvpx/vp8/encoder/temporal_filter.c
@@ -17,7 +17,7 @@
 #include "mcomp.h"
 #include "firstpass.h"
 #include "psnr.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vp8/common/extend.h"
 #include "ratectrl.h"
 #include "vp8/common/quant_common.h"
diff --git a/libvpx/vp8/encoder/tokenize.c b/libvpx/vp8/encoder/tokenize.c
index 3b5268b..11559a7 100644
--- a/libvpx/vp8/encoder/tokenize.c
+++ b/libvpx/vp8/encoder/tokenize.c
@@ -20,7 +20,7 @@
 /* Global event counters used for accumulating statistics across several
    compressions, then generating context.c = initial stats. */
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 _int64 context_counters[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];
 #endif
 void vp8_stuff_mb(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t) ;
@@ -413,7 +413,7 @@
 }
 
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 
 void init_context_counters(void)
 {
diff --git a/libvpx/vp8/encoder/tokenize.h b/libvpx/vp8/encoder/tokenize.h
index c2d1438..1e6cea1 100644
--- a/libvpx/vp8/encoder/tokenize.h
+++ b/libvpx/vp8/encoder/tokenize.h
@@ -33,7 +33,7 @@
 
 int rd_cost_mby(MACROBLOCKD *);
 
-#ifdef ENTROPY_STATS
+#ifdef VP8_ENTROPY_STATS
 void init_context_counters();
 void print_context_counters();
 
diff --git a/libvpx/vp8/encoder/asm_enc_offsets.c b/libvpx/vp8/encoder/vp8_asm_enc_offsets.c
similarity index 100%
rename from libvpx/vp8/encoder/asm_enc_offsets.c
rename to libvpx/vp8/encoder/vp8_asm_enc_offsets.c
diff --git a/libvpx/vp8/encoder/x86/dct_sse2.asm b/libvpx/vp8/encoder/x86/dct_sse2.asm
index d880ce0..d06bca5 100644
--- a/libvpx/vp8/encoder/x86/dct_sse2.asm
+++ b/libvpx/vp8/encoder/x86/dct_sse2.asm
@@ -29,7 +29,7 @@
     movsxd      rax, dword ptr arg(2)
     lea         rcx, [rsi + rax*2]
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     %define     input       rcx
     %define     output      rdx
     %define     pitch       r8
@@ -53,7 +53,7 @@
     RESTORE_GOT
     pop         rbp
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     RESTORE_XMM
   %endif
 %endif
diff --git a/libvpx/vp8/encoder/x86/denoising_sse2.c b/libvpx/vp8/encoder/x86/denoising_sse2.c
index c1ac6c1..cceb826 100644
--- a/libvpx/vp8/encoder/x86/denoising_sse2.c
+++ b/libvpx/vp8/encoder/x86/denoising_sse2.c
@@ -12,9 +12,10 @@
 #include "vp8/common/reconinter.h"
 #include "vpx/vpx_integer.h"
 #include "vpx_mem/vpx_mem.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 
 #include <emmintrin.h>
+#include "vpx_ports/emmintrin_compat.h"
 
 union sum_union {
     __m128i v;
diff --git a/libvpx/vp8/encoder/x86/quantize_sse2.asm b/libvpx/vp8/encoder/x86/quantize_sse2.asm
deleted file mode 100644
index 724e54c..0000000
--- a/libvpx/vp8/encoder/x86/quantize_sse2.asm
+++ /dev/null
@@ -1,386 +0,0 @@
-;
-;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-;  Use of this source code is governed by a BSD-style license and patent
-;  grant that can be found in the LICENSE file in the root of the source
-;  tree. All contributing project authors may be found in the AUTHORS
-;  file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-%include "asm_enc_offsets.asm"
-
-
-; void vp8_regular_quantize_b_sse2 | arg
-;  (BLOCK  *b,                     |  0
-;   BLOCKD *d)                     |  1
-
-global sym(vp8_regular_quantize_b_sse2) PRIVATE
-sym(vp8_regular_quantize_b_sse2):
-    push        rbp
-    mov         rbp, rsp
-    SAVE_XMM 7
-    GET_GOT     rbx
-
-%if ABI_IS_32BIT
-    push        rdi
-    push        rsi
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    push        rdi
-    push        rsi
-  %endif
-%endif
-
-    ALIGN_STACK 16, rax
-    %define zrun_zbin_boost   0  ;  8
-    %define abs_minus_zbin    8  ; 32
-    %define temp_qcoeff       40 ; 32
-    %define qcoeff            72 ; 32
-    %define stack_size        104
-    sub         rsp, stack_size
-    ; end prolog
-
-%if ABI_IS_32BIT
-    mov         rdi, arg(0)                 ; BLOCK *b
-    mov         rsi, arg(1)                 ; BLOCKD *d
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    mov         rdi, rcx                    ; BLOCK *b
-    mov         rsi, rdx                    ; BLOCKD *d
-  %else
-    ;mov         rdi, rdi                    ; BLOCK *b
-    ;mov         rsi, rsi                    ; BLOCKD *d
-  %endif
-%endif
-
-    mov         rdx, [rdi + vp8_block_coeff] ; coeff_ptr
-    mov         rcx, [rdi + vp8_block_zbin] ; zbin_ptr
-    movd        xmm7, [rdi + vp8_block_zbin_extra] ; zbin_oq_value
-
-    ; z
-    movdqa      xmm0, [rdx]
-    movdqa      xmm4, [rdx + 16]
-    mov         rdx, [rdi + vp8_block_round] ; round_ptr
-
-    pshuflw     xmm7, xmm7, 0
-    punpcklwd   xmm7, xmm7                  ; duplicated zbin_oq_value
-
-    movdqa      xmm1, xmm0
-    movdqa      xmm5, xmm4
-
-    ; sz
-    psraw       xmm0, 15
-    psraw       xmm4, 15
-
-    ; (z ^ sz)
-    pxor        xmm1, xmm0
-    pxor        xmm5, xmm4
-
-    ; x = abs(z)
-    psubw       xmm1, xmm0
-    psubw       xmm5, xmm4
-
-    movdqa      xmm2, [rcx]
-    movdqa      xmm3, [rcx + 16]
-    mov         rcx, [rdi + vp8_block_quant] ; quant_ptr
-
-    ; *zbin_ptr + zbin_oq_value
-    paddw       xmm2, xmm7
-    paddw       xmm3, xmm7
-
-    ; x - (*zbin_ptr + zbin_oq_value)
-    psubw       xmm1, xmm2
-    psubw       xmm5, xmm3
-    movdqa      [rsp + abs_minus_zbin], xmm1
-    movdqa      [rsp + abs_minus_zbin + 16], xmm5
-
-    ; add (zbin_ptr + zbin_oq_value) back
-    paddw       xmm1, xmm2
-    paddw       xmm5, xmm3
-
-    movdqa      xmm2, [rdx]
-    movdqa      xmm6, [rdx + 16]
-
-    movdqa      xmm3, [rcx]
-    movdqa      xmm7, [rcx + 16]
-
-    ; x + round
-    paddw       xmm1, xmm2
-    paddw       xmm5, xmm6
-
-    ; y = x * quant_ptr >> 16
-    pmulhw      xmm3, xmm1
-    pmulhw      xmm7, xmm5
-
-    ; y += x
-    paddw       xmm1, xmm3
-    paddw       xmm5, xmm7
-
-    movdqa      [rsp + temp_qcoeff], xmm1
-    movdqa      [rsp + temp_qcoeff + 16], xmm5
-
-    pxor        xmm6, xmm6
-    ; zero qcoeff
-    movdqa      [rsp + qcoeff], xmm6
-    movdqa      [rsp + qcoeff + 16], xmm6
-
-    mov         rdx, [rdi + vp8_block_zrun_zbin_boost] ; zbin_boost_ptr
-    mov         rax, [rdi + vp8_block_quant_shift] ; quant_shift_ptr
-    mov         [rsp + zrun_zbin_boost], rdx
-
-%macro ZIGZAG_LOOP 1
-    ; x
-    movsx       ecx, WORD PTR[rsp + abs_minus_zbin + %1 * 2]
-
-    ; if (x >= zbin)
-    sub         cx, WORD PTR[rdx]           ; x - zbin
-    lea         rdx, [rdx + 2]              ; zbin_boost_ptr++
-    jl          .rq_zigzag_loop_%1           ; x < zbin
-
-    movsx       edi, WORD PTR[rsp + temp_qcoeff + %1 * 2]
-
-    ; downshift by quant_shift[rc]
-    movsx       cx, BYTE PTR[rax + %1]      ; quant_shift_ptr[rc]
-    sar         edi, cl                     ; also sets Z bit
-    je          .rq_zigzag_loop_%1           ; !y
-    mov         WORD PTR[rsp + qcoeff + %1 * 2], di ;qcoeff_ptr[rc] = temp_qcoeff[rc]
-    mov         rdx, [rsp + zrun_zbin_boost] ; reset to b->zrun_zbin_boost
-.rq_zigzag_loop_%1:
-%endmacro
-; in vp8_default_zig_zag1d order: see vp8/common/entropy.c
-ZIGZAG_LOOP  0
-ZIGZAG_LOOP  1
-ZIGZAG_LOOP  4
-ZIGZAG_LOOP  8
-ZIGZAG_LOOP  5
-ZIGZAG_LOOP  2
-ZIGZAG_LOOP  3
-ZIGZAG_LOOP  6
-ZIGZAG_LOOP  9
-ZIGZAG_LOOP 12
-ZIGZAG_LOOP 13
-ZIGZAG_LOOP 10
-ZIGZAG_LOOP  7
-ZIGZAG_LOOP 11
-ZIGZAG_LOOP 14
-ZIGZAG_LOOP 15
-
-    movdqa      xmm2, [rsp + qcoeff]
-    movdqa      xmm3, [rsp + qcoeff + 16]
-
-    mov         rcx, [rsi + vp8_blockd_dequant] ; dequant_ptr
-    mov         rdi, [rsi + vp8_blockd_dqcoeff] ; dqcoeff_ptr
-
-    ; y ^ sz
-    pxor        xmm2, xmm0
-    pxor        xmm3, xmm4
-    ; x = (y ^ sz) - sz
-    psubw       xmm2, xmm0
-    psubw       xmm3, xmm4
-
-    ; dequant
-    movdqa      xmm0, [rcx]
-    movdqa      xmm1, [rcx + 16]
-
-    mov         rcx, [rsi + vp8_blockd_qcoeff] ; qcoeff_ptr
-
-    pmullw      xmm0, xmm2
-    pmullw      xmm1, xmm3
-
-    movdqa      [rcx], xmm2        ; store qcoeff
-    movdqa      [rcx + 16], xmm3
-    movdqa      [rdi], xmm0        ; store dqcoeff
-    movdqa      [rdi + 16], xmm1
-
-    mov         rcx, [rsi + vp8_blockd_eob]
-
-    ; select the last value (in zig_zag order) for EOB
-    pcmpeqw     xmm2, xmm6
-    pcmpeqw     xmm3, xmm6
-    ; !
-    pcmpeqw     xmm6, xmm6
-    pxor        xmm2, xmm6
-    pxor        xmm3, xmm6
-    ; mask inv_zig_zag
-    pand        xmm2, [GLOBAL(inv_zig_zag)]
-    pand        xmm3, [GLOBAL(inv_zig_zag + 16)]
-    ; select the max value
-    pmaxsw      xmm2, xmm3
-    pshufd      xmm3, xmm2, 00001110b
-    pmaxsw      xmm2, xmm3
-    pshuflw     xmm3, xmm2, 00001110b
-    pmaxsw      xmm2, xmm3
-    pshuflw     xmm3, xmm2, 00000001b
-    pmaxsw      xmm2, xmm3
-    movd        eax, xmm2
-    and         eax, 0xff
-
-    mov         BYTE PTR [rcx], al          ; store eob
-
-    ; begin epilog
-    add         rsp, stack_size
-    pop         rsp
-%if ABI_IS_32BIT
-    pop         rsi
-    pop         rdi
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    pop         rsi
-    pop         rdi
-  %endif
-%endif
-    RESTORE_GOT
-    RESTORE_XMM
-    pop         rbp
-    ret
-
-; void vp8_fast_quantize_b_sse2 | arg
-;  (BLOCK  *b,                  |  0
-;   BLOCKD *d)                  |  1
-
-global sym(vp8_fast_quantize_b_sse2) PRIVATE
-sym(vp8_fast_quantize_b_sse2):
-    push        rbp
-    mov         rbp, rsp
-    GET_GOT     rbx
-
-%if ABI_IS_32BIT
-    push        rdi
-    push        rsi
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    push        rdi
-    push        rsi
-  %else
-    ; these registers are used for passing arguments
-  %endif
-%endif
-
-    ; end prolog
-
-%if ABI_IS_32BIT
-    mov         rdi, arg(0)                 ; BLOCK *b
-    mov         rsi, arg(1)                 ; BLOCKD *d
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    mov         rdi, rcx                    ; BLOCK *b
-    mov         rsi, rdx                    ; BLOCKD *d
-  %else
-    ;mov         rdi, rdi                    ; BLOCK *b
-    ;mov         rsi, rsi                    ; BLOCKD *d
-  %endif
-%endif
-
-    mov         rax, [rdi + vp8_block_coeff]
-    mov         rcx, [rdi + vp8_block_round]
-    mov         rdx, [rdi + vp8_block_quant_fast]
-
-    ; z = coeff
-    movdqa      xmm0, [rax]
-    movdqa      xmm4, [rax + 16]
-
-    ; dup z so we can save sz
-    movdqa      xmm1, xmm0
-    movdqa      xmm5, xmm4
-
-    ; sz = z >> 15
-    psraw       xmm0, 15
-    psraw       xmm4, 15
-
-    ; x = abs(z) = (z ^ sz) - sz
-    pxor        xmm1, xmm0
-    pxor        xmm5, xmm4
-    psubw       xmm1, xmm0
-    psubw       xmm5, xmm4
-
-    ; x += round
-    paddw       xmm1, [rcx]
-    paddw       xmm5, [rcx + 16]
-
-    mov         rax, [rsi + vp8_blockd_qcoeff]
-    mov         rcx, [rsi + vp8_blockd_dequant]
-    mov         rdi, [rsi + vp8_blockd_dqcoeff]
-
-    ; y = x * quant >> 16
-    pmulhw      xmm1, [rdx]
-    pmulhw      xmm5, [rdx + 16]
-
-    ; x = (y ^ sz) - sz
-    pxor        xmm1, xmm0
-    pxor        xmm5, xmm4
-    psubw       xmm1, xmm0
-    psubw       xmm5, xmm4
-
-    ; qcoeff = x
-    movdqa      [rax], xmm1
-    movdqa      [rax + 16], xmm5
-
-    ; x * dequant
-    movdqa      xmm2, xmm1
-    movdqa      xmm3, xmm5
-    pmullw      xmm2, [rcx]
-    pmullw      xmm3, [rcx + 16]
-
-    ; dqcoeff = x * dequant
-    movdqa      [rdi], xmm2
-    movdqa      [rdi + 16], xmm3
-
-    pxor        xmm4, xmm4                  ;clear all bits
-    pcmpeqw     xmm1, xmm4
-    pcmpeqw     xmm5, xmm4
-
-    pcmpeqw     xmm4, xmm4                  ;set all bits
-    pxor        xmm1, xmm4
-    pxor        xmm5, xmm4
-
-    pand        xmm1, [GLOBAL(inv_zig_zag)]
-    pand        xmm5, [GLOBAL(inv_zig_zag + 16)]
-
-    pmaxsw      xmm1, xmm5
-
-    mov         rcx, [rsi + vp8_blockd_eob]
-
-    ; now down to 8
-    pshufd      xmm5, xmm1, 00001110b
-
-    pmaxsw      xmm1, xmm5
-
-    ; only 4 left
-    pshuflw     xmm5, xmm1, 00001110b
-
-    pmaxsw      xmm1, xmm5
-
-    ; okay, just 2!
-    pshuflw     xmm5, xmm1, 00000001b
-
-    pmaxsw      xmm1, xmm5
-
-    movd        eax, xmm1
-    and         eax, 0xff
-
-    mov         BYTE PTR [rcx], al          ; store eob
-
-    ; begin epilog
-%if ABI_IS_32BIT
-    pop         rsi
-    pop         rdi
-%else
-  %ifidn __OUTPUT_FORMAT__,x64
-    pop         rsi
-    pop         rdi
-  %endif
-%endif
-
-    RESTORE_GOT
-    pop         rbp
-    ret
-
-SECTION_RODATA
-align 16
-inv_zig_zag:
-  dw 0x0001, 0x0002, 0x0006, 0x0007
-  dw 0x0003, 0x0005, 0x0008, 0x000d
-  dw 0x0004, 0x0009, 0x000c, 0x000e
-  dw 0x000a, 0x000b, 0x000f, 0x0010
diff --git a/libvpx/vp8/encoder/x86/quantize_sse2.c b/libvpx/vp8/encoder/x86/quantize_sse2.c
new file mode 100644
index 0000000..f495bf2
--- /dev/null
+++ b/libvpx/vp8/encoder/x86/quantize_sse2.c
@@ -0,0 +1,229 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_config.h"
+#include "vp8_rtcd.h"
+#include "vpx_ports/x86.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp8/encoder/block.h"
+#include "vp8/common/entropy.h" /* vp8_default_inv_zig_zag */
+
+#include <mmintrin.h> /* MMX */
+#include <xmmintrin.h> /* SSE */
+#include <emmintrin.h> /* SSE2 */
+
+#define SELECT_EOB(i, z) \
+    do { \
+        short boost = *zbin_boost_ptr; \
+        int cmp = (x[z] < boost) | (y[z] == 0); \
+        zbin_boost_ptr++; \
+        if (cmp) \
+            goto select_eob_end_##i; \
+        qcoeff_ptr[z] = y[z]; \
+        eob = i; \
+        zbin_boost_ptr = b->zrun_zbin_boost; \
+        select_eob_end_##i:; \
+    } while (0)
+
+void vp8_regular_quantize_b_sse2(BLOCK *b, BLOCKD *d)
+{
+    char eob = 0;
+    short *zbin_boost_ptr  = b->zrun_zbin_boost;
+    short *qcoeff_ptr      = d->qcoeff;
+    DECLARE_ALIGNED_ARRAY(16, short, x, 16);
+    DECLARE_ALIGNED_ARRAY(16, short, y, 16);
+
+    __m128i sz0, x0, sz1, x1, y0, y1, x_minus_zbin0, x_minus_zbin1;
+    __m128i quant_shift0 = _mm_load_si128((__m128i *)(b->quant_shift));
+    __m128i quant_shift1 = _mm_load_si128((__m128i *)(b->quant_shift + 8));
+    __m128i z0 = _mm_load_si128((__m128i *)(b->coeff));
+    __m128i z1 = _mm_load_si128((__m128i *)(b->coeff+8));
+    __m128i zbin_extra = _mm_cvtsi32_si128(b->zbin_extra);
+    __m128i zbin0 = _mm_load_si128((__m128i *)(b->zbin));
+    __m128i zbin1 = _mm_load_si128((__m128i *)(b->zbin + 8));
+    __m128i round0 = _mm_load_si128((__m128i *)(b->round));
+    __m128i round1 = _mm_load_si128((__m128i *)(b->round + 8));
+    __m128i quant0 = _mm_load_si128((__m128i *)(b->quant));
+    __m128i quant1 = _mm_load_si128((__m128i *)(b->quant + 8));
+    __m128i dequant0 = _mm_load_si128((__m128i *)(d->dequant));
+    __m128i dequant1 = _mm_load_si128((__m128i *)(d->dequant + 8));
+
+    vpx_memset(qcoeff_ptr, 0, 32);
+
+    /* Duplicate to all lanes. */
+    zbin_extra = _mm_shufflelo_epi16(zbin_extra, 0);
+    zbin_extra = _mm_unpacklo_epi16(zbin_extra, zbin_extra);
+
+    /* Sign of z: z >> 15 */
+    sz0 = _mm_srai_epi16(z0, 15);
+    sz1 = _mm_srai_epi16(z1, 15);
+
+    /* x = abs(z): (z ^ sz) - sz */
+    x0 = _mm_xor_si128(z0, sz0);
+    x1 = _mm_xor_si128(z1, sz1);
+    x0 = _mm_sub_epi16(x0, sz0);
+    x1 = _mm_sub_epi16(x1, sz1);
+
+    /* zbin[] + zbin_extra */
+    zbin0 = _mm_add_epi16(zbin0, zbin_extra);
+    zbin1 = _mm_add_epi16(zbin1, zbin_extra);
+
+    /* In C x is compared to zbin where zbin = zbin[] + boost + extra. Rebalance
+     * the equation because boost is the only value which can change:
+     * x - (zbin[] + extra) >= boost */
+    x_minus_zbin0 = _mm_sub_epi16(x0, zbin0);
+    x_minus_zbin1 = _mm_sub_epi16(x1, zbin1);
+
+    _mm_store_si128((__m128i *)(x), x_minus_zbin0);
+    _mm_store_si128((__m128i *)(x + 8), x_minus_zbin1);
+
+    /* All the remaining calculations are valid whether they are done now with
+     * simd or later inside the loop one at a time. */
+    x0 = _mm_add_epi16(x0, round0);
+    x1 = _mm_add_epi16(x1, round1);
+
+    y0 = _mm_mulhi_epi16(x0, quant0);
+    y1 = _mm_mulhi_epi16(x1, quant1);
+
+    y0 = _mm_add_epi16(y0, x0);
+    y1 = _mm_add_epi16(y1, x1);
+
+    /* Instead of shifting each value independently we convert the scaling
+     * factor with 1 << (16 - shift) so we can use multiply/return high half. */
+    y0 = _mm_mulhi_epi16(y0, quant_shift0);
+    y1 = _mm_mulhi_epi16(y1, quant_shift1);
+
+    /* Return the sign: (y ^ sz) - sz */
+    y0 = _mm_xor_si128(y0, sz0);
+    y1 = _mm_xor_si128(y1, sz1);
+    y0 = _mm_sub_epi16(y0, sz0);
+    y1 = _mm_sub_epi16(y1, sz1);
+
+    _mm_store_si128((__m128i *)(y), y0);
+    _mm_store_si128((__m128i *)(y + 8), y1);
+
+    zbin_boost_ptr = b->zrun_zbin_boost;
+
+    /* The loop gets unrolled anyway. Avoid the vp8_default_zig_zag1d lookup. */
+    SELECT_EOB(1, 0);
+    SELECT_EOB(2, 1);
+    SELECT_EOB(3, 4);
+    SELECT_EOB(4, 8);
+    SELECT_EOB(5, 5);
+    SELECT_EOB(6, 2);
+    SELECT_EOB(7, 3);
+    SELECT_EOB(8, 6);
+    SELECT_EOB(9, 9);
+    SELECT_EOB(10, 12);
+    SELECT_EOB(11, 13);
+    SELECT_EOB(12, 10);
+    SELECT_EOB(13, 7);
+    SELECT_EOB(14, 11);
+    SELECT_EOB(15, 14);
+    SELECT_EOB(16, 15);
+
+    y0 = _mm_load_si128((__m128i *)(d->qcoeff));
+    y1 = _mm_load_si128((__m128i *)(d->qcoeff + 8));
+
+    /* dqcoeff = qcoeff * dequant */
+    y0 = _mm_mullo_epi16(y0, dequant0);
+    y1 = _mm_mullo_epi16(y1, dequant1);
+
+    _mm_store_si128((__m128i *)(d->dqcoeff), y0);
+    _mm_store_si128((__m128i *)(d->dqcoeff + 8), y1);
+
+    *d->eob = eob;
+}
+
+void vp8_fast_quantize_b_sse2(BLOCK *b, BLOCKD *d)
+{
+  __m128i z0 = _mm_load_si128((__m128i *)(b->coeff));
+  __m128i z1 = _mm_load_si128((__m128i *)(b->coeff + 8));
+  __m128i round0 = _mm_load_si128((__m128i *)(b->round));
+  __m128i round1 = _mm_load_si128((__m128i *)(b->round + 8));
+  __m128i quant_fast0 = _mm_load_si128((__m128i *)(b->quant_fast));
+  __m128i quant_fast1 = _mm_load_si128((__m128i *)(b->quant_fast + 8));
+  __m128i dequant0 = _mm_load_si128((__m128i *)(d->dequant));
+  __m128i dequant1 = _mm_load_si128((__m128i *)(d->dequant + 8));
+  __m128i inv_zig_zag0 = _mm_load_si128((const __m128i *)(vp8_default_inv_zig_zag));
+  __m128i inv_zig_zag1 = _mm_load_si128((const __m128i *)(vp8_default_inv_zig_zag + 8));
+
+  __m128i sz0, sz1, x0, x1, y0, y1, xdq0, xdq1, zeros, ones;
+
+  /* sign of z: z >> 15 */
+  sz0 = _mm_srai_epi16(z0, 15);
+  sz1 = _mm_srai_epi16(z1, 15);
+
+  /* x = abs(z): (z ^ sz) - sz */
+  x0 = _mm_xor_si128(z0, sz0);
+  x1 = _mm_xor_si128(z1, sz1);
+  x0 = _mm_sub_epi16(x0, sz0);
+  x1 = _mm_sub_epi16(x1, sz1);
+
+  /* x += round */
+  x0 = _mm_add_epi16(x0, round0);
+  x1 = _mm_add_epi16(x1, round1);
+
+  /* y = (x * quant) >> 16 */
+  y0 = _mm_mulhi_epi16(x0, quant_fast0);
+  y1 = _mm_mulhi_epi16(x1, quant_fast1);
+
+  /* x = abs(y) = (y ^ sz) - sz */
+  y0 = _mm_xor_si128(y0, sz0);
+  y1 = _mm_xor_si128(y1, sz1);
+  x0 = _mm_sub_epi16(y0, sz0);
+  x1 = _mm_sub_epi16(y1, sz1);
+
+  /* qcoeff = x */
+  _mm_store_si128((__m128i *)(d->qcoeff), x0);
+  _mm_store_si128((__m128i *)(d->qcoeff + 8), x1);
+
+  /* x * dequant */
+  xdq0 = _mm_mullo_epi16(x0, dequant0);
+  xdq1 = _mm_mullo_epi16(x1, dequant1);
+
+  /* dqcoeff = x * dequant */
+  _mm_store_si128((__m128i *)(d->dqcoeff), xdq0);
+  _mm_store_si128((__m128i *)(d->dqcoeff + 8), xdq1);
+
+  /* build a mask for the zig zag */
+  zeros = _mm_setzero_si128();
+
+  x0 = _mm_cmpeq_epi16(x0, zeros);
+  x1 = _mm_cmpeq_epi16(x1, zeros);
+
+  ones = _mm_cmpeq_epi16(zeros, zeros);
+
+  x0 = _mm_xor_si128(x0, ones);
+  x1 = _mm_xor_si128(x1, ones);
+
+  x0 = _mm_and_si128(x0, inv_zig_zag0);
+  x1 = _mm_and_si128(x1, inv_zig_zag1);
+
+  x0 = _mm_max_epi16(x0, x1);
+
+  /* now down to 8 */
+  x1 = _mm_shuffle_epi32(x0, 0xE); // 0b00001110
+
+  x0 = _mm_max_epi16(x0, x1);
+
+  /* only 4 left */
+  x1 = _mm_shufflelo_epi16(x0, 0xE); // 0b00001110
+
+  x0 = _mm_max_epi16(x0, x1);
+
+  /* okay, just 2! */
+  x1 = _mm_shufflelo_epi16(x0, 0x1); // 0b00000001
+
+  x0 = _mm_max_epi16(x0, x1);
+
+  *d->eob = 0xFF & _mm_cvtsi128_si32(x0);
+}
diff --git a/libvpx/vp8/encoder/x86/quantize_sse4.asm b/libvpx/vp8/encoder/x86/quantize_sse4.asm
index f0e5d40..dbd171b 100644
--- a/libvpx/vp8/encoder/x86/quantize_sse4.asm
+++ b/libvpx/vp8/encoder/x86/quantize_sse4.asm
@@ -9,7 +9,7 @@
 
 
 %include "vpx_ports/x86_abi_support.asm"
-%include "asm_enc_offsets.asm"
+%include "vp8_asm_enc_offsets.asm"
 
 
 ; void vp8_regular_quantize_b_sse4 | arg
@@ -31,7 +31,7 @@
     %define stack_size 32
     sub         rsp, stack_size
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     SAVE_XMM 8, u
     push        rdi
     push        rsi
@@ -43,7 +43,7 @@
     mov         rdi, arg(0)                 ; BLOCK *b
     mov         rsi, arg(1)                 ; BLOCKD *d
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     mov         rdi, rcx                    ; BLOCK *b
     mov         rsi, rdx                    ; BLOCKD *d
   %else
@@ -240,7 +240,7 @@
     pop         rbp
 %else
   %undef xmm5
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     pop         rsi
     pop         rdi
     RESTORE_XMM
diff --git a/libvpx/vp8/encoder/x86/quantize_ssse3.asm b/libvpx/vp8/encoder/x86/quantize_ssse3.asm
index dd526f4..7b1dc11 100644
--- a/libvpx/vp8/encoder/x86/quantize_ssse3.asm
+++ b/libvpx/vp8/encoder/x86/quantize_ssse3.asm
@@ -9,7 +9,7 @@
 
 
 %include "vpx_ports/x86_abi_support.asm"
-%include "asm_enc_offsets.asm"
+%include "vp8_asm_enc_offsets.asm"
 
 
 ; void vp8_fast_quantize_b_ssse3 | arg
@@ -27,7 +27,7 @@
     push        rdi
     push        rsi
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     push        rdi
     push        rsi
   %endif
@@ -38,7 +38,7 @@
     mov         rdi, arg(0)                 ; BLOCK *b
     mov         rsi, arg(1)                 ; BLOCKD *d
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     mov         rdi, rcx                    ; BLOCK *b
     mov         rsi, rdx                    ; BLOCKD *d
   %else
@@ -122,7 +122,7 @@
     pop         rsi
     pop         rdi
 %else
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     pop         rsi
     pop         rdi
   %endif
diff --git a/libvpx/vp8/encoder/x86/temporal_filter_apply_sse2.asm b/libvpx/vp8/encoder/x86/temporal_filter_apply_sse2.asm
index ce9d983..bd92b39 100644
--- a/libvpx/vp8/encoder/x86/temporal_filter_apply_sse2.asm
+++ b/libvpx/vp8/encoder/x86/temporal_filter_apply_sse2.asm
@@ -50,7 +50,7 @@
         ; 0x8000 >> (16 - strength)
         mov         rdx,            16
         sub         rdx,            arg(4) ; 16 - strength
-        movd        xmm4,           rdx    ; can't use rdx w/ shift
+        movq        xmm4,           rdx    ; can't use rdx w/ shift
         movdqa      xmm5,           [GLOBAL(_const_top_bit)]
         psrlw       xmm5,           xmm4
         movdqa      [rsp + rounding_bit], xmm5
diff --git a/libvpx/vp8/encoder/x86/vp8_enc_stubs_mmx.c b/libvpx/vp8/encoder/x86/vp8_enc_stubs_mmx.c
index da25f52..cf3d8ca 100644
--- a/libvpx/vp8/encoder/x86/vp8_enc_stubs_mmx.c
+++ b/libvpx/vp8/encoder/x86/vp8_enc_stubs_mmx.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_ports/x86.h"
 #include "vp8/encoder/block.h"
 
diff --git a/libvpx/vp8/encoder/x86/vp8_enc_stubs_sse2.c b/libvpx/vp8/encoder/x86/vp8_enc_stubs_sse2.c
index 68db815..3dfbee3 100644
--- a/libvpx/vp8/encoder/x86/vp8_enc_stubs_sse2.c
+++ b/libvpx/vp8/encoder/x86/vp8_enc_stubs_sse2.c
@@ -10,7 +10,7 @@
 
 
 #include "vpx_config.h"
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx_ports/x86.h"
 #include "vp8/encoder/block.h"
 
diff --git a/libvpx/vp8/vp8_common.mk b/libvpx/vp8/vp8_common.mk
index a328f46..cde2651 100644
--- a/libvpx/vp8/vp8_common.mk
+++ b/libvpx/vp8/vp8_common.mk
@@ -14,7 +14,6 @@
 VP8_COMMON_SRCS-yes += common/onyx.h
 VP8_COMMON_SRCS-yes += common/onyxd.h
 VP8_COMMON_SRCS-yes += common/alloccommon.c
-VP8_COMMON_SRCS-yes += common/asm_com_offsets.c
 VP8_COMMON_SRCS-yes += common/blockd.c
 VP8_COMMON_SRCS-yes += common/coefupdateprobs.h
 VP8_COMMON_SRCS-yes += common/debugmodes.c
@@ -67,6 +66,7 @@
 VP8_COMMON_SRCS-yes += common/swapyv12buffer.c
 VP8_COMMON_SRCS-yes += common/variance_c.c
 VP8_COMMON_SRCS-yes += common/variance.h
+VP8_COMMON_SRCS-yes += common/vp8_asm_com_offsets.c
 VP8_COMMON_SRCS-yes += common/vp8_entropymodedata.h
 
 
@@ -191,3 +191,8 @@
 VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance8x8_neon$(ASM)
 VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance16x16_neon$(ASM)
 VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance16x16s_neon$(ASM)
+
+$(eval $(call asm_offsets_template,\
+         vp8_asm_com_offsets.asm, $(VP8_PREFIX)common/vp8_asm_com_offsets.c))
+
+$(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.sh))
diff --git a/libvpx/vp8/vp8_cx_iface.c b/libvpx/vp8/vp8_cx_iface.c
index eeac3a8..4531d5a 100644
--- a/libvpx/vp8/vp8_cx_iface.c
+++ b/libvpx/vp8/vp8_cx_iface.c
@@ -9,7 +9,7 @@
  */
 
 
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/vpx_codec.h"
 #include "vpx/internal/vpx_codec_internal.h"
 #include "vpx_version.h"
@@ -582,7 +582,7 @@
 
     struct VP8_COMP *optr;
 
-    vpx_rtcd();
+    vp8_rtcd();
 
     if (!ctx->priv)
     {
@@ -684,6 +684,8 @@
     yv12->u_buffer = img->planes[VPX_PLANE_U];
     yv12->v_buffer = img->planes[VPX_PLANE_V];
 
+    yv12->y_crop_width  = img->d_w;
+    yv12->y_crop_height = img->d_h;
     yv12->y_width  = img->d_w;
     yv12->y_height = img->d_h;
     yv12->uv_width = (1 + yv12->y_width) / 2;
@@ -1178,7 +1180,9 @@
     {
         int res;
         vpx_scaling_mode_t scalemode = *(vpx_scaling_mode_t *)data ;
-        res = vp8_set_internal_size(ctx->cpi, scalemode.h_scaling_mode, scalemode.v_scaling_mode);
+        res = vp8_set_internal_size(ctx->cpi,
+                                    (VPX_SCALING)scalemode.h_scaling_mode,
+                                    (VPX_SCALING)scalemode.v_scaling_mode);
 
         if (!res)
         {
diff --git a/libvpx/vp8/vp8_dx_iface.c b/libvpx/vp8/vp8_dx_iface.c
index c13d697..c826f69 100644
--- a/libvpx/vp8/vp8_dx_iface.c
+++ b/libvpx/vp8/vp8_dx_iface.c
@@ -11,7 +11,7 @@
 
 #include <stdlib.h>
 #include <string.h>
-#include "vpx_rtcd.h"
+#include "vp8_rtcd.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx/vp8dx.h"
 #include "vpx/internal/vpx_codec_internal.h"
@@ -64,7 +64,6 @@
     vp8_stream_info_t       si;
     int                     defer_alloc;
     int                     decoder_init;
-    struct VP8D_COMP       *pbi;
     int                     postproc_cfg_set;
     vp8_postproc_cfg_t      postproc_cfg;
 #if CONFIG_POSTPROC_VISUALIZER
@@ -74,9 +73,13 @@
     int                     dbg_color_b_modes_flag;
     int                     dbg_display_mv_flag;
 #endif
+    vp8_decrypt_cb          *decrypt_cb;
+    void                    *decrypt_state;
     vpx_image_t             img;
     int                     img_setup;
+    struct frame_buffers    yv12_frame_buffers;
     void                    *user_priv;
+    FRAGMENT_DATA           fragments;
 };
 
 static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t flags)
@@ -163,6 +166,8 @@
 
     ctx->priv->alg_priv->mmaps[0] = *mmap;
     ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si);
+    ctx->priv->alg_priv->decrypt_cb = NULL;
+    ctx->priv->alg_priv->decrypt_state = NULL;
     ctx->priv->init_flags = ctx->init_flags;
 
     if (ctx->config.dec)
@@ -194,7 +199,7 @@
     vpx_codec_err_t        res = VPX_CODEC_OK;
     (void) data;
 
-    vpx_rtcd();
+    vp8_rtcd();
 
     /* This function only allocates space for the vpx_codec_alg_priv_t
      * structure. More memory may be required at the time the stream
@@ -210,14 +215,37 @@
         mmap.flags = vp8_mem_req_segs[0].flags;
 
         res = vp8_mmap_alloc(&mmap);
+        if (res != VPX_CODEC_OK) return res;
 
-        if (!res)
-        {
-            vp8_init_ctx(ctx, &mmap);
+        vp8_init_ctx(ctx, &mmap);
 
-            ctx->priv->alg_priv->defer_alloc = 1;
-            /*post processing level initialized to do nothing */
-        }
+        /* initialize number of fragments to zero */
+        ctx->priv->alg_priv->fragments.count = 0;
+        /* is input fragments enabled? */
+        ctx->priv->alg_priv->fragments.enabled =
+                (ctx->priv->alg_priv->base.init_flags &
+                    VPX_CODEC_USE_INPUT_FRAGMENTS);
+
+        ctx->priv->alg_priv->defer_alloc = 1;
+        /*post processing level initialized to do nothing */
+    }
+
+    ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads =
+            (ctx->priv->alg_priv->base.init_flags &
+                    VPX_CODEC_USE_FRAME_THREADING);
+
+    /* for now, disable frame threading */
+    ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads = 0;
+
+    if(ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads &&
+            (( ctx->priv->alg_priv->base.init_flags &
+                            VPX_CODEC_USE_ERROR_CONCEALMENT)
+                    || ( ctx->priv->alg_priv->base.init_flags &
+                            VPX_CODEC_USE_INPUT_FRAGMENTS) ) )
+    {
+        /* row-based threading, error concealment, and input fragments will
+         * not be supported when using frame-based threading */
+        res = VPX_CODEC_INVALID_PARAM;
     }
 
     return res;
@@ -227,7 +255,7 @@
 {
     int i;
 
-    vp8dx_remove_decompressor(ctx->pbi);
+    vp8_remove_decoder_instances(&ctx->yv12_frame_buffers);
 
     for (i = NELEMENTS(ctx->mmaps) - 1; i >= 0; i--)
     {
@@ -238,14 +266,18 @@
     return VPX_CODEC_OK;
 }
 
-static vpx_codec_err_t vp8_peek_si(const uint8_t         *data,
-                                   unsigned int           data_sz,
-                                   vpx_codec_stream_info_t *si)
+static vpx_codec_err_t vp8_peek_si_internal(const uint8_t *data,
+                                            unsigned int data_sz,
+                                            vpx_codec_stream_info_t *si,
+                                            vp8_decrypt_cb *decrypt_cb,
+                                            void *decrypt_state)
 {
     vpx_codec_err_t res = VPX_CODEC_OK;
 
     if(data + data_sz <= data)
+    {
         res = VPX_CODEC_INVALID_PARAM;
+    }
     else
     {
         /* Parse uncompresssed part of key frame header.
@@ -254,30 +286,44 @@
          * 4 bytes:- including image width and height in the lowest 14 bits
          *           of each 2-byte value.
          */
+        uint8_t clear_buffer[10];
+        const uint8_t *clear = data;
+        if (decrypt_cb)
+        {
+            int n = data_sz > 10 ? 10 : data_sz;
+            decrypt_cb(decrypt_state, data, clear_buffer, n);
+            clear = clear_buffer;
+        }
         si->is_kf = 0;
 
-        if (data_sz >= 10 && !(data[0] & 0x01))  /* I-Frame */
+        if (data_sz >= 10 && !(clear[0] & 0x01))  /* I-Frame */
         {
-            const uint8_t *c = data + 3;
             si->is_kf = 1;
 
             /* vet via sync code */
-            if (c[0] != 0x9d || c[1] != 0x01 || c[2] != 0x2a)
+            if (clear[3] != 0x9d || clear[4] != 0x01 || clear[5] != 0x2a)
                 res = VPX_CODEC_UNSUP_BITSTREAM;
 
-            si->w = (c[3] | (c[4] << 8)) & 0x3fff;
-            si->h = (c[5] | (c[6] << 8)) & 0x3fff;
+            si->w = (clear[6] | (clear[7] << 8)) & 0x3fff;
+            si->h = (clear[8] | (clear[9] << 8)) & 0x3fff;
 
             /*printf("w=%d, h=%d\n", si->w, si->h);*/
             if (!(si->h | si->w))
                 res = VPX_CODEC_UNSUP_BITSTREAM;
         }
         else
+        {
             res = VPX_CODEC_UNSUP_BITSTREAM;
+        }
     }
 
     return res;
+}
 
+static vpx_codec_err_t vp8_peek_si(const uint8_t *data,
+                                   unsigned int data_sz,
+                                   vpx_codec_stream_info_t *si) {
+    return vp8_peek_si_internal(data, data_sz, si, NULL, NULL);
 }
 
 static vpx_codec_err_t vp8_get_si(vpx_codec_alg_priv_t    *ctx,
@@ -343,6 +389,47 @@
     img->self_allocd = 0;
 }
 
+static int
+update_fragments(vpx_codec_alg_priv_t  *ctx,
+                 const uint8_t         *data,
+                 unsigned int           data_sz,
+                 vpx_codec_err_t       *res)
+{
+    *res = VPX_CODEC_OK;
+
+    if (ctx->fragments.count == 0)
+    {
+        /* New frame, reset fragment pointers and sizes */
+        vpx_memset((void*)ctx->fragments.ptrs, 0, sizeof(ctx->fragments.ptrs));
+        vpx_memset(ctx->fragments.sizes, 0, sizeof(ctx->fragments.sizes));
+    }
+    if (ctx->fragments.enabled && !(data == NULL && data_sz == 0))
+    {
+        /* Store a pointer to this fragment and return. We haven't
+         * received the complete frame yet, so we will wait with decoding.
+         */
+        ctx->fragments.ptrs[ctx->fragments.count] = data;
+        ctx->fragments.sizes[ctx->fragments.count] = data_sz;
+        ctx->fragments.count++;
+        if (ctx->fragments.count > (1 << EIGHT_PARTITION) + 1)
+        {
+            ctx->fragments.count = 0;
+            *res = VPX_CODEC_INVALID_PARAM;
+            return -1;
+        }
+        return 0;
+    }
+
+    if (!ctx->fragments.enabled)
+    {
+        ctx->fragments.ptrs[0] = data;
+        ctx->fragments.sizes[0] = data_sz;
+        ctx->fragments.count = 1;
+    }
+
+    return 1;
+}
+
 static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t  *ctx,
                                   const uint8_t         *data,
                                   unsigned int            data_sz,
@@ -353,6 +440,11 @@
     unsigned int resolution_change = 0;
     unsigned int w, h;
 
+
+    /* Update the input fragment data */
+    if(update_fragments(ctx, data, data_sz, &res) <= 0)
+        return res;
+
     /* Determine the stream parameters. Note that we rely on peek_si to
      * validate that we have a buffer that does not wrap around the top
      * of the heap.
@@ -360,7 +452,8 @@
     w = ctx->si.w;
     h = ctx->si.h;
 
-    res = ctx->base.iface->dec.peek_si(data, data_sz, &ctx->si);
+    res = vp8_peek_si_internal(ctx->fragments.ptrs[0], ctx->fragments.sizes[0],
+                               &ctx->si, ctx->decrypt_cb, ctx->decrypt_state);
 
     if((res == VPX_CODEC_UNSUP_BITSTREAM) && !ctx->si.is_kf)
     {
@@ -412,7 +505,6 @@
         if (!res)
         {
             VP8D_CONFIG oxcf;
-            struct VP8D_COMP* optr;
 
             oxcf.Width = ctx->si.w;
             oxcf.Height = ctx->si.h;
@@ -421,10 +513,6 @@
             oxcf.max_threads = ctx->cfg.threads;
             oxcf.error_concealment =
                     (ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT);
-            oxcf.input_fragments =
-                    (ctx->base.init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS);
-
-            optr = vp8dx_create_decompressor(&oxcf);
 
             /* If postprocessing was enabled by the application and a
              * configuration has not been provided, default it.
@@ -438,20 +526,19 @@
                 ctx->postproc_cfg.noise_level = 0;
             }
 
-            if (!optr)
-                res = VPX_CODEC_ERROR;
-            else
-                ctx->pbi = optr;
+            res = vp8_create_decoder_instances(&ctx->yv12_frame_buffers, &oxcf);
+            ctx->yv12_frame_buffers.pbi[0]->decrypt_cb = ctx->decrypt_cb;
+            ctx->yv12_frame_buffers.pbi[0]->decrypt_state = ctx->decrypt_state;
         }
 
         ctx->decoder_init = 1;
     }
 
-    if (!res && ctx->pbi)
+    if (!res)
     {
+        VP8D_COMP *pbi = ctx->yv12_frame_buffers.pbi[0];
         if(resolution_change)
         {
-            VP8D_COMP *pbi = ctx->pbi;
             VP8_COMMON *const pc = & pbi->common;
             MACROBLOCKD *const xd  = & pbi->mb;
 #if CONFIG_MULTITHREAD
@@ -541,15 +628,20 @@
             pbi->common.error.setjmp = 0;
 
             /* required to get past the first get_free_fb() call */
-            ctx->pbi->common.fb_idx_ref_cnt[0] = 0;
+            pbi->common.fb_idx_ref_cnt[0] = 0;
         }
 
+        /* update the pbi fragment data */
+        pbi->fragments = ctx->fragments;
+
         ctx->user_priv = user_priv;
-        if (vp8dx_receive_compressed_data(ctx->pbi, data_sz, data, deadline))
+        if (vp8dx_receive_compressed_data(pbi, data_sz, data, deadline))
         {
-            VP8D_COMP *pbi = (VP8D_COMP *)ctx->pbi;
             res = update_error_state(ctx, &pbi->common.error);
         }
+
+        /* get ready for the next series of fragments */
+        ctx->fragments.count = 0;
     }
 
     return res;
@@ -563,7 +655,7 @@
     /* iter acts as a flip flop, so an image is only returned on the first
      * call to get_frame.
      */
-    if (!(*iter))
+    if (!(*iter) && ctx->yv12_frame_buffers.pbi[0])
     {
         YV12_BUFFER_CONFIG sd;
         int64_t time_stamp = 0, time_end_stamp = 0;
@@ -590,7 +682,8 @@
 #endif
         }
 
-        if (0 == vp8dx_get_raw_frame(ctx->pbi, &sd, &time_stamp, &time_end_stamp, &flags))
+        if (0 == vp8dx_get_raw_frame(ctx->yv12_frame_buffers.pbi[0], &sd,
+                                     &time_stamp, &time_end_stamp, &flags))
         {
             yuvconfig2image(&ctx->img, &sd, ctx->user_priv);
 
@@ -693,6 +786,8 @@
     yv12->u_buffer = img->planes[VPX_PLANE_U];
     yv12->v_buffer = img->planes[VPX_PLANE_V];
 
+    yv12->y_crop_width  = img->d_w;
+    yv12->y_crop_height = img->d_h;
     yv12->y_width  = img->d_w;
     yv12->y_height = img->d_h;
     yv12->uv_width = yv12->y_width / 2;
@@ -715,14 +810,15 @@
 
     vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
 
-    if (data)
+    if (data && !ctx->yv12_frame_buffers.use_frame_threads)
     {
         vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
         YV12_BUFFER_CONFIG sd;
 
         image2yuvconfig(&frame->img, &sd);
 
-        return vp8dx_set_reference(ctx->pbi, frame->frame_type, &sd);
+        return vp8dx_set_reference(ctx->yv12_frame_buffers.pbi[0],
+                                   frame->frame_type, &sd);
     }
     else
         return VPX_CODEC_INVALID_PARAM;
@@ -736,14 +832,15 @@
 
     vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
 
-    if (data)
+    if (data && !ctx->yv12_frame_buffers.use_frame_threads)
     {
         vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
         YV12_BUFFER_CONFIG sd;
 
         image2yuvconfig(&frame->img, &sd);
 
-        return vp8dx_get_reference(ctx->pbi, frame->frame_type, &sd);
+        return vp8dx_get_reference(ctx->yv12_frame_buffers.pbi[0],
+                                   frame->frame_type, &sd);
     }
     else
         return VPX_CODEC_INVALID_PARAM;
@@ -799,10 +896,11 @@
                                                 va_list args)
 {
     int *update_info = va_arg(args, int *);
-    VP8D_COMP *pbi = (VP8D_COMP *)ctx->pbi;
 
-    if (update_info)
+    if (update_info && !ctx->yv12_frame_buffers.use_frame_threads)
     {
+        VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0];
+
         *update_info = pbi->common.refresh_alt_ref_frame * (int) VP8_ALTR_FRAME
             + pbi->common.refresh_golden_frame * (int) VP8_GOLD_FRAME
             + pbi->common.refresh_last_frame * (int) VP8_LAST_FRAME;
@@ -819,11 +917,11 @@
                                               va_list args)
 {
     int *ref_info = va_arg(args, int *);
-    VP8D_COMP *pbi = (VP8D_COMP *)ctx->pbi;
-    VP8_COMMON *oci = &pbi->common;
 
-    if (ref_info)
+    if (ref_info && !ctx->yv12_frame_buffers.use_frame_threads)
     {
+        VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0];
+        VP8_COMMON *oci = &pbi->common;
         *ref_info =
             (vp8dx_references_buffer( oci, ALTREF_FRAME )?VP8_ALTR_FRAME:0) |
             (vp8dx_references_buffer( oci, GOLDEN_FRAME )?VP8_GOLD_FRAME:0) |
@@ -841,10 +939,10 @@
 {
 
     int *corrupted = va_arg(args, int *);
+    VP8D_COMP *pbi = (VP8D_COMP *)ctx->yv12_frame_buffers.pbi[0];
 
-    if (corrupted)
+    if (corrupted && pbi)
     {
-        VP8D_COMP *pbi = (VP8D_COMP *)ctx->pbi;
         *corrupted = pbi->common.frame_to_show->corrupted;
 
         return VPX_CODEC_OK;
@@ -854,6 +952,25 @@
 
 }
 
+static vpx_codec_err_t vp8_set_decryptor(vpx_codec_alg_priv_t *ctx,
+                                         int ctrl_id,
+                                         va_list args)
+{
+    vp8_decrypt_init *init = va_arg(args, vp8_decrypt_init *);
+
+    if (init)
+    {
+        ctx->decrypt_cb = init->decrypt_cb;
+        ctx->decrypt_state = init->decrypt_state;
+    }
+    else
+    {
+        ctx->decrypt_cb = NULL;
+        ctx->decrypt_state = NULL;
+    }
+    return VPX_CODEC_OK;
+}
+
 vpx_codec_ctrl_fn_map_t vp8_ctf_maps[] =
 {
     {VP8_SET_REFERENCE,             vp8_set_reference},
@@ -866,6 +983,7 @@
     {VP8D_GET_LAST_REF_UPDATES,     vp8_get_last_ref_updates},
     {VP8D_GET_FRAME_CORRUPTED,      vp8_get_frame_corrupted},
     {VP8D_GET_LAST_REF_USED,        vp8_get_last_ref_frame},
+    {VP8D_SET_DECRYPTOR,            vp8_set_decryptor},
     { -1, NULL},
 };
 
diff --git a/libvpx/vp8/vp8cx.mk b/libvpx/vp8/vp8cx.mk
index 0ae2f10..cd091f3 100644
--- a/libvpx/vp8/vp8cx.mk
+++ b/libvpx/vp8/vp8cx.mk
@@ -9,8 +9,6 @@
 ##
 
 
-include $(SRC_PATH_BARE)/$(VP8_PREFIX)vp8_common.mk
-
 VP8_CX_EXPORTS += exports_enc
 
 VP8_CX_SRCS-yes += $(VP8_COMMON_SRCS-yes)
@@ -26,7 +24,6 @@
 
 VP8_CX_SRCS-yes += vp8_cx_iface.c
 
-VP8_CX_SRCS-yes += encoder/asm_enc_offsets.c
 VP8_CX_SRCS-yes += encoder/defaultcoefcounts.h
 VP8_CX_SRCS-yes += encoder/bitstream.c
 VP8_CX_SRCS-yes += encoder/boolhuff.c
@@ -80,6 +77,7 @@
 VP8_CX_SRCS-yes += encoder/temporal_filter.c
 VP8_CX_SRCS-$(CONFIG_MULTI_RES_ENCODING) += encoder/mr_dissim.c
 VP8_CX_SRCS-$(CONFIG_MULTI_RES_ENCODING) += encoder/mr_dissim.h
+VP8_CX_SRCS-yes += encoder/vp8_asm_enc_offsets.c
 
 ifeq ($(CONFIG_REALTIME_ONLY),yes)
 VP8_CX_SRCS_REMOVE-yes += encoder/firstpass.c
@@ -91,13 +89,10 @@
 VP8_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp8_enc_stubs_mmx.c
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/fwalsh_sse2.asm
-VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.asm
+VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.c
 
 ifeq ($(CONFIG_TEMPORAL_DENOISING),yes)
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/denoising_sse2.c
-ifeq ($(HAVE_SSE2),yes)
-vp8/encoder/x86/denoising_sse2.c.o: CFLAGS += -msse2
-endif
 endif
 
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm
@@ -113,5 +108,7 @@
 VP8_CX_SRCS_REMOVE-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
 endif
 
-
 VP8_CX_SRCS-yes := $(filter-out $(VP8_CX_SRCS_REMOVE-yes),$(VP8_CX_SRCS-yes))
+
+$(eval $(call asm_offsets_template,\
+         vp8_asm_enc_offsets.asm, $(VP8_PREFIX)encoder/vp8_asm_enc_offsets.c))
diff --git a/libvpx/vp8/vp8dx.mk b/libvpx/vp8/vp8dx.mk
index dd39190..c26f42d 100644
--- a/libvpx/vp8/vp8dx.mk
+++ b/libvpx/vp8/vp8dx.mk
@@ -9,8 +9,6 @@
 ##
 
 
-include $(SRC_PATH_BARE)/$(VP8_PREFIX)vp8_common.mk
-
 VP8_DX_EXPORTS += exports_dec
 
 VP8_DX_SRCS-yes += $(VP8_COMMON_SRCS-yes)
@@ -22,31 +20,6 @@
 
 VP8_DX_SRCS-yes += vp8_dx_iface.c
 
-# common
-#define ARM
-#define DISABLE_THREAD
-
-#INCLUDES += algo/vpx_common/vpx_mem/include
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += decoder
-
-
-
-# decoder
-#define ARM
-#define DISABLE_THREAD
-
-#INCLUDES += algo/vpx_common/vpx_mem/include
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += common
-#INCLUDES += decoder
-
-VP8_DX_SRCS-yes += decoder/asm_dec_offsets.c
 VP8_DX_SRCS-yes += decoder/dboolhuff.c
 VP8_DX_SRCS-yes += decoder/decodemv.c
 VP8_DX_SRCS-yes += decoder/decodframe.c
@@ -62,5 +35,9 @@
 VP8_DX_SRCS-yes += decoder/treereader.h
 VP8_DX_SRCS-yes += decoder/onyxd_if.c
 VP8_DX_SRCS-$(CONFIG_MULTITHREAD) += decoder/threading.c
+VP8_DX_SRCS-yes += decoder/vp8_asm_dec_offsets.c
 
 VP8_DX_SRCS-yes := $(filter-out $(VP8_DX_SRCS_REMOVE-yes),$(VP8_DX_SRCS-yes))
+
+$(eval $(call asm_offsets_template,\
+         vp8_asm_dec_offsets.asm, $(VP8_PREFIX)decoder/vp8_asm_dec_offsets.c))
diff --git a/libvpx/vp8_multi_resolution_encoder.c b/libvpx/vp8_multi_resolution_encoder.c
index eae36a4..4c29056 100644
--- a/libvpx/vp8_multi_resolution_encoder.c
+++ b/libvpx/vp8_multi_resolution_encoder.c
@@ -216,7 +216,7 @@
      * If target bitrate for highest-resolution level is set to 0,
      * (i.e. target_bitrate[0]=0), we skip encoding at that level.
      */
-    unsigned int         target_bitrate[NUM_ENCODERS]={1400, 500, 100};
+    unsigned int         target_bitrate[NUM_ENCODERS]={1000, 500, 100};
     /* Enter the frame rate of the input video */
     int                  framerate = 30;
     /* Set down-sampling factor for each resolution level.
@@ -351,27 +351,26 @@
         if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
             die_codec(&codec[i], "Failed to set cpu_used");
     }
-    /* Set static thresh for highest-resolution encoder. Set it to 1000 for
-     * better performance. */
+
+    /* Set static threshold. */
+    for ( i=0; i<NUM_ENCODERS; i++)
     {
-        unsigned int static_thresh = 1000;
-        if(vpx_codec_control(&codec[0], VP8E_SET_STATIC_THRESHOLD, static_thresh))
-            die_codec(&codec[0], "Failed to set static threshold");
-    }
-    /* Set static thresh = 0 for other encoders for better quality */
-    for ( i=1; i<NUM_ENCODERS; i++)
-    {
-        unsigned int static_thresh = 0;
+        unsigned int static_thresh = 1;
         if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh))
             die_codec(&codec[i], "Failed to set static threshold");
     }
+
     /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
-    for ( i=0; i< NUM_ENCODERS; i++)
+    /* Enable denoising for the highest-resolution encoder. */
+    if(vpx_codec_control(&codec[0], VP8E_SET_NOISE_SENSITIVITY, 1))
+        die_codec(&codec[0], "Failed to set noise_sensitivity");
+    for ( i=1; i< NUM_ENCODERS; i++)
     {
         if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0))
             die_codec(&codec[i], "Failed to set noise_sensitivity");
     }
 
+
     frame_avail = 1;
     got_data = 0;
 
diff --git a/libvpx/vp9/common/generic/vp9_systemdependent.c b/libvpx/vp9/common/generic/vp9_systemdependent.c
new file mode 100644
index 0000000..79092cd
--- /dev/null
+++ b/libvpx/vp9/common/generic/vp9_systemdependent.c
@@ -0,0 +1,18 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "./vpx_config.h"
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+void vp9_machine_specific_config(VP9_COMMON *ctx) {
+  vp9_rtcd();
+}
diff --git a/libvpx/vp9/common/vp9_alloccommon.c b/libvpx/vp9/common/vp9_alloccommon.c
new file mode 100644
index 0000000..2660344
--- /dev/null
+++ b/libvpx/vp9/common/vp9_alloccommon.c
@@ -0,0 +1,205 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "./vpx_config.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_systemdependent.h"
+
+void vp9_update_mode_info_border(VP9_COMMON *cm, MODE_INFO *mi) {
+  const int stride = cm->mode_info_stride;
+  int i;
+
+  // Clear down top border row
+  vpx_memset(mi, 0, sizeof(MODE_INFO) * stride);
+
+  // Clear left border column
+  for (i = 1; i < cm->mi_rows + 1; i++)
+    vpx_memset(&mi[i * stride], 0, sizeof(MODE_INFO));
+}
+
+void vp9_update_mode_info_in_image(VP9_COMMON *cm, MODE_INFO *mi) {
+  int i, j;
+
+  // For each in image mode_info element set the in image flag to 1
+  for (i = 0; i < cm->mi_rows; i++) {
+    MODE_INFO *ptr = mi;
+    for (j = 0; j < cm->mi_cols; j++) {
+      ptr->mbmi.mb_in_image = 1;
+      ptr++;  // Next element in the row
+    }
+
+    // Step over border element at start of next row
+    mi += cm->mode_info_stride;
+  }
+}
+
+void vp9_free_frame_buffers(VP9_COMMON *oci) {
+  int i;
+
+  for (i = 0; i < NUM_YV12_BUFFERS; i++)
+    vp9_free_frame_buffer(&oci->yv12_fb[i]);
+
+  vp9_free_frame_buffer(&oci->temp_scale_frame);
+  vp9_free_frame_buffer(&oci->post_proc_buffer);
+
+  vpx_free(oci->mip);
+  vpx_free(oci->prev_mip);
+  vpx_free(oci->above_seg_context);
+
+  vpx_free(oci->above_context[0]);
+  for (i = 0; i < MAX_MB_PLANE; i++)
+    oci->above_context[i] = 0;
+  oci->mip = 0;
+  oci->prev_mip = 0;
+  oci->above_seg_context = 0;
+}
+
+static void set_mb_mi(VP9_COMMON *cm, int aligned_width, int aligned_height) {
+  cm->mb_cols = (aligned_width + 8) >> 4;
+  cm->mb_rows = (aligned_height + 8) >> 4;
+  cm->MBs = cm->mb_rows * cm->mb_cols;
+
+  cm->mi_cols = aligned_width >> LOG2_MI_SIZE;
+  cm->mi_rows = aligned_height >> LOG2_MI_SIZE;
+  cm->mode_info_stride = cm->mi_cols + 64 / MI_SIZE;
+}
+
+static void setup_mi(VP9_COMMON *cm) {
+  cm->mi = cm->mip + cm->mode_info_stride + 1;
+  cm->prev_mi = cm->prev_mip + cm->mode_info_stride + 1;
+
+  vpx_memset(cm->mip, 0,
+             cm->mode_info_stride * (cm->mi_rows + 1) * sizeof(MODE_INFO));
+
+  vp9_update_mode_info_border(cm, cm->mip);
+  vp9_update_mode_info_in_image(cm, cm->mi);
+
+  vp9_update_mode_info_border(cm, cm->prev_mip);
+  vp9_update_mode_info_in_image(cm, cm->prev_mi);
+}
+
+int vp9_alloc_frame_buffers(VP9_COMMON *oci, int width, int height) {
+  int i, mi_cols;
+
+  // Our internal buffers are always multiples of 16
+  const int aligned_width = multiple8(width);
+  const int aligned_height = multiple8(height);
+  const int ss_x = oci->subsampling_x;
+  const int ss_y = oci->subsampling_y;
+
+  vp9_free_frame_buffers(oci);
+
+  for (i = 0; i < NUM_YV12_BUFFERS; i++) {
+    oci->fb_idx_ref_cnt[i] = 0;
+    if (vp9_alloc_frame_buffer(&oci->yv12_fb[i], width, height, ss_x, ss_y,
+                               VP9BORDERINPIXELS) < 0)
+      goto fail;
+  }
+
+  oci->new_fb_idx = NUM_YV12_BUFFERS - 1;
+  oci->fb_idx_ref_cnt[oci->new_fb_idx] = 1;
+
+  for (i = 0; i < ALLOWED_REFS_PER_FRAME; i++)
+    oci->active_ref_idx[i] = i;
+
+  for (i = 0; i < NUM_REF_FRAMES; i++) {
+    oci->ref_frame_map[i] = i;
+    oci->fb_idx_ref_cnt[i] = 1;
+  }
+
+  if (vp9_alloc_frame_buffer(&oci->temp_scale_frame, width, 16, ss_x, ss_y,
+                             VP9BORDERINPIXELS) < 0)
+    goto fail;
+
+  if (vp9_alloc_frame_buffer(&oci->post_proc_buffer, width, height, ss_x, ss_y,
+                             VP9BORDERINPIXELS) < 0)
+    goto fail;
+
+  set_mb_mi(oci, aligned_width, aligned_height);
+
+  // Allocation
+  oci->mip = vpx_calloc(oci->mode_info_stride * (oci->mi_rows + 64 / MI_SIZE),
+                        sizeof(MODE_INFO));
+  if (!oci->mip)
+    goto fail;
+
+  oci->prev_mip = vpx_calloc(oci->mode_info_stride *
+                             (oci->mi_rows + 64 / MI_SIZE),
+                             sizeof(MODE_INFO));
+  if (!oci->prev_mip)
+    goto fail;
+
+  setup_mi(oci);
+
+  // FIXME(jkoleszar): allocate subsampled arrays for U/V once subsampling
+  // information is exposed at this level
+  mi_cols = mi_cols_aligned_to_sb(oci);
+
+  // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
+  // block where mi unit size is 8x8.
+# if CONFIG_ALPHA
+  oci->above_context[0] = vpx_calloc(sizeof(ENTROPY_CONTEXT) * 8 * mi_cols, 1);
+#else
+  oci->above_context[0] = vpx_calloc(sizeof(ENTROPY_CONTEXT) * 6 * mi_cols, 1);
+#endif
+  if (!oci->above_context[0])
+    goto fail;
+
+  for (i = 1; i < MAX_MB_PLANE; i++)
+    oci->above_context[i] =
+        oci->above_context[0] + i * sizeof(ENTROPY_CONTEXT) * 2 * mi_cols;
+
+  oci->above_seg_context = vpx_calloc(sizeof(PARTITION_CONTEXT) * mi_cols, 1);
+  if (!oci->above_seg_context)
+    goto fail;
+
+  return 0;
+
+ fail:
+  vp9_free_frame_buffers(oci);
+  return 1;
+}
+
+void vp9_create_common(VP9_COMMON *oci) {
+  vp9_machine_specific_config(oci);
+
+  vp9_init_mbmode_probs(oci);
+
+  oci->txfm_mode = ONLY_4X4;
+  oci->comp_pred_mode = HYBRID_PREDICTION;
+  oci->clr_type = REG_YUV;
+
+  // Initialize reference frame sign bias structure to defaults
+  vpx_memset(oci->ref_frame_sign_bias, 0, sizeof(oci->ref_frame_sign_bias));
+}
+
+void vp9_remove_common(VP9_COMMON *oci) {
+  vp9_free_frame_buffers(oci);
+}
+
+void vp9_initialize_common() {
+  vp9_coef_tree_initialize();
+  vp9_entropy_mode_init();
+  vp9_entropy_mv_init();
+}
+
+void vp9_update_frame_size(VP9_COMMON *cm) {
+  const int aligned_width = multiple8(cm->width);
+  const int aligned_height = multiple8(cm->height);
+
+  set_mb_mi(cm, aligned_width, aligned_height);
+  setup_mi(cm);
+}
diff --git a/libvpx/vp9/common/vp9_alloccommon.h b/libvpx/vp9/common/vp9_alloccommon.h
new file mode 100644
index 0000000..8bf5ed1
--- /dev/null
+++ b/libvpx/vp9/common/vp9_alloccommon.h
@@ -0,0 +1,31 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_VP9_ALLOCCOMMON_H_
+#define VP9_COMMON_VP9_ALLOCCOMMON_H_
+
+#include "vp9/common/vp9_onyxc_int.h"
+
+void vp9_initialize_common();
+
+void vp9_update_mode_info_border(VP9_COMMON *cpi, MODE_INFO *mi);
+void vp9_update_mode_info_in_image(VP9_COMMON *cpi, MODE_INFO *mi);
+
+void vp9_create_common(VP9_COMMON *oci);
+void vp9_remove_common(VP9_COMMON *oci);
+
+int vp9_alloc_frame_buffers(VP9_COMMON *oci, int width, int height);
+void vp9_free_frame_buffers(VP9_COMMON *oci);
+
+
+void vp9_update_frame_size(VP9_COMMON *cm);
+
+#endif  // VP9_COMMON_VP9_ALLOCCOMMON_H_
diff --git a/libvpx/vp9/common/vp9_asm_com_offsets.c b/libvpx/vp9/common/vp9_asm_com_offsets.c
new file mode 100644
index 0000000..94ccb6e
--- /dev/null
+++ b/libvpx/vp9/common/vp9_asm_com_offsets.c
@@ -0,0 +1,21 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_config.h"
+#include "vpx/vpx_codec.h"
+#include "vpx_ports/asm_offsets.h"
+
+BEGIN
+
+END
+
+/* add asserts for any offset that is not supported by assembly code */
+/* add asserts for any size that is not supported by assembly code */
diff --git a/libvpx/vp9/common/vp9_blockd.h b/libvpx/vp9/common/vp9_blockd.h
new file mode 100644
index 0000000..37d29af
--- /dev/null
+++ b/libvpx/vp9/common/vp9_blockd.h
@@ -0,0 +1,904 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_VP9_BLOCKD_H_
+#define VP9_COMMON_VP9_BLOCKD_H_
+
+#include "./vpx_config.h"
+#include "vpx_scale/yv12config.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/vp9_mv.h"
+#include "vp9/common/vp9_treecoder.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_enums.h"
+
+#define BLOCK_SIZE_GROUPS   4
+#define MAX_MB_SEGMENTS     8
+#define MB_SEG_TREE_PROBS   (MAX_MB_SEGMENTS-1)
+
+#define PREDICTION_PROBS 3
+
+#define MBSKIP_CONTEXTS 3
+
+#define MAX_REF_LF_DELTAS       4
+#define MAX_MODE_LF_DELTAS      2
+
+/* Segment Feature Masks */
+#define SEGMENT_DELTADATA   0
+#define SEGMENT_ABSDATA     1
+#define MAX_MV_REF_CANDIDATES 2
+
+#define INTRA_INTER_CONTEXTS 4
+#define COMP_INTER_CONTEXTS 5
+#define REF_CONTEXTS 5
+
+typedef enum {
+  PLANE_TYPE_Y_WITH_DC,
+  PLANE_TYPE_UV,
+} PLANE_TYPE;
+
+typedef char ENTROPY_CONTEXT;
+
+typedef char PARTITION_CONTEXT;
+
+static INLINE int combine_entropy_contexts(ENTROPY_CONTEXT a,
+                                           ENTROPY_CONTEXT b) {
+  return (a != 0) + (b != 0);
+}
+
+typedef enum {
+  KEY_FRAME = 0,
+  INTER_FRAME = 1,
+  NUM_FRAME_TYPES,
+} FRAME_TYPE;
+
+typedef enum {
+  EIGHTTAP_SMOOTH,
+  EIGHTTAP,
+  EIGHTTAP_SHARP,
+  BILINEAR,
+  SWITCHABLE  /* should be the last one */
+} INTERPOLATIONFILTERTYPE;
+
+typedef enum {
+  DC_PRED,         // Average of above and left pixels
+  V_PRED,          // Vertical
+  H_PRED,          // Horizontal
+  D45_PRED,        // Directional 45  deg = round(arctan(1/1) * 180/pi)
+  D135_PRED,       // Directional 135 deg = 180 - 45
+  D117_PRED,       // Directional 117 deg = 180 - 63
+  D153_PRED,       // Directional 153 deg = 180 - 27
+  D27_PRED,        // Directional 27  deg = round(arctan(1/2) * 180/pi)
+  D63_PRED,        // Directional 63  deg = round(arctan(2/1) * 180/pi)
+  TM_PRED,         // True-motion
+  NEARESTMV,
+  NEARMV,
+  ZEROMV,
+  NEWMV,
+  MB_MODE_COUNT
+} MB_PREDICTION_MODE;
+
+static INLINE int is_inter_mode(MB_PREDICTION_MODE mode) {
+  return mode >= NEARESTMV && mode <= NEWMV;
+}
+
+// Segment level features.
+typedef enum {
+  SEG_LVL_ALT_Q = 0,               // Use alternate Quantizer ....
+  SEG_LVL_ALT_LF = 1,              // Use alternate loop filter value...
+  SEG_LVL_REF_FRAME = 2,           // Optional Segment reference frame
+  SEG_LVL_SKIP = 3,                // Optional Segment (0,0) + skip mode
+  SEG_LVL_MAX = 4                  // Number of MB level features supported
+} SEG_LVL_FEATURES;
+
+// Segment level features.
+typedef enum {
+  TX_4X4 = 0,                      // 4x4 dct transform
+  TX_8X8 = 1,                      // 8x8 dct transform
+  TX_16X16 = 2,                    // 16x16 dct transform
+  TX_32X32 = 3,                    // 32x32 dct transform
+  TX_SIZE_MAX_SB,                  // Number of transforms available to SBs
+} TX_SIZE;
+
+typedef enum {
+  DCT_DCT   = 0,                      // DCT  in both horizontal and vertical
+  ADST_DCT  = 1,                      // ADST in vertical, DCT in horizontal
+  DCT_ADST  = 2,                      // DCT  in vertical, ADST in horizontal
+  ADST_ADST = 3                       // ADST in both directions
+} TX_TYPE;
+
+#define VP9_INTRA_MODES (TM_PRED + 1)
+
+#define VP9_INTER_MODES (1 + NEWMV - NEARESTMV)
+
+#define WHT_UPSCALE_FACTOR 2
+
+#define TX_SIZE_PROBS  6  // (TX_SIZE_MAX_SB * (TX_SIZE_MAX_SB - 1) / 2)
+
+#define get_tx_probs(c, b) ((b) < BLOCK_SIZE_MB16X16 ? \
+                            (c)->fc.tx_probs_8x8p :    \
+                            (b) < BLOCK_SIZE_SB32X32 ? \
+                            (c)->fc.tx_probs_16x16p : (c)->fc.tx_probs_32x32p)
+
+/* For keyframes, intra block modes are predicted by the (already decoded)
+   modes for the Y blocks to the left and above us; for interframes, there
+   is a single probability table. */
+
+union b_mode_info {
+  struct {
+    MB_PREDICTION_MODE first;
+  } as_mode;
+  int_mv as_mv[2];  // first, second inter predictor motion vectors
+};
+
+typedef enum {
+  NONE = -1,
+  INTRA_FRAME = 0,
+  LAST_FRAME = 1,
+  GOLDEN_FRAME = 2,
+  ALTREF_FRAME = 3,
+  MAX_REF_FRAMES = 4
+} MV_REFERENCE_FRAME;
+
+static INLINE int b_width_log2(BLOCK_SIZE_TYPE sb_type) {
+  switch (sb_type) {
+    case BLOCK_SIZE_SB4X8:
+    case BLOCK_SIZE_AB4X4: return 0;
+    case BLOCK_SIZE_SB8X4:
+    case BLOCK_SIZE_SB8X8:
+    case BLOCK_SIZE_SB8X16: return 1;
+    case BLOCK_SIZE_SB16X8:
+    case BLOCK_SIZE_MB16X16:
+    case BLOCK_SIZE_SB16X32: return 2;
+    case BLOCK_SIZE_SB32X16:
+    case BLOCK_SIZE_SB32X32:
+    case BLOCK_SIZE_SB32X64: return 3;
+    case BLOCK_SIZE_SB64X32:
+    case BLOCK_SIZE_SB64X64: return 4;
+    default: assert(0);
+      return -1;
+  }
+}
+
+static INLINE int b_height_log2(BLOCK_SIZE_TYPE sb_type) {
+  switch (sb_type) {
+    case BLOCK_SIZE_SB8X4:
+    case BLOCK_SIZE_AB4X4: return 0;
+    case BLOCK_SIZE_SB4X8:
+    case BLOCK_SIZE_SB8X8:
+    case BLOCK_SIZE_SB16X8: return 1;
+    case BLOCK_SIZE_SB8X16:
+    case BLOCK_SIZE_MB16X16:
+    case BLOCK_SIZE_SB32X16: return 2;
+    case BLOCK_SIZE_SB16X32:
+    case BLOCK_SIZE_SB32X32:
+    case BLOCK_SIZE_SB64X32: return 3;
+    case BLOCK_SIZE_SB32X64:
+    case BLOCK_SIZE_SB64X64: return 4;
+    default: assert(0);
+      return -1;
+  }
+}
+
+static INLINE int mi_width_log2(BLOCK_SIZE_TYPE sb_type) {
+  int a = b_width_log2(sb_type) - 1;
+  // align 4x4 block to mode_info
+  if (a < 0)
+    a = 0;
+  assert(a >= 0);
+  return a;
+}
+
+static INLINE int mi_height_log2(BLOCK_SIZE_TYPE sb_type) {
+  int a = b_height_log2(sb_type) - 1;
+  if (a < 0)
+    a = 0;
+  assert(a >= 0);
+  return a;
+}
+
+typedef struct {
+  MB_PREDICTION_MODE mode, uv_mode;
+  MV_REFERENCE_FRAME ref_frame[2];
+  TX_SIZE txfm_size;
+  int_mv mv[2]; // for each reference frame used
+  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
+  int_mv best_mv, best_second_mv;
+
+  int mb_mode_context[MAX_REF_FRAMES];
+
+  unsigned char mb_skip_coeff;                                /* does this mb has coefficients at all, 1=no coefficients, 0=need decode tokens */
+  unsigned char segment_id;           // Segment id for current frame
+
+  // Flags used for prediction status of various bistream signals
+  unsigned char seg_id_predicted;
+
+  // Indicates if the mb is part of the image (1) vs border (0)
+  // This can be useful in determining whether the MB provides
+  // a valid predictor
+  unsigned char mb_in_image;
+
+  INTERPOLATIONFILTERTYPE interp_filter;
+
+  BLOCK_SIZE_TYPE sb_type;
+} MB_MODE_INFO;
+
+typedef struct {
+  MB_MODE_INFO mbmi;
+  union b_mode_info bmi[4];
+} MODE_INFO;
+
+#define VP9_REF_SCALE_SHIFT 14
+struct scale_factors {
+  int x_scale_fp;   // horizontal fixed point scale factor
+  int y_scale_fp;   // vertical fixed point scale factor
+  int x_offset_q4;
+  int x_step_q4;
+  int y_offset_q4;
+  int y_step_q4;
+
+  int (*scale_value_x)(int val, const struct scale_factors *scale);
+  int (*scale_value_y)(int val, const struct scale_factors *scale);
+  void (*set_scaled_offsets)(struct scale_factors *scale, int row, int col);
+  int_mv32 (*scale_mv_q3_to_q4)(const int_mv *src_mv,
+                                const struct scale_factors *scale);
+  int32_t (*scale_mv_component_q4)(int mv_q4, int scale_fp, int offset_q4);
+
+  convolve_fn_t predict[2][2][2];  // horiz, vert, avg
+};
+
+#if CONFIG_ALPHA
+enum { MAX_MB_PLANE = 4 };
+#else
+enum { MAX_MB_PLANE = 3 };
+#endif
+
+struct buf_2d {
+  uint8_t *buf;
+  int stride;
+};
+
+struct macroblockd_plane {
+  DECLARE_ALIGNED(16, int16_t,  qcoeff[64 * 64]);
+  DECLARE_ALIGNED(16, int16_t,  dqcoeff[64 * 64]);
+  DECLARE_ALIGNED(16, uint16_t, eobs[256]);
+  PLANE_TYPE plane_type;
+  int subsampling_x;
+  int subsampling_y;
+  struct buf_2d dst;
+  struct buf_2d pre[2];
+  int16_t *dequant;
+  ENTROPY_CONTEXT *above_context;
+  ENTROPY_CONTEXT *left_context;
+};
+
+#define BLOCK_OFFSET(x, i, n) ((x) + (i) * (n))
+
+typedef struct macroblockd {
+  struct macroblockd_plane plane[MAX_MB_PLANE];
+
+  struct scale_factors scale_factor[2];
+  struct scale_factors scale_factor_uv[2];
+
+  MODE_INFO *prev_mode_info_context;
+  MODE_INFO *mode_info_context;
+  int mode_info_stride;
+
+  FRAME_TYPE frame_type;
+
+  int up_available;
+  int left_available;
+  int right_available;
+
+  // partition contexts
+  PARTITION_CONTEXT *above_seg_context;
+  PARTITION_CONTEXT *left_seg_context;
+
+  /* 0 (disable) 1 (enable) segmentation */
+  unsigned char segmentation_enabled;
+
+  /* 0 (do not update) 1 (update) the macroblock segmentation map. */
+  unsigned char update_mb_segmentation_map;
+
+  /* 0 (do not update) 1 (update) the macroblock segmentation feature data. */
+  unsigned char update_mb_segmentation_data;
+
+  /* 0 (do not update) 1 (update) the macroblock segmentation feature data. */
+  unsigned char mb_segment_abs_delta;
+
+  /* Per frame flags that define which MB level features (such as quantizer or loop filter level) */
+  /* are enabled and when enabled the proabilities used to decode the per MB flags in MB_MODE_INFO */
+
+  // Probability Tree used to code Segment number
+  vp9_prob mb_segment_tree_probs[MB_SEG_TREE_PROBS];
+
+  // Segment features
+  int16_t segment_feature_data[MAX_MB_SEGMENTS][SEG_LVL_MAX];
+  unsigned int segment_feature_mask[MAX_MB_SEGMENTS];
+
+  /* mode_based Loop filter adjustment */
+  unsigned char mode_ref_lf_delta_enabled;
+  unsigned char mode_ref_lf_delta_update;
+
+  /* Delta values have the range +/- MAX_LOOP_FILTER */
+  /* 0 = Intra, Last, GF, ARF */
+  signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
+  /* 0 = Intra, Last, GF, ARF */
+  signed char ref_lf_deltas[MAX_REF_LF_DELTAS];
+  /* 0 = ZERO_MV, MV */
+  signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
+  /* 0 = ZERO_MV, MV */
+  signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];
+
+  /* Distance of MB away from frame edges */
+  int mb_to_left_edge;
+  int mb_to_right_edge;
+  int mb_to_top_edge;
+  int mb_to_bottom_edge;
+
+  unsigned int frames_since_golden;
+  unsigned int frames_till_alt_ref_frame;
+
+  int lossless;
+  /* Inverse transform function pointers. */
+  void (*inv_txm4x4_1_add)(int16_t *input, uint8_t *dest, int stride);
+  void (*inv_txm4x4_add)(int16_t *input, uint8_t *dest, int stride);
+  void (*itxm_add)(int16_t *input, uint8_t *dest, int stride, int eob);
+
+  struct subpix_fn_table  subpix;
+
+  int allow_high_precision_mv;
+
+  int corrupted;
+
+  int sb_index;   // index of 32x32 block inside the 64x64 block
+  int mb_index;   // index of 16x16 block inside the 32x32 block
+  int b_index;    // index of 8x8 block inside the 16x16 block
+  int ab_index;   // index of 4x4 block inside the 8x8 block
+  int q_index;
+
+} MACROBLOCKD;
+
+static int *get_sb_index(MACROBLOCKD *xd, BLOCK_SIZE_TYPE subsize) {
+  switch (subsize) {
+    case BLOCK_SIZE_SB64X64:
+    case BLOCK_SIZE_SB64X32:
+    case BLOCK_SIZE_SB32X64:
+    case BLOCK_SIZE_SB32X32:
+      return &xd->sb_index;
+    case BLOCK_SIZE_SB32X16:
+    case BLOCK_SIZE_SB16X32:
+    case BLOCK_SIZE_MB16X16:
+      return &xd->mb_index;
+    case BLOCK_SIZE_SB16X8:
+    case BLOCK_SIZE_SB8X16:
+    case BLOCK_SIZE_SB8X8:
+      return &xd->b_index;
+    case BLOCK_SIZE_SB8X4:
+    case BLOCK_SIZE_SB4X8:
+    case BLOCK_SIZE_AB4X4:
+      return &xd->ab_index;
+    default:
+      assert(0);
+      return NULL;
+  }
+}
+
+static INLINE void update_partition_context(MACROBLOCKD *xd,
+                                            BLOCK_SIZE_TYPE sb_type,
+                                            BLOCK_SIZE_TYPE sb_size) {
+  int bsl = b_width_log2(sb_size), bs = (1 << bsl) / 2;
+  int bwl = b_width_log2(sb_type);
+  int bhl = b_height_log2(sb_type);
+  int boffset = b_width_log2(BLOCK_SIZE_SB64X64) - bsl;
+  int i;
+
+  // update the partition context at the end notes. set partition bits
+  // of block sizes larger than the current one to be one, and partition
+  // bits of smaller block sizes to be zero.
+  if ((bwl == bsl) && (bhl == bsl)) {
+    for (i = 0; i < bs; i++)
+      xd->left_seg_context[i] = ~(0xf << boffset);
+    for (i = 0; i < bs; i++)
+      xd->above_seg_context[i] = ~(0xf << boffset);
+  } else if ((bwl == bsl) && (bhl < bsl)) {
+    for (i = 0; i < bs; i++)
+      xd->left_seg_context[i] = ~(0xe << boffset);
+    for (i = 0; i < bs; i++)
+      xd->above_seg_context[i] = ~(0xf << boffset);
+  }  else if ((bwl < bsl) && (bhl == bsl)) {
+    for (i = 0; i < bs; i++)
+      xd->left_seg_context[i] = ~(0xf << boffset);
+    for (i = 0; i < bs; i++)
+      xd->above_seg_context[i] = ~(0xe << boffset);
+  } else if ((bwl < bsl) && (bhl < bsl)) {
+    for (i = 0; i < bs; i++)
+      xd->left_seg_context[i] = ~(0xe << boffset);
+    for (i = 0; i < bs; i++)
+      xd->above_seg_context[i] = ~(0xe << boffset);
+  } else {
+    assert(0);
+  }
+}
+
+static INLINE int partition_plane_context(MACROBLOCKD *xd,
+                                          BLOCK_SIZE_TYPE sb_type) {
+  int bsl = mi_width_log2(sb_type), bs = 1 << bsl;
+  int above = 0, left = 0, i;
+  int boffset = mi_width_log2(BLOCK_SIZE_SB64X64) - bsl;
+
+  assert(mi_width_log2(sb_type) == mi_height_log2(sb_type));
+  assert(bsl >= 0);
+  assert(boffset >= 0);
+
+  for (i = 0; i < bs; i++)
+    above |= (xd->above_seg_context[i] & (1 << boffset));
+  for (i = 0; i < bs; i++)
+    left |= (xd->left_seg_context[i] & (1 << boffset));
+
+  above = (above > 0);
+  left  = (left > 0);
+
+  return (left * 2 + above) + bsl * PARTITION_PLOFFSET;
+}
+
+static BLOCK_SIZE_TYPE get_subsize(BLOCK_SIZE_TYPE bsize,
+                                   PARTITION_TYPE partition) {
+  BLOCK_SIZE_TYPE subsize;
+  switch (partition) {
+    case PARTITION_NONE:
+      subsize = bsize;
+      break;
+    case PARTITION_HORZ:
+      if (bsize == BLOCK_SIZE_SB64X64)
+        subsize = BLOCK_SIZE_SB64X32;
+      else if (bsize == BLOCK_SIZE_SB32X32)
+        subsize = BLOCK_SIZE_SB32X16;
+      else if (bsize == BLOCK_SIZE_MB16X16)
+        subsize = BLOCK_SIZE_SB16X8;
+      else if (bsize == BLOCK_SIZE_SB8X8)
+        subsize = BLOCK_SIZE_SB8X4;
+      else
+        assert(0);
+      break;
+    case PARTITION_VERT:
+      if (bsize == BLOCK_SIZE_SB64X64)
+        subsize = BLOCK_SIZE_SB32X64;
+      else if (bsize == BLOCK_SIZE_SB32X32)
+        subsize = BLOCK_SIZE_SB16X32;
+      else if (bsize == BLOCK_SIZE_MB16X16)
+        subsize = BLOCK_SIZE_SB8X16;
+      else if (bsize == BLOCK_SIZE_SB8X8)
+        subsize = BLOCK_SIZE_SB4X8;
+      else
+        assert(0);
+      break;
+    case PARTITION_SPLIT:
+      if (bsize == BLOCK_SIZE_SB64X64)
+        subsize = BLOCK_SIZE_SB32X32;
+      else if (bsize == BLOCK_SIZE_SB32X32)
+        subsize = BLOCK_SIZE_MB16X16;
+      else if (bsize == BLOCK_SIZE_MB16X16)
+        subsize = BLOCK_SIZE_SB8X8;
+      else if (bsize == BLOCK_SIZE_SB8X8)
+        subsize = BLOCK_SIZE_AB4X4;
+      else
+        assert(0);
+      break;
+    default:
+      assert(0);
+  }
+  return subsize;
+}
+
+// transform mapping
+static TX_TYPE txfm_map(MB_PREDICTION_MODE bmode) {
+  switch (bmode) {
+    case TM_PRED :
+    case D135_PRED :
+      return ADST_ADST;
+
+    case V_PRED :
+    case D117_PRED :
+    case D63_PRED:
+      return ADST_DCT;
+
+    case H_PRED :
+    case D153_PRED :
+    case D27_PRED :
+      return DCT_ADST;
+
+    default:
+      return DCT_DCT;
+  }
+}
+
+static TX_TYPE get_tx_type_4x4(const MACROBLOCKD *xd, int ib) {
+  TX_TYPE tx_type;
+  MODE_INFO *mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+  if (xd->lossless || mbmi->ref_frame[0] != INTRA_FRAME)
+    return DCT_DCT;
+  if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+    tx_type = txfm_map(mi->bmi[ib].as_mode.first);
+  } else {
+    assert(mbmi->mode <= TM_PRED);
+    tx_type = txfm_map(mbmi->mode);
+  }
+  return tx_type;
+}
+
+static TX_TYPE get_tx_type_8x8(const MACROBLOCKD *xd, int ib) {
+  TX_TYPE tx_type = DCT_DCT;
+  if (xd->mode_info_context->mbmi.mode <= TM_PRED) {
+    tx_type = txfm_map(xd->mode_info_context->mbmi.mode);
+  }
+  return tx_type;
+}
+
+static TX_TYPE get_tx_type_16x16(const MACROBLOCKD *xd, int ib) {
+  TX_TYPE tx_type = DCT_DCT;
+  if (xd->mode_info_context->mbmi.mode <= TM_PRED) {
+    tx_type = txfm_map(xd->mode_info_context->mbmi.mode);
+  }
+  return tx_type;
+}
+
+void vp9_setup_block_dptrs(MACROBLOCKD *xd,
+                           int subsampling_x, int subsampling_y);
+
+static TX_SIZE get_uv_tx_size(const MB_MODE_INFO *mbmi) {
+  const TX_SIZE size = mbmi->txfm_size;
+
+  switch (mbmi->sb_type) {
+    case BLOCK_SIZE_SB64X64:
+      return size;
+    case BLOCK_SIZE_SB64X32:
+    case BLOCK_SIZE_SB32X64:
+    case BLOCK_SIZE_SB32X32:
+      if (size == TX_32X32)
+        return TX_16X16;
+      else
+        return size;
+    case BLOCK_SIZE_SB32X16:
+    case BLOCK_SIZE_SB16X32:
+    case BLOCK_SIZE_MB16X16:
+      if (size == TX_16X16)
+        return TX_8X8;
+      else
+        return size;
+    default:
+      return TX_4X4;
+  }
+
+  return size;
+}
+
+struct plane_block_idx {
+  int plane;
+  int block;
+};
+
+// TODO(jkoleszar): returning a struct so it can be used in a const context,
+// expect to refactor this further later.
+static INLINE struct plane_block_idx plane_block_idx(int y_blocks,
+                                                     int b_idx) {
+  const int v_offset = y_blocks * 5 / 4;
+  struct plane_block_idx res;
+
+  if (b_idx < y_blocks) {
+    res.plane = 0;
+    res.block = b_idx;
+  } else if (b_idx < v_offset) {
+    res.plane = 1;
+    res.block = b_idx - y_blocks;
+  } else {
+    assert(b_idx < y_blocks * 3 / 2);
+    res.plane = 2;
+    res.block = b_idx - v_offset;
+  }
+  return res;
+}
+
+static INLINE int plane_block_width(BLOCK_SIZE_TYPE bsize,
+                                    const struct macroblockd_plane* plane) {
+  return 4 << (b_width_log2(bsize) - plane->subsampling_x);
+}
+
+static INLINE int plane_block_height(BLOCK_SIZE_TYPE bsize,
+                                     const struct macroblockd_plane* plane) {
+  return 4 << (b_height_log2(bsize) - plane->subsampling_y);
+}
+
+typedef void (*foreach_transformed_block_visitor)(int plane, int block,
+                                                  BLOCK_SIZE_TYPE bsize,
+                                                  int ss_txfrm_size,
+                                                  void *arg);
+
+static INLINE void foreach_transformed_block_in_plane(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize, int plane,
+    foreach_transformed_block_visitor visit, void *arg) {
+  const int bw = b_width_log2(bsize), bh = b_height_log2(bsize);
+
+  // block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
+  // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
+  // transform size varies per plane, look it up in a common way.
+  const MB_MODE_INFO* mbmi = &xd->mode_info_context->mbmi;
+  const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi)
+                                : mbmi->txfm_size;
+  const int block_size_b = bw + bh;
+  const int txfrm_size_b = tx_size * 2;
+
+  // subsampled size of the block
+  const int ss_sum = xd->plane[plane].subsampling_x
+      + xd->plane[plane].subsampling_y;
+  const int ss_block_size = block_size_b - ss_sum;
+
+  const int step = 1 << txfrm_size_b;
+
+  int i;
+
+  assert(txfrm_size_b <= block_size_b);
+  assert(txfrm_size_b <= ss_block_size);
+
+  // If mb_to_right_edge is < 0 we are in a situation in which
+  // the current block size extends into the UMV and we won't
+  // visit the sub blocks that are wholly within the UMV.
+  if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
+    int r, c;
+    const int sw = bw - xd->plane[plane].subsampling_x;
+    const int sh = bh - xd->plane[plane].subsampling_y;
+    int max_blocks_wide = 1 << sw;
+    int max_blocks_high = 1 << sh;
+
+    // xd->mb_to_right_edge is in units of pixels * 8.  This converts
+    // it to 4x4 block sizes.
+    if (xd->mb_to_right_edge < 0)
+      max_blocks_wide +=
+          + (xd->mb_to_right_edge >> (5 + xd->plane[plane].subsampling_x));
+
+    if (xd->mb_to_bottom_edge < 0)
+      max_blocks_high +=
+          + (xd->mb_to_bottom_edge >> (5 + xd->plane[plane].subsampling_y));
+
+    i = 0;
+    // Unlike the normal case - in here we have to keep track of the
+    // row and column of the blocks we use so that we know if we are in
+    // the unrestricted motion border..
+    for (r = 0; r < (1 << sh); r += (1 << tx_size)) {
+      for (c = 0; c < (1 << sw); c += (1 << tx_size)) {
+        if (r < max_blocks_high && c < max_blocks_wide)
+          visit(plane, i, bsize, txfrm_size_b, arg);
+        i += step;
+      }
+    }
+  } else {
+    for (i = 0; i < (1 << ss_block_size); i += step) {
+      visit(plane, i, bsize, txfrm_size_b, arg);
+    }
+  }
+}
+
+static INLINE void foreach_transformed_block(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize,
+    foreach_transformed_block_visitor visit, void *arg) {
+  int plane;
+
+  for (plane = 0; plane < MAX_MB_PLANE; plane++) {
+    foreach_transformed_block_in_plane(xd, bsize, plane,
+                                       visit, arg);
+  }
+}
+
+static INLINE void foreach_transformed_block_uv(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize,
+    foreach_transformed_block_visitor visit, void *arg) {
+  int plane;
+
+  for (plane = 1; plane < MAX_MB_PLANE; plane++) {
+    foreach_transformed_block_in_plane(xd, bsize, plane,
+                                       visit, arg);
+  }
+}
+
+// TODO(jkoleszar): In principle, pred_w, pred_h are unnecessary, as we could
+// calculate the subsampled BLOCK_SIZE_TYPE, but that type isn't defined for
+// sizes smaller than 16x16 yet.
+typedef void (*foreach_predicted_block_visitor)(int plane, int block,
+                                                BLOCK_SIZE_TYPE bsize,
+                                                int pred_w, int pred_h,
+                                                void *arg);
+static INLINE void foreach_predicted_block_in_plane(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize, int plane,
+    foreach_predicted_block_visitor visit, void *arg) {
+  int i, x, y;
+
+  // block sizes in number of 4x4 blocks log 2 ("*_b")
+  // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
+  // subsampled size of the block
+  const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int bhl = b_height_log2(bsize) - xd->plane[plane].subsampling_y;
+
+  // size of the predictor to use.
+  int pred_w, pred_h;
+
+  if (xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+    assert(bsize == BLOCK_SIZE_SB8X8);
+    pred_w = 0;
+    pred_h = 0;
+  } else {
+    pred_w = bwl;
+    pred_h = bhl;
+  }
+  assert(pred_w <= bwl);
+  assert(pred_h <= bhl);
+
+  // visit each subblock in raster order
+  i = 0;
+  for (y = 0; y < 1 << bhl; y += 1 << pred_h) {
+    for (x = 0; x < 1 << bwl; x += 1 << pred_w) {
+      visit(plane, i, bsize, pred_w, pred_h, arg);
+      i += 1 << pred_w;
+    }
+    i += (1 << (bwl + pred_h)) - (1 << bwl);
+  }
+}
+static INLINE void foreach_predicted_block(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize,
+    foreach_predicted_block_visitor visit, void *arg) {
+  int plane;
+
+  for (plane = 0; plane < MAX_MB_PLANE; plane++) {
+    foreach_predicted_block_in_plane(xd, bsize, plane, visit, arg);
+  }
+}
+static INLINE void foreach_predicted_block_uv(
+    const MACROBLOCKD* const xd, BLOCK_SIZE_TYPE bsize,
+    foreach_predicted_block_visitor visit, void *arg) {
+  int plane;
+
+  for (plane = 1; plane < MAX_MB_PLANE; plane++) {
+    foreach_predicted_block_in_plane(xd, bsize, plane, visit, arg);
+  }
+}
+static int raster_block_offset(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize,
+                               int plane, int block, int stride) {
+  const int bw = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int y = 4 * (block >> bw), x = 4 * (block & ((1 << bw) - 1));
+  return y * stride + x;
+}
+static int16_t* raster_block_offset_int16(MACROBLOCKD *xd,
+                                         BLOCK_SIZE_TYPE bsize,
+                                         int plane, int block, int16_t *base) {
+  const int stride = plane_block_width(bsize, &xd->plane[plane]);
+  return base + raster_block_offset(xd, bsize, plane, block, stride);
+}
+static uint8_t* raster_block_offset_uint8(MACROBLOCKD *xd,
+                                         BLOCK_SIZE_TYPE bsize,
+                                         int plane, int block,
+                                         uint8_t *base, int stride) {
+  return base + raster_block_offset(xd, bsize, plane, block, stride);
+}
+
+static int txfrm_block_to_raster_block(MACROBLOCKD *xd,
+                                       BLOCK_SIZE_TYPE bsize,
+                                       int plane, int block,
+                                       int ss_txfrm_size) {
+  const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int txwl = ss_txfrm_size / 2;
+  const int tx_cols_lg2 = bwl - txwl;
+  const int tx_cols = 1 << tx_cols_lg2;
+  const int raster_mb = block >> ss_txfrm_size;
+  const int x = (raster_mb & (tx_cols - 1)) << (txwl);
+  const int y = raster_mb >> tx_cols_lg2 << (txwl);
+  return x + (y << bwl);
+}
+
+static void txfrm_block_to_raster_xy(MACROBLOCKD *xd,
+                                     BLOCK_SIZE_TYPE bsize,
+                                     int plane, int block,
+                                     int ss_txfrm_size,
+                                     int *x, int *y) {
+  const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int txwl = ss_txfrm_size / 2;
+  const int tx_cols_lg2 = bwl - txwl;
+  const int tx_cols = 1 << tx_cols_lg2;
+  const int raster_mb = block >> ss_txfrm_size;
+  *x = (raster_mb & (tx_cols - 1)) << (txwl);
+  *y = raster_mb >> tx_cols_lg2 << (txwl);
+}
+
+static void extend_for_intra(MACROBLOCKD* const xd, int plane, int block,
+                             BLOCK_SIZE_TYPE bsize, int ss_txfrm_size) {
+  const int bw = plane_block_width(bsize, &xd->plane[plane]);
+  const int bh = plane_block_height(bsize, &xd->plane[plane]);
+  int x, y;
+  txfrm_block_to_raster_xy(xd, bsize, plane, block, ss_txfrm_size, &x, &y);
+  x = x * 4 - 1;
+  y = y * 4 - 1;
+  // Copy a pixel into the umv if we are in a situation where the block size
+  // extends into the UMV.
+  // TODO(JBB): Should be able to do the full extend in place so we don't have
+  // to do this multiple times.
+  if (xd->mb_to_right_edge < 0) {
+    int umv_border_start = bw
+        + (xd->mb_to_right_edge >> (3 + xd->plane[plane].subsampling_x));
+
+    if (x + bw > umv_border_start)
+      vpx_memset(
+          xd->plane[plane].dst.buf + y * xd->plane[plane].dst.stride
+              + umv_border_start,
+          *(xd->plane[plane].dst.buf + y * xd->plane[plane].dst.stride
+              + umv_border_start - 1),
+          bw);
+  }
+  if (xd->mb_to_bottom_edge < 0) {
+    int umv_border_start = bh
+        + (xd->mb_to_bottom_edge >> (3 + xd->plane[plane].subsampling_y));
+    int i;
+    uint8_t c = *(xd->plane[plane].dst.buf
+        + (umv_border_start - 1) * xd->plane[plane].dst.stride + x);
+
+    uint8_t *d = xd->plane[plane].dst.buf
+        + umv_border_start * xd->plane[plane].dst.stride + x;
+
+    if (y + bh > umv_border_start)
+      for (i = 0; i < bh; i++, d += xd->plane[plane].dst.stride)
+        *d = c;
+  }
+}
+static void set_contexts_on_border(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize,
+                                   int plane, int ss_tx_size, int eob, int aoff,
+                                   int loff, ENTROPY_CONTEXT *A,
+                                   ENTROPY_CONTEXT *L) {
+  const int bw = b_width_log2(bsize), bh = b_height_log2(bsize);
+  const int sw = bw - xd->plane[plane].subsampling_x;
+  const int sh = bh - xd->plane[plane].subsampling_y;
+  int mi_blocks_wide = 1 << sw;
+  int mi_blocks_high = 1 << sh;
+  int tx_size_in_blocks = (1 << ss_tx_size);
+  int above_contexts = tx_size_in_blocks;
+  int left_contexts = tx_size_in_blocks;
+  int pt;
+
+  // xd->mb_to_right_edge is in units of pixels * 8.  This converts
+  // it to 4x4 block sizes.
+  if (xd->mb_to_right_edge < 0) {
+    mi_blocks_wide += (xd->mb_to_right_edge
+        >> (5 + xd->plane[plane].subsampling_x));
+  }
+
+  // this code attempts to avoid copying into contexts that are outside
+  // our border.  Any blocks that do are set to 0...
+  if (above_contexts + aoff > mi_blocks_wide)
+    above_contexts = mi_blocks_wide - aoff;
+
+  if (xd->mb_to_bottom_edge < 0) {
+    mi_blocks_high += (xd->mb_to_bottom_edge
+        >> (5 + xd->plane[plane].subsampling_y));
+  }
+  if (left_contexts + loff > mi_blocks_high) {
+    left_contexts = mi_blocks_high - loff;
+  }
+
+  for (pt = 0; pt < above_contexts; pt++)
+    A[pt] = eob > 0;
+  for (pt = above_contexts; pt < (1 << ss_tx_size); pt++)
+    A[pt] = 0;
+  for (pt = 0; pt < left_contexts; pt++)
+    L[pt] = eob > 0;
+  for (pt = left_contexts; pt < (1 << ss_tx_size); pt++)
+    L[pt] = 0;
+}
+
+
+#endif  // VP9_COMMON_VP9_BLOCKD_H_
diff --git a/libvpx/vp9/common/vp9_common.h b/libvpx/vp9/common/vp9_common.h
new file mode 100644
index 0000000..0d7babf
--- /dev/null
+++ b/libvpx/vp9/common/vp9_common.h
@@ -0,0 +1,68 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_COMMON_H_
+#define VP9_COMMON_VP9_COMMON_H_
+
+/* Interface header for common constant data structures and lookup tables */
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx/vpx_integer.h"
+
+#define MIN(x, y) (((x) < (y)) ? (x) : (y))
+#define MAX(x, y) (((x) > (y)) ? (x) : (y))
+
+#define ROUND_POWER_OF_TWO(value, n) (((value) + (1 << ((n) - 1))) >> (n))
+
+/* If we don't want to use ROUND_POWER_OF_TWO macro
+static INLINE int16_t round_power_of_two(int16_t value, int n) {
+  return (value + (1 << (n - 1))) >> n;
+}*/
+
+// Only need this for fixed-size arrays, for structs just assign.
+#define vp9_copy(dest, src) {            \
+    assert(sizeof(dest) == sizeof(src)); \
+    vpx_memcpy(dest, src, sizeof(src));  \
+  }
+
+// Use this for variably-sized arrays.
+#define vp9_copy_array(dest, src, n) {       \
+    assert(sizeof(*dest) == sizeof(*src));   \
+    vpx_memcpy(dest, src, n * sizeof(*src)); \
+  }
+
+#define vp9_zero(dest) vpx_memset(&dest, 0, sizeof(dest));
+#define vp9_zero_array(dest, n) vpx_memset(dest, 0, n * sizeof(*dest));
+
+static INLINE uint8_t clip_pixel(int val) {
+  return (val > 255) ? 255u : (val < 0) ? 0u : val;
+}
+
+static INLINE int clamp(int value, int low, int high) {
+  return value < low ? low : (value > high ? high : value);
+}
+
+static INLINE double fclamp(double value, double low, double high) {
+  return value < low ? low : (value > high ? high : value);
+}
+
+static INLINE int multiple8(int value) {
+  return (value + 7) & ~7;
+}
+
+#define SYNC_CODE_0 0x49
+#define SYNC_CODE_1 0x83
+#define SYNC_CODE_2 0x42
+
+
+#endif  // VP9_COMMON_VP9_COMMON_H_
diff --git a/libvpx/vp9/common/vp9_convolve.c b/libvpx/vp9/common/vp9_convolve.c
new file mode 100644
index 0000000..46ae503
--- /dev/null
+++ b/libvpx/vp9/common/vp9_convolve.c
@@ -0,0 +1,376 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "vp9/common/vp9_convolve.h"
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+#define VP9_FILTER_WEIGHT 128
+#define VP9_FILTER_SHIFT  7
+
+/* Assume a bank of 16 filters to choose from. There are two implementations
+ * for filter wrapping behavior, since we want to be able to pick which filter
+ * to start with. We could either:
+ *
+ * 1) make filter_ a pointer to the base of the filter array, and then add an
+ *    additional offset parameter, to choose the starting filter.
+ * 2) use a pointer to 2 periods worth of filters, so that even if the original
+ *    phase offset is at 15/16, we'll have valid data to read. The filter
+ *    tables become [32][8], and the second half is duplicated.
+ * 3) fix the alignment of the filter tables, so that we know the 0/16 is
+ *    always 256 byte aligned.
+ *
+ * Implementations 2 and 3 are likely preferable, as they avoid an extra 2
+ * parameters, and switching between them is trivial, with the
+ * ALIGN_FILTERS_256 macro, below.
+ */
+ #define ALIGN_FILTERS_256 1
+
+static void convolve_horiz_c(const uint8_t *src, int src_stride,
+                             uint8_t *dst, int dst_stride,
+                             const int16_t *filter_x0, int x_step_q4,
+                             const int16_t *filter_y, int y_step_q4,
+                             int w, int h, int taps) {
+  int x, y, k, sum;
+  const int16_t *filter_x_base = filter_x0;
+
+#if ALIGN_FILTERS_256
+  filter_x_base = (const int16_t *)(((intptr_t)filter_x0) & ~(intptr_t)0xff);
+#endif
+
+  /* Adjust base pointer address for this source line */
+  src -= taps / 2 - 1;
+
+  for (y = 0; y < h; ++y) {
+    /* Pointer to filter to use */
+    const int16_t *filter_x = filter_x0;
+
+    /* Initial phase offset */
+    int x0_q4 = (filter_x - filter_x_base) / taps;
+    int x_q4 = x0_q4;
+
+    for (x = 0; x < w; ++x) {
+      /* Per-pixel src offset */
+      int src_x = (x_q4 - x0_q4) >> 4;
+
+      for (sum = 0, k = 0; k < taps; ++k) {
+        sum += src[src_x + k] * filter_x[k];
+      }
+      sum += (VP9_FILTER_WEIGHT >> 1);
+      dst[x] = clip_pixel(sum >> VP9_FILTER_SHIFT);
+
+      /* Adjust source and filter to use for the next pixel */
+      x_q4 += x_step_q4;
+      filter_x = filter_x_base + (x_q4 & 0xf) * taps;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void convolve_avg_horiz_c(const uint8_t *src, int src_stride,
+                                 uint8_t *dst, int dst_stride,
+                                 const int16_t *filter_x0, int x_step_q4,
+                                 const int16_t *filter_y, int y_step_q4,
+                                 int w, int h, int taps) {
+  int x, y, k, sum;
+  const int16_t *filter_x_base = filter_x0;
+
+#if ALIGN_FILTERS_256
+  filter_x_base = (const int16_t *)(((intptr_t)filter_x0) & ~(intptr_t)0xff);
+#endif
+
+  /* Adjust base pointer address for this source line */
+  src -= taps / 2 - 1;
+
+  for (y = 0; y < h; ++y) {
+    /* Pointer to filter to use */
+    const int16_t *filter_x = filter_x0;
+
+    /* Initial phase offset */
+    int x0_q4 = (filter_x - filter_x_base) / taps;
+    int x_q4 = x0_q4;
+
+    for (x = 0; x < w; ++x) {
+      /* Per-pixel src offset */
+      int src_x = (x_q4 - x0_q4) >> 4;
+
+      for (sum = 0, k = 0; k < taps; ++k) {
+        sum += src[src_x + k] * filter_x[k];
+      }
+      sum += (VP9_FILTER_WEIGHT >> 1);
+      dst[x] = (dst[x] + clip_pixel(sum >> VP9_FILTER_SHIFT) + 1) >> 1;
+
+      /* Adjust source and filter to use for the next pixel */
+      x_q4 += x_step_q4;
+      filter_x = filter_x_base + (x_q4 & 0xf) * taps;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void convolve_vert_c(const uint8_t *src, int src_stride,
+                            uint8_t *dst, int dst_stride,
+                            const int16_t *filter_x, int x_step_q4,
+                            const int16_t *filter_y0, int y_step_q4,
+                            int w, int h, int taps) {
+  int x, y, k, sum;
+
+  const int16_t *filter_y_base = filter_y0;
+
+#if ALIGN_FILTERS_256
+  filter_y_base = (const int16_t *)(((intptr_t)filter_y0) & ~(intptr_t)0xff);
+#endif
+
+  /* Adjust base pointer address for this source column */
+  src -= src_stride * (taps / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    /* Pointer to filter to use */
+    const int16_t *filter_y = filter_y0;
+
+    /* Initial phase offset */
+    int y0_q4 = (filter_y - filter_y_base) / taps;
+    int y_q4 = y0_q4;
+
+    for (y = 0; y < h; ++y) {
+      /* Per-pixel src offset */
+      int src_y = (y_q4 - y0_q4) >> 4;
+
+      for (sum = 0, k = 0; k < taps; ++k) {
+        sum += src[(src_y + k) * src_stride] * filter_y[k];
+      }
+      sum += (VP9_FILTER_WEIGHT >> 1);
+      dst[y * dst_stride] = clip_pixel(sum >> VP9_FILTER_SHIFT);
+
+      /* Adjust source and filter to use for the next pixel */
+      y_q4 += y_step_q4;
+      filter_y = filter_y_base + (y_q4 & 0xf) * taps;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void convolve_avg_vert_c(const uint8_t *src, int src_stride,
+                                uint8_t *dst, int dst_stride,
+                                const int16_t *filter_x, int x_step_q4,
+                                const int16_t *filter_y0, int y_step_q4,
+                                int w, int h, int taps) {
+  int x, y, k, sum;
+
+  const int16_t *filter_y_base = filter_y0;
+
+#if ALIGN_FILTERS_256
+  filter_y_base = (const int16_t *)(((intptr_t)filter_y0) & ~(intptr_t)0xff);
+#endif
+
+  /* Adjust base pointer address for this source column */
+  src -= src_stride * (taps / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    /* Pointer to filter to use */
+    const int16_t *filter_y = filter_y0;
+
+    /* Initial phase offset */
+    int y0_q4 = (filter_y - filter_y_base) / taps;
+    int y_q4 = y0_q4;
+
+    for (y = 0; y < h; ++y) {
+      /* Per-pixel src offset */
+      int src_y = (y_q4 - y0_q4) >> 4;
+
+      for (sum = 0, k = 0; k < taps; ++k) {
+        sum += src[(src_y + k) * src_stride] * filter_y[k];
+      }
+      sum += (VP9_FILTER_WEIGHT >> 1);
+      dst[y * dst_stride] =
+          (dst[y * dst_stride] + clip_pixel(sum >> VP9_FILTER_SHIFT) + 1) >> 1;
+
+      /* Adjust source and filter to use for the next pixel */
+      y_q4 += y_step_q4;
+      filter_y = filter_y_base + (y_q4 & 0xf) * taps;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void convolve_c(const uint8_t *src, int src_stride,
+                       uint8_t *dst, int dst_stride,
+                       const int16_t *filter_x, int x_step_q4,
+                       const int16_t *filter_y, int y_step_q4,
+                       int w, int h, int taps) {
+  /* Fixed size intermediate buffer places limits on parameters.
+   * Maximum intermediate_height is 135, for y_step_q4 == 32,
+   * h == 64, taps == 8.
+   */
+  uint8_t temp[64 * 135];
+  int intermediate_height = ((h * y_step_q4) >> 4) + taps - 1;
+
+  assert(w <= 64);
+  assert(h <= 64);
+  assert(taps <= 8);
+  assert(y_step_q4 <= 32);
+
+  if (intermediate_height < h)
+    intermediate_height = h;
+
+  convolve_horiz_c(src - src_stride * (taps / 2 - 1), src_stride,
+                   temp, 64,
+                   filter_x, x_step_q4, filter_y, y_step_q4,
+                   w, intermediate_height, taps);
+  convolve_vert_c(temp + 64 * (taps / 2 - 1), 64, dst, dst_stride,
+                  filter_x, x_step_q4, filter_y, y_step_q4,
+                  w, h, taps);
+}
+
+static void convolve_avg_c(const uint8_t *src, int src_stride,
+                           uint8_t *dst, int dst_stride,
+                           const int16_t *filter_x, int x_step_q4,
+                           const int16_t *filter_y, int y_step_q4,
+                           int w, int h, int taps) {
+  /* Fixed size intermediate buffer places limits on parameters.
+   * Maximum intermediate_height is 135, for y_step_q4 == 32,
+   * h == 64, taps == 8.
+   */
+  uint8_t temp[64 * 135];
+  int intermediate_height = ((h * y_step_q4) >> 4) + taps - 1;
+
+  assert(w <= 64);
+  assert(h <= 64);
+  assert(taps <= 8);
+  assert(y_step_q4 <= 32);
+
+  if (intermediate_height < h)
+    intermediate_height = h;
+
+  convolve_horiz_c(src - src_stride * (taps / 2 - 1), src_stride,
+                   temp, 64,
+                   filter_x, x_step_q4, filter_y, y_step_q4,
+                   w, intermediate_height, taps);
+  convolve_avg_vert_c(temp + 64 * (taps / 2 - 1), 64, dst, dst_stride,
+                      filter_x, x_step_q4, filter_y, y_step_q4,
+                      w, h, taps);
+}
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride,
+                           uint8_t *dst, int dst_stride,
+                           const int16_t *filter_x, int x_step_q4,
+                           const int16_t *filter_y, int y_step_q4,
+                           int w, int h) {
+  convolve_horiz_c(src, src_stride, dst, dst_stride,
+                   filter_x, x_step_q4, filter_y, y_step_q4,
+                   w, h, 8);
+}
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride,
+                               uint8_t *dst, int dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h) {
+  convolve_avg_horiz_c(src, src_stride, dst, dst_stride,
+                       filter_x, x_step_q4, filter_y, y_step_q4,
+                       w, h, 8);
+}
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride,
+                          uint8_t *dst, int dst_stride,
+                          const int16_t *filter_x, int x_step_q4,
+                          const int16_t *filter_y, int y_step_q4,
+                          int w, int h) {
+  convolve_vert_c(src, src_stride, dst, dst_stride,
+                  filter_x, x_step_q4, filter_y, y_step_q4,
+                  w, h, 8);
+}
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride,
+                              uint8_t *dst, int dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h) {
+  convolve_avg_vert_c(src, src_stride, dst, dst_stride,
+                      filter_x, x_step_q4, filter_y, y_step_q4,
+                      w, h, 8);
+}
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride,
+                     uint8_t *dst, int dst_stride,
+                     const int16_t *filter_x, int x_step_q4,
+                     const int16_t *filter_y, int y_step_q4,
+                     int w, int h) {
+  convolve_c(src, src_stride, dst, dst_stride,
+             filter_x, x_step_q4, filter_y, y_step_q4,
+             w, h, 8);
+}
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride,
+                         uint8_t *dst, int dst_stride,
+                         const int16_t *filter_x, int x_step_q4,
+                         const int16_t *filter_y, int y_step_q4,
+                         int w, int h) {
+  /* Fixed size intermediate buffer places limits on parameters. */
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp, 64 * 64);
+  assert(w <= 64);
+  assert(h <= 64);
+
+  vp9_convolve8(src, src_stride,
+                temp, 64,
+                filter_x, x_step_q4,
+                filter_y, y_step_q4,
+                w, h);
+  vp9_convolve_avg(temp, 64,
+                   dst, dst_stride,
+                   NULL, 0, /* These unused parameter should be removed! */
+                   NULL, 0, /* These unused parameter should be removed! */
+                   w, h);
+}
+
+void vp9_convolve_copy(const uint8_t *src, int src_stride,
+                       uint8_t *dst, int dst_stride,
+                       const int16_t *filter_x, int filter_x_stride,
+                       const int16_t *filter_y, int filter_y_stride,
+                       int w, int h) {
+  if (w == 16 && h == 16) {
+    vp9_copy_mem16x16(src, src_stride, dst, dst_stride);
+  } else if (w == 8 && h == 8) {
+    vp9_copy_mem8x8(src, src_stride, dst, dst_stride);
+  } else if (w == 8 && h == 4) {
+    vp9_copy_mem8x4(src, src_stride, dst, dst_stride);
+  } else {
+    int r;
+
+    for (r = h; r > 0; --r) {
+      memcpy(dst, src, w);
+      src += src_stride;
+      dst += dst_stride;
+    }
+  }
+}
+
+void vp9_convolve_avg(const uint8_t *src, int src_stride,
+                      uint8_t *dst, int dst_stride,
+                      const int16_t *filter_x, int filter_x_stride,
+                      const int16_t *filter_y, int filter_y_stride,
+                      int w, int h) {
+  int x, y;
+
+  for (y = 0; y < h; ++y) {
+    for (x = 0; x < w; ++x) {
+      dst[x] = (dst[x] + src[x] + 1) >> 1;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
diff --git a/libvpx/vp9/common/vp9_convolve.h b/libvpx/vp9/common/vp9_convolve.h
new file mode 100644
index 0000000..0596080
--- /dev/null
+++ b/libvpx/vp9/common/vp9_convolve.h
@@ -0,0 +1,41 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef VP9_COMMON_CONVOLVE_H_
+#define VP9_COMMON_CONVOLVE_H_
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+
+typedef void (*convolve_fn_t)(const uint8_t *src, int src_stride,
+                              uint8_t *dst, int dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h);
+
+// Not a convolution, a block copy conforming to the convolution prototype
+void vp9_convolve_copy(const uint8_t *src, int src_stride,
+                       uint8_t *dst, int dst_stride,
+                       const int16_t *filter_x, int x_step_q4,
+                       const int16_t *filter_y, int y_step_q4,
+                       int w, int h);
+
+// Not a convolution, a block average conforming to the convolution prototype
+void vp9_convolve_avg(const uint8_t *src, int src_stride,
+                      uint8_t *dst, int dst_stride,
+                      const int16_t *filter_x, int x_step_q4,
+                      const int16_t *filter_y, int y_step_q4,
+                      int w, int h);
+
+struct subpix_fn_table {
+  const int16_t (*filter_x)[8];
+  const int16_t (*filter_y)[8];
+};
+
+#endif  // VP9_COMMON_CONVOLVE_H_
diff --git a/libvpx/vp9/common/vp9_debugmodes.c b/libvpx/vp9/common/vp9_debugmodes.c
new file mode 100644
index 0000000..5841f80
--- /dev/null
+++ b/libvpx/vp9/common/vp9_debugmodes.c
@@ -0,0 +1,137 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdio.h>
+
+#include "vp9/common/vp9_blockd.h"
+
+void vp9_print_modes_and_motion_vectors(MODE_INFO *mi, int rows, int cols,
+                                        int frame, char *file) {
+  int mi_row;
+  int mi_col;
+  int mi_index = 0;
+  FILE *mvs = fopen(file, "a");
+
+  // Print out the macroblock Y modes
+  fprintf(mvs, "SB Types for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%2d ", mi[mi_index].mbmi.sb_type);
+
+      mi_index++;
+    }
+
+    fprintf(mvs, "\n");
+    mi_index += 8;
+  }
+
+  // Print out the macroblock Y modes
+  fprintf(mvs, "Mb Modes for Frame %d\n", frame);
+  mi_index = 0;
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%2d ", mi[mi_index].mbmi.mode);
+
+      mi_index++;
+    }
+
+    fprintf(mvs, "\n");
+    mi_index += 8;
+  }
+
+  fprintf(mvs, "\n");
+
+  mi_index = 0;
+  fprintf(mvs, "Mb mv ref for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%2d ", mi[mi_index].mbmi.ref_frame[0]);
+
+      mi_index++;
+    }
+
+    fprintf(mvs, "\n");
+    mi_index += 8;
+  }
+  fprintf(mvs, "\n");
+
+  mi_index = 0;
+  fprintf(mvs, "Mb mv ref for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%4d:%4d ", mi[mi_index].mbmi.mv[0].as_mv.row,
+              mi[mi_index].mbmi.mv[0].as_mv.col);
+
+      mi_index++;
+    }
+
+    fprintf(mvs, "\n");
+    mi_index += 8;
+  }
+
+  fprintf(mvs, "\n");
+
+  /* print out the macroblock txform sizes */
+  mi_index = 0;
+  fprintf(mvs, "TXFM size for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%2d ", mi[mi_index].mbmi.txfm_size);
+
+      mi_index++;
+    }
+
+    mi_index += 8;
+    fprintf(mvs, "\n");
+  }
+
+  fprintf(mvs, "\n");
+
+  /* print out the macroblock UV modes */
+  mi_index = 0;
+  fprintf(mvs, "UV Modes for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%2d ", mi[mi_index].mbmi.uv_mode);
+
+      mi_index++;
+    }
+
+    mi_index += 8;
+    fprintf(mvs, "\n");
+  }
+
+  fprintf(mvs, "\n");
+
+  /* print out the macroblock mvs */
+  mi_index = 0;
+  fprintf(mvs, "MVs for Frame %d\n", frame);
+
+  for (mi_row = 0; mi_row < rows; mi_row++) {
+    for (mi_col = 0; mi_col < cols; mi_col++) {
+      fprintf(mvs, "%5d:%-5d", mi[mi_index].mbmi.mv[0].as_mv.row / 2,
+              mi[mi_index].mbmi.mv[0].as_mv.col / 2);
+
+      mi_index++;
+    }
+
+    mi_index += 8;
+    fprintf(mvs, "\n");
+  }
+
+  fprintf(mvs, "\n");
+
+  fclose(mvs);
+}
diff --git a/libvpx/vp9/common/vp9_default_coef_probs.h b/libvpx/vp9/common/vp9_default_coef_probs.h
new file mode 100644
index 0000000..1954093
--- /dev/null
+++ b/libvpx/vp9/common/vp9_default_coef_probs.h
@@ -0,0 +1,1384 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+*/
+
+
+/*Generated file, included by vp9_entropy.c*/
+
+#if CONFIG_BALANCED_COEFTREE
+static const vp9_coeff_probs_model default_coef_probs_4x4[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   6, 213, 178 },
+        {  26, 113, 132 },
+        {  34,  17,  68 }
+      }, { /* Coeff Band 1 */
+        {  66,  96, 178 },
+        {  63,  96, 174 },
+        {  67,  54, 154 },
+        {  62,  28, 126 },
+        {  48,   9,  84 },
+        {  20,   1,  32 }
+      }, { /* Coeff Band 2 */
+        {  64, 144, 206 },
+        {  70,  99, 191 },
+        {  69,  36, 152 },
+        {  55,   9, 106 },
+        {  35,   1,  60 },
+        {  14,   1,  22 }
+      }, { /* Coeff Band 3 */
+        {  82, 154, 222 },
+        {  83, 112, 205 },
+        {  81,  31, 164 },
+        {  62,   7, 118 },
+        {  42,   1,  74 },
+        {  18,   1,  30 }
+      }, { /* Coeff Band 4 */
+        {  52, 179, 233 },
+        {  64, 132, 214 },
+        {  73,  36, 170 },
+        {  59,   8, 116 },
+        {  38,   1,  65 },
+        {  15,   1,  26 }
+      }, { /* Coeff Band 5 */
+        {  29, 175, 238 },
+        {  26, 169, 223 },
+        {  41,  80, 182 },
+        {  39,  32, 127 },
+        {  26,  10,  69 },
+        {  11,   2,  28 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  21, 226, 234 },
+        {  52, 182, 212 },
+        {  80, 112, 177 }
+      }, { /* Coeff Band 1 */
+        { 111, 164, 243 },
+        {  88, 152, 231 },
+        {  90,  43, 186 },
+        {  70,  12, 132 },
+        {  44,   2,  76 },
+        {  19,   1,  33 }
+      }, { /* Coeff Band 2 */
+        {  96, 185, 246 },
+        {  99, 127, 231 },
+        {  88,  21, 177 },
+        {  64,   5, 122 },
+        {  38,   1,  69 },
+        {  18,   1,  30 }
+      }, { /* Coeff Band 3 */
+        {  84, 206, 249 },
+        {  94, 147, 237 },
+        {  95,  33, 187 },
+        {  71,   8, 131 },
+        {  47,   1,  83 },
+        {  26,   1,  44 }
+      }, { /* Coeff Band 4 */
+        {  38, 221, 252 },
+        {  58, 177, 241 },
+        {  78,  46, 188 },
+        {  59,   9, 122 },
+        {  34,   1,  66 },
+        {  18,   1,  34 }
+      }, { /* Coeff Band 5 */
+        {  21, 216, 253 },
+        {  21, 206, 244 },
+        {  42,  93, 200 },
+        {  43,  41, 146 },
+        {  36,  13,  93 },
+        {  31,   1,  55 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   7, 213, 219 },
+        {  23, 139, 182 },
+        {  38,  60, 125 }
+      }, { /* Coeff Band 1 */
+        {  69, 156, 220 },
+        {  52, 178, 213 },
+        {  69, 111, 190 },
+        {  69,  58, 155 },
+        {  58,  21, 104 },
+        {  39,   7,  60 }
+      }, { /* Coeff Band 2 */
+        {  68, 189, 228 },
+        {  70, 158, 221 },
+        {  83,  64, 189 },
+        {  73,  18, 141 },
+        {  48,   4,  88 },
+        {  23,   1,  41 }
+      }, { /* Coeff Band 3 */
+        {  99, 194, 236 },
+        {  91, 138, 224 },
+        {  91,  53, 189 },
+        {  74,  20, 142 },
+        {  48,   6,  90 },
+        {  22,   1,  41 }
+      }, { /* Coeff Band 4 */
+        {  52, 203, 244 },
+        {  60, 168, 231 },
+        {  75,  62, 189 },
+        {  61,  18, 132 },
+        {  38,   4,  72 },
+        {  17,   1,  39 }
+      }, { /* Coeff Band 5 */
+        {  33, 192, 247 },
+        {  31, 185, 234 },
+        {  46,  85, 185 },
+        {  39,  35, 132 },
+        {  28,  15,  80 },
+        {  13,   5,  38 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {   5, 247, 246 },
+        {  28, 209, 228 },
+        {  65, 137, 203 }
+      }, { /* Coeff Band 1 */
+        {  69, 208, 250 },
+        {  54, 207, 242 },
+        {  81,  92, 204 },
+        {  70,  54, 153 },
+        {  58,  40, 108 },
+        {  58,  35,  71 }
+      }, { /* Coeff Band 2 */
+        {  65, 215, 250 },
+        {  72, 185, 239 },
+        {  92,  50, 197 },
+        {  75,  14, 147 },
+        {  49,   2,  99 },
+        {  26,   1,  53 }
+      }, { /* Coeff Band 3 */
+        {  70, 220, 251 },
+        {  76, 186, 241 },
+        {  90,  65, 198 },
+        {  75,  26, 151 },
+        {  58,  12, 112 },
+        {  34,   6,  49 }
+      }, { /* Coeff Band 4 */
+        {  34, 224, 253 },
+        {  44, 204, 245 },
+        {  69,  85, 204 },
+        {  64,  31, 150 },
+        {  44,   2,  78 },
+        {   1,   1, 128 }
+      }, { /* Coeff Band 5 */
+        {  25, 216, 253 },
+        {  21, 215, 248 },
+        {  47, 108, 214 },
+        {  47,  48, 160 },
+        {  26,  20,  90 },
+        {  64, 171, 128 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_8x8[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   9, 203, 199 },
+        {  26,  92, 128 },
+        {  28,  11,  55 }
+      }, { /* Coeff Band 1 */
+        {  99,  54, 160 },
+        {  78,  99, 155 },
+        {  80,  44, 138 },
+        {  71,  17, 115 },
+        {  51,   5,  80 },
+        {  27,   1,  40 }
+      }, { /* Coeff Band 2 */
+        { 135,  81, 190 },
+        { 113,  61, 182 },
+        {  93,  16, 153 },
+        {  70,   4, 115 },
+        {  41,   1,  68 },
+        {  16,   1,  27 }
+      }, { /* Coeff Band 3 */
+        { 155, 103, 214 },
+        { 129,  48, 199 },
+        {  95,  10, 159 },
+        {  63,   1, 110 },
+        {  32,   1,  58 },
+        {  12,   1,  21 }
+      }, { /* Coeff Band 4 */
+        { 163, 149, 231 },
+        { 137,  69, 213 },
+        {  95,  11, 164 },
+        {  62,   3, 108 },
+        {  32,   1,  57 },
+        {  13,   1,  22 }
+      }, { /* Coeff Band 5 */
+        { 136, 189, 239 },
+        { 123, 102, 223 },
+        {  97,  19, 170 },
+        {  66,   4, 111 },
+        {  38,   1,  60 },
+        {  18,   1,  26 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  24, 226, 244 },
+        {  54, 178, 211 },
+        {  80,  74, 152 }
+      }, { /* Coeff Band 1 */
+        { 145, 153, 236 },
+        { 101, 163, 223 },
+        { 108,  50, 187 },
+        {  90,  22, 145 },
+        {  66,   8,  97 },
+        {  42,   4,  50 }
+      }, { /* Coeff Band 2 */
+        { 150, 159, 238 },
+        { 128,  90, 218 },
+        {  94,   9, 163 },
+        {  64,   3, 110 },
+        {  34,   1,  61 },
+        {  13,   1,  24 }
+      }, { /* Coeff Band 3 */
+        { 151, 162, 242 },
+        { 135,  80, 222 },
+        {  93,   9, 166 },
+        {  61,   3, 111 },
+        {  31,   1,  59 },
+        {  12,   1,  22 }
+      }, { /* Coeff Band 4 */
+        { 161, 170, 245 },
+        { 140,  84, 228 },
+        {  99,   8, 174 },
+        {  64,   1, 116 },
+        {  34,   1,  63 },
+        {  14,   1,  26 }
+      }, { /* Coeff Band 5 */
+        { 138, 197, 246 },
+        { 127, 109, 233 },
+        { 100,  16, 179 },
+        {  66,   3, 119 },
+        {  37,   1,  66 },
+        {  16,   1,  30 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   6, 216, 212 },
+        {  25, 134, 171 },
+        {  43,  48, 118 }
+      }, { /* Coeff Band 1 */
+        {  93, 112, 209 },
+        {  66, 159, 206 },
+        {  82,  78, 184 },
+        {  75,  28, 148 },
+        {  46,   4,  82 },
+        {  18,   1,  28 }
+      }, { /* Coeff Band 2 */
+        { 108, 148, 220 },
+        {  90, 130, 216 },
+        {  92,  40, 186 },
+        {  73,  10, 135 },
+        {  46,   1,  79 },
+        {  20,   1,  35 }
+      }, { /* Coeff Band 3 */
+        { 125, 173, 232 },
+        { 109, 117, 223 },
+        {  97,  31, 183 },
+        {  71,   7, 127 },
+        {  44,   1,  76 },
+        {  21,   1,  36 }
+      }, { /* Coeff Band 4 */
+        { 133, 195, 236 },
+        { 112, 121, 224 },
+        {  97,  23, 178 },
+        {  69,   3, 122 },
+        {  42,   1,  72 },
+        {  19,   1,  34 }
+      }, { /* Coeff Band 5 */
+        { 132, 180, 238 },
+        { 119, 102, 225 },
+        { 101,  18, 179 },
+        {  71,   3, 124 },
+        {  42,   1,  70 },
+        {  17,   1,  28 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {   5, 242, 250 },
+        {  26, 198, 226 },
+        {  58,  98, 168 }
+      }, { /* Coeff Band 1 */
+        {  82, 201, 246 },
+        {  50, 219, 237 },
+        {  94, 107, 205 },
+        {  89,  61, 167 },
+        {  77,  31, 131 },
+        {  57,  14,  91 }
+      }, { /* Coeff Band 2 */
+        {  99, 202, 247 },
+        {  96, 165, 234 },
+        { 100,  31, 190 },
+        {  72,   8, 131 },
+        {  41,   1,  72 },
+        {  14,   1,  24 }
+      }, { /* Coeff Band 3 */
+        { 108, 204, 248 },
+        { 107, 156, 235 },
+        { 103,  27, 186 },
+        {  71,   4, 124 },
+        {  39,   1,  66 },
+        {  14,   1,  19 }
+      }, { /* Coeff Band 4 */
+        { 120, 211, 248 },
+        { 118, 149, 234 },
+        { 107,  19, 182 },
+        {  72,   3, 126 },
+        {  40,   1,  69 },
+        {  16,   1,  24 }
+      }, { /* Coeff Band 5 */
+        { 127, 199, 245 },
+        { 122, 125, 232 },
+        { 112,  20, 186 },
+        {  82,   3, 136 },
+        {  55,   1,  88 },
+        {  10,   1,  38 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_16x16[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {  25,   9, 101 },
+        {  25,   2,  67 },
+        {  15,   1,  28 }
+      }, { /* Coeff Band 1 */
+        {  67,  30, 118 },
+        {  61,  56, 116 },
+        {  60,  31, 105 },
+        {  52,  11,  85 },
+        {  34,   2,  54 },
+        {  14,   1,  22 }
+      }, { /* Coeff Band 2 */
+        { 107,  58, 149 },
+        {  92,  53, 147 },
+        {  78,  14, 123 },
+        {  56,   3,  87 },
+        {  35,   1,  56 },
+        {  17,   1,  27 }
+      }, { /* Coeff Band 3 */
+        { 142,  61, 171 },
+        { 111,  30, 162 },
+        {  80,   4, 128 },
+        {  53,   1,  87 },
+        {  31,   1,  52 },
+        {  14,   1,  24 }
+      }, { /* Coeff Band 4 */
+        { 171,  73, 200 },
+        { 129,  28, 184 },
+        {  86,   3, 140 },
+        {  54,   1,  90 },
+        {  28,   1,  49 },
+        {  12,   1,  21 }
+      }, { /* Coeff Band 5 */
+        { 193, 129, 227 },
+        { 148,  28, 200 },
+        {  90,   2, 144 },
+        {  53,   1,  90 },
+        {  28,   1,  50 },
+        {  13,   1,  22 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  60,   7, 234 },
+        {  64,   4, 184 },
+        {  56,   1, 104 }
+      }, { /* Coeff Band 1 */
+        { 150, 111, 210 },
+        {  87, 185, 202 },
+        { 101,  81, 177 },
+        {  90,  34, 142 },
+        {  67,  11,  95 },
+        {  38,   2,  51 }
+      }, { /* Coeff Band 2 */
+        { 153, 139, 218 },
+        { 120,  72, 195 },
+        {  90,  11, 147 },
+        {  63,   3, 101 },
+        {  39,   1,  61 },
+        {  20,   1,  33 }
+      }, { /* Coeff Band 3 */
+        { 171, 132, 223 },
+        { 131,  56, 200 },
+        {  92,   6, 147 },
+        {  58,   1,  95 },
+        {  32,   1,  52 },
+        {  14,   1,  23 }
+      }, { /* Coeff Band 4 */
+        { 183, 137, 227 },
+        { 139,  48, 204 },
+        {  91,   3, 148 },
+        {  55,   1,  91 },
+        {  28,   1,  47 },
+        {  13,   1,  21 }
+      }, { /* Coeff Band 5 */
+        { 198, 149, 234 },
+        { 153,  32, 208 },
+        {  95,   2, 148 },
+        {  55,   1,  90 },
+        {  30,   1,  51 },
+        {  16,   1,  25 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   7, 209, 217 },
+        {  31, 106, 151 },
+        {  40,  21,  86 }
+      }, { /* Coeff Band 1 */
+        { 101,  71, 184 },
+        {  74, 131, 177 },
+        {  88,  50, 158 },
+        {  78,  16, 129 },
+        {  51,   2,  82 },
+        {  18,   1,  29 }
+      }, { /* Coeff Band 2 */
+        { 116, 115, 199 },
+        { 102,  88, 191 },
+        {  94,  22, 160 },
+        {  74,   6, 122 },
+        {  47,   1,  77 },
+        {  18,   1,  30 }
+      }, { /* Coeff Band 3 */
+        { 157, 124, 210 },
+        { 130,  53, 201 },
+        { 102,  10, 165 },
+        {  73,   1, 120 },
+        {  42,   1,  69 },
+        {  16,   1,  27 }
+      }, { /* Coeff Band 4 */
+        { 174, 147, 225 },
+        { 134,  67, 212 },
+        { 100,  10, 168 },
+        {  66,   1, 111 },
+        {  36,   1,  60 },
+        {  16,   1,  27 }
+      }, { /* Coeff Band 5 */
+        { 185, 165, 232 },
+        { 147,  56, 214 },
+        { 105,   5, 165 },
+        {  66,   1, 108 },
+        {  35,   1,  59 },
+        {  16,   1,  27 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {   3, 232, 245 },
+        {  18, 162, 210 },
+        {  38,  64, 131 }
+      }, { /* Coeff Band 1 */
+        {  84, 187, 239 },
+        {  35, 231, 231 },
+        {  82, 150, 209 },
+        {  87,  97, 181 },
+        {  81,  64, 151 },
+        {  67,  60, 119 }
+      }, { /* Coeff Band 2 */
+        { 107, 185, 239 },
+        { 100, 149, 224 },
+        { 107,  34, 185 },
+        {  83,  12, 141 },
+        {  49,   4,  92 },
+        {  21,   1,  40 }
+      }, { /* Coeff Band 3 */
+        { 125, 184, 243 },
+        { 121, 127, 228 },
+        { 113,  25, 185 },
+        {  82,   6, 134 },
+        {  48,   1,  82 },
+        {  26,   1,  38 }
+      }, { /* Coeff Band 4 */
+        { 143, 185, 245 },
+        { 133, 115, 231 },
+        { 114,  14, 184 },
+        {  77,   3, 126 },
+        {  43,   1,  68 },
+        {  34,   1,  40 }
+      }, { /* Coeff Band 5 */
+        { 170, 194, 241 },
+        { 151,  80, 226 },
+        { 118,   9, 180 },
+        {  81,   1, 130 },
+        {  51,   1,  78 },
+        {  18,   1,  49 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_32x32[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {  29,  42, 137 },
+        {  26,   3,  60 },
+        {  13,   1,  23 }
+      }, { /* Coeff Band 1 */
+        {  69,  36, 122 },
+        {  63,  57, 123 },
+        {  60,  33, 112 },
+        {  52,  11,  90 },
+        {  32,   2,  52 },
+        {  10,   1,  15 }
+      }, { /* Coeff Band 2 */
+        { 107,  55, 143 },
+        {  86,  69, 143 },
+        {  74,  24, 116 },
+        {  52,   5,  78 },
+        {  29,   1,  44 },
+        {  12,   1,  18 }
+      }, { /* Coeff Band 3 */
+        { 137,  71, 160 },
+        { 107,  34, 152 },
+        {  73,   6, 114 },
+        {  44,   1,  69 },
+        {  25,   1,  40 },
+        {  12,   1,  18 }
+      }, { /* Coeff Band 4 */
+        { 165,  70, 174 },
+        { 118,  24, 159 },
+        {  74,   3, 117 },
+        {  45,   1,  73 },
+        {  26,   1,  43 },
+        {  12,   1,  19 }
+      }, { /* Coeff Band 5 */
+        { 220,  93, 223 },
+        { 153,  10, 187 },
+        {  86,   2, 131 },
+        {  49,   1,  79 },
+        {  26,   1,  43 },
+        {  12,   1,  20 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  30,  58, 227 },
+        {  35,  10, 172 },
+        {  24,  23, 112 }
+      }, { /* Coeff Band 1 */
+        { 117, 145, 219 },
+        {  51, 221, 216 },
+        {  75, 169, 196 },
+        {  88,  96, 165 },
+        {  77,  43, 117 },
+        {  53,  18,  60 }
+      }, { /* Coeff Band 2 */
+        { 128, 176, 225 },
+        { 108, 114, 202 },
+        {  92,  19, 152 },
+        {  65,   4, 103 },
+        {  38,   1,  61 },
+        {  19,   1,  30 }
+      }, { /* Coeff Band 3 */
+        { 146, 184, 228 },
+        { 122,  95, 205 },
+        {  92,  11, 149 },
+        {  62,   1,  98 },
+        {  35,   1,  57 },
+        {  17,   1,  26 }
+      }, { /* Coeff Band 4 */
+        { 165, 192, 230 },
+        { 132,  81, 206 },
+        {  93,   6, 147 },
+        {  58,   1,  94 },
+        {  32,   1,  52 },
+        {  15,   1,  24 }
+      }, { /* Coeff Band 5 */
+        { 204, 223, 234 },
+        { 156,  49, 204 },
+        {  97,   3, 145 },
+        {  59,   1,  92 },
+        {  33,   1,  52 },
+        {  15,   1,  24 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   7, 184, 200 },
+        {  25,  67, 113 },
+        {  30,   9,  59 }
+      }, { /* Coeff Band 1 */
+        {  92,  42, 158 },
+        {  65, 121, 159 },
+        {  77,  56, 146 },
+        {  70,  22, 120 },
+        {  47,   4,  76 },
+        {  18,   1,  26 }
+      }, { /* Coeff Band 2 */
+        { 113,  81, 177 },
+        {  96,  75, 167 },
+        {  84,  24, 136 },
+        {  63,   8, 100 },
+        {  37,   1,  58 },
+        {  13,   1,  19 }
+      }, { /* Coeff Band 3 */
+        { 147,  85, 194 },
+        { 119,  36, 178 },
+        {  88,   8, 139 },
+        {  59,   1,  93 },
+        {  31,   1,  49 },
+        {  10,   1,  18 }
+      }, { /* Coeff Band 4 */
+        { 169, 108, 210 },
+        { 131,  41, 191 },
+        {  92,   5, 144 },
+        {  56,   1,  88 },
+        {  29,   1,  47 },
+        {  14,   1,  22 }
+      }, { /* Coeff Band 5 */
+        { 210, 106, 223 },
+        { 148,  14, 192 },
+        {  89,   2, 138 },
+        {  52,   1,  84 },
+        {  29,   1,  47 },
+        {  14,   1,  23 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {   3, 207, 245 },
+        {  12, 102, 213 },
+        {  18,  33, 144 }
+      }, { /* Coeff Band 1 */
+        {  85, 205, 245 },
+        {  18, 249, 242 },
+        {  59, 221, 229 },
+        {  91, 166, 213 },
+        {  88, 117, 183 },
+        {  70,  95, 149 }
+      }, { /* Coeff Band 2 */
+        { 114, 193, 241 },
+        { 104, 155, 221 },
+        { 100,  33, 181 },
+        {  78,  10, 132 },
+        {  43,   2,  75 },
+        {  15,   1,  48 }
+      }, { /* Coeff Band 3 */
+        { 118, 198, 244 },
+        { 117, 142, 224 },
+        { 111,  25, 179 },
+        {  83,   4, 134 },
+        {  57,   1,  84 },
+        {   1,   1,   1 }
+      }, { /* Coeff Band 4 */
+        { 144, 201, 248 },
+        { 136, 130, 234 },
+        { 124,  12, 188 },
+        {  83,   1, 130 },
+        {  61,   1,  66 },
+        {  64, 171, 128 }
+      }, { /* Coeff Band 5 */
+        { 174, 227, 250 },
+        { 165, 118, 242 },
+        { 132,  21, 197 },
+        {  84,   3, 134 },
+        {  70,   1,  69 },
+        {   1,   1,   1 }
+      }
+    }
+  }
+};
+#else
+static const vp9_coeff_probs_model default_coef_probs_4x4[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 195,  29, 183 },
+        {  84,  49, 136 },
+        {   8,  42,  71 }
+      }, { /* Coeff Band 1 */
+        {  31, 107, 169 },
+        {  35,  99, 159 },
+        {  17,  82, 140 },
+        {   8,  66, 114 },
+        {   2,  44,  76 },
+        {   1,  19,  32 }
+      }, { /* Coeff Band 2 */
+        {  40, 132, 201 },
+        {  29, 114, 187 },
+        {  13,  91, 157 },
+        {   7,  75, 127 },
+        {   3,  58,  95 },
+        {   1,  28,  47 }
+      }, { /* Coeff Band 3 */
+        {  69, 142, 221 },
+        {  42, 122, 201 },
+        {  15,  91, 159 },
+        {   6,  67, 121 },
+        {   1,  42,  77 },
+        {   1,  17,  31 }
+      }, { /* Coeff Band 4 */
+        { 102, 148, 228 },
+        {  67, 117, 204 },
+        {  17,  82, 154 },
+        {   6,  59, 114 },
+        {   2,  39,  75 },
+        {   1,  15,  29 }
+      }, { /* Coeff Band 5 */
+        { 156,  57, 233 },
+        { 119,  57, 212 },
+        {  58,  48, 163 },
+        {  29,  40, 124 },
+        {  12,  30,  81 },
+        {   3,  12,  31 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 191, 107, 226 },
+        { 124, 117, 204 },
+        {  25,  99, 155 }
+      }, { /* Coeff Band 1 */
+        {  29, 148, 210 },
+        {  37, 126, 194 },
+        {   8,  93, 157 },
+        {   2,  68, 118 },
+        {   1,  39,  69 },
+        {   1,  17,  33 }
+      }, { /* Coeff Band 2 */
+        {  41, 151, 213 },
+        {  27, 123, 193 },
+        {   3,  82, 144 },
+        {   1,  58, 105 },
+        {   1,  32,  60 },
+        {   1,  13,  26 }
+      }, { /* Coeff Band 3 */
+        {  59, 159, 220 },
+        {  23, 126, 198 },
+        {   4,  88, 151 },
+        {   1,  66, 114 },
+        {   1,  38,  71 },
+        {   1,  18,  34 }
+      }, { /* Coeff Band 4 */
+        { 114, 136, 232 },
+        {  51, 114, 207 },
+        {  11,  83, 155 },
+        {   3,  56, 105 },
+        {   1,  33,  65 },
+        {   1,  17,  34 }
+      }, { /* Coeff Band 5 */
+        { 149,  65, 234 },
+        { 121,  57, 215 },
+        {  61,  49, 166 },
+        {  28,  36, 114 },
+        {  12,  25,  76 },
+        {   3,  16,  42 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 214,  49, 220 },
+        { 132,  63, 188 },
+        {  42,  65, 137 }
+      }, { /* Coeff Band 1 */
+        {  85, 137, 221 },
+        { 104, 131, 216 },
+        {  49, 111, 192 },
+        {  21,  87, 155 },
+        {   2,  49,  87 },
+        {   1,  16,  28 }
+      }, { /* Coeff Band 2 */
+        {  89, 163, 230 },
+        {  90, 137, 220 },
+        {  29, 100, 183 },
+        {  10,  70, 135 },
+        {   2,  42,  81 },
+        {   1,  17,  33 }
+      }, { /* Coeff Band 3 */
+        { 108, 167, 237 },
+        {  55, 133, 222 },
+        {  15,  97, 179 },
+        {   4,  72, 135 },
+        {   1,  45,  85 },
+        {   1,  19,  38 }
+      }, { /* Coeff Band 4 */
+        { 124, 146, 240 },
+        {  66, 124, 224 },
+        {  17,  88, 175 },
+        {   4,  58, 122 },
+        {   1,  36,  75 },
+        {   1,  18,  37 }
+      }, { /* Coeff Band 5 */
+        { 141,  79, 241 },
+        { 126,  70, 227 },
+        {  66,  58, 182 },
+        {  30,  44, 136 },
+        {  12,  34,  96 },
+        {   2,  20,  47 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 229,  99, 249 },
+        { 143, 111, 235 },
+        {  46, 109, 192 }
+      }, { /* Coeff Band 1 */
+        {  82, 158, 236 },
+        {  94, 146, 224 },
+        {  25, 117, 191 },
+        {   9,  87, 149 },
+        {   3,  56,  99 },
+        {   1,  33,  57 }
+      }, { /* Coeff Band 2 */
+        {  83, 167, 237 },
+        {  68, 145, 222 },
+        {  10, 103, 177 },
+        {   2,  72, 131 },
+        {   1,  41,  79 },
+        {   1,  20,  39 }
+      }, { /* Coeff Band 3 */
+        {  99, 167, 239 },
+        {  47, 141, 224 },
+        {  10, 104, 178 },
+        {   2,  73, 133 },
+        {   1,  44,  85 },
+        {   1,  22,  47 }
+      }, { /* Coeff Band 4 */
+        { 127, 145, 243 },
+        {  71, 129, 228 },
+        {  17,  93, 177 },
+        {   3,  61, 124 },
+        {   1,  41,  84 },
+        {   1,  21,  52 }
+      }, { /* Coeff Band 5 */
+        { 157,  78, 244 },
+        { 140,  72, 231 },
+        {  69,  58, 184 },
+        {  31,  44, 137 },
+        {  14,  38, 105 },
+        {   8,  23,  61 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_8x8[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 125,  34, 187 },
+        {  52,  41, 133 },
+        {   6,  31,  56 }
+      }, { /* Coeff Band 1 */
+        {  37, 109, 153 },
+        {  51, 102, 147 },
+        {  23,  87, 128 },
+        {   8,  67, 101 },
+        {   1,  41,  63 },
+        {   1,  19,  29 }
+      }, { /* Coeff Band 2 */
+        {  31, 154, 185 },
+        {  17, 127, 175 },
+        {   6,  96, 145 },
+        {   2,  73, 114 },
+        {   1,  51,  82 },
+        {   1,  28,  45 }
+      }, { /* Coeff Band 3 */
+        {  23, 163, 200 },
+        {  10, 131, 185 },
+        {   2,  93, 148 },
+        {   1,  67, 111 },
+        {   1,  41,  69 },
+        {   1,  14,  24 }
+      }, { /* Coeff Band 4 */
+        {  29, 176, 217 },
+        {  12, 145, 201 },
+        {   3, 101, 156 },
+        {   1,  69, 111 },
+        {   1,  39,  63 },
+        {   1,  14,  23 }
+      }, { /* Coeff Band 5 */
+        {  57, 192, 233 },
+        {  25, 154, 215 },
+        {   6, 109, 167 },
+        {   3,  78, 118 },
+        {   1,  48,  69 },
+        {   1,  21,  29 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 202, 105, 245 },
+        { 108, 106, 216 },
+        {  18,  90, 144 }
+      }, { /* Coeff Band 1 */
+        {  33, 172, 219 },
+        {  64, 149, 206 },
+        {  14, 117, 177 },
+        {   5,  90, 141 },
+        {   2,  61,  95 },
+        {   1,  37,  57 }
+      }, { /* Coeff Band 2 */
+        {  33, 179, 220 },
+        {  11, 140, 198 },
+        {   1,  89, 148 },
+        {   1,  60, 104 },
+        {   1,  33,  57 },
+        {   1,  12,  21 }
+      }, { /* Coeff Band 3 */
+        {  30, 181, 221 },
+        {   8, 141, 198 },
+        {   1,  87, 145 },
+        {   1,  58, 100 },
+        {   1,  31,  55 },
+        {   1,  12,  20 }
+      }, { /* Coeff Band 4 */
+        {  32, 186, 224 },
+        {   7, 142, 198 },
+        {   1,  86, 143 },
+        {   1,  58, 100 },
+        {   1,  31,  55 },
+        {   1,  12,  22 }
+      }, { /* Coeff Band 5 */
+        {  57, 192, 227 },
+        {  20, 143, 204 },
+        {   3,  96, 154 },
+        {   1,  68, 112 },
+        {   1,  42,  69 },
+        {   1,  19,  32 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 212,  35, 215 },
+        { 113,  47, 169 },
+        {  29,  48, 105 }
+      }, { /* Coeff Band 1 */
+        {  74, 129, 203 },
+        { 106, 120, 203 },
+        {  49, 107, 178 },
+        {  19,  84, 144 },
+        {   4,  50,  84 },
+        {   1,  15,  25 }
+      }, { /* Coeff Band 2 */
+        {  71, 172, 217 },
+        {  44, 141, 209 },
+        {  15, 102, 173 },
+        {   6,  76, 133 },
+        {   2,  51,  89 },
+        {   1,  24,  42 }
+      }, { /* Coeff Band 3 */
+        {  64, 185, 231 },
+        {  31, 148, 216 },
+        {   8, 103, 175 },
+        {   3,  74, 131 },
+        {   1,  46,  81 },
+        {   1,  18,  30 }
+      }, { /* Coeff Band 4 */
+        {  65, 196, 235 },
+        {  25, 157, 221 },
+        {   5, 105, 174 },
+        {   1,  67, 120 },
+        {   1,  38,  69 },
+        {   1,  15,  30 }
+      }, { /* Coeff Band 5 */
+        {  65, 204, 238 },
+        {  30, 156, 224 },
+        {   7, 107, 177 },
+        {   2,  70, 124 },
+        {   1,  42,  73 },
+        {   1,  18,  34 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 225,  86, 251 },
+        { 144, 104, 235 },
+        {  42,  99, 181 }
+      }, { /* Coeff Band 1 */
+        {  85, 175, 239 },
+        { 112, 165, 229 },
+        {  29, 136, 200 },
+        {  12, 103, 162 },
+        {   6,  77, 123 },
+        {   2,  53,  84 }
+      }, { /* Coeff Band 2 */
+        {  75, 183, 239 },
+        {  30, 155, 221 },
+        {   3, 106, 171 },
+        {   1,  74, 128 },
+        {   1,  44,  76 },
+        {   1,  17,  28 }
+      }, { /* Coeff Band 3 */
+        {  73, 185, 240 },
+        {  27, 159, 222 },
+        {   2, 107, 172 },
+        {   1,  75, 127 },
+        {   1,  42,  73 },
+        {   1,  17,  29 }
+      }, { /* Coeff Band 4 */
+        {  62, 190, 238 },
+        {  21, 159, 222 },
+        {   2, 107, 172 },
+        {   1,  72, 122 },
+        {   1,  40,  71 },
+        {   1,  18,  32 }
+      }, { /* Coeff Band 5 */
+        {  61, 199, 240 },
+        {  27, 161, 226 },
+        {   4, 113, 180 },
+        {   1,  76, 129 },
+        {   1,  46,  80 },
+        {   1,  23,  41 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_16x16[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {   7,  27, 153 },
+        {   5,  30,  95 },
+        {   1,  16,  30 }
+      }, { /* Coeff Band 1 */
+        {  50,  75, 127 },
+        {  57,  75, 124 },
+        {  27,  67, 108 },
+        {  10,  54,  86 },
+        {   1,  33,  52 },
+        {   1,  12,  18 }
+      }, { /* Coeff Band 2 */
+        {  43, 125, 151 },
+        {  26, 108, 148 },
+        {   7,  83, 122 },
+        {   2,  59,  89 },
+        {   1,  38,  60 },
+        {   1,  17,  27 }
+      }, { /* Coeff Band 3 */
+        {  23, 144, 163 },
+        {  13, 112, 154 },
+        {   2,  75, 117 },
+        {   1,  50,  81 },
+        {   1,  31,  51 },
+        {   1,  14,  23 }
+      }, { /* Coeff Band 4 */
+        {  18, 162, 185 },
+        {   6, 123, 171 },
+        {   1,  78, 125 },
+        {   1,  51,  86 },
+        {   1,  31,  54 },
+        {   1,  14,  23 }
+      }, { /* Coeff Band 5 */
+        {  15, 199, 227 },
+        {   3, 150, 204 },
+        {   1,  91, 146 },
+        {   1,  55,  95 },
+        {   1,  30,  53 },
+        {   1,  11,  20 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  19,  55, 240 },
+        {  19,  59, 196 },
+        {   3,  52, 105 }
+      }, { /* Coeff Band 1 */
+        {  41, 166, 207 },
+        { 104, 153, 199 },
+        {  31, 123, 181 },
+        {  14, 101, 152 },
+        {   5,  72, 106 },
+        {   1,  36,  52 }
+      }, { /* Coeff Band 2 */
+        {  35, 176, 211 },
+        {  12, 131, 190 },
+        {   2,  88, 144 },
+        {   1,  60, 101 },
+        {   1,  36,  60 },
+        {   1,  16,  28 }
+      }, { /* Coeff Band 3 */
+        {  28, 183, 213 },
+        {   8, 134, 191 },
+        {   1,  86, 142 },
+        {   1,  56,  96 },
+        {   1,  30,  53 },
+        {   1,  12,  20 }
+      }, { /* Coeff Band 4 */
+        {  20, 190, 215 },
+        {   4, 135, 192 },
+        {   1,  84, 139 },
+        {   1,  53,  91 },
+        {   1,  28,  49 },
+        {   1,  11,  20 }
+      }, { /* Coeff Band 5 */
+        {  13, 196, 216 },
+        {   2, 137, 192 },
+        {   1,  86, 143 },
+        {   1,  57,  99 },
+        {   1,  32,  56 },
+        {   1,  13,  24 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 211,  29, 217 },
+        {  96,  47, 156 },
+        {  22,  43,  87 }
+      }, { /* Coeff Band 1 */
+        {  78, 120, 193 },
+        { 111, 116, 186 },
+        {  46, 102, 164 },
+        {  15,  80, 128 },
+        {   2,  49,  76 },
+        {   1,  18,  28 }
+      }, { /* Coeff Band 2 */
+        {  71, 161, 203 },
+        {  42, 132, 192 },
+        {  10,  98, 150 },
+        {   3,  69, 109 },
+        {   1,  44,  70 },
+        {   1,  18,  29 }
+      }, { /* Coeff Band 3 */
+        {  57, 186, 211 },
+        {  30, 140, 196 },
+        {   4,  93, 146 },
+        {   1,  62, 102 },
+        {   1,  38,  65 },
+        {   1,  16,  27 }
+      }, { /* Coeff Band 4 */
+        {  47, 199, 217 },
+        {  14, 145, 196 },
+        {   1,  88, 142 },
+        {   1,  57,  98 },
+        {   1,  36,  62 },
+        {   1,  15,  26 }
+      }, { /* Coeff Band 5 */
+        {  26, 219, 229 },
+        {   5, 155, 207 },
+        {   1,  94, 151 },
+        {   1,  60, 104 },
+        {   1,  36,  62 },
+        {   1,  16,  28 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 233,  29, 248 },
+        { 146,  47, 220 },
+        {  43,  52, 140 }
+      }, { /* Coeff Band 1 */
+        { 100, 163, 232 },
+        { 179, 161, 222 },
+        {  63, 142, 204 },
+        {  37, 113, 174 },
+        {  26,  89, 137 },
+        {  18,  68,  97 }
+      }, { /* Coeff Band 2 */
+        {  85, 181, 230 },
+        {  32, 146, 209 },
+        {   7, 100, 164 },
+        {   3,  71, 121 },
+        {   1,  45,  77 },
+        {   1,  18,  30 }
+      }, { /* Coeff Band 3 */
+        {  65, 187, 230 },
+        {  20, 148, 207 },
+        {   2,  97, 159 },
+        {   1,  68, 116 },
+        {   1,  40,  70 },
+        {   1,  14,  29 }
+      }, { /* Coeff Band 4 */
+        {  40, 194, 227 },
+        {   8, 147, 204 },
+        {   1,  94, 155 },
+        {   1,  65, 112 },
+        {   1,  39,  66 },
+        {   1,  14,  26 }
+      }, { /* Coeff Band 5 */
+        {  16, 208, 228 },
+        {   3, 151, 207 },
+        {   1,  98, 160 },
+        {   1,  67, 117 },
+        {   1,  41,  74 },
+        {   1,  17,  31 }
+      }
+    }
+  }
+};
+static const vp9_coeff_probs_model default_coef_probs_32x32[BLOCK_TYPES] = {
+  { /* block Type 0 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        {  17,  38, 140 },
+        {   7,  34,  80 },
+        {   1,  17,  29 }
+      }, { /* Coeff Band 1 */
+        {  37,  75, 128 },
+        {  41,  76, 128 },
+        {  26,  66, 116 },
+        {  12,  52,  94 },
+        {   2,  32,  55 },
+        {   1,  10,  16 }
+      }, { /* Coeff Band 2 */
+        {  50, 127, 154 },
+        {  37, 109, 152 },
+        {  16,  82, 121 },
+        {   5,  59,  85 },
+        {   1,  35,  54 },
+        {   1,  13,  20 }
+      }, { /* Coeff Band 3 */
+        {  40, 142, 167 },
+        {  17, 110, 157 },
+        {   2,  71, 112 },
+        {   1,  44,  72 },
+        {   1,  27,  45 },
+        {   1,  11,  17 }
+      }, { /* Coeff Band 4 */
+        {  30, 175, 188 },
+        {   9, 124, 169 },
+        {   1,  74, 116 },
+        {   1,  48,  78 },
+        {   1,  30,  49 },
+        {   1,  11,  18 }
+      }, { /* Coeff Band 5 */
+        {  10, 222, 223 },
+        {   2, 150, 194 },
+        {   1,  83, 128 },
+        {   1,  48,  79 },
+        {   1,  27,  45 },
+        {   1,  11,  17 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        {  36,  41, 235 },
+        {  29,  36, 193 },
+        {  10,  27, 111 }
+      }, { /* Coeff Band 1 */
+        {  85, 165, 222 },
+        { 177, 162, 215 },
+        { 110, 135, 195 },
+        {  57, 113, 168 },
+        {  23,  83, 120 },
+        {  10,  49,  61 }
+      }, { /* Coeff Band 2 */
+        {  85, 190, 223 },
+        {  36, 139, 200 },
+        {   5,  90, 146 },
+        {   1,  60, 103 },
+        {   1,  38,  65 },
+        {   1,  18,  30 }
+      }, { /* Coeff Band 3 */
+        {  72, 202, 223 },
+        {  23, 141, 199 },
+        {   2,  86, 140 },
+        {   1,  56,  97 },
+        {   1,  36,  61 },
+        {   1,  16,  27 }
+      }, { /* Coeff Band 4 */
+        {  55, 218, 225 },
+        {  13, 145, 200 },
+        {   1,  86, 141 },
+        {   1,  57,  99 },
+        {   1,  35,  61 },
+        {   1,  13,  22 }
+      }, { /* Coeff Band 5 */
+        {  15, 235, 212 },
+        {   1, 132, 184 },
+        {   1,  84, 139 },
+        {   1,  57,  97 },
+        {   1,  34,  56 },
+        {   1,  14,  23 }
+      }
+    }
+  }, { /* block Type 1 */
+    { /* Intra */
+      { /* Coeff Band 0 */
+        { 181,  21, 201 },
+        {  61,  37, 123 },
+        {  10,  38,  71 }
+      }, { /* Coeff Band 1 */
+        {  47, 106, 172 },
+        {  95, 104, 173 },
+        {  42,  93, 159 },
+        {  18,  77, 131 },
+        {   4,  50,  81 },
+        {   1,  17,  23 }
+      }, { /* Coeff Band 2 */
+        {  62, 147, 199 },
+        {  44, 130, 189 },
+        {  28, 102, 154 },
+        {  18,  75, 115 },
+        {   2,  44,  65 },
+        {   1,  12,  19 }
+      }, { /* Coeff Band 3 */
+        {  55, 153, 210 },
+        {  24, 130, 194 },
+        {   3,  93, 146 },
+        {   1,  61,  97 },
+        {   1,  31,  50 },
+        {   1,  10,  16 }
+      }, { /* Coeff Band 4 */
+        {  49, 186, 223 },
+        {  17, 148, 204 },
+        {   1,  96, 142 },
+        {   1,  53,  83 },
+        {   1,  26,  44 },
+        {   1,  11,  17 }
+      }, { /* Coeff Band 5 */
+        {  13, 217, 212 },
+        {   2, 136, 180 },
+        {   1,  78, 124 },
+        {   1,  50,  83 },
+        {   1,  29,  49 },
+        {   1,  14,  23 }
+      }
+    }, { /* Inter */
+      { /* Coeff Band 0 */
+        { 197,  13, 247 },
+        {  82,  17, 222 },
+        {  25,  17, 162 }
+      }, { /* Coeff Band 1 */
+        { 126, 186, 247 },
+        { 234, 191, 243 },
+        { 176, 177, 234 },
+        { 104, 158, 220 },
+        {  66, 128, 186 },
+        {  55,  90, 137 }
+      }, { /* Coeff Band 2 */
+        { 111, 197, 242 },
+        {  46, 158, 219 },
+        {   9, 104, 171 },
+        {   2,  65, 125 },
+        {   1,  44,  80 },
+        {   1,  17,  91 }
+      }, { /* Coeff Band 3 */
+        { 104, 208, 245 },
+        {  39, 168, 224 },
+        {   3, 109, 162 },
+        {   1,  79, 124 },
+        {   1,  50, 102 },
+        {   1,  43, 102 }
+      }, { /* Coeff Band 4 */
+        {  84, 220, 246 },
+        {  31, 177, 231 },
+        {   2, 115, 180 },
+        {   1,  79, 134 },
+        {   1,  55,  77 },
+        {   1,  60,  79 }
+      }, { /* Coeff Band 5 */
+        {  43, 243, 240 },
+        {   8, 180, 217 },
+        {   1, 115, 166 },
+        {   1,  84, 121 },
+        {   1,  51,  67 },
+        {   1,  16,   6 }
+      }
+    }
+  }
+};
+#endif
diff --git a/libvpx/vp9/common/vp9_entropy.c b/libvpx/vp9/common/vp9_entropy.c
new file mode 100644
index 0000000..080867e
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropy.c
@@ -0,0 +1,737 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx/vpx_integer.h"
+
+DECLARE_ALIGNED(16, const uint8_t, vp9_norm[256]) = {
+  0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
+  3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+};
+
+DECLARE_ALIGNED(16, const uint8_t,
+                vp9_coefband_trans_8x8plus[MAXBAND_INDEX + 1]) = {
+  0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
+  4, 4, 4, 4, 4, 5
+};
+
+DECLARE_ALIGNED(16, const uint8_t,
+                vp9_coefband_trans_4x4[MAXBAND_INDEX + 1]) = {
+  0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5,
+  5, 5, 5, 5, 5, 5
+};
+
+DECLARE_ALIGNED(16, const uint8_t, vp9_pt_energy_class[MAX_ENTROPY_TOKENS]) = {
+  0, 1, 2, 3, 3, 4, 4, 5, 5, 5, 5, 5
+};
+
+DECLARE_ALIGNED(16, const int, vp9_default_scan_4x4[16]) = {
+  0,  4,  1,  5,
+  8,  2, 12,  9,
+  3,  6, 13, 10,
+  7, 14, 11, 15,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_col_scan_4x4[16]) = {
+  0,  4,  8,  1,
+  12,  5,  9,  2,
+  13,  6, 10,  3,
+  7, 14, 11, 15,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_row_scan_4x4[16]) = {
+  0,  1,  4,  2,
+  5,  3,  6,  8,
+  9,  7, 12, 10,
+  13, 11, 14, 15,
+};
+
+DECLARE_ALIGNED(64, const int, vp9_default_scan_8x8[64]) = {
+  0,  8,  1, 16,  9,  2, 17, 24,
+  10,  3, 18, 25, 32, 11,  4, 26,
+  33, 19, 40, 12, 34, 27,  5, 41,
+  20, 48, 13, 35, 42, 28, 21,  6,
+  49, 56, 36, 43, 29,  7, 14, 50,
+  57, 44, 22, 37, 15, 51, 58, 30,
+  45, 23, 52, 59, 38, 31, 60, 53,
+  46, 39, 61, 54, 47, 62, 55, 63,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_col_scan_8x8[64]) = {
+  0,  8, 16,  1, 24,  9, 32, 17,
+  2, 40, 25, 10, 33, 18, 48,  3,
+  26, 41, 11, 56, 19, 34,  4, 49,
+  27, 42, 12, 35, 20, 57, 50, 28,
+  5, 43, 13, 36, 58, 51, 21, 44,
+  6, 29, 59, 37, 14, 52, 22,  7,
+  45, 60, 30, 15, 38, 53, 23, 46,
+  31, 61, 39, 54, 47, 62, 55, 63,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_row_scan_8x8[64]) = {
+  0,  1,  2,  8,  9,  3, 16, 10,
+  4, 17, 11, 24,  5, 18, 25, 12,
+  19, 26, 32,  6, 13, 20, 33, 27,
+  7, 34, 40, 21, 28, 41, 14, 35,
+  48, 42, 29, 36, 49, 22, 43, 15,
+  56, 37, 50, 44, 30, 57, 23, 51,
+  58, 45, 38, 52, 31, 59, 53, 46,
+  60, 39, 61, 47, 54, 55, 62, 63,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_default_scan_16x16[256]) = {
+  0,  16,   1,  32,  17,   2,  48,  33,  18,   3,  64,  34,  49,  19,  65,  80,
+  50,   4,  35,  66,  20,  81,  96,  51,   5,  36,  82,  97,  67, 112,  21,  52,
+  98,  37,  83, 113,   6,  68, 128,  53,  22,  99, 114,  84,   7, 129,  38,  69,
+  100, 115, 144, 130,  85,  54,  23,   8, 145,  39,  70, 116, 101, 131, 160, 146,
+  55,  86,  24,  71, 132, 117, 161,  40,   9, 102, 147, 176, 162,  87,  56,  25,
+  133, 118, 177, 148,  72, 103,  41, 163,  10, 192, 178,  88,  57, 134, 149, 119,
+  26, 164,  73, 104, 193,  42, 179, 208,  11, 135,  89, 165, 120, 150,  58, 194,
+  180,  27,  74, 209, 105, 151, 136,  43,  90, 224, 166, 195, 181, 121, 210,  59,
+  12, 152, 106, 167, 196,  75, 137, 225, 211, 240, 182, 122,  91,  28, 197,  13,
+  226, 168, 183, 153,  44, 212, 138, 107, 241,  60,  29, 123, 198, 184, 227, 169,
+  242,  76, 213, 154,  45,  92,  14, 199, 139,  61, 228, 214, 170, 185, 243, 108,
+  77, 155,  30,  15, 200, 229, 124, 215, 244,  93,  46, 186, 171, 201, 109, 140,
+  230,  62, 216, 245,  31, 125,  78, 156, 231,  47, 187, 202, 217,  94, 246, 141,
+  63, 232, 172, 110, 247, 157,  79, 218, 203, 126, 233, 188, 248,  95, 173, 142,
+  219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159, 251,
+  190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239, 255,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_col_scan_16x16[256]) = {
+  0,  16,  32,  48,   1,  64,  17,  80,  33,  96,  49,   2,  65, 112,  18,  81,
+  34, 128,  50,  97,   3,  66, 144,  19, 113,  35,  82, 160,  98,  51, 129,   4,
+  67, 176,  20, 114, 145,  83,  36,  99, 130,  52, 192,   5, 161,  68, 115,  21,
+  146,  84, 208, 177,  37, 131, 100,  53, 162, 224,  69,   6, 116, 193, 147,  85,
+  22, 240, 132,  38, 178, 101, 163,  54, 209, 117,  70,   7, 148, 194,  86, 179,
+  225,  23, 133,  39, 164,   8, 102, 210, 241,  55, 195, 118, 149,  71, 180,  24,
+  87, 226, 134, 165, 211,  40, 103,  56,  72, 150, 196, 242, 119,   9, 181, 227,
+  88, 166,  25, 135,  41, 104, 212,  57, 151, 197, 120,  73, 243, 182, 136, 167,
+  213,  89,  10, 228, 105, 152, 198,  26,  42, 121, 183, 244, 168,  58, 137, 229,
+  74, 214,  90, 153, 199, 184,  11, 106, 245,  27, 122, 230, 169,  43, 215,  59,
+  200, 138, 185, 246,  75,  12,  91, 154, 216, 231, 107,  28,  44, 201, 123, 170,
+  60, 247, 232,  76, 139,  13,  92, 217, 186, 248, 155, 108,  29, 124,  45, 202,
+  233, 171,  61,  14,  77, 140,  15, 249,  93,  30, 187, 156, 218,  46, 109, 125,
+  62, 172,  78, 203,  31, 141, 234,  94,  47, 188,  63, 157, 110, 250, 219,  79,
+  126, 204, 173, 142,  95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205, 236,
+  159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239, 255,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_row_scan_16x16[256]) = {
+  0,   1,   2,  16,   3,  17,   4,  18,  32,   5,  33,  19,   6,  34,  48,  20,
+  49,   7,  35,  21,  50,  64,   8,  36,  65,  22,  51,  37,  80,   9,  66,  52,
+  23,  38,  81,  67,  10,  53,  24,  82,  68,  96,  39,  11,  54,  83,  97,  69,
+  25,  98,  84,  40, 112,  55,  12,  70,  99, 113,  85,  26,  41,  56, 114, 100,
+  13,  71, 128,  86,  27, 115, 101, 129,  42,  57,  72, 116,  14,  87, 130, 102,
+  144,  73, 131, 117,  28,  58,  15,  88,  43, 145, 103, 132, 146, 118,  74, 160,
+  89, 133, 104,  29,  59, 147, 119,  44, 161, 148,  90, 105, 134, 162, 120, 176,
+  75, 135, 149,  30,  60, 163, 177,  45, 121,  91, 106, 164, 178, 150, 192, 136,
+  165, 179,  31, 151, 193,  76, 122,  61, 137, 194, 107, 152, 180, 208,  46, 166,
+  167, 195,  92, 181, 138, 209, 123, 153, 224, 196,  77, 168, 210, 182, 240, 108,
+  197,  62, 154, 225, 183, 169, 211,  47, 139,  93, 184, 226, 212, 241, 198, 170,
+  124, 155, 199,  78, 213, 185, 109, 227, 200,  63, 228, 242, 140, 214, 171, 186,
+  156, 229, 243, 125,  94, 201, 244, 215, 216, 230, 141, 187, 202,  79, 172, 110,
+  157, 245, 217, 231,  95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111, 158,
+  188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220, 175,
+  190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254, 255,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_default_scan_32x32[1024]) = {
+  0,   32,    1,   64,   33,    2,   96,   65,   34,  128,    3,   97,   66,  160,  129,   35,   98,    4,   67,  130,  161,  192,   36,   99,  224,    5,  162,  193,   68,  131,   37,  100,
+  225,  194,  256,  163,   69,  132,    6,  226,  257,  288,  195,  101,  164,   38,  258,    7,  227,  289,  133,  320,   70,  196,  165,  290,  259,  228,   39,  321,  102,  352,    8,  197,
+  71,  134,  322,  291,  260,  353,  384,  229,  166,  103,   40,  354,  323,  292,  135,  385,  198,  261,   72,    9,  416,  167,  386,  355,  230,  324,  104,  293,   41,  417,  199,  136,
+  262,  387,  448,  325,  356,   10,   73,  418,  231,  168,  449,  294,  388,  105,  419,  263,   42,  200,  357,  450,  137,  480,   74,  326,  232,   11,  389,  169,  295,  420,  106,  451,
+  481,  358,  264,  327,  201,   43,  138,  512,  482,  390,  296,  233,  170,  421,   75,  452,  359,   12,  513,  265,  483,  328,  107,  202,  514,  544,  422,  391,  453,  139,   44,  234,
+  484,  297,  360,  171,   76,  515,  545,  266,  329,  454,   13,  423,  203,  108,  546,  485,  576,  298,  235,  140,  361,  330,  172,  547,   45,  455,  267,  577,  486,   77,  204,  362,
+  608,   14,  299,  578,  109,  236,  487,  609,  331,  141,  579,   46,   15,  173,  610,  363,   78,  205,   16,  110,  237,  611,  142,   47,  174,   79,  206,   17,  111,  238,   48,  143,
+  80,  175,  112,  207,   49,   18,  239,   81,  113,   19,   50,   82,  114,   51,   83,  115,  640,  516,  392,  268,  144,   20,  672,  641,  548,  517,  424,  393,  300,  269,  176,  145,
+  52,   21,  704,  673,  642,  580,  549,  518,  456,  425,  394,  332,  301,  270,  208,  177,  146,   84,   53,   22,  736,  705,  674,  643,  612,  581,  550,  519,  488,  457,  426,  395,
+  364,  333,  302,  271,  240,  209,  178,  147,  116,   85,   54,   23,  737,  706,  675,  613,  582,  551,  489,  458,  427,  365,  334,  303,  241,  210,  179,  117,   86,   55,  738,  707,
+  614,  583,  490,  459,  366,  335,  242,  211,  118,   87,  739,  615,  491,  367,  243,  119,  768,  644,  520,  396,  272,  148,   24,  800,  769,  676,  645,  552,  521,  428,  397,  304,
+  273,  180,  149,   56,   25,  832,  801,  770,  708,  677,  646,  584,  553,  522,  460,  429,  398,  336,  305,  274,  212,  181,  150,   88,   57,   26,  864,  833,  802,  771,  740,  709,
+  678,  647,  616,  585,  554,  523,  492,  461,  430,  399,  368,  337,  306,  275,  244,  213,  182,  151,  120,   89,   58,   27,  865,  834,  803,  741,  710,  679,  617,  586,  555,  493,
+  462,  431,  369,  338,  307,  245,  214,  183,  121,   90,   59,  866,  835,  742,  711,  618,  587,  494,  463,  370,  339,  246,  215,  122,   91,  867,  743,  619,  495,  371,  247,  123,
+  896,  772,  648,  524,  400,  276,  152,   28,  928,  897,  804,  773,  680,  649,  556,  525,  432,  401,  308,  277,  184,  153,   60,   29,  960,  929,  898,  836,  805,  774,  712,  681,
+  650,  588,  557,  526,  464,  433,  402,  340,  309,  278,  216,  185,  154,   92,   61,   30,  992,  961,  930,  899,  868,  837,  806,  775,  744,  713,  682,  651,  620,  589,  558,  527,
+  496,  465,  434,  403,  372,  341,  310,  279,  248,  217,  186,  155,  124,   93,   62,   31,  993,  962,  931,  869,  838,  807,  745,  714,  683,  621,  590,  559,  497,  466,  435,  373,
+  342,  311,  249,  218,  187,  125,   94,   63,  994,  963,  870,  839,  746,  715,  622,  591,  498,  467,  374,  343,  250,  219,  126,   95,  995,  871,  747,  623,  499,  375,  251,  127,
+  900,  776,  652,  528,  404,  280,  156,  932,  901,  808,  777,  684,  653,  560,  529,  436,  405,  312,  281,  188,  157,  964,  933,  902,  840,  809,  778,  716,  685,  654,  592,  561,
+  530,  468,  437,  406,  344,  313,  282,  220,  189,  158,  996,  965,  934,  903,  872,  841,  810,  779,  748,  717,  686,  655,  624,  593,  562,  531,  500,  469,  438,  407,  376,  345,
+  314,  283,  252,  221,  190,  159,  997,  966,  935,  873,  842,  811,  749,  718,  687,  625,  594,  563,  501,  470,  439,  377,  346,  315,  253,  222,  191,  998,  967,  874,  843,  750,
+  719,  626,  595,  502,  471,  378,  347,  254,  223,  999,  875,  751,  627,  503,  379,  255,  904,  780,  656,  532,  408,  284,  936,  905,  812,  781,  688,  657,  564,  533,  440,  409,
+  316,  285,  968,  937,  906,  844,  813,  782,  720,  689,  658,  596,  565,  534,  472,  441,  410,  348,  317,  286, 1000,  969,  938,  907,  876,  845,  814,  783,  752,  721,  690,  659,
+  628,  597,  566,  535,  504,  473,  442,  411,  380,  349,  318,  287, 1001,  970,  939,  877,  846,  815,  753,  722,  691,  629,  598,  567,  505,  474,  443,  381,  350,  319, 1002,  971,
+  878,  847,  754,  723,  630,  599,  506,  475,  382,  351, 1003,  879,  755,  631,  507,  383,  908,  784,  660,  536,  412,  940,  909,  816,  785,  692,  661,  568,  537,  444,  413,  972,
+  941,  910,  848,  817,  786,  724,  693,  662,  600,  569,  538,  476,  445,  414, 1004,  973,  942,  911,  880,  849,  818,  787,  756,  725,  694,  663,  632,  601,  570,  539,  508,  477,
+  446,  415, 1005,  974,  943,  881,  850,  819,  757,  726,  695,  633,  602,  571,  509,  478,  447, 1006,  975,  882,  851,  758,  727,  634,  603,  510,  479, 1007,  883,  759,  635,  511,
+  912,  788,  664,  540,  944,  913,  820,  789,  696,  665,  572,  541,  976,  945,  914,  852,  821,  790,  728,  697,  666,  604,  573,  542, 1008,  977,  946,  915,  884,  853,  822,  791,
+  760,  729,  698,  667,  636,  605,  574,  543, 1009,  978,  947,  885,  854,  823,  761,  730,  699,  637,  606,  575, 1010,  979,  886,  855,  762,  731,  638,  607, 1011,  887,  763,  639,
+  916,  792,  668,  948,  917,  824,  793,  700,  669,  980,  949,  918,  856,  825,  794,  732,  701,  670, 1012,  981,  950,  919,  888,  857,  826,  795,  764,  733,  702,  671, 1013,  982,
+  951,  889,  858,  827,  765,  734,  703, 1014,  983,  890,  859,  766,  735, 1015,  891,  767,  920,  796,  952,  921,  828,  797,  984,  953,  922,  860,  829,  798, 1016,  985,  954,  923,
+  892,  861,  830,  799, 1017,  986,  955,  893,  862,  831, 1018,  987,  894,  863, 1019,  895,  924,  956,  925,  988,  957,  926, 1020,  989,  958,  927, 1021,  990,  959, 1022,  991, 1023,
+};
+
+/* Array indices are identical to previously-existing CONTEXT_NODE indices */
+
+const vp9_tree_index vp9_coef_tree[ 22] =     /* corresponding _CONTEXT_NODEs */
+{
+#if CONFIG_BALANCED_COEFTREE
+  -ZERO_TOKEN, 2,                             /* 0 = ZERO */
+  -DCT_EOB_TOKEN, 4,                          /* 1 = EOB  */
+#else
+  -DCT_EOB_TOKEN, 2,                          /* 0 = EOB */
+  -ZERO_TOKEN, 4,                             /* 1 = ZERO */
+#endif
+  -ONE_TOKEN, 6,                              /* 2 = ONE */
+  8, 12,                                      /* 3 = LOW_VAL */
+  -TWO_TOKEN, 10,                            /* 4 = TWO */
+  -THREE_TOKEN, -FOUR_TOKEN,                /* 5 = THREE */
+  14, 16,                                   /* 6 = HIGH_LOW */
+  -DCT_VAL_CATEGORY1, -DCT_VAL_CATEGORY2,   /* 7 = CAT_ONE */
+  18, 20,                                   /* 8 = CAT_THREEFOUR */
+  -DCT_VAL_CATEGORY3, -DCT_VAL_CATEGORY4,   /* 9 = CAT_THREE */
+  -DCT_VAL_CATEGORY5, -DCT_VAL_CATEGORY6    /* 10 = CAT_FIVE */
+};
+
+struct vp9_token vp9_coef_encodings[MAX_ENTROPY_TOKENS];
+
+/* Trees for extra bits.  Probabilities are constant and
+   do not depend on previously encoded bits */
+
+static const vp9_prob Pcat1[] = { 159};
+static const vp9_prob Pcat2[] = { 165, 145};
+static const vp9_prob Pcat3[] = { 173, 148, 140};
+static const vp9_prob Pcat4[] = { 176, 155, 140, 135};
+static const vp9_prob Pcat5[] = { 180, 157, 141, 134, 130};
+static const vp9_prob Pcat6[] = {
+  254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129
+};
+
+const vp9_tree_index vp9_coefmodel_tree[6] = {
+#if CONFIG_BALANCED_COEFTREE
+  -ZERO_TOKEN, 2,
+  -DCT_EOB_MODEL_TOKEN, 4,
+#else
+  -DCT_EOB_MODEL_TOKEN, 2,                      /* 0 = EOB */
+  -ZERO_TOKEN, 4,                               /* 1 = ZERO */
+#endif
+  -ONE_TOKEN, -TWO_TOKEN,
+};
+
+// Model obtained from a 2-sided zero-centerd distribuition derived
+// from a Pareto distribution. The cdf of the distribution is:
+// cdf(x) = 0.5 + 0.5 * sgn(x) * [1 - {alpha/(alpha + |x|)} ^ beta]
+//
+// For a given beta and a given probablity of the 1-node, the alpha
+// is first solved, and then the {alpha, beta} pair is used to generate
+// the probabilities for the rest of the nodes.
+
+// beta = 8
+const vp9_prob vp9_modelcoefprobs_pareto8[COEFPROB_MODELS][MODEL_NODES] = {
+  {  3,  86, 128,   6,  86,  23,  88,  29},
+  {  9,  86, 129,  17,  88,  61,  94,  76},
+  { 15,  87, 129,  28,  89,  93, 100, 110},
+  { 20,  88, 130,  38,  91, 118, 106, 136},
+  { 26,  89, 131,  48,  92, 139, 111, 156},
+  { 31,  90, 131,  58,  94, 156, 117, 171},
+  { 37,  90, 132,  66,  95, 171, 122, 184},
+  { 42,  91, 132,  75,  97, 183, 127, 194},
+  { 47,  92, 133,  83,  98, 193, 132, 202},
+  { 52,  93, 133,  90, 100, 201, 137, 208},
+  { 57,  94, 134,  98, 101, 208, 142, 214},
+  { 62,  94, 135, 105, 103, 214, 146, 218},
+  { 66,  95, 135, 111, 104, 219, 151, 222},
+  { 71,  96, 136, 117, 106, 224, 155, 225},
+  { 76,  97, 136, 123, 107, 227, 159, 228},
+  { 80,  98, 137, 129, 109, 231, 162, 231},
+  { 84,  98, 138, 134, 110, 234, 166, 233},
+  { 89,  99, 138, 140, 112, 236, 170, 235},
+  { 93, 100, 139, 145, 113, 238, 173, 236},
+  { 97, 101, 140, 149, 115, 240, 176, 238},
+  {101, 102, 140, 154, 116, 242, 179, 239},
+  {105, 103, 141, 158, 118, 243, 182, 240},
+  {109, 104, 141, 162, 119, 244, 185, 241},
+  {113, 104, 142, 166, 120, 245, 187, 242},
+  {116, 105, 143, 170, 122, 246, 190, 243},
+  {120, 106, 143, 173, 123, 247, 192, 244},
+  {123, 107, 144, 177, 125, 248, 195, 244},
+  {127, 108, 145, 180, 126, 249, 197, 245},
+  {130, 109, 145, 183, 128, 249, 199, 245},
+  {134, 110, 146, 186, 129, 250, 201, 246},
+  {137, 111, 147, 189, 131, 251, 203, 246},
+  {140, 112, 147, 192, 132, 251, 205, 247},
+  {143, 113, 148, 194, 133, 251, 207, 247},
+  {146, 114, 149, 197, 135, 252, 208, 248},
+  {149, 115, 149, 199, 136, 252, 210, 248},
+  {152, 115, 150, 201, 138, 252, 211, 248},
+  {155, 116, 151, 204, 139, 253, 213, 249},
+  {158, 117, 151, 206, 140, 253, 214, 249},
+  {161, 118, 152, 208, 142, 253, 216, 249},
+  {163, 119, 153, 210, 143, 253, 217, 249},
+  {166, 120, 153, 212, 144, 254, 218, 250},
+  {168, 121, 154, 213, 146, 254, 220, 250},
+  {171, 122, 155, 215, 147, 254, 221, 250},
+  {173, 123, 155, 217, 148, 254, 222, 250},
+  {176, 124, 156, 218, 150, 254, 223, 250},
+  {178, 125, 157, 220, 151, 254, 224, 251},
+  {180, 126, 157, 221, 152, 254, 225, 251},
+  {183, 127, 158, 222, 153, 254, 226, 251},
+  {185, 128, 159, 224, 155, 255, 227, 251},
+  {187, 129, 160, 225, 156, 255, 228, 251},
+  {189, 131, 160, 226, 157, 255, 228, 251},
+  {191, 132, 161, 227, 159, 255, 229, 251},
+  {193, 133, 162, 228, 160, 255, 230, 252},
+  {195, 134, 163, 230, 161, 255, 231, 252},
+  {197, 135, 163, 231, 162, 255, 231, 252},
+  {199, 136, 164, 232, 163, 255, 232, 252},
+  {201, 137, 165, 233, 165, 255, 233, 252},
+  {202, 138, 166, 233, 166, 255, 233, 252},
+  {204, 139, 166, 234, 167, 255, 234, 252},
+  {206, 140, 167, 235, 168, 255, 235, 252},
+  {207, 141, 168, 236, 169, 255, 235, 252},
+  {209, 142, 169, 237, 171, 255, 236, 252},
+  {210, 144, 169, 237, 172, 255, 236, 252},
+  {212, 145, 170, 238, 173, 255, 237, 252},
+  {214, 146, 171, 239, 174, 255, 237, 253},
+  {215, 147, 172, 240, 175, 255, 238, 253},
+  {216, 148, 173, 240, 176, 255, 238, 253},
+  {218, 149, 173, 241, 177, 255, 239, 253},
+  {219, 150, 174, 241, 179, 255, 239, 253},
+  {220, 152, 175, 242, 180, 255, 240, 253},
+  {222, 153, 176, 242, 181, 255, 240, 253},
+  {223, 154, 177, 243, 182, 255, 240, 253},
+  {224, 155, 178, 244, 183, 255, 241, 253},
+  {225, 156, 178, 244, 184, 255, 241, 253},
+  {226, 158, 179, 244, 185, 255, 242, 253},
+  {228, 159, 180, 245, 186, 255, 242, 253},
+  {229, 160, 181, 245, 187, 255, 242, 253},
+  {230, 161, 182, 246, 188, 255, 243, 253},
+  {231, 163, 183, 246, 189, 255, 243, 253},
+  {232, 164, 184, 247, 190, 255, 243, 253},
+  {233, 165, 185, 247, 191, 255, 244, 253},
+  {234, 166, 185, 247, 192, 255, 244, 253},
+  {235, 168, 186, 248, 193, 255, 244, 253},
+  {236, 169, 187, 248, 194, 255, 244, 253},
+  {236, 170, 188, 248, 195, 255, 245, 253},
+  {237, 171, 189, 249, 196, 255, 245, 254},
+  {238, 173, 190, 249, 197, 255, 245, 254},
+  {239, 174, 191, 249, 198, 255, 245, 254},
+  {240, 175, 192, 249, 199, 255, 246, 254},
+  {240, 177, 193, 250, 200, 255, 246, 254},
+  {241, 178, 194, 250, 201, 255, 246, 254},
+  {242, 179, 195, 250, 202, 255, 246, 254},
+  {242, 181, 196, 250, 203, 255, 247, 254},
+  {243, 182, 197, 251, 204, 255, 247, 254},
+  {244, 184, 198, 251, 205, 255, 247, 254},
+  {244, 185, 199, 251, 206, 255, 247, 254},
+  {245, 186, 200, 251, 207, 255, 247, 254},
+  {246, 188, 201, 252, 207, 255, 248, 254},
+  {246, 189, 202, 252, 208, 255, 248, 254},
+  {247, 191, 203, 252, 209, 255, 248, 254},
+  {247, 192, 204, 252, 210, 255, 248, 254},
+  {248, 194, 205, 252, 211, 255, 248, 254},
+  {248, 195, 206, 252, 212, 255, 249, 254},
+  {249, 197, 207, 253, 213, 255, 249, 254},
+  {249, 198, 208, 253, 214, 255, 249, 254},
+  {250, 200, 210, 253, 215, 255, 249, 254},
+  {250, 201, 211, 253, 215, 255, 249, 254},
+  {250, 203, 212, 253, 216, 255, 249, 254},
+  {251, 204, 213, 253, 217, 255, 250, 254},
+  {251, 206, 214, 254, 218, 255, 250, 254},
+  {252, 207, 216, 254, 219, 255, 250, 254},
+  {252, 209, 217, 254, 220, 255, 250, 254},
+  {252, 211, 218, 254, 221, 255, 250, 254},
+  {253, 213, 219, 254, 222, 255, 250, 254},
+  {253, 214, 221, 254, 223, 255, 250, 254},
+  {253, 216, 222, 254, 224, 255, 251, 254},
+  {253, 218, 224, 254, 225, 255, 251, 254},
+  {254, 220, 225, 254, 225, 255, 251, 254},
+  {254, 222, 227, 255, 226, 255, 251, 254},
+  {254, 224, 228, 255, 227, 255, 251, 254},
+  {254, 226, 230, 255, 228, 255, 251, 254},
+  {255, 228, 231, 255, 230, 255, 251, 254},
+  {255, 230, 233, 255, 231, 255, 252, 254},
+  {255, 232, 235, 255, 232, 255, 252, 254},
+  {255, 235, 237, 255, 233, 255, 252, 254},
+  {255, 238, 240, 255, 235, 255, 252, 255},
+  {255, 241, 243, 255, 236, 255, 252, 254},
+  {255, 246, 247, 255, 239, 255, 253, 255}
+};
+
+static void extend_model_to_full_distribution(vp9_prob p,
+                                              vp9_prob *tree_probs) {
+  const int l = ((p - 1) / 2);
+  const vp9_prob (*model)[MODEL_NODES];
+  model = vp9_modelcoefprobs_pareto8;
+  if (p & 1) {
+    vpx_memcpy(tree_probs + UNCONSTRAINED_NODES,
+               model[l], MODEL_NODES * sizeof(vp9_prob));
+  } else {
+    // interpolate
+    int i;
+    for (i = UNCONSTRAINED_NODES; i < ENTROPY_NODES; ++i)
+      tree_probs[i] = (model[l][i - UNCONSTRAINED_NODES] +
+                       model[l + 1][i - UNCONSTRAINED_NODES]) >> 1;
+  }
+}
+
+void vp9_model_to_full_probs(const vp9_prob *model, vp9_prob *full) {
+  if (full != model)
+    vpx_memcpy(full, model, sizeof(vp9_prob) * UNCONSTRAINED_NODES);
+  extend_model_to_full_distribution(model[PIVOT_NODE], full);
+}
+
+void vp9_model_to_full_probs_sb(
+    vp9_prob model[COEF_BANDS][PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES],
+    vp9_prob full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES]) {
+  int c, p;
+  for (c = 0; c < COEF_BANDS; ++c)
+    for (p = 0; p < PREV_COEF_CONTEXTS; ++p) {
+      vp9_model_to_full_probs(model[c][p], full[c][p]);
+    }
+}
+
+static vp9_tree_index cat1[2], cat2[4], cat3[6], cat4[8], cat5[10], cat6[28];
+
+static void init_bit_tree(vp9_tree_index *p, int n) {
+  int i = 0;
+
+  while (++i < n) {
+    p[0] = p[1] = i << 1;
+    p += 2;
+  }
+
+  p[0] = p[1] = 0;
+}
+
+static void init_bit_trees() {
+  init_bit_tree(cat1, 1);
+  init_bit_tree(cat2, 2);
+  init_bit_tree(cat3, 3);
+  init_bit_tree(cat4, 4);
+  init_bit_tree(cat5, 5);
+  init_bit_tree(cat6, 14);
+}
+
+vp9_extra_bit vp9_extra_bits[12] = {
+  { 0, 0, 0, 0},
+  { 0, 0, 0, 1},
+  { 0, 0, 0, 2},
+  { 0, 0, 0, 3},
+  { 0, 0, 0, 4},
+  { cat1, Pcat1, 1, 5},
+  { cat2, Pcat2, 2, 7},
+  { cat3, Pcat3, 3, 11},
+  { cat4, Pcat4, 4, 19},
+  { cat5, Pcat5, 5, 35},
+  { cat6, Pcat6, 14, 67},
+  { 0, 0, 0, 0}
+};
+
+#include "vp9/common/vp9_default_coef_probs.h"
+
+// This function updates and then returns n AC coefficient context
+// This is currently a placeholder function to allow experimentation
+// using various context models based on the energy earlier tokens
+// within the current block.
+//
+// For now it just returns the previously used context.
+#define MAX_NEIGHBORS 2
+int vp9_get_coef_context(const int *scan, const int *neighbors,
+                         int nb_pad, uint8_t *token_cache, int c, int l) {
+  int eob = l;
+  assert(nb_pad == MAX_NEIGHBORS);
+  if (c == eob) {
+    return 0;
+  } else {
+    int ctx;
+    assert(neighbors[MAX_NEIGHBORS * c + 0] >= 0);
+    if (neighbors[MAX_NEIGHBORS * c + 1] >= 0) {
+      ctx = (1 + token_cache[scan[neighbors[MAX_NEIGHBORS * c + 0]]] +
+             token_cache[scan[neighbors[MAX_NEIGHBORS * c + 1]]]) >> 1;
+    } else {
+      ctx = token_cache[scan[neighbors[MAX_NEIGHBORS * c + 0]]];
+    }
+    return ctx;
+  }
+};
+
+void vp9_default_coef_probs(VP9_COMMON *pc) {
+  vpx_memcpy(pc->fc.coef_probs[TX_4X4], default_coef_probs_4x4,
+             sizeof(pc->fc.coef_probs[TX_4X4]));
+  vpx_memcpy(pc->fc.coef_probs[TX_8X8], default_coef_probs_8x8,
+             sizeof(pc->fc.coef_probs[TX_8X8]));
+  vpx_memcpy(pc->fc.coef_probs[TX_16X16], default_coef_probs_16x16,
+             sizeof(pc->fc.coef_probs[TX_16X16]));
+  vpx_memcpy(pc->fc.coef_probs[TX_32X32], default_coef_probs_32x32,
+             sizeof(pc->fc.coef_probs[TX_32X32]));
+}
+
+// Neighborhood 5-tuples for various scans and blocksizes,
+// in {top, left, topleft, topright, bottomleft} order
+// for each position in raster scan order.
+// -1 indicates the neighbor does not exist.
+DECLARE_ALIGNED(16, int,
+                vp9_default_scan_4x4_neighbors[16 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_col_scan_4x4_neighbors[16 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_row_scan_4x4_neighbors[16 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_col_scan_8x8_neighbors[64 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_row_scan_8x8_neighbors[64 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_default_scan_8x8_neighbors[64 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_col_scan_16x16_neighbors[256 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_row_scan_16x16_neighbors[256 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_default_scan_16x16_neighbors[256 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int,
+                vp9_default_scan_32x32_neighbors[1024 * MAX_NEIGHBORS]);
+
+static int find_in_scan(const int *scan, int l, int idx) {
+  int n, l2 = l * l;
+  for (n = 0; n < l2; n++) {
+    int rc = scan[n];
+    if (rc == idx)
+      return  n;
+  }
+  assert(0);
+  return -1;
+}
+static void init_scan_neighbors(const int *scan, int l, int *neighbors,
+                                int max_neighbors) {
+  int l2 = l * l;
+  int n, i, j;
+
+  for (n = 0; n < l2; n++) {
+    int rc = scan[n];
+    assert(max_neighbors == MAX_NEIGHBORS);
+    i = rc / l;
+    j = rc % l;
+    if (i > 0 && j > 0) {
+      // col/row scan is used for adst/dct, and generally means that
+      // energy decreases to zero much faster in the dimension in
+      // which ADST is used compared to the direction in which DCT
+      // is used. Likewise, we find much higher correlation between
+      // coefficients within the direction in which DCT is used.
+      // Therefore, if we use ADST/DCT, prefer the DCT neighbor coeff
+      // as a context. If ADST or DCT is used in both directions, we
+      // use the combination of the two as a context.
+      int a = find_in_scan(scan, l, (i - 1) * l + j);
+      int b = find_in_scan(scan, l,  i      * l + j - 1);
+      if (scan == vp9_col_scan_4x4 || scan == vp9_col_scan_8x8 ||
+          scan == vp9_col_scan_16x16) {
+        neighbors[max_neighbors * n + 0] = a;
+        neighbors[max_neighbors * n + 1] = -1;
+      } else if (scan == vp9_row_scan_4x4 || scan == vp9_row_scan_8x8 ||
+                 scan == vp9_row_scan_16x16) {
+        neighbors[max_neighbors * n + 0] = b;
+        neighbors[max_neighbors * n + 1] = -1;
+      } else {
+        neighbors[max_neighbors * n + 0] = a;
+        neighbors[max_neighbors * n + 1] = b;
+      }
+    } else if (i > 0) {
+      neighbors[max_neighbors * n + 0] = find_in_scan(scan, l, (i - 1) * l + j);
+      neighbors[max_neighbors * n + 1] = -1;
+    } else if (j > 0) {
+      neighbors[max_neighbors * n + 0] =
+          find_in_scan(scan, l,  i      * l + j - 1);
+      neighbors[max_neighbors * n + 1] = -1;
+    } else {
+      assert(n == 0);
+      // dc predictor doesn't use previous tokens
+      neighbors[max_neighbors * n + 0] = -1;
+    }
+    assert(neighbors[max_neighbors * n + 0] < n);
+  }
+}
+
+void vp9_init_neighbors() {
+  init_scan_neighbors(vp9_default_scan_4x4, 4,
+                      vp9_default_scan_4x4_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_row_scan_4x4, 4,
+                      vp9_row_scan_4x4_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_col_scan_4x4, 4,
+                      vp9_col_scan_4x4_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_default_scan_8x8, 8,
+                      vp9_default_scan_8x8_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_row_scan_8x8, 8,
+                      vp9_row_scan_8x8_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_col_scan_8x8, 8,
+                      vp9_col_scan_8x8_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_default_scan_16x16, 16,
+                      vp9_default_scan_16x16_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_row_scan_16x16, 16,
+                      vp9_row_scan_16x16_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_col_scan_16x16, 16,
+                      vp9_col_scan_16x16_neighbors, MAX_NEIGHBORS);
+  init_scan_neighbors(vp9_default_scan_32x32, 32,
+                      vp9_default_scan_32x32_neighbors, MAX_NEIGHBORS);
+}
+
+const int *vp9_get_coef_neighbors_handle(const int *scan, int *pad) {
+  if (scan == vp9_default_scan_4x4) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_default_scan_4x4_neighbors;
+  } else if (scan == vp9_row_scan_4x4) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_row_scan_4x4_neighbors;
+  } else if (scan == vp9_col_scan_4x4) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_col_scan_4x4_neighbors;
+  } else if (scan == vp9_default_scan_8x8) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_default_scan_8x8_neighbors;
+  } else if (scan == vp9_row_scan_8x8) {
+    *pad = 2;
+    return vp9_row_scan_8x8_neighbors;
+  } else if (scan == vp9_col_scan_8x8) {
+    *pad = 2;
+    return vp9_col_scan_8x8_neighbors;
+  } else if (scan == vp9_default_scan_16x16) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_default_scan_16x16_neighbors;
+  } else if (scan == vp9_row_scan_16x16) {
+    *pad = 2;
+    return vp9_row_scan_16x16_neighbors;
+  } else if (scan == vp9_col_scan_16x16) {
+    *pad = 2;
+    return vp9_col_scan_16x16_neighbors;
+  } else if (scan == vp9_default_scan_32x32) {
+    *pad = MAX_NEIGHBORS;
+    return vp9_default_scan_32x32_neighbors;
+  } else {
+    assert(0);
+    return NULL;
+  }
+}
+
+void vp9_coef_tree_initialize() {
+  vp9_init_neighbors();
+  init_bit_trees();
+  vp9_tokens_from_tree(vp9_coef_encodings, vp9_coef_tree);
+}
+
+// #define COEF_COUNT_TESTING
+
+#define COEF_COUNT_SAT 24
+#define COEF_MAX_UPDATE_FACTOR 112
+#define COEF_COUNT_SAT_KEY 24
+#define COEF_MAX_UPDATE_FACTOR_KEY 112
+#define COEF_COUNT_SAT_AFTER_KEY 24
+#define COEF_MAX_UPDATE_FACTOR_AFTER_KEY 128
+
+void vp9_full_to_model_count(unsigned int *model_count,
+                             unsigned int *full_count) {
+  int n;
+  model_count[ZERO_TOKEN] = full_count[ZERO_TOKEN];
+  model_count[ONE_TOKEN] = full_count[ONE_TOKEN];
+  model_count[TWO_TOKEN] = full_count[TWO_TOKEN];
+  for (n = THREE_TOKEN; n < DCT_EOB_TOKEN; ++n)
+    model_count[TWO_TOKEN] += full_count[n];
+  model_count[DCT_EOB_MODEL_TOKEN] = full_count[DCT_EOB_TOKEN];
+}
+
+void vp9_full_to_model_counts(
+    vp9_coeff_count_model *model_count, vp9_coeff_count *full_count) {
+  int i, j, k, l;
+  for (i = 0; i < BLOCK_TYPES; ++i)
+    for (j = 0; j < REF_TYPES; ++j)
+      for (k = 0; k < COEF_BANDS; ++k)
+        for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
+          if (l >= 3 && k == 0)
+            continue;
+          vp9_full_to_model_count(model_count[i][j][k][l],
+                                  full_count[i][j][k][l]);
+        }
+}
+
+static void adapt_coef_probs(VP9_COMMON *cm, TX_SIZE txfm_size,
+                             int count_sat, int update_factor) {
+  vp9_coeff_probs_model *dst_coef_probs = cm->fc.coef_probs[txfm_size];
+  vp9_coeff_probs_model *pre_coef_probs = cm->fc.pre_coef_probs[txfm_size];
+  vp9_coeff_count_model *coef_counts = cm->fc.coef_counts[txfm_size];
+  unsigned int (*eob_branch_count)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS] =
+      cm->fc.eob_branch_counts[txfm_size];
+  int t, i, j, k, l, count;
+  int factor;
+  unsigned int branch_ct[UNCONSTRAINED_NODES][2];
+  vp9_prob coef_probs[UNCONSTRAINED_NODES];
+  int entropy_nodes_adapt = UNCONSTRAINED_NODES;
+
+  for (i = 0; i < BLOCK_TYPES; ++i)
+    for (j = 0; j < REF_TYPES; ++j)
+      for (k = 0; k < COEF_BANDS; ++k)
+        for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
+          if (l >= 3 && k == 0)
+            continue;
+          vp9_tree_probs_from_distribution(
+              vp9_coefmodel_tree,
+              coef_probs, branch_ct,
+              coef_counts[i][j][k][l], 0);
+#if CONFIG_BALANCED_COEFTREE
+          branch_ct[1][1] = eob_branch_count[i][j][k][l] - branch_ct[1][0];
+          coef_probs[1] = get_binary_prob(branch_ct[1][0], branch_ct[1][1]);
+#else
+          branch_ct[0][1] = eob_branch_count[i][j][k][l] - branch_ct[0][0];
+          coef_probs[0] = get_binary_prob(branch_ct[0][0], branch_ct[0][1]);
+#endif
+          for (t = 0; t < entropy_nodes_adapt; ++t) {
+            count = branch_ct[t][0] + branch_ct[t][1];
+            count = count > count_sat ? count_sat : count;
+            factor = (update_factor * count / count_sat);
+            dst_coef_probs[i][j][k][l][t] =
+                weighted_prob(pre_coef_probs[i][j][k][l][t],
+                              coef_probs[t], factor);
+          }
+        }
+}
+
+void vp9_adapt_coef_probs(VP9_COMMON *cm) {
+  TX_SIZE t;
+  int count_sat;
+  int update_factor; /* denominator 256 */
+
+  if ((cm->frame_type == KEY_FRAME) || cm->intra_only) {
+    update_factor = COEF_MAX_UPDATE_FACTOR_KEY;
+    count_sat = COEF_COUNT_SAT_KEY;
+  } else if (cm->last_frame_type == KEY_FRAME) {
+    update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY;  /* adapt quickly */
+    count_sat = COEF_COUNT_SAT_AFTER_KEY;
+  } else {
+    update_factor = COEF_MAX_UPDATE_FACTOR;
+    count_sat = COEF_COUNT_SAT;
+  }
+  for (t = TX_4X4; t <= TX_32X32; t++)
+    adapt_coef_probs(cm, t, count_sat, update_factor);
+}
diff --git a/libvpx/vp9/common/vp9_entropy.h b/libvpx/vp9/common/vp9_entropy.h
new file mode 100644
index 0000000..7f2bf3d
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropy.h
@@ -0,0 +1,225 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ENTROPY_H_
+#define VP9_COMMON_VP9_ENTROPY_H_
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_treecoder.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_common.h"
+
+/* Coefficient token alphabet */
+
+#define ZERO_TOKEN              0       /* 0         Extra Bits 0+0 */
+#define ONE_TOKEN               1       /* 1         Extra Bits 0+1 */
+#define TWO_TOKEN               2       /* 2         Extra Bits 0+1 */
+#define THREE_TOKEN             3       /* 3         Extra Bits 0+1 */
+#define FOUR_TOKEN              4       /* 4         Extra Bits 0+1 */
+#define DCT_VAL_CATEGORY1       5       /* 5-6       Extra Bits 1+1 */
+#define DCT_VAL_CATEGORY2       6       /* 7-10      Extra Bits 2+1 */
+#define DCT_VAL_CATEGORY3       7       /* 11-18     Extra Bits 3+1 */
+#define DCT_VAL_CATEGORY4       8       /* 19-34     Extra Bits 4+1 */
+#define DCT_VAL_CATEGORY5       9       /* 35-66     Extra Bits 5+1 */
+#define DCT_VAL_CATEGORY6       10      /* 67+       Extra Bits 14+1 */
+#define DCT_EOB_TOKEN           11      /* EOB       Extra Bits 0+0 */
+#define MAX_ENTROPY_TOKENS      12
+#define ENTROPY_NODES           11
+#define EOSB_TOKEN              127     /* Not signalled, encoder only */
+
+#define INTER_MODE_CONTEXTS     7
+
+extern const vp9_tree_index vp9_coef_tree[];
+
+#define DCT_EOB_MODEL_TOKEN     3      /* EOB       Extra Bits 0+0 */
+extern const vp9_tree_index vp9_coefmodel_tree[];
+
+extern struct vp9_token vp9_coef_encodings[MAX_ENTROPY_TOKENS];
+
+typedef struct {
+  vp9_tree_p tree;
+  const vp9_prob *prob;
+  int len;
+  int base_val;
+} vp9_extra_bit;
+
+extern vp9_extra_bit vp9_extra_bits[12];    /* indexed by token value */
+
+#define PROB_UPDATE_BASELINE_COST   7
+
+#define MAX_PROB                255
+#define DCT_MAX_VALUE           16384
+
+/* Coefficients are predicted via a 3-dimensional probability table. */
+
+/* Outside dimension.  0 = Y with DC, 1 = UV */
+#define BLOCK_TYPES 2
+#define REF_TYPES 2  // intra=0, inter=1
+
+/* Middle dimension reflects the coefficient position within the transform. */
+#define COEF_BANDS 6
+
+/* Inside dimension is measure of nearby complexity, that reflects the energy
+   of nearby coefficients are nonzero.  For the first coefficient (DC, unless
+   block type is 0), we look at the (already encoded) blocks above and to the
+   left of the current block.  The context index is then the number (0,1,or 2)
+   of these blocks having nonzero coefficients.
+   After decoding a coefficient, the measure is determined by the size of the
+   most recently decoded coefficient.
+   Note that the intuitive meaning of this measure changes as coefficients
+   are decoded, e.g., prior to the first token, a zero means that my neighbors
+   are empty while, after the first token, because of the use of end-of-block,
+   a zero means we just decoded a zero and hence guarantees that a non-zero
+   coefficient will appear later in this block.  However, this shift
+   in meaning is perfectly OK because our context depends also on the
+   coefficient band (and since zigzag positions 0, 1, and 2 are in
+   distinct bands). */
+
+/*# define DC_TOKEN_CONTEXTS        3*/ /* 00, 0!0, !0!0 */
+#define PREV_COEF_CONTEXTS          6
+
+// #define ENTROPY_STATS
+
+typedef unsigned int vp9_coeff_count[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS]
+                                    [MAX_ENTROPY_TOKENS];
+typedef unsigned int vp9_coeff_stats[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS]
+                                    [ENTROPY_NODES][2];
+typedef vp9_prob vp9_coeff_probs[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS]
+                                [ENTROPY_NODES];
+
+#define SUBEXP_PARAM                4   /* Subexponential code parameter */
+#define MODULUS_PARAM               13  /* Modulus parameter */
+
+struct VP9Common;
+void vp9_default_coef_probs(struct VP9Common *);
+extern DECLARE_ALIGNED(16, const int, vp9_default_scan_4x4[16]);
+
+extern DECLARE_ALIGNED(16, const int, vp9_col_scan_4x4[16]);
+extern DECLARE_ALIGNED(16, const int, vp9_row_scan_4x4[16]);
+
+extern DECLARE_ALIGNED(64, const int, vp9_default_scan_8x8[64]);
+
+extern DECLARE_ALIGNED(16, const int, vp9_col_scan_8x8[64]);
+extern DECLARE_ALIGNED(16, const int, vp9_row_scan_8x8[64]);
+
+extern DECLARE_ALIGNED(16, const int, vp9_default_scan_16x16[256]);
+
+extern DECLARE_ALIGNED(16, const int, vp9_col_scan_16x16[256]);
+extern DECLARE_ALIGNED(16, const int, vp9_row_scan_16x16[256]);
+
+extern DECLARE_ALIGNED(16, const int, vp9_default_scan_32x32[1024]);
+
+void vp9_coef_tree_initialize(void);
+void vp9_adapt_coef_probs(struct VP9Common *);
+
+static INLINE void vp9_reset_sb_tokens_context(MACROBLOCKD* const xd,
+                                               BLOCK_SIZE_TYPE bsize) {
+  /* Clear entropy contexts */
+  const int bw = 1 << b_width_log2(bsize);
+  const int bh = 1 << b_height_log2(bsize);
+  int i;
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    vpx_memset(xd->plane[i].above_context, 0,
+               sizeof(ENTROPY_CONTEXT) * bw >> xd->plane[i].subsampling_x);
+    vpx_memset(xd->plane[i].left_context, 0,
+               sizeof(ENTROPY_CONTEXT) * bh >> xd->plane[i].subsampling_y);
+  }
+}
+
+// This is the index in the scan order beyond which all coefficients for
+// 8x8 transform and above are in the top band.
+// For 4x4 blocks the index is less but to keep things common the lookup
+// table for 4x4 is padded out to this index.
+#define MAXBAND_INDEX 21
+
+extern const uint8_t vp9_coefband_trans_8x8plus[MAXBAND_INDEX + 1];
+extern const uint8_t vp9_coefband_trans_4x4[MAXBAND_INDEX + 1];
+
+
+static int get_coef_band(const uint8_t * band_translate, int coef_index) {
+  return (coef_index > MAXBAND_INDEX)
+    ? (COEF_BANDS-1) : band_translate[coef_index];
+}
+
+extern int vp9_get_coef_context(const int *scan, const int *neighbors,
+                                int nb_pad, uint8_t *token_cache, int c, int l);
+const int *vp9_get_coef_neighbors_handle(const int *scan, int *pad);
+
+
+// 128 lists of probabilities are stored for the following ONE node probs:
+// 1, 3, 5, 7, ..., 253, 255
+// In between probabilities are interpolated linearly
+
+#define COEFPROB_MODELS             128
+
+#define UNCONSTRAINED_NODES         3
+#define MODEL_NODES                 (ENTROPY_NODES - UNCONSTRAINED_NODES)
+
+#define PIVOT_NODE                  2   // which node is pivot
+
+typedef vp9_prob vp9_coeff_probs_model[REF_TYPES][COEF_BANDS]
+                                      [PREV_COEF_CONTEXTS]
+                                      [UNCONSTRAINED_NODES];
+
+typedef unsigned int vp9_coeff_count_model[REF_TYPES][COEF_BANDS]
+                                          [PREV_COEF_CONTEXTS]
+                                          [UNCONSTRAINED_NODES + 1];
+typedef unsigned int vp9_coeff_stats_model[REF_TYPES][COEF_BANDS]
+                                          [PREV_COEF_CONTEXTS]
+                                          [UNCONSTRAINED_NODES][2];
+extern void vp9_full_to_model_count(unsigned int *model_count,
+                                    unsigned int *full_count);
+extern void vp9_full_to_model_counts(
+    vp9_coeff_count_model *model_count, vp9_coeff_count *full_count);
+
+void vp9_model_to_full_probs(const vp9_prob *model, vp9_prob *full);
+
+void vp9_model_to_full_probs_sb(
+    vp9_prob model[COEF_BANDS][PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES],
+    vp9_prob full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES]);
+
+extern const vp9_prob vp9_modelcoefprobs[COEFPROB_MODELS][ENTROPY_NODES - 1];
+
+static INLINE const int* get_scan_4x4(TX_TYPE tx_type) {
+  switch (tx_type) {
+    case ADST_DCT:
+      return vp9_row_scan_4x4;
+    case DCT_ADST:
+      return vp9_col_scan_4x4;
+    default:
+      return vp9_default_scan_4x4;
+  }
+}
+
+static INLINE const int* get_scan_8x8(TX_TYPE tx_type) {
+  switch (tx_type) {
+    case ADST_DCT:
+      return vp9_row_scan_8x8;
+    case DCT_ADST:
+      return vp9_col_scan_8x8;
+    default:
+      return vp9_default_scan_8x8;
+  }
+}
+
+static INLINE const int* get_scan_16x16(TX_TYPE tx_type) {
+  switch (tx_type) {
+    case ADST_DCT:
+      return vp9_row_scan_16x16;
+    case DCT_ADST:
+      return vp9_col_scan_16x16;
+    default:
+      return vp9_default_scan_16x16;
+  }
+}
+
+enum { VP9_COEF_UPDATE_PROB = 252 };
+
+#endif  // VP9_COMMON_VP9_ENTROPY_H_
diff --git a/libvpx/vp9/common/vp9_entropymode.c b/libvpx/vp9/common/vp9_entropymode.c
new file mode 100644
index 0000000..3302814
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropymode.c
@@ -0,0 +1,535 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_modecont.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vpx_mem/vpx_mem.h"
+
+static const vp9_prob default_kf_uv_probs[VP9_INTRA_MODES]
+                                         [VP9_INTRA_MODES - 1] = {
+  { 144,  11,  54, 157, 195, 130,  46,  58, 108 } /* y = dc */,
+  { 118,  15, 123, 148, 131, 101,  44,  93, 131 } /* y = v */,
+  { 113,  12,  23, 188, 226, 142,  26,  32, 125 } /* y = h */,
+  { 120,  11,  50, 123, 163, 135,  64,  77, 103 } /* y = d45 */,
+  { 113,   9,  36, 155, 111, 157,  32,  44, 161 } /* y = d135 */,
+  { 116,   9,  55, 176,  76,  96,  37,  61, 149 } /* y = d117 */,
+  { 115,   9,  28, 141, 161, 167,  21,  25, 193 } /* y = d153 */,
+  { 120,  12,  32, 145, 195, 142,  32,  38,  86 } /* y = d27 */,
+  { 116,  12,  64, 120, 140, 125,  49, 115, 121 } /* y = d63 */,
+  { 102,  19,  66, 162, 182, 122,  35,  59, 128 } /* y = tm */
+};
+
+static const vp9_prob default_if_y_probs[BLOCK_SIZE_GROUPS]
+                                        [VP9_INTRA_MODES - 1] = {
+  {  65,  32,  18, 144, 162, 194,  41,  51,  98 } /* block_size < 8x8 */,
+  { 132,  68,  18, 165, 217, 196,  45,  40,  78 } /* block_size < 16x16 */,
+  { 173,  80,  19, 176, 240, 193,  64,  35,  46 } /* block_size < 32x32 */,
+  { 221, 135,  38, 194, 248, 121,  96,  85,  29 } /* block_size >= 32x32 */
+};
+
+static const vp9_prob default_if_uv_probs[VP9_INTRA_MODES]
+                                         [VP9_INTRA_MODES - 1] = {
+  { 120,   7,  76, 176, 208, 126,  28,  54, 103 } /* y = dc */,
+  {  48,  12, 154, 155, 139,  90,  34, 117, 119 } /* y = v */,
+  {  67,   6,  25, 204, 243, 158,  13,  21,  96 } /* y = h */,
+  {  97,   5,  44, 131, 176, 139,  48,  68,  97 } /* y = d45 */,
+  {  83,   5,  42, 156, 111, 152,  26,  49, 152 } /* y = d135 */,
+  {  80,   5,  58, 178,  74,  83,  33,  62, 145 } /* y = d117 */,
+  {  86,   5,  32, 154, 192, 168,  14,  22, 163 } /* y = d153 */,
+  {  85,   5,  32, 156, 216, 148,  19,  29,  73 } /* y = d27 */,
+  {  77,   7,  64, 116, 132, 122,  37, 126, 120 } /* y = d63 */,
+  { 101,  21, 107, 181, 192, 103,  19,  67, 125 } /* y = tm */
+};
+
+const vp9_prob vp9_partition_probs[NUM_FRAME_TYPES][NUM_PARTITION_CONTEXTS]
+                                  [PARTITION_TYPES - 1] = {
+  { /* frame_type = keyframe */
+    /* 8x8 -> 4x4 */
+    { 158,  97,  94 } /* a/l both not split */,
+    {  93,  24,  99 } /* a split, l not split */,
+    {  85, 119,  44 } /* l split, a not split */,
+    {  62,  59,  67 } /* a/l both split */,
+    /* 16x16 -> 8x8 */
+    { 149,  53,  53 } /* a/l both not split */,
+    {  94,  20,  48 } /* a split, l not split */,
+    {  83,  53,  24 } /* l split, a not split */,
+    {  52,  18,  18 } /* a/l both split */,
+    /* 32x32 -> 16x16 */
+    { 150,  40,  39 } /* a/l both not split */,
+    {  78,  12,  26 } /* a split, l not split */,
+    {  67,  33,  11 } /* l split, a not split */,
+    {  24,   7,   5 } /* a/l both split */,
+    /* 64x64 -> 32x32 */
+    { 174,  35,  49 } /* a/l both not split */,
+    {  68,  11,  27 } /* a split, l not split */,
+    {  57,  15,   9 } /* l split, a not split */,
+    {  12,   3,   3 } /* a/l both split */
+  }, { /* frame_type = interframe */
+    /* 8x8 -> 4x4 */
+    { 199, 122, 141 } /* a/l both not split */,
+    { 147,  63, 159 } /* a split, l not split */,
+    { 148, 133, 118 } /* l split, a not split */,
+    { 121, 104, 114 } /* a/l both split */,
+    /* 16x16 -> 8x8 */
+    { 174,  73,  87 } /* a/l both not split */,
+    {  92,  41,  83 } /* a split, l not split */,
+    {  82,  99,  50 } /* l split, a not split */,
+    {  53,  39,  39 } /* a/l both split */,
+    /* 32x32 -> 16x16 */
+    { 177,  58,  59 } /* a/l both not split */,
+    {  68,  26,  63 } /* a split, l not split */,
+    {  52,  79,  25 } /* l split, a not split */,
+    {  17,  14,  12 } /* a/l both split */,
+    /* 64x64 -> 32x32 */
+    { 222,  34,  30 } /* a/l both not split */,
+    {  72,  16,  44 } /* a split, l not split */,
+    {  58,  32,  12 } /* l split, a not split */,
+    {  10,   7,   6 } /* a/l both split */
+  }
+};
+
+/* Array indices are identical to previously-existing INTRAMODECONTEXTNODES. */
+const vp9_tree_index vp9_intra_mode_tree[VP9_INTRA_MODES * 2 - 2] = {
+  -DC_PRED, 2,                      /* 0 = DC_NODE */
+  -TM_PRED, 4,                      /* 1 = TM_NODE */
+  -V_PRED, 6,                       /* 2 = V_NODE */
+  8, 12,                            /* 3 = COM_NODE */
+  -H_PRED, 10,                      /* 4 = H_NODE */
+  -D135_PRED, -D117_PRED,           /* 5 = D135_NODE */
+  -D45_PRED, 14,                    /* 6 = D45_NODE */
+  -D63_PRED, 16,                    /* 7 = D63_NODE */
+  -D153_PRED, -D27_PRED             /* 8 = D153_NODE */
+};
+
+const vp9_tree_index vp9_sb_mv_ref_tree[6] = {
+  -ZEROMV, 2,
+  -NEARESTMV, 4,
+  -NEARMV, -NEWMV
+};
+
+const vp9_tree_index vp9_partition_tree[6] = {
+  -PARTITION_NONE, 2,
+  -PARTITION_HORZ, 4,
+  -PARTITION_VERT, -PARTITION_SPLIT
+};
+
+struct vp9_token vp9_intra_mode_encodings[VP9_INTRA_MODES];
+
+struct vp9_token vp9_sb_mv_ref_encoding_array[VP9_INTER_MODES];
+
+struct vp9_token vp9_partition_encodings[PARTITION_TYPES];
+
+static const vp9_prob default_intra_inter_p[INTRA_INTER_CONTEXTS] = {
+  9, 102, 187, 225
+};
+
+static const vp9_prob default_comp_inter_p[COMP_INTER_CONTEXTS] = {
+  239, 183, 119,  96,  41
+};
+
+static const vp9_prob default_comp_ref_p[REF_CONTEXTS] = {
+  50, 126, 123, 221, 226
+};
+
+static const vp9_prob default_single_ref_p[REF_CONTEXTS][2] = {
+  {  33,  16 },
+  {  77,  74 },
+  { 142, 142 },
+  { 172, 170 },
+  { 238, 247 }
+};
+
+const vp9_prob vp9_default_tx_probs_32x32p[TX_SIZE_CONTEXTS]
+                                          [TX_SIZE_MAX_SB - 1] = {
+  { 3, 136, 37, },
+  { 5, 52, 13, },
+};
+const vp9_prob vp9_default_tx_probs_16x16p[TX_SIZE_CONTEXTS]
+                                          [TX_SIZE_MAX_SB - 2] = {
+  { 20, 152, },
+  { 15, 101, },
+};
+const vp9_prob vp9_default_tx_probs_8x8p[TX_SIZE_CONTEXTS]
+                                        [TX_SIZE_MAX_SB - 3] = {
+  { 100, },
+  { 66, },
+};
+
+void tx_counts_to_branch_counts_32x32(unsigned int *tx_count_32x32p,
+                                      unsigned int (*ct_32x32p)[2]) {
+  ct_32x32p[0][0] = tx_count_32x32p[TX_4X4];
+  ct_32x32p[0][1] = tx_count_32x32p[TX_8X8] +
+                    tx_count_32x32p[TX_16X16] +
+                    tx_count_32x32p[TX_32X32];
+  ct_32x32p[1][0] = tx_count_32x32p[TX_8X8];
+  ct_32x32p[1][1] = tx_count_32x32p[TX_16X16] +
+                    tx_count_32x32p[TX_32X32];
+  ct_32x32p[2][0] = tx_count_32x32p[TX_16X16];
+  ct_32x32p[2][1] = tx_count_32x32p[TX_32X32];
+}
+
+void tx_counts_to_branch_counts_16x16(unsigned int *tx_count_16x16p,
+                                      unsigned int (*ct_16x16p)[2]) {
+  ct_16x16p[0][0] = tx_count_16x16p[TX_4X4];
+  ct_16x16p[0][1] = tx_count_16x16p[TX_8X8] +
+                    tx_count_16x16p[TX_16X16];
+  ct_16x16p[1][0] = tx_count_16x16p[TX_8X8];
+  ct_16x16p[1][1] = tx_count_16x16p[TX_16X16];
+}
+
+void tx_counts_to_branch_counts_8x8(unsigned int *tx_count_8x8p,
+                                    unsigned int (*ct_8x8p)[2]) {
+  ct_8x8p[0][0] =   tx_count_8x8p[TX_4X4];
+  ct_8x8p[0][1] =   tx_count_8x8p[TX_8X8];
+}
+
+const vp9_prob vp9_default_mbskip_probs[MBSKIP_CONTEXTS] = {
+  192, 128, 64
+};
+
+void vp9_init_mbmode_probs(VP9_COMMON *x) {
+  vpx_memcpy(x->fc.uv_mode_prob, default_if_uv_probs,
+             sizeof(default_if_uv_probs));
+  vpx_memcpy(x->kf_uv_mode_prob, default_kf_uv_probs,
+             sizeof(default_kf_uv_probs));
+  vpx_memcpy(x->fc.y_mode_prob, default_if_y_probs,
+             sizeof(default_if_y_probs));
+
+  vpx_memcpy(x->fc.switchable_interp_prob, vp9_switchable_interp_prob,
+             sizeof(vp9_switchable_interp_prob));
+
+  vpx_memcpy(x->fc.partition_prob, vp9_partition_probs,
+             sizeof(vp9_partition_probs));
+
+  vpx_memcpy(x->fc.intra_inter_prob, default_intra_inter_p,
+             sizeof(default_intra_inter_p));
+  vpx_memcpy(x->fc.comp_inter_prob, default_comp_inter_p,
+             sizeof(default_comp_inter_p));
+  vpx_memcpy(x->fc.comp_ref_prob, default_comp_ref_p,
+             sizeof(default_comp_ref_p));
+  vpx_memcpy(x->fc.single_ref_prob, default_single_ref_p,
+             sizeof(default_single_ref_p));
+  vpx_memcpy(x->fc.tx_probs_32x32p, vp9_default_tx_probs_32x32p,
+             sizeof(vp9_default_tx_probs_32x32p));
+  vpx_memcpy(x->fc.tx_probs_16x16p, vp9_default_tx_probs_16x16p,
+             sizeof(vp9_default_tx_probs_16x16p));
+  vpx_memcpy(x->fc.tx_probs_8x8p, vp9_default_tx_probs_8x8p,
+             sizeof(vp9_default_tx_probs_8x8p));
+  vpx_memcpy(x->fc.mbskip_probs, vp9_default_mbskip_probs,
+             sizeof(vp9_default_mbskip_probs));
+}
+
+const vp9_tree_index vp9_switchable_interp_tree[VP9_SWITCHABLE_FILTERS*2-2] = {
+  -0, 2,
+  -1, -2
+};
+struct vp9_token vp9_switchable_interp_encodings[VP9_SWITCHABLE_FILTERS];
+const INTERPOLATIONFILTERTYPE vp9_switchable_interp[VP9_SWITCHABLE_FILTERS] = {
+  EIGHTTAP, EIGHTTAP_SMOOTH, EIGHTTAP_SHARP};
+const int vp9_switchable_interp_map[SWITCHABLE+1] = {1, 0, 2, -1, -1};
+const vp9_prob vp9_switchable_interp_prob [VP9_SWITCHABLE_FILTERS+1]
+                                          [VP9_SWITCHABLE_FILTERS-1] = {
+  { 235, 162, },
+  { 36, 255, },
+  { 34, 3, },
+  { 149, 144, },
+};
+
+// Indicates if the filter is interpolating or non-interpolating
+const int vp9_is_interpolating_filter[SWITCHABLE + 1] = {1, 1, 1, 1, -1};
+
+void vp9_entropy_mode_init() {
+  vp9_tokens_from_tree(vp9_intra_mode_encodings, vp9_intra_mode_tree);
+  vp9_tokens_from_tree(vp9_switchable_interp_encodings,
+                       vp9_switchable_interp_tree);
+  vp9_tokens_from_tree(vp9_partition_encodings, vp9_partition_tree);
+
+  vp9_tokens_from_tree_offset(vp9_sb_mv_ref_encoding_array,
+                              vp9_sb_mv_ref_tree, NEARESTMV);
+}
+
+void vp9_init_mode_contexts(VP9_COMMON *pc) {
+  vpx_memset(pc->fc.inter_mode_counts, 0, sizeof(pc->fc.inter_mode_counts));
+  vpx_memcpy(pc->fc.inter_mode_probs,
+             vp9_default_inter_mode_probs,
+             sizeof(vp9_default_inter_mode_probs));
+}
+
+void vp9_accum_mv_refs(VP9_COMMON *pc,
+                       MB_PREDICTION_MODE m,
+                       const int context) {
+  unsigned int (*inter_mode_counts)[VP9_INTER_MODES - 1][2] =
+      pc->fc.inter_mode_counts;
+
+  if (m == ZEROMV) {
+    ++inter_mode_counts[context][0][0];
+  } else {
+    ++inter_mode_counts[context][0][1];
+    if (m == NEARESTMV) {
+      ++inter_mode_counts[context][1][0];
+    } else {
+      ++inter_mode_counts[context][1][1];
+      if (m == NEARMV) {
+        ++inter_mode_counts[context][2][0];
+      } else {
+        ++inter_mode_counts[context][2][1];
+      }
+    }
+  }
+}
+
+#define MVREF_COUNT_SAT 20
+#define MVREF_MAX_UPDATE_FACTOR 128
+void vp9_adapt_mode_context(VP9_COMMON *pc) {
+  int i, j;
+  unsigned int (*inter_mode_counts)[VP9_INTER_MODES - 1][2] =
+      pc->fc.inter_mode_counts;
+  vp9_prob (*mode_context)[VP9_INTER_MODES - 1] = pc->fc.inter_mode_probs;
+
+  for (j = 0; j < INTER_MODE_CONTEXTS; j++) {
+    for (i = 0; i < VP9_INTER_MODES - 1; i++) {
+      int count = inter_mode_counts[j][i][0] + inter_mode_counts[j][i][1];
+      int factor;
+      count = count > MVREF_COUNT_SAT ? MVREF_COUNT_SAT : count;
+      factor = (MVREF_MAX_UPDATE_FACTOR * count / MVREF_COUNT_SAT);
+      mode_context[j][i] = weighted_prob(
+          pc->fc.pre_inter_mode_probs[j][i],
+          get_binary_prob(inter_mode_counts[j][i][0],
+                          inter_mode_counts[j][i][1]),
+          factor);
+    }
+  }
+}
+
+#define MODE_COUNT_SAT 20
+#define MODE_MAX_UPDATE_FACTOR 128
+static int update_mode_ct(vp9_prob pre_prob, vp9_prob prob,
+                          unsigned int branch_ct[2]) {
+  int factor, count = branch_ct[0] + branch_ct[1];
+  count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
+  factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
+  return weighted_prob(pre_prob, prob, factor);
+}
+
+static void update_mode_probs(int n_modes,
+                              const vp9_tree_index *tree, unsigned int *cnt,
+                              vp9_prob *pre_probs, vp9_prob *dst_probs,
+                              unsigned int tok0_offset) {
+#define MAX_PROBS 32
+  vp9_prob probs[MAX_PROBS];
+  unsigned int branch_ct[MAX_PROBS][2];
+  int t;
+
+  assert(n_modes - 1 < MAX_PROBS);
+  vp9_tree_probs_from_distribution(tree, probs, branch_ct, cnt, tok0_offset);
+  for (t = 0; t < n_modes - 1; ++t)
+    dst_probs[t] = update_mode_ct(pre_probs[t], probs[t], branch_ct[t]);
+}
+
+static int update_mode_ct2(vp9_prob pre_prob, unsigned int branch_ct[2]) {
+  return update_mode_ct(pre_prob, get_binary_prob(branch_ct[0],
+                                                  branch_ct[1]), branch_ct);
+}
+
+// #define MODE_COUNT_TESTING
+void vp9_adapt_mode_probs(VP9_COMMON *cm) {
+  int i, j;
+  FRAME_CONTEXT *fc = &cm->fc;
+#ifdef MODE_COUNT_TESTING
+  int t;
+
+  printf("static const unsigned int\nymode_counts"
+         "[VP9_INTRA_MODES] = {\n");
+  for (t = 0; t < VP9_INTRA_MODES; ++t)
+    printf("%d, ", fc->ymode_counts[t]);
+  printf("};\n");
+  printf("static const unsigned int\nuv_mode_counts"
+         "[VP9_INTRA_MODES] [VP9_INTRA_MODES] = {\n");
+  for (i = 0; i < VP9_INTRA_MODES; ++i) {
+    printf("  {");
+    for (t = 0; t < VP9_INTRA_MODES; ++t)
+      printf("%d, ", fc->uv_mode_counts[i][t]);
+    printf("},\n");
+  }
+  printf("};\n");
+  printf("static const unsigned int\nbmode_counts"
+         "[VP9_NKF_BINTRAMODES] = {\n");
+  for (t = 0; t < VP9_NKF_BINTRAMODES; ++t)
+    printf("%d, ", fc->bmode_counts[t]);
+  printf("};\n");
+  printf("static const unsigned int\ni8x8_mode_counts"
+         "[VP9_I8X8_MODES] = {\n");
+  for (t = 0; t < VP9_I8X8_MODES; ++t)
+    printf("%d, ", fc->i8x8_mode_counts[t]);
+  printf("};\n");
+  printf("static const unsigned int\nmbsplit_counts"
+         "[VP9_NUMMBSPLITS] = {\n");
+  for (t = 0; t < VP9_NUMMBSPLITS; ++t)
+    printf("%d, ", fc->mbsplit_counts[t]);
+  printf("};\n");
+#endif
+
+  for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
+    fc->intra_inter_prob[i] = update_mode_ct2(fc->pre_intra_inter_prob[i],
+                                              fc->intra_inter_count[i]);
+  for (i = 0; i < COMP_INTER_CONTEXTS; i++)
+    fc->comp_inter_prob[i] = update_mode_ct2(fc->pre_comp_inter_prob[i],
+                                             fc->comp_inter_count[i]);
+  for (i = 0; i < REF_CONTEXTS; i++)
+    fc->comp_ref_prob[i] = update_mode_ct2(fc->pre_comp_ref_prob[i],
+                                           fc->comp_ref_count[i]);
+  for (i = 0; i < REF_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      fc->single_ref_prob[i][j] = update_mode_ct2(fc->pre_single_ref_prob[i][j],
+                                                  fc->single_ref_count[i][j]);
+
+  for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
+    update_mode_probs(VP9_INTRA_MODES, vp9_intra_mode_tree,
+                      fc->y_mode_counts[i], fc->pre_y_mode_prob[i],
+                      fc->y_mode_prob[i], 0);
+
+  for (i = 0; i < VP9_INTRA_MODES; ++i)
+    update_mode_probs(VP9_INTRA_MODES, vp9_intra_mode_tree,
+                      fc->uv_mode_counts[i], fc->pre_uv_mode_prob[i],
+                      fc->uv_mode_prob[i], 0);
+
+  for (i = 0; i < NUM_PARTITION_CONTEXTS; i++)
+    update_mode_probs(PARTITION_TYPES, vp9_partition_tree,
+                      fc->partition_counts[i], fc->pre_partition_prob[i],
+                      fc->partition_prob[INTER_FRAME][i], 0);
+
+  if (cm->mcomp_filter_type == SWITCHABLE) {
+    for (i = 0; i <= VP9_SWITCHABLE_FILTERS; i++) {
+      update_mode_probs(VP9_SWITCHABLE_FILTERS, vp9_switchable_interp_tree,
+                        fc->switchable_interp_count[i],
+                        fc->pre_switchable_interp_prob[i],
+                        fc->switchable_interp_prob[i], 0);
+    }
+  }
+  if (cm->txfm_mode == TX_MODE_SELECT) {
+    int j;
+    unsigned int branch_ct_8x8p[TX_SIZE_MAX_SB - 3][2];
+    unsigned int branch_ct_16x16p[TX_SIZE_MAX_SB - 2][2];
+    unsigned int branch_ct_32x32p[TX_SIZE_MAX_SB - 1][2];
+    for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+      tx_counts_to_branch_counts_8x8(cm->fc.tx_count_8x8p[i],
+                                     branch_ct_8x8p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j) {
+        int factor;
+        int count = branch_ct_8x8p[j][0] + branch_ct_8x8p[j][1];
+        vp9_prob prob = get_binary_prob(branch_ct_8x8p[j][0],
+                                        branch_ct_8x8p[j][1]);
+        count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
+        factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
+        cm->fc.tx_probs_8x8p[i][j] = weighted_prob(
+            cm->fc.pre_tx_probs_8x8p[i][j], prob, factor);
+      }
+    }
+    for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+      tx_counts_to_branch_counts_16x16(cm->fc.tx_count_16x16p[i],
+                                       branch_ct_16x16p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j) {
+        int factor;
+        int count = branch_ct_16x16p[j][0] + branch_ct_16x16p[j][1];
+        vp9_prob prob = get_binary_prob(branch_ct_16x16p[j][0],
+                                        branch_ct_16x16p[j][1]);
+        count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
+        factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
+        cm->fc.tx_probs_16x16p[i][j] = weighted_prob(
+            cm->fc.pre_tx_probs_16x16p[i][j], prob, factor);
+      }
+    }
+    for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+      tx_counts_to_branch_counts_32x32(cm->fc.tx_count_32x32p[i],
+                                       branch_ct_32x32p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j) {
+        int factor;
+        int count = branch_ct_32x32p[j][0] + branch_ct_32x32p[j][1];
+        vp9_prob prob = get_binary_prob(branch_ct_32x32p[j][0],
+                                        branch_ct_32x32p[j][1]);
+        count = count > MODE_COUNT_SAT ? MODE_COUNT_SAT : count;
+        factor = (MODE_MAX_UPDATE_FACTOR * count / MODE_COUNT_SAT);
+        cm->fc.tx_probs_32x32p[i][j] = weighted_prob(
+            cm->fc.pre_tx_probs_32x32p[i][j], prob, factor);
+      }
+    }
+  }
+  for (i = 0; i < MBSKIP_CONTEXTS; ++i)
+    fc->mbskip_probs[i] = update_mode_ct2(fc->pre_mbskip_probs[i],
+                                          fc->mbskip_count[i]);
+}
+
+static void set_default_lf_deltas(MACROBLOCKD *xd) {
+  xd->mode_ref_lf_delta_enabled = 1;
+  xd->mode_ref_lf_delta_update = 1;
+
+  xd->ref_lf_deltas[INTRA_FRAME] = 1;
+  xd->ref_lf_deltas[LAST_FRAME] = 0;
+  xd->ref_lf_deltas[GOLDEN_FRAME] = -1;
+  xd->ref_lf_deltas[ALTREF_FRAME] = -1;
+
+  xd->mode_lf_deltas[0] = 0;              // Zero
+  xd->mode_lf_deltas[1] = 0;               // New mv
+}
+
+void vp9_setup_past_independence(VP9_COMMON *cm, MACROBLOCKD *xd) {
+  // Reset the segment feature data to the default stats:
+  // Features disabled, 0, with delta coding (Default state).
+  int i;
+  vp9_clearall_segfeatures(xd);
+  xd->mb_segment_abs_delta = SEGMENT_DELTADATA;
+  if (cm->last_frame_seg_map)
+    vpx_memset(cm->last_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
+
+  // Reset the mode ref deltas for loop filter
+  vpx_memset(xd->last_ref_lf_deltas, 0, sizeof(xd->last_ref_lf_deltas));
+  vpx_memset(xd->last_mode_lf_deltas, 0, sizeof(xd->last_mode_lf_deltas));
+  set_default_lf_deltas(xd);
+
+  vp9_default_coef_probs(cm);
+  vp9_init_mbmode_probs(cm);
+  vpx_memcpy(cm->kf_y_mode_prob, vp9_kf_default_bmode_probs,
+             sizeof(vp9_kf_default_bmode_probs));
+  vp9_init_mv_probs(cm);
+
+  // To force update of the sharpness
+  cm->last_sharpness_level = -1;
+
+  vp9_init_mode_contexts(cm);
+
+  if ((cm->frame_type == KEY_FRAME) ||
+      cm->error_resilient_mode || (cm->reset_frame_context == 3)) {
+    // Reset all frame contexts.
+    for (i = 0; i < NUM_FRAME_CONTEXTS; ++i)
+      vpx_memcpy(&cm->frame_contexts[i], &cm->fc, sizeof(cm->fc));
+  } else if (cm->reset_frame_context == 2) {
+    // Reset only the frame context specified in the frame header.
+    vpx_memcpy(&cm->frame_contexts[cm->frame_context_idx], &cm->fc,
+               sizeof(cm->fc));
+  }
+
+  vpx_memset(cm->prev_mip, 0,
+             cm->mode_info_stride * (cm->mi_rows + 1) * sizeof(MODE_INFO));
+  vpx_memset(cm->mip, 0,
+             cm->mode_info_stride * (cm->mi_rows + 1) * sizeof(MODE_INFO));
+
+  vp9_update_mode_info_border(cm, cm->mip);
+  vp9_update_mode_info_in_image(cm, cm->mi);
+
+  vp9_update_mode_info_border(cm, cm->prev_mip);
+  vp9_update_mode_info_in_image(cm, cm->prev_mi);
+
+  vpx_memset(cm->ref_frame_sign_bias, 0, sizeof(cm->ref_frame_sign_bias));
+
+  cm->frame_context_idx = 0;
+}
diff --git a/libvpx/vp9/common/vp9_entropymode.h b/libvpx/vp9/common/vp9_entropymode.h
new file mode 100644
index 0000000..aa8aec7
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropymode.h
@@ -0,0 +1,96 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ENTROPYMODE_H_
+#define VP9_COMMON_VP9_ENTROPYMODE_H_
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_treecoder.h"
+
+#define SUBMVREF_COUNT 5
+#define TX_SIZE_CONTEXTS 2
+
+#define VP9_MODE_UPDATE_PROB  252
+
+// #define MODE_STATS
+
+extern int vp9_mv_cont(const int_mv *l, const int_mv *a);
+
+
+extern const vp9_prob vp9_kf_default_bmode_probs[VP9_INTRA_MODES]
+                                                [VP9_INTRA_MODES]
+                                                [VP9_INTRA_MODES - 1];
+
+extern const vp9_tree_index vp9_intra_mode_tree[];
+extern const vp9_tree_index  vp9_sb_mv_ref_tree[];
+
+extern struct vp9_token vp9_intra_mode_encodings[VP9_INTRA_MODES];
+
+/* Inter mode values do not start at zero */
+
+extern struct vp9_token vp9_sb_mv_ref_encoding_array[VP9_INTER_MODES];
+
+// probability models for partition information
+extern const vp9_tree_index  vp9_partition_tree[];
+extern struct vp9_token vp9_partition_encodings[PARTITION_TYPES];
+extern const vp9_prob vp9_partition_probs[NUM_FRAME_TYPES]
+                                         [NUM_PARTITION_CONTEXTS]
+                                         [PARTITION_TYPES - 1];
+
+void vp9_entropy_mode_init(void);
+
+struct VP9Common;
+
+/* sets up common features to forget past dependence */
+void vp9_setup_past_independence(struct VP9Common *cm, MACROBLOCKD *xd);
+
+void vp9_init_mbmode_probs(struct VP9Common *x);
+
+extern void vp9_init_mode_contexts(struct VP9Common *pc);
+
+extern void vp9_adapt_mode_context(struct VP9Common *pc);
+
+extern void vp9_accum_mv_refs(struct VP9Common *pc,
+                              MB_PREDICTION_MODE m,
+                              const int context);
+
+void vp9_adapt_mode_probs(struct VP9Common *);
+
+#define VP9_SWITCHABLE_FILTERS 3 /* number of switchable filters */
+
+extern const  INTERPOLATIONFILTERTYPE vp9_switchable_interp
+                  [VP9_SWITCHABLE_FILTERS];
+
+extern const  int vp9_switchable_interp_map[SWITCHABLE + 1];
+
+extern const  int vp9_is_interpolating_filter[SWITCHABLE + 1];
+
+extern const  vp9_tree_index vp9_switchable_interp_tree
+                  [2 * (VP9_SWITCHABLE_FILTERS - 1)];
+
+extern struct vp9_token vp9_switchable_interp_encodings[VP9_SWITCHABLE_FILTERS];
+
+extern const  vp9_prob vp9_switchable_interp_prob[VP9_SWITCHABLE_FILTERS + 1]
+                                                 [VP9_SWITCHABLE_FILTERS - 1];
+
+extern const vp9_prob vp9_default_tx_probs_32x32p[TX_SIZE_CONTEXTS]
+                                                 [TX_SIZE_MAX_SB - 1];
+extern const vp9_prob vp9_default_tx_probs_16x16p[TX_SIZE_CONTEXTS]
+                                                 [TX_SIZE_MAX_SB - 2];
+extern const vp9_prob vp9_default_tx_probs_8x8p[TX_SIZE_CONTEXTS]
+                                               [TX_SIZE_MAX_SB - 3];
+
+extern void tx_counts_to_branch_counts_32x32(unsigned int *tx_count_32x32p,
+                                             unsigned int (*ct_32x32p)[2]);
+extern void tx_counts_to_branch_counts_16x16(unsigned int *tx_count_16x16p,
+                                             unsigned int (*ct_16x16p)[2]);
+extern void tx_counts_to_branch_counts_8x8(unsigned int *tx_count_8x8p,
+                                           unsigned int (*ct_8x8p)[2]);
+#endif  // VP9_COMMON_VP9_ENTROPYMODE_H_
diff --git a/libvpx/vp9/common/vp9_entropymv.c b/libvpx/vp9/common/vp9_entropymv.c
new file mode 100644
index 0000000..e07e43c
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropymv.c
@@ -0,0 +1,452 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_entropymv.h"
+
+//#define MV_COUNT_TESTING
+
+#define MV_COUNT_SAT 20
+#define MV_MAX_UPDATE_FACTOR 128
+
+/* Integer pel reference mv threshold for use of high-precision 1/8 mv */
+#define COMPANDED_MVREF_THRESH    8
+
+/* Smooth or bias the mv-counts before prob computation */
+/* #define SMOOTH_MV_COUNTS */
+
+const vp9_tree_index vp9_mv_joint_tree[2 * MV_JOINTS - 2] = {
+  -MV_JOINT_ZERO, 2,
+  -MV_JOINT_HNZVZ, 4,
+  -MV_JOINT_HZVNZ, -MV_JOINT_HNZVNZ
+};
+struct vp9_token vp9_mv_joint_encodings[MV_JOINTS];
+
+const vp9_tree_index vp9_mv_class_tree[2 * MV_CLASSES - 2] = {
+  -MV_CLASS_0, 2,
+  -MV_CLASS_1, 4,
+  6, 8,
+  -MV_CLASS_2, -MV_CLASS_3,
+  10, 12,
+  -MV_CLASS_4, -MV_CLASS_5,
+  -MV_CLASS_6, 14,
+  16, 18,
+  -MV_CLASS_7, -MV_CLASS_8,
+  -MV_CLASS_9, -MV_CLASS_10,
+};
+struct vp9_token vp9_mv_class_encodings[MV_CLASSES];
+
+const vp9_tree_index vp9_mv_class0_tree [2 * CLASS0_SIZE - 2] = {
+  -0, -1,
+};
+struct vp9_token vp9_mv_class0_encodings[CLASS0_SIZE];
+
+const vp9_tree_index vp9_mv_fp_tree [2 * 4 - 2] = {
+  -0, 2,
+  -1, 4,
+  -2, -3
+};
+struct vp9_token vp9_mv_fp_encodings[4];
+
+const nmv_context vp9_default_nmv_context = {
+  {32, 64, 96},
+  {
+    { /* vert component */
+      128,                                                  /* sign */
+      {224, 144, 192, 168, 192, 176, 192, 198, 198, 245},   /* class */
+      {216},                                                /* class0 */
+      {136, 140, 148, 160, 176, 192, 224, 234, 234, 240},   /* bits */
+      {{128, 128, 64}, {96, 112, 64}},                      /* class0_fp */
+      {64, 96, 64},                                         /* fp */
+      160,                                                  /* class0_hp bit */
+      128,                                                  /* hp */
+    },
+    { /* hor component */
+      128,                                                  /* sign */
+      {216, 128, 176, 160, 176, 176, 192, 198, 198, 208},   /* class */
+      {208},                                                /* class0 */
+      {136, 140, 148, 160, 176, 192, 224, 234, 234, 240},   /* bits */
+      {{128, 128, 64}, {96, 112, 64}},                      /* class0_fp */
+      {64, 96, 64},                                         /* fp */
+      160,                                                  /* class0_hp bit */
+      128,                                                  /* hp */
+    }
+  },
+};
+
+MV_JOINT_TYPE vp9_get_mv_joint(const MV *mv) {
+  if (mv->row == 0 && mv->col == 0)
+    return MV_JOINT_ZERO;
+  else if (mv->row == 0 && mv->col != 0)
+    return MV_JOINT_HNZVZ;
+  else if (mv->row != 0 && mv->col == 0)
+    return MV_JOINT_HZVNZ;
+  else
+    return MV_JOINT_HNZVNZ;
+}
+
+#define mv_class_base(c) ((c) ? (CLASS0_SIZE << (c + 2)) : 0)
+
+MV_CLASS_TYPE vp9_get_mv_class(int z, int *offset) {
+  MV_CLASS_TYPE c;
+  if      (z < CLASS0_SIZE * 8)    c = MV_CLASS_0;
+  else if (z < CLASS0_SIZE * 16)   c = MV_CLASS_1;
+  else if (z < CLASS0_SIZE * 32)   c = MV_CLASS_2;
+  else if (z < CLASS0_SIZE * 64)   c = MV_CLASS_3;
+  else if (z < CLASS0_SIZE * 128)  c = MV_CLASS_4;
+  else if (z < CLASS0_SIZE * 256)  c = MV_CLASS_5;
+  else if (z < CLASS0_SIZE * 512)  c = MV_CLASS_6;
+  else if (z < CLASS0_SIZE * 1024) c = MV_CLASS_7;
+  else if (z < CLASS0_SIZE * 2048) c = MV_CLASS_8;
+  else if (z < CLASS0_SIZE * 4096) c = MV_CLASS_9;
+  else if (z < CLASS0_SIZE * 8192) c = MV_CLASS_10;
+  else assert(0);
+  if (offset)
+    *offset = z - mv_class_base(c);
+  return c;
+}
+
+int vp9_use_nmv_hp(const MV *ref) {
+  return (abs(ref->row) >> 3) < COMPANDED_MVREF_THRESH &&
+         (abs(ref->col) >> 3) < COMPANDED_MVREF_THRESH;
+}
+
+int vp9_get_mv_mag(MV_CLASS_TYPE c, int offset) {
+  return mv_class_base(c) + offset;
+}
+
+static void increment_nmv_component_count(int v,
+                                          nmv_component_counts *mvcomp,
+                                          int incr,
+                                          int usehp) {
+  assert (v != 0);            /* should not be zero */
+  mvcomp->mvcount[MV_MAX + v] += incr;
+}
+
+static void increment_nmv_component(int v,
+                                    nmv_component_counts *mvcomp,
+                                    int incr,
+                                    int usehp) {
+  int s, z, c, o, d, e, f;
+  if (!incr)
+    return;
+  assert (v != 0);            /* should not be zero */
+  s = v < 0;
+  mvcomp->sign[s] += incr;
+  z = (s ? -v : v) - 1;       /* magnitude - 1 */
+
+  c = vp9_get_mv_class(z, &o);
+  mvcomp->classes[c] += incr;
+
+  d = (o >> 3);               /* int mv data */
+  f = (o >> 1) & 3;           /* fractional pel mv data */
+  e = (o & 1);                /* high precision mv data */
+  if (c == MV_CLASS_0) {
+    mvcomp->class0[d] += incr;
+  } else {
+    int i;
+    int b = c + CLASS0_BITS - 1;  // number of bits
+    for (i = 0; i < b; ++i)
+      mvcomp->bits[i][((d >> i) & 1)] += incr;
+  }
+
+  /* Code the fractional pel bits */
+  if (c == MV_CLASS_0) {
+    mvcomp->class0_fp[d][f] += incr;
+  } else {
+    mvcomp->fp[f] += incr;
+  }
+
+  /* Code the high precision bit */
+  if (usehp) {
+    if (c == MV_CLASS_0) {
+      mvcomp->class0_hp[e] += incr;
+    } else {
+      mvcomp->hp[e] += incr;
+    }
+  }
+}
+
+#ifdef SMOOTH_MV_COUNTS
+static void smooth_counts(nmv_component_counts *mvcomp) {
+  static const int flen = 3;  // (filter_length + 1) / 2
+  static const int fval[] = {8, 3, 1};
+  static const int fvalbits = 4;
+  int i;
+  unsigned int smvcount[MV_VALS];
+  vpx_memcpy(smvcount, mvcomp->mvcount, sizeof(smvcount));
+  smvcount[MV_MAX] = (smvcount[MV_MAX - 1] + smvcount[MV_MAX + 1]) >> 1;
+  for (i = flen - 1; i <= MV_VALS - flen; ++i) {
+    int j, s = smvcount[i] * fval[0];
+    for (j = 1; j < flen; ++j)
+      s += (smvcount[i - j] + smvcount[i + j]) * fval[j];
+    mvcomp->mvcount[i] = (s + (1 << (fvalbits - 1))) >> fvalbits;
+  }
+}
+#endif
+
+static void counts_to_context(nmv_component_counts *mvcomp, int usehp) {
+  int v;
+  vpx_memset(mvcomp->sign, 0, sizeof(nmv_component_counts) - sizeof(mvcomp->mvcount));
+  for (v = 1; v <= MV_MAX; v++) {
+    increment_nmv_component(-v, mvcomp, mvcomp->mvcount[MV_MAX - v], usehp);
+    increment_nmv_component( v, mvcomp, mvcomp->mvcount[MV_MAX + v], usehp);
+  }
+}
+
+void vp9_increment_nmv(const MV *mv, const MV *ref, nmv_context_counts *mvctx,
+                       int usehp) {
+  const MV_JOINT_TYPE j = vp9_get_mv_joint(mv);
+  mvctx->joints[j]++;
+  usehp = usehp && vp9_use_nmv_hp(ref);
+  if (mv_joint_vertical(j))
+    increment_nmv_component_count(mv->row, &mvctx->comps[0], 1, usehp);
+
+  if (mv_joint_horizontal(j))
+    increment_nmv_component_count(mv->col, &mvctx->comps[1], 1, usehp);
+}
+
+static void adapt_prob(vp9_prob *dest, vp9_prob prep, unsigned int ct[2]) {
+  const int count = MIN(ct[0] + ct[1], MV_COUNT_SAT);
+  if (count) {
+    const vp9_prob newp = get_binary_prob(ct[0], ct[1]);
+    const int factor = MV_MAX_UPDATE_FACTOR * count / MV_COUNT_SAT;
+    *dest = weighted_prob(prep, newp, factor);
+  } else {
+    *dest = prep;
+  }
+}
+
+void vp9_counts_process(nmv_context_counts *nmv_count, int usehp) {
+  counts_to_context(&nmv_count->comps[0], usehp);
+  counts_to_context(&nmv_count->comps[1], usehp);
+}
+
+void vp9_counts_to_nmv_context(
+    nmv_context_counts *nmv_count,
+    nmv_context *prob,
+    int usehp,
+    unsigned int (*branch_ct_joint)[2],
+    unsigned int (*branch_ct_sign)[2],
+    unsigned int (*branch_ct_classes)[MV_CLASSES - 1][2],
+    unsigned int (*branch_ct_class0)[CLASS0_SIZE - 1][2],
+    unsigned int (*branch_ct_bits)[MV_OFFSET_BITS][2],
+    unsigned int (*branch_ct_class0_fp)[CLASS0_SIZE][4 - 1][2],
+    unsigned int (*branch_ct_fp)[4 - 1][2],
+    unsigned int (*branch_ct_class0_hp)[2],
+    unsigned int (*branch_ct_hp)[2]) {
+  int i, j, k;
+  vp9_counts_process(nmv_count, usehp);
+  vp9_tree_probs_from_distribution(vp9_mv_joint_tree,
+                                   prob->joints,
+                                   branch_ct_joint,
+                                   nmv_count->joints, 0);
+  for (i = 0; i < 2; ++i) {
+    const uint32_t s0 = nmv_count->comps[i].sign[0];
+    const uint32_t s1 = nmv_count->comps[i].sign[1];
+
+    prob->comps[i].sign = get_binary_prob(s0, s1);
+    branch_ct_sign[i][0] = s0;
+    branch_ct_sign[i][1] = s1;
+    vp9_tree_probs_from_distribution(vp9_mv_class_tree,
+                                     prob->comps[i].classes,
+                                     branch_ct_classes[i],
+                                     nmv_count->comps[i].classes, 0);
+    vp9_tree_probs_from_distribution(vp9_mv_class0_tree,
+                                     prob->comps[i].class0,
+                                     branch_ct_class0[i],
+                                     nmv_count->comps[i].class0, 0);
+    for (j = 0; j < MV_OFFSET_BITS; ++j) {
+      const uint32_t b0 = nmv_count->comps[i].bits[j][0];
+      const uint32_t b1 = nmv_count->comps[i].bits[j][1];
+
+      prob->comps[i].bits[j] = get_binary_prob(b0, b1);
+      branch_ct_bits[i][j][0] = b0;
+      branch_ct_bits[i][j][1] = b1;
+    }
+  }
+  for (i = 0; i < 2; ++i) {
+    for (k = 0; k < CLASS0_SIZE; ++k) {
+      vp9_tree_probs_from_distribution(vp9_mv_fp_tree,
+                                       prob->comps[i].class0_fp[k],
+                                       branch_ct_class0_fp[i][k],
+                                       nmv_count->comps[i].class0_fp[k], 0);
+    }
+    vp9_tree_probs_from_distribution(vp9_mv_fp_tree,
+                                     prob->comps[i].fp,
+                                     branch_ct_fp[i],
+                                     nmv_count->comps[i].fp, 0);
+  }
+  if (usehp) {
+    for (i = 0; i < 2; ++i) {
+      const uint32_t c0_hp0 = nmv_count->comps[i].class0_hp[0];
+      const uint32_t c0_hp1 = nmv_count->comps[i].class0_hp[1];
+      const uint32_t hp0 = nmv_count->comps[i].hp[0];
+      const uint32_t hp1 = nmv_count->comps[i].hp[1];
+
+      prob->comps[i].class0_hp = get_binary_prob(c0_hp0, c0_hp1);
+      branch_ct_class0_hp[i][0] = c0_hp0;
+      branch_ct_class0_hp[i][1] = c0_hp1;
+
+      prob->comps[i].hp = get_binary_prob(hp0, hp1);
+      branch_ct_hp[i][0] = hp0;
+      branch_ct_hp[i][1] = hp1;
+    }
+  }
+}
+
+static unsigned int adapt_probs(unsigned int i,
+                                vp9_tree tree,
+                                vp9_prob this_probs[],
+                                const vp9_prob last_probs[],
+                                const unsigned int num_events[]) {
+  vp9_prob this_prob;
+
+  const uint32_t left = tree[i] <= 0
+          ? num_events[-tree[i]]
+          : adapt_probs(tree[i], tree, this_probs, last_probs, num_events);
+
+  const uint32_t right = tree[i + 1] <= 0
+          ? num_events[-tree[i + 1]]
+          : adapt_probs(tree[i + 1], tree, this_probs, last_probs, num_events);
+
+  uint32_t weight = left + right;
+  if (weight) {
+    this_prob = get_binary_prob(left, right);
+    weight = weight > MV_COUNT_SAT ? MV_COUNT_SAT : weight;
+    this_prob = weighted_prob(last_probs[i >> 1], this_prob,
+                              MV_MAX_UPDATE_FACTOR * weight / MV_COUNT_SAT);
+  } else {
+    this_prob = last_probs[i >> 1];
+  }
+  this_probs[i >> 1] = this_prob;
+  return left + right;
+}
+
+
+void vp9_adapt_nmv_probs(VP9_COMMON *cm, int usehp) {
+  int i, j;
+#ifdef MV_COUNT_TESTING
+  printf("joints count: ");
+  for (j = 0; j < MV_JOINTS; ++j) printf("%d ", cm->fc.NMVcount.joints[j]);
+  printf("\n"); fflush(stdout);
+  printf("signs count:\n");
+  for (i = 0; i < 2; ++i)
+    printf("%d/%d ", cm->fc.NMVcount.comps[i].sign[0], cm->fc.NMVcount.comps[i].sign[1]);
+  printf("\n"); fflush(stdout);
+  printf("classes count:\n");
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < MV_CLASSES; ++j)
+      printf("%d ", cm->fc.NMVcount.comps[i].classes[j]);
+    printf("\n"); fflush(stdout);
+  }
+  printf("class0 count:\n");
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j)
+      printf("%d ", cm->fc.NMVcount.comps[i].class0[j]);
+    printf("\n"); fflush(stdout);
+  }
+  printf("bits count:\n");
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < MV_OFFSET_BITS; ++j)
+      printf("%d/%d ", cm->fc.NMVcount.comps[i].bits[j][0],
+                       cm->fc.NMVcount.comps[i].bits[j][1]);
+    printf("\n"); fflush(stdout);
+  }
+  printf("class0_fp count:\n");
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      printf("{");
+      for (k = 0; k < 4; ++k)
+        printf("%d ", cm->fc.NMVcount.comps[i].class0_fp[j][k]);
+      printf("}, ");
+    }
+    printf("\n"); fflush(stdout);
+  }
+  printf("fp count:\n");
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < 4; ++j)
+      printf("%d ", cm->fc.NMVcount.comps[i].fp[j]);
+    printf("\n"); fflush(stdout);
+  }
+  if (usehp) {
+    printf("class0_hp count:\n");
+    for (i = 0; i < 2; ++i)
+      printf("%d/%d ", cm->fc.NMVcount.comps[i].class0_hp[0],
+                       cm->fc.NMVcount.comps[i].class0_hp[1]);
+    printf("\n"); fflush(stdout);
+    printf("hp count:\n");
+    for (i = 0; i < 2; ++i)
+      printf("%d/%d ", cm->fc.NMVcount.comps[i].hp[0],
+                       cm->fc.NMVcount.comps[i].hp[1]);
+    printf("\n"); fflush(stdout);
+  }
+#endif
+#ifdef SMOOTH_MV_COUNTS
+  smooth_counts(&cm->fc.NMVcount.comps[0]);
+  smooth_counts(&cm->fc.NMVcount.comps[1]);
+#endif
+  vp9_counts_process(&cm->fc.NMVcount, usehp);
+
+  adapt_probs(0, vp9_mv_joint_tree,
+              cm->fc.nmvc.joints, cm->fc.pre_nmvc.joints,
+              cm->fc.NMVcount.joints);
+
+  for (i = 0; i < 2; ++i) {
+    adapt_prob(&cm->fc.nmvc.comps[i].sign,
+               cm->fc.pre_nmvc.comps[i].sign,
+               cm->fc.NMVcount.comps[i].sign);
+    adapt_probs(0, vp9_mv_class_tree,
+                cm->fc.nmvc.comps[i].classes, cm->fc.pre_nmvc.comps[i].classes,
+                cm->fc.NMVcount.comps[i].classes);
+    adapt_probs(0, vp9_mv_class0_tree,
+                cm->fc.nmvc.comps[i].class0, cm->fc.pre_nmvc.comps[i].class0,
+                cm->fc.NMVcount.comps[i].class0);
+    for (j = 0; j < MV_OFFSET_BITS; ++j) {
+      adapt_prob(&cm->fc.nmvc.comps[i].bits[j],
+                 cm->fc.pre_nmvc.comps[i].bits[j],
+                 cm->fc.NMVcount.comps[i].bits[j]);
+    }
+  }
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      adapt_probs(0, vp9_mv_fp_tree,
+                  cm->fc.nmvc.comps[i].class0_fp[j],
+                  cm->fc.pre_nmvc.comps[i].class0_fp[j],
+                  cm->fc.NMVcount.comps[i].class0_fp[j]);
+    }
+    adapt_probs(0, vp9_mv_fp_tree,
+                cm->fc.nmvc.comps[i].fp,
+                cm->fc.pre_nmvc.comps[i].fp,
+                cm->fc.NMVcount.comps[i].fp);
+  }
+  if (usehp) {
+    for (i = 0; i < 2; ++i) {
+      adapt_prob(&cm->fc.nmvc.comps[i].class0_hp,
+                 cm->fc.pre_nmvc.comps[i].class0_hp,
+                 cm->fc.NMVcount.comps[i].class0_hp);
+      adapt_prob(&cm->fc.nmvc.comps[i].hp,
+                 cm->fc.pre_nmvc.comps[i].hp,
+                 cm->fc.NMVcount.comps[i].hp);
+    }
+  }
+}
+
+void vp9_entropy_mv_init() {
+  vp9_tokens_from_tree(vp9_mv_joint_encodings, vp9_mv_joint_tree);
+  vp9_tokens_from_tree(vp9_mv_class_encodings, vp9_mv_class_tree);
+  vp9_tokens_from_tree(vp9_mv_class0_encodings, vp9_mv_class0_tree);
+  vp9_tokens_from_tree(vp9_mv_fp_encodings, vp9_mv_fp_tree);
+}
+
+void vp9_init_mv_probs(VP9_COMMON *cm) {
+  vpx_memcpy(&cm->fc.nmvc, &vp9_default_nmv_context, sizeof(nmv_context));
+}
diff --git a/libvpx/vp9/common/vp9_entropymv.h b/libvpx/vp9/common/vp9_entropymv.h
new file mode 100644
index 0000000..15994a6
--- /dev/null
+++ b/libvpx/vp9/common/vp9_entropymv.h
@@ -0,0 +1,142 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_VP9_ENTROPYMV_H_
+#define VP9_COMMON_VP9_ENTROPYMV_H_
+
+#include "vp9/common/vp9_treecoder.h"
+#include "vpx_config.h"
+#include "vp9/common/vp9_blockd.h"
+
+struct VP9Common;
+
+void vp9_entropy_mv_init();
+void vp9_init_mv_probs(struct VP9Common *cm);
+
+void vp9_adapt_nmv_probs(struct VP9Common *cm, int usehp);
+int vp9_use_nmv_hp(const MV *ref);
+
+#define VP9_NMV_UPDATE_PROB  252
+
+//#define MV_GROUP_UPDATE
+
+#define LOW_PRECISION_MV_UPDATE  /* Use 7 bit forward update */
+
+/* Symbols for coding which components are zero jointly */
+#define MV_JOINTS     4
+typedef enum {
+  MV_JOINT_ZERO = 0,             /* Zero vector */
+  MV_JOINT_HNZVZ = 1,            /* Vert zero, hor nonzero */
+  MV_JOINT_HZVNZ = 2,            /* Hor zero, vert nonzero */
+  MV_JOINT_HNZVNZ = 3,           /* Both components nonzero */
+} MV_JOINT_TYPE;
+
+static INLINE int mv_joint_vertical(MV_JOINT_TYPE type) {
+  return type == MV_JOINT_HZVNZ || type == MV_JOINT_HNZVNZ;
+}
+
+static INLINE int mv_joint_horizontal(MV_JOINT_TYPE type) {
+  return type == MV_JOINT_HNZVZ || type == MV_JOINT_HNZVNZ;
+}
+
+extern const vp9_tree_index vp9_mv_joint_tree[2 * MV_JOINTS - 2];
+extern struct vp9_token vp9_mv_joint_encodings[MV_JOINTS];
+
+/* Symbols for coding magnitude class of nonzero components */
+#define MV_CLASSES     11
+typedef enum {
+  MV_CLASS_0 = 0,      /* (0, 2]     integer pel */
+  MV_CLASS_1 = 1,      /* (2, 4]     integer pel */
+  MV_CLASS_2 = 2,      /* (4, 8]     integer pel */
+  MV_CLASS_3 = 3,      /* (8, 16]    integer pel */
+  MV_CLASS_4 = 4,      /* (16, 32]   integer pel */
+  MV_CLASS_5 = 5,      /* (32, 64]   integer pel */
+  MV_CLASS_6 = 6,      /* (64, 128]  integer pel */
+  MV_CLASS_7 = 7,      /* (128, 256] integer pel */
+  MV_CLASS_8 = 8,      /* (256, 512] integer pel */
+  MV_CLASS_9 = 9,      /* (512, 1024] integer pel */
+  MV_CLASS_10 = 10,    /* (1024,2048] integer pel */
+} MV_CLASS_TYPE;
+
+extern const vp9_tree_index vp9_mv_class_tree[2 * MV_CLASSES - 2];
+extern struct vp9_token vp9_mv_class_encodings[MV_CLASSES];
+
+#define CLASS0_BITS    1  /* bits at integer precision for class 0 */
+#define CLASS0_SIZE    (1 << CLASS0_BITS)
+#define MV_OFFSET_BITS (MV_CLASSES + CLASS0_BITS - 2)
+
+#define MV_MAX_BITS    (MV_CLASSES + CLASS0_BITS + 2)
+#define MV_MAX         ((1 << MV_MAX_BITS) - 1)
+#define MV_VALS        ((MV_MAX << 1) + 1)
+
+extern const vp9_tree_index vp9_mv_class0_tree[2 * CLASS0_SIZE - 2];
+extern struct vp9_token vp9_mv_class0_encodings[CLASS0_SIZE];
+
+extern const vp9_tree_index vp9_mv_fp_tree[2 * 4 - 2];
+extern struct vp9_token vp9_mv_fp_encodings[4];
+
+typedef struct {
+  vp9_prob sign;
+  vp9_prob classes[MV_CLASSES - 1];
+  vp9_prob class0[CLASS0_SIZE - 1];
+  vp9_prob bits[MV_OFFSET_BITS];
+  vp9_prob class0_fp[CLASS0_SIZE][4 - 1];
+  vp9_prob fp[4 - 1];
+  vp9_prob class0_hp;
+  vp9_prob hp;
+} nmv_component;
+
+typedef struct {
+  vp9_prob joints[MV_JOINTS - 1];
+  nmv_component comps[2];
+} nmv_context;
+
+MV_JOINT_TYPE vp9_get_mv_joint(const MV *mv);
+MV_CLASS_TYPE vp9_get_mv_class(int z, int *offset);
+int vp9_get_mv_mag(MV_CLASS_TYPE c, int offset);
+
+
+typedef struct {
+  unsigned int mvcount[MV_VALS];
+  unsigned int sign[2];
+  unsigned int classes[MV_CLASSES];
+  unsigned int class0[CLASS0_SIZE];
+  unsigned int bits[MV_OFFSET_BITS][2];
+  unsigned int class0_fp[CLASS0_SIZE][4];
+  unsigned int fp[4];
+  unsigned int class0_hp[2];
+  unsigned int hp[2];
+} nmv_component_counts;
+
+typedef struct {
+  unsigned int joints[MV_JOINTS];
+  nmv_component_counts comps[2];
+} nmv_context_counts;
+
+void vp9_increment_nmv(const MV *mv, const MV *ref, nmv_context_counts *mvctx,
+                       int usehp);
+extern const nmv_context vp9_default_nmv_context;
+void vp9_counts_to_nmv_context(
+    nmv_context_counts *NMVcount,
+    nmv_context *prob,
+    int usehp,
+    unsigned int (*branch_ct_joint)[2],
+    unsigned int (*branch_ct_sign)[2],
+    unsigned int (*branch_ct_classes)[MV_CLASSES - 1][2],
+    unsigned int (*branch_ct_class0)[CLASS0_SIZE - 1][2],
+    unsigned int (*branch_ct_bits)[MV_OFFSET_BITS][2],
+    unsigned int (*branch_ct_class0_fp)[CLASS0_SIZE][4 - 1][2],
+    unsigned int (*branch_ct_fp)[4 - 1][2],
+    unsigned int (*branch_ct_class0_hp)[2],
+    unsigned int (*branch_ct_hp)[2]);
+void vp9_counts_process(nmv_context_counts *NMVcount, int usehp);
+
+#endif  // VP9_COMMON_VP9_ENTROPYMV_H_
diff --git a/libvpx/vp9/common/vp9_enums.h b/libvpx/vp9/common/vp9_enums.h
new file mode 100644
index 0000000..e18d353
--- /dev/null
+++ b/libvpx/vp9/common/vp9_enums.h
@@ -0,0 +1,49 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ENUMS_H_
+#define VP9_COMMON_VP9_ENUMS_H_
+
+#include "./vpx_config.h"
+
+#define LOG2_MI_SIZE 3
+
+#define MI_SIZE (1 << LOG2_MI_SIZE)
+#define MI_MASK ((64 >> LOG2_MI_SIZE) - 1)
+
+typedef enum BLOCK_SIZE_TYPE {
+  BLOCK_SIZE_AB4X4,
+  BLOCK_SIZE_SB4X8,
+  BLOCK_SIZE_SB8X4,
+  BLOCK_SIZE_SB8X8,
+  BLOCK_SIZE_SB8X16,
+  BLOCK_SIZE_SB16X8,
+  BLOCK_SIZE_MB16X16,
+  BLOCK_SIZE_SB16X32,
+  BLOCK_SIZE_SB32X16,
+  BLOCK_SIZE_SB32X32,
+  BLOCK_SIZE_SB32X64,
+  BLOCK_SIZE_SB64X32,
+  BLOCK_SIZE_SB64X64,
+  BLOCK_SIZE_TYPES
+} BLOCK_SIZE_TYPE;
+
+typedef enum PARTITION_TYPE {
+  PARTITION_NONE,
+  PARTITION_HORZ,
+  PARTITION_VERT,
+  PARTITION_SPLIT,
+  PARTITION_TYPES
+} PARTITION_TYPE;
+
+#define PARTITION_PLOFFSET   4  // number of probability models per block size
+#define NUM_PARTITION_CONTEXTS (4 * PARTITION_PLOFFSET)
+
+#endif  // VP9_COMMON_VP9_ENUMS_H_
diff --git a/libvpx/vp9/common/vp9_extend.c b/libvpx/vp9/common/vp9_extend.c
new file mode 100644
index 0000000..95ec590
--- /dev/null
+++ b/libvpx/vp9/common/vp9_extend.c
@@ -0,0 +1,133 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_extend.h"
+#include "vpx_mem/vpx_mem.h"
+
+static void copy_and_extend_plane(const uint8_t *src, int src_pitch,
+                                  uint8_t *dst, int dst_pitch,
+                                  int w, int h,
+                                  int extend_top, int extend_left,
+                                  int extend_bottom, int extend_right) {
+  int i, linesize;
+
+  // copy the left and right most columns out
+  const uint8_t *src_ptr1 = src;
+  const uint8_t *src_ptr2 = src + w - 1;
+  uint8_t *dst_ptr1 = dst - extend_left;
+  uint8_t *dst_ptr2 = dst + w;
+
+  for (i = 0; i < h; i++) {
+    vpx_memset(dst_ptr1, src_ptr1[0], extend_left);
+    vpx_memcpy(dst_ptr1 + extend_left, src_ptr1, w);
+    vpx_memset(dst_ptr2, src_ptr2[0], extend_right);
+    src_ptr1 += src_pitch;
+    src_ptr2 += src_pitch;
+    dst_ptr1 += dst_pitch;
+    dst_ptr2 += dst_pitch;
+  }
+
+  // Now copy the top and bottom lines into each line of the respective
+  // borders
+  src_ptr1 = dst - extend_left;
+  src_ptr2 = dst + dst_pitch * (h - 1) - extend_left;
+  dst_ptr1 = dst + dst_pitch * (-extend_top) - extend_left;
+  dst_ptr2 = dst + dst_pitch * (h) - extend_left;
+  linesize = extend_left + extend_right + w;
+
+  for (i = 0; i < extend_top; i++) {
+    vpx_memcpy(dst_ptr1, src_ptr1, linesize);
+    dst_ptr1 += dst_pitch;
+  }
+
+  for (i = 0; i < extend_bottom; i++) {
+    vpx_memcpy(dst_ptr2, src_ptr2, linesize);
+    dst_ptr2 += dst_pitch;
+  }
+}
+
+void vp9_copy_and_extend_frame(const YV12_BUFFER_CONFIG *src,
+                               YV12_BUFFER_CONFIG *dst) {
+  const int et_y = dst->border;
+  const int el_y = dst->border;
+  const int eb_y = dst->border + dst->y_height - src->y_height;
+  const int er_y = dst->border + dst->y_width - src->y_width;
+
+  const int et_uv = dst->border >> (dst->uv_height != dst->y_height);
+  const int el_uv = dst->border >> (dst->uv_width != dst->y_width);
+  const int eb_uv = et_uv + dst->uv_height - src->uv_height;
+  const int er_uv = el_uv + dst->uv_width - src->uv_width;
+
+#if CONFIG_ALPHA
+  const int et_a = dst->border >> (dst->alpha_height != dst->y_height);
+  const int el_a = dst->border >> (dst->alpha_width != dst->y_width);
+  const int eb_a = et_a + dst->alpha_height - src->alpha_height;
+  const int er_a = el_a + dst->alpha_width - src->alpha_width;
+
+  copy_and_extend_plane(src->alpha_buffer, src->alpha_stride,
+                        dst->alpha_buffer, dst->alpha_stride,
+                        src->alpha_width, src->alpha_height,
+                        et_a, el_a, eb_a, er_a);
+#endif
+
+  copy_and_extend_plane(src->y_buffer, src->y_stride,
+                        dst->y_buffer, dst->y_stride,
+                        src->y_width, src->y_height,
+                        et_y, el_y, eb_y, er_y);
+
+  copy_and_extend_plane(src->u_buffer, src->uv_stride,
+                        dst->u_buffer, dst->uv_stride,
+                        src->uv_width, src->uv_height,
+                        et_uv, el_uv, eb_uv, er_uv);
+
+  copy_and_extend_plane(src->v_buffer, src->uv_stride,
+                        dst->v_buffer, dst->uv_stride,
+                        src->uv_width, src->uv_height,
+                        et_uv, el_uv, eb_uv, er_uv);
+}
+
+void vp9_copy_and_extend_frame_with_rect(const YV12_BUFFER_CONFIG *src,
+                                         YV12_BUFFER_CONFIG *dst,
+                                         int srcy, int srcx,
+                                         int srch, int srcw) {
+  // If the side is not touching the bounder then don't extend.
+  const int et_y = srcy ? 0 : dst->border;
+  const int el_y = srcx ? 0 : dst->border;
+  const int eb_y = srcy + srch != src->y_height ? 0 :
+                      dst->border + dst->y_height - src->y_height;
+  const int er_y = srcx + srcw != src->y_width ? 0 :
+                      dst->border + dst->y_width - src->y_width;
+  const int src_y_offset = srcy * src->y_stride + srcx;
+  const int dst_y_offset = srcy * dst->y_stride + srcx;
+
+  const int et_uv = (et_y + 1) >> 1;
+  const int el_uv = (el_y + 1) >> 1;
+  const int eb_uv = (eb_y + 1) >> 1;
+  const int er_uv = (er_y + 1) >> 1;
+  const int src_uv_offset = ((srcy * src->uv_stride) >> 1) + (srcx >> 1);
+  const int dst_uv_offset = ((srcy * dst->uv_stride) >> 1) + (srcx >> 1);
+  const int srch_uv = (srch + 1) >> 1;
+  const int srcw_uv = (srcw + 1) >> 1;
+
+  copy_and_extend_plane(src->y_buffer + src_y_offset, src->y_stride,
+                        dst->y_buffer + dst_y_offset, dst->y_stride,
+                        srcw, srch,
+                        et_y, el_y, eb_y, er_y);
+
+  copy_and_extend_plane(src->u_buffer + src_uv_offset, src->uv_stride,
+                        dst->u_buffer + dst_uv_offset, dst->uv_stride,
+                        srcw_uv, srch_uv,
+                        et_uv, el_uv, eb_uv, er_uv);
+
+  copy_and_extend_plane(src->v_buffer + src_uv_offset, src->uv_stride,
+                        dst->v_buffer + dst_uv_offset, dst->uv_stride,
+                        srcw_uv, srch_uv,
+                        et_uv, el_uv, eb_uv, er_uv);
+}
diff --git a/libvpx/vp9/common/vp9_extend.h b/libvpx/vp9/common/vp9_extend.h
new file mode 100644
index 0000000..7ff79b7
--- /dev/null
+++ b/libvpx/vp9/common/vp9_extend.h
@@ -0,0 +1,25 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_EXTEND_H_
+#define VP9_COMMON_VP9_EXTEND_H_
+
+#include "vpx_scale/yv12config.h"
+#include "vpx/vpx_integer.h"
+
+
+void vp9_copy_and_extend_frame(const YV12_BUFFER_CONFIG *src,
+                               YV12_BUFFER_CONFIG *dst);
+
+void vp9_copy_and_extend_frame_with_rect(const YV12_BUFFER_CONFIG *src,
+                                         YV12_BUFFER_CONFIG *dst,
+                                         int srcy, int srcx,
+                                         int srch, int srcw);
+#endif  // VP9_COMMON_VP9_EXTEND_H_
diff --git a/libvpx/vp9/common/vp9_filter.c b/libvpx/vp9/common/vp9_filter.c
new file mode 100644
index 0000000..e5503cd
--- /dev/null
+++ b/libvpx/vp9/common/vp9_filter.c
@@ -0,0 +1,97 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+#include "vp9/common/vp9_filter.h"
+#include "vpx_ports/mem.h"
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+
+DECLARE_ALIGNED(256, const int16_t, vp9_bilinear_filters[SUBPEL_SHIFTS][8]) = {
+  { 0, 0, 0, 128,   0, 0, 0, 0 },
+  { 0, 0, 0, 120,   8, 0, 0, 0 },
+  { 0, 0, 0, 112,  16, 0, 0, 0 },
+  { 0, 0, 0, 104,  24, 0, 0, 0 },
+  { 0, 0, 0,  96,  32, 0, 0, 0 },
+  { 0, 0, 0,  88,  40, 0, 0, 0 },
+  { 0, 0, 0,  80,  48, 0, 0, 0 },
+  { 0, 0, 0,  72,  56, 0, 0, 0 },
+  { 0, 0, 0,  64,  64, 0, 0, 0 },
+  { 0, 0, 0,  56,  72, 0, 0, 0 },
+  { 0, 0, 0,  48,  80, 0, 0, 0 },
+  { 0, 0, 0,  40,  88, 0, 0, 0 },
+  { 0, 0, 0,  32,  96, 0, 0, 0 },
+  { 0, 0, 0,  24, 104, 0, 0, 0 },
+  { 0, 0, 0,  16, 112, 0, 0, 0 },
+  { 0, 0, 0,   8, 120, 0, 0, 0 }
+};
+
+DECLARE_ALIGNED(256, const int16_t, vp9_sub_pel_filters_8[SUBPEL_SHIFTS][8]) = {
+  /* Lagrangian interpolation filter */
+  { 0,   0,   0, 128,   0,   0,   0,  0},
+  { 0,   1,  -5, 126,   8,  -3,   1,  0},
+  { -1,   3, -10, 122,  18,  -6,   2,  0},
+  { -1,   4, -13, 118,  27,  -9,   3, -1},
+  { -1,   4, -16, 112,  37, -11,   4, -1},
+  { -1,   5, -18, 105,  48, -14,   4, -1},
+  { -1,   5, -19,  97,  58, -16,   5, -1},
+  { -1,   6, -19,  88,  68, -18,   5, -1},
+  { -1,   6, -19,  78,  78, -19,   6, -1},
+  { -1,   5, -18,  68,  88, -19,   6, -1},
+  { -1,   5, -16,  58,  97, -19,   5, -1},
+  { -1,   4, -14,  48, 105, -18,   5, -1},
+  { -1,   4, -11,  37, 112, -16,   4, -1},
+  { -1,   3,  -9,  27, 118, -13,   4, -1},
+  { 0,   2,  -6,  18, 122, -10,   3, -1},
+  { 0,   1,  -3,   8, 126,  -5,   1,  0}
+};
+
+DECLARE_ALIGNED(256, const int16_t, vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][8])
+    = {
+  /* dct based filter */
+  {0,   0,   0, 128,   0,   0,   0, 0},
+  {-1,   3,  -7, 127,   8,  -3,   1, 0},
+  {-2,   5, -13, 125,  17,  -6,   3, -1},
+  {-3,   7, -17, 121,  27, -10,   5, -2},
+  {-4,   9, -20, 115,  37, -13,   6, -2},
+  {-4,  10, -23, 108,  48, -16,   8, -3},
+  {-4,  10, -24, 100,  59, -19,   9, -3},
+  {-4,  11, -24,  90,  70, -21,  10, -4},
+  {-4,  11, -23,  80,  80, -23,  11, -4},
+  {-4,  10, -21,  70,  90, -24,  11, -4},
+  {-3,   9, -19,  59, 100, -24,  10, -4},
+  {-3,   8, -16,  48, 108, -23,  10, -4},
+  {-2,   6, -13,  37, 115, -20,   9, -4},
+  {-2,   5, -10,  27, 121, -17,   7, -3},
+  {-1,   3,  -6,  17, 125, -13,   5, -2},
+  {0,   1,  -3,   8, 127,  -7,   3, -1}
+};
+
+DECLARE_ALIGNED(256, const int16_t,
+                vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][8]) = {
+  /* freqmultiplier = 0.5 */
+  { 0,  0,  0, 128,  0,  0,  0,  0},
+  {-3, -1, 32,  64, 38,  1, -3,  0},
+  {-2, -2, 29,  63, 41,  2, -3,  0},
+  {-2, -2, 26,  63, 43,  4, -4,  0},
+  {-2, -3, 24,  62, 46,  5, -4,  0},
+  {-2, -3, 21,  60, 49,  7, -4,  0},
+  {-1, -4, 18,  59, 51,  9, -4,  0},
+  {-1, -4, 16,  57, 53, 12, -4, -1},
+  {-1, -4, 14,  55, 55, 14, -4, -1},
+  {-1, -4, 12,  53, 57, 16, -4, -1},
+  { 0, -4,  9,  51, 59, 18, -4, -1},
+  { 0, -4,  7,  49, 60, 21, -3, -2},
+  { 0, -4,  5,  46, 62, 24, -3, -2},
+  { 0, -4,  4,  43, 63, 26, -2, -2},
+  { 0, -3,  2,  41, 63, 29, -2, -2},
+  { 0, -3,  1,  38, 64, 32, -1, -3}
+};
diff --git a/libvpx/vp9/common/vp9_filter.h b/libvpx/vp9/common/vp9_filter.h
new file mode 100644
index 0000000..1ccfdaa
--- /dev/null
+++ b/libvpx/vp9/common/vp9_filter.h
@@ -0,0 +1,37 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_FILTER_H_
+#define VP9_COMMON_VP9_FILTER_H_
+
+#include "vpx_config.h"
+#include "vpx_scale/yv12config.h"
+#include "vpx/vpx_integer.h"
+
+#define BLOCK_HEIGHT_WIDTH 4
+#define VP9_FILTER_WEIGHT 128
+#define VP9_FILTER_SHIFT  7
+
+#define SUBPEL_SHIFTS 16
+
+extern const int16_t vp9_bilinear_filters[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_6[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][8];
+
+// The VP9_BILINEAR_FILTERS_2TAP macro returns a pointer to the bilinear
+// filter kernel as a 2 tap filter.
+#define BF_LENGTH (sizeof(vp9_bilinear_filters[0]) / \
+                   sizeof(vp9_bilinear_filters[0][0]))
+#define BF_OFFSET (BF_LENGTH / 2 - 1)
+#define VP9_BILINEAR_FILTERS_2TAP(x) (vp9_bilinear_filters[x] + BF_OFFSET)
+
+#endif  // VP9_COMMON_VP9_FILTER_H_
diff --git a/libvpx/vp9/common/vp9_findnearmv.c b/libvpx/vp9/common/vp9_findnearmv.c
new file mode 100644
index 0000000..a692271
--- /dev/null
+++ b/libvpx/vp9/common/vp9_findnearmv.c
@@ -0,0 +1,89 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <limits.h>
+
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_mvref_common.h"
+#include "vp9/common/vp9_sadmxn.h"
+
+static void lower_mv_precision(int_mv *mv, int usehp) {
+  if (!usehp || !vp9_use_nmv_hp(&mv->as_mv)) {
+    if (mv->as_mv.row & 1)
+      mv->as_mv.row += (mv->as_mv.row > 0 ? -1 : 1);
+    if (mv->as_mv.col & 1)
+      mv->as_mv.col += (mv->as_mv.col > 0 ? -1 : 1);
+  }
+}
+
+
+void vp9_find_best_ref_mvs(MACROBLOCKD *xd,
+                           int_mv *mvlist,
+                           int_mv *nearest,
+                           int_mv *near) {
+  int i;
+  // Make sure all the candidates are properly clamped etc
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
+    lower_mv_precision(&mvlist[i], xd->allow_high_precision_mv);
+    clamp_mv2(&mvlist[i], xd);
+  }
+  *nearest = mvlist[0];
+  *near = mvlist[1];
+}
+
+void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
+                                   int_mv *dst_nearest,
+                                   int_mv *dst_near,
+                                   int block_idx, int ref_idx) {
+  int_mv dst_list[MAX_MV_REF_CANDIDATES];
+  int_mv mv_list[MAX_MV_REF_CANDIDATES];
+  MODE_INFO *mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+
+  assert(ref_idx == 0 || ref_idx == 1);
+  assert(MAX_MV_REF_CANDIDATES == 2);  // makes code here slightly easier
+
+  vp9_find_mv_refs_idx(cm, xd, xd->mode_info_context,
+                       xd->prev_mode_info_context,
+                       mbmi->ref_frame[ref_idx],
+                       mv_list, cm->ref_frame_sign_bias, block_idx);
+
+  dst_list[1].as_int = 0;
+  if (block_idx == 0) {
+    memcpy(dst_list, mv_list, MAX_MV_REF_CANDIDATES * sizeof(int_mv));
+  } else if (block_idx == 1 || block_idx == 2) {
+    int dst = 0, n;
+    union b_mode_info *bmi = mi->bmi;
+
+    dst_list[dst++].as_int = bmi[0].as_mv[ref_idx].as_int;
+    for (n = 0; dst < MAX_MV_REF_CANDIDATES &&
+                n < MAX_MV_REF_CANDIDATES; n++)
+      if (mv_list[n].as_int != dst_list[0].as_int)
+        dst_list[dst++].as_int = mv_list[n].as_int;
+  } else {
+    int dst = 0, n;
+    union b_mode_info *bmi = mi->bmi;
+
+    assert(block_idx == 3);
+    dst_list[dst++].as_int = bmi[2].as_mv[ref_idx].as_int;
+    if (dst_list[0].as_int != bmi[1].as_mv[ref_idx].as_int)
+      dst_list[dst++].as_int = bmi[1].as_mv[ref_idx].as_int;
+    if (dst < MAX_MV_REF_CANDIDATES &&
+        dst_list[0].as_int != bmi[0].as_mv[ref_idx].as_int)
+      dst_list[dst++].as_int = bmi[0].as_mv[ref_idx].as_int;
+    for (n = 0; dst < MAX_MV_REF_CANDIDATES &&
+                n < MAX_MV_REF_CANDIDATES; n++)
+      if (mv_list[n].as_int != dst_list[0].as_int)
+        dst_list[dst++].as_int = mv_list[n].as_int;
+  }
+
+  dst_nearest->as_int = dst_list[0].as_int;
+  dst_near->as_int = dst_list[1].as_int;
+}
diff --git a/libvpx/vp9/common/vp9_findnearmv.h b/libvpx/vp9/common/vp9_findnearmv.h
new file mode 100644
index 0000000..d4ae210
--- /dev/null
+++ b/libvpx/vp9/common/vp9_findnearmv.h
@@ -0,0 +1,116 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_VP9_FINDNEARMV_H_
+#define VP9_COMMON_VP9_FINDNEARMV_H_
+
+#include "vp9/common/vp9_mv.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_treecoder.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+#define LEFT_TOP_MARGIN     ((VP9BORDERINPIXELS - VP9_INTERP_EXTEND) << 3)
+#define RIGHT_BOTTOM_MARGIN ((VP9BORDERINPIXELS - VP9_INTERP_EXTEND) << 3)
+
+// check a list of motion vectors by sad score using a number rows of pixels
+// above and a number cols of pixels in the left to select the one with best
+// score to use as ref motion vector
+void vp9_find_best_ref_mvs(MACROBLOCKD *xd,
+                           int_mv *mvlist,
+                           int_mv *nearest,
+                           int_mv *near);
+
+static void mv_bias(int refmb_ref_frame_sign_bias, int refframe,
+                    int_mv *mvp, const int *ref_frame_sign_bias) {
+  MV xmv = mvp->as_mv;
+
+  if (refmb_ref_frame_sign_bias != ref_frame_sign_bias[refframe]) {
+    xmv.row *= -1;
+    xmv.col *= -1;
+  }
+
+  mvp->as_mv = xmv;
+}
+
+// TODO(jingning): this mv clamping function should be block size dependent.
+static void clamp_mv(int_mv *mv,
+                     int mb_to_left_edge,
+                     int mb_to_right_edge,
+                     int mb_to_top_edge,
+                     int mb_to_bottom_edge) {
+  mv->as_mv.col = clamp(mv->as_mv.col, mb_to_left_edge, mb_to_right_edge);
+  mv->as_mv.row = clamp(mv->as_mv.row, mb_to_top_edge, mb_to_bottom_edge);
+}
+
+static int clamp_mv2(int_mv *mv, const MACROBLOCKD *xd) {
+  int_mv tmp_mv;
+  tmp_mv.as_int = mv->as_int;
+  clamp_mv(mv,
+           xd->mb_to_left_edge - LEFT_TOP_MARGIN,
+           xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN,
+           xd->mb_to_top_edge - LEFT_TOP_MARGIN,
+           xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN);
+  return tmp_mv.as_int != mv->as_int;
+}
+
+static int check_mv_bounds(int_mv *mv,
+                           int mb_to_left_edge, int mb_to_right_edge,
+                           int mb_to_top_edge, int mb_to_bottom_edge) {
+  return mv->as_mv.col < mb_to_left_edge ||
+         mv->as_mv.col > mb_to_right_edge ||
+         mv->as_mv.row < mb_to_top_edge ||
+         mv->as_mv.row > mb_to_bottom_edge;
+}
+
+void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *pc,
+                                   MACROBLOCKD *xd,
+                                   int_mv *dst_nearest,
+                                   int_mv *dst_near,
+                                   int block_idx, int ref_idx);
+
+static MB_PREDICTION_MODE left_block_mode(const MODE_INFO *cur_mb, int b) {
+  // FIXME(rbultje, jingning): temporary hack because jenkins doesn't
+  // understand this condition. This will go away soon.
+  if (b == 0 || b == 2) {
+    /* On L edge, get from MB to left of us */
+    --cur_mb;
+
+    if (cur_mb->mbmi.ref_frame[0] != INTRA_FRAME) {
+      return DC_PRED;
+    } else if (cur_mb->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+      return ((cur_mb->bmi + 1 + b)->as_mode.first);
+    } else {
+      return cur_mb->mbmi.mode;
+    }
+  }
+  assert(b == 1 || b == 3);
+  return (cur_mb->bmi + b - 1)->as_mode.first;
+}
+
+static MB_PREDICTION_MODE above_block_mode(const MODE_INFO *cur_mb,
+                                          int b, int mi_stride) {
+  if (!(b >> 1)) {
+    /* On top edge, get from MB above us */
+    cur_mb -= mi_stride;
+
+    if (cur_mb->mbmi.ref_frame[0] != INTRA_FRAME) {
+      return DC_PRED;
+    } else if (cur_mb->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+      return ((cur_mb->bmi + 2 + b)->as_mode.first);
+    } else {
+      return cur_mb->mbmi.mode;
+    }
+  }
+
+  return (cur_mb->bmi + b - 2)->as_mode.first;
+}
+
+#endif  // VP9_COMMON_VP9_FINDNEARMV_H_
diff --git a/libvpx/vp9/common/vp9_idct.c b/libvpx/vp9/common/vp9_idct.c
new file mode 100644
index 0000000..dcc7f03
--- /dev/null
+++ b/libvpx/vp9/common/vp9_idct.c
@@ -0,0 +1,1310 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <math.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_idct.h"
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
+   0.5 shifts per pixel. */
+  int i;
+  int16_t output[16];
+  int a1, b1, c1, d1, e1;
+  int16_t *ip = input;
+  int16_t *op = output;
+
+  for (i = 0; i < 4; i++) {
+    a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+    c1 = ip[1] >> WHT_UPSCALE_FACTOR;
+    d1 = ip[2] >> WHT_UPSCALE_FACTOR;
+    b1 = ip[3] >> WHT_UPSCALE_FACTOR;
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+    op[0] = a1;
+    op[1] = b1;
+    op[2] = c1;
+    op[3] = d1;
+    ip += 4;
+    op += 4;
+  }
+
+  ip = output;
+  for (i = 0; i < 4; i++) {
+    a1 = ip[4 * 0];
+    c1 = ip[4 * 1];
+    d1 = ip[4 * 2];
+    b1 = ip[4 * 3];
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+    dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
+    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + b1);
+    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + c1);
+    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + d1);
+
+    ip++;
+    dest++;
+  }
+}
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
+  int i;
+  int a1, e1;
+  int16_t tmp[4];
+  int16_t *ip = in;
+  int16_t *op = tmp;
+
+  a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+  e1 = a1 >> 1;
+  a1 -= e1;
+  op[0] = a1;
+  op[1] = op[2] = op[3] = e1;
+
+  ip = tmp;
+  for (i = 0; i < 4; i++) {
+    e1 = ip[0] >> 1;
+    a1 = ip[0] - e1;
+    dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
+    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + e1);
+    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + e1);
+    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + e1);
+    ip++;
+    dest++;
+  }
+}
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output) {
+  int16_t step[4];
+  int temp1, temp2;
+  // stage 1
+  temp1 = (input[0] + input[2]) * cospi_16_64;
+  temp2 = (input[0] - input[2]) * cospi_16_64;
+  step[0] = dct_const_round_shift(temp1);
+  step[1] = dct_const_round_shift(temp2);
+  temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
+  temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
+  step[2] = dct_const_round_shift(temp1);
+  step[3] = dct_const_round_shift(temp2);
+
+  // stage 2
+  output[0] = step[0] + step[3];
+  output[1] = step[1] + step[2];
+  output[2] = step[1] - step[2];
+  output[3] = step[0] - step[3];
+}
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+  int16_t out[4 * 4];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[4], temp_out[4];
+
+  // Rows
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = input[j];
+    vp9_idct4_1d(temp_in, outptr);
+    input += 4;
+    outptr += 4;
+  }
+
+  // Columns
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = out[j * 4 + i];
+    vp9_idct4_1d(temp_in, temp_out);
+    for (j = 0; j < 4; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+  int i;
+  int a1;
+  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  a1 = ROUND_POWER_OF_TWO(out, 4);
+
+  for (i = 0; i < 4; i++) {
+    dest[0] = clip_pixel(dest[0] + a1);
+    dest[1] = clip_pixel(dest[1] + a1);
+    dest[2] = clip_pixel(dest[2] + a1);
+    dest[3] = clip_pixel(dest[3] + a1);
+    dest += dest_stride;
+  }
+}
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr,
+                            uint8_t *dst_ptr, int pitch, int stride) {
+  int a1;
+  int r, c;
+  int16_t out = dct_const_round_shift(input_dc * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  a1 = ROUND_POWER_OF_TWO(out, 4);
+
+  for (r = 0; r < 4; r++) {
+    for (c = 0; c < 4; c++)
+      dst_ptr[c] = clip_pixel(a1 + pred_ptr[c]);
+
+    dst_ptr += stride;
+    pred_ptr += pitch;
+  }
+}
+
+static void idct8_1d(int16_t *input, int16_t *output) {
+  int16_t step1[8], step2[8];
+  int temp1, temp2;
+  // stage 1
+  step1[0] = input[0];
+  step1[2] = input[4];
+  step1[1] = input[2];
+  step1[3] = input[6];
+  temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
+  temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
+  step1[4] = dct_const_round_shift(temp1);
+  step1[7] = dct_const_round_shift(temp2);
+  temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
+  temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+
+  // stage 2 & stage 3 - even half
+  vp9_idct4_1d(step1, step1);
+
+  // stage 2 - odd half
+  step2[4] = step1[4] + step1[5];
+  step2[5] = step1[4] - step1[5];
+  step2[6] = -step1[6] + step1[7];
+  step2[7] = step1[6] + step1[7];
+
+  // stage 3 -odd half
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+  step1[7] = step2[7];
+
+  // stage 4
+  output[0] = step1[0] + step1[7];
+  output[1] = step1[1] + step1[6];
+  output[2] = step1[2] + step1[5];
+  output[3] = step1[3] + step1[4];
+  output[4] = step1[3] - step1[4];
+  output[5] = step1[2] - step1[5];
+  output[6] = step1[1] - step1[6];
+  output[7] = step1[0] - step1[7];
+}
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+  int16_t out[8 * 8];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[8], temp_out[8];
+
+  // First transform rows
+  for (i = 0; i < 8; ++i) {
+    idct8_1d(input, outptr);
+    input += 8;
+    outptr += 8;
+  }
+
+  // Then transform columns
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    idct8_1d(temp_in, temp_out);
+    for (j = 0; j < 8; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+static void iadst4_1d(int16_t *input, int16_t *output) {
+  int s0, s1, s2, s3, s4, s5, s6, s7;
+
+  int x0 = input[0];
+  int x1 = input[1];
+  int x2 = input[2];
+  int x3 = input[3];
+
+  if (!(x0 | x1 | x2 | x3)) {
+    output[0] = output[1] = output[2] = output[3] = 0;
+    return;
+  }
+
+  s0 = sinpi_1_9 * x0;
+  s1 = sinpi_2_9 * x0;
+  s2 = sinpi_3_9 * x1;
+  s3 = sinpi_4_9 * x2;
+  s4 = sinpi_1_9 * x2;
+  s5 = sinpi_2_9 * x3;
+  s6 = sinpi_4_9 * x3;
+  s7 = x0 - x2 + x3;
+
+  x0 = s0 + s3 + s5;
+  x1 = s1 - s4 - s6;
+  x2 = sinpi_3_9 * s7;
+  x3 = s2;
+
+  s0 = x0 + x3;
+  s1 = x1 + x3;
+  s2 = x2;
+  s3 = x0 + x1 - x3;
+
+  // 1-D transform scaling factor is sqrt(2).
+  // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
+  // + 1b (addition) = 29b.
+  // Hence the output bit depth is 15b.
+  output[0] = dct_const_round_shift(s0);
+  output[1] = dct_const_round_shift(s1);
+  output[2] = dct_const_round_shift(s2);
+  output[3] = dct_const_round_shift(s3);
+}
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride,
+                            int tx_type) {
+  const transform_2d IHT_4[] = {
+    { vp9_idct4_1d,  vp9_idct4_1d  },  // DCT_DCT  = 0
+    { iadst4_1d, vp9_idct4_1d  },      // ADST_DCT = 1
+    { vp9_idct4_1d,  iadst4_1d },      // DCT_ADST = 2
+    { iadst4_1d, iadst4_1d }           // ADST_ADST = 3
+  };
+
+  int i, j;
+  int16_t out[4 * 4];
+  int16_t *outptr = out;
+  int16_t temp_in[4], temp_out[4];
+
+  // inverse transform row vectors
+  for (i = 0; i < 4; ++i) {
+    IHT_4[tx_type].rows(input, outptr);
+    input  += 4;
+    outptr += 4;
+  }
+
+  // inverse transform column vectors
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = out[j * 4 + i];
+    IHT_4[tx_type].cols(temp_in, temp_out);
+    for (j = 0; j < 4; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+static void iadst8_1d(int16_t *input, int16_t *output) {
+  int s0, s1, s2, s3, s4, s5, s6, s7;
+
+  int x0 = input[7];
+  int x1 = input[0];
+  int x2 = input[5];
+  int x3 = input[2];
+  int x4 = input[3];
+  int x5 = input[4];
+  int x6 = input[1];
+  int x7 = input[6];
+
+  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
+    output[0] = output[1] = output[2] = output[3] = output[4]
+              = output[5] = output[6] = output[7] = 0;
+    return;
+  }
+
+  // stage 1
+  s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
+  s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
+  s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
+  s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
+  s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
+  s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
+  s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
+  s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
+
+  x0 = dct_const_round_shift(s0 + s4);
+  x1 = dct_const_round_shift(s1 + s5);
+  x2 = dct_const_round_shift(s2 + s6);
+  x3 = dct_const_round_shift(s3 + s7);
+  x4 = dct_const_round_shift(s0 - s4);
+  x5 = dct_const_round_shift(s1 - s5);
+  x6 = dct_const_round_shift(s2 - s6);
+  x7 = dct_const_round_shift(s3 - s7);
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 =  cospi_8_64  * x4 + cospi_24_64 * x5;
+  s5 =  cospi_24_64 * x4 - cospi_8_64  * x5;
+  s6 = -cospi_24_64 * x6 + cospi_8_64  * x7;
+  s7 =  cospi_8_64  * x6 + cospi_24_64 * x7;
+
+  x0 = s0 + s2;
+  x1 = s1 + s3;
+  x2 = s0 - s2;
+  x3 = s1 - s3;
+  x4 = dct_const_round_shift(s4 + s6);
+  x5 = dct_const_round_shift(s5 + s7);
+  x6 = dct_const_round_shift(s4 - s6);
+  x7 = dct_const_round_shift(s5 - s7);
+
+  // stage 3
+  s2 = cospi_16_64 * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (x6 - x7);
+
+  x2 = dct_const_round_shift(s2);
+  x3 = dct_const_round_shift(s3);
+  x6 = dct_const_round_shift(s6);
+  x7 = dct_const_round_shift(s7);
+
+  output[0] =  x0;
+  output[1] = -x4;
+  output[2] =  x6;
+  output[3] = -x2;
+  output[4] =  x3;
+  output[5] = -x7;
+  output[6] =  x5;
+  output[7] = -x1;
+}
+
+static const transform_2d IHT_8[] = {
+  { idct8_1d,  idct8_1d  },  // DCT_DCT  = 0
+  { iadst8_1d, idct8_1d  },  // ADST_DCT = 1
+  { idct8_1d,  iadst8_1d },  // DCT_ADST = 2
+  { iadst8_1d, iadst8_1d }   // ADST_ADST = 3
+};
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride,
+                            int tx_type) {
+  int i, j;
+  int16_t out[8 * 8];
+  int16_t *outptr = out;
+  int16_t temp_in[8], temp_out[8];
+  const transform_2d ht = IHT_8[tx_type];
+
+  // inverse transform row vectors
+  for (i = 0; i < 8; ++i) {
+    ht.rows(input, outptr);
+    input += 8;
+    outptr += 8;
+  }
+
+  // inverse transform column vectors
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    ht.cols(temp_in, temp_out);
+    for (j = 0; j < 8; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * dest_stride + i]);  }
+}
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest,
+                                int dest_stride) {
+  int16_t out[8 * 8];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[8], temp_out[8];
+
+  vpx_memset(out, 0, sizeof(out));
+  // First transform rows
+  // only first 4 row has non-zero coefs
+  for (i = 0; i < 4; ++i) {
+    idct8_1d(input, outptr);
+    input += 8;
+    outptr += 8;
+  }
+
+  // Then transform columns
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    idct8_1d(temp_in, temp_out);
+    for (j = 0; j < 8; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output) {
+  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  output[0] = ROUND_POWER_OF_TWO(out, 5);
+}
+
+static void idct16_1d(int16_t *input, int16_t *output) {
+  int16_t step1[16], step2[16];
+  int temp1, temp2;
+
+  // stage 1
+  step1[0] = input[0/2];
+  step1[1] = input[16/2];
+  step1[2] = input[8/2];
+  step1[3] = input[24/2];
+  step1[4] = input[4/2];
+  step1[5] = input[20/2];
+  step1[6] = input[12/2];
+  step1[7] = input[28/2];
+  step1[8] = input[2/2];
+  step1[9] = input[18/2];
+  step1[10] = input[10/2];
+  step1[11] = input[26/2];
+  step1[12] = input[6/2];
+  step1[13] = input[22/2];
+  step1[14] = input[14/2];
+  step1[15] = input[30/2];
+
+  // stage 2
+  step2[0] = step1[0];
+  step2[1] = step1[1];
+  step2[2] = step1[2];
+  step2[3] = step1[3];
+  step2[4] = step1[4];
+  step2[5] = step1[5];
+  step2[6] = step1[6];
+  step2[7] = step1[7];
+
+  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
+  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
+  step2[8] = dct_const_round_shift(temp1);
+  step2[15] = dct_const_round_shift(temp2);
+
+  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
+  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
+  step2[9] = dct_const_round_shift(temp1);
+  step2[14] = dct_const_round_shift(temp2);
+
+  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
+  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+
+  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
+  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
+  step2[11] = dct_const_round_shift(temp1);
+  step2[12] = dct_const_round_shift(temp2);
+
+  // stage 3
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[2];
+  step1[3] = step2[3];
+
+  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
+  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
+  step1[4] = dct_const_round_shift(temp1);
+  step1[7] = dct_const_round_shift(temp2);
+  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
+  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+
+  step1[8] = step2[8] + step2[9];
+  step1[9] = step2[8] - step2[9];
+  step1[10] = -step2[10] + step2[11];
+  step1[11] = step2[10] + step2[11];
+  step1[12] = step2[12] + step2[13];
+  step1[13] = step2[12] - step2[13];
+  step1[14] = -step2[14] + step2[15];
+  step1[15] = step2[14] + step2[15];
+
+  temp1 = (step1[0] + step1[1]) * cospi_16_64;
+  temp2 = (step1[0] - step1[1]) * cospi_16_64;
+  step2[0] = dct_const_round_shift(temp1);
+  step2[1] = dct_const_round_shift(temp2);
+  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
+  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
+  step2[2] = dct_const_round_shift(temp1);
+  step2[3] = dct_const_round_shift(temp2);
+  step2[4] = step1[4] + step1[5];
+  step2[5] = step1[4] - step1[5];
+  step2[6] = -step1[6] + step1[7];
+  step2[7] = step1[6] + step1[7];
+
+  step2[8] = step1[8];
+  step2[15] = step1[15];
+  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
+  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
+  step2[9] = dct_const_round_shift(temp1);
+  step2[14] = dct_const_round_shift(temp2);
+  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
+  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+
+  // stage 5
+  step1[0] = step2[0] + step2[3];
+  step1[1] = step2[1] + step2[2];
+  step1[2] = step2[1] - step2[2];
+  step1[3] = step2[0] - step2[3];
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+  step1[7] = step2[7];
+
+  step1[8] = step2[8] + step2[11];
+  step1[9] = step2[9] + step2[10];
+  step1[10] = step2[9] - step2[10];
+  step1[11] = step2[8] - step2[11];
+  step1[12] = -step2[12] + step2[15];
+  step1[13] = -step2[13] + step2[14];
+  step1[14] = step2[13] + step2[14];
+  step1[15] = step2[12] + step2[15];
+
+  // stage 6
+  step2[0] = step1[0] + step1[7];
+  step2[1] = step1[1] + step1[6];
+  step2[2] = step1[2] + step1[5];
+  step2[3] = step1[3] + step1[4];
+  step2[4] = step1[3] - step1[4];
+  step2[5] = step1[2] - step1[5];
+  step2[6] = step1[1] - step1[6];
+  step2[7] = step1[0] - step1[7];
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
+  temp2 = (step1[10] + step1[13]) * cospi_16_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
+  temp2 = (step1[11] + step1[12]) * cospi_16_64;
+  step2[11] = dct_const_round_shift(temp1);
+  step2[12] = dct_const_round_shift(temp2);
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+
+  // stage 7
+  output[0] = step2[0] + step2[15];
+  output[1] = step2[1] + step2[14];
+  output[2] = step2[2] + step2[13];
+  output[3] = step2[3] + step2[12];
+  output[4] = step2[4] + step2[11];
+  output[5] = step2[5] + step2[10];
+  output[6] = step2[6] + step2[9];
+  output[7] = step2[7] + step2[8];
+  output[8] = step2[7] - step2[8];
+  output[9] = step2[6] - step2[9];
+  output[10] = step2[5] - step2[10];
+  output[11] = step2[4] - step2[11];
+  output[12] = step2[3] - step2[12];
+  output[13] = step2[2] - step2[13];
+  output[14] = step2[1] - step2[14];
+  output[15] = step2[0] - step2[15];
+}
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+  int16_t out[16 * 16];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[16], temp_out[16];
+
+  // First transform rows
+  for (i = 0; i < 16; ++i) {
+    idct16_1d(input, outptr);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Then transform columns
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j * 16 + i];
+    idct16_1d(temp_in, temp_out);
+    for (j = 0; j < 16; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+void iadst16_1d(int16_t *input, int16_t *output) {
+  int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
+
+  int x0 = input[15];
+  int x1 = input[0];
+  int x2 = input[13];
+  int x3 = input[2];
+  int x4 = input[11];
+  int x5 = input[4];
+  int x6 = input[9];
+  int x7 = input[6];
+  int x8 = input[7];
+  int x9 = input[8];
+  int x10 = input[5];
+  int x11 = input[10];
+  int x12 = input[3];
+  int x13 = input[12];
+  int x14 = input[1];
+  int x15 = input[14];
+
+  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8
+           | x9 | x10 | x11 | x12 | x13 | x14 | x15)) {
+    output[0] = output[1] = output[2] = output[3] = output[4]
+              = output[5] = output[6] = output[7] = output[8]
+              = output[9] = output[10] = output[11] = output[12]
+              = output[13] = output[14] = output[15] = 0;
+    return;
+  }
+
+  // stage 1
+  s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
+  s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
+  s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
+  s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
+  s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
+  s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
+  s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
+  s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
+  s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
+  s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
+  s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
+  s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
+  s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
+  s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
+  s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
+  s15 = x14 * cospi_3_64  - x15 * cospi_29_64;
+
+  x0 = dct_const_round_shift(s0 + s8);
+  x1 = dct_const_round_shift(s1 + s9);
+  x2 = dct_const_round_shift(s2 + s10);
+  x3 = dct_const_round_shift(s3 + s11);
+  x4 = dct_const_round_shift(s4 + s12);
+  x5 = dct_const_round_shift(s5 + s13);
+  x6 = dct_const_round_shift(s6 + s14);
+  x7 = dct_const_round_shift(s7 + s15);
+  x8  = dct_const_round_shift(s0 - s8);
+  x9  = dct_const_round_shift(s1 - s9);
+  x10 = dct_const_round_shift(s2 - s10);
+  x11 = dct_const_round_shift(s3 - s11);
+  x12 = dct_const_round_shift(s4 - s12);
+  x13 = dct_const_round_shift(s5 - s13);
+  x14 = dct_const_round_shift(s6 - s14);
+  x15 = dct_const_round_shift(s7 - s15);
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4;
+  s5 = x5;
+  s6 = x6;
+  s7 = x7;
+  s8 =    x8 * cospi_4_64   + x9 * cospi_28_64;
+  s9 =    x8 * cospi_28_64  - x9 * cospi_4_64;
+  s10 =   x10 * cospi_20_64 + x11 * cospi_12_64;
+  s11 =   x10 * cospi_12_64 - x11 * cospi_20_64;
+  s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
+  s13 =   x12 * cospi_4_64  + x13 * cospi_28_64;
+  s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
+  s15 =   x14 * cospi_20_64 + x15 * cospi_12_64;
+
+  x0 = s0 + s4;
+  x1 = s1 + s5;
+  x2 = s2 + s6;
+  x3 = s3 + s7;
+  x4 = s0 - s4;
+  x5 = s1 - s5;
+  x6 = s2 - s6;
+  x7 = s3 - s7;
+  x8 = dct_const_round_shift(s8 + s12);
+  x9 = dct_const_round_shift(s9 + s13);
+  x10 = dct_const_round_shift(s10 + s14);
+  x11 = dct_const_round_shift(s11 + s15);
+  x12 = dct_const_round_shift(s8 - s12);
+  x13 = dct_const_round_shift(s9 - s13);
+  x14 = dct_const_round_shift(s10 - s14);
+  x15 = dct_const_round_shift(s11 - s15);
+
+  // stage 3
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4 * cospi_8_64  + x5 * cospi_24_64;
+  s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
+  s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
+  s7 =   x6 * cospi_8_64  + x7 * cospi_24_64;
+  s8 = x8;
+  s9 = x9;
+  s10 = x10;
+  s11 = x11;
+  s12 = x12 * cospi_8_64  + x13 * cospi_24_64;
+  s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
+  s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
+  s15 =   x14 * cospi_8_64  + x15 * cospi_24_64;
+
+  x0 = s0 + s2;
+  x1 = s1 + s3;
+  x2 = s0 - s2;
+  x3 = s1 - s3;
+  x4 = dct_const_round_shift(s4 + s6);
+  x5 = dct_const_round_shift(s5 + s7);
+  x6 = dct_const_round_shift(s4 - s6);
+  x7 = dct_const_round_shift(s5 - s7);
+  x8 = s8 + s10;
+  x9 = s9 + s11;
+  x10 = s8 - s10;
+  x11 = s9 - s11;
+  x12 = dct_const_round_shift(s12 + s14);
+  x13 = dct_const_round_shift(s13 + s15);
+  x14 = dct_const_round_shift(s12 - s14);
+  x15 = dct_const_round_shift(s13 - s15);
+
+  // stage 4
+  s2 = (- cospi_16_64) * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (- x6 + x7);
+  s10 = cospi_16_64 * (x10 + x11);
+  s11 = cospi_16_64 * (- x10 + x11);
+  s14 = (- cospi_16_64) * (x14 + x15);
+  s15 = cospi_16_64 * (x14 - x15);
+
+  x2 = dct_const_round_shift(s2);
+  x3 = dct_const_round_shift(s3);
+  x6 = dct_const_round_shift(s6);
+  x7 = dct_const_round_shift(s7);
+  x10 = dct_const_round_shift(s10);
+  x11 = dct_const_round_shift(s11);
+  x14 = dct_const_round_shift(s14);
+  x15 = dct_const_round_shift(s15);
+
+  output[0] =  x0;
+  output[1] = -x8;
+  output[2] =  x12;
+  output[3] = -x4;
+  output[4] =  x6;
+  output[5] =  x14;
+  output[6] =  x10;
+  output[7] =  x2;
+  output[8] =  x3;
+  output[9] =  x11;
+  output[10] =  x15;
+  output[11] =  x7;
+  output[12] =  x5;
+  output[13] = -x13;
+  output[14] =  x9;
+  output[15] = -x1;
+}
+
+static const transform_2d IHT_16[] = {
+  { idct16_1d,  idct16_1d  },  // DCT_DCT  = 0
+  { iadst16_1d, idct16_1d  },  // ADST_DCT = 1
+  { idct16_1d,  iadst16_1d },  // DCT_ADST = 2
+  { iadst16_1d, iadst16_1d }   // ADST_ADST = 3
+};
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride,
+                              int tx_type) {
+  int i, j;
+  int16_t out[16 * 16];
+  int16_t *outptr = out;
+  int16_t temp_in[16], temp_out[16];
+  const transform_2d ht = IHT_16[tx_type];
+
+  // Rows
+  for (i = 0; i < 16; ++i) {
+    ht.rows(input, outptr);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Columns
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j * 16 + i];
+    ht.cols(temp_in, temp_out);
+    for (j = 0; j < 16; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * dest_stride + i]);  }
+}
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest,
+                                  int dest_stride) {
+  int16_t out[16 * 16];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[16], temp_out[16];
+
+  /* First transform rows. Since all non-zero dct coefficients are in
+   * upper-left 4x4 area, we only need to calculate first 4 rows here.
+   */
+  vpx_memset(out, 0, sizeof(out));
+  for (i = 0; i < 4; ++i) {
+    idct16_1d(input, outptr);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Then transform columns
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j*16 + i];
+    idct16_1d(temp_in, temp_out);
+    for (j = 0; j < 16; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output) {
+  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  output[0] = ROUND_POWER_OF_TWO(out, 6);
+}
+
+static void idct32_1d(int16_t *input, int16_t *output) {
+  int16_t step1[32], step2[32];
+  int temp1, temp2;
+
+  // stage 1
+  step1[0] = input[0];
+  step1[1] = input[16];
+  step1[2] = input[8];
+  step1[3] = input[24];
+  step1[4] = input[4];
+  step1[5] = input[20];
+  step1[6] = input[12];
+  step1[7] = input[28];
+  step1[8] = input[2];
+  step1[9] = input[18];
+  step1[10] = input[10];
+  step1[11] = input[26];
+  step1[12] = input[6];
+  step1[13] = input[22];
+  step1[14] = input[14];
+  step1[15] = input[30];
+
+  temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
+  temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
+  step1[16] = dct_const_round_shift(temp1);
+  step1[31] = dct_const_round_shift(temp2);
+
+  temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
+  temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
+  step1[17] = dct_const_round_shift(temp1);
+  step1[30] = dct_const_round_shift(temp2);
+
+  temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
+  temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
+  step1[18] = dct_const_round_shift(temp1);
+  step1[29] = dct_const_round_shift(temp2);
+
+  temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
+  temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
+  step1[19] = dct_const_round_shift(temp1);
+  step1[28] = dct_const_round_shift(temp2);
+
+  temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
+  temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
+  step1[20] = dct_const_round_shift(temp1);
+  step1[27] = dct_const_round_shift(temp2);
+
+  temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
+  temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
+  step1[21] = dct_const_round_shift(temp1);
+  step1[26] = dct_const_round_shift(temp2);
+
+  temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
+  temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
+  step1[22] = dct_const_round_shift(temp1);
+  step1[25] = dct_const_round_shift(temp2);
+
+  temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
+  temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
+  step1[23] = dct_const_round_shift(temp1);
+  step1[24] = dct_const_round_shift(temp2);
+
+  // stage 2
+  step2[0] = step1[0];
+  step2[1] = step1[1];
+  step2[2] = step1[2];
+  step2[3] = step1[3];
+  step2[4] = step1[4];
+  step2[5] = step1[5];
+  step2[6] = step1[6];
+  step2[7] = step1[7];
+
+  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
+  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
+  step2[8] = dct_const_round_shift(temp1);
+  step2[15] = dct_const_round_shift(temp2);
+
+  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
+  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
+  step2[9] = dct_const_round_shift(temp1);
+  step2[14] = dct_const_round_shift(temp2);
+
+  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
+  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+
+  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
+  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
+  step2[11] = dct_const_round_shift(temp1);
+  step2[12] = dct_const_round_shift(temp2);
+
+  step2[16] = step1[16] + step1[17];
+  step2[17] = step1[16] - step1[17];
+  step2[18] = -step1[18] + step1[19];
+  step2[19] = step1[18] + step1[19];
+  step2[20] = step1[20] + step1[21];
+  step2[21] = step1[20] - step1[21];
+  step2[22] = -step1[22] + step1[23];
+  step2[23] = step1[22] + step1[23];
+  step2[24] = step1[24] + step1[25];
+  step2[25] = step1[24] - step1[25];
+  step2[26] = -step1[26] + step1[27];
+  step2[27] = step1[26] + step1[27];
+  step2[28] = step1[28] + step1[29];
+  step2[29] = step1[28] - step1[29];
+  step2[30] = -step1[30] + step1[31];
+  step2[31] = step1[30] + step1[31];
+
+  // stage 3
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[2];
+  step1[3] = step2[3];
+
+  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
+  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
+  step1[4] = dct_const_round_shift(temp1);
+  step1[7] = dct_const_round_shift(temp2);
+  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
+  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+
+  step1[8] = step2[8] + step2[9];
+  step1[9] = step2[8] - step2[9];
+  step1[10] = -step2[10] + step2[11];
+  step1[11] = step2[10] + step2[11];
+  step1[12] = step2[12] + step2[13];
+  step1[13] = step2[12] - step2[13];
+  step1[14] = -step2[14] + step2[15];
+  step1[15] = step2[14] + step2[15];
+
+  step1[16] = step2[16];
+  step1[31] = step2[31];
+  temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64;
+  temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64;
+  step1[17] = dct_const_round_shift(temp1);
+  step1[30] = dct_const_round_shift(temp2);
+  temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64;
+  temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64;
+  step1[18] = dct_const_round_shift(temp1);
+  step1[29] = dct_const_round_shift(temp2);
+  step1[19] = step2[19];
+  step1[20] = step2[20];
+  temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64;
+  temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64;
+  step1[21] = dct_const_round_shift(temp1);
+  step1[26] = dct_const_round_shift(temp2);
+  temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64;
+  temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64;
+  step1[22] = dct_const_round_shift(temp1);
+  step1[25] = dct_const_round_shift(temp2);
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[27] = step2[27];
+  step1[28] = step2[28];
+
+  // stage 4
+  temp1 = (step1[0] + step1[1]) * cospi_16_64;
+  temp2 = (step1[0] - step1[1]) * cospi_16_64;
+  step2[0] = dct_const_round_shift(temp1);
+  step2[1] = dct_const_round_shift(temp2);
+  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
+  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
+  step2[2] = dct_const_round_shift(temp1);
+  step2[3] = dct_const_round_shift(temp2);
+  step2[4] = step1[4] + step1[5];
+  step2[5] = step1[4] - step1[5];
+  step2[6] = -step1[6] + step1[7];
+  step2[7] = step1[6] + step1[7];
+
+  step2[8] = step1[8];
+  step2[15] = step1[15];
+  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
+  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
+  step2[9] = dct_const_round_shift(temp1);
+  step2[14] = dct_const_round_shift(temp2);
+  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
+  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+
+  step2[16] = step1[16] + step1[19];
+  step2[17] = step1[17] + step1[18];
+  step2[18] = step1[17] - step1[18];
+  step2[19] = step1[16] - step1[19];
+  step2[20] = -step1[20] + step1[23];
+  step2[21] = -step1[21] + step1[22];
+  step2[22] = step1[21] + step1[22];
+  step2[23] = step1[20] + step1[23];
+
+  step2[24] = step1[24] + step1[27];
+  step2[25] = step1[25] + step1[26];
+  step2[26] = step1[25] - step1[26];
+  step2[27] = step1[24] - step1[27];
+  step2[28] = -step1[28] + step1[31];
+  step2[29] = -step1[29] + step1[30];
+  step2[30] = step1[29] + step1[30];
+  step2[31] = step1[28] + step1[31];
+
+  // stage 5
+  step1[0] = step2[0] + step2[3];
+  step1[1] = step2[1] + step2[2];
+  step1[2] = step2[1] - step2[2];
+  step1[3] = step2[0] - step2[3];
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = dct_const_round_shift(temp1);
+  step1[6] = dct_const_round_shift(temp2);
+  step1[7] = step2[7];
+
+  step1[8] = step2[8] + step2[11];
+  step1[9] = step2[9] + step2[10];
+  step1[10] = step2[9] - step2[10];
+  step1[11] = step2[8] - step2[11];
+  step1[12] = -step2[12] + step2[15];
+  step1[13] = -step2[13] + step2[14];
+  step1[14] = step2[13] + step2[14];
+  step1[15] = step2[12] + step2[15];
+
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64;
+  temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64;
+  step1[18] = dct_const_round_shift(temp1);
+  step1[29] = dct_const_round_shift(temp2);
+  temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64;
+  temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64;
+  step1[19] = dct_const_round_shift(temp1);
+  step1[28] = dct_const_round_shift(temp2);
+  temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64;
+  temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64;
+  step1[20] = dct_const_round_shift(temp1);
+  step1[27] = dct_const_round_shift(temp2);
+  temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64;
+  temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64;
+  step1[21] = dct_const_round_shift(temp1);
+  step1[26] = dct_const_round_shift(temp2);
+  step1[22] = step2[22];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[25] = step2[25];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+
+  // stage 6
+  step2[0] = step1[0] + step1[7];
+  step2[1] = step1[1] + step1[6];
+  step2[2] = step1[2] + step1[5];
+  step2[3] = step1[3] + step1[4];
+  step2[4] = step1[3] - step1[4];
+  step2[5] = step1[2] - step1[5];
+  step2[6] = step1[1] - step1[6];
+  step2[7] = step1[0] - step1[7];
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
+  temp2 = (step1[10] + step1[13]) * cospi_16_64;
+  step2[10] = dct_const_round_shift(temp1);
+  step2[13] = dct_const_round_shift(temp2);
+  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
+  temp2 = (step1[11] + step1[12]) * cospi_16_64;
+  step2[11] = dct_const_round_shift(temp1);
+  step2[12] = dct_const_round_shift(temp2);
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+
+  step2[16] = step1[16] + step1[23];
+  step2[17] = step1[17] + step1[22];
+  step2[18] = step1[18] + step1[21];
+  step2[19] = step1[19] + step1[20];
+  step2[20] = step1[19] - step1[20];
+  step2[21] = step1[18] - step1[21];
+  step2[22] = step1[17] - step1[22];
+  step2[23] = step1[16] - step1[23];
+
+  step2[24] = -step1[24] + step1[31];
+  step2[25] = -step1[25] + step1[30];
+  step2[26] = -step1[26] + step1[29];
+  step2[27] = -step1[27] + step1[28];
+  step2[28] = step1[27] + step1[28];
+  step2[29] = step1[26] + step1[29];
+  step2[30] = step1[25] + step1[30];
+  step2[31] = step1[24] + step1[31];
+
+  // stage 7
+  step1[0] = step2[0] + step2[15];
+  step1[1] = step2[1] + step2[14];
+  step1[2] = step2[2] + step2[13];
+  step1[3] = step2[3] + step2[12];
+  step1[4] = step2[4] + step2[11];
+  step1[5] = step2[5] + step2[10];
+  step1[6] = step2[6] + step2[9];
+  step1[7] = step2[7] + step2[8];
+  step1[8] = step2[7] - step2[8];
+  step1[9] = step2[6] - step2[9];
+  step1[10] = step2[5] - step2[10];
+  step1[11] = step2[4] - step2[11];
+  step1[12] = step2[3] - step2[12];
+  step1[13] = step2[2] - step2[13];
+  step1[14] = step2[1] - step2[14];
+  step1[15] = step2[0] - step2[15];
+
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[18] = step2[18];
+  step1[19] = step2[19];
+  temp1 = (-step2[20] + step2[27]) * cospi_16_64;
+  temp2 = (step2[20] + step2[27]) * cospi_16_64;
+  step1[20] = dct_const_round_shift(temp1);
+  step1[27] = dct_const_round_shift(temp2);
+  temp1 = (-step2[21] + step2[26]) * cospi_16_64;
+  temp2 = (step2[21] + step2[26]) * cospi_16_64;
+  step1[21] = dct_const_round_shift(temp1);
+  step1[26] = dct_const_round_shift(temp2);
+  temp1 = (-step2[22] + step2[25]) * cospi_16_64;
+  temp2 = (step2[22] + step2[25]) * cospi_16_64;
+  step1[22] = dct_const_round_shift(temp1);
+  step1[25] = dct_const_round_shift(temp2);
+  temp1 = (-step2[23] + step2[24]) * cospi_16_64;
+  temp2 = (step2[23] + step2[24]) * cospi_16_64;
+  step1[23] = dct_const_round_shift(temp1);
+  step1[24] = dct_const_round_shift(temp2);
+  step1[28] = step2[28];
+  step1[29] = step2[29];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+
+  // final stage
+  output[0] = step1[0] + step1[31];
+  output[1] = step1[1] + step1[30];
+  output[2] = step1[2] + step1[29];
+  output[3] = step1[3] + step1[28];
+  output[4] = step1[4] + step1[27];
+  output[5] = step1[5] + step1[26];
+  output[6] = step1[6] + step1[25];
+  output[7] = step1[7] + step1[24];
+  output[8] = step1[8] + step1[23];
+  output[9] = step1[9] + step1[22];
+  output[10] = step1[10] + step1[21];
+  output[11] = step1[11] + step1[20];
+  output[12] = step1[12] + step1[19];
+  output[13] = step1[13] + step1[18];
+  output[14] = step1[14] + step1[17];
+  output[15] = step1[15] + step1[16];
+  output[16] = step1[15] - step1[16];
+  output[17] = step1[14] - step1[17];
+  output[18] = step1[13] - step1[18];
+  output[19] = step1[12] - step1[19];
+  output[20] = step1[11] - step1[20];
+  output[21] = step1[10] - step1[21];
+  output[22] = step1[9] - step1[22];
+  output[23] = step1[8] - step1[23];
+  output[24] = step1[7] - step1[24];
+  output[25] = step1[6] - step1[25];
+  output[26] = step1[5] - step1[26];
+  output[27] = step1[4] - step1[27];
+  output[28] = step1[3] - step1[28];
+  output[29] = step1[2] - step1[29];
+  output[30] = step1[1] - step1[30];
+  output[31] = step1[0] - step1[31];
+}
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+  int16_t out[32 * 32];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[32], temp_out[32];
+
+  // Rows
+  for (i = 0; i < 32; ++i) {
+    idct32_1d(input, outptr);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    idct32_1d(temp_in, temp_out);
+    for (j = 0; j < 32; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * dest_stride + i]);
+  }
+}
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output) {
+  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  output[0] = ROUND_POWER_OF_TWO(out, 6);
+}
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest,
+                                  int dest_stride) {
+  int16_t out[32 * 32];
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[32], temp_out[32];
+
+  /* First transform rows. Since all non-zero dct coefficients are in
+   * upper-left 4x4 area, we only need to calculate first 4 rows here.
+   */
+  vpx_memset(out, 0, sizeof(out));
+  for (i = 0; i < 4; ++i) {
+    idct32_1d(input, outptr);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    idct32_1d(temp_in, temp_out);
+    for (j = 0; j < 32; ++j)
+      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * dest_stride + i]);
+  }
+}
diff --git a/libvpx/vp9/common/vp9_idct.h b/libvpx/vp9/common/vp9_idct.h
new file mode 100644
index 0000000..64f14c9
--- /dev/null
+++ b/libvpx/vp9/common/vp9_idct.h
@@ -0,0 +1,80 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_IDCT_H_
+#define VP9_COMMON_VP9_IDCT_H_
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+
+
+// Constants and Macros used by all idct/dct functions
+#define DCT_CONST_BITS 14
+#define DCT_CONST_ROUNDING  (1 << (DCT_CONST_BITS - 1))
+
+#define pair_set_epi16(a, b) \
+  _mm_set1_epi32(((uint16_t)(a)) + (((uint16_t)(b)) << 16))
+
+// Constants are round(16384 * cos(k*Pi/64)) where k = 1 to 31.
+// Note: sin(k*Pi/64) = cos((32-k)*Pi/64)
+static const int cospi_1_64  = 16364;
+static const int cospi_2_64  = 16305;
+static const int cospi_3_64  = 16207;
+static const int cospi_4_64  = 16069;
+static const int cospi_5_64  = 15893;
+static const int cospi_6_64  = 15679;
+static const int cospi_7_64  = 15426;
+static const int cospi_8_64  = 15137;
+static const int cospi_9_64  = 14811;
+static const int cospi_10_64 = 14449;
+static const int cospi_11_64 = 14053;
+static const int cospi_12_64 = 13623;
+static const int cospi_13_64 = 13160;
+static const int cospi_14_64 = 12665;
+static const int cospi_15_64 = 12140;
+static const int cospi_16_64 = 11585;
+static const int cospi_17_64 = 11003;
+static const int cospi_18_64 = 10394;
+static const int cospi_19_64 = 9760;
+static const int cospi_20_64 = 9102;
+static const int cospi_21_64 = 8423;
+static const int cospi_22_64 = 7723;
+static const int cospi_23_64 = 7005;
+static const int cospi_24_64 = 6270;
+static const int cospi_25_64 = 5520;
+static const int cospi_26_64 = 4756;
+static const int cospi_27_64 = 3981;
+static const int cospi_28_64 = 3196;
+static const int cospi_29_64 = 2404;
+static const int cospi_30_64 = 1606;
+static const int cospi_31_64 = 804;
+
+//  16384 * sqrt(2) * sin(kPi/9) * 2 / 3
+static const int sinpi_1_9 = 5283;
+static const int sinpi_2_9 = 9929;
+static const int sinpi_3_9 = 13377;
+static const int sinpi_4_9 = 15212;
+
+static INLINE int dct_const_round_shift(int input) {
+  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  assert(INT16_MIN <= rv && rv <= INT16_MAX);
+  return rv;
+}
+
+typedef void (*transform_1d)(int16_t*, int16_t*);
+
+typedef struct {
+  transform_1d cols, rows;  // vertical and horizontal
+} transform_2d;
+
+#endif  // VP9_COMMON_VP9_IDCT_H_
diff --git a/libvpx/vp9/common/vp9_implicit_segmentation.c b/libvpx/vp9/common/vp9_implicit_segmentation.c
new file mode 100644
index 0000000..2a1d35f
--- /dev/null
+++ b/libvpx/vp9/common/vp9_implicit_segmentation.c
@@ -0,0 +1,253 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_onyxc_int.h"
+
+#define MAX_REGIONS 24000
+#ifndef NULL
+#define NULL 0
+#endif
+
+#define min_mbs_in_region 3
+
+// this linked list structure holds equivalences for connected
+// component labeling
+struct list_el {
+  int label;
+  int seg_value;
+  int count;
+  struct list_el *next;
+};
+typedef struct list_el item;
+
+// connected colorsegments
+typedef struct {
+  int min_x;
+  int min_y;
+  int max_x;
+  int max_y;
+  int64_t sum_x;
+  int64_t sum_y;
+  int pixels;
+  int seg_value;
+  int label;
+} segment_info;
+
+
+typedef enum {
+  SEGMENT_MODE,
+  SEGMENT_MV,
+  SEGMENT_REFFRAME,
+  SEGMENT_SKIPPED
+} SEGMENT_TYPE;
+
+
+// this merges the two equivalence lists and
+// then makes sure that every label points to the same
+// equivalence list
+void merge(item *labels, int u, int v) {
+  item *a = labels[u].next;
+  item *b = labels[v].next;
+  item c;
+  item *it = &c;
+  int count;
+
+  // check if they are already merged
+  if (u == v || a == b)
+    return;
+
+  count = a->count + b->count;
+
+  // merge 2 sorted linked lists.
+  while (a != NULL && b != NULL) {
+    if (a->label < b->label) {
+      it->next = a;
+      a = a->next;
+    } else {
+      it->next = b;
+      b = b->next;
+    }
+
+    it = it->next;
+  }
+
+  if (a == NULL)
+    it->next = b;
+  else
+    it->next = a;
+
+  it = c.next;
+
+  // make sure every equivalence in the linked list points to this new ll
+  while (it != NULL) {
+    labels[it->label].next = c.next;
+    it = it->next;
+  }
+  c.next->count = count;
+
+}
+
+void segment_via_mode_info(VP9_COMMON *oci, int how) {
+  MODE_INFO *mi = oci->mi;
+  int i, j;
+  int mb_index = 0;
+
+  int label = 1;
+  int pitch = oci->mb_cols;
+
+  // holds linked list equivalences
+  // the max should probably be allocated at a higher level in oci
+  item equivalences[MAX_REGIONS];
+  int eq_ptr = 0;
+  item labels[MAX_REGIONS];
+  segment_info segments[MAX_REGIONS];
+  int label_count = 1;
+  int labeling[400 * 300];
+  int *lp = labeling;
+
+  label_count = 1;
+  memset(labels, 0, sizeof(labels));
+  memset(segments, 0, sizeof(segments));
+
+  /* Go through each macroblock first pass labelling */
+  for (i = 0; i < oci->mb_rows; i++, lp += pitch) {
+    for (j = 0; j < oci->mb_cols; j++) {
+      // int above seg_value, left seg_value, this seg_value...
+      int a = -1, l = -1, n = -1;
+
+      // above label, left label
+      int al = -1, ll = -1;
+      if (i) {
+        al = lp[j - pitch];
+        a = labels[al].next->seg_value;
+      }
+      if (j) {
+        ll = lp[j - 1];
+        l = labels[ll].next->seg_value;
+      }
+
+      // what setting are we going to do the implicit segmentation on
+      switch (how) {
+        case SEGMENT_MODE:
+          n = mi[mb_index].mbmi.mode;
+          break;
+        case SEGMENT_MV:
+          n = mi[mb_index].mbmi.mv[0].as_int;
+          if (mi[mb_index].mbmi.ref_frame[0] == INTRA_FRAME)
+            n = -9999999;
+          break;
+        case SEGMENT_REFFRAME:
+          n = mi[mb_index].mbmi.ref_frame[0];
+          break;
+        case SEGMENT_SKIPPED:
+          n = mi[mb_index].mbmi.mb_skip_coeff;
+          break;
+      }
+
+      // above and left both have the same seg_value
+      if (n == a && n == l) {
+        // pick the lowest label
+        lp[j] = (al < ll ? al : ll);
+        labels[lp[j]].next->count++;
+
+        // merge the above and left equivalencies
+        merge(labels, al, ll);
+      }
+      // this matches above seg_value
+      else if (n == a) {
+        // give it the same label as above
+        lp[j] = al;
+        labels[al].next->count++;
+      }
+      // this matches left seg_value
+      else if (n == l) {
+        // give it the same label as above
+        lp[j] = ll;
+        labels[ll].next->count++;
+      } else {
+        // new label doesn't match either
+        item *e = &labels[label];
+        item *nl = &equivalences[eq_ptr++];
+        lp[j] = label;
+        nl->label = label;
+        nl->next = 0;
+        nl->seg_value = n;
+        nl->count = 1;
+        e->next = nl;
+        label++;
+      }
+      mb_index++;
+    }
+    mb_index++;
+  }
+  lp = labeling;
+
+  // give new labels to regions
+  for (i = 1; i < label; i++)
+    if (labels[i].next->count > min_mbs_in_region &&
+        labels[labels[i].next->label].label == 0) {
+      segment_info *cs = &segments[label_count];
+      cs->label = label_count;
+      labels[labels[i].next->label].label = label_count++;
+      labels[labels[i].next->label].seg_value = labels[i].next->seg_value;
+      cs->seg_value = labels[labels[i].next->label].seg_value;
+      cs->min_x = oci->mb_cols;
+      cs->min_y = oci->mb_rows;
+      cs->max_x = 0;
+      cs->max_y = 0;
+      cs->sum_x = 0;
+      cs->sum_y = 0;
+      cs->pixels = 0;
+    }
+
+  lp = labeling;
+
+  // this is just to gather stats...
+  for (i = 0; i < oci->mb_rows; i++, lp += pitch) {
+    for (j = 0; j < oci->mb_cols; j++) {
+      const int old_lab = labels[lp[j]].next->label;
+      const int lab = labels[old_lab].label;
+      segment_info *cs = &segments[lab];
+
+      cs->min_x = MIN(cs->min_x, j);
+      cs->max_x = MAX(cs->max_x, j);
+      cs->min_y = MIN(cs->min_y, i);
+      cs->max_y = MAX(cs->max_y, i);
+      cs->sum_x += j;
+      cs->sum_y += i;
+      cs->pixels++;
+
+      lp[j] = lab;
+      mb_index++;
+    }
+    mb_index++;
+  }
+
+  {
+    lp = labeling;
+    printf("labelling \n");
+    mb_index = 0;
+    for (i = 0; i < oci->mb_rows; i++, lp += pitch) {
+      for (j = 0; j < oci->mb_cols; j++) {
+        printf("%4d", lp[j]);
+      }
+      printf("            ");
+      for (j = 0; j < oci->mb_cols; j++, mb_index++) {
+        // printf("%3d",mi[mb_index].mbmi.mode );
+        printf("%4d:%4d", mi[mb_index].mbmi.mv[0].as_mv.row,
+            mi[mb_index].mbmi.mv[0].as_mv.col);
+      }
+      printf("\n");
+      ++mb_index;
+    }
+    printf("\n");
+  }
+}
+
diff --git a/libvpx/vp9/common/vp9_loopfilter.c b/libvpx/vp9/common/vp9_loopfilter.c
new file mode 100644
index 0000000..7b3f0be
--- /dev/null
+++ b/libvpx/vp9/common/vp9_loopfilter.c
@@ -0,0 +1,407 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/common/vp9_seg_common.h"
+
+static void lf_init_lut(loop_filter_info_n *lfi) {
+  lfi->mode_lf_lut[DC_PRED] = 0;
+  lfi->mode_lf_lut[D45_PRED] = 0;
+  lfi->mode_lf_lut[D135_PRED] = 0;
+  lfi->mode_lf_lut[D117_PRED] = 0;
+  lfi->mode_lf_lut[D153_PRED] = 0;
+  lfi->mode_lf_lut[D27_PRED] = 0;
+  lfi->mode_lf_lut[D63_PRED] = 0;
+  lfi->mode_lf_lut[V_PRED] = 0;
+  lfi->mode_lf_lut[H_PRED] = 0;
+  lfi->mode_lf_lut[TM_PRED] = 0;
+  lfi->mode_lf_lut[ZEROMV]  = 0;
+  lfi->mode_lf_lut[NEARESTMV] = 1;
+  lfi->mode_lf_lut[NEARMV] = 1;
+  lfi->mode_lf_lut[NEWMV] = 1;
+}
+
+void vp9_loop_filter_update_sharpness(loop_filter_info_n *lfi,
+                                      int sharpness_lvl) {
+  int i;
+
+  /* For each possible value for the loop filter fill out limits */
+  for (i = 0; i <= MAX_LOOP_FILTER; i++) {
+    int filt_lvl = i;
+    int block_inside_limit = 0;
+
+    /* Set loop filter paramaeters that control sharpness. */
+    block_inside_limit = filt_lvl >> (sharpness_lvl > 0);
+    block_inside_limit = block_inside_limit >> (sharpness_lvl > 4);
+
+    if (sharpness_lvl > 0) {
+      if (block_inside_limit > (9 - sharpness_lvl))
+        block_inside_limit = (9 - sharpness_lvl);
+    }
+
+    if (block_inside_limit < 1)
+      block_inside_limit = 1;
+
+    vpx_memset(lfi->lim[i], block_inside_limit, SIMD_WIDTH);
+    vpx_memset(lfi->blim[i], (2 * filt_lvl + block_inside_limit),
+               SIMD_WIDTH);
+    vpx_memset(lfi->mblim[i], (2 * (filt_lvl + 2) + block_inside_limit),
+               SIMD_WIDTH);
+  }
+}
+
+void vp9_loop_filter_init(VP9_COMMON *cm) {
+  loop_filter_info_n *lfi = &cm->lf_info;
+  int i;
+
+  // init limits for given sharpness
+  vp9_loop_filter_update_sharpness(lfi, cm->sharpness_level);
+  cm->last_sharpness_level = cm->sharpness_level;
+
+  // init LUT for lvl  and hev thr picking
+  lf_init_lut(lfi);
+
+  // init hev threshold const vectors
+  for (i = 0; i < 4; i++)
+    vpx_memset(lfi->hev_thr[i], i, SIMD_WIDTH);
+}
+
+void vp9_loop_filter_frame_init(VP9_COMMON *cm,
+                                MACROBLOCKD *xd,
+                                int default_filt_lvl) {
+  int seg,    // segment number
+      ref,    // index in ref_lf_deltas
+      mode;   // index in mode_lf_deltas
+  // n_shift is the a multiplier for lf_deltas
+  // the multiplier is 1 for when filter_lvl is between 0 and 31;
+  // 2 when filter_lvl is between 32 and 63
+  int n_shift = default_filt_lvl >> 5;
+
+  loop_filter_info_n *lfi = &cm->lf_info;
+
+  /* update limits if sharpness has changed */
+  // printf("vp9_loop_filter_frame_init %d\n", default_filt_lvl);
+  // printf("sharpness level: %d [%d]\n",
+  //        cm->sharpness_level, cm->last_sharpness_level);
+  if (cm->last_sharpness_level != cm->sharpness_level) {
+    vp9_loop_filter_update_sharpness(lfi, cm->sharpness_level);
+    cm->last_sharpness_level = cm->sharpness_level;
+  }
+
+  for (seg = 0; seg < MAX_MB_SEGMENTS; seg++) {
+    int lvl_seg = default_filt_lvl;
+    int lvl_ref, lvl_mode;
+
+
+    // Set the baseline filter values for each segment
+    if (vp9_segfeature_active(xd, seg, SEG_LVL_ALT_LF)) {
+      /* Abs value */
+      if (xd->mb_segment_abs_delta == SEGMENT_ABSDATA) {
+        lvl_seg = vp9_get_segdata(xd, seg, SEG_LVL_ALT_LF);
+      } else { /* Delta Value */
+        lvl_seg += vp9_get_segdata(xd, seg, SEG_LVL_ALT_LF);
+        lvl_seg = clamp(lvl_seg, 0, 63);
+      }
+    }
+
+    if (!xd->mode_ref_lf_delta_enabled) {
+      /* we could get rid of this if we assume that deltas are set to
+       * zero when not in use; encoder always uses deltas
+       */
+      vpx_memset(lfi->lvl[seg][0], lvl_seg, 4 * 4);
+      continue;
+    }
+
+    lvl_ref = lvl_seg;
+
+    /* INTRA_FRAME */
+    ref = INTRA_FRAME;
+
+    /* Apply delta for reference frame */
+    lvl_ref += xd->ref_lf_deltas[ref] << n_shift;
+
+    mode = 0; /* all the rest of Intra modes */
+    lvl_mode = lvl_ref;
+    lfi->lvl[seg][ref][mode] = clamp(lvl_mode, 0, 63);
+
+    /* LAST, GOLDEN, ALT */
+    for (ref = 1; ref < MAX_REF_FRAMES; ref++) {
+      int lvl_ref = lvl_seg;
+
+      /* Apply delta for reference frame */
+      lvl_ref += xd->ref_lf_deltas[ref] << n_shift;
+
+      /* Apply delta for Inter modes */
+      for (mode = 0; mode < MAX_MODE_LF_DELTAS; mode++) {
+        lvl_mode = lvl_ref + (xd->mode_lf_deltas[mode] << n_shift);
+        lfi->lvl[seg][ref][mode] = clamp(lvl_mode, 0, 63);
+      }
+    }
+  }
+}
+
+static int build_lfi(const VP9_COMMON *cm, const MB_MODE_INFO *mbmi,
+                      struct loop_filter_info *lfi) {
+  const loop_filter_info_n *lfi_n = &cm->lf_info;
+  int mode = mbmi->mode;
+  int mode_index = lfi_n->mode_lf_lut[mode];
+  int seg = mbmi->segment_id;
+  int ref_frame = mbmi->ref_frame[0];
+  int filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
+
+  if (filter_level) {
+    const int hev_index = filter_level >> 4;
+    lfi->mblim = lfi_n->mblim[filter_level];
+    lfi->blim = lfi_n->blim[filter_level];
+    lfi->lim = lfi_n->lim[filter_level];
+    lfi->hev_thr = lfi_n->hev_thr[hev_index];
+    return 1;
+  }
+  return 0;
+}
+
+static void filter_selectively_vert(uint8_t *s, int pitch,
+                                    unsigned int mask_16x16,
+                                    unsigned int mask_8x8,
+                                    unsigned int mask_4x4,
+                                    unsigned int mask_4x4_int,
+                                    const struct loop_filter_info *lfi) {
+  unsigned int mask;
+
+  for (mask = mask_16x16 | mask_8x8 | mask_4x4; mask; mask >>= 1) {
+    if (mask & 1) {
+      if (mask_16x16 & 1) {
+        vp9_mb_lpf_vertical_edge_w(s, pitch, lfi->mblim, lfi->lim,
+                                   lfi->hev_thr);
+        assert(!(mask_8x8 & 1));
+        assert(!(mask_4x4 & 1));
+        assert(!(mask_4x4_int & 1));
+      } else if (mask_8x8 & 1) {
+        vp9_mbloop_filter_vertical_edge(s, pitch, lfi->mblim, lfi->lim,
+                                        lfi->hev_thr, 1);
+        assert(!(mask_16x16 & 1));
+        assert(!(mask_4x4 & 1));
+      } else if (mask_4x4 & 1) {
+        vp9_loop_filter_vertical_edge(s, pitch, lfi->mblim, lfi->lim,
+                                      lfi->hev_thr, 1);
+        assert(!(mask_16x16 & 1));
+        assert(!(mask_8x8 & 1));
+      } else {
+        assert(0);
+      }
+
+      if (mask_4x4_int & 1)
+        vp9_loop_filter_vertical_edge(s + 4, pitch, lfi->mblim, lfi->lim,
+                                      lfi->hev_thr, 1);
+    }
+    s += 8;
+    lfi++;
+    mask_16x16 >>= 1;
+    mask_8x8 >>= 1;
+    mask_4x4 >>= 1;
+    mask_4x4_int >>= 1;
+  }
+}
+
+static void filter_selectively_horiz(uint8_t *s, int pitch,
+                                     unsigned int mask_16x16,
+                                     unsigned int mask_8x8,
+                                     unsigned int mask_4x4,
+                                     unsigned int mask_4x4_int,
+                                     int only_4x4_1,
+                                     const struct loop_filter_info *lfi) {
+  unsigned int mask;
+
+  for (mask = mask_16x16 | mask_8x8 | mask_4x4; mask; mask >>= 1) {
+    if (mask & 1) {
+      if (!only_4x4_1) {
+        if (mask_16x16 & 1) {
+          vp9_mb_lpf_horizontal_edge_w(s, pitch, lfi->mblim, lfi->lim,
+                                       lfi->hev_thr);
+          assert(!(mask_8x8 & 1));
+          assert(!(mask_4x4 & 1));
+          assert(!(mask_4x4_int & 1));
+        } else if (mask_8x8 & 1) {
+          vp9_mbloop_filter_horizontal_edge(s, pitch, lfi->mblim, lfi->lim,
+                                            lfi->hev_thr, 1);
+          assert(!(mask_16x16 & 1));
+          assert(!(mask_4x4 & 1));
+        } else if (mask_4x4 & 1) {
+          vp9_loop_filter_horizontal_edge(s, pitch, lfi->mblim, lfi->lim,
+                                          lfi->hev_thr, 1);
+          assert(!(mask_16x16 & 1));
+          assert(!(mask_8x8 & 1));
+        } else {
+          assert(0);
+        }
+      }
+
+      if (mask_4x4_int & 1)
+        vp9_loop_filter_horizontal_edge(s + 4 * pitch, pitch, lfi->mblim,
+                                        lfi->lim, lfi->hev_thr, 1);
+    }
+    s += 8;
+    lfi++;
+    mask_16x16 >>= 1;
+    mask_8x8 >>= 1;
+    mask_4x4 >>= 1;
+    mask_4x4_int >>= 1;
+  }
+}
+
+static void filter_block_plane(VP9_COMMON *cm, MACROBLOCKD *xd,
+                               int plane, int mi_row, int mi_col) {
+  const int ss_x = xd->plane[plane].subsampling_x;
+  const int ss_y = xd->plane[plane].subsampling_y;
+  const int row_step = 1 << xd->plane[plane].subsampling_y;
+  const int col_step = 1 << xd->plane[plane].subsampling_x;
+  struct buf_2d * const dst = &xd->plane[plane].dst;
+  uint8_t* const dst0 = dst->buf;
+  MODE_INFO* const mi0 = xd->mode_info_context;
+  unsigned int mask_16x16[64 / MI_SIZE] = {0};
+  unsigned int mask_8x8[64 / MI_SIZE] = {0};
+  unsigned int mask_4x4[64 / MI_SIZE] = {0};
+  unsigned int mask_4x4_int[64 / MI_SIZE] = {0};
+  struct loop_filter_info lfi[64 / MI_SIZE][64 / MI_SIZE];
+  int r, c;
+
+  for (r = 0; r < 64 / MI_SIZE && mi_row + r < cm->mi_rows; r += row_step) {
+    unsigned int mask_16x16_c = 0;
+    unsigned int mask_8x8_c = 0;
+    unsigned int mask_4x4_c = 0;
+    unsigned int border_mask;
+
+    // Determine the vertical edges that need filtering
+    for (c = 0; c < 64 / MI_SIZE && mi_col + c < cm->mi_cols; c += col_step) {
+      const MODE_INFO * const mi = xd->mode_info_context;
+      const int skip_this = mi[c].mbmi.mb_skip_coeff
+                            && mi[c].mbmi.ref_frame[0] != INTRA_FRAME;
+      // left edge of current unit is block/partition edge -> no skip
+      const int block_edge_left = b_width_log2(mi[c].mbmi.sb_type) ?
+          !(c & ((1 << (b_width_log2(mi[c].mbmi.sb_type)-1)) - 1)) : 1;
+      const int skip_this_c = skip_this && !block_edge_left;
+      // top edge of current unit is block/partition edge -> no skip
+      const int block_edge_above = b_height_log2(mi[c].mbmi.sb_type) ?
+          !(r & ((1 << (b_height_log2(mi[c].mbmi.sb_type)-1)) - 1)) : 1;
+      const int skip_this_r = skip_this && !block_edge_above;
+      const TX_SIZE tx_size = plane ? get_uv_tx_size(&mi[c].mbmi)
+                                    : mi[c].mbmi.txfm_size;
+      const int skip_border_4x4_c = ss_x && mi_col + c == cm->mi_cols - 1;
+      const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1;
+
+      // Filter level can vary per MI
+      if (!build_lfi(cm, &mi[c].mbmi,
+                     lfi[r] + (c >> xd->plane[plane].subsampling_x)))
+        continue;
+
+      // Build masks based on the transform size of each block
+      if (tx_size == TX_32X32) {
+        if (!skip_this_c && ((c >> ss_x) & 3) == 0) {
+          if (!skip_border_4x4_c)
+            mask_16x16_c |= 1 << (c >> ss_x);
+          else
+            mask_8x8_c |= 1 << (c >> ss_x);
+        }
+        if (!skip_this_r && ((r >> ss_y) & 3) == 0) {
+          if (!skip_border_4x4_r)
+            mask_16x16[r] |= 1 << (c >> ss_x);
+          else
+            mask_8x8[r] |= 1 << (c >> ss_x);
+        }
+      } else if (tx_size == TX_16X16) {
+        if (!skip_this_c && ((c >> ss_x) & 1) == 0) {
+          if (!skip_border_4x4_c)
+            mask_16x16_c |= 1 << (c >> ss_x);
+          else
+            mask_8x8_c |= 1 << (c >> ss_x);
+        }
+        if (!skip_this_r && ((r >> ss_y) & 1) == 0) {
+          if (!skip_border_4x4_r)
+            mask_16x16[r] |= 1 << (c >> ss_x);
+          else
+            mask_8x8[r] |= 1 << (c >> ss_x);
+        }
+      } else {
+        // force 8x8 filtering on 32x32 boundaries
+        if (!skip_this_c) {
+          if (tx_size == TX_8X8 || ((c >> ss_x) & 3) == 0)
+            mask_8x8_c |= 1 << (c >> ss_x);
+          else
+            mask_4x4_c |= 1 << (c >> ss_x);
+        }
+
+        if (!skip_this_r) {
+          if (tx_size == TX_8X8 || ((r >> ss_y) & 3) == 0)
+            mask_8x8[r] |= 1 << (c >> ss_x);
+          else
+            mask_4x4[r] |= 1 << (c >> ss_x);
+        }
+
+        if (!skip_this && tx_size < TX_8X8 && !skip_border_4x4_c)
+          mask_4x4_int[r] |= 1 << (c >> ss_x);
+      }
+    }
+
+    // Disable filtering on the leftmost column
+    border_mask = ~(mi_col == 0);
+    filter_selectively_vert(dst->buf, dst->stride,
+                            mask_16x16_c & border_mask,
+                            mask_8x8_c & border_mask,
+                            mask_4x4_c & border_mask,
+                            mask_4x4_int[r], lfi[r]);
+    dst->buf += 8 * dst->stride;
+    xd->mode_info_context += cm->mode_info_stride * row_step;
+  }
+
+  // Now do horizontal pass
+  dst->buf = dst0;
+  xd->mode_info_context = mi0;
+  for (r = 0; r < 64 / MI_SIZE && mi_row + r < cm->mi_rows; r += row_step) {
+    const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1;
+    const unsigned int mask_4x4_int_r = skip_border_4x4_r ? 0 : mask_4x4_int[r];
+
+    filter_selectively_horiz(dst->buf, dst->stride,
+                             mask_16x16[r],
+                             mask_8x8[r],
+                             mask_4x4[r],
+                             mask_4x4_int_r, mi_row + r == 0, lfi[r]);
+    dst->buf += 8 * dst->stride;
+    xd->mode_info_context += cm->mode_info_stride * row_step;
+  }
+}
+
+void vp9_loop_filter_frame(VP9_COMMON *cm,
+                           MACROBLOCKD *xd,
+                           int frame_filter_level,
+                           int y_only) {
+  int mi_row, mi_col;
+
+  // Initialize the loop filter for this frame.
+  vp9_loop_filter_frame_init(cm, xd, frame_filter_level);
+
+  for (mi_row = 0; mi_row < cm->mi_rows; mi_row += 64 / MI_SIZE) {
+    MODE_INFO* const mi = cm->mi + mi_row * cm->mode_info_stride;
+
+    for (mi_col = 0; mi_col < cm->mi_cols; mi_col += 64 / MI_SIZE) {
+      int plane;
+
+      setup_dst_planes(xd, cm->frame_to_show, mi_row, mi_col);
+      for (plane = 0; plane < (y_only ? 1 : MAX_MB_PLANE); plane++) {
+        xd->mode_info_context = mi + mi_col;
+        filter_block_plane(cm, xd, plane, mi_row, mi_col);
+      }
+    }
+  }
+}
diff --git a/libvpx/vp9/common/vp9_loopfilter.h b/libvpx/vp9/common/vp9_loopfilter.h
new file mode 100644
index 0000000..ce954c0
--- /dev/null
+++ b/libvpx/vp9/common/vp9_loopfilter.h
@@ -0,0 +1,85 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_LOOPFILTER_H_
+#define VP9_COMMON_VP9_LOOPFILTER_H_
+
+#include "vpx_ports/mem.h"
+#include "vpx_config.h"
+#include "vp9/common/vp9_blockd.h"
+
+#define MAX_LOOP_FILTER 63
+#define SIMD_WIDTH 16
+
+/* Need to align this structure so when it is declared and
+ * passed it can be loaded into vector registers.
+ */
+typedef struct {
+  DECLARE_ALIGNED(SIMD_WIDTH, unsigned char,
+                  mblim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
+  DECLARE_ALIGNED(SIMD_WIDTH, unsigned char,
+                  blim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
+  DECLARE_ALIGNED(SIMD_WIDTH, unsigned char,
+                  lim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
+  DECLARE_ALIGNED(SIMD_WIDTH, unsigned char,
+                  hev_thr[4][SIMD_WIDTH]);
+  unsigned char lvl[MAX_MB_SEGMENTS][4][4];
+  unsigned char mode_lf_lut[MB_MODE_COUNT];
+} loop_filter_info_n;
+
+struct loop_filter_info {
+  const unsigned char *mblim;
+  const unsigned char *blim;
+  const unsigned char *lim;
+  const unsigned char *hev_thr;
+};
+
+#define prototype_loopfilter(sym) \
+  void sym(uint8_t *src, int pitch, const unsigned char *blimit, \
+           const unsigned char *limit, const unsigned char *thresh, int count)
+
+#define prototype_loopfilter_block(sym) \
+  void sym(uint8_t *y, uint8_t *u, uint8_t *v, \
+           int ystride, int uv_stride, struct loop_filter_info *lfi)
+
+#if ARCH_X86 || ARCH_X86_64
+#include "x86/vp9_loopfilter_x86.h"
+#endif
+
+typedef void loop_filter_uvfunction(uint8_t *u,   /* source pointer */
+                                    int p,              /* pitch */
+                                    const unsigned char *blimit,
+                                    const unsigned char *limit,
+                                    const unsigned char *thresh,
+                                    uint8_t *v);
+
+/* assorted loopfilter functions which get used elsewhere */
+struct VP9Common;
+struct macroblockd;
+
+void vp9_loop_filter_init(struct VP9Common *cm);
+
+void vp9_loop_filter_frame_init(struct VP9Common *cm,
+                                struct macroblockd *mbd,
+                                int default_filt_lvl);
+
+void vp9_loop_filter_frame(struct VP9Common *cm,
+                           struct macroblockd *mbd,
+                           int filter_level,
+                           int y_only);
+
+void vp9_loop_filter_partial_frame(struct VP9Common *cm,
+                                   struct macroblockd *mbd,
+                                   int default_filt_lvl);
+
+void vp9_loop_filter_update_sharpness(loop_filter_info_n *lfi,
+                                      int sharpness_lvl);
+
+#endif  // VP9_COMMON_VP9_LOOPFILTER_H_
diff --git a/libvpx/vp9/common/vp9_loopfilter_filters.c b/libvpx/vp9/common/vp9_loopfilter_filters.c
new file mode 100644
index 0000000..0efbcaf
--- /dev/null
+++ b/libvpx/vp9/common/vp9_loopfilter_filters.c
@@ -0,0 +1,308 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+static INLINE int8_t signed_char_clamp(int t) {
+  return (int8_t)clamp(t, -128, 127);
+}
+
+// should we apply any filter at all: 11111111 yes, 00000000 no
+static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit,
+                                 uint8_t p3, uint8_t p2,
+                                 uint8_t p1, uint8_t p0,
+                                 uint8_t q0, uint8_t q1,
+                                 uint8_t q2, uint8_t q3) {
+  int8_t mask = 0;
+  mask |= (abs(p3 - p2) > limit) * -1;
+  mask |= (abs(p2 - p1) > limit) * -1;
+  mask |= (abs(p1 - p0) > limit) * -1;
+  mask |= (abs(q1 - q0) > limit) * -1;
+  mask |= (abs(q2 - q1) > limit) * -1;
+  mask |= (abs(q3 - q2) > limit) * -1;
+  mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+  return ~mask;
+}
+
+// is there high edge variance internal edge: 11111111 yes, 00000000 no
+static INLINE int8_t hevmask(uint8_t thresh, uint8_t p1, uint8_t p0,
+                             uint8_t q0, uint8_t q1) {
+  int8_t hev = 0;
+  hev  |= (abs(p1 - p0) > thresh) * -1;
+  hev  |= (abs(q1 - q0) > thresh) * -1;
+  return hev;
+}
+
+static INLINE void filter(int8_t mask, uint8_t hev, uint8_t *op1,
+                          uint8_t *op0, uint8_t *oq0, uint8_t *oq1) {
+  int8_t filter1, filter2;
+
+  const int8_t ps1 = (int8_t) *op1 ^ 0x80;
+  const int8_t ps0 = (int8_t) *op0 ^ 0x80;
+  const int8_t qs0 = (int8_t) *oq0 ^ 0x80;
+  const int8_t qs1 = (int8_t) *oq1 ^ 0x80;
+
+  // add outer taps if we have high edge variance
+  int8_t filter = signed_char_clamp(ps1 - qs1) & hev;
+
+  // inner taps
+  filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask;
+
+  // save bottom 3 bits so that we round one side +4 and the other +3
+  // if it equals 4 we'll set to adjust by -1 to account for the fact
+  // we'd round 3 the other way
+  filter1 = signed_char_clamp(filter + 4) >> 3;
+  filter2 = signed_char_clamp(filter + 3) >> 3;
+
+  *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80;
+  *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80;
+
+  // outer tap adjustments
+  filter = ((filter1 + 1) >> 1) & ~hev;
+
+  *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80;
+  *op1 = signed_char_clamp(ps1 + filter) ^ 0x80;
+}
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int p /* pitch */,
+                                       const uint8_t *blimit,
+                                       const uint8_t *limit,
+                                       const uint8_t *thresh,
+                                       int count) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint8_t q0 = s[0 * p],  q1 = s[1 * p],  q2 = s[2 * p],  q3 = s[3 * p];
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(*thresh, p1, p0, q0, q1);
+    filter(mask, hev, s - 2 * p, s - 1 * p, s, s + 1 * p);
+    ++s;
+  }
+}
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch,
+                                     const uint8_t *blimit,
+                                     const uint8_t *limit,
+                                     const uint8_t *thresh,
+                                     int count) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint8_t q0 = s[0],  q1 = s[1],  q2 = s[2],  q3 = s[3];
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(*thresh, p1, p0, q0, q1);
+    filter(mask, hev, s - 2, s - 1, s, s + 1);
+    s += pitch;
+  }
+}
+
+static INLINE int8_t flatmask4(uint8_t thresh,
+                               uint8_t p3, uint8_t p2,
+                               uint8_t p1, uint8_t p0,
+                               uint8_t q0, uint8_t q1,
+                               uint8_t q2, uint8_t q3) {
+  int8_t flat = 0;
+  flat |= (abs(p1 - p0) > thresh) * -1;
+  flat |= (abs(q1 - q0) > thresh) * -1;
+  flat |= (abs(p0 - p2) > thresh) * -1;
+  flat |= (abs(q0 - q2) > thresh) * -1;
+  flat |= (abs(p3 - p0) > thresh) * -1;
+  flat |= (abs(q3 - q0) > thresh) * -1;
+  return ~flat;
+}
+static INLINE signed char flatmask5(uint8_t thresh,
+                                    uint8_t p4, uint8_t p3, uint8_t p2,
+                                    uint8_t p1, uint8_t p0,
+                                    uint8_t q0, uint8_t q1, uint8_t q2,
+                                    uint8_t q3, uint8_t q4) {
+  int8_t flat = 0;
+  flat |= (abs(p4 - p0) > thresh) * -1;
+  flat |= (abs(q4 - q0) > thresh) * -1;
+  flat = ~flat;
+  return flat & flatmask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3);
+}
+
+
+static INLINE void mbfilter(int8_t mask, uint8_t hev, uint8_t flat,
+                            uint8_t *op3, uint8_t *op2,
+                            uint8_t *op1, uint8_t *op0,
+                            uint8_t *oq0, uint8_t *oq1,
+                            uint8_t *oq2, uint8_t *oq3) {
+  // use a 7 tap filter [1, 1, 1, 2, 1, 1, 1] for flat line
+  if (flat && mask) {
+    const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
+    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;
+
+    *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + p2 + p2 + p1 + p0 + q0, 3);
+    *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + p1 + p1 + p0 + q0 + q1, 3);
+    *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 + p0 + q0 + q1 + q2, 3);
+    *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + q0 + q0 + q1 + q2 + q3, 3);
+    *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + q1 + q1 + q2 + q3 + q3, 3);
+    *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + q2 + q2 + q3 + q3 + q3, 3);
+  } else {
+    filter(mask, hev, op1,  op0, oq0, oq1);
+  }
+}
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int p,
+                                         const uint8_t *blimit,
+                                         const uint8_t *limit,
+                                         const uint8_t *thresh,
+                                         int count) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
+
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(*thresh, p1, p0, q0, q1);
+    const int8_t flat = flatmask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
+    mbfilter(mask, hev, flat,
+             s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p,
+             s,         s + 1 * p, s + 2 * p, s + 3 * p);
+    ++s;
+  }
+}
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch,
+                                       const uint8_t *blimit,
+                                       const uint8_t *limit,
+                                       const uint8_t *thresh,
+                                       int count) {
+  int i;
+
+  for (i = 0; i < 8 * count; ++i) {
+    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(thresh[0], p1, p0, q0, q1);
+    const int8_t flat = flatmask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
+    mbfilter(mask, hev, flat, s - 4, s - 3, s - 2, s - 1,
+                              s,     s + 1, s + 2, s + 3);
+    s += pitch;
+  }
+}
+
+static INLINE void wide_mbfilter(int8_t mask, uint8_t hev,
+                                 uint8_t flat, uint8_t flat2,
+                                 uint8_t *op7, uint8_t *op6, uint8_t *op5,
+                                 uint8_t *op4, uint8_t *op3, uint8_t *op2,
+                                 uint8_t *op1, uint8_t *op0, uint8_t *oq0,
+                                 uint8_t *oq1, uint8_t *oq2, uint8_t *oq3,
+                                 uint8_t *oq4, uint8_t *oq5, uint8_t *oq6,
+                                 uint8_t *oq7) {
+  // use a 15 tap filter [1,1,1,1,1,1,1,2,1,1,1,1,1,1,1] for flat line
+  if (flat2 && flat && mask) {
+    const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4,
+                  p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
+
+    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3,
+                  q4 = *oq4, q5 = *oq5, q6 = *oq6, q7 = *oq7;
+
+    *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0, 4);
+    *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1, 4);
+    *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2, 4);
+    *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3, 4);
+    *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4, 4);
+    *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5, 4);
+    *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 +
+                              q0 + q1 + q2 + q3 + q4 + q5 + q6, 4);
+    *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4);
+    *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4);
+    *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4);
+    *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
+    *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
+    *oq5 = ROUND_POWER_OF_TWO(p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
+    *oq6 = ROUND_POWER_OF_TWO(p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
+  } else {
+    mbfilter(mask, hev, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3);
+  }
+}
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int p,
+                                    const uint8_t *blimit,
+                                    const uint8_t *limit,
+                                    const uint8_t *thresh) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8; ++i) {
+    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(*thresh, p1, p0, q0, q1);
+    const int8_t flat = flatmask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t flat2 = flatmask5(1,
+                             s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0,
+                             q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p]);
+
+    wide_mbfilter(mask, hev, flat, flat2,
+                  s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p,
+                  s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p,
+                  s,         s + 1 * p, s + 2 * p, s + 3 * p,
+                  s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p);
+
+    ++s;
+  }
+}
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int p,
+                                  const uint8_t *blimit,
+                                  const uint8_t *limit,
+                                  const uint8_t *thresh) {
+  int i;
+
+  for (i = 0; i < 8; ++i) {
+    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint8_t q0 = s[0], q1 = s[1],  q2 = s[2], q3 = s[3];
+    const int8_t mask = filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t hev = hevmask(*thresh, p1, p0, q0, q1);
+    const int8_t flat = flatmask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t flat2 = flatmask5(1, s[-8], s[-7], s[-6], s[-5], p0,
+                                   q0, s[4], s[5], s[6], s[7]);
+
+    wide_mbfilter(mask, hev, flat, flat2,
+                  s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1,
+                  s,     s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7);
+    s += p;
+  }
+}
diff --git a/libvpx/vp9/common/vp9_maskingmv.c b/libvpx/vp9/common/vp9_maskingmv.c
new file mode 100644
index 0000000..326201b
--- /dev/null
+++ b/libvpx/vp9/common/vp9_maskingmv.c
@@ -0,0 +1,803 @@
+/*
+ ============================================================================
+ Name        : vp9_maskingmv.c
+ Author      : jimbankoski
+ Version     :
+ Copyright   : Your copyright notice
+ Description : Hello World in C, Ansi-style
+ ============================================================================
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+unsigned int vp9_sad16x16_sse3(
+  unsigned char *src_ptr,
+  int  src_stride,
+  unsigned char *ref_ptr,
+  int  ref_stride,
+  int  max_err);
+
+int vp8_growmaskmb_sse3(
+  unsigned char *om,
+  unsigned char *nm);
+
+void vp8_makemask_sse3(
+  unsigned char *y,
+  unsigned char *u,
+  unsigned char *v,
+  unsigned char *ym,
+  int yp,
+  int uvp,
+  int ys,
+  int us,
+  int vs,
+  int yt,
+  int ut,
+  int vt);
+
+unsigned int vp9_sad16x16_unmasked_wmt(
+  unsigned char *src_ptr,
+  int  src_stride,
+  unsigned char *ref_ptr,
+  int  ref_stride,
+  unsigned char *mask);
+
+unsigned int vp9_sad16x16_masked_wmt(
+  unsigned char *src_ptr,
+  int  src_stride,
+  unsigned char *ref_ptr,
+  int  ref_stride,
+  unsigned char *mask);
+
+unsigned int vp8_masked_predictor_wmt(
+  unsigned char *masked,
+  unsigned char *unmasked,
+  int  src_stride,
+  unsigned char *dst_ptr,
+  int  dst_stride,
+  unsigned char *mask);
+unsigned int vp8_masked_predictor_uv_wmt(
+  unsigned char *masked,
+  unsigned char *unmasked,
+  int  src_stride,
+  unsigned char *dst_ptr,
+  int  dst_stride,
+  unsigned char *mask);
+unsigned int vp8_uv_from_y_mask(
+  unsigned char *ymask,
+  unsigned char *uvmask);
+int yp = 16;
+unsigned char sxy[] = {
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 80, 120, 120, 90, 90, 90, 90, 90, 80, 120, 120, 90, 90, 90, 90, 90
+};
+
+unsigned char sts[] = {
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+};
+unsigned char str[] = {
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
+};
+
+unsigned char y[] = {
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40,
+  60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40,
+  40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40, 40,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40,
+  40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40, 40,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40,
+  40, 40, 40, 60, 60, 60, 60, 40, 40, 40, 40, 60, 60, 60, 60, 40
+};
+int uvp = 8;
+unsigned char u[] = {
+  90, 80, 70, 70, 90, 90, 90, 17,
+  90, 80, 70, 70, 90, 90, 90, 17,
+  84, 70, 70, 90, 90, 90, 17, 17,
+  84, 70, 70, 90, 90, 90, 17, 17,
+  80, 70, 70, 90, 90, 90, 17, 17,
+  90, 80, 70, 70, 90, 90, 90, 17,
+  90, 80, 70, 70, 90, 90, 90, 17,
+  90, 80, 70, 70, 90, 90, 90, 17
+};
+
+unsigned char v[] = {
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80,
+  80, 80, 80, 80, 80, 80, 80, 80
+};
+
+unsigned char ym[256];
+unsigned char uvm[64];
+typedef struct {
+  unsigned char y;
+  unsigned char yt;
+  unsigned char u;
+  unsigned char ut;
+  unsigned char v;
+  unsigned char vt;
+  unsigned char use;
+} COLOR_SEG_ELEMENT;
+
+/*
+COLOR_SEG_ELEMENT segmentation[]=
+{
+    { 60,4,80,17,80,10, 1},
+    { 40,4,15,10,80,10, 1},
+};
+*/
+
+COLOR_SEG_ELEMENT segmentation[] = {
+  { 79, 44, 92, 44, 237, 60, 1},
+};
+
+unsigned char pixel_mask(unsigned char y, unsigned char u, unsigned char v,
+                         COLOR_SEG_ELEMENT sgm[],
+                         int c) {
+  COLOR_SEG_ELEMENT *s = sgm;
+  unsigned char m = 0;
+  int i;
+  for (i = 0; i < c; i++, s++)
+    m |= (abs(y - s->y) < s->yt &&
+          abs(u - s->u) < s->ut &&
+          abs(v - s->v) < s->vt ? 255 : 0);
+
+  return m;
+}
+int neighbors[256][8];
+int makeneighbors(void) {
+  int i, j;
+  for (i = 0; i < 256; i++) {
+    int r = (i >> 4), c = (i & 15);
+    int ni = 0;
+    for (j = 0; j < 8; j++)
+      neighbors[i][j] = i;
+    for (j = 0; j < 256; j++) {
+      int nr = (j >> 4), nc = (j & 15);
+      if (abs(nr - r) < 2 && abs(nc - c) < 2)
+        neighbors[i][ni++] = j;
+    }
+  }
+  return 0;
+}
+void grow_ymask(unsigned char *ym) {
+  unsigned char nym[256];
+  int i, j;
+
+  for (i = 0; i < 256; i++) {
+    nym[i] = ym[i];
+    for (j = 0; j < 8; j++) {
+      nym[i] |= ym[neighbors[i][j]];
+    }
+  }
+  for (i = 0; i < 256; i++)
+    ym[i] = nym[i];
+}
+
+void make_mb_mask(unsigned char *y, unsigned char *u, unsigned char *v,
+                  unsigned char *ym, unsigned char *uvm,
+                  int yp, int uvp,
+                  COLOR_SEG_ELEMENT sgm[],
+                  int count) {
+  int r, c;
+  unsigned char *oym = ym;
+
+  memset(ym, 20, 256);
+  for (r = 0; r < 8; r++, uvm += 8, u += uvp, v += uvp, y += (yp + yp), ym += 32)
+    for (c = 0; c < 8; c++) {
+      int y1 = y[c << 1];
+      int u1 = u[c];
+      int v1 = v[c];
+      int m = pixel_mask(y1, u1, v1, sgm, count);
+      uvm[c] = m;
+      ym[c << 1] = uvm[c]; // = pixel_mask(y[c<<1],u[c],v[c],sgm,count);
+      ym[(c << 1) + 1] = pixel_mask(y[1 + (c << 1)], u[c], v[c], sgm, count);
+      ym[(c << 1) + 16] = pixel_mask(y[yp + (c << 1)], u[c], v[c], sgm, count);
+      ym[(c << 1) + 17] = pixel_mask(y[1 + yp + (c << 1)], u[c], v[c], sgm, count);
+    }
+  grow_ymask(oym);
+}
+
+int masked_sad(unsigned char *src, int p, unsigned char *dst, int dp,
+               unsigned char *ym) {
+  int i, j;
+  unsigned sad = 0;
+  for (i = 0; i < 16; i++, src += p, dst += dp, ym += 16)
+    for (j = 0; j < 16; j++)
+      if (ym[j])
+        sad += abs(src[j] - dst[j]);
+
+  return sad;
+}
+
+int compare_masks(unsigned char *sym, unsigned char *ym) {
+  int i, j;
+  unsigned sad = 0;
+  for (i = 0; i < 16; i++, sym += 16, ym += 16)
+    for (j = 0; j < 16; j++)
+      sad += (sym[j] != ym[j] ? 1 : 0);
+
+  return sad;
+}
+
+int unmasked_sad(unsigned char *src, int p, unsigned char *dst, int dp,
+                 unsigned char *ym) {
+  int i, j;
+  unsigned sad = 0;
+  for (i = 0; i < 16; i++, src += p, dst += dp, ym += 16)
+    for (j = 0; j < 16; j++)
+      if (!ym[j])
+        sad += abs(src[j] - dst[j]);
+
+  return sad;
+}
+
+int masked_motion_search(unsigned char *y, unsigned char *u, unsigned char *v,
+                         int yp, int uvp,
+                         unsigned char *dy, unsigned char *du, unsigned char *dv,
+                         int dyp, int duvp,
+                         COLOR_SEG_ELEMENT sgm[],
+                         int count,
+                         int *mi,
+                         int *mj,
+                         int *ui,
+                         int *uj,
+                         int *wm) {
+  int i, j;
+
+  unsigned char ym[256];
+  unsigned char uvm[64];
+  unsigned char dym[256];
+  unsigned char duvm[64];
+  unsigned int e = 0;
+  int beste = 256;
+  int bmi = -32, bmj = -32;
+  int bui = -32, buj = -32;
+  int beste1 = 256;
+  int bmi1 = -32, bmj1 = -32;
+  int bui1 = -32, buj1 = -32;
+  int obeste;
+
+  // first try finding best mask and then unmasked
+  beste = 0xffffffff;
+
+  // find best unmasked mv
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    unsigned char *duz = i / 2 * duvp + du;
+    unsigned char *dvz = i / 2 * duvp + dv;
+    for (j = -32; j < 32; j++) {
+      // 0,0  masked destination
+      make_mb_mask(dyz + j, duz + j / 2, dvz + j / 2, dym, duvm, dyp, duvp, sgm, count);
+
+      e = unmasked_sad(y, yp, dyz + j, dyp, dym);
+
+      if (e < beste) {
+        bui = i;
+        buj = j;
+        beste = e;
+      }
+    }
+  }
+  // bui=0;buj=0;
+  // best mv masked destination
+  make_mb_mask(dy + bui * dyp + buj, du + bui / 2 * duvp + buj / 2, dv + bui / 2 * duvp + buj / 2,
+               dym, duvm, dyp, duvp, sgm, count);
+
+  obeste = beste;
+  beste = 0xffffffff;
+
+  // find best masked
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    for (j = -32; j < 32; j++) {
+      e = masked_sad(y, yp, dyz + j, dyp, dym);
+
+      if (e < beste) {
+        bmi = i;
+        bmj = j;
+        beste = e;
+      }
+    }
+  }
+  beste1 = beste + obeste;
+  bmi1 = bmi;
+  bmj1 = bmj;
+  bui1 = bui;
+  buj1 = buj;
+
+  beste = 0xffffffff;
+  // source mask
+  make_mb_mask(y, u, v, ym, uvm, yp, uvp, sgm, count);
+
+  // find best mask
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    unsigned char *duz = i / 2 * duvp + du;
+    unsigned char *dvz = i / 2 * duvp + dv;
+    for (j = -32; j < 32; j++) {
+      // 0,0  masked destination
+      make_mb_mask(dyz + j, duz + j / 2, dvz + j / 2, dym, duvm, dyp, duvp, sgm, count);
+
+      e = compare_masks(ym, dym);
+
+      if (e < beste) {
+        bmi = i;
+        bmj = j;
+        beste = e;
+      }
+    }
+  }
+
+
+  // best mv masked destination
+  make_mb_mask(dy + bmi * dyp + bmj, du + bmi / 2 * duvp + bmj / 2, dv + bmi / 2 * duvp + bmj / 2,
+               dym, duvm, dyp, duvp, sgm, count);
+
+  obeste = masked_sad(y, yp, dy + bmi * dyp + bmj, dyp, dym);
+
+  beste = 0xffffffff;
+
+  // find best unmasked mv
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    for (j = -32; j < 32; j++) {
+      e = unmasked_sad(y, yp, dyz + j, dyp, dym);
+
+      if (e < beste) {
+        bui = i;
+        buj = j;
+        beste = e;
+      }
+    }
+  }
+  beste += obeste;
+
+
+  if (beste < beste1) {
+    *mi = bmi;
+    *mj = bmj;
+    *ui = bui;
+    *uj = buj;
+    *wm = 1;
+  } else {
+    *mi = bmi1;
+    *mj = bmj1;
+    *ui = bui1;
+    *uj = buj1;
+    *wm = 0;
+
+  }
+  return 0;
+}
+
+int predict(unsigned char *src, int p, unsigned char *dst, int dp,
+            unsigned char *ym, unsigned char *prd) {
+  int i, j;
+  for (i = 0; i < 16; i++, src += p, dst += dp, ym += 16, prd += 16)
+    for (j = 0; j < 16; j++)
+      prd[j] = (ym[j] ? src[j] : dst[j]);
+  return 0;
+}
+
+int fast_masked_motion_search(unsigned char *y, unsigned char *u, unsigned char *v,
+                              int yp, int uvp,
+                              unsigned char *dy, unsigned char *du, unsigned char *dv,
+                              int dyp, int duvp,
+                              COLOR_SEG_ELEMENT sgm[],
+                              int count,
+                              int *mi,
+                              int *mj,
+                              int *ui,
+                              int *uj,
+                              int *wm) {
+  int i, j;
+
+  unsigned char ym[256];
+  unsigned char ym2[256];
+  unsigned char uvm[64];
+  unsigned char dym2[256];
+  unsigned char dym[256];
+  unsigned char duvm[64];
+  unsigned int e = 0;
+  int beste = 256;
+  int bmi = -32, bmj = -32;
+  int bui = -32, buj = -32;
+  int beste1 = 256;
+  int bmi1 = -32, bmj1 = -32;
+  int bui1 = -32, buj1 = -32;
+  int obeste;
+
+  // first try finding best mask and then unmasked
+  beste = 0xffffffff;
+
+#if 0
+  for (i = 0; i < 16; i++) {
+    unsigned char *dy = i * yp + y;
+    for (j = 0; j < 16; j++)
+      printf("%2x", dy[j]);
+    printf("\n");
+  }
+  printf("\n");
+
+  for (i = -32; i < 48; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    for (j = -32; j < 48; j++)
+      printf("%2x", dyz[j]);
+    printf("\n");
+  }
+#endif
+
+  // find best unmasked mv
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    unsigned char *duz = i / 2 * duvp + du;
+    unsigned char *dvz = i / 2 * duvp + dv;
+    for (j = -32; j < 32; j++) {
+      // 0,0  masked destination
+      vp8_makemask_sse3(dyz + j, duz + j / 2, dvz + j / 2, dym, dyp, duvp,
+                        sgm[0].y, sgm[0].u, sgm[0].v,
+                        sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+      vp8_growmaskmb_sse3(dym, dym2);
+
+      e = vp9_sad16x16_unmasked_wmt(y, yp, dyz + j, dyp, dym2);
+
+      if (e < beste) {
+        bui = i;
+        buj = j;
+        beste = e;
+      }
+    }
+  }
+  // bui=0;buj=0;
+  // best mv masked destination
+
+  vp8_makemask_sse3(dy + bui * dyp + buj, du + bui / 2 * duvp + buj / 2, dv + bui / 2 * duvp + buj / 2,
+                    dym, dyp, duvp,
+                    sgm[0].y, sgm[0].u, sgm[0].v,
+                    sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+  vp8_growmaskmb_sse3(dym, dym2);
+
+  obeste = beste;
+  beste = 0xffffffff;
+
+  // find best masked
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    for (j = -32; j < 32; j++) {
+      e = vp9_sad16x16_masked_wmt(y, yp, dyz + j, dyp, dym2);
+      if (e < beste) {
+        bmi = i;
+        bmj = j;
+        beste = e;
+      }
+    }
+  }
+  beste1 = beste + obeste;
+  bmi1 = bmi;
+  bmj1 = bmj;
+  bui1 = bui;
+  buj1 = buj;
+
+  // source mask
+  vp8_makemask_sse3(y, u, v,
+                    ym, yp, uvp,
+                    sgm[0].y, sgm[0].u, sgm[0].v,
+                    sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+  vp8_growmaskmb_sse3(ym, ym2);
+
+  // find best mask
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    unsigned char *duz = i / 2 * duvp + du;
+    unsigned char *dvz = i / 2 * duvp + dv;
+    for (j = -32; j < 32; j++) {
+      // 0,0  masked destination
+      vp8_makemask_sse3(dyz + j, duz + j / 2, dvz + j / 2, dym, dyp, duvp,
+                        sgm[0].y, sgm[0].u, sgm[0].v,
+                        sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+      vp8_growmaskmb_sse3(dym, dym2);
+
+      e = compare_masks(ym2, dym2);
+
+      if (e < beste) {
+        bmi = i;
+        bmj = j;
+        beste = e;
+      }
+    }
+  }
+
+  vp8_makemask_sse3(dy + bmi * dyp + bmj, du + bmi / 2 * duvp + bmj / 2, dv + bmi / 2 * duvp + bmj / 2,
+                    dym, dyp, duvp,
+                    sgm[0].y, sgm[0].u, sgm[0].v,
+                    sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+  vp8_growmaskmb_sse3(dym, dym2);
+
+  obeste = vp9_sad16x16_masked_wmt(y, yp, dy + bmi * dyp + bmj, dyp, dym2);
+
+  beste = 0xffffffff;
+
+  // find best unmasked mv
+  for (i = -32; i < 32; i++) {
+    unsigned char *dyz = i * dyp + dy;
+    for (j = -32; j < 32; j++) {
+      e = vp9_sad16x16_unmasked_wmt(y, yp, dyz + j, dyp, dym2);
+
+      if (e < beste) {
+        bui = i;
+        buj = j;
+        beste = e;
+      }
+    }
+  }
+  beste += obeste;
+
+  if (beste < beste1) {
+    *mi = bmi;
+    *mj = bmj;
+    *ui = bui;
+    *uj = buj;
+    *wm = 1;
+  } else {
+    *mi = bmi1;
+    *mj = bmj1;
+    *ui = bui1;
+    *uj = buj1;
+    *wm = 0;
+    beste = beste1;
+
+  }
+  return beste;
+}
+
+int predict_all(unsigned char *ym, unsigned char *um, unsigned char *vm,
+                int ymp, int uvmp,
+                unsigned char *yp, unsigned char *up, unsigned char *vp,
+                int ypp, int uvpp,
+                COLOR_SEG_ELEMENT sgm[],
+                int count,
+                int mi,
+                int mj,
+                int ui,
+                int uj,
+                int wm) {
+  int i, j;
+  unsigned char dym[256];
+  unsigned char dym2[256];
+  unsigned char duvm[64];
+  unsigned char *yu = ym, *uu = um, *vu = vm;
+
+  unsigned char *dym3 = dym2;
+
+  ym += mi * ymp + mj;
+  um += mi / 2 * uvmp + mj / 2;
+  vm += mi / 2 * uvmp + mj / 2;
+
+  yu += ui * ymp + uj;
+  uu += ui / 2 * uvmp + uj / 2;
+  vu += ui / 2 * uvmp + uj / 2;
+
+  // best mv masked destination
+  if (wm)
+    vp8_makemask_sse3(ym, um, vm, dym, ymp, uvmp,
+                      sgm[0].y, sgm[0].u, sgm[0].v,
+                      sgm[0].yt, sgm[0].ut, sgm[0].vt);
+  else
+    vp8_makemask_sse3(yu, uu, vu, dym, ymp, uvmp,
+                      sgm[0].y, sgm[0].u, sgm[0].v,
+                      sgm[0].yt, sgm[0].ut, sgm[0].vt);
+
+  vp8_growmaskmb_sse3(dym, dym2);
+  vp8_masked_predictor_wmt(ym, yu, ymp, yp, ypp, dym3);
+  vp8_uv_from_y_mask(dym3, duvm);
+  vp8_masked_predictor_uv_wmt(um, uu, uvmp, up, uvpp, duvm);
+  vp8_masked_predictor_uv_wmt(vm, vu, uvmp, vp, uvpp, duvm);
+
+  return 0;
+}
+
+unsigned char f0p[1280 * 720 * 3 / 2];
+unsigned char f1p[1280 * 720 * 3 / 2];
+unsigned char prd[1280 * 720 * 3 / 2];
+unsigned char msk[1280 * 720 * 3 / 2];
+
+
+int mainz(int argc, char *argv[]) {
+
+  FILE *f = fopen(argv[1], "rb");
+  FILE *g = fopen(argv[2], "wb");
+  int w = atoi(argv[3]), h = atoi(argv[4]);
+  int y_stride = w, uv_stride = w / 2;
+  int r, c;
+  unsigned char *f0 = f0p, *f1 = f1p, *t;
+  unsigned char ym[256], uvm[64];
+  unsigned char ym2[256], uvm2[64];
+  unsigned char ym3[256], uvm3[64];
+  int a, b;
+
+  COLOR_SEG_ELEMENT last = { 20, 20, 20, 20, 230, 20, 1}, best;
+#if 0
+  makeneighbors();
+  COLOR_SEG_ELEMENT segmentation[] = {
+    { 60, 4, 80, 17, 80, 10, 1},
+    { 40, 4, 15, 10, 80, 10, 1},
+  };
+  make_mb_mask(y, u, v, ym2, uvm2, 16, 8, segmentation, 1);
+
+  vp8_makemask_sse3(y, u, v, ym, (int) 16, (int) 8,
+                    (int) segmentation[0].y, (int) segmentation[0].u, (int) segmentation[0].v,
+                    segmentation[0].yt, segmentation[0].ut, segmentation[0].vt);
+
+  vp8_growmaskmb_sse3(ym, ym3);
+
+  a = vp9_sad16x16_masked_wmt(str, 16, sts, 16, ym3);
+  b = vp9_sad16x16_unmasked_wmt(str, 16, sts, 16, ym3);
+
+  vp8_masked_predictor_wmt(str, sts, 16, ym, 16, ym3);
+
+  vp8_uv_from_y_mask(ym3, uvm3);
+
+  return 4;
+#endif
+  makeneighbors();
+
+
+  memset(prd, 128, w * h * 3 / 2);
+
+  fread(f0, w * h * 3 / 2, 1, f);
+
+  while (!feof(f)) {
+    unsigned char *ys = f1, *yd = f0, *yp = prd;
+    unsigned char *us = f1 + w * h, *ud = f0 + w * h, *up = prd + w * h;
+    unsigned char *vs = f1 + w * h * 5 / 4, *vd = f0 + w * h * 5 / 4, *vp = prd + w * h * 5 / 4;
+    fread(f1, w * h * 3 / 2, 1, f);
+
+    ys += 32 * y_stride;
+    yd += 32 * y_stride;
+    yp += 32 * y_stride;
+    us += 16 * uv_stride;
+    ud += 16 * uv_stride;
+    up += 16 * uv_stride;
+    vs += 16 * uv_stride;
+    vd += 16 * uv_stride;
+    vp += 16 * uv_stride;
+    for (r = 32; r < h - 32; r += 16,
+         ys += 16 * w, yd += 16 * w, yp += 16 * w,
+         us += 8 * uv_stride, ud += 8 * uv_stride, up += 8 * uv_stride,
+         vs += 8 * uv_stride, vd += 8 * uv_stride, vp += 8 * uv_stride) {
+      for (c = 32; c < w - 32; c += 16) {
+        int mi, mj, ui, uj, wm;
+        int bmi, bmj, bui, buj, bwm;
+        unsigned char ym[256];
+
+        if (vp9_sad16x16_sse3(ys + c, y_stride, yd + c, y_stride, 0xffff) == 0)
+          bmi = bmj = bui = buj = bwm = 0;
+        else {
+          COLOR_SEG_ELEMENT cs[5];
+          int j;
+          unsigned int beste = 0xfffffff;
+          unsigned int bestj = 0;
+
+          // try color from last mb segmentation
+          cs[0] = last;
+
+          // try color segs from 4 pixels in mb recon as segmentation
+          cs[1].y = yd[c + y_stride + 1];
+          cs[1].u = ud[c / 2 + uv_stride];
+          cs[1].v = vd[c / 2 + uv_stride];
+          cs[1].yt = cs[1].ut = cs[1].vt = 20;
+          cs[2].y = yd[c + w + 14];
+          cs[2].u = ud[c / 2 + uv_stride + 7];
+          cs[2].v = vd[c / 2 + uv_stride + 7];
+          cs[2].yt = cs[2].ut = cs[2].vt = 20;
+          cs[3].y = yd[c + w * 14 + 1];
+          cs[3].u = ud[c / 2 + uv_stride * 7];
+          cs[3].v = vd[c / 2 + uv_stride * 7];
+          cs[3].yt = cs[3].ut = cs[3].vt = 20;
+          cs[4].y = yd[c + w * 14 + 14];
+          cs[4].u = ud[c / 2 + uv_stride * 7 + 7];
+          cs[4].v = vd[c / 2 + uv_stride * 7 + 7];
+          cs[4].yt = cs[4].ut = cs[4].vt = 20;
+
+          for (j = 0; j < 5; j++) {
+            int e;
+
+            e = fast_masked_motion_search(
+                  ys + c, us + c / 2, vs + c / 2, y_stride, uv_stride,
+                  yd + c, ud + c / 2, vd + c / 2, y_stride, uv_stride,
+                  &cs[j], 1, &mi, &mj, &ui, &uj, &wm);
+
+            if (e < beste) {
+              bmi = mi;
+              bmj = mj;
+              bui = ui;
+              buj = uj, bwm = wm;
+              bestj = j;
+              beste = e;
+            }
+          }
+          best = cs[bestj];
+          // best = segmentation[0];
+          last = best;
+        }
+        predict_all(yd + c, ud + c / 2, vd + c / 2, w, uv_stride,
+                    yp + c, up + c / 2, vp + c / 2, w, uv_stride,
+                    &best, 1, bmi, bmj, bui, buj, bwm);
+
+      }
+    }
+    fwrite(prd, w * h * 3 / 2, 1, g);
+    t = f0;
+    f0 = f1;
+    f1 = t;
+
+  }
+  fclose(f);
+  fclose(g);
+  return 0;
+}
diff --git a/libvpx/vp9/common/vp9_mbpitch.c b/libvpx/vp9/common/vp9_mbpitch.c
new file mode 100644
index 0000000..3cf37ff
--- /dev/null
+++ b/libvpx/vp9/common/vp9_mbpitch.c
@@ -0,0 +1,28 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_blockd.h"
+
+void vp9_setup_block_dptrs(MACROBLOCKD *mb,
+                           int subsampling_x, int subsampling_y) {
+  int i;
+
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    mb->plane[i].plane_type = i ? PLANE_TYPE_UV : PLANE_TYPE_Y_WITH_DC;
+    mb->plane[i].subsampling_x = i ? subsampling_x : 0;
+    mb->plane[i].subsampling_y = i ? subsampling_y : 0;
+  }
+#if CONFIG_ALPHA
+  // TODO(jkoleszar): Using the Y w/h for now
+  mb->plane[3].subsampling_x = 0;
+  mb->plane[3].subsampling_y = 0;
+#endif
+}
diff --git a/libvpx/vp9/common/vp9_modecont.c b/libvpx/vp9/common/vp9_modecont.c
new file mode 100644
index 0000000..5d92cfa
--- /dev/null
+++ b/libvpx/vp9/common/vp9_modecont.c
@@ -0,0 +1,23 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_modecont.h"
+
+const vp9_prob vp9_default_inter_mode_probs[INTER_MODE_CONTEXTS]
+                                           [VP9_INTER_MODES - 1] = {
+  {2,       173,   34},  // 0 = both zero mv
+  {7,       145,   85},  // 1 = one zero mv + one a predicted mv
+  {7,       166,   63},  // 2 = two predicted mvs
+  {7,       94,    66},  // 3 = one predicted/zero and one new mv
+  {8,       64,    46},  // 4 = two new mvs
+  {17,      81,    31},  // 5 = one intra neighbour + x
+  {25,      29,    30},  // 6 = two intra neighbours
+};
diff --git a/libvpx/vp9/common/vp9_modecont.h b/libvpx/vp9/common/vp9_modecont.h
new file mode 100644
index 0000000..3ec6079
--- /dev/null
+++ b/libvpx/vp9/common/vp9_modecont.h
@@ -0,0 +1,19 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_MODECONT_H_
+#define VP9_COMMON_VP9_MODECONT_H_
+
+#include "vp9/common/vp9_entropy.h"
+
+extern const vp9_prob vp9_default_inter_mode_probs[INTER_MODE_CONTEXTS]
+                                                  [VP9_INTER_MODES - 1];
+
+#endif  // VP9_COMMON_VP9_MODECONT_H_
diff --git a/libvpx/vp9/common/vp9_modecontext.c b/libvpx/vp9/common/vp9_modecontext.c
new file mode 100644
index 0000000..a79ab2a
--- /dev/null
+++ b/libvpx/vp9/common/vp9_modecontext.c
@@ -0,0 +1,128 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_entropymode.h"
+
+const vp9_prob vp9_kf_default_bmode_probs[VP9_INTRA_MODES]
+                                         [VP9_INTRA_MODES]
+                                         [VP9_INTRA_MODES - 1] = {
+  { /* above = dc */
+    { 137,  30,  42, 148, 151, 207,  70,  52,  91 } /* left = dc */,
+    {  92,  45, 102, 136, 116, 180,  74,  90, 100 } /* left = v */,
+    {  73,  32,  19, 187, 222, 215,  46,  34, 100 } /* left = h */,
+    {  91,  30,  32, 116, 121, 186,  93,  86,  94 } /* left = d45 */,
+    {  72,  35,  36, 149,  68, 206,  68,  63, 105 } /* left = d135 */,
+    {  73,  31,  28, 138,  57, 124,  55, 122, 151 } /* left = d117 */,
+    {  67,  23,  21, 140, 126, 197,  40,  37, 171 } /* left = d153 */,
+    {  86,  27,  28, 128, 154, 212,  45,  43,  53 } /* left = d27 */,
+    {  74,  32,  27, 107,  86, 160,  63, 134, 102 } /* left = d63 */,
+    {  59,  67,  44, 140, 161, 202,  78,  67, 119 } /* left = tm */
+  }, { /* above = v */
+    {  63,  36, 126, 146, 123, 158,  60,  90,  96 } /* left = dc */,
+    {  43,  46, 168, 134, 107, 128,  69, 142,  92 } /* left = v */,
+    {  44,  29,  68, 159, 201, 177,  50,  57,  77 } /* left = h */,
+    {  58,  38,  76, 114,  97, 172,  78, 133,  92 } /* left = d45 */,
+    {  46,  41,  76, 140,  63, 184,  69, 112,  57 } /* left = d135 */,
+    {  38,  32,  85, 140,  46, 112,  54, 151, 133 } /* left = d117 */,
+    {  39,  27,  61, 131, 110, 175,  44,  75, 136 } /* left = d153 */,
+    {  52,  30,  74, 113, 130, 175,  51,  64,  58 } /* left = d27 */,
+    {  47,  35,  80, 100,  74, 143,  64, 163,  74 } /* left = d63 */,
+    {  36,  61, 116, 114, 128, 162,  80, 125,  82 } /* left = tm */
+  }, { /* above = h */
+    {  82,  26,  26, 171, 208, 204,  44,  32, 105 } /* left = dc */,
+    {  55,  44,  68, 166, 179, 192,  57,  57, 108 } /* left = v */,
+    {  42,  26,  11, 199, 241, 228,  23,  15,  85 } /* left = h */,
+    {  68,  42,  19, 131, 160, 199,  55,  52,  83 } /* left = d45 */,
+    {  58,  50,  25, 139, 115, 232,  39,  52, 118 } /* left = d135 */,
+    {  50,  35,  33, 153, 104, 162,  64,  59, 131 } /* left = d117 */,
+    {  44,  24,  16, 150, 177, 202,  33,  19, 156 } /* left = d153 */,
+    {  55,  27,  12, 153, 203, 218,  26,  27,  49 } /* left = d27 */,
+    {  53,  49,  21, 110, 116, 168,  59,  80,  76 } /* left = d63 */,
+    {  38,  72,  19, 168, 203, 212,  50,  50, 107 } /* left = tm */
+  }, { /* above = d45 */
+    { 103,  26,  36, 129, 132, 201,  83,  80,  93 } /* left = dc */,
+    {  59,  38,  83, 112, 103, 162,  98, 136,  90 } /* left = v */,
+    {  62,  30,  23, 158, 200, 207,  59,  57,  50 } /* left = h */,
+    {  67,  30,  29,  84,  86, 191, 102,  91,  59 } /* left = d45 */,
+    {  60,  32,  33, 112,  71, 220,  64,  89, 104 } /* left = d135 */,
+    {  53,  26,  34, 130,  56, 149,  84, 120, 103 } /* left = d117 */,
+    {  53,  21,  23, 133, 109, 210,  56,  77, 172 } /* left = d153 */,
+    {  77,  19,  29, 112, 142, 228,  55,  66,  36 } /* left = d27 */,
+    {  61,  29,  29,  93,  97, 165,  83, 175, 162 } /* left = d63 */,
+    {  47,  47,  43, 114, 137, 181, 100,  99,  95 } /* left = tm */
+  }, { /* above = d135 */
+    {  69,  23,  29, 128,  83, 199,  46,  44, 101 } /* left = dc */,
+    {  53,  40,  55, 139,  69, 183,  61,  80, 110 } /* left = v */,
+    {  40,  29,  19, 161, 180, 207,  43,  24,  91 } /* left = h */,
+    {  60,  34,  19, 105,  61, 198,  53,  64,  89 } /* left = d45 */,
+    {  52,  31,  22, 158,  40, 209,  58,  62,  89 } /* left = d135 */,
+    {  44,  31,  29, 147,  46, 158,  56, 102, 198 } /* left = d117 */,
+    {  35,  19,  12, 135,  87, 209,  41,  45, 167 } /* left = d153 */,
+    {  55,  25,  21, 118,  95, 215,  38,  39,  66 } /* left = d27 */,
+    {  51,  38,  25, 113,  58, 164,  70,  93,  97 } /* left = d63 */,
+    {  47,  54,  34, 146, 108, 203,  72, 103, 151 } /* left = tm */
+  }, { /* above = d117 */
+    {  64,  19,  37, 156,  66, 138,  49,  95, 133 } /* left = dc */,
+    {  46,  27,  80, 150,  55, 124,  55, 121, 135 } /* left = v */,
+    {  36,  23,  27, 165, 149, 166,  54,  64, 118 } /* left = h */,
+    {  53,  21,  36, 131,  63, 163,  60, 109,  81 } /* left = d45 */,
+    {  40,  26,  35, 154,  40, 185,  51,  97, 123 } /* left = d135 */,
+    {  35,  19,  34, 179,  19,  97,  48, 129, 124 } /* left = d117 */,
+    {  36,  20,  26, 136,  62, 164,  33,  77, 154 } /* left = d153 */,
+    {  45,  18,  32, 130,  90, 157,  40,  79,  91 } /* left = d27 */,
+    {  45,  26,  28, 129,  45, 129,  49, 147, 123 } /* left = d63 */,
+    {  38,  44,  51, 136,  74, 162,  57,  97, 121 } /* left = tm */
+  }, { /* above = d153 */
+    {  75,  17,  22, 136, 138, 185,  32,  34, 166 } /* left = dc */,
+    {  56,  39,  58, 133, 117, 173,  48,  53, 187 } /* left = v */,
+    {  35,  21,  12, 161, 212, 207,  20,  23, 145 } /* left = h */,
+    {  56,  29,  19, 117, 109, 181,  55,  68, 112 } /* left = d45 */,
+    {  47,  29,  17, 153,  64, 220,  59,  51, 114 } /* left = d135 */,
+    {  46,  16,  24, 136,  76, 147,  41,  64, 172 } /* left = d117 */,
+    {  34,  17,  11, 108, 152, 187,  13,  15, 209 } /* left = d153 */,
+    {  51,  24,  14, 115, 133, 209,  32,  26, 104 } /* left = d27 */,
+    {  55,  30,  18, 122,  79, 179,  44,  88, 116 } /* left = d63 */,
+    {  37,  49,  25, 129, 168, 164,  41,  54, 148 } /* left = tm */
+  }, { /* above = d27 */
+    {  82,  22,  32, 127, 143, 213,  39,  41,  70 } /* left = dc */,
+    {  62,  44,  61, 123, 105, 189,  48,  57,  64 } /* left = v */,
+    {  47,  25,  17, 175, 222, 220,  24,  30,  86 } /* left = h */,
+    {  68,  36,  17, 106, 102, 206,  59,  74,  74 } /* left = d45 */,
+    {  57,  39,  23, 151,  68, 216,  55,  63,  58 } /* left = d135 */,
+    {  49,  30,  35, 141,  70, 168,  82,  40, 115 } /* left = d117 */,
+    {  51,  25,  15, 136, 129, 202,  38,  35, 139 } /* left = d153 */,
+    {  68,  26,  16, 111, 141, 215,  29,  28,  28 } /* left = d27 */,
+    {  59,  39,  19, 114,  75, 180,  77, 104,  42 } /* left = d63 */,
+    {  40,  61,  26, 126, 152, 206,  61,  59,  93 } /* left = tm */
+  }, { /* above = d63 */
+    {  78,  23,  39, 111, 117, 170,  74, 124,  94 } /* left = dc */,
+    {  48,  34,  86, 101,  92, 146,  78, 179, 134 } /* left = v */,
+    {  47,  22,  24, 138, 187, 178,  68,  69,  59 } /* left = h */,
+    {  56,  25,  33, 105, 112, 187,  95, 177, 129 } /* left = d45 */,
+    {  48,  31,  27, 114,  63, 183,  82, 116,  56 } /* left = d135 */,
+    {  43,  28,  37, 121,  63, 123,  61, 192, 169 } /* left = d117 */,
+    {  42,  17,  24, 109,  97, 177,  56,  76, 122 } /* left = d153 */,
+    {  58,  18,  28, 105, 139, 182,  70,  92,  63 } /* left = d27 */,
+    {  46,  23,  32,  74,  86, 150,  67, 183,  88 } /* left = d63 */,
+    {  36,  38,  48,  92, 122, 165,  88, 137,  91 } /* left = tm */
+  }, { /* above = tm */
+    {  65,  70,  60, 155, 159, 199,  61,  60,  81 } /* left = dc */,
+    {  44,  78, 115, 132, 119, 173,  71, 112,  93 } /* left = v */,
+    {  39,  38,  21, 184, 227, 206,  42,  32,  64 } /* left = h */,
+    {  58,  47,  36, 124, 137, 193,  80,  82,  78 } /* left = d45 */,
+    {  49,  50,  35, 144,  95, 205,  63,  78,  59 } /* left = d135 */,
+    {  41,  53,  52, 148,  71, 142,  65, 128,  51 } /* left = d117 */,
+    {  40,  36,  28, 143, 143, 202,  40,  55, 137 } /* left = d153 */,
+    {  52,  34,  29, 129, 183, 227,  42,  35,  43 } /* left = d27 */,
+    {  42,  44,  44, 104, 105, 164,  64, 130,  80 } /* left = d63 */,
+    {  43,  81,  53, 140, 169, 204,  68,  84,  72 } /* left = tm */
+  }
+};
diff --git a/libvpx/vp9/common/vp9_mv.h b/libvpx/vp9/common/vp9_mv.h
new file mode 100644
index 0000000..a1eef46
--- /dev/null
+++ b/libvpx/vp9/common/vp9_mv.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_MV_H_
+#define VP9_COMMON_VP9_MV_H_
+
+#include "vpx/vpx_integer.h"
+
+typedef struct {
+  int16_t row;
+  int16_t col;
+} MV;
+
+typedef union int_mv {
+  uint32_t as_int;
+  MV as_mv;
+} int_mv; /* facilitates faster equality tests and copies */
+
+struct mv32 {
+  int32_t row;
+  int32_t col;
+};
+
+typedef union int_mv32 {
+  uint64_t    as_int;
+  struct mv32 as_mv;
+} int_mv32; /* facilitates faster equality tests and copies */
+
+#endif  // VP9_COMMON_VP9_MV_H_
diff --git a/libvpx/vp9/common/vp9_mvref_common.c b/libvpx/vp9/common/vp9_mvref_common.c
new file mode 100644
index 0000000..78fb2f0
--- /dev/null
+++ b/libvpx/vp9/common/vp9_mvref_common.c
@@ -0,0 +1,306 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_mvref_common.h"
+
+#define MVREF_NEIGHBOURS 8
+static int mv_ref_blocks[BLOCK_SIZE_TYPES][MVREF_NEIGHBOURS][2] = {
+  // SB4X4
+  {{0, -1}, {-1, 0}, {-1, -1}, {0, -2}, {-2, 0}, {-1, -2}, {-2, -1}, {-2, -2}},
+  // SB4X8
+  {{0, -1}, {-1, 0}, {-1, -1}, {0, -2}, {-2, 0}, {-1, -2}, {-2, -1}, {-2, -2}},
+  // SB8X4
+  {{0, -1}, {-1, 0}, {-1, -1}, {0, -2}, {-2, 0}, {-1, -2}, {-2, -1}, {-2, -2}},
+  // SB8X8
+  {{0, -1}, {-1, 0}, {-1, -1}, {0, -2}, {-2, 0}, {-1, -2}, {-2, -1}, {-2, -2}},
+  // SB8X16
+  {{-1, 0}, {0, -1}, {-1, 1}, {-1, -1}, {-2, 0}, {0, -2}, {-1, -2}, {-2, -1}},
+  // SB16X8
+  {{0, -1}, {-1, 0}, {1, -1}, {-1, -1}, {0, -2}, {-2, 0}, {-2, -1}, {-1, -2}},
+  // SB16X16
+  {{0, -1}, {-1, 0}, {1, -1}, {-1, 1}, {-1, -1}, {0, -3}, {-3, 0}, {-3, -3}},
+  // SB16X32
+  {{-1, 0}, {0, -1}, {-1, 2}, {-1, -1}, {1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
+  // SB32X16
+  {{0, -1}, {-1, 0}, {2, -1}, {-1, -1}, {-1, 1}, {0, -3}, {-3, 0}, {-3, -3}},
+  // SB32X32
+  {{1, -1}, {-1, 1}, {2, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {-3, -3}},
+  // SB32X64
+  {{-1, 0}, {0, -1}, {-1, 4}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-1, 2}},
+  // SB64X32
+  {{0, -1}, {-1, 0}, {4, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {2, -1}},
+  // SB64X64
+  {{3, -1}, {-1, 3}, {4, -1}, {-1, 4}, {-1, -1}, {0, -1}, {-1, 0}, {6, -1}}
+};
+// clamp_mv_ref
+#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units
+
+static void clamp_mv_ref(const MACROBLOCKD *xd, int_mv *mv) {
+  mv->as_mv.col = clamp(mv->as_mv.col, xd->mb_to_left_edge - MV_BORDER,
+                                       xd->mb_to_right_edge + MV_BORDER);
+  mv->as_mv.row = clamp(mv->as_mv.row, xd->mb_to_top_edge - MV_BORDER,
+                                       xd->mb_to_bottom_edge + MV_BORDER);
+}
+
+// Gets a candidate reference motion vector from the given mode info
+// structure if one exists that matches the given reference frame.
+static int get_matching_candidate(const MODE_INFO *candidate_mi,
+                                  MV_REFERENCE_FRAME ref_frame,
+                                  int_mv *c_mv, int block_idx) {
+  if (ref_frame == candidate_mi->mbmi.ref_frame[0]) {
+    if (block_idx >= 0 && candidate_mi->mbmi.sb_type < BLOCK_SIZE_SB8X8)
+      c_mv->as_int = candidate_mi->bmi[block_idx].as_mv[0].as_int;
+    else
+      c_mv->as_int = candidate_mi->mbmi.mv[0].as_int;
+  } else if (ref_frame == candidate_mi->mbmi.ref_frame[1]) {
+    if (block_idx >= 0 && candidate_mi->mbmi.sb_type < BLOCK_SIZE_SB8X8)
+      c_mv->as_int = candidate_mi->bmi[block_idx].as_mv[1].as_int;
+    else
+      c_mv->as_int = candidate_mi->mbmi.mv[1].as_int;
+  } else {
+    return 0;
+  }
+
+  return 1;
+}
+
+// Gets candidate reference motion vector(s) from the given mode info
+// structure if they exists and do NOT match the given reference frame.
+static void get_non_matching_candidates(const MODE_INFO *candidate_mi,
+                                        MV_REFERENCE_FRAME ref_frame,
+                                        MV_REFERENCE_FRAME *c_ref_frame,
+                                        int_mv *c_mv,
+                                        MV_REFERENCE_FRAME *c2_ref_frame,
+                                        int_mv *c2_mv) {
+
+  c_mv->as_int = 0;
+  c2_mv->as_int = 0;
+  *c_ref_frame = INTRA_FRAME;
+  *c2_ref_frame = INTRA_FRAME;
+
+  // If first candidate not valid neither will be.
+  if (candidate_mi->mbmi.ref_frame[0] > INTRA_FRAME) {
+    // First candidate
+    if (candidate_mi->mbmi.ref_frame[0] != ref_frame) {
+      *c_ref_frame = candidate_mi->mbmi.ref_frame[0];
+      c_mv->as_int = candidate_mi->mbmi.mv[0].as_int;
+    }
+
+    // Second candidate
+    if ((candidate_mi->mbmi.ref_frame[1] > INTRA_FRAME) &&
+        (candidate_mi->mbmi.ref_frame[1] != ref_frame) &&
+        (candidate_mi->mbmi.mv[1].as_int != candidate_mi->mbmi.mv[0].as_int)) {
+      *c2_ref_frame = candidate_mi->mbmi.ref_frame[1];
+      c2_mv->as_int = candidate_mi->mbmi.mv[1].as_int;
+    }
+  }
+}
+
+
+// Performs mv sign inversion if indicated by the reference frame combination.
+static void scale_mv(MACROBLOCKD *xd, MV_REFERENCE_FRAME this_ref_frame,
+                     MV_REFERENCE_FRAME candidate_ref_frame,
+                     int_mv *candidate_mv, int *ref_sign_bias) {
+
+  // Sign inversion where appropriate.
+  if (ref_sign_bias[candidate_ref_frame] != ref_sign_bias[this_ref_frame]) {
+    candidate_mv->as_mv.row = -candidate_mv->as_mv.row;
+    candidate_mv->as_mv.col = -candidate_mv->as_mv.col;
+  }
+}
+
+// Add a candidate mv.
+// Discard if it has already been seen.
+static void add_candidate_mv(int_mv *mv_list,  int *mv_scores,
+                             int *candidate_count, int_mv candidate_mv,
+                             int weight) {
+  if (*candidate_count == 0) {
+    mv_list[0].as_int = candidate_mv.as_int;
+    mv_scores[0] = weight;
+    *candidate_count += 1;
+  } else if ((*candidate_count == 1) &&
+             (candidate_mv.as_int != mv_list[0].as_int)) {
+    mv_list[1].as_int = candidate_mv.as_int;
+    mv_scores[1] = weight;
+    *candidate_count += 1;
+  }
+}
+
+// This function searches the neighbourhood of a given MB/SB
+// to try and find candidate reference vectors.
+//
+void vp9_find_mv_refs_idx(VP9_COMMON *cm, MACROBLOCKD *xd, MODE_INFO *here,
+                          MODE_INFO *lf_here, MV_REFERENCE_FRAME ref_frame,
+                          int_mv *mv_ref_list, int *ref_sign_bias,
+                          int block_idx) {
+  int i;
+  MODE_INFO *candidate_mi;
+  MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
+  int_mv c_refmv;
+  int_mv c2_refmv;
+  MV_REFERENCE_FRAME c_ref_frame;
+  MV_REFERENCE_FRAME c2_ref_frame;
+  int candidate_scores[MAX_MV_REF_CANDIDATES];
+  int refmv_count = 0;
+  int split_count = 0;
+  int (*mv_ref_search)[2];
+  const int mi_col = get_mi_col(xd);
+  const int mi_row = get_mi_row(xd);
+  int intra_count = 0;
+  int zero_count = 0;
+  int newmv_count = 0;
+  int x_idx = 0, y_idx = 0;
+
+  // Blank the reference vector lists and other local structures.
+  vpx_memset(mv_ref_list, 0, sizeof(int_mv) * MAX_MV_REF_CANDIDATES);
+  vpx_memset(candidate_scores, 0, sizeof(candidate_scores));
+
+  mv_ref_search = mv_ref_blocks[mbmi->sb_type];
+  if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+    x_idx = block_idx & 1;
+    y_idx = block_idx >> 1;
+  }
+
+  // We first scan for candidate vectors that match the current reference frame
+  // Look at nearest neigbours
+  for (i = 0; i < 2; ++i) {
+    const int mi_search_col = mi_col + mv_ref_search[i][0];
+    const int mi_search_row = mi_row + mv_ref_search[i][1];
+    if ((mi_search_col >= cm->cur_tile_mi_col_start) &&
+        (mi_search_col < cm->cur_tile_mi_col_end) &&
+        (mi_search_row >= 0) && (mi_search_row < cm->mi_rows)) {
+      int b;
+
+      candidate_mi = here + mv_ref_search[i][0] +
+                     (mv_ref_search[i][1] * xd->mode_info_stride);
+
+      if (block_idx >= 0) {
+        if (mv_ref_search[i][0])
+          b = 1 + y_idx * 2;
+        else
+          b = 2 + x_idx;
+      } else {
+        b = -1;
+      }
+      if (get_matching_candidate(candidate_mi, ref_frame, &c_refmv, b)) {
+        add_candidate_mv(mv_ref_list, candidate_scores,
+                         &refmv_count, c_refmv, 16);
+      }
+      split_count += (candidate_mi->mbmi.sb_type < BLOCK_SIZE_SB8X8 &&
+                      candidate_mi->mbmi.ref_frame[0] != INTRA_FRAME);
+
+      // Count number of neihgbours coded intra and zeromv
+      intra_count += (candidate_mi->mbmi.mode < NEARESTMV);
+      zero_count += (candidate_mi->mbmi.mode == ZEROMV);
+      newmv_count += (candidate_mi->mbmi.mode >= NEWMV);
+    }
+  }
+
+  // More distant neigbours
+  for (i = 2; (i < MVREF_NEIGHBOURS) &&
+              (refmv_count < MAX_MV_REF_CANDIDATES); ++i) {
+    const int mi_search_col = mi_col + mv_ref_search[i][0];
+    const int mi_search_row = mi_row + mv_ref_search[i][1];
+    if ((mi_search_col >= cm->cur_tile_mi_col_start) &&
+        (mi_search_col < cm->cur_tile_mi_col_end) &&
+        (mi_search_row >= 0) && (mi_search_row < cm->mi_rows)) {
+      candidate_mi = here + mv_ref_search[i][0] +
+                     (mv_ref_search[i][1] * xd->mode_info_stride);
+
+      if (get_matching_candidate(candidate_mi, ref_frame, &c_refmv, -1)) {
+        add_candidate_mv(mv_ref_list, candidate_scores,
+                         &refmv_count, c_refmv, 16);
+      }
+    }
+  }
+
+  // Look in the last frame if it exists
+  if (lf_here && (refmv_count < MAX_MV_REF_CANDIDATES)) {
+    candidate_mi = lf_here;
+    if (get_matching_candidate(candidate_mi, ref_frame, &c_refmv, -1)) {
+      add_candidate_mv(mv_ref_list, candidate_scores,
+                       &refmv_count, c_refmv, 16);
+    }
+  }
+
+  // If we have not found enough candidates consider ones where the
+  // reference frame does not match. Break out when we have
+  // MAX_MV_REF_CANDIDATES candidates.
+  // Look first at spatial neighbours
+  for (i = 0; (i < MVREF_NEIGHBOURS) &&
+              (refmv_count < MAX_MV_REF_CANDIDATES); ++i) {
+    const int mi_search_col = mi_col + mv_ref_search[i][0];
+    const int mi_search_row = mi_row + mv_ref_search[i][1];
+    if ((mi_search_col >= cm->cur_tile_mi_col_start) &&
+        (mi_search_col < cm->cur_tile_mi_col_end) &&
+        (mi_search_row >= 0) && (mi_search_row < cm->mi_rows)) {
+      candidate_mi = here + mv_ref_search[i][0] +
+                     (mv_ref_search[i][1] * xd->mode_info_stride);
+
+      get_non_matching_candidates(candidate_mi, ref_frame,
+                                  &c_ref_frame, &c_refmv,
+                                  &c2_ref_frame, &c2_refmv);
+
+      if (c_ref_frame != INTRA_FRAME) {
+        scale_mv(xd, ref_frame, c_ref_frame, &c_refmv, ref_sign_bias);
+        add_candidate_mv(mv_ref_list, candidate_scores,
+                         &refmv_count, c_refmv, 1);
+      }
+
+      if (c2_ref_frame != INTRA_FRAME) {
+        scale_mv(xd, ref_frame, c2_ref_frame, &c2_refmv, ref_sign_bias);
+        add_candidate_mv(mv_ref_list, candidate_scores,
+                         &refmv_count, c2_refmv, 1);
+      }
+    }
+  }
+
+  // Look at the last frame if it exists
+  if (lf_here && (refmv_count < MAX_MV_REF_CANDIDATES)) {
+    candidate_mi = lf_here;
+    get_non_matching_candidates(candidate_mi, ref_frame,
+                                &c_ref_frame, &c_refmv,
+                                &c2_ref_frame, &c2_refmv);
+
+    if (c_ref_frame != INTRA_FRAME) {
+      scale_mv(xd, ref_frame, c_ref_frame, &c_refmv, ref_sign_bias);
+      add_candidate_mv(mv_ref_list, candidate_scores,
+                       &refmv_count, c_refmv, 1);
+    }
+
+    if (c2_ref_frame != INTRA_FRAME) {
+      scale_mv(xd, ref_frame, c2_ref_frame, &c2_refmv, ref_sign_bias);
+      add_candidate_mv(mv_ref_list, candidate_scores,
+                       &refmv_count, c2_refmv, 1);
+    }
+  }
+
+  if (!intra_count) {
+    if (!newmv_count) {
+      // 0 = both zero mv
+      // 1 = one zero mv + one a predicted mv
+      // 2 = two predicted mvs
+      mbmi->mb_mode_context[ref_frame] = 2 - zero_count;
+    } else {
+      // 3 = one predicted/zero and one new mv
+      // 4 = two new mvs
+      mbmi->mb_mode_context[ref_frame] = 2 + newmv_count;
+    }
+  } else {
+    // 5 = one intra neighbour + x
+    // 6 = two intra neighbours
+    mbmi->mb_mode_context[ref_frame] = 4 + intra_count;
+  }
+
+  // Clamp vectors
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
+    clamp_mv_ref(xd, &mv_ref_list[i]);
+  }
+}
diff --git a/libvpx/vp9/common/vp9_mvref_common.h b/libvpx/vp9/common/vp9_mvref_common.h
new file mode 100644
index 0000000..7290f10
--- /dev/null
+++ b/libvpx/vp9/common/vp9_mvref_common.h
@@ -0,0 +1,37 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_blockd.h"
+
+#ifndef VP9_COMMON_VP9_MVREF_COMMON_H_
+#define VP9_COMMON_VP9_MVREF_COMMON_H_
+
+void vp9_find_mv_refs_idx(VP9_COMMON *cm,
+                          MACROBLOCKD *xd,
+                          MODE_INFO *here,
+                          MODE_INFO *lf_here,
+                          MV_REFERENCE_FRAME ref_frame,
+                          int_mv *mv_ref_list,
+                          int *ref_sign_bias,
+                          int block_idx);
+
+static INLINE void vp9_find_mv_refs(VP9_COMMON *cm,
+                                    MACROBLOCKD *xd,
+                                    MODE_INFO *here,
+                                    MODE_INFO *lf_here,
+                                    MV_REFERENCE_FRAME ref_frame,
+                                    int_mv *mv_ref_list,
+                                    int *ref_sign_bias) {
+  vp9_find_mv_refs_idx(cm, xd, here, lf_here, ref_frame,
+                       mv_ref_list, ref_sign_bias, -1);
+}
+
+#endif  // VP9_COMMON_VP9_MVREF_COMMON_H_
diff --git a/libvpx/vp9/common/vp9_onyx.h b/libvpx/vp9/common/vp9_onyx.h
new file mode 100644
index 0000000..b85b889
--- /dev/null
+++ b/libvpx/vp9/common/vp9_onyx.h
@@ -0,0 +1,247 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ONYX_H_
+#define VP9_COMMON_VP9_ONYX_H_
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+#include "./vpx_config.h"
+#include "vpx/internal/vpx_codec_internal.h"
+#include "vpx/vp8cx.h"
+#include "vpx_scale/yv12config.h"
+#include "vp9/common/vp9_ppflags.h"
+
+#define MAX_MB_SEGMENTS 8
+
+  typedef int *VP9_PTR;
+
+  /* Create/destroy static data structures. */
+
+  typedef enum {
+    NORMAL      = 0,
+    FOURFIVE    = 1,
+    THREEFIVE   = 2,
+    ONETWO      = 3
+
+  } VPX_SCALING;
+
+  typedef enum {
+    VP9_LAST_FLAG = 1,
+    VP9_GOLD_FLAG = 2,
+    VP9_ALT_FLAG = 4
+  } VP9_REFFRAME;
+
+
+  typedef enum {
+    USAGE_STREAM_FROM_SERVER    = 0x0,
+    USAGE_LOCAL_FILE_PLAYBACK   = 0x1,
+    USAGE_CONSTRAINED_QUALITY   = 0x2
+  } END_USAGE;
+
+
+  typedef enum {
+    MODE_GOODQUALITY    = 0x1,
+    MODE_BESTQUALITY    = 0x2,
+    MODE_FIRSTPASS      = 0x3,
+    MODE_SECONDPASS     = 0x4,
+    MODE_SECONDPASS_BEST = 0x5,
+  } MODE;
+
+  typedef enum {
+    FRAMEFLAGS_KEY    = 1,
+    FRAMEFLAGS_GOLDEN = 2,
+    FRAMEFLAGS_ALTREF = 4,
+  } FRAMETYPE_FLAGS;
+
+
+#include <assert.h>
+  static INLINE void Scale2Ratio(int mode, int *hr, int *hs) {
+    switch (mode) {
+      case    NORMAL:
+        *hr = 1;
+        *hs = 1;
+        break;
+      case    FOURFIVE:
+        *hr = 4;
+        *hs = 5;
+        break;
+      case    THREEFIVE:
+        *hr = 3;
+        *hs = 5;
+        break;
+      case    ONETWO:
+        *hr = 1;
+        *hs = 2;
+        break;
+      default:
+        *hr = 1;
+        *hs = 1;
+        assert(0);
+        break;
+    }
+  }
+
+  typedef struct {
+    int version;  // 4 versions of bitstream defined:
+                  //   0 - best quality/slowest decode,
+                  //   3 - lowest quality/fastest decode
+    int width;  // width of data passed to the compressor
+    int height;  // height of data passed to the compressor
+    double frame_rate;       // set to passed in framerate
+    int64_t target_bandwidth;    // bandwidth to be used in kilobits per second
+
+    int noise_sensitivity;   // parameter used for applying pre processing blur: recommendation 0
+    int Sharpness;          // parameter used for sharpening output: recommendation 0:
+    int cpu_used;
+    unsigned int rc_max_intra_bitrate_pct;
+
+    // mode ->
+    // (0)=Realtime/Live Encoding. This mode is optimized for realtim encoding (for example, capturing
+    //    a television signal or feed from a live camera). ( speed setting controls how fast )
+    // (1)=Good Quality Fast Encoding. The encoder balances quality with the amount of time it takes to
+    //    encode the output. ( speed setting controls how fast )
+    // (2)=One Pass - Best Quality. The encoder places priority on the quality of the output over encoding
+    //    speed. The output is compressed at the highest possible quality. This option takes the longest
+    //    amount of time to encode. ( speed setting ignored )
+    // (3)=Two Pass - First Pass. The encoder generates a file of statistics for use in the second encoding
+    //    pass. ( speed setting controls how fast )
+    // (4)=Two Pass - Second Pass. The encoder uses the statistics that were generated in the first encoding
+    //    pass to create the compressed output. ( speed setting controls how fast )
+    // (5)=Two Pass - Second Pass Best.  The encoder uses the statistics that were generated in the first
+    //    encoding pass to create the compressed output using the highest possible quality, and taking a
+    //    longer amount of time to encode.. ( speed setting ignored )
+    int Mode;               //
+
+    // Key Framing Operations
+    int auto_key;            // automatically detect cut scenes and set the keyframes
+    int key_freq;            // maximum distance to key frame.
+
+    int allow_lag;           // allow lagged compression (if 0 lagin frames is ignored)
+    int lag_in_frames;        // how many frames lag before we start encoding
+
+    // ----------------------------------------------------------------
+    // DATARATE CONTROL OPTIONS
+
+    int end_usage; // vbr or cbr
+
+    // buffer targeting aggressiveness
+    int under_shoot_pct;
+    int over_shoot_pct;
+
+    // buffering parameters
+    int64_t starting_buffer_level;  // in seconds
+    int64_t optimal_buffer_level;
+    int64_t maximum_buffer_size;
+
+    // controlling quality
+    int fixed_q;
+    int worst_allowed_q;
+    int best_allowed_q;
+    int cq_level;
+    int lossless;
+
+    // two pass datarate control
+    int two_pass_vbrbias;        // two pass datarate control tweaks
+    int two_pass_vbrmin_section;
+    int two_pass_vbrmax_section;
+    // END DATARATE CONTROL OPTIONS
+    // ----------------------------------------------------------------
+
+
+    // these parameters aren't to be used in final build don't use!!!
+    int play_alternate;
+    int alt_freq;
+
+    int encode_breakout;  // early breakout encode threshold : for video conf recommend 800
+
+    /* Bitfield defining the error resiliency features to enable.
+     * Can provide decodable frames after losses in previous
+     * frames and decodable partitions after losses in the same frame.
+     */
+    unsigned int error_resilient_mode;
+
+    /* Bitfield defining the parallel decoding mode where the
+     * decoding in successive frames may be conducted in parallel
+     * just by decoding the frame headers.
+     */
+    unsigned int frame_parallel_decoding_mode;
+
+    int arnr_max_frames;
+    int arnr_strength;
+    int arnr_type;
+
+    int tile_columns;
+    int tile_rows;
+
+    struct vpx_fixed_buf         two_pass_stats_in;
+    struct vpx_codec_pkt_list  *output_pkt_list;
+
+    vp8e_tuning tuning;
+  } VP9_CONFIG;
+
+
+  void vp9_initialize_enc();
+
+  VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf);
+  void vp9_remove_compressor(VP9_PTR *comp);
+
+  void vp9_change_config(VP9_PTR onyx, VP9_CONFIG *oxcf);
+
+// receive a frames worth of data caller can assume that a copy of this frame is made
+// and not just a copy of the pointer..
+  int vp9_receive_raw_frame(VP9_PTR comp, unsigned int frame_flags,
+                            YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
+                            int64_t end_time_stamp);
+
+  int vp9_get_compressed_data(VP9_PTR comp, unsigned int *frame_flags,
+                              unsigned long *size, unsigned char *dest,
+                              int64_t *time_stamp, int64_t *time_end,
+                              int flush);
+
+  int vp9_get_preview_raw_frame(VP9_PTR comp, YV12_BUFFER_CONFIG *dest,
+                                vp9_ppflags_t *flags);
+
+  int vp9_use_as_reference(VP9_PTR comp, int ref_frame_flags);
+
+  int vp9_update_reference(VP9_PTR comp, int ref_frame_flags);
+
+  int vp9_copy_reference_enc(VP9_PTR comp, VP9_REFFRAME ref_frame_flag,
+                             YV12_BUFFER_CONFIG *sd);
+
+  int vp9_get_reference_enc(VP9_PTR ptr, int index, YV12_BUFFER_CONFIG **fb);
+
+  int vp9_set_reference_enc(VP9_PTR comp, VP9_REFFRAME ref_frame_flag,
+                            YV12_BUFFER_CONFIG *sd);
+
+  int vp9_update_entropy(VP9_PTR comp, int update);
+
+  int vp9_set_roimap(VP9_PTR comp, unsigned char *map,
+                     unsigned int rows, unsigned int cols,
+                     int delta_q[MAX_MB_SEGMENTS],
+                     int delta_lf[MAX_MB_SEGMENTS],
+                     unsigned int threshold[MAX_MB_SEGMENTS]);
+
+  int vp9_set_active_map(VP9_PTR comp, unsigned char *map,
+                         unsigned int rows, unsigned int cols);
+
+  int vp9_set_internal_size(VP9_PTR comp,
+                            VPX_SCALING horiz_mode, VPX_SCALING vert_mode);
+
+  int vp9_get_quantizer(VP9_PTR c);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // VP9_COMMON_VP9_ONYX_H_
diff --git a/libvpx/vp9/common/vp9_onyxc_int.h b/libvpx/vp9/common/vp9_onyxc_int.h
new file mode 100644
index 0000000..0d8b0f4
--- /dev/null
+++ b/libvpx/vp9/common/vp9_onyxc_int.h
@@ -0,0 +1,371 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ONYXC_INT_H_
+#define VP9_COMMON_VP9_ONYXC_INT_H_
+
+#include "vpx_config.h"
+#include "vpx/internal/vpx_codec_internal.h"
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_quant_common.h"
+
+#if CONFIG_POSTPROC
+#include "vp9/common/vp9_postproc.h"
+#endif
+
+/* Create/destroy static data structures. */
+
+// Define the number of candidate reference buffers.
+#define NUM_REF_FRAMES 8
+#define NUM_REF_FRAMES_LG2 3
+
+#define ALLOWED_REFS_PER_FRAME 3
+
+// 1 scratch frame for the new frame, 3 for scaled references on the encoder
+// TODO(jkoleszar): These 3 extra references could probably come from the
+// normal reference pool.
+#define NUM_YV12_BUFFERS (NUM_REF_FRAMES + 4)
+
+#define NUM_FRAME_CONTEXTS_LG2 2
+#define NUM_FRAME_CONTEXTS (1 << NUM_FRAME_CONTEXTS_LG2)
+
+#define MAX_LAG_BUFFERS 25
+
+typedef struct frame_contexts {
+  vp9_prob y_mode_prob[BLOCK_SIZE_GROUPS][VP9_INTRA_MODES - 1];
+  vp9_prob uv_mode_prob[VP9_INTRA_MODES][VP9_INTRA_MODES - 1];
+  vp9_prob partition_prob[NUM_FRAME_TYPES][NUM_PARTITION_CONTEXTS]
+                         [PARTITION_TYPES - 1];
+
+  nmv_context nmvc;
+  nmv_context pre_nmvc;
+  /* interframe intra mode probs */
+  vp9_prob pre_y_mode_prob[BLOCK_SIZE_GROUPS][VP9_INTRA_MODES - 1];
+  vp9_prob pre_uv_mode_prob[VP9_INTRA_MODES][VP9_INTRA_MODES - 1];
+  vp9_prob pre_partition_prob[NUM_PARTITION_CONTEXTS][PARTITION_TYPES - 1];
+  /* interframe intra mode probs */
+  unsigned int y_mode_counts[BLOCK_SIZE_GROUPS][VP9_INTRA_MODES];
+  unsigned int uv_mode_counts[VP9_INTRA_MODES][VP9_INTRA_MODES];
+  unsigned int partition_counts[NUM_PARTITION_CONTEXTS][PARTITION_TYPES];
+
+  vp9_coeff_probs_model coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  vp9_coeff_probs_model pre_coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  vp9_coeff_count_model coef_counts[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  unsigned int eob_branch_counts[TX_SIZE_MAX_SB][BLOCK_TYPES][REF_TYPES]
+                                [COEF_BANDS][PREV_COEF_CONTEXTS];
+
+  nmv_context_counts NMVcount;
+  vp9_prob switchable_interp_prob[VP9_SWITCHABLE_FILTERS + 1]
+                                 [VP9_SWITCHABLE_FILTERS - 1];
+  vp9_prob pre_switchable_interp_prob[VP9_SWITCHABLE_FILTERS + 1]
+      [VP9_SWITCHABLE_FILTERS - 1];
+  unsigned int switchable_interp_count[VP9_SWITCHABLE_FILTERS + 1]
+                                      [VP9_SWITCHABLE_FILTERS];
+
+  vp9_prob inter_mode_probs[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1];
+  vp9_prob pre_inter_mode_probs[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1];
+  unsigned int inter_mode_counts[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1][2];
+
+  vp9_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
+  vp9_prob comp_inter_prob[COMP_INTER_CONTEXTS];
+  vp9_prob single_ref_prob[REF_CONTEXTS][2];
+  vp9_prob comp_ref_prob[REF_CONTEXTS];
+  vp9_prob pre_intra_inter_prob[INTRA_INTER_CONTEXTS];
+  vp9_prob pre_comp_inter_prob[COMP_INTER_CONTEXTS];
+  vp9_prob pre_single_ref_prob[REF_CONTEXTS][2];
+  vp9_prob pre_comp_ref_prob[REF_CONTEXTS];
+  unsigned int intra_inter_count[INTRA_INTER_CONTEXTS][2];
+  unsigned int comp_inter_count[COMP_INTER_CONTEXTS][2];
+  unsigned int single_ref_count[REF_CONTEXTS][2][2];
+  unsigned int comp_ref_count[REF_CONTEXTS][2];
+
+  vp9_prob tx_probs_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
+  vp9_prob tx_probs_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
+  vp9_prob tx_probs_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 3];
+  vp9_prob pre_tx_probs_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
+  vp9_prob pre_tx_probs_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
+  vp9_prob pre_tx_probs_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 3];
+  unsigned int tx_count_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB];
+  unsigned int tx_count_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
+  unsigned int tx_count_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
+
+  vp9_prob mbskip_probs[MBSKIP_CONTEXTS];
+  vp9_prob pre_mbskip_probs[MBSKIP_CONTEXTS];
+  unsigned int mbskip_count[MBSKIP_CONTEXTS][2];
+} FRAME_CONTEXT;
+
+typedef enum {
+  SINGLE_PREDICTION_ONLY = 0,
+  COMP_PREDICTION_ONLY   = 1,
+  HYBRID_PREDICTION      = 2,
+  NB_PREDICTION_TYPES    = 3,
+} COMPPREDMODE_TYPE;
+
+typedef enum {
+  ONLY_4X4            = 0,
+  ALLOW_8X8           = 1,
+  ALLOW_16X16         = 2,
+  ALLOW_32X32         = 3,
+  TX_MODE_SELECT      = 4,
+  NB_TXFM_MODES       = 5,
+} TXFM_MODE;
+
+typedef struct VP9Common {
+  struct vpx_internal_error_info  error;
+
+  DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][2]);
+  DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][2]);
+#if CONFIG_ALPHA
+  DECLARE_ALIGNED(16, int16_t, a_dequant[QINDEX_RANGE][2]);
+#endif
+
+  int width;
+  int height;
+  int display_width;
+  int display_height;
+  int last_width;
+  int last_height;
+
+  // TODO(jkoleszar): this implies chroma ss right now, but could vary per
+  // plane. Revisit as part of the future change to YV12_BUFFER_CONFIG to
+  // support additional planes.
+  int subsampling_x;
+  int subsampling_y;
+
+  YUV_TYPE clr_type;
+
+  YV12_BUFFER_CONFIG *frame_to_show;
+
+  YV12_BUFFER_CONFIG yv12_fb[NUM_YV12_BUFFERS];
+  int fb_idx_ref_cnt[NUM_YV12_BUFFERS]; /* reference counts */
+  int ref_frame_map[NUM_REF_FRAMES]; /* maps fb_idx to reference slot */
+
+  // TODO(jkoleszar): could expand active_ref_idx to 4, with 0 as intra, and
+  // roll new_fb_idx into it.
+
+  // Each frame can reference ALLOWED_REFS_PER_FRAME buffers
+  int active_ref_idx[ALLOWED_REFS_PER_FRAME];
+  struct scale_factors active_ref_scale[ALLOWED_REFS_PER_FRAME];
+  int new_fb_idx;
+
+
+  YV12_BUFFER_CONFIG post_proc_buffer;
+  YV12_BUFFER_CONFIG temp_scale_frame;
+
+
+  FRAME_TYPE last_frame_type;  /* Save last frame's frame type for motion search. */
+  FRAME_TYPE frame_type;
+
+  int show_frame;
+  int last_show_frame;
+
+  // Flag signaling that the frame is encoded using only INTRA modes.
+  int intra_only;
+
+  // Flag signaling that the frame context should be reset to default values.
+  // 0 or 1 implies don't reset, 2 reset just the context specified in the
+  // frame header, 3 reset all contexts.
+  int reset_frame_context;
+
+  int frame_flags;
+  // MBs, mb_rows/cols is in 16-pixel units; mi_rows/cols is in
+  // MODE_INFO (8-pixel) units.
+  int MBs;
+  int mb_rows, mi_rows;
+  int mb_cols, mi_cols;
+  int mode_info_stride;
+
+  /* profile settings */
+  TXFM_MODE txfm_mode;
+
+  int base_qindex;
+  int last_kf_gf_q;  /* Q used on the last GF or KF */
+
+  int y_dc_delta_q;
+  int uv_dc_delta_q;
+  int uv_ac_delta_q;
+#if CONFIG_ALPHA
+  int a_dc_delta_q;
+  int a_ac_delta_q;
+#endif
+
+  unsigned int frames_since_golden;
+  unsigned int frames_till_alt_ref_frame;
+
+  /* We allocate a MODE_INFO struct for each macroblock, together with
+     an extra row on top and column on the left to simplify prediction. */
+
+  MODE_INFO *mip; /* Base of allocated array */
+  MODE_INFO *mi;  /* Corresponds to upper left visible macroblock */
+  MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
+  MODE_INFO *prev_mi;  /* 'mi' from last frame (points into prev_mip) */
+
+
+  // Persistent mb segment id map used in prediction.
+  unsigned char *last_frame_seg_map;
+
+  INTERPOLATIONFILTERTYPE mcomp_filter_type;
+
+  loop_filter_info_n lf_info;
+
+  int filter_level;
+  int last_sharpness_level;
+  int sharpness_level;
+
+  int refresh_frame_context;    /* Two state 0 = NO, 1 = YES */
+
+  int ref_frame_sign_bias[MAX_REF_FRAMES];    /* Two state 0, 1 */
+
+  /* Y,U,V */
+  ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
+  ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
+
+  // partition contexts
+  PARTITION_CONTEXT *above_seg_context;
+  PARTITION_CONTEXT left_seg_context[8];
+
+  /* keyframe block modes are predicted by their above, left neighbors */
+
+  vp9_prob kf_y_mode_prob[VP9_INTRA_MODES]
+                         [VP9_INTRA_MODES]
+                         [VP9_INTRA_MODES - 1];
+  vp9_prob kf_uv_mode_prob[VP9_INTRA_MODES] [VP9_INTRA_MODES - 1];
+
+  // Context probabilities when using predictive coding of segment id
+  vp9_prob segment_pred_probs[PREDICTION_PROBS];
+  unsigned char temporal_update;
+
+  // Context probabilities for reference frame prediction
+  int allow_comp_inter_inter;
+  MV_REFERENCE_FRAME comp_fixed_ref;
+  MV_REFERENCE_FRAME comp_var_ref[2];
+  COMPPREDMODE_TYPE comp_pred_mode;
+
+  FRAME_CONTEXT fc;  /* this frame entropy */
+  FRAME_CONTEXT frame_contexts[NUM_FRAME_CONTEXTS];
+  unsigned int  frame_context_idx; /* Context to use/update */
+
+  unsigned int current_video_frame;
+  int near_boffset[3];
+  int version;
+
+  double bitrate;
+  double framerate;
+
+#if CONFIG_POSTPROC
+  struct postproc_state  postproc_state;
+#endif
+
+  int error_resilient_mode;
+  int frame_parallel_decoding_mode;
+
+  int tile_columns, log2_tile_columns;
+  int cur_tile_mi_col_start, cur_tile_mi_col_end, cur_tile_col_idx;
+  int tile_rows, log2_tile_rows;
+  int cur_tile_mi_row_start, cur_tile_mi_row_end, cur_tile_row_idx;
+} VP9_COMMON;
+
+static int get_free_fb(VP9_COMMON *cm) {
+  int i;
+  for (i = 0; i < NUM_YV12_BUFFERS; i++)
+    if (cm->fb_idx_ref_cnt[i] == 0)
+      break;
+
+  assert(i < NUM_YV12_BUFFERS);
+  cm->fb_idx_ref_cnt[i] = 1;
+  return i;
+}
+
+static void ref_cnt_fb(int *buf, int *idx, int new_idx) {
+  if (buf[*idx] > 0)
+    buf[*idx]--;
+
+  *idx = new_idx;
+
+  buf[new_idx]++;
+}
+
+static int mi_cols_aligned_to_sb(VP9_COMMON *cm) {
+  return 2 * ((cm->mb_cols + 3) & ~3);
+}
+
+static INLINE void set_partition_seg_context(VP9_COMMON *cm,
+                                             MACROBLOCKD *xd,
+                                             int mi_row, int mi_col) {
+  xd->above_seg_context = cm->above_seg_context + mi_col;
+  xd->left_seg_context  = cm->left_seg_context + (mi_row & MI_MASK);
+}
+
+static int check_bsize_coverage(VP9_COMMON *cm, MACROBLOCKD *xd,
+                                int mi_row, int mi_col,
+                                BLOCK_SIZE_TYPE bsize) {
+  int bsl = mi_width_log2(bsize), bs = 1 << bsl;
+  int ms = bs / 2;
+
+  if ((mi_row + ms < cm->mi_rows) && (mi_col + ms < cm->mi_cols))
+    return 0;
+
+  // frame width/height are multiples of 8, hence 8x8 block should always
+  // pass the above check
+  assert(bsize > BLOCK_SIZE_SB8X8);
+
+  // return the node index in the prob tree for binary coding
+  // only allow horizontal/split partition types
+  if ((mi_col + ms < cm->mi_cols) && (mi_row + ms >= cm->mi_rows))
+    return 1;
+  // only allow vertical/split partition types
+  if ((mi_row + ms < cm->mi_rows) && (mi_col + ms >= cm->mi_cols))
+    return 2;
+
+  return -1;
+}
+
+static void set_mi_row_col(VP9_COMMON *cm, MACROBLOCKD *xd,
+                       int mi_row, int bh,
+                       int mi_col, int bw) {
+  xd->mb_to_top_edge    = -((mi_row * MI_SIZE) << 3);
+  xd->mb_to_bottom_edge = ((cm->mi_rows - bh - mi_row) * MI_SIZE) << 3;
+  xd->mb_to_left_edge   = -((mi_col * MI_SIZE) << 3);
+  xd->mb_to_right_edge  = ((cm->mi_cols - bw - mi_col) * MI_SIZE) << 3;
+
+  // Are edges available for intra prediction?
+  xd->up_available    = (mi_row != 0);
+  xd->left_available  = (mi_col > cm->cur_tile_mi_col_start);
+  xd->right_available = (mi_col + bw < cm->cur_tile_mi_col_end);
+}
+
+static int get_mi_row(const MACROBLOCKD *xd) {
+  return ((-xd->mb_to_top_edge) >> (3 + LOG2_MI_SIZE));
+}
+
+static int get_mi_col(const MACROBLOCKD *xd) {
+  return ((-xd->mb_to_left_edge) >> (3 + LOG2_MI_SIZE));
+}
+
+static int get_token_alloc(int mb_rows, int mb_cols) {
+  return mb_rows * mb_cols * (48 * 16 + 4);
+}
+
+static void set_prev_mi(VP9_COMMON *cm) {
+  const int use_prev_in_find_mv_refs = cm->width == cm->last_width &&
+                                       cm->height == cm->last_height &&
+                                       !cm->error_resilient_mode &&
+                                       !cm->intra_only &&
+                                       cm->last_show_frame;
+  // Special case: set prev_mi to NULL when the previous mode info
+  // context cannot be used.
+  cm->prev_mi = use_prev_in_find_mv_refs ?
+                  cm->prev_mip + cm->mode_info_stride + 1 : NULL;
+}
+#endif  // VP9_COMMON_VP9_ONYXC_INT_H_
diff --git a/libvpx/vp9/common/vp9_postproc.c b/libvpx/vp9/common/vp9_postproc.c
new file mode 100644
index 0000000..4282ddd
--- /dev/null
+++ b/libvpx/vp9/common/vp9_postproc.c
@@ -0,0 +1,1017 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "./vpx_config.h"
+#include "vpx_scale/yv12config.h"
+#include "vp9/common/vp9_postproc.h"
+#include "vp9/common/vp9_textblit.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "./vp9_rtcd.h"
+#include "./vpx_scale_rtcd.h"
+
+
+#include <math.h>
+#include <stdlib.h>
+#include <stdio.h>
+
+#define RGB_TO_YUV(t)                                            \
+  ( (0.257*(float)(t >> 16))  + (0.504*(float)(t >> 8 & 0xff)) + \
+    (0.098*(float)(t & 0xff)) + 16),                             \
+  (-(0.148*(float)(t >> 16))  - (0.291*(float)(t >> 8 & 0xff)) + \
+    (0.439*(float)(t & 0xff)) + 128),                            \
+  ( (0.439*(float)(t >> 16))  - (0.368*(float)(t >> 8 & 0xff)) - \
+    (0.071*(float)(t & 0xff)) + 128)
+
+/* global constants */
+#if 0 && CONFIG_POSTPROC_VISUALIZER
+static const unsigned char MB_PREDICTION_MODE_colors[MB_MODE_COUNT][3] = {
+  { RGB_TO_YUV(0x98FB98) },   /* PaleGreen */
+  { RGB_TO_YUV(0x00FF00) },   /* Green */
+  { RGB_TO_YUV(0xADFF2F) },   /* GreenYellow */
+  { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
+  { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
+  { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
+  { RGB_TO_YUV(0x008F8F) },   /* Dark Cyan */
+  { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
+  { RGB_TO_YUV(0x8F0000) },   /* Dark Red */
+  { RGB_TO_YUV(0x228B22) },   /* ForestGreen */
+  { RGB_TO_YUV(0x006400) },   /* DarkGreen */
+  { RGB_TO_YUV(0x98F5FF) },   /* Cadet Blue */
+  { RGB_TO_YUV(0x6CA6CD) },   /* Sky Blue */
+  { RGB_TO_YUV(0x00008B) },   /* Dark blue */
+  { RGB_TO_YUV(0x551A8B) },   /* Purple */
+  { RGB_TO_YUV(0xFF0000) }    /* Red */
+  { RGB_TO_YUV(0xCC33FF) },   /* Magenta */
+};
+
+static const unsigned char B_PREDICTION_MODE_colors[VP9_INTRA_MODES][3] = {
+  { RGB_TO_YUV(0x6633ff) },   /* Purple */
+  { RGB_TO_YUV(0xcc33ff) },   /* Magenta */
+  { RGB_TO_YUV(0xff33cc) },   /* Pink */
+  { RGB_TO_YUV(0xff3366) },   /* Coral */
+  { RGB_TO_YUV(0x3366ff) },   /* Blue */
+  { RGB_TO_YUV(0xed00f5) },   /* Dark Blue */
+  { RGB_TO_YUV(0x2e00b8) },   /* Dark Purple */
+  { RGB_TO_YUV(0xff6633) },   /* Orange */
+  { RGB_TO_YUV(0x33ccff) },   /* Light Blue */
+  { RGB_TO_YUV(0x8ab800) },   /* Green */
+  { RGB_TO_YUV(0xffcc33) },   /* Light Orange */
+  { RGB_TO_YUV(0x33ffcc) },   /* Aqua */
+  { RGB_TO_YUV(0x66ff33) },   /* Light Green */
+  { RGB_TO_YUV(0xccff33) },   /* Yellow */
+};
+
+static const unsigned char MV_REFERENCE_FRAME_colors[MAX_REF_FRAMES][3] = {
+  { RGB_TO_YUV(0x00ff00) },   /* Blue */
+  { RGB_TO_YUV(0x0000ff) },   /* Green */
+  { RGB_TO_YUV(0xffff00) },   /* Yellow */
+  { RGB_TO_YUV(0xff0000) },   /* Red */
+};
+#endif
+
+static const short kernel5[] = {
+  1, 1, 4, 1, 1
+};
+
+const short vp9_rv[] = {
+  8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
+  0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
+  10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
+  8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
+  8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
+  1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
+  3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
+  11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
+  14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
+  4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
+  7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
+  0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
+  8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
+  3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
+  3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
+  13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
+  5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
+  9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
+  4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
+  3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
+  11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
+  5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
+  0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
+  10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
+  4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
+  0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
+  8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
+  3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
+  3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
+  13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
+  5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
+  9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
+  4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
+  3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
+  11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
+  5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
+  0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
+  10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
+  4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
+  3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
+  11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
+  14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
+  5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
+  0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
+};
+
+
+/****************************************************************************
+ */
+void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
+                                     uint8_t *dst_ptr,
+                                     int src_pixels_per_line,
+                                     int dst_pixels_per_line,
+                                     int rows,
+                                     int cols,
+                                     int flimit) {
+  uint8_t const *p_src;
+  uint8_t *p_dst;
+  int row;
+  int col;
+  int i;
+  int v;
+  int pitch = src_pixels_per_line;
+  uint8_t d[8];
+  (void)dst_pixels_per_line;
+
+  for (row = 0; row < rows; row++) {
+    /* post_proc_down for one row */
+    p_src = src_ptr;
+    p_dst = dst_ptr;
+
+    for (col = 0; col < cols; col++) {
+
+      int kernel = 4;
+      int v = p_src[col];
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i * pitch]) > flimit)
+          goto down_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i * pitch];
+      }
+
+      v = (kernel >> 3);
+    down_skip_convolve:
+      p_dst[col] = v;
+    }
+
+    /* now post_proc_across */
+    p_src = dst_ptr;
+    p_dst = dst_ptr;
+
+    for (i = 0; i < 8; i++)
+      d[i] = p_src[i];
+
+    for (col = 0; col < cols; col++) {
+      int kernel = 4;
+      v = p_src[col];
+
+      d[col & 7] = v;
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i]) > flimit)
+          goto across_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i];
+      }
+
+      d[col & 7] = (kernel >> 3);
+    across_skip_convolve:
+
+      if (col >= 2)
+        p_dst[col - 2] = d[(col - 2) & 7];
+    }
+
+    /* handle the last two pixels */
+    p_dst[col - 2] = d[(col - 2) & 7];
+    p_dst[col - 1] = d[(col - 1) & 7];
+
+
+    /* next row */
+    src_ptr += pitch;
+    dst_ptr += pitch;
+  }
+}
+
+static int q2mbl(int x) {
+  if (x < 20) x = 20;
+
+  x = 50 + (x - 50) * 10 / 8;
+  return x * x / 3;
+}
+
+void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
+                                 int rows, int cols, int flimit) {
+  int r, c, i;
+
+  uint8_t *s = src;
+  uint8_t d[16];
+
+
+  for (r = 0; r < rows; r++) {
+    int sumsq = 0;
+    int sum   = 0;
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i] * s[i];
+      sum   += s[i];
+      d[i + 8] = 0;
+    }
+
+    for (c = 0; c < cols + 8; c++) {
+      int x = s[c + 7] - s[c - 8];
+      int y = s[c + 7] + s[c - 8];
+
+      sum  += x;
+      sumsq += x * y;
+
+      d[c & 15] = s[c];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[c & 15] = (8 + sum + s[c]) >> 4;
+      }
+
+      s[c - 8] = d[(c - 8) & 15];
+    }
+
+    s += pitch;
+  }
+}
+
+void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
+                            int rows, int cols, int flimit) {
+  int r, c, i;
+  const short *rv3 = &vp9_rv[63 & rand()];
+
+  for (c = 0; c < cols; c++) {
+    uint8_t *s = &dst[c];
+    int sumsq = 0;
+    int sum   = 0;
+    uint8_t d[16];
+    const short *rv2 = rv3 + ((c * 17) & 127);
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i * pitch] * s[i * pitch];
+      sum   += s[i * pitch];
+    }
+
+    for (r = 0; r < rows + 8; r++) {
+      sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
+      sum  += s[7 * pitch] - s[-8 * pitch];
+      d[r & 15] = s[0];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
+      }
+
+      s[-8 * pitch] = d[(r - 8) & 15];
+      s += pitch;
+    }
+  }
+}
+
+static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
+                                       YV12_BUFFER_CONFIG   *post,
+                                       int                   q,
+                                       int                   low_var_thresh,
+                                       int                   flag) {
+  double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
+  int ppl = (int)(level + .5);
+  (void) low_var_thresh;
+  (void) flag;
+
+  vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
+                                source->y_stride, post->y_stride,
+                                source->y_height, source->y_width, ppl);
+
+  vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
+                            post->y_width, q2mbl(q));
+
+  vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
+                       post->y_width, q2mbl(q));
+
+  vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
+                                source->uv_stride, post->uv_stride,
+                                source->uv_height, source->uv_width, ppl);
+  vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
+                                source->uv_stride, post->uv_stride,
+                                source->uv_height, source->uv_width, ppl);
+}
+
+void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
+                 int q) {
+  const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
+                        + 0.0065 + 0.5);
+  int i;
+
+  const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                                  src->alpha_buffer};
+  const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                              src->alpha_stride};
+  const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
+                             src->alpha_width};
+  const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
+                              src->alpha_height};
+
+  uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
+                            dst->alpha_buffer};
+  const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
+                              dst->alpha_stride};
+
+  for (i = 0; i < MAX_MB_PLANE; ++i)
+    vp9_post_proc_down_and_across(srcs[i], dsts[i],
+                                  src_strides[i], dst_strides[i],
+                                  src_heights[i], src_widths[i], ppl);
+}
+
+void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
+                 int q) {
+  const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
+                        + 0.0065 + 0.5);
+  int i;
+
+  const uint8_t *const srcs[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                                  src->alpha_buffer};
+  const int src_strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                              src->alpha_stride};
+  const int src_widths[4] = {src->y_width, src->uv_width, src->uv_width,
+                             src->alpha_width};
+  const int src_heights[4] = {src->y_height, src->uv_height, src->uv_height,
+                              src->alpha_height};
+
+  uint8_t *const dsts[4] = {dst->y_buffer, dst->u_buffer, dst->v_buffer,
+                            dst->alpha_buffer};
+  const int dst_strides[4] = {dst->y_stride, dst->uv_stride, dst->uv_stride,
+                              dst->alpha_stride};
+
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+    const int src_stride = src_strides[i];
+    const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
+    const int src_width = src_widths[i] - 4;
+    const int src_height = src_heights[i] - 4;
+
+    const int dst_stride = dst_strides[i];
+    uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
+
+    vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
+                                  src_height, src_width, ppl);
+  }
+}
+
+double vp9_gaussian(double sigma, double mu, double x) {
+  return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
+         (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
+}
+
+static void fillrd(struct postproc_state *state, int q, int a) {
+  char char_dist[300];
+
+  double sigma;
+  int ai = a, qi = q, i;
+
+  vp9_clear_system_state();
+
+  sigma = ai + .5 + .6 * (63 - qi) / 63.0;
+
+  /* set up a lookup table of 256 entries that matches
+   * a gaussian distribution with sigma determined by q.
+   */
+  {
+    double i;
+    int next, j;
+
+    next = 0;
+
+    for (i = -32; i < 32; i++) {
+      int a = (int)(.5 + 256 * vp9_gaussian(sigma, 0, i));
+
+      if (a) {
+        for (j = 0; j < a; j++) {
+          char_dist[next + j] = (char) i;
+        }
+
+        next = next + j;
+      }
+
+    }
+
+    for (next = next; next < 256; next++)
+      char_dist[next] = 0;
+  }
+
+  for (i = 0; i < 3072; i++) {
+    state->noise[i] = char_dist[rand() & 0xff];
+  }
+
+  for (i = 0; i < 16; i++) {
+    state->blackclamp[i] = -char_dist[0];
+    state->whiteclamp[i] = -char_dist[0];
+    state->bothclamp[i] = -2 * char_dist[0];
+  }
+
+  state->last_q = q;
+  state->last_noise = a;
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : plane_add_noise_c
+ *
+ *  INPUTS        : unsigned char *Start  starting address of buffer to
+ *                                        add gaussian noise to
+ *                  unsigned int width    width of plane
+ *                  unsigned int height   height of plane
+ *                  int  pitch    distance between subsequent lines of frame
+ *                  int  q        quantizer used to determine amount of noise
+ *                                  to add
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void.
+ *
+ *  FUNCTION      : adds gaussian noise to a plane of pixels
+ *
+ *  SPECIAL NOTES : None.
+ *
+ ****************************************************************************/
+void vp9_plane_add_noise_c(uint8_t *start, char *noise,
+                           char blackclamp[16],
+                           char whiteclamp[16],
+                           char bothclamp[16],
+                           unsigned int width, unsigned int height, int pitch) {
+  unsigned int i, j;
+
+  for (i = 0; i < height; i++) {
+    uint8_t *pos = start + i * pitch;
+    char  *ref = (char *)(noise + (rand() & 0xff));  // NOLINT
+
+    for (j = 0; j < width; j++) {
+      if (pos[j] < blackclamp[0])
+        pos[j] = blackclamp[0];
+
+      if (pos[j] > 255 + whiteclamp[0])
+        pos[j] = 255 + whiteclamp[0];
+
+      pos[j] += ref[j];
+    }
+  }
+}
+
+/* Blend the macro block with a solid colored square.  Leave the
+ * edges unblended to give distinction to macro blocks in areas
+ * filled with the same color block.
+ */
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v,
+                          int y1, int u1, int v1, int alpha, int stride) {
+  int i, j;
+  int y1_const = y1 * ((1 << 16) - alpha);
+  int u1_const = u1 * ((1 << 16) - alpha);
+  int v1_const = v1 * ((1 << 16) - alpha);
+
+  y += 2 * stride + 2;
+  for (i = 0; i < 12; i++) {
+    for (j = 0; j < 12; j++) {
+      y[j] = (y[j] * alpha + y1_const) >> 16;
+    }
+    y += stride;
+  }
+
+  stride >>= 1;
+
+  u += stride + 1;
+  v += stride + 1;
+
+  for (i = 0; i < 6; i++) {
+    for (j = 0; j < 6; j++) {
+      u[j] = (u[j] * alpha + u1_const) >> 16;
+      v[j] = (v[j] * alpha + v1_const) >> 16;
+    }
+    u += stride;
+    v += stride;
+  }
+}
+
+/* Blend only the edge of the macro block.  Leave center
+ * unblended to allow for other visualizations to be layered.
+ */
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v,
+                          int y1, int u1, int v1, int alpha, int stride) {
+  int i, j;
+  int y1_const = y1 * ((1 << 16) - alpha);
+  int u1_const = u1 * ((1 << 16) - alpha);
+  int v1_const = v1 * ((1 << 16) - alpha);
+
+  for (i = 0; i < 2; i++) {
+    for (j = 0; j < 16; j++) {
+      y[j] = (y[j] * alpha + y1_const) >> 16;
+    }
+    y += stride;
+  }
+
+  for (i = 0; i < 12; i++) {
+    y[0]  = (y[0] * alpha  + y1_const) >> 16;
+    y[1]  = (y[1] * alpha  + y1_const) >> 16;
+    y[14] = (y[14] * alpha + y1_const) >> 16;
+    y[15] = (y[15] * alpha + y1_const) >> 16;
+    y += stride;
+  }
+
+  for (i = 0; i < 2; i++) {
+    for (j = 0; j < 16; j++) {
+      y[j] = (y[j] * alpha + y1_const) >> 16;
+    }
+    y += stride;
+  }
+
+  stride >>= 1;
+
+  for (j = 0; j < 8; j++) {
+    u[j] = (u[j] * alpha + u1_const) >> 16;
+    v[j] = (v[j] * alpha + v1_const) >> 16;
+  }
+  u += stride;
+  v += stride;
+
+  for (i = 0; i < 6; i++) {
+    u[0] = (u[0] * alpha + u1_const) >> 16;
+    v[0] = (v[0] * alpha + v1_const) >> 16;
+
+    u[7] = (u[7] * alpha + u1_const) >> 16;
+    v[7] = (v[7] * alpha + v1_const) >> 16;
+
+    u += stride;
+    v += stride;
+  }
+
+  for (j = 0; j < 8; j++) {
+    u[j] = (u[j] * alpha + u1_const) >> 16;
+    v[j] = (v[j] * alpha + v1_const) >> 16;
+  }
+}
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v,
+                   int y1, int u1, int v1, int alpha, int stride) {
+  int i, j;
+  int y1_const = y1 * ((1 << 16) - alpha);
+  int u1_const = u1 * ((1 << 16) - alpha);
+  int v1_const = v1 * ((1 << 16) - alpha);
+
+  for (i = 0; i < 4; i++) {
+    for (j = 0; j < 4; j++) {
+      y[j] = (y[j] * alpha + y1_const) >> 16;
+    }
+    y += stride;
+  }
+
+  stride >>= 1;
+
+  for (i = 0; i < 2; i++) {
+    for (j = 0; j < 2; j++) {
+      u[j] = (u[j] * alpha + u1_const) >> 16;
+      v[j] = (v[j] * alpha + v1_const) >> 16;
+    }
+    u += stride;
+    v += stride;
+  }
+}
+
+static void constrain_line(int x0, int *x1, int y0, int *y1,
+                           int width, int height) {
+  int dx;
+  int dy;
+
+  if (*x1 > width) {
+    dx = *x1 - x0;
+    dy = *y1 - y0;
+
+    *x1 = width;
+    if (dx)
+      *y1 = ((width - x0) * dy) / dx + y0;
+  }
+  if (*x1 < 0) {
+    dx = *x1 - x0;
+    dy = *y1 - y0;
+
+    *x1 = 0;
+    if (dx)
+      *y1 = ((0 - x0) * dy) / dx + y0;
+  }
+  if (*y1 > height) {
+    dx = *x1 - x0;
+    dy = *y1 - y0;
+
+    *y1 = height;
+    if (dy)
+      *x1 = ((height - y0) * dx) / dy + x0;
+  }
+  if (*y1 < 0) {
+    dx = *x1 - x0;
+    dy = *y1 - y0;
+
+    *y1 = 0;
+    if (dy)
+      *x1 = ((0 - y0) * dx) / dy + x0;
+  }
+}
+
+int vp9_post_proc_frame(VP9_COMMON *oci, YV12_BUFFER_CONFIG *dest,
+                        vp9_ppflags_t *ppflags) {
+  int q = oci->filter_level * 10 / 6;
+  int flags = ppflags->post_proc_flag;
+  int deblock_level = ppflags->deblocking_level;
+  int noise_level = ppflags->noise_level;
+
+  if (!oci->frame_to_show)
+    return -1;
+
+  if (q > 63)
+    q = 63;
+
+  if (!flags) {
+    *dest = *oci->frame_to_show;
+    return 0;
+  }
+
+#if ARCH_X86||ARCH_X86_64
+  vpx_reset_mmx_state();
+#endif
+
+  if (flags & VP9D_DEMACROBLOCK) {
+    deblock_and_de_macro_block(oci->frame_to_show, &oci->post_proc_buffer,
+                               q + (deblock_level - 5) * 10, 1, 0);
+  } else if (flags & VP9D_DEBLOCK) {
+    vp9_deblock(oci->frame_to_show, &oci->post_proc_buffer, q);
+  } else {
+    vp8_yv12_copy_frame(oci->frame_to_show, &oci->post_proc_buffer);
+  }
+
+  if (flags & VP9D_ADDNOISE) {
+    if (oci->postproc_state.last_q != q
+        || oci->postproc_state.last_noise != noise_level) {
+      fillrd(&oci->postproc_state, 63 - q, noise_level);
+    }
+
+    vp9_plane_add_noise(oci->post_proc_buffer.y_buffer,
+                        oci->postproc_state.noise,
+                        oci->postproc_state.blackclamp,
+                        oci->postproc_state.whiteclamp,
+                        oci->postproc_state.bothclamp,
+                        oci->post_proc_buffer.y_width,
+                        oci->post_proc_buffer.y_height,
+                        oci->post_proc_buffer.y_stride);
+  }
+
+#if 0 && CONFIG_POSTPROC_VISUALIZER
+  if (flags & VP9D_DEBUG_TXT_FRAME_INFO) {
+    char message[512];
+    sprintf(message, "F%1dG%1dQ%3dF%3dP%d_s%dx%d",
+            (oci->frame_type == KEY_FRAME),
+            oci->refresh_golden_frame,
+            oci->base_qindex,
+            oci->filter_level,
+            flags,
+            oci->mb_cols, oci->mb_rows);
+    vp9_blit_text(message, oci->post_proc_buffer.y_buffer,
+                  oci->post_proc_buffer.y_stride);
+  }
+
+  if (flags & VP9D_DEBUG_TXT_MBLK_MODES) {
+    int i, j;
+    uint8_t *y_ptr;
+    YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
+    int mb_rows = post->y_height >> 4;
+    int mb_cols = post->y_width  >> 4;
+    int mb_index = 0;
+    MODE_INFO *mi = oci->mi;
+
+    y_ptr = post->y_buffer + 4 * post->y_stride + 4;
+
+    /* vp9_filter each macro block */
+    for (i = 0; i < mb_rows; i++) {
+      for (j = 0; j < mb_cols; j++) {
+        char zz[4];
+
+        sprintf(zz, "%c", mi[mb_index].mbmi.mode + 'a');
+
+        vp9_blit_text(zz, y_ptr, post->y_stride);
+        mb_index++;
+        y_ptr += 16;
+      }
+
+      mb_index++; /* border */
+      y_ptr += post->y_stride  * 16 - post->y_width;
+
+    }
+  }
+
+  if (flags & VP9D_DEBUG_TXT_DC_DIFF) {
+    int i, j;
+    uint8_t *y_ptr;
+    YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
+    int mb_rows = post->y_height >> 4;
+    int mb_cols = post->y_width  >> 4;
+    int mb_index = 0;
+    MODE_INFO *mi = oci->mi;
+
+    y_ptr = post->y_buffer + 4 * post->y_stride + 4;
+
+    /* vp9_filter each macro block */
+    for (i = 0; i < mb_rows; i++) {
+      for (j = 0; j < mb_cols; j++) {
+        char zz[4];
+        int dc_diff = !(mi[mb_index].mbmi.mode != I4X4_PRED &&
+                        mi[mb_index].mbmi.mode != SPLITMV &&
+                        mi[mb_index].mbmi.mb_skip_coeff);
+
+        if (oci->frame_type == KEY_FRAME)
+          sprintf(zz, "a");
+        else
+          sprintf(zz, "%c", dc_diff + '0');
+
+        vp9_blit_text(zz, y_ptr, post->y_stride);
+        mb_index++;
+        y_ptr += 16;
+      }
+
+      mb_index++; /* border */
+      y_ptr += post->y_stride  * 16 - post->y_width;
+
+    }
+  }
+
+  if (flags & VP9D_DEBUG_TXT_RATE_INFO) {
+    char message[512];
+    snprintf(message, sizeof(message),
+             "Bitrate: %10.2f frame_rate: %10.2f ",
+             oci->bitrate, oci->framerate);
+    vp9_blit_text(message, oci->post_proc_buffer.y_buffer,
+                  oci->post_proc_buffer.y_stride);
+  }
+
+  /* Draw motion vectors */
+  if ((flags & VP9D_DEBUG_DRAW_MV) && ppflags->display_mv_flag) {
+    YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
+    int width  = post->y_width;
+    int height = post->y_height;
+    uint8_t *y_buffer = oci->post_proc_buffer.y_buffer;
+    int y_stride = oci->post_proc_buffer.y_stride;
+    MODE_INFO *mi = oci->mi;
+    int x0, y0;
+
+    for (y0 = 0; y0 < height; y0 += 16) {
+      for (x0 = 0; x0 < width; x0 += 16) {
+        int x1, y1;
+
+        if (!(ppflags->display_mv_flag & (1 << mi->mbmi.mode))) {
+          mi++;
+          continue;
+        }
+
+        if (mi->mbmi.mode == SPLITMV) {
+          switch (mi->mbmi.partitioning) {
+            case PARTITIONING_16X8 : {  /* mv_top_bottom */
+              union b_mode_info *bmi = &mi->bmi[0];
+              MV *mv = &bmi->mv.as_mv;
+
+              x1 = x0 + 8 + (mv->col >> 3);
+              y1 = y0 + 4 + (mv->row >> 3);
+
+              constrain_line(x0 + 8, &x1, y0 + 4, &y1, width, height);
+              vp9_blit_line(x0 + 8,  x1, y0 + 4,  y1, y_buffer, y_stride);
+
+              bmi = &mi->bmi[8];
+
+              x1 = x0 + 8 + (mv->col >> 3);
+              y1 = y0 + 12 + (mv->row >> 3);
+
+              constrain_line(x0 + 8, &x1, y0 + 12, &y1, width, height);
+              vp9_blit_line(x0 + 8,  x1, y0 + 12,  y1, y_buffer, y_stride);
+
+              break;
+            }
+            case PARTITIONING_8X16 : {  /* mv_left_right */
+              union b_mode_info *bmi = &mi->bmi[0];
+              MV *mv = &bmi->mv.as_mv;
+
+              x1 = x0 + 4 + (mv->col >> 3);
+              y1 = y0 + 8 + (mv->row >> 3);
+
+              constrain_line(x0 + 4, &x1, y0 + 8, &y1, width, height);
+              vp9_blit_line(x0 + 4,  x1, y0 + 8,  y1, y_buffer, y_stride);
+
+              bmi = &mi->bmi[2];
+
+              x1 = x0 + 12 + (mv->col >> 3);
+              y1 = y0 + 8 + (mv->row >> 3);
+
+              constrain_line(x0 + 12, &x1, y0 + 8, &y1, width, height);
+              vp9_blit_line(x0 + 12,  x1, y0 + 8,  y1, y_buffer, y_stride);
+
+              break;
+            }
+            case PARTITIONING_8X8 : {  /* mv_quarters   */
+              union b_mode_info *bmi = &mi->bmi[0];
+              MV *mv = &bmi->mv.as_mv;
+
+              x1 = x0 + 4 + (mv->col >> 3);
+              y1 = y0 + 4 + (mv->row >> 3);
+
+              constrain_line(x0 + 4, &x1, y0 + 4, &y1, width, height);
+              vp9_blit_line(x0 + 4,  x1, y0 + 4,  y1, y_buffer, y_stride);
+
+              bmi = &mi->bmi[2];
+
+              x1 = x0 + 12 + (mv->col >> 3);
+              y1 = y0 + 4 + (mv->row >> 3);
+
+              constrain_line(x0 + 12, &x1, y0 + 4, &y1, width, height);
+              vp9_blit_line(x0 + 12,  x1, y0 + 4,  y1, y_buffer, y_stride);
+
+              bmi = &mi->bmi[8];
+
+              x1 = x0 + 4 + (mv->col >> 3);
+              y1 = y0 + 12 + (mv->row >> 3);
+
+              constrain_line(x0 + 4, &x1, y0 + 12, &y1, width, height);
+              vp9_blit_line(x0 + 4,  x1, y0 + 12,  y1, y_buffer, y_stride);
+
+              bmi = &mi->bmi[10];
+
+              x1 = x0 + 12 + (mv->col >> 3);
+              y1 = y0 + 12 + (mv->row >> 3);
+
+              constrain_line(x0 + 12, &x1, y0 + 12, &y1, width, height);
+              vp9_blit_line(x0 + 12,  x1, y0 + 12,  y1, y_buffer, y_stride);
+              break;
+            }
+            case PARTITIONING_4X4:
+            default : {
+              union b_mode_info *bmi = mi->bmi;
+              int bx0, by0;
+
+              for (by0 = y0; by0 < (y0 + 16); by0 += 4) {
+                for (bx0 = x0; bx0 < (x0 + 16); bx0 += 4) {
+                  MV *mv = &bmi->mv.as_mv;
+
+                  x1 = bx0 + 2 + (mv->col >> 3);
+                  y1 = by0 + 2 + (mv->row >> 3);
+
+                  constrain_line(bx0 + 2, &x1, by0 + 2, &y1, width, height);
+                  vp9_blit_line(bx0 + 2,  x1, by0 + 2,  y1, y_buffer, y_stride);
+
+                  bmi++;
+                }
+              }
+            }
+          }
+        } else if (mi->mbmi.mode >= NEARESTMV) {
+          MV *mv = &mi->mbmi.mv.as_mv;
+          const int lx0 = x0 + 8;
+          const int ly0 = y0 + 8;
+
+          x1 = lx0 + (mv->col >> 3);
+          y1 = ly0 + (mv->row >> 3);
+
+          if (x1 != lx0 && y1 != ly0) {
+            constrain_line(lx0, &x1, ly0 - 1, &y1, width, height);
+            vp9_blit_line(lx0,  x1, ly0 - 1,  y1, y_buffer, y_stride);
+
+            constrain_line(lx0, &x1, ly0 + 1, &y1, width, height);
+            vp9_blit_line(lx0,  x1, ly0 + 1,  y1, y_buffer, y_stride);
+          } else
+            vp9_blit_line(lx0,  x1, ly0,  y1, y_buffer, y_stride);
+        }
+
+        mi++;
+      }
+      mi++;
+    }
+  }
+
+  /* Color in block modes */
+  if ((flags & VP9D_DEBUG_CLR_BLK_MODES)
+      && (ppflags->display_mb_modes_flag || ppflags->display_b_modes_flag)) {
+    int y, x;
+    YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
+    int width  = post->y_width;
+    int height = post->y_height;
+    uint8_t *y_ptr = oci->post_proc_buffer.y_buffer;
+    uint8_t *u_ptr = oci->post_proc_buffer.u_buffer;
+    uint8_t *v_ptr = oci->post_proc_buffer.v_buffer;
+    int y_stride = oci->post_proc_buffer.y_stride;
+    MODE_INFO *mi = oci->mi;
+
+    for (y = 0; y < height; y += 16) {
+      for (x = 0; x < width; x += 16) {
+        int Y = 0, U = 0, V = 0;
+
+        if (mi->mbmi.mode == I4X4_PRED &&
+            ((ppflags->display_mb_modes_flag & I4X4_PRED) ||
+             ppflags->display_b_modes_flag)) {
+          int by, bx;
+          uint8_t *yl, *ul, *vl;
+          union b_mode_info *bmi = mi->bmi;
+
+          yl = y_ptr + x;
+          ul = u_ptr + (x >> 1);
+          vl = v_ptr + (x >> 1);
+
+          for (by = 0; by < 16; by += 4) {
+            for (bx = 0; bx < 16; bx += 4) {
+              if ((ppflags->display_b_modes_flag & (1 << mi->mbmi.mode))
+                  || (ppflags->display_mb_modes_flag & I4X4_PRED)) {
+                Y = B_PREDICTION_MODE_colors[bmi->as_mode.first][0];
+                U = B_PREDICTION_MODE_colors[bmi->as_mode.first][1];
+                V = B_PREDICTION_MODE_colors[bmi->as_mode.first][2];
+
+                vp9_blend_b(yl + bx, ul + (bx >> 1), vl + (bx >> 1), Y, U, V,
+                    0xc000, y_stride);
+              }
+              bmi++;
+            }
+
+            yl += y_stride * 4;
+            ul += y_stride * 1;
+            vl += y_stride * 1;
+          }
+        } else if (ppflags->display_mb_modes_flag & (1 << mi->mbmi.mode)) {
+          Y = MB_PREDICTION_MODE_colors[mi->mbmi.mode][0];
+          U = MB_PREDICTION_MODE_colors[mi->mbmi.mode][1];
+          V = MB_PREDICTION_MODE_colors[mi->mbmi.mode][2];
+
+          vp9_blend_mb_inner(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
+                             Y, U, V, 0xc000, y_stride);
+        }
+
+        mi++;
+      }
+      y_ptr += y_stride * 16;
+      u_ptr += y_stride * 4;
+      v_ptr += y_stride * 4;
+
+      mi++;
+    }
+  }
+
+  /* Color in frame reference blocks */
+  if ((flags & VP9D_DEBUG_CLR_FRM_REF_BLKS) &&
+      ppflags->display_ref_frame_flag) {
+    int y, x;
+    YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
+    int width  = post->y_width;
+    int height = post->y_height;
+    uint8_t *y_ptr = oci->post_proc_buffer.y_buffer;
+    uint8_t *u_ptr = oci->post_proc_buffer.u_buffer;
+    uint8_t *v_ptr = oci->post_proc_buffer.v_buffer;
+    int y_stride = oci->post_proc_buffer.y_stride;
+    MODE_INFO *mi = oci->mi;
+
+    for (y = 0; y < height; y += 16) {
+      for (x = 0; x < width; x += 16) {
+        int Y = 0, U = 0, V = 0;
+
+        if (ppflags->display_ref_frame_flag & (1 << mi->mbmi.ref_frame)) {
+          Y = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][0];
+          U = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][1];
+          V = MV_REFERENCE_FRAME_colors[mi->mbmi.ref_frame][2];
+
+          vp9_blend_mb_outer(y_ptr + x, u_ptr + (x >> 1), v_ptr + (x >> 1),
+                             Y, U, V, 0xc000, y_stride);
+        }
+
+        mi++;
+      }
+      y_ptr += y_stride * 16;
+      u_ptr += y_stride * 4;
+      v_ptr += y_stride * 4;
+
+      mi++;
+    }
+  }
+#endif
+
+  *dest = oci->post_proc_buffer;
+
+  /* handle problem with extending borders */
+  dest->y_width = oci->width;
+  dest->y_height = oci->height;
+  dest->uv_height = dest->y_height / 2;
+
+  return 0;
+}
diff --git a/libvpx/vp9/common/vp9_postproc.h b/libvpx/vp9/common/vp9_postproc.h
new file mode 100644
index 0000000..2c0d333
--- /dev/null
+++ b/libvpx/vp9/common/vp9_postproc.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_VP9_POSTPROC_H_
+#define VP9_COMMON_VP9_POSTPROC_H_
+
+#include "vpx_ports/mem.h"
+
+struct postproc_state {
+  int last_q;
+  int last_noise;
+  char noise[3072];
+  DECLARE_ALIGNED(16, char, blackclamp[16]);
+  DECLARE_ALIGNED(16, char, whiteclamp[16]);
+  DECLARE_ALIGNED(16, char, bothclamp[16]);
+};
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_ppflags.h"
+
+int vp9_post_proc_frame(struct VP9Common *oci, YV12_BUFFER_CONFIG *dest,
+                        vp9_ppflags_t *flags);
+
+void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, int q);
+
+void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, int q);
+
+#endif  // VP9_COMMON_VP9_POSTPROC_H_
diff --git a/libvpx/vp9/common/vp9_ppflags.h b/libvpx/vp9/common/vp9_ppflags.h
new file mode 100644
index 0000000..561c930
--- /dev/null
+++ b/libvpx/vp9/common/vp9_ppflags.h
@@ -0,0 +1,38 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_PPFLAGS_H_
+#define VP9_COMMON_VP9_PPFLAGS_H_
+
+enum {
+  VP9D_NOFILTERING            = 0,
+  VP9D_DEBLOCK                = 1 << 0,
+  VP9D_DEMACROBLOCK           = 1 << 1,
+  VP9D_ADDNOISE               = 1 << 2,
+  VP9D_DEBUG_TXT_FRAME_INFO   = 1 << 3,
+  VP9D_DEBUG_TXT_MBLK_MODES   = 1 << 4,
+  VP9D_DEBUG_TXT_DC_DIFF      = 1 << 5,
+  VP9D_DEBUG_TXT_RATE_INFO    = 1 << 6,
+  VP9D_DEBUG_DRAW_MV          = 1 << 7,
+  VP9D_DEBUG_CLR_BLK_MODES    = 1 << 8,
+  VP9D_DEBUG_CLR_FRM_REF_BLKS = 1 << 9
+};
+
+typedef struct {
+  int post_proc_flag;
+  int deblocking_level;
+  int noise_level;
+  int display_ref_frame_flag;
+  int display_mb_modes_flag;
+  int display_b_modes_flag;
+  int display_mv_flag;
+} vp9_ppflags_t;
+
+#endif  // VP9_COMMON_VP9_PPFLAGS_H_
diff --git a/libvpx/vp9/common/vp9_pragmas.h b/libvpx/vp9/common/vp9_pragmas.h
new file mode 100644
index 0000000..f079161
--- /dev/null
+++ b/libvpx/vp9/common/vp9_pragmas.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_PRAGMAS_H_
+#define VP9_COMMON_VP9_PRAGMAS_H_
+
+#ifdef __INTEL_COMPILER
+#pragma warning(disable:997 1011 170)
+#endif
+
+#ifdef _MSC_VER
+#pragma warning(disable:4799)
+#endif
+
+#endif  // VP9_COMMON_VP9_PRAGMAS_H_
diff --git a/libvpx/vp9/common/vp9_pred_common.c b/libvpx/vp9/common/vp9_pred_common.c
new file mode 100644
index 0000000..17da4f2
--- /dev/null
+++ b/libvpx/vp9/common/vp9_pred_common.c
@@ -0,0 +1,520 @@
+
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <limits.h>
+
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_treecoder.h"
+
+// TBD prediction functions for various bitstream signals
+
+// Returns a context number for the given MB prediction signal
+unsigned char vp9_get_pred_context(const VP9_COMMON *const cm,
+                                   const MACROBLOCKD *const xd,
+                                   PRED_ID pred_id) {
+  int pred_context;
+  const MODE_INFO *const mi = xd->mode_info_context;
+  const MODE_INFO *const above_mi = mi - cm->mode_info_stride;
+  const MODE_INFO *const left_mi = mi - 1;
+  const int left_in_image = xd->left_available && left_mi->mbmi.mb_in_image;
+  const int above_in_image = xd->up_available && above_mi->mbmi.mb_in_image;
+  // Note:
+  // The mode info data structure has a one element border above and to the
+  // left of the entries correpsonding to real macroblocks.
+  // The prediction flags in these dummy entries are initialised to 0.
+  switch (pred_id) {
+    case PRED_SEG_ID:
+      pred_context = above_mi->mbmi.seg_id_predicted;
+      if (xd->left_available)
+        pred_context += left_mi->mbmi.seg_id_predicted;
+      break;
+
+    case PRED_MBSKIP:
+      pred_context = above_mi->mbmi.mb_skip_coeff;
+      if (xd->left_available)
+        pred_context += left_mi->mbmi.mb_skip_coeff;
+      break;
+
+    case PRED_SWITCHABLE_INTERP: {
+      // left
+      const int left_mv_pred = is_inter_mode(left_mi->mbmi.mode);
+      const int left_interp = left_in_image && left_mv_pred ?
+                    vp9_switchable_interp_map[left_mi->mbmi.interp_filter] :
+                    VP9_SWITCHABLE_FILTERS;
+
+      // above
+      const int above_mv_pred = is_inter_mode(above_mi->mbmi.mode);
+      const int above_interp = above_in_image && above_mv_pred ?
+                    vp9_switchable_interp_map[above_mi->mbmi.interp_filter] :
+                    VP9_SWITCHABLE_FILTERS;
+
+      assert(left_interp != -1);
+      assert(above_interp != -1);
+
+      if (left_interp == above_interp)
+        pred_context = left_interp;
+      else if (left_interp == VP9_SWITCHABLE_FILTERS &&
+               above_interp != VP9_SWITCHABLE_FILTERS)
+         pred_context = above_interp;
+      else if (left_interp != VP9_SWITCHABLE_FILTERS &&
+               above_interp == VP9_SWITCHABLE_FILTERS)
+        pred_context = left_interp;
+      else
+        pred_context = VP9_SWITCHABLE_FILTERS;
+
+      break;
+    }
+
+    case PRED_INTRA_INTER: {
+      if (above_in_image && left_in_image) {  // both edges available
+        if (left_mi->mbmi.ref_frame[0] == INTRA_FRAME &&
+            above_mi->mbmi.ref_frame[0] == INTRA_FRAME) {  // intra/intra (3)
+          pred_context = 3;
+        } else {  // intra/inter (1) or inter/inter (0)
+          pred_context = left_mi->mbmi.ref_frame[0] == INTRA_FRAME ||
+                         above_mi->mbmi.ref_frame[0] == INTRA_FRAME;
+        }
+      } else if (above_in_image || left_in_image) {  // one edge available
+        const MODE_INFO *edge = above_in_image ? above_mi : left_mi;
+
+        // inter: 0, intra: 2
+        pred_context = 2 * (edge->mbmi.ref_frame[0] == INTRA_FRAME);
+      } else {
+        pred_context = 0;
+      }
+      assert(pred_context >= 0 && pred_context < INTRA_INTER_CONTEXTS);
+      break;
+    }
+
+    case PRED_COMP_INTER_INTER: {
+      if (above_in_image && left_in_image) {  // both edges available
+        if (above_mi->mbmi.ref_frame[1] <= INTRA_FRAME &&
+            left_mi->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          // neither edge uses comp pred (0/1)
+          pred_context = ((above_mi->mbmi.ref_frame[0] == cm->comp_fixed_ref) ^
+                          (left_mi->mbmi.ref_frame[0] == cm->comp_fixed_ref));
+        } else if (above_mi->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          // one of two edges uses comp pred (2/3)
+          pred_context = 2 +
+              (above_mi->mbmi.ref_frame[0] == cm->comp_fixed_ref ||
+               above_mi->mbmi.ref_frame[0] == INTRA_FRAME);
+        } else if (left_mi->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          // one of two edges uses comp pred (2/3)
+          pred_context = 2 +
+              (left_mi->mbmi.ref_frame[0] == cm->comp_fixed_ref ||
+               left_mi->mbmi.ref_frame[0] == INTRA_FRAME);
+        } else {  // both edges use comp pred (4)
+          pred_context = 4;
+        }
+      } else if (above_in_image || left_in_image) {  // one edge available
+        const MODE_INFO *edge = above_in_image ? above_mi : left_mi;
+
+        if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          // edge does not use comp pred (0/1)
+          pred_context = edge->mbmi.ref_frame[0] == cm->comp_fixed_ref;
+        } else {  // edge uses comp pred (3)
+          pred_context = 3;
+        }
+      } else {  // no edges available (1)
+        pred_context = 1;
+      }
+      assert(pred_context >= 0 && pred_context < COMP_INTER_CONTEXTS);
+      break;
+    }
+
+    case PRED_COMP_REF_P: {
+      const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
+      const int var_ref_idx = !fix_ref_idx;
+
+      if (above_in_image && left_in_image) {  // both edges available
+        if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME &&
+            left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {  // intra/intra (2)
+          pred_context = 2;
+        } else if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME ||
+                   left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {  // intra/inter
+          const MODE_INFO *edge = above_mi->mbmi.ref_frame[0] == INTRA_FRAME ?
+                                  left_mi : above_mi;
+
+          if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {  // single pred (1/3)
+            pred_context = 1 +
+                2 * edge->mbmi.ref_frame[0] != cm->comp_var_ref[1];
+          } else {  // comp pred (1/3)
+            pred_context = 1 +
+                2 * edge->mbmi.ref_frame[var_ref_idx] != cm->comp_var_ref[1];
+          }
+        } else {  // inter/inter
+          int l_sg = left_mi->mbmi.ref_frame[1] <= INTRA_FRAME;
+          int a_sg = above_mi->mbmi.ref_frame[1] <= INTRA_FRAME;
+          MV_REFERENCE_FRAME vrfa = a_sg ? above_mi->mbmi.ref_frame[0] :
+              above_mi->mbmi.ref_frame[var_ref_idx];
+          MV_REFERENCE_FRAME vrfl = l_sg ? left_mi->mbmi.ref_frame[0] :
+              left_mi->mbmi.ref_frame[var_ref_idx];
+
+          if (vrfa == vrfl && cm->comp_var_ref[1] == vrfa) {
+            pred_context = 0;
+          } else if (l_sg && a_sg) {  // single/single
+            if ((vrfa == cm->comp_fixed_ref && vrfl == cm->comp_var_ref[0]) ||
+                (vrfl == cm->comp_fixed_ref && vrfa == cm->comp_var_ref[0])) {
+              pred_context = 4;
+            } else if (vrfa == vrfl) {
+              pred_context = 3;
+            } else {
+              pred_context = 1;
+            }
+          } else if (l_sg || a_sg) {  // single/comp
+            MV_REFERENCE_FRAME vrfc = l_sg ? vrfa : vrfl;
+            MV_REFERENCE_FRAME rfs = a_sg ? vrfa : vrfl;
+
+            if (vrfc == cm->comp_var_ref[1] && rfs != cm->comp_var_ref[1]) {
+              pred_context = 1;
+            } else if (rfs == cm->comp_var_ref[1] &&
+                       vrfc != cm->comp_var_ref[1]) {
+              pred_context = 2;
+            } else {
+              pred_context = 4;
+            }
+          } else if (vrfa == vrfl) {  // comp/comp
+            pred_context = 4;
+          } else {
+            pred_context = 2;
+          }
+        }
+      } else if (above_in_image || left_in_image) {  // one edge available
+        const MODE_INFO *edge = above_in_image ? above_mi : left_mi;
+
+        if (edge->mbmi.ref_frame[0] == INTRA_FRAME) {
+          pred_context = 2;
+        } else if (edge->mbmi.ref_frame[1] > INTRA_FRAME) {
+          pred_context =
+              4 * edge->mbmi.ref_frame[var_ref_idx] != cm->comp_var_ref[1];
+        } else {
+          pred_context = 3 * edge->mbmi.ref_frame[0] != cm->comp_var_ref[1];
+        }
+      } else {  // no edges available (2)
+        pred_context = 2;
+      }
+      assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
+      break;
+    }
+
+    case PRED_SINGLE_REF_P1: {
+      if (above_in_image && left_in_image) {  // both edges available
+        if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME &&
+            left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
+          pred_context = 2;
+        } else if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME ||
+                   left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
+          const MODE_INFO *edge = above_mi->mbmi.ref_frame[0] == INTRA_FRAME ?
+                                  left_mi : above_mi;
+
+          if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {
+            pred_context = 4 * (edge->mbmi.ref_frame[0] == LAST_FRAME);
+          } else {
+            pred_context = 1 + (edge->mbmi.ref_frame[0] == LAST_FRAME ||
+                                edge->mbmi.ref_frame[1] == LAST_FRAME);
+          }
+        } else if (above_mi->mbmi.ref_frame[1] <= INTRA_FRAME &&
+                   left_mi->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          pred_context = 2 * (above_mi->mbmi.ref_frame[0] == LAST_FRAME) +
+                         2 * (left_mi->mbmi.ref_frame[0] == LAST_FRAME);
+        } else if (above_mi->mbmi.ref_frame[1] > INTRA_FRAME &&
+                   left_mi->mbmi.ref_frame[1] > INTRA_FRAME) {
+          pred_context = 1 + (above_mi->mbmi.ref_frame[0] == LAST_FRAME ||
+                              above_mi->mbmi.ref_frame[1] == LAST_FRAME ||
+                              left_mi->mbmi.ref_frame[0] == LAST_FRAME ||
+                              left_mi->mbmi.ref_frame[1] == LAST_FRAME);
+        } else {
+          MV_REFERENCE_FRAME rfs = above_mi->mbmi.ref_frame[1] <= INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[0] : left_mi->mbmi.ref_frame[0];
+          MV_REFERENCE_FRAME crf1 = above_mi->mbmi.ref_frame[1] > INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[0] : left_mi->mbmi.ref_frame[0];
+          MV_REFERENCE_FRAME crf2 = above_mi->mbmi.ref_frame[1] > INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[1] : left_mi->mbmi.ref_frame[1];
+
+          if (rfs == LAST_FRAME) {
+            pred_context = 3 + (crf1 == LAST_FRAME || crf2 == LAST_FRAME);
+          } else {
+            pred_context = crf1 == LAST_FRAME || crf2 == LAST_FRAME;
+          }
+        }
+      } else if (above_in_image || left_in_image) {  // one edge available
+        const MODE_INFO *edge = above_in_image ? above_mi : left_mi;
+
+        if (edge->mbmi.ref_frame[0] == INTRA_FRAME) {
+          pred_context = 2;
+        } else if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          pred_context = 4 * (edge->mbmi.ref_frame[0] == LAST_FRAME);
+        } else {
+          pred_context = 1 + (edge->mbmi.ref_frame[0] == LAST_FRAME ||
+                              edge->mbmi.ref_frame[1] == LAST_FRAME);
+        }
+      } else {  // no edges available (2)
+        pred_context = 2;
+      }
+      assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
+      break;
+    }
+
+    case PRED_SINGLE_REF_P2: {
+      if (above_in_image && left_in_image) {  // both edges available
+        if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME &&
+            left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
+          pred_context = 2;
+        } else if (above_mi->mbmi.ref_frame[0] == INTRA_FRAME ||
+                   left_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
+          const MODE_INFO *edge = above_mi->mbmi.ref_frame[0] == INTRA_FRAME ?
+                                  left_mi : above_mi;
+
+          if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {
+            if (edge->mbmi.ref_frame[0] == LAST_FRAME) {
+              pred_context = 3;
+            } else {
+              pred_context = 4 * (edge->mbmi.ref_frame[0] == GOLDEN_FRAME);
+            }
+          } else {
+            pred_context = 1 + 2 * (edge->mbmi.ref_frame[0] == GOLDEN_FRAME ||
+                                    edge->mbmi.ref_frame[1] == GOLDEN_FRAME);
+          }
+        } else if (above_mi->mbmi.ref_frame[1] <= INTRA_FRAME &&
+                   left_mi->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          if (above_mi->mbmi.ref_frame[0] == LAST_FRAME &&
+              left_mi->mbmi.ref_frame[0] == LAST_FRAME) {
+            pred_context = 3;
+          } else if (above_mi->mbmi.ref_frame[0] == LAST_FRAME ||
+                     left_mi->mbmi.ref_frame[0] == LAST_FRAME) {
+            const MODE_INFO *edge = above_mi->mbmi.ref_frame[0] == LAST_FRAME ?
+                                    left_mi : above_mi;
+
+            pred_context = 4 * (edge->mbmi.ref_frame[0] == GOLDEN_FRAME);
+          } else {
+            pred_context = 2 * (above_mi->mbmi.ref_frame[0] == GOLDEN_FRAME) +
+                           2 * (left_mi->mbmi.ref_frame[0] == GOLDEN_FRAME);
+          }
+        } else if (above_mi->mbmi.ref_frame[1] > INTRA_FRAME &&
+                   left_mi->mbmi.ref_frame[1] > INTRA_FRAME) {
+          if (above_mi->mbmi.ref_frame[0] == left_mi->mbmi.ref_frame[0] &&
+              above_mi->mbmi.ref_frame[1] == left_mi->mbmi.ref_frame[1]) {
+            pred_context = 3 * (above_mi->mbmi.ref_frame[0] == GOLDEN_FRAME ||
+                                above_mi->mbmi.ref_frame[1] == GOLDEN_FRAME ||
+                                left_mi->mbmi.ref_frame[0] == GOLDEN_FRAME ||
+                                left_mi->mbmi.ref_frame[1] == GOLDEN_FRAME);
+          } else {
+            pred_context = 2;
+          }
+        } else {
+          MV_REFERENCE_FRAME rfs = above_mi->mbmi.ref_frame[1] <= INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[0] : left_mi->mbmi.ref_frame[0];
+          MV_REFERENCE_FRAME crf1 = above_mi->mbmi.ref_frame[1] > INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[0] : left_mi->mbmi.ref_frame[0];
+          MV_REFERENCE_FRAME crf2 = above_mi->mbmi.ref_frame[1] > INTRA_FRAME ?
+              above_mi->mbmi.ref_frame[1] : left_mi->mbmi.ref_frame[1];
+
+          if (rfs == GOLDEN_FRAME) {
+            pred_context = 3 + (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME);
+          } else if (rfs == ALTREF_FRAME) {
+            pred_context = crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME;
+          } else {
+            pred_context =
+                1 + 2 * (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME);
+          }
+        }
+      } else if (above_in_image || left_in_image) {  // one edge available
+        const MODE_INFO *edge = above_in_image ? above_mi : left_mi;
+
+        if (edge->mbmi.ref_frame[0] == INTRA_FRAME ||
+            (edge->mbmi.ref_frame[0] == LAST_FRAME &&
+             edge->mbmi.ref_frame[1] <= INTRA_FRAME)) {
+          pred_context = 2;
+        } else if (edge->mbmi.ref_frame[1] <= INTRA_FRAME) {
+          pred_context = 4 * (edge->mbmi.ref_frame[0] == GOLDEN_FRAME);
+        } else {
+          pred_context = 3 * (edge->mbmi.ref_frame[0] == GOLDEN_FRAME ||
+                              edge->mbmi.ref_frame[1] == GOLDEN_FRAME);
+        }
+      } else {  // no edges available (2)
+        pred_context = 2;
+      }
+      assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
+      break;
+    }
+
+    case PRED_TX_SIZE: {
+      int above_context, left_context;
+      int max_tx_size;
+      if (mi->mbmi.sb_type < BLOCK_SIZE_SB8X8)
+        max_tx_size = TX_4X4;
+      else if (mi->mbmi.sb_type < BLOCK_SIZE_MB16X16)
+        max_tx_size = TX_8X8;
+      else if (mi->mbmi.sb_type < BLOCK_SIZE_SB32X32)
+        max_tx_size = TX_16X16;
+      else
+        max_tx_size = TX_32X32;
+      above_context = left_context = max_tx_size;
+      if (above_in_image) {
+        above_context = (above_mi->mbmi.mb_skip_coeff ?
+                         max_tx_size : above_mi->mbmi.txfm_size);
+      }
+      if (left_in_image) {
+        left_context = (left_mi->mbmi.mb_skip_coeff ?
+                        max_tx_size : left_mi->mbmi.txfm_size);
+      }
+      if (!left_in_image) {
+        left_context = above_context;
+      }
+      if (!above_in_image) {
+        above_context = left_context;
+      }
+      pred_context = (above_context + left_context > max_tx_size);
+      break;
+    }
+
+    default:
+      assert(0);
+      pred_context = 0;  // *** add error trap code.
+      break;
+  }
+
+  return pred_context;
+}
+
+// This function returns a context probability for coding a given
+// prediction signal
+vp9_prob vp9_get_pred_prob(const VP9_COMMON *const cm,
+                          const MACROBLOCKD *const xd,
+                          PRED_ID pred_id) {
+  const int pred_context = vp9_get_pred_context(cm, xd, pred_id);
+
+  switch (pred_id) {
+    case PRED_SEG_ID:
+      return cm->segment_pred_probs[pred_context];
+    case PRED_MBSKIP:
+      return cm->fc.mbskip_probs[pred_context];
+    case PRED_INTRA_INTER:
+      return cm->fc.intra_inter_prob[pred_context];
+    case PRED_COMP_INTER_INTER:
+      return cm->fc.comp_inter_prob[pred_context];
+    case PRED_COMP_REF_P:
+      return cm->fc.comp_ref_prob[pred_context];
+    case PRED_SINGLE_REF_P1:
+      return cm->fc.single_ref_prob[pred_context][0];
+    case PRED_SINGLE_REF_P2:
+      return cm->fc.single_ref_prob[pred_context][1];
+    default:
+      assert(0);
+      return 128;  // *** add error trap code.
+  }
+}
+
+// This function returns a context probability ptr for coding a given
+// prediction signal
+const vp9_prob *vp9_get_pred_probs(const VP9_COMMON *const cm,
+                                   const MACROBLOCKD *const xd,
+                                   PRED_ID pred_id) {
+  const MODE_INFO *const mi = xd->mode_info_context;
+  const int pred_context = vp9_get_pred_context(cm, xd, pred_id);
+
+  switch (pred_id) {
+    case PRED_SWITCHABLE_INTERP:
+      return &cm->fc.switchable_interp_prob[pred_context][0];
+
+    case PRED_TX_SIZE:
+      if (mi->mbmi.sb_type < BLOCK_SIZE_MB16X16)
+        return cm->fc.tx_probs_8x8p[pred_context];
+      else if (mi->mbmi.sb_type < BLOCK_SIZE_SB32X32)
+        return cm->fc.tx_probs_16x16p[pred_context];
+      else
+        return cm->fc.tx_probs_32x32p[pred_context];
+
+    default:
+      assert(0);
+      return NULL;  // *** add error trap code.
+  }
+}
+
+// This function returns the status of the given prediction signal.
+// I.e. is the predicted value for the given signal correct.
+unsigned char vp9_get_pred_flag(const MACROBLOCKD *const xd,
+                                PRED_ID pred_id) {
+  switch (pred_id) {
+    case PRED_SEG_ID:
+      return xd->mode_info_context->mbmi.seg_id_predicted;
+    case PRED_MBSKIP:
+      return xd->mode_info_context->mbmi.mb_skip_coeff;
+    default:
+      assert(0);
+      return 0;  // *** add error trap code.
+  }
+}
+
+// This function sets the status of the given prediction signal.
+// I.e. is the predicted value for the given signal correct.
+void vp9_set_pred_flag(MACROBLOCKD *const xd,
+                       PRED_ID pred_id,
+                       unsigned char pred_flag) {
+  const int mis = xd->mode_info_stride;
+  BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+  const int bh = 1 << mi_height_log2(bsize);
+  const int bw = 1 << mi_width_log2(bsize);
+#define sub(a, b) (b) < 0 ? (a) + (b) : (a)
+  const int x_mis = sub(bw, xd->mb_to_right_edge >> (3 + LOG2_MI_SIZE));
+  const int y_mis = sub(bh, xd->mb_to_bottom_edge >> (3 + LOG2_MI_SIZE));
+#undef sub
+  int x, y;
+
+  switch (pred_id) {
+    case PRED_SEG_ID:
+      for (y = 0; y < y_mis; y++) {
+        for (x = 0; x < x_mis; x++) {
+          xd->mode_info_context[y * mis + x].mbmi.seg_id_predicted = pred_flag;
+        }
+      }
+      break;
+
+    case PRED_MBSKIP:
+      for (y = 0; y < y_mis; y++) {
+        for (x = 0; x < x_mis; x++) {
+          xd->mode_info_context[y * mis + x].mbmi.mb_skip_coeff = pred_flag;
+        }
+      }
+      break;
+
+    default:
+      assert(0);
+      // *** add error trap code.
+      break;
+  }
+}
+
+
+// The following contain the guts of the prediction code used to
+// peredict various bitstream signals.
+
+// Macroblock segment id prediction function
+int vp9_get_pred_mi_segid(VP9_COMMON *cm, BLOCK_SIZE_TYPE sb_type,
+                          int mi_row, int mi_col) {
+  const int mi_index = mi_row * cm->mi_cols + mi_col;
+  const int bw = 1 << mi_width_log2(sb_type);
+  const int bh = 1 << mi_height_log2(sb_type);
+  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = MIN(cm->mi_cols - mi_col, bw);
+  int segment_id = INT_MAX;
+  int x, y;
+
+  for (y = 0; y < ymis; y++) {
+    for (x = 0; x < xmis; x++) {
+      const int index = mi_index + (y * cm->mi_cols + x);
+      segment_id = MIN(segment_id, cm->last_frame_seg_map[index]);
+    }
+  }
+  return segment_id;
+}
diff --git a/libvpx/vp9/common/vp9_pred_common.h b/libvpx/vp9/common/vp9_pred_common.h
new file mode 100644
index 0000000..b728724
--- /dev/null
+++ b/libvpx/vp9/common/vp9_pred_common.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_PRED_COMMON_H_
+#define VP9_COMMON_VP9_PRED_COMMON_H_
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+// Predicted items
+typedef enum {
+  PRED_SEG_ID = 0,  // Segment identifier
+  PRED_MBSKIP = 1,
+  PRED_SWITCHABLE_INTERP = 2,
+  PRED_INTRA_INTER = 3,
+  PRED_COMP_INTER_INTER = 4,
+  PRED_SINGLE_REF_P1 = 5,
+  PRED_SINGLE_REF_P2 = 6,
+  PRED_COMP_REF_P = 7,
+  PRED_TX_SIZE = 8
+} PRED_ID;
+
+unsigned char vp9_get_pred_context(const VP9_COMMON *const cm,
+                                   const MACROBLOCKD *const xd,
+                                   PRED_ID pred_id);
+
+vp9_prob vp9_get_pred_prob(const VP9_COMMON *const cm,
+                           const MACROBLOCKD *const xd,
+                           PRED_ID pred_id);
+
+const vp9_prob *vp9_get_pred_probs(const VP9_COMMON *const cm,
+                                   const MACROBLOCKD *const xd,
+                                   PRED_ID pred_id);
+
+unsigned char vp9_get_pred_flag(const MACROBLOCKD *const xd,
+                                PRED_ID pred_id);
+
+void vp9_set_pred_flag(MACROBLOCKD *const xd,
+                       PRED_ID pred_id,
+                       unsigned char pred_flag);
+
+
+int vp9_get_pred_mi_segid(VP9_COMMON *cm, BLOCK_SIZE_TYPE sb_type,
+                          int mi_row, int mi_col);
+
+#endif  // VP9_COMMON_VP9_PRED_COMMON_H_
diff --git a/libvpx/vp9/common/vp9_quant_common.c b/libvpx/vp9/common/vp9_quant_common.c
new file mode 100644
index 0000000..295c8e7
--- /dev/null
+++ b/libvpx/vp9/common/vp9_quant_common.c
@@ -0,0 +1,69 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_seg_common.h"
+
+static int16_t dc_qlookup[QINDEX_RANGE];
+static int16_t ac_qlookup[QINDEX_RANGE];
+
+#define ACDC_MIN 8
+
+// TODO(dkovalev) move to common and reuse
+static double poly3(double a, double b, double c, double d, double x) {
+  return a*x*x*x + b*x*x + c*x + d;
+}
+
+void vp9_init_quant_tables() {
+  int i, val = 4;
+
+  // A "real" q of 1.0 forces lossless mode.
+  // In practice non lossless Q's between 1.0 and 2.0 (represented here by
+  // integer values from 5-7 give poor rd results (lower psnr and often
+  // larger size than the lossless encode. To block out those "not very useful"
+  // values we increment the ac and dc q lookup values by 4 after position 0.
+  ac_qlookup[0] = val;
+  dc_qlookup[0] = val;
+  val += 4;
+
+  for (i = 1; i < QINDEX_RANGE; i++) {
+    const int ac_val = val;
+
+    val = (int)(val * 1.01975);
+    if (val == ac_val)
+      ++val;
+
+    ac_qlookup[i] = (int16_t)ac_val;
+    dc_qlookup[i] = (int16_t)MAX(ACDC_MIN, poly3(0.000000305, -0.00065, 0.9,
+                                                 0.5, ac_val));
+  }
+}
+
+int16_t vp9_dc_quant(int qindex, int delta) {
+  return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
+}
+
+int16_t vp9_ac_quant(int qindex, int delta) {
+  return ac_qlookup[clamp(qindex + delta, 0, MAXQ)];
+}
+
+
+int vp9_get_qindex(MACROBLOCKD *xd, int segment_id, int base_qindex) {
+  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_ALT_Q)) {
+    const int data = vp9_get_segdata(xd, segment_id, SEG_LVL_ALT_Q);
+    return xd->mb_segment_abs_delta == SEGMENT_ABSDATA ?
+               data :  // Abs value
+               clamp(base_qindex + data, 0, MAXQ);  // Delta value
+  } else {
+    return base_qindex;
+  }
+}
+
diff --git a/libvpx/vp9/common/vp9_quant_common.h b/libvpx/vp9/common/vp9_quant_common.h
new file mode 100644
index 0000000..ded9426
--- /dev/null
+++ b/libvpx/vp9/common/vp9_quant_common.h
@@ -0,0 +1,28 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_QUANT_COMMON_H_
+#define VP9_COMMON_VP9_QUANT_COMMON_H_
+
+#include "vp9/common/vp9_blockd.h"
+
+#define MINQ 0
+#define MAXQ 255
+#define QINDEX_RANGE (MAXQ - MINQ + 1)
+#define QINDEX_BITS 8
+
+void vp9_init_quant_tables();
+
+int16_t vp9_dc_quant(int qindex, int delta);
+int16_t vp9_ac_quant(int qindex, int delta);
+
+int vp9_get_qindex(MACROBLOCKD *mb, int segment_id, int base_qindex);
+
+#endif  // VP9_COMMON_VP9_QUANT_COMMON_H_
diff --git a/libvpx/vp9/common/vp9_reconinter.c b/libvpx/vp9/common/vp9_reconinter.c
new file mode 100644
index 0000000..b28d333
--- /dev/null
+++ b/libvpx/vp9/common/vp9_reconinter.c
@@ -0,0 +1,528 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_filter.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/common/vp9_reconintra.h"
+
+static int scale_value_x_with_scaling(int val,
+                                      const struct scale_factors *scale) {
+  return (val * scale->x_scale_fp >> VP9_REF_SCALE_SHIFT);
+}
+
+static int scale_value_y_with_scaling(int val,
+                                      const struct scale_factors *scale) {
+  return (val * scale->y_scale_fp >> VP9_REF_SCALE_SHIFT);
+}
+
+static int unscaled_value(int val, const struct scale_factors *scale) {
+  (void) scale;
+  return val;
+}
+
+static int_mv32 mv_q3_to_q4_with_scaling(const int_mv *src_mv,
+                                         const struct scale_factors *scale) {
+  // returns mv * scale + offset
+  int_mv32 result;
+  const int32_t mv_row_q4 = src_mv->as_mv.row << 1;
+  const int32_t mv_col_q4 = src_mv->as_mv.col << 1;
+
+  result.as_mv.row = (mv_row_q4 * scale->y_scale_fp >> VP9_REF_SCALE_SHIFT)
+                      + scale->y_offset_q4;
+  result.as_mv.col = (mv_col_q4 * scale->x_scale_fp >> VP9_REF_SCALE_SHIFT)
+                      + scale->x_offset_q4;
+  return result;
+}
+
+static int_mv32 mv_q3_to_q4_without_scaling(const int_mv *src_mv,
+                                            const struct scale_factors *scale) {
+  // returns mv * scale + offset
+  int_mv32 result;
+
+  result.as_mv.row = src_mv->as_mv.row << 1;
+  result.as_mv.col = src_mv->as_mv.col << 1;
+  return result;
+}
+
+static int32_t mv_component_q4_with_scaling(int mv_q4, int scale_fp,
+                                            int offset_q4) {
+  int32_t scaled_mv;
+  // returns the scaled and offset value of the mv component.
+  scaled_mv = (mv_q4 * scale_fp >> VP9_REF_SCALE_SHIFT) + offset_q4;
+
+  return scaled_mv;
+}
+
+static int32_t mv_component_q4_without_scaling(int mv_q4, int scale_fp,
+                                               int offset_q4) {
+  // returns the scaled and offset value of the mv component.
+  (void)scale_fp;
+  (void)offset_q4;
+  return mv_q4;
+}
+
+static void set_offsets_with_scaling(struct scale_factors *scale,
+                                     int row, int col) {
+  const int x_q4 = 16 * col;
+  const int y_q4 = 16 * row;
+
+  scale->x_offset_q4 = (x_q4 * scale->x_scale_fp >> VP9_REF_SCALE_SHIFT) & 0xf;
+  scale->y_offset_q4 = (y_q4 * scale->y_scale_fp >> VP9_REF_SCALE_SHIFT) & 0xf;
+}
+
+static void set_offsets_without_scaling(struct scale_factors *scale,
+                                        int row, int col) {
+  scale->x_offset_q4 = 0;
+  scale->y_offset_q4 = 0;
+}
+
+static int get_fixed_point_scale_factor(int other_size, int this_size) {
+  // Calculate scaling factor once for each reference frame
+  // and use fixed point scaling factors in decoding and encoding routines.
+  // Hardware implementations can calculate scale factor in device driver
+  // and use multiplication and shifting on hardware instead of division.
+  return (other_size << VP9_REF_SCALE_SHIFT) / this_size;
+}
+
+void vp9_setup_scale_factors_for_frame(struct scale_factors *scale,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h) {
+  scale->x_scale_fp = get_fixed_point_scale_factor(other_w, this_w);
+  scale->x_offset_q4 = 0;  // calculated per-mb
+  scale->x_step_q4 = (16 * scale->x_scale_fp >> VP9_REF_SCALE_SHIFT);
+
+  scale->y_scale_fp = get_fixed_point_scale_factor(other_h, this_h);
+  scale->y_offset_q4 = 0;  // calculated per-mb
+  scale->y_step_q4 = (16 * scale->y_scale_fp >> VP9_REF_SCALE_SHIFT);
+
+  if ((other_w == this_w) && (other_h == this_h)) {
+    scale->scale_value_x = unscaled_value;
+    scale->scale_value_y = unscaled_value;
+    scale->set_scaled_offsets = set_offsets_without_scaling;
+    scale->scale_mv_q3_to_q4 = mv_q3_to_q4_without_scaling;
+    scale->scale_mv_component_q4 = mv_component_q4_without_scaling;
+  } else {
+    scale->scale_value_x = scale_value_x_with_scaling;
+    scale->scale_value_y = scale_value_y_with_scaling;
+    scale->set_scaled_offsets = set_offsets_with_scaling;
+    scale->scale_mv_q3_to_q4 = mv_q3_to_q4_with_scaling;
+    scale->scale_mv_component_q4 = mv_component_q4_with_scaling;
+  }
+
+  // TODO(agrange): Investigate the best choice of functions to use here
+  // for EIGHTTAP_SMOOTH. Since it is not interpolating, need to choose what
+  // to do at full-pel offsets. The current selection, where the filter is
+  // applied in one direction only, and not at all for 0,0, seems to give the
+  // best quality, but it may be worth trying an additional mode that does
+  // do the filtering on full-pel.
+  if (scale->x_step_q4 == 16) {
+    if (scale->y_step_q4 == 16) {
+      // No scaling in either direction.
+      scale->predict[0][0][0] = vp9_convolve_copy;
+      scale->predict[0][0][1] = vp9_convolve_avg;
+      scale->predict[0][1][0] = vp9_convolve8_vert;
+      scale->predict[0][1][1] = vp9_convolve8_avg_vert;
+      scale->predict[1][0][0] = vp9_convolve8_horiz;
+      scale->predict[1][0][1] = vp9_convolve8_avg_horiz;
+    } else {
+      // No scaling in x direction. Must always scale in the y direction.
+      scale->predict[0][0][0] = vp9_convolve8_vert;
+      scale->predict[0][0][1] = vp9_convolve8_avg_vert;
+      scale->predict[0][1][0] = vp9_convolve8_vert;
+      scale->predict[0][1][1] = vp9_convolve8_avg_vert;
+      scale->predict[1][0][0] = vp9_convolve8;
+      scale->predict[1][0][1] = vp9_convolve8_avg;
+    }
+  } else {
+    if (scale->y_step_q4 == 16) {
+      // No scaling in the y direction. Must always scale in the x direction.
+      scale->predict[0][0][0] = vp9_convolve8_horiz;
+      scale->predict[0][0][1] = vp9_convolve8_avg_horiz;
+      scale->predict[0][1][0] = vp9_convolve8;
+      scale->predict[0][1][1] = vp9_convolve8_avg;
+      scale->predict[1][0][0] = vp9_convolve8_horiz;
+      scale->predict[1][0][1] = vp9_convolve8_avg_horiz;
+    } else {
+      // Must always scale in both directions.
+      scale->predict[0][0][0] = vp9_convolve8;
+      scale->predict[0][0][1] = vp9_convolve8_avg;
+      scale->predict[0][1][0] = vp9_convolve8;
+      scale->predict[0][1][1] = vp9_convolve8_avg;
+      scale->predict[1][0][0] = vp9_convolve8;
+      scale->predict[1][0][1] = vp9_convolve8_avg;
+    }
+  }
+  // 2D subpel motion always gets filtered in both directions
+  scale->predict[1][1][0] = vp9_convolve8;
+  scale->predict[1][1][1] = vp9_convolve8_avg;
+}
+
+void vp9_setup_interp_filters(MACROBLOCKD *xd,
+                              INTERPOLATIONFILTERTYPE mcomp_filter_type,
+                              VP9_COMMON *cm) {
+  if (xd->mode_info_context) {
+    MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+
+    set_scale_factors(xd,
+                      mbmi->ref_frame[0] - 1,
+                      mbmi->ref_frame[1] - 1,
+                      cm->active_ref_scale);
+  }
+
+  switch (mcomp_filter_type) {
+    case EIGHTTAP:
+    case SWITCHABLE:
+      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8;
+      break;
+    case EIGHTTAP_SMOOTH:
+      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8lp;
+      break;
+    case EIGHTTAP_SHARP:
+      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8s;
+      break;
+    case BILINEAR:
+      xd->subpix.filter_x = xd->subpix.filter_y = vp9_bilinear_filters;
+      break;
+  }
+  assert(((intptr_t)xd->subpix.filter_x & 0xff) == 0);
+}
+
+void vp9_copy_mem16x16_c(const uint8_t *src,
+                         int src_stride,
+                         uint8_t *dst,
+                         int dst_stride) {
+  int r;
+
+  for (r = 0; r < 16; r++) {
+#if !(CONFIG_FAST_UNALIGNED)
+    dst[0] = src[0];
+    dst[1] = src[1];
+    dst[2] = src[2];
+    dst[3] = src[3];
+    dst[4] = src[4];
+    dst[5] = src[5];
+    dst[6] = src[6];
+    dst[7] = src[7];
+    dst[8] = src[8];
+    dst[9] = src[9];
+    dst[10] = src[10];
+    dst[11] = src[11];
+    dst[12] = src[12];
+    dst[13] = src[13];
+    dst[14] = src[14];
+    dst[15] = src[15];
+
+#else
+    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
+    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
+    ((uint32_t *)dst)[2] = ((const uint32_t *)src)[2];
+    ((uint32_t *)dst)[3] = ((const uint32_t *)src)[3];
+
+#endif
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vp9_copy_mem8x8_c(const uint8_t *src,
+                       int src_stride,
+                       uint8_t *dst,
+                       int dst_stride) {
+  int r;
+
+  for (r = 0; r < 8; r++) {
+#if !(CONFIG_FAST_UNALIGNED)
+    dst[0] = src[0];
+    dst[1] = src[1];
+    dst[2] = src[2];
+    dst[3] = src[3];
+    dst[4] = src[4];
+    dst[5] = src[5];
+    dst[6] = src[6];
+    dst[7] = src[7];
+#else
+    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
+    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
+#endif
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vp9_copy_mem8x4_c(const uint8_t *src,
+                       int src_stride,
+                       uint8_t *dst,
+                       int dst_stride) {
+  int r;
+
+  for (r = 0; r < 4; r++) {
+#if !(CONFIG_FAST_UNALIGNED)
+    dst[0] = src[0];
+    dst[1] = src[1];
+    dst[2] = src[2];
+    dst[3] = src[3];
+    dst[4] = src[4];
+    dst[5] = src[5];
+    dst[6] = src[6];
+    dst[7] = src[7];
+#else
+    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
+    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
+#endif
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
+                               uint8_t *dst, int dst_stride,
+                               const int_mv *mv_q3,
+                               const struct scale_factors *scale,
+                               int w, int h, int weight,
+                               const struct subpix_fn_table *subpix) {
+  int_mv32 mv = scale->scale_mv_q3_to_q4(mv_q3, scale);
+  src += (mv.as_mv.row >> 4) * src_stride + (mv.as_mv.col >> 4);
+  scale->predict[!!(mv.as_mv.col & 15)][!!(mv.as_mv.row & 15)][weight](
+      src, src_stride, dst, dst_stride,
+      subpix->filter_x[mv.as_mv.col & 15], scale->x_step_q4,
+      subpix->filter_y[mv.as_mv.row & 15], scale->y_step_q4,
+      w, h);
+}
+
+void vp9_build_inter_predictor_q4(const uint8_t *src, int src_stride,
+                                  uint8_t *dst, int dst_stride,
+                                  const int_mv *mv_q4,
+                                  const struct scale_factors *scale,
+                                  int w, int h, int weight,
+                                  const struct subpix_fn_table *subpix) {
+  const int scaled_mv_row_q4 = scale->scale_mv_component_q4(mv_q4->as_mv.row,
+                                                            scale->y_scale_fp,
+                                                            scale->y_offset_q4);
+  const int scaled_mv_col_q4 = scale->scale_mv_component_q4(mv_q4->as_mv.col,
+                                                            scale->x_scale_fp,
+                                                            scale->x_offset_q4);
+  const int subpel_x = scaled_mv_col_q4 & 15;
+  const int subpel_y = scaled_mv_row_q4 & 15;
+
+  src += (scaled_mv_row_q4 >> 4) * src_stride + (scaled_mv_col_q4 >> 4);
+  scale->predict[!!subpel_x][!!subpel_y][weight](
+      src, src_stride, dst, dst_stride,
+      subpix->filter_x[subpel_x], scale->x_step_q4,
+      subpix->filter_y[subpel_y], scale->y_step_q4,
+      w, h);
+}
+
+static INLINE int round_mv_comp_q4(int value) {
+  return (value < 0 ? value - 2 : value + 2) / 4;
+}
+
+static int mi_mv_pred_row_q4(MACROBLOCKD *mb, int idx) {
+  const int temp = mb->mode_info_context->bmi[0].as_mv[idx].as_mv.row +
+                   mb->mode_info_context->bmi[1].as_mv[idx].as_mv.row +
+                   mb->mode_info_context->bmi[2].as_mv[idx].as_mv.row +
+                   mb->mode_info_context->bmi[3].as_mv[idx].as_mv.row;
+  return round_mv_comp_q4(temp);
+}
+
+static int mi_mv_pred_col_q4(MACROBLOCKD *mb, int idx) {
+  const int temp = mb->mode_info_context->bmi[0].as_mv[idx].as_mv.col +
+                   mb->mode_info_context->bmi[1].as_mv[idx].as_mv.col +
+                   mb->mode_info_context->bmi[2].as_mv[idx].as_mv.col +
+                   mb->mode_info_context->bmi[3].as_mv[idx].as_mv.col;
+  return round_mv_comp_q4(temp);
+}
+
+// TODO(jkoleszar): yet another mv clamping function :-(
+MV clamp_mv_to_umv_border_sb(const MV *src_mv,
+    int bwl, int bhl, int ss_x, int ss_y,
+    int mb_to_left_edge, int mb_to_top_edge,
+    int mb_to_right_edge, int mb_to_bottom_edge) {
+  /* If the MV points so far into the UMV border that no visible pixels
+   * are used for reconstruction, the subpel part of the MV can be
+   * discarded and the MV limited to 16 pixels with equivalent results.
+   */
+  const int spel_left = (VP9_INTERP_EXTEND + (4 << bwl)) << 4;
+  const int spel_right = spel_left - (1 << 4);
+  const int spel_top = (VP9_INTERP_EXTEND + (4 << bhl)) << 4;
+  const int spel_bottom = spel_top - (1 << 4);
+  MV clamped_mv;
+
+  assert(ss_x <= 1);
+  assert(ss_y <= 1);
+  clamped_mv.col = clamp(src_mv->col << (1 - ss_x),
+                         (mb_to_left_edge << (1 - ss_x)) - spel_left,
+                         (mb_to_right_edge << (1 - ss_x)) + spel_right);
+  clamped_mv.row = clamp(src_mv->row << (1 - ss_y),
+                         (mb_to_top_edge << (1 - ss_y)) - spel_top,
+                         (mb_to_bottom_edge << (1 - ss_y)) + spel_bottom);
+  return clamped_mv;
+}
+
+struct build_inter_predictors_args {
+  MACROBLOCKD *xd;
+  int x;
+  int y;
+  uint8_t* dst[MAX_MB_PLANE];
+  int dst_stride[MAX_MB_PLANE];
+  uint8_t* pre[2][MAX_MB_PLANE];
+  int pre_stride[2][MAX_MB_PLANE];
+};
+static void build_inter_predictors(int plane, int block,
+                                   BLOCK_SIZE_TYPE bsize,
+                                   int pred_w, int pred_h,
+                                   void *argv) {
+  const struct build_inter_predictors_args* const arg = argv;
+  MACROBLOCKD * const xd = arg->xd;
+  const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int bhl = b_height_log2(bsize) - xd->plane[plane].subsampling_y;
+  const int bh = 4 << bhl, bw = 4 << bwl;
+  const int x = 4 * (block & ((1 << bwl) - 1)), y = 4 * (block >> bwl);
+  const int use_second_ref = xd->mode_info_context->mbmi.ref_frame[1] > 0;
+  int which_mv;
+
+  assert(x < bw);
+  assert(y < bh);
+  assert(xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8 ||
+         4 << pred_w == bw);
+  assert(xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8 ||
+         4 << pred_h == bh);
+
+  for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
+    // source
+    const uint8_t * const base_pre = arg->pre[which_mv][plane];
+    const int pre_stride = arg->pre_stride[which_mv][plane];
+    const uint8_t *const pre = base_pre +
+        scaled_buffer_offset(x, y, pre_stride, &xd->scale_factor[which_mv]);
+    struct scale_factors * const scale =
+      plane == 0 ? &xd->scale_factor[which_mv] : &xd->scale_factor_uv[which_mv];
+
+    // dest
+    uint8_t *const dst = arg->dst[plane] + arg->dst_stride[plane] * y + x;
+
+    // motion vector
+    const MV *mv;
+    MV split_chroma_mv;
+    int_mv clamped_mv;
+
+    if (xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+      if (plane == 0) {
+        mv = &xd->mode_info_context->bmi[block].as_mv[which_mv].as_mv;
+      } else {
+        // TODO(jkoleszar): All chroma MVs in SPLITMV mode are taken as the
+        // same MV (the average of the 4 luma MVs) but we could do something
+        // smarter for non-4:2:0. Just punt for now, pending the changes to get
+        // rid of SPLITMV mode entirely.
+        split_chroma_mv.row = mi_mv_pred_row_q4(xd, which_mv);
+        split_chroma_mv.col = mi_mv_pred_col_q4(xd, which_mv);
+        mv = &split_chroma_mv;
+      }
+    } else {
+      mv = &xd->mode_info_context->mbmi.mv[which_mv].as_mv;
+    }
+
+    /* TODO(jkoleszar): This clamping is done in the incorrect place for the
+     * scaling case. It needs to be done on the scaled MV, not the pre-scaling
+     * MV. Note however that it performs the subsampling aware scaling so
+     * that the result is always q4.
+     */
+    clamped_mv.as_mv = clamp_mv_to_umv_border_sb(mv, bwl, bhl,
+                                                 xd->plane[plane].subsampling_x,
+                                                 xd->plane[plane].subsampling_y,
+                                                 xd->mb_to_left_edge,
+                                                 xd->mb_to_top_edge,
+                                                 xd->mb_to_right_edge,
+                                                 xd->mb_to_bottom_edge);
+    scale->set_scaled_offsets(scale, arg->y + y, arg->x + x);
+
+    vp9_build_inter_predictor_q4(pre, pre_stride,
+                                 dst, arg->dst_stride[plane],
+                                 &clamped_mv, &xd->scale_factor[which_mv],
+                                 4 << pred_w, 4 << pred_h, which_mv,
+                                 &xd->subpix);
+  }
+}
+void vp9_build_inter_predictors_sby(MACROBLOCKD *xd,
+                                    int mi_row,
+                                    int mi_col,
+                                    BLOCK_SIZE_TYPE bsize) {
+  struct build_inter_predictors_args args = {
+    xd, mi_col * MI_SIZE, mi_row * MI_SIZE,
+    {xd->plane[0].dst.buf, NULL, NULL}, {xd->plane[0].dst.stride, 0, 0},
+    {{xd->plane[0].pre[0].buf, NULL, NULL},
+     {xd->plane[0].pre[1].buf, NULL, NULL}},
+    {{xd->plane[0].pre[0].stride, 0, 0}, {xd->plane[0].pre[1].stride, 0, 0}},
+  };
+
+  foreach_predicted_block_in_plane(xd, bsize, 0, build_inter_predictors, &args);
+}
+void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd,
+                                     int mi_row,
+                                     int mi_col,
+                                     BLOCK_SIZE_TYPE bsize) {
+  struct build_inter_predictors_args args = {
+    xd, mi_col * MI_SIZE, mi_row * MI_SIZE,
+#if CONFIG_ALPHA
+    {NULL, xd->plane[1].dst.buf, xd->plane[2].dst.buf,
+     xd->plane[3].dst.buf},
+    {0, xd->plane[1].dst.stride, xd->plane[1].dst.stride,
+     xd->plane[3].dst.stride},
+    {{NULL, xd->plane[1].pre[0].buf, xd->plane[2].pre[0].buf,
+      xd->plane[3].pre[0].buf},
+     {NULL, xd->plane[1].pre[1].buf, xd->plane[2].pre[1].buf,
+      xd->plane[3].pre[1].buf}},
+    {{0, xd->plane[1].pre[0].stride, xd->plane[1].pre[0].stride,
+      xd->plane[3].pre[0].stride},
+     {0, xd->plane[1].pre[1].stride, xd->plane[1].pre[1].stride,
+      xd->plane[3].pre[1].stride}},
+#else
+    {NULL, xd->plane[1].dst.buf, xd->plane[2].dst.buf},
+    {0, xd->plane[1].dst.stride, xd->plane[1].dst.stride},
+    {{NULL, xd->plane[1].pre[0].buf, xd->plane[2].pre[0].buf},
+     {NULL, xd->plane[1].pre[1].buf, xd->plane[2].pre[1].buf}},
+    {{0, xd->plane[1].pre[0].stride, xd->plane[1].pre[0].stride},
+     {0, xd->plane[1].pre[1].stride, xd->plane[1].pre[1].stride}},
+#endif
+  };
+  foreach_predicted_block_uv(xd, bsize, build_inter_predictors, &args);
+}
+void vp9_build_inter_predictors_sb(MACROBLOCKD *xd,
+                                   int mi_row, int mi_col,
+                                   BLOCK_SIZE_TYPE bsize) {
+
+  vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+  vp9_build_inter_predictors_sbuv(xd, mi_row, mi_col, bsize);
+}
+
+/*encoder only*/
+void vp9_build_inter4x4_predictors_mbuv(MACROBLOCKD *xd,
+                                        int mb_row, int mb_col) {
+  vp9_build_inter_predictors_sbuv(xd, mb_row, mb_col,
+                                  BLOCK_SIZE_MB16X16);
+}
+
+// TODO(dkovalev: find better place for this function)
+void vp9_setup_scale_factors(VP9_COMMON *cm, int i) {
+  const int ref = cm->active_ref_idx[i];
+  struct scale_factors *const sf = &cm->active_ref_scale[i];
+  if (ref >= NUM_YV12_BUFFERS) {
+    memset(sf, 0, sizeof(*sf));
+  } else {
+    YV12_BUFFER_CONFIG *const fb = &cm->yv12_fb[ref];
+    vp9_setup_scale_factors_for_frame(sf,
+                                      fb->y_crop_width, fb->y_crop_height,
+                                      cm->width, cm->height);
+  }
+}
+
diff --git a/libvpx/vp9/common/vp9_reconinter.h b/libvpx/vp9/common/vp9_reconinter.h
new file mode 100644
index 0000000..4e52185
--- /dev/null
+++ b/libvpx/vp9/common/vp9_reconinter.h
@@ -0,0 +1,130 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_RECONINTER_H_
+#define VP9_COMMON_VP9_RECONINTER_H_
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+struct subpix_fn_table;
+void vp9_build_inter_predictors_sby(MACROBLOCKD *xd,
+                                    int mb_row,
+                                    int mb_col,
+                                    BLOCK_SIZE_TYPE bsize);
+
+void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd,
+                                     int mb_row,
+                                     int mb_col,
+                                     BLOCK_SIZE_TYPE bsize);
+
+void vp9_build_inter_predictors_sb(MACROBLOCKD *mb,
+                                   int mb_row, int mb_col,
+                                   BLOCK_SIZE_TYPE bsize);
+
+void vp9_setup_interp_filters(MACROBLOCKD *xd,
+                              INTERPOLATIONFILTERTYPE filter,
+                              VP9_COMMON *cm);
+
+void vp9_setup_scale_factors_for_frame(struct scale_factors *scale,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h);
+
+void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
+                               uint8_t *dst, int dst_stride,
+                               const int_mv *mv_q3,
+                               const struct scale_factors *scale,
+                               int w, int h, int do_avg,
+                               const struct subpix_fn_table *subpix);
+
+void vp9_build_inter_predictor_q4(const uint8_t *src, int src_stride,
+                                  uint8_t *dst, int dst_stride,
+                                  const int_mv *mv_q4,
+                                  const struct scale_factors *scale,
+                                  int w, int h, int do_avg,
+                                  const struct subpix_fn_table *subpix);
+
+static int scaled_buffer_offset(int x_offset, int y_offset, int stride,
+                                const struct scale_factors *scale) {
+  const int x = scale ? scale->scale_value_x(x_offset, scale) : x_offset;
+  const int y = scale ? scale->scale_value_y(y_offset, scale) : y_offset;
+  return y * stride + x;
+}
+
+static void setup_pred_plane(struct buf_2d *dst,
+                             uint8_t *src, int stride,
+                             int mi_row, int mi_col,
+                             const struct scale_factors *scale,
+                             int subsampling_x, int subsampling_y) {
+  const int x = (MI_SIZE * mi_col) >> subsampling_x;
+  const int y = (MI_SIZE * mi_row) >> subsampling_y;
+  dst->buf = src + scaled_buffer_offset(x, y, stride, scale);
+  dst->stride = stride;
+}
+
+// TODO(jkoleszar): audit all uses of this that don't set mb_row, mb_col
+static void setup_dst_planes(MACROBLOCKD *xd,
+                             const YV12_BUFFER_CONFIG *src,
+                             int mi_row, int mi_col) {
+  uint8_t *buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                         src->alpha_buffer};
+  int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                    src->alpha_stride};
+  int i;
+
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+    struct macroblockd_plane *pd = &xd->plane[i];
+    setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
+                     pd->subsampling_x, pd->subsampling_y);
+  }
+}
+
+static void setup_pre_planes(MACROBLOCKD *xd,
+                             const YV12_BUFFER_CONFIG *src0,
+                             const YV12_BUFFER_CONFIG *src1,
+                             int mi_row, int mi_col,
+                             const struct scale_factors *scale,
+                             const struct scale_factors *scale_uv) {
+  const YV12_BUFFER_CONFIG *srcs[2] = {src0, src1};
+  int i, j;
+
+  for (i = 0; i < 2; ++i) {
+    const YV12_BUFFER_CONFIG *src = srcs[i];
+    if (src) {
+      uint8_t* buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                             src->alpha_buffer};
+      int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                        src->alpha_stride};
+
+      for (j = 0; j < MAX_MB_PLANE; ++j) {
+        struct macroblockd_plane *pd = &xd->plane[j];
+        const struct scale_factors *sf = j ? scale_uv : scale;
+        setup_pred_plane(&pd->pre[i],
+                         buffers[j], strides[j],
+                         mi_row, mi_col, sf ? &sf[i] : NULL,
+                         pd->subsampling_x, pd->subsampling_y);
+      }
+    }
+  }
+}
+
+static void set_scale_factors(MACROBLOCKD *xd,
+    int ref0, int ref1,
+    struct scale_factors scale_factor[MAX_REF_FRAMES]) {
+
+  xd->scale_factor[0] = scale_factor[ref0 >= 0 ? ref0 : 0];
+  xd->scale_factor[1] = scale_factor[ref1 >= 0 ? ref1 : 0];
+  xd->scale_factor_uv[0] = xd->scale_factor[0];
+  xd->scale_factor_uv[1] = xd->scale_factor[1];
+}
+
+void vp9_setup_scale_factors(VP9_COMMON *cm, int i);
+
+#endif  // VP9_COMMON_VP9_RECONINTER_H_
diff --git a/libvpx/vp9/common/vp9_reconintra.c b/libvpx/vp9/common/vp9_reconintra.c
new file mode 100644
index 0000000..85dfe51
--- /dev/null
+++ b/libvpx/vp9/common/vp9_reconintra.c
@@ -0,0 +1,357 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vpx_mem/vpx_mem.h"
+
+static void d27_predictor(uint8_t *ypred_ptr, int y_stride,
+                          int bw, int bh,
+                          uint8_t *yabove_row, uint8_t *yleft_col) {
+  int r, c;
+  // first column
+  for (r = 0; r < bh - 1; ++r) {
+      ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r] +
+                                                   yleft_col[r + 1], 1);
+  }
+  ypred_ptr[(bh - 1) * y_stride] = yleft_col[bh-1];
+  ypred_ptr++;
+  // second column
+  for (r = 0; r < bh - 2; ++r) {
+      ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r] +
+                                                   yleft_col[r + 1] * 2 +
+                                                   yleft_col[r + 2], 2);
+  }
+  ypred_ptr[(bh - 2) * y_stride] = ROUND_POWER_OF_TWO(yleft_col[bh - 2] +
+                                                      yleft_col[bh - 1] * 3,
+                                                      2);
+  ypred_ptr[(bh - 1) * y_stride] = yleft_col[bh-1];
+  ypred_ptr++;
+
+  // rest of last row
+  for (c = 0; c < bw - 2; ++c) {
+    ypred_ptr[(bh - 1) * y_stride + c] = yleft_col[bh-1];
+  }
+
+  for (r = bh - 2; r >= 0; --r) {
+    for (c = 0; c < bw - 2; ++c) {
+      ypred_ptr[r * y_stride + c] = ypred_ptr[(r + 1) * y_stride + c - 2];
+    }
+  }
+}
+
+static void d63_predictor(uint8_t *ypred_ptr, int y_stride,
+                          int bw, int bh,
+                          uint8_t *yabove_row, uint8_t *yleft_col) {
+  int r, c;
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
+      if (r & 1) {
+        ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[r/2 + c] +
+                                          yabove_row[r/2 + c + 1] * 2 +
+                                          yabove_row[r/2 + c + 2], 2);
+      } else {
+        ypred_ptr[c] =ROUND_POWER_OF_TWO(yabove_row[r/2 + c] +
+                                         yabove_row[r/2+ c + 1], 1);
+      }
+    }
+    ypred_ptr += y_stride;
+  }
+}
+
+static void d45_predictor(uint8_t *ypred_ptr, int y_stride,
+                          int bw, int bh,
+                          uint8_t *yabove_row, uint8_t *yleft_col) {
+  int r, c;
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
+      if (r + c + 2 < bw * 2)
+        ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[r + c] +
+                                          yabove_row[r + c + 1] * 2 +
+                                          yabove_row[r + c + 2], 2);
+      else
+        ypred_ptr[c] = yabove_row[bw * 2 - 1];
+    }
+    ypred_ptr += y_stride;
+  }
+}
+
+static void d117_predictor(uint8_t *ypred_ptr, int y_stride,
+                           int bw, int bh,
+                           uint8_t *yabove_row, uint8_t *yleft_col) {
+  int r, c;
+  // first row
+  for (c = 0; c < bw; c++)
+    ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 1] + yabove_row[c], 1);
+  ypred_ptr += y_stride;
+
+  // second row
+  ypred_ptr[0] = ROUND_POWER_OF_TWO(yleft_col[0] +
+                                    yabove_row[-1] * 2 +
+                                    yabove_row[0], 2);
+  for (c = 1; c < bw; c++)
+    ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 2] +
+                                      yabove_row[c - 1] * 2 +
+                                      yabove_row[c], 2);
+  ypred_ptr += y_stride;
+
+  // the rest of first col
+  ypred_ptr[0] = ROUND_POWER_OF_TWO(yabove_row[-1] +
+                                    yleft_col[0] * 2 +
+                                    yleft_col[1], 2);
+  for (r = 3; r < bh; ++r)
+    ypred_ptr[(r-2) * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 3] +
+                                                     yleft_col[r - 2] * 2 +
+                                                     yleft_col[r - 1], 2);
+  // the rest of the block
+  for (r = 2; r < bh; ++r) {
+    for (c = 1; c < bw; c++)
+      ypred_ptr[c] = ypred_ptr[-2 * y_stride + c - 1];
+    ypred_ptr += y_stride;
+  }
+}
+
+
+static void d135_predictor(uint8_t *ypred_ptr, int y_stride,
+                           int bw, int bh,
+                           uint8_t *yabove_row, uint8_t *yleft_col) {
+  int r, c;
+  ypred_ptr[0] = ROUND_POWER_OF_TWO(yleft_col[0] +
+                                    yabove_row[-1] * 2 +
+                                    yabove_row[0], 2);
+  for (c = 1; c < bw; c++)
+    ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 2] +
+                                      yabove_row[c - 1] * 2 +
+                                      yabove_row[c], 2);
+
+  ypred_ptr[y_stride] = ROUND_POWER_OF_TWO(yabove_row[-1] +
+                                           yleft_col[0] * 2 +
+                                           yleft_col[1], 2);
+  for (r = 2; r < bh; ++r)
+    ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 2] +
+                                                 yleft_col[r - 1] * 2 +
+                                                 yleft_col[r], 2);
+
+  ypred_ptr += y_stride;
+  for (r = 1; r < bh; ++r) {
+    for (c = 1; c < bw; c++)
+      ypred_ptr[c] = ypred_ptr[-y_stride + c - 1];
+    ypred_ptr += y_stride;
+  }
+}
+
+static void d153_predictor(uint8_t *ypred_ptr,
+                           int y_stride,
+                           int bw, int bh,
+                           uint8_t *yabove_row,
+                           uint8_t *yleft_col) {
+  int r, c;
+  ypred_ptr[0] = ROUND_POWER_OF_TWO(yabove_row[-1] + yleft_col[0], 1);
+  for (r = 1; r < bh; r++)
+    ypred_ptr[r * y_stride] =
+        ROUND_POWER_OF_TWO(yleft_col[r - 1] + yleft_col[r], 1);
+  ypred_ptr++;
+
+  ypred_ptr[0] = ROUND_POWER_OF_TWO(yleft_col[0] +
+                                    yabove_row[-1] * 2 +
+                                    yabove_row[0], 2);
+  ypred_ptr[y_stride] = ROUND_POWER_OF_TWO(yabove_row[-1] +
+                                           yleft_col[0] * 2 +
+                                           yleft_col[1], 2);
+  for (r = 2; r < bh; r++)
+    ypred_ptr[r * y_stride] = ROUND_POWER_OF_TWO(yleft_col[r - 2] +
+                                                 yleft_col[r - 1] * 2 +
+                                                 yleft_col[r], 2);
+  ypred_ptr++;
+
+  for (c = 0; c < bw - 2; c++)
+    ypred_ptr[c] = ROUND_POWER_OF_TWO(yabove_row[c - 1] +
+                                      yabove_row[c] * 2 +
+                                      yabove_row[c + 1], 2);
+  ypred_ptr += y_stride;
+  for (r = 1; r < bh; ++r) {
+    for (c = 0; c < bw - 2; c++)
+      ypred_ptr[c] = ypred_ptr[-y_stride + c - 2];
+    ypred_ptr += y_stride;
+  }
+}
+
+void vp9_build_intra_predictors(uint8_t *src, int src_stride,
+                                uint8_t *ypred_ptr,
+                                int y_stride, int mode,
+                                int bw, int bh,
+                                int up_available, int left_available,
+                                int right_available) {
+  int r, c, i;
+  uint8_t yleft_col[64], yabove_data[129], ytop_left;
+  uint8_t *yabove_row = yabove_data + 1;
+
+  // 127 127 127 .. 127 127 127 127 127 127
+  // 129  A   B  ..  Y   Z
+  // 129  C   D  ..  W   X
+  // 129  E   F  ..  U   V
+  // 129  G   H  ..  S   T   T   T   T   T
+  // ..
+
+  assert(bw == bh);
+
+  if (left_available) {
+    for (i = 0; i < bh; i++)
+      yleft_col[i] = src[i * src_stride - 1];
+  } else {
+    vpx_memset(yleft_col, 129, bh);
+  }
+
+  if (up_available) {
+    uint8_t *yabove_ptr = src - src_stride;
+    vpx_memcpy(yabove_row, yabove_ptr, bw);
+    if (bw == 4 && right_available)
+      vpx_memcpy(yabove_row + bw, yabove_ptr + bw, bw);
+    else
+      vpx_memset(yabove_row + bw, yabove_row[bw -1], bw);
+    ytop_left = left_available ? yabove_ptr[-1] : 129;
+  } else {
+    vpx_memset(yabove_row, 127, bw * 2);
+    ytop_left = 127;
+  }
+  yabove_row[-1] = ytop_left;
+
+  switch (mode) {
+    case DC_PRED: {
+      int i;
+      int expected_dc = 128;
+      int average = 0;
+      int count = 0;
+
+      if (up_available || left_available) {
+        if (up_available) {
+          for (i = 0; i < bw; i++)
+            average += yabove_row[i];
+          count += bw;
+        }
+        if (left_available) {
+          for (i = 0; i < bh; i++)
+            average += yleft_col[i];
+          count += bh;
+        }
+        expected_dc = (average + (count >> 1)) / count;
+      }
+      for (r = 0; r < bh; r++) {
+        vpx_memset(ypred_ptr, expected_dc, bw);
+        ypred_ptr += y_stride;
+      }
+    }
+    break;
+    case V_PRED:
+      for (r = 0; r < bh; r++) {
+        vpx_memcpy(ypred_ptr, yabove_row, bw);
+        ypred_ptr += y_stride;
+      }
+      break;
+    case H_PRED:
+      for (r = 0; r < bh; r++) {
+        vpx_memset(ypred_ptr, yleft_col[r], bw);
+        ypred_ptr += y_stride;
+      }
+      break;
+    case TM_PRED:
+      for (r = 0; r < bh; r++) {
+        for (c = 0; c < bw; c++)
+          ypred_ptr[c] = clip_pixel(yleft_col[r] + yabove_row[c] - ytop_left);
+        ypred_ptr += y_stride;
+      }
+      break;
+    case D45_PRED:
+      d45_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    case D135_PRED:
+      d135_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    case D117_PRED:
+      d117_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    case D153_PRED:
+      d153_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    case D27_PRED:
+      d27_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    case D63_PRED:
+      d63_predictor(ypred_ptr, y_stride, bw, bh, yabove_row, yleft_col);
+      break;
+    default:
+      break;
+  }
+}
+
+void vp9_build_intra_predictors_sby_s(MACROBLOCKD *xd,
+                                      BLOCK_SIZE_TYPE bsize) {
+  const struct macroblockd_plane* const pd = &xd->plane[0];
+  const int bw = plane_block_width(bsize, pd);
+  const int bh = plane_block_height(bsize, pd);
+  vp9_build_intra_predictors(pd->dst.buf, pd->dst.stride,
+                             pd->dst.buf, pd->dst.stride,
+                             xd->mode_info_context->mbmi.mode,
+                             bw, bh, xd->up_available, xd->left_available,
+                             0 /*xd->right_available*/);
+}
+
+void vp9_build_intra_predictors_sbuv_s(MACROBLOCKD *xd,
+                                       BLOCK_SIZE_TYPE bsize) {
+  const int bwl = b_width_log2(bsize), bw = 2 << bwl;
+  const int bhl = b_height_log2(bsize), bh = 2 << bhl;
+
+  vp9_build_intra_predictors(xd->plane[1].dst.buf, xd->plane[1].dst.stride,
+                             xd->plane[1].dst.buf, xd->plane[1].dst.stride,
+                             xd->mode_info_context->mbmi.uv_mode,
+                             bw, bh, xd->up_available,
+                             xd->left_available, 0 /*xd->right_available*/);
+  vp9_build_intra_predictors(xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                             xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                             xd->mode_info_context->mbmi.uv_mode,
+                             bw, bh, xd->up_available,
+                             xd->left_available, 0 /*xd->right_available*/);
+}
+
+void vp9_predict_intra_block(MACROBLOCKD *xd,
+                            int block_idx,
+                            int bwl_in,
+                            TX_SIZE tx_size,
+                            int mode,
+                            uint8_t *predictor, int pre_stride) {
+  const int bwl = bwl_in - tx_size;
+  const int wmask = (1 << bwl) - 1;
+  const int have_top = (block_idx >> bwl) || xd->up_available;
+  const int have_left = (block_idx & wmask) || xd->left_available;
+  const int have_right = ((block_idx & wmask) != wmask);
+  const int txfm_block_size = 4 << tx_size;
+
+  assert(bwl >= 0);
+  vp9_build_intra_predictors(predictor, pre_stride,
+                             predictor, pre_stride,
+                             mode,
+                             txfm_block_size,
+                             txfm_block_size,
+                             have_top, have_left,
+                             have_right);
+}
+
+void vp9_intra4x4_predict(MACROBLOCKD *xd,
+                          int block_idx,
+                          BLOCK_SIZE_TYPE bsize,
+                          int mode,
+                          uint8_t *predictor, int pre_stride) {
+  vp9_predict_intra_block(xd, block_idx, b_width_log2(bsize), TX_4X4,
+                          mode, predictor, pre_stride);
+}
diff --git a/libvpx/vp9/common/vp9_reconintra.h b/libvpx/vp9/common/vp9_reconintra.h
new file mode 100644
index 0000000..f5f5f42
--- /dev/null
+++ b/libvpx/vp9/common/vp9_reconintra.h
@@ -0,0 +1,30 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_RECONINTRA_H_
+#define VP9_COMMON_VP9_RECONINTRA_H_
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_blockd.h"
+
+MB_PREDICTION_MODE vp9_find_dominant_direction(uint8_t *ptr,
+                                               int stride, int n,
+                                               int tx, int ty);
+
+MB_PREDICTION_MODE vp9_find_bpred_context(MACROBLOCKD *xd, int block,
+                                          uint8_t *ptr, int stride);
+
+void vp9_predict_intra_block(MACROBLOCKD *xd,
+                            int block_idx,
+                            int bwl_in,
+                            TX_SIZE tx_size,
+                            int mode,
+                            uint8_t *predictor, int pre_stride);
+#endif  // VP9_COMMON_VP9_RECONINTRA_H_
diff --git a/libvpx/vp9/common/vp9_rtcd.c b/libvpx/vp9/common/vp9_rtcd.c
new file mode 100644
index 0000000..72613ae
--- /dev/null
+++ b/libvpx/vp9/common/vp9_rtcd.c
@@ -0,0 +1,20 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "vpx_config.h"
+#define RTCD_C
+#include "vp9_rtcd.h"
+#include "vpx_ports/vpx_once.h"
+
+void vpx_scale_rtcd(void);
+
+void vp9_rtcd() {
+    vpx_scale_rtcd();
+    once(setup_rtcd_internal);
+}
diff --git a/libvpx/vp9/common/vp9_rtcd_defs.sh b/libvpx/vp9/common/vp9_rtcd_defs.sh
new file mode 100644
index 0000000..a405aab
--- /dev/null
+++ b/libvpx/vp9/common/vp9_rtcd_defs.sh
@@ -0,0 +1,611 @@
+vp9_common_forward_decls() {
+cat <<EOF
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+EOF
+}
+forward_decls vp9_common_forward_decls
+
+#
+# Dequant
+#
+
+prototype void vp9_idct_add_16x16 "int16_t *input, uint8_t *dest, int stride, int eob"
+specialize vp9_idct_add_16x16
+
+prototype void vp9_idct_add_8x8 "int16_t *input, uint8_t *dest, int stride, int eob"
+specialize vp9_idct_add_8x8
+
+prototype void vp9_idct_add "int16_t *input, uint8_t *dest, int stride, int eob"
+specialize vp9_idct_add
+
+
+
+prototype void vp9_idct_add_32x32 "int16_t *q, uint8_t *dst, int stride, int eob"
+specialize vp9_idct_add_32x32
+
+#
+# RECON
+#
+prototype void vp9_copy_mem16x16 "const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch"
+specialize vp9_copy_mem16x16 mmx sse2 dspr2
+vp9_copy_mem16x16_dspr2=vp9_copy_mem16x16_dspr2
+
+prototype void vp9_copy_mem8x8 "const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch"
+specialize vp9_copy_mem8x8 mmx dspr2
+vp9_copy_mem8x8_dspr2=vp9_copy_mem8x8_dspr2
+
+prototype void vp9_copy_mem8x4 "const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch"
+specialize vp9_copy_mem8x4 mmx
+
+prototype void vp9_build_intra_predictors "uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available"
+specialize void vp9_build_intra_predictors
+
+prototype void vp9_build_intra_predictors_sby_s "struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize"
+specialize vp9_build_intra_predictors_sby_s
+
+prototype void vp9_build_intra_predictors_sbuv_s "struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize"
+specialize vp9_build_intra_predictors_sbuv_s
+
+prototype void vp9_intra4x4_predict "struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride"
+specialize vp9_intra4x4_predict;
+
+if [ "$CONFIG_VP9_DECODER" = "yes" ]; then
+prototype void vp9_add_constant_residual_8x8 "const int16_t diff, uint8_t *dest, int stride"
+specialize vp9_add_constant_residual_8x8 sse2
+
+prototype void vp9_add_constant_residual_16x16 "const int16_t diff, uint8_t *dest, int stride"
+specialize vp9_add_constant_residual_16x16 sse2
+
+prototype void vp9_add_constant_residual_32x32 "const int16_t diff, uint8_t *dest, int stride"
+specialize vp9_add_constant_residual_32x32 sse2
+fi
+
+#
+# Loopfilter
+#
+prototype void vp9_mb_lpf_vertical_edge_w "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh"
+specialize vp9_mb_lpf_vertical_edge_w sse2
+
+prototype void vp9_mbloop_filter_vertical_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
+specialize vp9_mbloop_filter_vertical_edge sse2
+
+prototype void vp9_loop_filter_vertical_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
+specialize vp9_loop_filter_vertical_edge mmx
+
+prototype void vp9_mb_lpf_horizontal_edge_w "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh"
+specialize vp9_mb_lpf_horizontal_edge_w sse2
+
+prototype void vp9_mbloop_filter_horizontal_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
+specialize vp9_mbloop_filter_horizontal_edge sse2
+
+prototype void vp9_loop_filter_horizontal_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
+specialize vp9_loop_filter_horizontal_edge mmx
+
+#
+# post proc
+#
+if [ "$CONFIG_POSTPROC" = "yes" ]; then
+prototype void vp9_mbpost_proc_down "uint8_t *dst, int pitch, int rows, int cols, int flimit"
+specialize vp9_mbpost_proc_down mmx sse2
+vp9_mbpost_proc_down_sse2=vp9_mbpost_proc_down_xmm
+
+prototype void vp9_mbpost_proc_across_ip "uint8_t *src, int pitch, int rows, int cols, int flimit"
+specialize vp9_mbpost_proc_across_ip sse2
+vp9_mbpost_proc_across_ip_sse2=vp9_mbpost_proc_across_ip_xmm
+
+prototype void vp9_post_proc_down_and_across "const uint8_t *src_ptr, uint8_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit"
+specialize vp9_post_proc_down_and_across mmx sse2
+vp9_post_proc_down_and_across_sse2=vp9_post_proc_down_and_across_xmm
+
+prototype void vp9_plane_add_noise "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch"
+specialize vp9_plane_add_noise mmx sse2
+vp9_plane_add_noise_sse2=vp9_plane_add_noise_wmt
+fi
+
+prototype void vp9_blend_mb_inner "uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride"
+specialize vp9_blend_mb_inner
+
+prototype void vp9_blend_mb_outer "uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride"
+specialize vp9_blend_mb_outer
+
+prototype void vp9_blend_b "uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride"
+specialize vp9_blend_b
+
+#
+# Sub Pixel Filters
+#
+prototype void vp9_convolve8 "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8 ssse3
+
+prototype void vp9_convolve8_horiz "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8_horiz ssse3
+
+prototype void vp9_convolve8_vert "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8_vert ssse3
+
+prototype void vp9_convolve8_avg "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8_avg ssse3
+
+prototype void vp9_convolve8_avg_horiz "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8_avg_horiz ssse3
+
+prototype void vp9_convolve8_avg_vert "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
+specialize vp9_convolve8_avg_vert ssse3
+
+#
+# dct
+#
+prototype void vp9_short_idct4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct4x4_1_add
+
+prototype void vp9_short_idct4x4_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct4x4_add sse2
+
+prototype void vp9_short_idct8x8_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct8x8_add sse2
+
+prototype void vp9_short_idct10_8x8_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct10_8x8_add sse2
+
+prototype void vp9_short_idct1_8x8 "int16_t *input, int16_t *output"
+specialize vp9_short_idct1_8x8
+
+prototype void vp9_short_idct16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct16x16_add sse2
+
+prototype void vp9_short_idct10_16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct10_16x16_add sse2
+
+prototype void vp9_short_idct1_16x16 "int16_t *input, int16_t *output"
+specialize vp9_short_idct1_16x16
+
+prototype void vp9_short_idct32x32_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct32x32_add sse2
+
+prototype void vp9_short_idct1_32x32 "int16_t *input, int16_t *output"
+specialize vp9_short_idct1_32x32
+
+prototype void vp9_short_idct10_32x32_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_idct10_32x32_add
+
+prototype void vp9_short_iht4x4_add "int16_t *input, uint8_t *dest, int dest_stride, int tx_type"
+specialize vp9_short_iht4x4_add
+
+prototype void vp9_short_iht8x8_add "int16_t *input, uint8_t *dest, int dest_stride, int tx_type"
+specialize vp9_short_iht8x8_add
+
+prototype void vp9_short_iht16x16_add "int16_t *input, uint8_t *output, int pitch, int tx_type"
+specialize vp9_short_iht16x16_add
+
+prototype void vp9_idct4_1d "int16_t *input, int16_t *output"
+specialize vp9_idct4_1d sse2
+# dct and add
+
+prototype void vp9_dc_only_idct_add "int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride"
+specialize vp9_dc_only_idct_add sse2
+
+prototype void vp9_short_iwalsh4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_iwalsh4x4_1_add
+
+prototype void vp9_short_iwalsh4x4_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_short_iwalsh4x4_add
+
+prototype unsigned int vp9_sad32x3 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad"
+specialize vp9_sad32x3
+
+prototype unsigned int vp9_sad3x32 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad"
+specialize vp9_sad3x32
+
+#
+# Encoder functions below this point.
+#
+if [ "$CONFIG_VP9_ENCODER" = "yes" ]; then
+
+
+# variance
+[ $arch = "x86_64" ] && mmx_x86_64=mmx && sse2_x86_64=sse2
+
+prototype unsigned int vp9_variance32x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance32x16 sse2
+
+prototype unsigned int vp9_variance16x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance16x32 sse2
+
+prototype unsigned int vp9_variance64x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance64x32 sse2
+
+prototype unsigned int vp9_variance32x64 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance32x64 sse2
+
+prototype unsigned int vp9_variance32x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance32x32 sse2
+
+prototype unsigned int vp9_variance64x64 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance64x64 sse2
+
+prototype unsigned int vp9_variance16x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance16x16 mmx sse2
+
+prototype unsigned int vp9_variance16x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance16x8 mmx sse2
+
+prototype unsigned int vp9_variance8x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance8x16 mmx sse2
+
+prototype unsigned int vp9_variance8x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance8x8 mmx sse2
+
+prototype void vp9_get_sse_sum_8x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum"
+specialize vp9_get_sse_sum_8x8 sse2
+vp9_get_sse_sum_8x8_sse2=vp9_get8x8var_sse2
+
+prototype unsigned int vp9_variance8x4 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance8x4 sse2
+
+prototype unsigned int vp9_variance4x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance4x8 sse2
+
+prototype unsigned int vp9_variance4x4 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance4x4 mmx sse2
+
+prototype unsigned int vp9_sub_pixel_variance64x64 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance64x64 sse2
+
+prototype unsigned int vp9_sub_pixel_avg_variance64x64 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance64x64
+
+prototype unsigned int vp9_sub_pixel_variance32x64 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance32x64
+
+prototype unsigned int vp9_sub_pixel_avg_variance32x64 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance32x64
+
+prototype unsigned int vp9_sub_pixel_variance64x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance64x32
+
+prototype unsigned int vp9_sub_pixel_avg_variance64x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance64x32
+
+prototype unsigned int vp9_sub_pixel_variance32x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance32x16
+
+prototype unsigned int vp9_sub_pixel_avg_variance32x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance32x16
+
+prototype unsigned int vp9_sub_pixel_variance16x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance16x32
+
+prototype unsigned int vp9_sub_pixel_avg_variance16x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance16x32
+
+prototype unsigned int vp9_sub_pixel_variance32x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance32x32 sse2
+
+prototype unsigned int vp9_sub_pixel_avg_variance32x32 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance32x32
+
+prototype unsigned int vp9_sub_pixel_variance16x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance16x16 sse2 mmx ssse3
+
+prototype unsigned int vp9_sub_pixel_avg_variance16x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance16x16
+
+prototype unsigned int vp9_sub_pixel_variance8x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance8x16 sse2 mmx
+vp9_sub_pixel_variance8x16_sse2=vp9_sub_pixel_variance8x16_wmt
+
+prototype unsigned int vp9_sub_pixel_avg_variance8x16 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance8x16
+
+prototype unsigned int vp9_sub_pixel_variance16x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance16x8 sse2 mmx ssse3
+vp9_sub_pixel_variance16x8_sse2=vp9_sub_pixel_variance16x8_ssse3;
+vp9_sub_pixel_variance16x8_sse2=vp9_sub_pixel_variance16x8_wmt
+
+prototype unsigned int vp9_sub_pixel_avg_variance16x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance16x8
+
+prototype unsigned int vp9_sub_pixel_variance8x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance8x8 sse2 mmx
+vp9_sub_pixel_variance8x8_sse2=vp9_sub_pixel_variance8x8_wmt
+
+prototype unsigned int vp9_sub_pixel_avg_variance8x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance8x8
+
+# TODO(jingning): need to convert 8x4/4x8 functions into mmx/sse form
+prototype unsigned int vp9_sub_pixel_variance8x4 "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance8x4
+
+prototype unsigned int vp9_sub_pixel_avg_variance8x4 "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance8x4
+
+prototype unsigned int vp9_sub_pixel_variance4x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance4x8
+
+prototype unsigned int vp9_sub_pixel_avg_variance4x8 "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance4x8
+
+prototype unsigned int vp9_sub_pixel_variance4x4 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_variance4x4 sse2 mmx
+vp9_sub_pixel_variance4x4_sse2=vp9_sub_pixel_variance4x4_wmt
+
+prototype unsigned int vp9_sub_pixel_avg_variance4x4 "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred"
+specialize vp9_sub_pixel_avg_variance4x4
+
+prototype unsigned int vp9_sad64x64 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad64x64 sse2
+
+prototype unsigned int vp9_sad32x64 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad32x64 sse2
+
+prototype unsigned int vp9_sad64x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad64x32 sse2
+
+prototype unsigned int vp9_sad32x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad32x16 sse2
+
+prototype unsigned int vp9_sad16x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad16x32 sse2
+
+prototype unsigned int vp9_sad32x32 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad32x32 sse2
+
+prototype unsigned int vp9_sad16x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad16x16 mmx sse2
+
+prototype unsigned int vp9_sad16x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad16x8 mmx sse2
+
+prototype unsigned int vp9_sad8x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad8x16 mmx sse2
+
+prototype unsigned int vp9_sad8x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad8x8 mmx sse2
+
+# TODO(jingning): need to covert these functions into mmx/sse2 form
+prototype unsigned int vp9_sad8x4 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad8x4 sse2
+
+prototype unsigned int vp9_sad4x8 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int max_sad"
+specialize vp9_sad4x8 sse
+
+prototype unsigned int vp9_sad4x4 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int max_sad"
+specialize vp9_sad4x4 mmx sse
+
+prototype unsigned int vp9_variance_halfpixvar16x16_h "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar16x16_h mmx sse2
+vp9_variance_halfpixvar16x16_h_sse2=vp9_variance_halfpixvar16x16_h_wmt
+
+prototype unsigned int vp9_variance_halfpixvar16x16_v "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar16x16_v mmx sse2
+vp9_variance_halfpixvar16x16_v_sse2=vp9_variance_halfpixvar16x16_v_wmt
+
+prototype unsigned int vp9_variance_halfpixvar16x16_hv "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar16x16_hv mmx sse2
+vp9_variance_halfpixvar16x16_hv_sse2=vp9_variance_halfpixvar16x16_hv_wmt
+
+prototype unsigned int vp9_variance_halfpixvar64x64_h "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar64x64_h
+
+prototype unsigned int vp9_variance_halfpixvar64x64_v "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar64x64_v
+
+prototype unsigned int vp9_variance_halfpixvar64x64_hv "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar64x64_hv
+
+prototype unsigned int vp9_variance_halfpixvar32x32_h "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar32x32_h
+
+prototype unsigned int vp9_variance_halfpixvar32x32_v "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar32x32_v
+
+prototype unsigned int vp9_variance_halfpixvar32x32_hv "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_variance_halfpixvar32x32_hv
+
+prototype void vp9_sad64x64x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad64x64x3
+
+prototype void vp9_sad32x32x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad32x32x3
+
+prototype void vp9_sad16x16x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad16x16x3 sse3 ssse3
+
+prototype void vp9_sad16x8x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad16x8x3 sse3 ssse3
+
+prototype void vp9_sad8x16x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad8x16x3 sse3
+
+prototype void vp9_sad8x8x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad8x8x3 sse3
+
+prototype void vp9_sad4x4x3 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad4x4x3 sse3
+
+prototype void vp9_sad64x64x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad64x64x8
+
+prototype void vp9_sad32x32x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad32x32x8
+
+prototype void vp9_sad16x16x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad16x16x8 sse4
+
+prototype void vp9_sad16x8x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad16x8x8 sse4
+
+prototype void vp9_sad8x16x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad8x16x8 sse4
+
+prototype void vp9_sad8x8x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad8x8x8 sse4
+
+prototype void vp9_sad8x4x8 "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array"
+specialize vp9_sad8x4x8
+
+prototype void vp9_sad4x8x8 "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array"
+specialize vp9_sad4x8x8
+
+prototype void vp9_sad4x4x8 "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array"
+specialize vp9_sad4x4x8 sse4
+
+prototype void vp9_sad64x64x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad64x64x4d sse2
+
+prototype void vp9_sad32x64x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad32x64x4d sse2
+
+prototype void vp9_sad64x32x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad64x32x4d sse2
+
+prototype void vp9_sad32x16x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad32x16x4d sse2
+
+prototype void vp9_sad16x32x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad16x32x4d sse2
+
+prototype void vp9_sad32x32x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad32x32x4d sse2
+
+prototype void vp9_sad16x16x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad16x16x4d sse2
+
+prototype void vp9_sad16x8x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad16x8x4d sse2
+
+prototype void vp9_sad8x16x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad8x16x4d sse2
+
+prototype void vp9_sad8x8x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad8x8x4d sse2
+
+# TODO(jingning): need to convert these 4x8/8x4 functions into sse2 form
+prototype void vp9_sad8x4x4d "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, unsigned int *sad_array"
+specialize vp9_sad8x4x4d sse2
+
+prototype void vp9_sad4x8x4d "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, unsigned int *sad_array"
+specialize vp9_sad4x8x4d sse
+
+prototype void vp9_sad4x4x4d "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array"
+specialize vp9_sad4x4x4d sse
+
+prototype unsigned int vp9_sub_pixel_mse16x16 "const uint8_t *src_ptr, int  src_pixels_per_line, int  xoffset, int  yoffset, const uint8_t *dst_ptr, int dst_pixels_per_line, unsigned int *sse"
+specialize vp9_sub_pixel_mse16x16 sse2 mmx
+
+prototype unsigned int vp9_mse16x16 "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse"
+specialize vp9_mse16x16 mmx sse2
+vp9_mse16x16_sse2=vp9_mse16x16_wmt
+
+prototype unsigned int vp9_mse8x16 "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse"
+specialize vp9_mse8x16
+
+prototype unsigned int vp9_mse16x8 "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse"
+specialize vp9_mse16x8
+
+prototype unsigned int vp9_mse8x8 "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse"
+specialize vp9_mse8x8
+
+prototype unsigned int vp9_sub_pixel_mse64x64 "const uint8_t *src_ptr, int  source_stride, int  xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_mse64x64
+
+prototype unsigned int vp9_sub_pixel_mse32x32 "const uint8_t *src_ptr, int  source_stride, int  xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
+specialize vp9_sub_pixel_mse32x32
+
+prototype unsigned int vp9_get_mb_ss "const int16_t *"
+specialize vp9_get_mb_ss mmx sse2
+# ENCODEMB INVOKE
+
+prototype int vp9_block_error "int16_t *coeff, int16_t *dqcoeff, int block_size"
+specialize vp9_block_error mmx sse2
+vp9_block_error_sse2=vp9_block_error_xmm
+
+#
+# Structured Similarity (SSIM)
+#
+if [ "$CONFIG_INTERNAL_STATS" = "yes" ]; then
+    [ $arch = "x86_64" ] && sse2_on_x86_64=sse2
+
+    prototype void vp9_ssim_parms_8x8 "uint8_t *s, int sp, uint8_t *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"
+    specialize vp9_ssim_parms_8x8 $sse2_on_x86_64
+
+    prototype void vp9_ssim_parms_16x16 "uint8_t *s, int sp, uint8_t *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"
+    specialize vp9_ssim_parms_16x16 $sse2_on_x86_64
+fi
+
+# fdct functions
+prototype void vp9_short_fht4x4 "int16_t *InputData, int16_t *OutputData, int pitch, int tx_type"
+specialize vp9_short_fht4x4
+
+prototype void vp9_short_fht8x8 "int16_t *InputData, int16_t *OutputData, int pitch, int tx_type"
+specialize vp9_short_fht8x8
+
+prototype void vp9_short_fht16x16 "int16_t *InputData, int16_t *OutputData, int pitch, int tx_type"
+specialize vp9_short_fht16x16
+
+prototype void vp9_short_fdct8x8 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct8x8 sse2
+
+prototype void vp9_short_fdct4x4 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct4x4 sse2
+
+prototype void vp9_short_fdct8x4 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct8x4 sse2
+
+prototype void vp9_short_fdct32x32 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct32x32
+
+prototype void vp9_short_fdct32x32_rd "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct32x32_rd
+
+prototype void vp9_short_fdct16x16 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_fdct16x16 sse2
+
+prototype void vp9_short_walsh4x4 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_walsh4x4
+
+prototype void vp9_short_walsh8x4 "int16_t *InputData, int16_t *OutputData, int pitch"
+specialize vp9_short_walsh8x4
+
+#
+# Motion search
+#
+prototype int vp9_full_search_sad "struct macroblock *x, union int_mv *ref_mv, int sad_per_bit, int distance, struct vp9_variance_vtable *fn_ptr, DEC_MVCOSTS, union int_mv *center_mv, int n"
+specialize vp9_full_search_sad sse3 sse4_1
+vp9_full_search_sad_sse3=vp9_full_search_sadx3
+vp9_full_search_sad_sse4_1=vp9_full_search_sadx8
+
+prototype int vp9_refining_search_sad "struct macroblock *x, union int_mv *ref_mv, int sad_per_bit, int distance, struct vp9_variance_vtable *fn_ptr, DEC_MVCOSTS, union int_mv *center_mv"
+specialize vp9_refining_search_sad sse3
+vp9_refining_search_sad_sse3=vp9_refining_search_sadx4
+
+prototype int vp9_diamond_search_sad "struct macroblock *x, union int_mv *ref_mv, union int_mv *best_mv, int search_param, int sad_per_bit, int *num00, struct vp9_variance_vtable *fn_ptr, DEC_MVCOSTS, union int_mv *center_mv"
+specialize vp9_diamond_search_sad sse3
+vp9_diamond_search_sad_sse3=vp9_diamond_search_sadx4
+
+prototype void vp9_temporal_filter_apply "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_size, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count"
+specialize vp9_temporal_filter_apply sse2
+
+prototype void vp9_yv12_copy_partial_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int fraction"
+specialize vp9_yv12_copy_partial_frame
+
+
+fi
+# end encoder functions
diff --git a/libvpx/vp9/common/vp9_sadmxn.h b/libvpx/vp9/common/vp9_sadmxn.h
new file mode 100644
index 0000000..b2dfd63
--- /dev/null
+++ b/libvpx/vp9/common/vp9_sadmxn.h
@@ -0,0 +1,38 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_SADMXN_H_
+#define VP9_COMMON_VP9_SADMXN_H_
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+
+static INLINE unsigned int sad_mx_n_c(const uint8_t *src_ptr,
+                                      int src_stride,
+                                      const uint8_t *ref_ptr,
+                                      int ref_stride,
+                                      int m,
+                                      int n) {
+  int r, c;
+  unsigned int sad = 0;
+
+  for (r = 0; r < n; r++) {
+    for (c = 0; c < m; c++) {
+      sad += abs(src_ptr[c] - ref_ptr[c]);
+    }
+
+    src_ptr += src_stride;
+    ref_ptr += ref_stride;
+  }
+
+  return sad;
+}
+
+#endif  // VP9_COMMON_VP9_SADMXN_H_
diff --git a/libvpx/vp9/common/vp9_seg_common.c b/libvpx/vp9/common/vp9_seg_common.c
new file mode 100644
index 0000000..df7747c
--- /dev/null
+++ b/libvpx/vp9/common/vp9_seg_common.c
@@ -0,0 +1,80 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_seg_common.h"
+
+static const int seg_feature_data_signed[SEG_LVL_MAX] = { 1, 1, 0, 0 };
+static const int seg_feature_data_max[SEG_LVL_MAX] = { MAXQ, 63, 3, 0 };
+
+// These functions provide access to new segment level features.
+// Eventually these function may be "optimized out" but for the moment,
+// the coding mechanism is still subject to change so these provide a
+// convenient single point of change.
+
+int vp9_segfeature_active(const MACROBLOCKD *xd, int segment_id,
+                          SEG_LVL_FEATURES feature_id) {
+  return xd->segmentation_enabled &&
+         (xd->segment_feature_mask[segment_id] & (1 << feature_id));
+}
+
+void vp9_clearall_segfeatures(MACROBLOCKD *xd) {
+  vpx_memset(xd->segment_feature_data, 0, sizeof(xd->segment_feature_data));
+  vpx_memset(xd->segment_feature_mask, 0, sizeof(xd->segment_feature_mask));
+}
+
+void vp9_enable_segfeature(MACROBLOCKD *xd, int segment_id,
+                           SEG_LVL_FEATURES feature_id) {
+  xd->segment_feature_mask[segment_id] |= 1 << feature_id;
+}
+
+void vp9_disable_segfeature(MACROBLOCKD *xd, int segment_id,
+                            SEG_LVL_FEATURES feature_id) {
+  xd->segment_feature_mask[segment_id] &= ~(1 << feature_id);
+}
+
+int vp9_seg_feature_data_max(SEG_LVL_FEATURES feature_id) {
+  return seg_feature_data_max[feature_id];
+}
+
+int vp9_is_segfeature_signed(SEG_LVL_FEATURES feature_id) {
+  return seg_feature_data_signed[feature_id];
+}
+
+void vp9_clear_segdata(MACROBLOCKD *xd, int segment_id,
+                       SEG_LVL_FEATURES feature_id) {
+  xd->segment_feature_data[segment_id][feature_id] = 0;
+}
+
+void vp9_set_segdata(MACROBLOCKD *xd, int segment_id,
+                     SEG_LVL_FEATURES feature_id, int seg_data) {
+  assert(seg_data <= seg_feature_data_max[feature_id]);
+  if (seg_data < 0) {
+    assert(seg_feature_data_signed[feature_id]);
+    assert(-seg_data <= seg_feature_data_max[feature_id]);
+  }
+
+  xd->segment_feature_data[segment_id][feature_id] = seg_data;
+}
+
+int vp9_get_segdata(const MACROBLOCKD *xd, int segment_id,
+                    SEG_LVL_FEATURES feature_id) {
+  return xd->segment_feature_data[segment_id][feature_id];
+}
+
+
+const vp9_tree_index vp9_segment_tree[14] = {
+  2,  4,  6,  8, 10, 12,
+  0, -1, -2, -3, -4, -5, -6, -7
+};
+
+
+// TBD? Functions to read and write segment data with range / validity checking
diff --git a/libvpx/vp9/common/vp9_seg_common.h b/libvpx/vp9/common/vp9_seg_common.h
new file mode 100644
index 0000000..74ba03c
--- /dev/null
+++ b/libvpx/vp9/common/vp9_seg_common.h
@@ -0,0 +1,51 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_blockd.h"
+
+#ifndef VP9_COMMON_VP9_SEG_COMMON_H_
+#define VP9_COMMON_VP9_SEG_COMMON_H_
+
+int vp9_segfeature_active(const MACROBLOCKD *xd,
+                          int segment_id,
+                          SEG_LVL_FEATURES feature_id);
+
+void vp9_clearall_segfeatures(MACROBLOCKD *xd);
+
+void vp9_enable_segfeature(MACROBLOCKD *xd,
+                           int segment_id,
+                           SEG_LVL_FEATURES feature_id);
+
+void vp9_disable_segfeature(MACROBLOCKD *xd,
+                            int segment_id,
+                            SEG_LVL_FEATURES feature_id);
+
+int vp9_seg_feature_data_max(SEG_LVL_FEATURES feature_id);
+
+int vp9_is_segfeature_signed(SEG_LVL_FEATURES feature_id);
+
+void vp9_clear_segdata(MACROBLOCKD *xd,
+                       int segment_id,
+                       SEG_LVL_FEATURES feature_id);
+
+void vp9_set_segdata(MACROBLOCKD *xd,
+                     int segment_id,
+                     SEG_LVL_FEATURES feature_id,
+                     int seg_data);
+
+int vp9_get_segdata(const MACROBLOCKD *xd,
+                    int segment_id,
+                    SEG_LVL_FEATURES feature_id);
+
+extern const vp9_tree_index vp9_segment_tree[14];
+
+#endif  // VP9_COMMON_VP9_SEG_COMMON_H_
+
diff --git a/libvpx/vp9/common/vp9_subpelvar.h b/libvpx/vp9/common/vp9_subpelvar.h
new file mode 100644
index 0000000..ad674f1
--- /dev/null
+++ b/libvpx/vp9/common/vp9_subpelvar.h
@@ -0,0 +1,148 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_SUBPELVAR_H_
+#define VP9_COMMON_VP9_SUBPELVAR_H_
+
+#include "vp9/common/vp9_filter.h"
+
+static void variance(const uint8_t *src_ptr,
+                     int  source_stride,
+                     const uint8_t *ref_ptr,
+                     int  recon_stride,
+                     int  w,
+                     int  h,
+                     unsigned int *sse,
+                     int *sum) {
+  int i, j;
+  int diff;
+
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      diff = src_ptr[j] - ref_ptr[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+
+    src_ptr += source_stride;
+    ref_ptr += recon_stride;
+  }
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : filter_block2d_bil_first_pass
+ *
+ *  INPUTS        : uint8_t  *src_ptr          : Pointer to source block.
+ *                  uint32_t src_pixels_per_line : Stride of input block.
+ *                  uint32_t pixel_step        : Offset between filter input samples (see notes).
+ *                  uint32_t output_height     : Input block height.
+ *                  uint32_t output_width      : Input block width.
+ *                  int32_t  *vp9_filter          : Array of 2 bi-linear filter taps.
+ *
+ *  OUTPUTS       : int32_t *output_ptr        : Pointer to filtered block.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Applies a 1-D 2-tap bi-linear filter to the source block in
+ *                  either horizontal or vertical direction to produce the
+ *                  filtered output block. Used to implement first-pass
+ *                  of 2-D separable filter.
+ *
+ *  SPECIAL NOTES : Produces int32_t output to retain precision for next pass.
+ *                  Two filter taps should sum to VP9_FILTER_WEIGHT.
+ *                  pixel_step defines whether the filter is applied
+ *                  horizontally (pixel_step=1) or vertically (pixel_step=stride).
+ *                  It defines the offset required to move from one input
+ *                  to the next.
+ *
+ ****************************************************************************/
+static void var_filter_block2d_bil_first_pass(const uint8_t *src_ptr,
+                                              uint16_t *output_ptr,
+                                              unsigned int src_pixels_per_line,
+                                              int pixel_step,
+                                              unsigned int output_height,
+                                              unsigned int output_width,
+                                              const int16_t *vp9_filter) {
+  unsigned int i, j;
+
+  for (i = 0; i < output_height; i++) {
+    for (j = 0; j < output_width; j++) {
+      // Apply bilinear filter
+      output_ptr[j] = (((int)src_ptr[0]          * vp9_filter[0]) +
+                       ((int)src_ptr[pixel_step] * vp9_filter[1]) +
+                       (VP9_FILTER_WEIGHT / 2)) >> VP9_FILTER_SHIFT;
+      src_ptr++;
+    }
+
+    // Next row...
+    src_ptr    += src_pixels_per_line - output_width;
+    output_ptr += output_width;
+  }
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : filter_block2d_bil_second_pass
+ *
+ *  INPUTS        : int32_t  *src_ptr          : Pointer to source block.
+ *                  uint32_t src_pixels_per_line : Stride of input block.
+ *                  uint32_t pixel_step        : Offset between filter input samples (see notes).
+ *                  uint32_t output_height     : Input block height.
+ *                  uint32_t output_width      : Input block width.
+ *                  int32_t  *vp9_filter          : Array of 2 bi-linear filter taps.
+ *
+ *  OUTPUTS       : uint16_t *output_ptr       : Pointer to filtered block.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Applies a 1-D 2-tap bi-linear filter to the source block in
+ *                  either horizontal or vertical direction to produce the
+ *                  filtered output block. Used to implement second-pass
+ *                  of 2-D separable filter.
+ *
+ *  SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass.
+ *                  Two filter taps should sum to VP9_FILTER_WEIGHT.
+ *                  pixel_step defines whether the filter is applied
+ *                  horizontally (pixel_step=1) or vertically (pixel_step=stride).
+ *                  It defines the offset required to move from one input
+ *                  to the next.
+ *
+ ****************************************************************************/
+static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
+                                               uint8_t *output_ptr,
+                                               unsigned int src_pixels_per_line,
+                                               unsigned int pixel_step,
+                                               unsigned int output_height,
+                                               unsigned int output_width,
+                                               const int16_t *vp9_filter) {
+  unsigned int  i, j;
+  int  Temp;
+
+  for (i = 0; i < output_height; i++) {
+    for (j = 0; j < output_width; j++) {
+      // Apply filter
+      Temp = ((int)src_ptr[0]         * vp9_filter[0]) +
+             ((int)src_ptr[pixel_step] * vp9_filter[1]) +
+             (VP9_FILTER_WEIGHT / 2);
+      output_ptr[j] = (unsigned int)(Temp >> VP9_FILTER_SHIFT);
+      src_ptr++;
+    }
+
+    // Next row...
+    src_ptr    += src_pixels_per_line - output_width;
+    output_ptr += output_width;
+  }
+}
+
+#endif  // VP9_COMMON_VP9_SUBPELVAR_H_
diff --git a/libvpx/vp9/common/vp9_systemdependent.h b/libvpx/vp9/common/vp9_systemdependent.h
new file mode 100644
index 0000000..1b9147e
--- /dev/null
+++ b/libvpx/vp9/common/vp9_systemdependent.h
@@ -0,0 +1,39 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
+#define VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
+
+#ifdef _MSC_VER
+#include <math.h>
+#endif
+
+#include "./vpx_config.h"
+#if ARCH_X86 || ARCH_X86_64
+void vpx_reset_mmx_state(void);
+#define vp9_clear_system_state() vpx_reset_mmx_state()
+#else
+#define vp9_clear_system_state()
+#endif
+
+#ifdef _MSC_VER
+// round is not defined in MSVC
+static int round(double x) {
+  if (x < 0)
+    return (int)ceil(x - 0.5);
+  else
+    return (int)floor(x + 0.5);
+}
+#endif
+
+struct VP9Common;
+void vp9_machine_specific_config(struct VP9Common *);
+
+#endif  // VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
diff --git a/libvpx/vp9/common/vp9_tapify.py b/libvpx/vp9/common/vp9_tapify.py
new file mode 100644
index 0000000..99529cf
--- /dev/null
+++ b/libvpx/vp9/common/vp9_tapify.py
@@ -0,0 +1,106 @@
+"""
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+"""
+#!/usr/bin/env python
+import sys,string,os,re,math,numpy
+scale = 2**16
+def dist(p1,p2):
+  x1,y1 = p1
+  x2,y2 = p2
+  if x1==x2 and y1==y2 :
+    return 1.0 
+  return 1/ math.sqrt((x1-x2)*(x1-x2)+(y1-y2)*(y1-y2))
+
+def gettaps(p):
+  def l(b):
+    return int(math.floor(b))
+  def h(b):
+    return int(math.ceil(b))
+  def t(b,p,s):
+    return int((scale*dist(b,p)+s/2)/s)
+  r,c = p
+  ul=[l(r),l(c)]
+  ur=[l(r),h(c)]
+  ll=[h(r),l(c)]
+  lr=[h(r),h(c)]
+  sum = dist(ul,p)+dist(ur,p)+dist(ll,p)+dist(lr,p)
+  t4 = scale - t(ul,p,sum) - t(ur,p,sum) - t(ll,p,sum);
+  return [[ul,t(ul,p,sum)],[ur,t(ur,p,sum)],
+          [ll,t(ll,p,sum)],[lr,t4]]
+
+def print_mb_taps(angle,blocksize):
+  theta = angle / 57.2957795;
+  affine = [[math.cos(theta),-math.sin(theta)],
+            [math.sin(theta),math.cos(theta)]]
+  radius = (float(blocksize)-1)/2
+  print " // angle of",angle,"degrees"
+  for y in range(blocksize) :
+    for x in range(blocksize) :
+      r,c = numpy.dot(affine,[y-radius, x-radius])
+      tps = gettaps([r+radius,c+radius])
+      for t in tps :
+        p,t = t
+        tr,tc = p
+        print " %2d, %2d, %5d, " % (tr,tc,t,),
+      print " // %2d,%2d " % (y,x)
+
+i=float(sys.argv[1])
+while  i <= float(sys.argv[2]) :
+  print_mb_taps(i,float(sys.argv[4]))
+  i=i+float(sys.argv[3])
+"""
+
+taps = []
+pt=dict()
+ptr=dict()
+for y in range(16) :
+  for x in range(16) :
+    r,c = numpy.dot(affine,[y-7.5, x-7.5])
+    tps = gettaps([r+7.5,c+7.5])
+    j=0
+    for tp in tps : 
+      p,i = tp
+      r,c = p
+      pt[y,x,j]= [p,i]
+      try: 
+        ptr[r,j,c].append([y,x])
+      except:
+        ptr[r,j,c]=[[y,x]]
+      j = j+1 
+
+for key in sorted(pt.keys()) :
+  print key,pt[key]
+
+lr = -99
+lj = -99 
+lc = 0
+
+shuf=""
+mask=""
+for r,j,c in sorted(ptr.keys()) :
+  for y,x in ptr[r,j,c] :
+    if lr != r or lj != j :
+      print "shuf_"+str(lr)+"_"+str(lj)+"_"+shuf.ljust(16,"0"), lc
+      shuf=""
+      lc = 0
+    for i in range(lc,c-1) :
+      shuf = shuf +"0"
+    shuf = shuf + hex(x)[2]
+    lc =c
+    break
+  lr = r
+  lj = j
+#  print r,j,c,ptr[r,j,c]    
+#  print 
+
+for r,j,c in sorted(ptr.keys()) :
+  for y,x in ptr[r,j,c] :
+    print r,j,c,y,x 
+    break
+"""
diff --git a/libvpx/vp9/common/vp9_textblit.c b/libvpx/vp9/common/vp9_textblit.c
new file mode 100644
index 0000000..60e95e0
--- /dev/null
+++ b/libvpx/vp9/common/vp9_textblit.c
@@ -0,0 +1,120 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdlib.h>
+
+#include "vp9/common/vp9_textblit.h"
+
+static const int font[] = {
+  0x0, 0x5C00, 0x8020, 0xAFABEA, 0xD7EC0, 0x1111111, 0x1855740, 0x18000,
+  0x45C0, 0x74400, 0x51140, 0x23880, 0xC4000, 0x21080, 0x80000, 0x111110,
+  0xE9D72E, 0x87E40, 0x12AD732, 0xAAD62A, 0x4F94C4, 0x4D6B7, 0x456AA,
+  0x3E8423, 0xAAD6AA, 0xAAD6A2, 0x2800, 0x2A00, 0x8A880, 0x52940, 0x22A20,
+  0x15422, 0x6AD62E, 0x1E4A53E, 0xAAD6BF, 0x8C62E, 0xE8C63F, 0x118D6BF,
+  0x1094BF, 0xCAC62E, 0x1F2109F, 0x118FE31, 0xF8C628, 0x8A89F, 0x108421F,
+  0x1F1105F, 0x1F4105F, 0xE8C62E, 0x2294BF, 0x164C62E, 0x12694BF, 0x8AD6A2,
+  0x10FC21, 0x1F8421F, 0x744107, 0xF8220F, 0x1151151, 0x117041, 0x119D731,
+  0x47E0, 0x1041041, 0xFC400, 0x10440, 0x1084210, 0x820
+};
+
+static void plot(int x, int y, unsigned char *image, int pitch) {
+  image[x + y * pitch] ^= 255;
+}
+
+void vp9_blit_text(const char *msg, unsigned char *address, const int pitch) {
+  int letter_bitmap;
+  unsigned char *output_pos = address;
+  int colpos = 0;
+
+  while (msg[colpos] != 0) {
+    char letter = msg[colpos];
+    int fontcol, fontrow;
+
+    if (letter <= 'Z' && letter >= ' ')
+      letter_bitmap = font[letter - ' '];
+    else if (letter <= 'z' && letter >= 'a')
+      letter_bitmap = font[letter - 'a' + 'A' - ' '];
+    else
+      letter_bitmap = font[0];
+
+    for (fontcol = 6; fontcol >= 0; fontcol--)
+      for (fontrow = 0; fontrow < 5; fontrow++)
+        output_pos[fontrow * pitch + fontcol] =
+          ((letter_bitmap >> (fontcol * 5)) & (1 << fontrow) ? 255 : 0);
+
+    output_pos += 7;
+    colpos++;
+  }
+}
+
+
+
+/* Bresenham line algorithm */
+void vp9_blit_line(int x0, int x1, int y0, int y1, unsigned char *image,
+                   int pitch) {
+  int steep = abs(y1 - y0) > abs(x1 - x0);
+  int deltax, deltay;
+  int error, ystep, y, x;
+
+  if (steep) {
+    int t;
+    t = x0;
+    x0 = y0;
+    y0 = t;
+
+    t = x1;
+    x1 = y1;
+    y1 = t;
+  }
+
+  if (x0 > x1) {
+    int t;
+    t = x0;
+    x0 = x1;
+    x1 = t;
+
+    t = y0;
+    y0 = y1;
+    y1 = t;
+  }
+
+  deltax = x1 - x0;
+  deltay = abs(y1 - y0);
+  error  = deltax / 2;
+
+  y = y0;
+
+  if (y0 < y1)
+    ystep = 1;
+  else
+    ystep = -1;
+
+  if (steep) {
+    for (x = x0; x <= x1; x++) {
+      plot(y, x, image, pitch);
+
+      error = error - deltay;
+      if (error < 0) {
+        y = y + ystep;
+        error = error + deltax;
+      }
+    }
+  } else {
+    for (x = x0; x <= x1; x++) {
+      plot(x, y, image, pitch);
+
+      error = error - deltay;
+      if (error < 0) {
+        y = y + ystep;
+        error = error + deltax;
+      }
+    }
+  }
+}
diff --git a/libvpx/vp9/common/vp9_textblit.h b/libvpx/vp9/common/vp9_textblit.h
new file mode 100644
index 0000000..c968628
--- /dev/null
+++ b/libvpx/vp9/common/vp9_textblit.h
@@ -0,0 +1,19 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_TEXTBLIT_H_
+#define VP9_COMMON_VP9_TEXTBLIT_H_
+
+void vp9_blit_text(const char *msg, unsigned char *address, int pitch);
+
+void vp9_blit_line(int x0, int x1, int y0, int y1, unsigned char *image,
+                   int pitch);
+
+#endif  // VP9_COMMON_VP9_TEXTBLIT_H_
diff --git a/libvpx/vp9/common/vp9_tile_common.c b/libvpx/vp9/common/vp9_tile_common.c
new file mode 100644
index 0000000..95296ad
--- /dev/null
+++ b/libvpx/vp9/common/vp9_tile_common.c
@@ -0,0 +1,63 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_tile_common.h"
+
+#define MIN_TILE_WIDTH 256
+#define MAX_TILE_WIDTH 4096
+#define MIN_TILE_WIDTH_SBS (MIN_TILE_WIDTH >> 6)
+#define MAX_TILE_WIDTH_SBS (MAX_TILE_WIDTH >> 6)
+
+static void vp9_get_tile_offsets(VP9_COMMON *cm, int *min_tile_off,
+                                 int *max_tile_off, int tile_idx,
+                                 int log2_n_tiles, int n_mis) {
+  const int n_sbs = (n_mis + 7) >> 3;
+  const int sb_off1 =  (tile_idx      * n_sbs) >> log2_n_tiles;
+  const int sb_off2 = ((tile_idx + 1) * n_sbs) >> log2_n_tiles;
+
+  *min_tile_off = MIN(sb_off1 << 3, n_mis);
+  *max_tile_off = MIN(sb_off2 << 3, n_mis);
+}
+
+void vp9_get_tile_col_offsets(VP9_COMMON *cm, int tile_col_idx) {
+  cm->cur_tile_col_idx = tile_col_idx;
+  vp9_get_tile_offsets(cm, &cm->cur_tile_mi_col_start,
+                       &cm->cur_tile_mi_col_end, tile_col_idx,
+                       cm->log2_tile_columns, cm->mi_cols);
+}
+
+void vp9_get_tile_row_offsets(VP9_COMMON *cm, int tile_row_idx) {
+  cm->cur_tile_row_idx = tile_row_idx;
+  vp9_get_tile_offsets(cm, &cm->cur_tile_mi_row_start,
+                       &cm->cur_tile_mi_row_end, tile_row_idx,
+                       cm->log2_tile_rows, cm->mi_rows);
+}
+
+
+void vp9_get_tile_n_bits(VP9_COMMON *cm, int *min_log2_n_tiles_ptr,
+                         int *delta_log2_n_tiles) {
+  const int sb_cols = (cm->mb_cols + 3) >> 2;
+  int min_log2_n_tiles, max_log2_n_tiles;
+
+  for (max_log2_n_tiles = 0;
+       (sb_cols >> max_log2_n_tiles) >= MIN_TILE_WIDTH_SBS;
+       max_log2_n_tiles++) {}
+  max_log2_n_tiles--;
+  if (max_log2_n_tiles <  0)
+    max_log2_n_tiles = 0;
+
+  for (min_log2_n_tiles = 0;
+       (MAX_TILE_WIDTH_SBS << min_log2_n_tiles) < sb_cols;
+       min_log2_n_tiles++) {}
+
+  assert(max_log2_n_tiles >= min_log2_n_tiles);
+  *min_log2_n_tiles_ptr = min_log2_n_tiles;
+  *delta_log2_n_tiles = max_log2_n_tiles - min_log2_n_tiles;
+}
diff --git a/libvpx/vp9/common/vp9_tile_common.h b/libvpx/vp9/common/vp9_tile_common.h
new file mode 100644
index 0000000..7ea3772
--- /dev/null
+++ b/libvpx/vp9/common/vp9_tile_common.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_TILE_COMMON_H_
+#define VP9_COMMON_VP9_TILE_COMMON_H_
+
+#include "vp9/common/vp9_onyxc_int.h"
+
+void vp9_get_tile_col_offsets(VP9_COMMON *cm, int tile_col_idx);
+
+void vp9_get_tile_row_offsets(VP9_COMMON *cm, int tile_row_idx);
+
+void vp9_get_tile_n_bits(VP9_COMMON *cm, int *min_log2_n_tiles,
+                         int *delta_log2_n_tiles);
+
+#endif  // VP9_COMMON_VP9_TILE_COMMON_H_
diff --git a/libvpx/vp9/common/vp9_treecoder.c b/libvpx/vp9/common/vp9_treecoder.c
new file mode 100644
index 0000000..531fa75
--- /dev/null
+++ b/libvpx/vp9/common/vp9_treecoder.c
@@ -0,0 +1,78 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_config.h"
+
+#if defined(CONFIG_DEBUG) && CONFIG_DEBUG
+#include <assert.h>
+#endif
+
+#include "vp9/common/vp9_treecoder.h"
+
+static void tree2tok(struct vp9_token *const p, vp9_tree t,
+                    int i, int v, int l) {
+  v += v;
+  ++l;
+
+  do {
+    const vp9_tree_index j = t[i++];
+
+    if (j <= 0) {
+      p[-j].value = v;
+      p[-j].len = l;
+    } else
+      tree2tok(p, t, j, v, l);
+  } while (++v & 1);
+}
+
+void vp9_tokens_from_tree(struct vp9_token *p, vp9_tree t) {
+  tree2tok(p, t, 0, 0, 0);
+}
+
+void vp9_tokens_from_tree_offset(struct vp9_token *p, vp9_tree t,
+                                 int offset) {
+  tree2tok(p - offset, t, 0, 0, 0);
+}
+
+static unsigned int convert_distribution(unsigned int i,
+                                         vp9_tree tree,
+                                         vp9_prob probs[],
+                                         unsigned int branch_ct[][2],
+                                         const unsigned int num_events[],
+                                         unsigned int tok0_offset) {
+  unsigned int left, right;
+
+  if (tree[i] <= 0) {
+    left = num_events[-tree[i] - tok0_offset];
+  } else {
+    left = convert_distribution(tree[i], tree, probs, branch_ct,
+                                num_events, tok0_offset);
+  }
+  if (tree[i + 1] <= 0)
+    right = num_events[-tree[i + 1] - tok0_offset];
+  else
+    right = convert_distribution(tree[i + 1], tree, probs, branch_ct,
+                                 num_events, tok0_offset);
+
+  probs[i>>1] = get_binary_prob(left, right);
+  branch_ct[i>>1][0] = left;
+  branch_ct[i>>1][1] = right;
+  return left + right;
+}
+
+void vp9_tree_probs_from_distribution(
+  vp9_tree tree,
+  vp9_prob probs          [ /* n-1 */ ],
+  unsigned int branch_ct       [ /* n-1 */ ] [2],
+  const unsigned int num_events[ /* n */ ],
+  unsigned int tok0_offset) {
+  convert_distribution(0, tree, probs, branch_ct, num_events, tok0_offset);
+}
diff --git a/libvpx/vp9/common/vp9_treecoder.h b/libvpx/vp9/common/vp9_treecoder.h
new file mode 100644
index 0000000..ebcd411
--- /dev/null
+++ b/libvpx/vp9/common/vp9_treecoder.h
@@ -0,0 +1,82 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_TREECODER_H_
+#define VP9_COMMON_VP9_TREECODER_H_
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+
+typedef uint8_t vp9_prob;
+
+#define vp9_prob_half ((vp9_prob) 128)
+
+typedef int8_t vp9_tree_index;
+
+#define vp9_complement(x) (255 - x)
+
+/* We build coding trees compactly in arrays.
+   Each node of the tree is a pair of vp9_tree_indices.
+   Array index often references a corresponding probability table.
+   Index <= 0 means done encoding/decoding and value = -Index,
+   Index > 0 means need another bit, specification at index.
+   Nonnegative indices are always even;  processing begins at node 0. */
+
+typedef const vp9_tree_index vp9_tree[], *vp9_tree_p;
+
+struct vp9_token {
+  int value;
+  int len;
+};
+
+/* Construct encoding array from tree. */
+
+void vp9_tokens_from_tree(struct vp9_token*, vp9_tree);
+void vp9_tokens_from_tree_offset(struct vp9_token*, vp9_tree, int offset);
+
+/* Convert array of token occurrence counts into a table of probabilities
+   for the associated binary encoding tree.  Also writes count of branches
+   taken for each node on the tree; this facilitiates decisions as to
+   probability updates. */
+
+void vp9_tree_probs_from_distribution(vp9_tree tree,
+                                      vp9_prob probs[ /* n - 1 */ ],
+                                      unsigned int branch_ct[ /* n - 1 */ ][2],
+                                      const unsigned int num_events[ /* n */ ],
+                                      unsigned int tok0_offset);
+
+static INLINE vp9_prob clip_prob(int p) {
+  return (p > 255) ? 255u : (p < 1) ? 1u : p;
+}
+
+// int64 is not needed for normal frame level calculations.
+// However when outputing entropy stats accumulated over many frames
+// or even clips we can overflow int math.
+#ifdef ENTROPY_STATS
+static INLINE vp9_prob get_prob(int num, int den) {
+  return (den == 0) ? 128u : clip_prob(((int64_t)num * 256 + (den >> 1)) / den);
+}
+#else
+static INLINE vp9_prob get_prob(int num, int den) {
+  return (den == 0) ? 128u : clip_prob((num * 256 + (den >> 1)) / den);
+}
+#endif
+
+static INLINE vp9_prob get_binary_prob(int n0, int n1) {
+  return get_prob(n0, n0 + n1);
+}
+
+/* this function assumes prob1 and prob2 are already within [1,255] range */
+static INLINE vp9_prob weighted_prob(int prob1, int prob2, int factor) {
+  return ROUND_POWER_OF_TWO(prob1 * (256 - factor) + prob2 * factor, 8);
+}
+
+#endif  // VP9_COMMON_VP9_TREECODER_H_
diff --git a/libvpx/vp9/common/x86/vp9_asm_stubs.c b/libvpx/vp9/common/x86/vp9_asm_stubs.c
new file mode 100644
index 0000000..2b66834
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_asm_stubs.c
@@ -0,0 +1,318 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vpx_ports/mem.h"
+///////////////////////////////////////////////////////////////////////////
+// the mmx function that does the bilinear filtering and var calculation //
+// int one pass                                                          //
+///////////////////////////////////////////////////////////////////////////
+DECLARE_ALIGNED(16, const short, vp9_bilinear_filters_mmx[16][8]) = {
+  { 128, 128, 128, 128,  0,  0,  0,  0 },
+  { 120, 120, 120, 120,  8,  8,  8,  8 },
+  { 112, 112, 112, 112, 16, 16, 16, 16 },
+  { 104, 104, 104, 104, 24, 24, 24, 24 },
+  {  96, 96, 96, 96, 32, 32, 32, 32 },
+  {  88, 88, 88, 88, 40, 40, 40, 40 },
+  {  80, 80, 80, 80, 48, 48, 48, 48 },
+  {  72, 72, 72, 72, 56, 56, 56, 56 },
+  {  64, 64, 64, 64, 64, 64, 64, 64 },
+  {  56, 56, 56, 56, 72, 72, 72, 72 },
+  {  48, 48, 48, 48, 80, 80, 80, 80 },
+  {  40, 40, 40, 40, 88, 88, 88, 88 },
+  {  32, 32, 32, 32, 96, 96, 96, 96 },
+  {  24, 24, 24, 24, 104, 104, 104, 104 },
+  {  16, 16, 16, 16, 112, 112, 112, 112 },
+  {   8,  8,  8,  8, 120, 120, 120, 120 }
+};
+
+#if HAVE_SSSE3
+void vp9_filter_block1d16_v8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d16_h8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d8_v8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d8_h8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d4_v8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d4_h8_ssse3(const unsigned char *src_ptr,
+                                   const unsigned int src_pitch,
+                                   unsigned char *output_ptr,
+                                   unsigned int out_pitch,
+                                   unsigned int output_height,
+                                   const short *filter);
+
+void vp9_filter_block1d16_v8_avg_ssse3(const unsigned char *src_ptr,
+                                       const unsigned int src_pitch,
+                                       unsigned char *output_ptr,
+                                       unsigned int out_pitch,
+                                       unsigned int output_height,
+                                       const short *filter);
+
+void vp9_filter_block1d16_h8_avg_ssse3(const unsigned char *src_ptr,
+                                       const unsigned int src_pitch,
+                                       unsigned char *output_ptr,
+                                       unsigned int out_pitch,
+                                       unsigned int output_height,
+                                       const short *filter);
+
+void vp9_filter_block1d8_v8_avg_ssse3(const unsigned char *src_ptr,
+                                     const unsigned int src_pitch,
+                                     unsigned char *output_ptr,
+                                     unsigned int out_pitch,
+                                     unsigned int output_height,
+                                     const short *filter);
+
+void vp9_filter_block1d8_h8_avg_ssse3(const unsigned char *src_ptr,
+                                     const unsigned int src_pitch,
+                                     unsigned char *output_ptr,
+                                     unsigned int out_pitch,
+                                     unsigned int output_height,
+                                     const short *filter);
+
+void vp9_filter_block1d4_v8_avg_ssse3(const unsigned char *src_ptr,
+                                     const unsigned int src_pitch,
+                                     unsigned char *output_ptr,
+                                     unsigned int out_pitch,
+                                     unsigned int output_height,
+                                     const short *filter);
+
+void vp9_filter_block1d4_h8_avg_ssse3(const unsigned char *src_ptr,
+                                     const unsigned int src_pitch,
+                                     unsigned char *output_ptr,
+                                     unsigned int out_pitch,
+                                     unsigned int output_height,
+                                     const short *filter);
+
+void vp9_convolve8_horiz_ssse3(const uint8_t *src, int src_stride,
+                               uint8_t *dst, int dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h) {
+  if (x_step_q4 == 16 && filter_x[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_h8_ssse3(src, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_x);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_h8_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_h8_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+                          filter_x, x_step_q4, filter_y, y_step_q4,
+                          w, h);
+  }
+}
+
+void vp9_convolve8_vert_ssse3(const uint8_t *src, int src_stride,
+                              uint8_t *dst, int dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h) {
+  if (y_step_q4 == 16 && filter_y[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_v8_ssse3(src - src_stride * 3, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_y);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_v8_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_v8_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
+                         filter_x, x_step_q4, filter_y, y_step_q4,
+                         w, h);
+  }
+}
+
+void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, int src_stride,
+                               uint8_t *dst, int dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h) {
+  if (x_step_q4 == 16 && filter_x[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_h8_avg_ssse3(src, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_x);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_h8_avg_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_h8_avg_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+                              filter_x, x_step_q4, filter_y, y_step_q4,
+                              w, h);
+  }
+}
+
+void vp9_convolve8_avg_vert_ssse3(const uint8_t *src, int src_stride,
+                              uint8_t *dst, int dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h) {
+  if (y_step_q4 == 16 && filter_y[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_v8_avg_ssse3(src - src_stride * 3, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_y);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_v8_avg_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_v8_avg_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+                             filter_x, x_step_q4, filter_y, y_step_q4,
+                             w, h);
+  }
+}
+
+void vp9_convolve8_ssse3(const uint8_t *src, int src_stride,
+                         uint8_t *dst, int dst_stride,
+                         const int16_t *filter_x, int x_step_q4,
+                         const int16_t *filter_y, int y_step_q4,
+                         int w, int h) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64*71);
+
+  assert(w <= 64);
+  assert(h <= 64);
+  if (x_step_q4 == 16 && y_step_q4 == 16) {
+    vp9_convolve8_horiz_ssse3(src - 3 * src_stride, src_stride, fdata2, 64,
+                              filter_x, x_step_q4, filter_y, y_step_q4,
+                              w, h + 7);
+    vp9_convolve8_vert_ssse3(fdata2 + 3 * 64, 64, dst, dst_stride,
+                             filter_x, x_step_q4, filter_y, y_step_q4, w, h);
+  } else {
+    vp9_convolve8_c(src, src_stride, dst, dst_stride,
+                    filter_x, x_step_q4, filter_y, y_step_q4, w, h);
+  }
+}
+
+void vp9_convolve8_avg_ssse3(const uint8_t *src, int src_stride,
+                         uint8_t *dst, int dst_stride,
+                         const int16_t *filter_x, int x_step_q4,
+                         const int16_t *filter_y, int y_step_q4,
+                         int w, int h) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64*71);
+
+  assert(w <= 64);
+  assert(h <= 64);
+  if (x_step_q4 == 16 && y_step_q4 == 16) {
+    vp9_convolve8_horiz_ssse3(src - 3 * src_stride, src_stride, fdata2, 64,
+                              filter_x, x_step_q4, filter_y, y_step_q4,
+                              w, h + 7);
+    vp9_convolve8_avg_vert_ssse3(fdata2 + 3 * 64, 64, dst, dst_stride,
+                                 filter_x, x_step_q4, filter_y, y_step_q4,
+                                 w, h);
+  } else {
+    vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
+                        filter_x, x_step_q4, filter_y, y_step_q4, w, h);
+  }
+}
+#endif
diff --git a/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c b/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c
new file mode 100644
index 0000000..599dcff
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -0,0 +1,1985 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <emmintrin.h>  // SSE2
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_idct.h"
+
+// In order to improve performance, clip absolute diff values to [0, 255],
+// which allows to keep the additions/subtractions in 8 bits.
+void vp9_dc_only_idct_add_sse2(int input_dc, uint8_t *pred_ptr,
+                               uint8_t *dst_ptr, int pitch, int stride) {
+  int a1;
+  int16_t out;
+  uint8_t abs_diff;
+  __m128i p0, p1, p2, p3;
+  unsigned int extended_diff;
+  __m128i diff;
+
+  out = dct_const_round_shift(input_dc * cospi_16_64);
+  out = dct_const_round_shift(out * cospi_16_64);
+  a1 = ROUND_POWER_OF_TWO(out, 4);
+
+  // Read prediction data.
+  p0 = _mm_cvtsi32_si128 (*(const int *)(pred_ptr + 0 * pitch));
+  p1 = _mm_cvtsi32_si128 (*(const int *)(pred_ptr + 1 * pitch));
+  p2 = _mm_cvtsi32_si128 (*(const int *)(pred_ptr + 2 * pitch));
+  p3 = _mm_cvtsi32_si128 (*(const int *)(pred_ptr + 3 * pitch));
+
+  // Unpack prediction data, and store 4x4 array in 1 XMM register.
+  p0 = _mm_unpacklo_epi32(p0, p1);
+  p2 = _mm_unpacklo_epi32(p2, p3);
+  p0 = _mm_unpacklo_epi64(p0, p2);
+
+  // Clip dc value to [0, 255] range. Then, do addition or subtraction
+  // according to its sign.
+  if (a1 >= 0) {
+    abs_diff = (a1 > 255) ? 255 : a1;
+    extended_diff = abs_diff * 0x01010101u;
+    diff = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_diff), 0);
+
+    p1 = _mm_adds_epu8(p0, diff);
+  } else {
+    abs_diff = (a1 < -255) ? 255 : -a1;
+    extended_diff = abs_diff * 0x01010101u;
+    diff = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_diff), 0);
+
+    p1 = _mm_subs_epu8(p0, diff);
+  }
+
+  // Store results to dst.
+  *(int *)dst_ptr = _mm_cvtsi128_si32(p1);
+  dst_ptr += stride;
+
+  p1 = _mm_srli_si128(p1, 4);
+  *(int *)dst_ptr = _mm_cvtsi128_si32(p1);
+  dst_ptr += stride;
+
+  p1 = _mm_srli_si128(p1, 4);
+  *(int *)dst_ptr = _mm_cvtsi128_si32(p1);
+  dst_ptr += stride;
+
+  p1 = _mm_srli_si128(p1, 4);
+  *(int *)dst_ptr = _mm_cvtsi128_si32(p1);
+}
+
+void vp9_short_idct4x4_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i eight = _mm_set1_epi16(8);
+  const __m128i cst = _mm_setr_epi16((int16_t)cospi_16_64, (int16_t)cospi_16_64,
+                                    (int16_t)cospi_16_64, (int16_t)-cospi_16_64,
+                                    (int16_t)cospi_24_64, (int16_t)-cospi_8_64,
+                                    (int16_t)cospi_8_64, (int16_t)cospi_24_64);
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  __m128i input0, input1, input2, input3;
+
+  // Rows
+  input0 = _mm_loadl_epi64((__m128i *)input);
+  input1 = _mm_loadl_epi64((__m128i *)(input + 4));
+  input2 = _mm_loadl_epi64((__m128i *)(input + 8));
+  input3 = _mm_loadl_epi64((__m128i *)(input + 12));
+
+  // Construct i3, i1, i3, i1, i2, i0, i2, i0
+  input0 = _mm_shufflelo_epi16(input0, 0xd8);
+  input1 = _mm_shufflelo_epi16(input1, 0xd8);
+  input2 = _mm_shufflelo_epi16(input2, 0xd8);
+  input3 = _mm_shufflelo_epi16(input3, 0xd8);
+
+  input0 = _mm_unpacklo_epi32(input0, input0);
+  input1 = _mm_unpacklo_epi32(input1, input1);
+  input2 = _mm_unpacklo_epi32(input2, input2);
+  input3 = _mm_unpacklo_epi32(input3, input3);
+
+  // Stage 1
+  input0 = _mm_madd_epi16(input0, cst);
+  input1 = _mm_madd_epi16(input1, cst);
+  input2 = _mm_madd_epi16(input2, cst);
+  input3 = _mm_madd_epi16(input3, cst);
+
+  input0 = _mm_add_epi32(input0, rounding);
+  input1 = _mm_add_epi32(input1, rounding);
+  input2 = _mm_add_epi32(input2, rounding);
+  input3 = _mm_add_epi32(input3, rounding);
+
+  input0 = _mm_srai_epi32(input0, DCT_CONST_BITS);
+  input1 = _mm_srai_epi32(input1, DCT_CONST_BITS);
+  input2 = _mm_srai_epi32(input2, DCT_CONST_BITS);
+  input3 = _mm_srai_epi32(input3, DCT_CONST_BITS);
+
+  // Stage 2
+  input0 = _mm_packs_epi32(input0, zero);
+  input1 = _mm_packs_epi32(input1, zero);
+  input2 = _mm_packs_epi32(input2, zero);
+  input3 = _mm_packs_epi32(input3, zero);
+
+  // Transpose
+  input1 = _mm_unpacklo_epi16(input0, input1);
+  input3 = _mm_unpacklo_epi16(input2, input3);
+  input0 = _mm_unpacklo_epi32(input1, input3);
+  input1 = _mm_unpackhi_epi32(input1, input3);
+
+  // Switch column2, column 3, and then, we got:
+  // input2: column1, column 0;  input3: column2, column 3.
+  input1 = _mm_shuffle_epi32(input1, 0x4e);
+  input2 = _mm_add_epi16(input0, input1);
+  input3 = _mm_sub_epi16(input0, input1);
+
+  // Columns
+  // Construct i3, i1, i3, i1, i2, i0, i2, i0
+  input0 = _mm_shufflelo_epi16(input2, 0xd8);
+  input1 = _mm_shufflehi_epi16(input2, 0xd8);
+  input2 = _mm_shufflehi_epi16(input3, 0xd8);
+  input3 = _mm_shufflelo_epi16(input3, 0xd8);
+
+  input0 = _mm_unpacklo_epi32(input0, input0);
+  input1 = _mm_unpackhi_epi32(input1, input1);
+  input2 = _mm_unpackhi_epi32(input2, input2);
+  input3 = _mm_unpacklo_epi32(input3, input3);
+
+  // Stage 1
+  input0 = _mm_madd_epi16(input0, cst);
+  input1 = _mm_madd_epi16(input1, cst);
+  input2 = _mm_madd_epi16(input2, cst);
+  input3 = _mm_madd_epi16(input3, cst);
+
+  input0 = _mm_add_epi32(input0, rounding);
+  input1 = _mm_add_epi32(input1, rounding);
+  input2 = _mm_add_epi32(input2, rounding);
+  input3 = _mm_add_epi32(input3, rounding);
+
+  input0 = _mm_srai_epi32(input0, DCT_CONST_BITS);
+  input1 = _mm_srai_epi32(input1, DCT_CONST_BITS);
+  input2 = _mm_srai_epi32(input2, DCT_CONST_BITS);
+  input3 = _mm_srai_epi32(input3, DCT_CONST_BITS);
+
+  // Stage 2
+  input0 = _mm_packs_epi32(input0, zero);
+  input1 = _mm_packs_epi32(input1, zero);
+  input2 = _mm_packs_epi32(input2, zero);
+  input3 = _mm_packs_epi32(input3, zero);
+
+  // Transpose
+  input1 = _mm_unpacklo_epi16(input0, input1);
+  input3 = _mm_unpacklo_epi16(input2, input3);
+  input0 = _mm_unpacklo_epi32(input1, input3);
+  input1 = _mm_unpackhi_epi32(input1, input3);
+
+  // Switch column2, column 3, and then, we got:
+  // input2: column1, column 0;  input3: column2, column 3.
+  input1 = _mm_shuffle_epi32(input1, 0x4e);
+  input2 = _mm_add_epi16(input0, input1);
+  input3 = _mm_sub_epi16(input0, input1);
+
+  // Final round and shift
+  input2 = _mm_add_epi16(input2, eight);
+  input3 = _mm_add_epi16(input3, eight);
+
+  input2 = _mm_srai_epi16(input2, 4);
+  input3 = _mm_srai_epi16(input3, 4);
+
+#define RECON_AND_STORE4X4(dest, in_x) \
+  {                                                     \
+      __m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest)); \
+      d0 = _mm_unpacklo_epi8(d0, zero); \
+      d0 = _mm_add_epi16(in_x, d0); \
+      d0 = _mm_packus_epi16(d0, d0); \
+      *(int *)dest = _mm_cvtsi128_si32(d0); \
+      dest += stride; \
+  }
+
+  input0 = _mm_srli_si128(input2, 8);
+  input1 = _mm_srli_si128(input3, 8);
+
+  RECON_AND_STORE4X4(dest, input2);
+  RECON_AND_STORE4X4(dest, input0);
+  RECON_AND_STORE4X4(dest, input1);
+  RECON_AND_STORE4X4(dest, input3);
+}
+
+void vp9_idct4_1d_sse2(int16_t *input, int16_t *output) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i c1 = _mm_setr_epi16((int16_t)cospi_16_64, (int16_t)cospi_16_64,
+                                    (int16_t)cospi_16_64, (int16_t)-cospi_16_64,
+                                    (int16_t)cospi_24_64, (int16_t)-cospi_8_64,
+                                    (int16_t)cospi_8_64, (int16_t)cospi_24_64);
+  const __m128i c2 = _mm_setr_epi16(1, 1, 1, 1, 1, -1, 1, -1);
+
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  __m128i in, temp;
+
+  // Load input data.
+  in = _mm_loadl_epi64((__m128i *)input);
+
+  // Construct i3, i1, i3, i1, i2, i0, i2, i0
+  in = _mm_shufflelo_epi16(in, 0xd8);
+  in = _mm_unpacklo_epi32(in, in);
+
+  // Stage 1
+  in = _mm_madd_epi16(in, c1);
+  in = _mm_add_epi32(in, rounding);
+  in = _mm_srai_epi32(in, DCT_CONST_BITS);
+  in = _mm_packs_epi32(in, zero);
+
+  // Stage 2
+  temp = _mm_shufflelo_epi16(in, 0x9c);
+  in = _mm_shufflelo_epi16(in, 0xc9);
+  in = _mm_unpacklo_epi64(temp, in);
+  in = _mm_madd_epi16(in, c2);
+  in = _mm_packs_epi32(in, zero);
+
+  // Store results
+  _mm_storel_epi64((__m128i *)output, in);
+}
+
+#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, \
+                      out0, out1, out2, out3, out4, out5, out6, out7) \
+  {                                                     \
+    const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \
+    const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \
+    const __m128i tr0_2 = _mm_unpackhi_epi16(in0, in1); \
+    const __m128i tr0_3 = _mm_unpackhi_epi16(in2, in3); \
+    const __m128i tr0_4 = _mm_unpacklo_epi16(in4, in5); \
+    const __m128i tr0_5 = _mm_unpacklo_epi16(in6, in7); \
+    const __m128i tr0_6 = _mm_unpackhi_epi16(in4, in5); \
+    const __m128i tr0_7 = _mm_unpackhi_epi16(in6, in7); \
+                                                        \
+    const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \
+    const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3); \
+    const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); \
+    const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3); \
+    const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); \
+    const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7); \
+    const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); \
+    const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7); \
+                                                            \
+    out0 = _mm_unpacklo_epi64(tr1_0, tr1_4); \
+    out1 = _mm_unpackhi_epi64(tr1_0, tr1_4); \
+    out2 = _mm_unpacklo_epi64(tr1_2, tr1_6); \
+    out3 = _mm_unpackhi_epi64(tr1_2, tr1_6); \
+    out4 = _mm_unpacklo_epi64(tr1_1, tr1_5); \
+    out5 = _mm_unpackhi_epi64(tr1_1, tr1_5); \
+    out6 = _mm_unpacklo_epi64(tr1_3, tr1_7); \
+    out7 = _mm_unpackhi_epi64(tr1_3, tr1_7); \
+  }
+
+#define TRANSPOSE_4X8(in0, in1, in2, in3, in4, in5, in6, in7, \
+                      out0, out1, out2, out3, out4, out5, out6, out7) \
+  {                                                     \
+    const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \
+    const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \
+    const __m128i tr0_4 = _mm_unpacklo_epi16(in4, in5); \
+    const __m128i tr0_5 = _mm_unpacklo_epi16(in6, in7); \
+                                                        \
+    const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \
+    const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); \
+    const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); \
+    const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); \
+                                                            \
+    out0 = _mm_unpacklo_epi64(tr1_0, tr1_4); \
+    out1 = _mm_unpackhi_epi64(tr1_0, tr1_4); \
+    out2 = _mm_unpacklo_epi64(tr1_2, tr1_6); \
+    out3 = _mm_unpackhi_epi64(tr1_2, tr1_6); \
+    out4 = out5 = out6 = out7 = zero; \
+  }
+
+#define TRANSPOSE_8X4(in0, in1, in2, in3, out0, out1, out2, out3) \
+  {                                                     \
+    const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \
+    const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \
+    const __m128i tr0_2 = _mm_unpackhi_epi16(in0, in1); \
+    const __m128i tr0_3 = _mm_unpackhi_epi16(in2, in3); \
+                                                        \
+    in0 = _mm_unpacklo_epi32(tr0_0, tr0_1);  /* i1 i0 */  \
+    in1 = _mm_unpackhi_epi32(tr0_0, tr0_1);  /* i3 i2 */  \
+    in2 = _mm_unpacklo_epi32(tr0_2, tr0_3);  /* i5 i4 */  \
+    in3 = _mm_unpackhi_epi32(tr0_2, tr0_3);  /* i7 i6 */  \
+  }
+
+// Define Macro for multiplying elements by constants and adding them together.
+#define MULTIPLICATION_AND_ADD(lo_0, hi_0, lo_1, hi_1, \
+                               cst0, cst1, cst2, cst3, res0, res1, res2, res3) \
+  {   \
+      tmp0 = _mm_madd_epi16(lo_0, cst0); \
+      tmp1 = _mm_madd_epi16(hi_0, cst0); \
+      tmp2 = _mm_madd_epi16(lo_0, cst1); \
+      tmp3 = _mm_madd_epi16(hi_0, cst1); \
+      tmp4 = _mm_madd_epi16(lo_1, cst2); \
+      tmp5 = _mm_madd_epi16(hi_1, cst2); \
+      tmp6 = _mm_madd_epi16(lo_1, cst3); \
+      tmp7 = _mm_madd_epi16(hi_1, cst3); \
+      \
+      tmp0 = _mm_add_epi32(tmp0, rounding); \
+      tmp1 = _mm_add_epi32(tmp1, rounding); \
+      tmp2 = _mm_add_epi32(tmp2, rounding); \
+      tmp3 = _mm_add_epi32(tmp3, rounding); \
+      tmp4 = _mm_add_epi32(tmp4, rounding); \
+      tmp5 = _mm_add_epi32(tmp5, rounding); \
+      tmp6 = _mm_add_epi32(tmp6, rounding); \
+      tmp7 = _mm_add_epi32(tmp7, rounding); \
+      \
+      tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \
+      tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \
+      tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \
+      tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \
+      tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS); \
+      tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS); \
+      tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS); \
+      tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS); \
+      \
+      res0 = _mm_packs_epi32(tmp0, tmp1); \
+      res1 = _mm_packs_epi32(tmp2, tmp3); \
+      res2 = _mm_packs_epi32(tmp4, tmp5); \
+      res3 = _mm_packs_epi32(tmp6, tmp7); \
+  }
+
+#define IDCT8x8_1D  \
+  /* Stage1 */      \
+  { \
+    const __m128i lo_17 = _mm_unpacklo_epi16(in1, in7); \
+    const __m128i hi_17 = _mm_unpackhi_epi16(in1, in7); \
+    const __m128i lo_35 = _mm_unpacklo_epi16(in3, in5); \
+    const __m128i hi_35 = _mm_unpackhi_epi16(in3, in5); \
+    \
+    MULTIPLICATION_AND_ADD(lo_17, hi_17, lo_35, hi_35, stg1_0, \
+                          stg1_1, stg1_2, stg1_3, stp1_4,      \
+                          stp1_7, stp1_5, stp1_6)              \
+  } \
+    \
+  /* Stage2 */ \
+  { \
+    const __m128i lo_04 = _mm_unpacklo_epi16(in0, in4); \
+    const __m128i hi_04 = _mm_unpackhi_epi16(in0, in4); \
+    const __m128i lo_26 = _mm_unpacklo_epi16(in2, in6); \
+    const __m128i hi_26 = _mm_unpackhi_epi16(in2, in6); \
+    \
+    MULTIPLICATION_AND_ADD(lo_04, hi_04, lo_26, hi_26, stg2_0, \
+                           stg2_1, stg2_2, stg2_3, stp2_0,     \
+                           stp2_1, stp2_2, stp2_3)             \
+    \
+    stp2_4 = _mm_adds_epi16(stp1_4, stp1_5); \
+    stp2_5 = _mm_subs_epi16(stp1_4, stp1_5); \
+    stp2_6 = _mm_subs_epi16(stp1_7, stp1_6); \
+    stp2_7 = _mm_adds_epi16(stp1_7, stp1_6); \
+  } \
+    \
+  /* Stage3 */ \
+  { \
+    const __m128i lo_56 = _mm_unpacklo_epi16(stp2_6, stp2_5); \
+    const __m128i hi_56 = _mm_unpackhi_epi16(stp2_6, stp2_5); \
+    \
+    stp1_0 = _mm_adds_epi16(stp2_0, stp2_3); \
+    stp1_1 = _mm_adds_epi16(stp2_1, stp2_2); \
+    stp1_2 = _mm_subs_epi16(stp2_1, stp2_2); \
+    stp1_3 = _mm_subs_epi16(stp2_0, stp2_3); \
+    \
+    tmp0 = _mm_madd_epi16(lo_56, stg2_1); \
+    tmp1 = _mm_madd_epi16(hi_56, stg2_1); \
+    tmp2 = _mm_madd_epi16(lo_56, stg2_0); \
+    tmp3 = _mm_madd_epi16(hi_56, stg2_0); \
+    \
+    tmp0 = _mm_add_epi32(tmp0, rounding); \
+    tmp1 = _mm_add_epi32(tmp1, rounding); \
+    tmp2 = _mm_add_epi32(tmp2, rounding); \
+    tmp3 = _mm_add_epi32(tmp3, rounding); \
+    \
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \
+    tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \
+    tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \
+    \
+    stp1_5 = _mm_packs_epi32(tmp0, tmp1); \
+    stp1_6 = _mm_packs_epi32(tmp2, tmp3); \
+  } \
+  \
+  /* Stage4  */ \
+  in0 = _mm_adds_epi16(stp1_0, stp2_7); \
+  in1 = _mm_adds_epi16(stp1_1, stp1_6); \
+  in2 = _mm_adds_epi16(stp1_2, stp1_5); \
+  in3 = _mm_adds_epi16(stp1_3, stp2_4); \
+  in4 = _mm_subs_epi16(stp1_3, stp2_4); \
+  in5 = _mm_subs_epi16(stp1_2, stp1_5); \
+  in6 = _mm_subs_epi16(stp1_1, stp1_6); \
+  in7 = _mm_subs_epi16(stp1_0, stp2_7);
+
+#define RECON_AND_STORE(dest, in_x) \
+  {                                                     \
+     __m128i d0 = _mm_loadl_epi64((__m128i *)(dest)); \
+      d0 = _mm_unpacklo_epi8(d0, zero); \
+      in_x = _mm_add_epi16(in_x, d0); \
+      in_x = _mm_packus_epi16(in_x, in_x); \
+      _mm_storel_epi64((__m128i *)(dest), in_x); \
+      dest += stride; \
+  }
+
+void vp9_short_idct8x8_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i final_rounding = _mm_set1_epi16(1<<4);
+  const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+  const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
+  const __m128i stg1_2 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+  const __m128i stg1_3 = pair_set_epi16(cospi_12_64, cospi_20_64);
+  const __m128i stg2_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
+  const __m128i stg2_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i stg2_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+  const __m128i stg2_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
+
+  __m128i in0, in1, in2, in3, in4, in5, in6, in7;
+  __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7;
+  __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7;
+  __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  int i;
+
+  // Load input data.
+  in0 = _mm_load_si128((__m128i *)input);
+  in1 = _mm_load_si128((__m128i *)(input + 8 * 1));
+  in2 = _mm_load_si128((__m128i *)(input + 8 * 2));
+  in3 = _mm_load_si128((__m128i *)(input + 8 * 3));
+  in4 = _mm_load_si128((__m128i *)(input + 8 * 4));
+  in5 = _mm_load_si128((__m128i *)(input + 8 * 5));
+  in6 = _mm_load_si128((__m128i *)(input + 8 * 6));
+  in7 = _mm_load_si128((__m128i *)(input + 8 * 7));
+
+  // 2-D
+  for (i = 0; i < 2; i++) {
+    // 8x8 Transpose is copied from vp9_short_fdct8x8_sse2()
+    TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
+                  in4, in5, in6, in7);
+
+    // 4-stage 1D idct8x8
+    IDCT8x8_1D
+  }
+
+  // Final rounding and shift
+  in0 = _mm_adds_epi16(in0, final_rounding);
+  in1 = _mm_adds_epi16(in1, final_rounding);
+  in2 = _mm_adds_epi16(in2, final_rounding);
+  in3 = _mm_adds_epi16(in3, final_rounding);
+  in4 = _mm_adds_epi16(in4, final_rounding);
+  in5 = _mm_adds_epi16(in5, final_rounding);
+  in6 = _mm_adds_epi16(in6, final_rounding);
+  in7 = _mm_adds_epi16(in7, final_rounding);
+
+  in0 = _mm_srai_epi16(in0, 5);
+  in1 = _mm_srai_epi16(in1, 5);
+  in2 = _mm_srai_epi16(in2, 5);
+  in3 = _mm_srai_epi16(in3, 5);
+  in4 = _mm_srai_epi16(in4, 5);
+  in5 = _mm_srai_epi16(in5, 5);
+  in6 = _mm_srai_epi16(in6, 5);
+  in7 = _mm_srai_epi16(in7, 5);
+
+  RECON_AND_STORE(dest, in0);
+  RECON_AND_STORE(dest, in1);
+  RECON_AND_STORE(dest, in2);
+  RECON_AND_STORE(dest, in3);
+  RECON_AND_STORE(dest, in4);
+  RECON_AND_STORE(dest, in5);
+  RECON_AND_STORE(dest, in6);
+  RECON_AND_STORE(dest, in7);
+}
+
+void vp9_short_idct10_8x8_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i final_rounding = _mm_set1_epi16(1<<4);
+  const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+  const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
+  const __m128i stg1_2 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+  const __m128i stg1_3 = pair_set_epi16(cospi_12_64, cospi_20_64);
+  const __m128i stg2_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
+  const __m128i stg2_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i stg2_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+  const __m128i stg2_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
+  const __m128i stg3_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+
+  __m128i in0, in1, in2, in3, in4, in5, in6, in7;
+  __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7;
+  __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7;
+  __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+
+  // Rows. Load 4-row input data.
+  in0 = _mm_load_si128((__m128i *)input);
+  in1 = _mm_load_si128((__m128i *)(input + 8 * 1));
+  in2 = _mm_load_si128((__m128i *)(input + 8 * 2));
+  in3 = _mm_load_si128((__m128i *)(input + 8 * 3));
+
+  // 8x4 Transpose
+  TRANSPOSE_8X4(in0, in1, in2, in3, in0, in1, in2, in3)
+
+  // Stage1
+  {
+    const __m128i lo_17 = _mm_unpackhi_epi16(in0, in3);
+    const __m128i lo_35 = _mm_unpackhi_epi16(in1, in2);
+
+    tmp0 = _mm_madd_epi16(lo_17, stg1_0);
+    tmp2 = _mm_madd_epi16(lo_17, stg1_1);
+    tmp4 = _mm_madd_epi16(lo_35, stg1_2);
+    tmp6 = _mm_madd_epi16(lo_35, stg1_3);
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+
+    stp1_4 = _mm_packs_epi32(tmp0, zero);
+    stp1_7 = _mm_packs_epi32(tmp2, zero);
+    stp1_5 = _mm_packs_epi32(tmp4, zero);
+    stp1_6 = _mm_packs_epi32(tmp6, zero);
+  }
+
+  // Stage2
+  {
+    const __m128i lo_04 = _mm_unpacklo_epi16(in0, in2);
+    const __m128i lo_26 = _mm_unpacklo_epi16(in1, in3);
+
+    tmp0 = _mm_madd_epi16(lo_04, stg2_0);
+    tmp2 = _mm_madd_epi16(lo_04, stg2_1);
+    tmp4 = _mm_madd_epi16(lo_26, stg2_2);
+    tmp6 = _mm_madd_epi16(lo_26, stg2_3);
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+
+    stp2_0 = _mm_packs_epi32(tmp0, zero);
+    stp2_1 = _mm_packs_epi32(tmp2, zero);
+    stp2_2 = _mm_packs_epi32(tmp4, zero);
+    stp2_3 = _mm_packs_epi32(tmp6, zero);
+
+    stp2_4 = _mm_adds_epi16(stp1_4, stp1_5);
+    stp2_5 = _mm_subs_epi16(stp1_4, stp1_5);
+    stp2_6 = _mm_subs_epi16(stp1_7, stp1_6);
+    stp2_7 = _mm_adds_epi16(stp1_7, stp1_6);
+  }
+
+  // Stage3
+  {
+    const __m128i lo_56 = _mm_unpacklo_epi16(stp2_5, stp2_6);
+    stp1_0 = _mm_adds_epi16(stp2_0, stp2_3);
+    stp1_1 = _mm_adds_epi16(stp2_1, stp2_2);
+    stp1_2 = _mm_subs_epi16(stp2_1, stp2_2);
+    stp1_3 = _mm_subs_epi16(stp2_0, stp2_3);
+
+    tmp0 = _mm_madd_epi16(lo_56, stg3_0);
+    tmp2 = _mm_madd_epi16(lo_56, stg2_0);  // stg3_1 = stg2_0
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+
+    stp1_5 = _mm_packs_epi32(tmp0, zero);
+    stp1_6 = _mm_packs_epi32(tmp2, zero);
+  }
+
+  // Stage4
+  in0 = _mm_adds_epi16(stp1_0, stp2_7);
+  in1 = _mm_adds_epi16(stp1_1, stp1_6);
+  in2 = _mm_adds_epi16(stp1_2, stp1_5);
+  in3 = _mm_adds_epi16(stp1_3, stp2_4);
+  in4 = _mm_subs_epi16(stp1_3, stp2_4);
+  in5 = _mm_subs_epi16(stp1_2, stp1_5);
+  in6 = _mm_subs_epi16(stp1_1, stp1_6);
+  in7 = _mm_subs_epi16(stp1_0, stp2_7);
+
+  // Columns. 4x8 Transpose
+  TRANSPOSE_4X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
+                in4, in5, in6, in7)
+
+  // 1D idct8x8
+  IDCT8x8_1D
+
+  // Final rounding and shift
+  in0 = _mm_adds_epi16(in0, final_rounding);
+  in1 = _mm_adds_epi16(in1, final_rounding);
+  in2 = _mm_adds_epi16(in2, final_rounding);
+  in3 = _mm_adds_epi16(in3, final_rounding);
+  in4 = _mm_adds_epi16(in4, final_rounding);
+  in5 = _mm_adds_epi16(in5, final_rounding);
+  in6 = _mm_adds_epi16(in6, final_rounding);
+  in7 = _mm_adds_epi16(in7, final_rounding);
+
+  in0 = _mm_srai_epi16(in0, 5);
+  in1 = _mm_srai_epi16(in1, 5);
+  in2 = _mm_srai_epi16(in2, 5);
+  in3 = _mm_srai_epi16(in3, 5);
+  in4 = _mm_srai_epi16(in4, 5);
+  in5 = _mm_srai_epi16(in5, 5);
+  in6 = _mm_srai_epi16(in6, 5);
+  in7 = _mm_srai_epi16(in7, 5);
+
+  RECON_AND_STORE(dest, in0);
+  RECON_AND_STORE(dest, in1);
+  RECON_AND_STORE(dest, in2);
+  RECON_AND_STORE(dest, in3);
+  RECON_AND_STORE(dest, in4);
+  RECON_AND_STORE(dest, in5);
+  RECON_AND_STORE(dest, in6);
+  RECON_AND_STORE(dest, in7);
+}
+
+#define IDCT16x16_1D \
+  /* Stage2 */ \
+  { \
+    const __m128i lo_1_15 = _mm_unpacklo_epi16(in1, in15); \
+    const __m128i hi_1_15 = _mm_unpackhi_epi16(in1, in15); \
+    const __m128i lo_9_7 = _mm_unpacklo_epi16(in9, in7);   \
+    const __m128i hi_9_7 = _mm_unpackhi_epi16(in9, in7);   \
+    const __m128i lo_5_11 = _mm_unpacklo_epi16(in5, in11); \
+    const __m128i hi_5_11 = _mm_unpackhi_epi16(in5, in11); \
+    const __m128i lo_13_3 = _mm_unpacklo_epi16(in13, in3); \
+    const __m128i hi_13_3 = _mm_unpackhi_epi16(in13, in3); \
+    \
+    MULTIPLICATION_AND_ADD(lo_1_15, hi_1_15, lo_9_7, hi_9_7, \
+                           stg2_0, stg2_1, stg2_2, stg2_3, \
+                           stp2_8, stp2_15, stp2_9, stp2_14) \
+    \
+    MULTIPLICATION_AND_ADD(lo_5_11, hi_5_11, lo_13_3, hi_13_3, \
+                           stg2_4, stg2_5, stg2_6, stg2_7, \
+                           stp2_10, stp2_13, stp2_11, stp2_12) \
+  } \
+    \
+  /* Stage3 */ \
+  { \
+    const __m128i lo_2_14 = _mm_unpacklo_epi16(in2, in14); \
+    const __m128i hi_2_14 = _mm_unpackhi_epi16(in2, in14); \
+    const __m128i lo_10_6 = _mm_unpacklo_epi16(in10, in6); \
+    const __m128i hi_10_6 = _mm_unpackhi_epi16(in10, in6); \
+    \
+    MULTIPLICATION_AND_ADD(lo_2_14, hi_2_14, lo_10_6, hi_10_6, \
+                           stg3_0, stg3_1, stg3_2, stg3_3, \
+                           stp1_4, stp1_7, stp1_5, stp1_6) \
+    \
+    stp1_8_0 = _mm_add_epi16(stp2_8, stp2_9);  \
+    stp1_9 = _mm_sub_epi16(stp2_8, stp2_9);    \
+    stp1_10 = _mm_sub_epi16(stp2_11, stp2_10); \
+    stp1_11 = _mm_add_epi16(stp2_11, stp2_10); \
+    \
+    stp1_12_0 = _mm_add_epi16(stp2_12, stp2_13); \
+    stp1_13 = _mm_sub_epi16(stp2_12, stp2_13); \
+    stp1_14 = _mm_sub_epi16(stp2_15, stp2_14); \
+    stp1_15 = _mm_add_epi16(stp2_15, stp2_14); \
+  } \
+  \
+  /* Stage4 */ \
+  { \
+    const __m128i lo_0_8 = _mm_unpacklo_epi16(in0, in8); \
+    const __m128i hi_0_8 = _mm_unpackhi_epi16(in0, in8); \
+    const __m128i lo_4_12 = _mm_unpacklo_epi16(in4, in12); \
+    const __m128i hi_4_12 = _mm_unpackhi_epi16(in4, in12); \
+    \
+    const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14); \
+    const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14); \
+    const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \
+    const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \
+    \
+    MULTIPLICATION_AND_ADD(lo_0_8, hi_0_8, lo_4_12, hi_4_12, \
+                           stg4_0, stg4_1, stg4_2, stg4_3, \
+                           stp2_0, stp2_1, stp2_2, stp2_3) \
+    \
+    stp2_4 = _mm_add_epi16(stp1_4, stp1_5); \
+    stp2_5 = _mm_sub_epi16(stp1_4, stp1_5); \
+    stp2_6 = _mm_sub_epi16(stp1_7, stp1_6); \
+    stp2_7 = _mm_add_epi16(stp1_7, stp1_6); \
+    \
+    MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, \
+                           stg4_4, stg4_5, stg4_6, stg4_7, \
+                           stp2_9, stp2_14, stp2_10, stp2_13) \
+  } \
+    \
+  /* Stage5 */ \
+  { \
+    const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5); \
+    const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5); \
+    \
+    stp1_0 = _mm_add_epi16(stp2_0, stp2_3); \
+    stp1_1 = _mm_add_epi16(stp2_1, stp2_2); \
+    stp1_2 = _mm_sub_epi16(stp2_1, stp2_2); \
+    stp1_3 = _mm_sub_epi16(stp2_0, stp2_3); \
+    \
+    tmp0 = _mm_madd_epi16(lo_6_5, stg4_1); \
+    tmp1 = _mm_madd_epi16(hi_6_5, stg4_1); \
+    tmp2 = _mm_madd_epi16(lo_6_5, stg4_0); \
+    tmp3 = _mm_madd_epi16(hi_6_5, stg4_0); \
+    \
+    tmp0 = _mm_add_epi32(tmp0, rounding); \
+    tmp1 = _mm_add_epi32(tmp1, rounding); \
+    tmp2 = _mm_add_epi32(tmp2, rounding); \
+    tmp3 = _mm_add_epi32(tmp3, rounding); \
+    \
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS); \
+    tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS); \
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS); \
+    tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS); \
+    \
+    stp1_5 = _mm_packs_epi32(tmp0, tmp1); \
+    stp1_6 = _mm_packs_epi32(tmp2, tmp3); \
+    \
+    stp1_8 = _mm_add_epi16(stp1_8_0, stp1_11);  \
+    stp1_9 = _mm_add_epi16(stp2_9, stp2_10);    \
+    stp1_10 = _mm_sub_epi16(stp2_9, stp2_10);   \
+    stp1_11 = _mm_sub_epi16(stp1_8_0, stp1_11); \
+    \
+    stp1_12 = _mm_sub_epi16(stp1_15, stp1_12_0); \
+    stp1_13 = _mm_sub_epi16(stp2_14, stp2_13);   \
+    stp1_14 = _mm_add_epi16(stp2_14, stp2_13);   \
+    stp1_15 = _mm_add_epi16(stp1_15, stp1_12_0); \
+  } \
+    \
+  /* Stage6 */ \
+  { \
+    const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13); \
+    const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13); \
+    const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12); \
+    const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12); \
+    \
+    stp2_0 = _mm_add_epi16(stp1_0, stp2_7); \
+    stp2_1 = _mm_add_epi16(stp1_1, stp1_6); \
+    stp2_2 = _mm_add_epi16(stp1_2, stp1_5); \
+    stp2_3 = _mm_add_epi16(stp1_3, stp2_4); \
+    stp2_4 = _mm_sub_epi16(stp1_3, stp2_4); \
+    stp2_5 = _mm_sub_epi16(stp1_2, stp1_5); \
+    stp2_6 = _mm_sub_epi16(stp1_1, stp1_6); \
+    stp2_7 = _mm_sub_epi16(stp1_0, stp2_7); \
+    \
+    MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12, \
+                           stg6_0, stg4_0, stg6_0, stg4_0, \
+                           stp2_10, stp2_13, stp2_11, stp2_12) \
+  }
+
+void vp9_short_idct16x16_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i final_rounding = _mm_set1_epi16(1<<5);
+  const __m128i zero = _mm_setzero_si128();
+
+  const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64);
+  const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64);
+  const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64);
+  const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64);
+  const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64);
+  const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64);
+  const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64);
+  const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64);
+
+  const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+  const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
+  const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64);
+  const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64);
+
+  const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
+  const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+  const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
+  const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
+  const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+
+  const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+
+  __m128i in0 = zero, in1 = zero, in2 = zero, in3 = zero, in4 = zero,
+          in5 = zero, in6 = zero, in7 = zero, in8 = zero, in9 = zero,
+          in10 = zero, in11 = zero, in12 = zero, in13 = zero,
+          in14 = zero, in15 = zero;
+  __m128i l0 = zero, l1 = zero, l2 = zero, l3 = zero, l4 = zero, l5 = zero,
+          l6 = zero, l7 = zero, l8 = zero, l9 = zero, l10 = zero, l11 = zero,
+          l12 = zero, l13 = zero, l14 = zero, l15 = zero;
+  __m128i r0 = zero, r1 = zero, r2 = zero, r3 = zero, r4 = zero, r5 = zero,
+          r6 = zero, r7 = zero, r8 = zero, r9 = zero, r10 = zero, r11 = zero,
+          r12 = zero, r13 = zero, r14 = zero, r15 = zero;
+  __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
+          stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
+          stp1_8_0, stp1_12_0;
+  __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7,
+          stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15;
+  __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  int i;
+
+  // We work on a 8x16 block each time, and loop 4 times for 2-D 16x16 idct.
+  for (i = 0; i < 4; i++) {
+    // 1-D idct
+    if (i < 2) {
+      if (i == 1) input += 128;
+
+      // Load input data.
+      in0 = _mm_load_si128((__m128i *)input);
+      in8 = _mm_load_si128((__m128i *)(input + 8 * 1));
+      in1 = _mm_load_si128((__m128i *)(input + 8 * 2));
+      in9 = _mm_load_si128((__m128i *)(input + 8 * 3));
+      in2 = _mm_load_si128((__m128i *)(input + 8 * 4));
+      in10 = _mm_load_si128((__m128i *)(input + 8 * 5));
+      in3 = _mm_load_si128((__m128i *)(input + 8 * 6));
+      in11 = _mm_load_si128((__m128i *)(input + 8 * 7));
+      in4 = _mm_load_si128((__m128i *)(input + 8 * 8));
+      in12 = _mm_load_si128((__m128i *)(input + 8 * 9));
+      in5 = _mm_load_si128((__m128i *)(input + 8 * 10));
+      in13 = _mm_load_si128((__m128i *)(input + 8 * 11));
+      in6 = _mm_load_si128((__m128i *)(input + 8 * 12));
+      in14 = _mm_load_si128((__m128i *)(input + 8 * 13));
+      in7 = _mm_load_si128((__m128i *)(input + 8 * 14));
+      in15 = _mm_load_si128((__m128i *)(input + 8 * 15));
+
+      TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
+                    in4, in5, in6, in7);
+      TRANSPOSE_8X8(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
+                    in10, in11, in12, in13, in14, in15);
+    }
+
+    if (i == 2) {
+      TRANSPOSE_8X8(l0, l1, l2, l3, l4, l5, l6, l7, in0, in1, in2, in3, in4,
+                    in5, in6, in7);
+      TRANSPOSE_8X8(r0, r1, r2, r3, r4, r5, r6, r7, in8, in9, in10, in11, in12,
+                    in13, in14, in15);
+    }
+
+    if (i == 3) {
+      TRANSPOSE_8X8(l8, l9, l10, l11, l12, l13, l14, l15, in0, in1, in2, in3,
+                    in4, in5, in6, in7);
+      TRANSPOSE_8X8(r8, r9, r10, r11, r12, r13, r14, r15, in8, in9, in10, in11,
+                    in12, in13, in14, in15);
+    }
+
+    IDCT16x16_1D
+
+    // Stage7
+    if (i == 0) {
+      // Left 8x16
+      l0 = _mm_add_epi16(stp2_0, stp1_15);
+      l1 = _mm_add_epi16(stp2_1, stp1_14);
+      l2 = _mm_add_epi16(stp2_2, stp2_13);
+      l3 = _mm_add_epi16(stp2_3, stp2_12);
+      l4 = _mm_add_epi16(stp2_4, stp2_11);
+      l5 = _mm_add_epi16(stp2_5, stp2_10);
+      l6 = _mm_add_epi16(stp2_6, stp1_9);
+      l7 = _mm_add_epi16(stp2_7, stp1_8);
+      l8 = _mm_sub_epi16(stp2_7, stp1_8);
+      l9 = _mm_sub_epi16(stp2_6, stp1_9);
+      l10 = _mm_sub_epi16(stp2_5, stp2_10);
+      l11 = _mm_sub_epi16(stp2_4, stp2_11);
+      l12 = _mm_sub_epi16(stp2_3, stp2_12);
+      l13 = _mm_sub_epi16(stp2_2, stp2_13);
+      l14 = _mm_sub_epi16(stp2_1, stp1_14);
+      l15 = _mm_sub_epi16(stp2_0, stp1_15);
+    } else if (i == 1) {
+      // Right 8x16
+      r0 = _mm_add_epi16(stp2_0, stp1_15);
+      r1 = _mm_add_epi16(stp2_1, stp1_14);
+      r2 = _mm_add_epi16(stp2_2, stp2_13);
+      r3 = _mm_add_epi16(stp2_3, stp2_12);
+      r4 = _mm_add_epi16(stp2_4, stp2_11);
+      r5 = _mm_add_epi16(stp2_5, stp2_10);
+      r6 = _mm_add_epi16(stp2_6, stp1_9);
+      r7 = _mm_add_epi16(stp2_7, stp1_8);
+      r8 = _mm_sub_epi16(stp2_7, stp1_8);
+      r9 = _mm_sub_epi16(stp2_6, stp1_9);
+      r10 = _mm_sub_epi16(stp2_5, stp2_10);
+      r11 = _mm_sub_epi16(stp2_4, stp2_11);
+      r12 = _mm_sub_epi16(stp2_3, stp2_12);
+      r13 = _mm_sub_epi16(stp2_2, stp2_13);
+      r14 = _mm_sub_epi16(stp2_1, stp1_14);
+      r15 = _mm_sub_epi16(stp2_0, stp1_15);
+    } else {
+      // 2-D
+      in0 = _mm_add_epi16(stp2_0, stp1_15);
+      in1 = _mm_add_epi16(stp2_1, stp1_14);
+      in2 = _mm_add_epi16(stp2_2, stp2_13);
+      in3 = _mm_add_epi16(stp2_3, stp2_12);
+      in4 = _mm_add_epi16(stp2_4, stp2_11);
+      in5 = _mm_add_epi16(stp2_5, stp2_10);
+      in6 = _mm_add_epi16(stp2_6, stp1_9);
+      in7 = _mm_add_epi16(stp2_7, stp1_8);
+      in8 = _mm_sub_epi16(stp2_7, stp1_8);
+      in9 = _mm_sub_epi16(stp2_6, stp1_9);
+      in10 = _mm_sub_epi16(stp2_5, stp2_10);
+      in11 = _mm_sub_epi16(stp2_4, stp2_11);
+      in12 = _mm_sub_epi16(stp2_3, stp2_12);
+      in13 = _mm_sub_epi16(stp2_2, stp2_13);
+      in14 = _mm_sub_epi16(stp2_1, stp1_14);
+      in15 = _mm_sub_epi16(stp2_0, stp1_15);
+
+      // Final rounding and shift
+      in0 = _mm_adds_epi16(in0, final_rounding);
+      in1 = _mm_adds_epi16(in1, final_rounding);
+      in2 = _mm_adds_epi16(in2, final_rounding);
+      in3 = _mm_adds_epi16(in3, final_rounding);
+      in4 = _mm_adds_epi16(in4, final_rounding);
+      in5 = _mm_adds_epi16(in5, final_rounding);
+      in6 = _mm_adds_epi16(in6, final_rounding);
+      in7 = _mm_adds_epi16(in7, final_rounding);
+      in8 = _mm_adds_epi16(in8, final_rounding);
+      in9 = _mm_adds_epi16(in9, final_rounding);
+      in10 = _mm_adds_epi16(in10, final_rounding);
+      in11 = _mm_adds_epi16(in11, final_rounding);
+      in12 = _mm_adds_epi16(in12, final_rounding);
+      in13 = _mm_adds_epi16(in13, final_rounding);
+      in14 = _mm_adds_epi16(in14, final_rounding);
+      in15 = _mm_adds_epi16(in15, final_rounding);
+
+      in0 = _mm_srai_epi16(in0, 6);
+      in1 = _mm_srai_epi16(in1, 6);
+      in2 = _mm_srai_epi16(in2, 6);
+      in3 = _mm_srai_epi16(in3, 6);
+      in4 = _mm_srai_epi16(in4, 6);
+      in5 = _mm_srai_epi16(in5, 6);
+      in6 = _mm_srai_epi16(in6, 6);
+      in7 = _mm_srai_epi16(in7, 6);
+      in8 = _mm_srai_epi16(in8, 6);
+      in9 = _mm_srai_epi16(in9, 6);
+      in10 = _mm_srai_epi16(in10, 6);
+      in11 = _mm_srai_epi16(in11, 6);
+      in12 = _mm_srai_epi16(in12, 6);
+      in13 = _mm_srai_epi16(in13, 6);
+      in14 = _mm_srai_epi16(in14, 6);
+      in15 = _mm_srai_epi16(in15, 6);
+
+      RECON_AND_STORE(dest, in0);
+      RECON_AND_STORE(dest, in1);
+      RECON_AND_STORE(dest, in2);
+      RECON_AND_STORE(dest, in3);
+      RECON_AND_STORE(dest, in4);
+      RECON_AND_STORE(dest, in5);
+      RECON_AND_STORE(dest, in6);
+      RECON_AND_STORE(dest, in7);
+      RECON_AND_STORE(dest, in8);
+      RECON_AND_STORE(dest, in9);
+      RECON_AND_STORE(dest, in10);
+      RECON_AND_STORE(dest, in11);
+      RECON_AND_STORE(dest, in12);
+      RECON_AND_STORE(dest, in13);
+      RECON_AND_STORE(dest, in14);
+      RECON_AND_STORE(dest, in15);
+
+      dest += 8 - (stride * 16);
+    }
+  }
+}
+
+void vp9_short_idct10_16x16_add_sse2(int16_t *input, uint8_t *dest,
+                                     int stride) {
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i final_rounding = _mm_set1_epi16(1<<5);
+  const __m128i zero = _mm_setzero_si128();
+
+  const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64);
+  const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64);
+  const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64);
+  const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64);
+  const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64);
+  const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64);
+  const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64);
+  const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64);
+
+  const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+  const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
+  const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64);
+  const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64);
+
+  const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
+  const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+  const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
+  const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
+  const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+
+  const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+
+  __m128i in0 = zero, in1 = zero, in2 = zero, in3 = zero, in4 = zero,
+          in5 = zero, in6 = zero, in7 = zero, in8 = zero, in9 = zero,
+          in10 = zero, in11 = zero, in12 = zero, in13 = zero,
+          in14 = zero, in15 = zero;
+  __m128i l0 = zero, l1 = zero, l2 = zero, l3 = zero, l4 = zero, l5 = zero,
+          l6 = zero, l7 = zero, l8 = zero, l9 = zero, l10 = zero, l11 = zero,
+          l12 = zero, l13 = zero, l14 = zero, l15 = zero;
+
+  __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
+          stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
+          stp1_8_0, stp1_12_0;
+  __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7,
+          stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15;
+  __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  int i;
+  // 1-D idct. Load input data.
+  in0 = _mm_load_si128((__m128i *)input);
+  in8 = _mm_load_si128((__m128i *)(input + 8 * 1));
+  in1 = _mm_load_si128((__m128i *)(input + 8 * 2));
+  in9 = _mm_load_si128((__m128i *)(input + 8 * 3));
+  in2 = _mm_load_si128((__m128i *)(input + 8 * 4));
+  in10 = _mm_load_si128((__m128i *)(input + 8 * 5));
+  in3 = _mm_load_si128((__m128i *)(input + 8 * 6));
+  in11 = _mm_load_si128((__m128i *)(input + 8 * 7));
+
+  TRANSPOSE_8X4(in0, in1, in2, in3, in0, in1, in2, in3);
+  TRANSPOSE_8X4(in8, in9, in10, in11, in8, in9, in10, in11);
+
+  // Stage2
+  {
+    const __m128i lo_1_15 = _mm_unpackhi_epi16(in0, in11);
+    const __m128i lo_9_7 = _mm_unpackhi_epi16(in8, in3);
+    const __m128i lo_5_11 = _mm_unpackhi_epi16(in2, in9);
+    const __m128i lo_13_3 = _mm_unpackhi_epi16(in10, in1);
+
+    tmp0 = _mm_madd_epi16(lo_1_15, stg2_0);
+    tmp2 = _mm_madd_epi16(lo_1_15, stg2_1);
+    tmp4 = _mm_madd_epi16(lo_9_7, stg2_2);
+    tmp6 = _mm_madd_epi16(lo_9_7, stg2_3);
+    tmp1 = _mm_madd_epi16(lo_5_11, stg2_4);
+    tmp3 = _mm_madd_epi16(lo_5_11, stg2_5);
+    tmp5 = _mm_madd_epi16(lo_13_3, stg2_6);
+    tmp7 = _mm_madd_epi16(lo_13_3, stg2_7);
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+    tmp1 = _mm_add_epi32(tmp1, rounding);
+    tmp3 = _mm_add_epi32(tmp3, rounding);
+    tmp5 = _mm_add_epi32(tmp5, rounding);
+    tmp7 = _mm_add_epi32(tmp7, rounding);
+
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+    tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
+    tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
+    tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
+    tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
+
+    stp2_8 = _mm_packs_epi32(tmp0, zero);
+    stp2_15 = _mm_packs_epi32(tmp2, zero);
+    stp2_9 = _mm_packs_epi32(tmp4, zero);
+    stp2_14 = _mm_packs_epi32(tmp6, zero);
+
+    stp2_10 = _mm_packs_epi32(tmp1, zero);
+    stp2_13 = _mm_packs_epi32(tmp3, zero);
+    stp2_11 = _mm_packs_epi32(tmp5, zero);
+    stp2_12 = _mm_packs_epi32(tmp7, zero);
+  }
+
+  // Stage3
+  {
+    const __m128i lo_2_14 = _mm_unpacklo_epi16(in1, in11);
+    const __m128i lo_10_6 = _mm_unpacklo_epi16(in9, in3);
+
+    tmp0 = _mm_madd_epi16(lo_2_14, stg3_0);
+    tmp2 = _mm_madd_epi16(lo_2_14, stg3_1);
+    tmp4 = _mm_madd_epi16(lo_10_6, stg3_2);
+    tmp6 = _mm_madd_epi16(lo_10_6, stg3_3);
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+
+    stp1_4 = _mm_packs_epi32(tmp0, zero);
+    stp1_7 = _mm_packs_epi32(tmp2, zero);
+    stp1_5 = _mm_packs_epi32(tmp4, zero);
+    stp1_6 = _mm_packs_epi32(tmp6, zero);
+
+    stp1_8_0 = _mm_add_epi16(stp2_8, stp2_9);
+    stp1_9 = _mm_sub_epi16(stp2_8, stp2_9);
+    stp1_10 = _mm_sub_epi16(stp2_11, stp2_10);
+    stp1_11 = _mm_add_epi16(stp2_11, stp2_10);
+
+    stp1_12_0 = _mm_add_epi16(stp2_12, stp2_13);
+    stp1_13 = _mm_sub_epi16(stp2_12, stp2_13);
+    stp1_14 = _mm_sub_epi16(stp2_15, stp2_14);
+    stp1_15 = _mm_add_epi16(stp2_15, stp2_14);
+  }
+
+  // Stage4
+  {
+    const __m128i lo_0_8 = _mm_unpacklo_epi16(in0, in8);
+    const __m128i lo_4_12 = _mm_unpacklo_epi16(in2, in10);
+    const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14);
+    const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
+
+    tmp0 = _mm_madd_epi16(lo_0_8, stg4_0);
+    tmp2 = _mm_madd_epi16(lo_0_8, stg4_1);
+    tmp4 = _mm_madd_epi16(lo_4_12, stg4_2);
+    tmp6 = _mm_madd_epi16(lo_4_12, stg4_3);
+    tmp1 = _mm_madd_epi16(lo_9_14, stg4_4);
+    tmp3 = _mm_madd_epi16(lo_9_14, stg4_5);
+    tmp5 = _mm_madd_epi16(lo_10_13, stg4_6);
+    tmp7 = _mm_madd_epi16(lo_10_13, stg4_7);
+
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+    tmp1 = _mm_add_epi32(tmp1, rounding);
+    tmp3 = _mm_add_epi32(tmp3, rounding);
+    tmp5 = _mm_add_epi32(tmp5, rounding);
+    tmp7 = _mm_add_epi32(tmp7, rounding);
+
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+    tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
+    tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
+    tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
+    tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
+
+    stp2_0 = _mm_packs_epi32(tmp0, zero);
+    stp2_1 = _mm_packs_epi32(tmp2, zero);
+    stp2_2 = _mm_packs_epi32(tmp4, zero);
+    stp2_3 = _mm_packs_epi32(tmp6, zero);
+    stp2_9 = _mm_packs_epi32(tmp1, zero);
+    stp2_14 = _mm_packs_epi32(tmp3, zero);
+    stp2_10 = _mm_packs_epi32(tmp5, zero);
+    stp2_13 = _mm_packs_epi32(tmp7, zero);
+
+    stp2_4 = _mm_add_epi16(stp1_4, stp1_5);
+    stp2_5 = _mm_sub_epi16(stp1_4, stp1_5);
+    stp2_6 = _mm_sub_epi16(stp1_7, stp1_6);
+    stp2_7 = _mm_add_epi16(stp1_7, stp1_6);
+  }
+
+  // Stage5 and Stage6
+  {
+    stp1_0 = _mm_add_epi16(stp2_0, stp2_3);
+    stp1_1 = _mm_add_epi16(stp2_1, stp2_2);
+    stp1_2 = _mm_sub_epi16(stp2_1, stp2_2);
+    stp1_3 = _mm_sub_epi16(stp2_0, stp2_3);
+
+    stp1_8 = _mm_add_epi16(stp1_8_0, stp1_11);
+    stp1_9 = _mm_add_epi16(stp2_9, stp2_10);
+    stp1_10 = _mm_sub_epi16(stp2_9, stp2_10);
+    stp1_11 = _mm_sub_epi16(stp1_8_0, stp1_11);
+
+    stp1_12 = _mm_sub_epi16(stp1_15, stp1_12_0);
+    stp1_13 = _mm_sub_epi16(stp2_14, stp2_13);
+    stp1_14 = _mm_add_epi16(stp2_14, stp2_13);
+    stp1_15 = _mm_add_epi16(stp1_15, stp1_12_0);
+  }
+
+  // Stage6
+  {
+    const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5);
+    const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
+    const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12);
+
+    tmp1 = _mm_madd_epi16(lo_6_5, stg4_1);
+    tmp3 = _mm_madd_epi16(lo_6_5, stg4_0);
+    tmp0 = _mm_madd_epi16(lo_10_13, stg6_0);
+    tmp2 = _mm_madd_epi16(lo_10_13, stg4_0);
+    tmp4 = _mm_madd_epi16(lo_11_12, stg6_0);
+    tmp6 = _mm_madd_epi16(lo_11_12, stg4_0);
+
+    tmp1 = _mm_add_epi32(tmp1, rounding);
+    tmp3 = _mm_add_epi32(tmp3, rounding);
+    tmp0 = _mm_add_epi32(tmp0, rounding);
+    tmp2 = _mm_add_epi32(tmp2, rounding);
+    tmp4 = _mm_add_epi32(tmp4, rounding);
+    tmp6 = _mm_add_epi32(tmp6, rounding);
+
+    tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
+    tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
+    tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+    tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+    tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
+    tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
+
+    stp1_5 = _mm_packs_epi32(tmp1, zero);
+    stp1_6 = _mm_packs_epi32(tmp3, zero);
+    stp2_10 = _mm_packs_epi32(tmp0, zero);
+    stp2_13 = _mm_packs_epi32(tmp2, zero);
+    stp2_11 = _mm_packs_epi32(tmp4, zero);
+    stp2_12 = _mm_packs_epi32(tmp6, zero);
+
+    stp2_0 = _mm_add_epi16(stp1_0, stp2_7);
+    stp2_1 = _mm_add_epi16(stp1_1, stp1_6);
+    stp2_2 = _mm_add_epi16(stp1_2, stp1_5);
+    stp2_3 = _mm_add_epi16(stp1_3, stp2_4);
+    stp2_4 = _mm_sub_epi16(stp1_3, stp2_4);
+    stp2_5 = _mm_sub_epi16(stp1_2, stp1_5);
+    stp2_6 = _mm_sub_epi16(stp1_1, stp1_6);
+    stp2_7 = _mm_sub_epi16(stp1_0, stp2_7);
+  }
+
+  // Stage7. Left 8x16 only.
+  l0 = _mm_add_epi16(stp2_0, stp1_15);
+  l1 = _mm_add_epi16(stp2_1, stp1_14);
+  l2 = _mm_add_epi16(stp2_2, stp2_13);
+  l3 = _mm_add_epi16(stp2_3, stp2_12);
+  l4 = _mm_add_epi16(stp2_4, stp2_11);
+  l5 = _mm_add_epi16(stp2_5, stp2_10);
+  l6 = _mm_add_epi16(stp2_6, stp1_9);
+  l7 = _mm_add_epi16(stp2_7, stp1_8);
+  l8 = _mm_sub_epi16(stp2_7, stp1_8);
+  l9 = _mm_sub_epi16(stp2_6, stp1_9);
+  l10 = _mm_sub_epi16(stp2_5, stp2_10);
+  l11 = _mm_sub_epi16(stp2_4, stp2_11);
+  l12 = _mm_sub_epi16(stp2_3, stp2_12);
+  l13 = _mm_sub_epi16(stp2_2, stp2_13);
+  l14 = _mm_sub_epi16(stp2_1, stp1_14);
+  l15 = _mm_sub_epi16(stp2_0, stp1_15);
+
+  // 2-D idct. We do 2 8x16 blocks.
+  for (i = 0; i < 2; i++) {
+    if (i == 0)
+      TRANSPOSE_4X8(l0, l1, l2, l3, l4, l5, l6, l7, in0, in1, in2, in3, in4,
+                    in5, in6, in7);
+
+    if (i == 1)
+      TRANSPOSE_4X8(l8, l9, l10, l11, l12, l13, l14, l15, in0, in1, in2, in3,
+                    in4, in5, in6, in7);
+
+    in8 = in9 = in10 = in11 = in12 = in13 = in14 = in15 = zero;
+
+    IDCT16x16_1D
+
+    // Stage7
+    in0 = _mm_add_epi16(stp2_0, stp1_15);
+    in1 = _mm_add_epi16(stp2_1, stp1_14);
+    in2 = _mm_add_epi16(stp2_2, stp2_13);
+    in3 = _mm_add_epi16(stp2_3, stp2_12);
+    in4 = _mm_add_epi16(stp2_4, stp2_11);
+    in5 = _mm_add_epi16(stp2_5, stp2_10);
+    in6 = _mm_add_epi16(stp2_6, stp1_9);
+    in7 = _mm_add_epi16(stp2_7, stp1_8);
+    in8 = _mm_sub_epi16(stp2_7, stp1_8);
+    in9 = _mm_sub_epi16(stp2_6, stp1_9);
+    in10 = _mm_sub_epi16(stp2_5, stp2_10);
+    in11 = _mm_sub_epi16(stp2_4, stp2_11);
+    in12 = _mm_sub_epi16(stp2_3, stp2_12);
+    in13 = _mm_sub_epi16(stp2_2, stp2_13);
+    in14 = _mm_sub_epi16(stp2_1, stp1_14);
+    in15 = _mm_sub_epi16(stp2_0, stp1_15);
+
+    // Final rounding and shift
+    in0 = _mm_adds_epi16(in0, final_rounding);
+    in1 = _mm_adds_epi16(in1, final_rounding);
+    in2 = _mm_adds_epi16(in2, final_rounding);
+    in3 = _mm_adds_epi16(in3, final_rounding);
+    in4 = _mm_adds_epi16(in4, final_rounding);
+    in5 = _mm_adds_epi16(in5, final_rounding);
+    in6 = _mm_adds_epi16(in6, final_rounding);
+    in7 = _mm_adds_epi16(in7, final_rounding);
+    in8 = _mm_adds_epi16(in8, final_rounding);
+    in9 = _mm_adds_epi16(in9, final_rounding);
+    in10 = _mm_adds_epi16(in10, final_rounding);
+    in11 = _mm_adds_epi16(in11, final_rounding);
+    in12 = _mm_adds_epi16(in12, final_rounding);
+    in13 = _mm_adds_epi16(in13, final_rounding);
+    in14 = _mm_adds_epi16(in14, final_rounding);
+    in15 = _mm_adds_epi16(in15, final_rounding);
+
+    in0 = _mm_srai_epi16(in0, 6);
+    in1 = _mm_srai_epi16(in1, 6);
+    in2 = _mm_srai_epi16(in2, 6);
+    in3 = _mm_srai_epi16(in3, 6);
+    in4 = _mm_srai_epi16(in4, 6);
+    in5 = _mm_srai_epi16(in5, 6);
+    in6 = _mm_srai_epi16(in6, 6);
+    in7 = _mm_srai_epi16(in7, 6);
+    in8 = _mm_srai_epi16(in8, 6);
+    in9 = _mm_srai_epi16(in9, 6);
+    in10 = _mm_srai_epi16(in10, 6);
+    in11 = _mm_srai_epi16(in11, 6);
+    in12 = _mm_srai_epi16(in12, 6);
+    in13 = _mm_srai_epi16(in13, 6);
+    in14 = _mm_srai_epi16(in14, 6);
+    in15 = _mm_srai_epi16(in15, 6);
+
+    RECON_AND_STORE(dest, in0);
+    RECON_AND_STORE(dest, in1);
+    RECON_AND_STORE(dest, in2);
+    RECON_AND_STORE(dest, in3);
+    RECON_AND_STORE(dest, in4);
+    RECON_AND_STORE(dest, in5);
+    RECON_AND_STORE(dest, in6);
+    RECON_AND_STORE(dest, in7);
+    RECON_AND_STORE(dest, in8);
+    RECON_AND_STORE(dest, in9);
+    RECON_AND_STORE(dest, in10);
+    RECON_AND_STORE(dest, in11);
+    RECON_AND_STORE(dest, in12);
+    RECON_AND_STORE(dest, in13);
+    RECON_AND_STORE(dest, in14);
+    RECON_AND_STORE(dest, in15);
+
+    dest += 8 - (stride * 16);
+  }
+}
+
+void vp9_short_idct32x32_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+  const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i final_rounding = _mm_set1_epi16(1<<5);
+
+  // idct constants for each stage
+  const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64);
+  const __m128i stg1_1 = pair_set_epi16(cospi_1_64, cospi_31_64);
+  const __m128i stg1_2 = pair_set_epi16(cospi_15_64, -cospi_17_64);
+  const __m128i stg1_3 = pair_set_epi16(cospi_17_64, cospi_15_64);
+  const __m128i stg1_4 = pair_set_epi16(cospi_23_64, -cospi_9_64);
+  const __m128i stg1_5 = pair_set_epi16(cospi_9_64, cospi_23_64);
+  const __m128i stg1_6 = pair_set_epi16(cospi_7_64, -cospi_25_64);
+  const __m128i stg1_7 = pair_set_epi16(cospi_25_64, cospi_7_64);
+  const __m128i stg1_8 = pair_set_epi16(cospi_27_64, -cospi_5_64);
+  const __m128i stg1_9 = pair_set_epi16(cospi_5_64, cospi_27_64);
+  const __m128i stg1_10 = pair_set_epi16(cospi_11_64, -cospi_21_64);
+  const __m128i stg1_11 = pair_set_epi16(cospi_21_64, cospi_11_64);
+  const __m128i stg1_12 = pair_set_epi16(cospi_19_64, -cospi_13_64);
+  const __m128i stg1_13 = pair_set_epi16(cospi_13_64, cospi_19_64);
+  const __m128i stg1_14 = pair_set_epi16(cospi_3_64, -cospi_29_64);
+  const __m128i stg1_15 = pair_set_epi16(cospi_29_64, cospi_3_64);
+
+  const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64);
+  const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64);
+  const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64);
+  const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64);
+  const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64);
+  const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64);
+  const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64);
+  const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64);
+
+  const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+  const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
+  const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64);
+  const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64);
+  const __m128i stg3_4 = pair_set_epi16(-cospi_4_64, cospi_28_64);
+  const __m128i stg3_5 = pair_set_epi16(cospi_28_64, cospi_4_64);
+  const __m128i stg3_6 = pair_set_epi16(-cospi_28_64, -cospi_4_64);
+  const __m128i stg3_8 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+  const __m128i stg3_9 = pair_set_epi16(cospi_12_64, cospi_20_64);
+  const __m128i stg3_10 = pair_set_epi16(-cospi_12_64, -cospi_20_64);
+
+  const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
+  const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+  const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
+  const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
+
+  const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+
+  __m128i in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11, in12,
+          in13, in14, in15, in16, in17, in18, in19, in20, in21, in22, in23,
+          in24, in25, in26, in27, in28, in29, in30, in31;
+  __m128i col[128];
+  __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
+          stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
+          stp1_16, stp1_17, stp1_18, stp1_19, stp1_20, stp1_21, stp1_22,
+          stp1_23, stp1_24, stp1_25, stp1_26, stp1_27, stp1_28, stp1_29,
+          stp1_30, stp1_31;
+  __m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7,
+          stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15,
+          stp2_16, stp2_17, stp2_18, stp2_19, stp2_20, stp2_21, stp2_22,
+          stp2_23, stp2_24, stp2_25, stp2_26, stp2_27, stp2_28, stp2_29,
+          stp2_30, stp2_31;
+  __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+  int i, j;
+
+  // We work on a 8x32 block each time, and loop 8 times for 2-D 32x32 idct.
+  for (i = 0; i < 8; i++) {
+    if (i < 4) {
+      // First 1-D idct
+      // Load input data.
+      in0 = _mm_load_si128((__m128i *)input);
+      in8 = _mm_load_si128((__m128i *)(input + 8 * 1));
+      in16 = _mm_load_si128((__m128i *)(input + 8 * 2));
+      in24 = _mm_load_si128((__m128i *)(input + 8 * 3));
+      in1 = _mm_load_si128((__m128i *)(input + 8 * 4));
+      in9 = _mm_load_si128((__m128i *)(input + 8 * 5));
+      in17 = _mm_load_si128((__m128i *)(input + 8 * 6));
+      in25 = _mm_load_si128((__m128i *)(input + 8 * 7));
+      in2 = _mm_load_si128((__m128i *)(input + 8 * 8));
+      in10 = _mm_load_si128((__m128i *)(input + 8 * 9));
+      in18 = _mm_load_si128((__m128i *)(input + 8 * 10));
+      in26 = _mm_load_si128((__m128i *)(input + 8 * 11));
+      in3 = _mm_load_si128((__m128i *)(input + 8 * 12));
+      in11 = _mm_load_si128((__m128i *)(input + 8 * 13));
+      in19 = _mm_load_si128((__m128i *)(input + 8 * 14));
+      in27 = _mm_load_si128((__m128i *)(input + 8 * 15));
+
+      in4 = _mm_load_si128((__m128i *)(input + 8 * 16));
+      in12 = _mm_load_si128((__m128i *)(input + 8 * 17));
+      in20 = _mm_load_si128((__m128i *)(input + 8 * 18));
+      in28 = _mm_load_si128((__m128i *)(input + 8 * 19));
+      in5 = _mm_load_si128((__m128i *)(input + 8 * 20));
+      in13 = _mm_load_si128((__m128i *)(input + 8 * 21));
+      in21 = _mm_load_si128((__m128i *)(input + 8 * 22));
+      in29 = _mm_load_si128((__m128i *)(input + 8 * 23));
+      in6 = _mm_load_si128((__m128i *)(input + 8 * 24));
+      in14 = _mm_load_si128((__m128i *)(input + 8 * 25));
+      in22 = _mm_load_si128((__m128i *)(input + 8 * 26));
+      in30 = _mm_load_si128((__m128i *)(input + 8 * 27));
+      in7 = _mm_load_si128((__m128i *)(input + 8 * 28));
+      in15 = _mm_load_si128((__m128i *)(input + 8 * 29));
+      in23 = _mm_load_si128((__m128i *)(input + 8 * 30));
+      in31 = _mm_load_si128((__m128i *)(input + 8 * 31));
+
+      input += 256;
+
+      // Transpose 32x8 block to 8x32 block
+      TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
+                    in4, in5, in6, in7);
+      TRANSPOSE_8X8(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
+                    in10, in11, in12, in13, in14, in15);
+      TRANSPOSE_8X8(in16, in17, in18, in19, in20, in21, in22, in23, in16, in17,
+                    in18, in19, in20, in21, in22, in23);
+      TRANSPOSE_8X8(in24, in25, in26, in27, in28, in29, in30, in31, in24, in25,
+                    in26, in27, in28, in29, in30, in31);
+    } else {
+      // Second 1-D idct
+      j = i - 4;
+
+      // Transpose 32x8 block to 8x32 block
+      TRANSPOSE_8X8(col[j * 8 + 0], col[j * 8 + 1], col[j * 8 + 2],
+                    col[j * 8 + 3], col[j * 8 + 4], col[j * 8 + 5],
+                    col[j * 8 + 6], col[j * 8 + 7], in0, in1, in2, in3, in4,
+                    in5, in6, in7);
+      j += 4;
+      TRANSPOSE_8X8(col[j * 8 + 0], col[j * 8 + 1], col[j * 8 + 2],
+                    col[j * 8 + 3], col[j * 8 + 4], col[j * 8 + 5],
+                    col[j * 8 + 6], col[j * 8 + 7], in8, in9, in10,
+                    in11, in12, in13, in14, in15);
+      j += 4;
+      TRANSPOSE_8X8(col[j * 8 + 0], col[j * 8 + 1], col[j * 8 + 2],
+                    col[j * 8 + 3], col[j * 8 + 4], col[j * 8 + 5],
+                    col[j * 8 + 6], col[j * 8 + 7], in16, in17, in18,
+                    in19, in20, in21, in22, in23);
+      j += 4;
+      TRANSPOSE_8X8(col[j * 8 + 0], col[j * 8 + 1], col[j * 8 + 2],
+                    col[j * 8 + 3], col[j * 8 + 4], col[j * 8 + 5],
+                    col[j * 8 + 6], col[j * 8 + 7], in24, in25, in26, in27,
+                    in28, in29, in30, in31);
+    }
+
+    // Stage1
+    {
+      const __m128i lo_1_31 = _mm_unpacklo_epi16(in1, in31);
+      const __m128i hi_1_31 = _mm_unpackhi_epi16(in1, in31);
+      const __m128i lo_17_15 = _mm_unpacklo_epi16(in17, in15);
+      const __m128i hi_17_15 = _mm_unpackhi_epi16(in17, in15);
+
+      const __m128i lo_9_23 = _mm_unpacklo_epi16(in9, in23);
+      const __m128i hi_9_23 = _mm_unpackhi_epi16(in9, in23);
+      const __m128i lo_25_7= _mm_unpacklo_epi16(in25, in7);
+      const __m128i hi_25_7 = _mm_unpackhi_epi16(in25, in7);
+
+      const __m128i lo_5_27 = _mm_unpacklo_epi16(in5, in27);
+      const __m128i hi_5_27 = _mm_unpackhi_epi16(in5, in27);
+      const __m128i lo_21_11 = _mm_unpacklo_epi16(in21, in11);
+      const __m128i hi_21_11 = _mm_unpackhi_epi16(in21, in11);
+
+      const __m128i lo_13_19 = _mm_unpacklo_epi16(in13, in19);
+      const __m128i hi_13_19 = _mm_unpackhi_epi16(in13, in19);
+      const __m128i lo_29_3 = _mm_unpacklo_epi16(in29, in3);
+      const __m128i hi_29_3 = _mm_unpackhi_epi16(in29, in3);
+
+      MULTIPLICATION_AND_ADD(lo_1_31, hi_1_31, lo_17_15, hi_17_15, stg1_0,
+                             stg1_1, stg1_2, stg1_3, stp1_16, stp1_31,
+                             stp1_17, stp1_30)
+      MULTIPLICATION_AND_ADD(lo_9_23, hi_9_23, lo_25_7, hi_25_7, stg1_4,
+                             stg1_5, stg1_6, stg1_7, stp1_18, stp1_29,
+                             stp1_19, stp1_28)
+      MULTIPLICATION_AND_ADD(lo_5_27, hi_5_27, lo_21_11, hi_21_11, stg1_8,
+                             stg1_9, stg1_10, stg1_11, stp1_20, stp1_27,
+                             stp1_21, stp1_26)
+      MULTIPLICATION_AND_ADD(lo_13_19, hi_13_19, lo_29_3, hi_29_3, stg1_12,
+                             stg1_13, stg1_14, stg1_15, stp1_22, stp1_25,
+                             stp1_23, stp1_24)
+    }
+
+    // Stage2
+    {
+      const __m128i lo_2_30 = _mm_unpacklo_epi16(in2, in30);
+      const __m128i hi_2_30 = _mm_unpackhi_epi16(in2, in30);
+      const __m128i lo_18_14 = _mm_unpacklo_epi16(in18, in14);
+      const __m128i hi_18_14 = _mm_unpackhi_epi16(in18, in14);
+
+      const __m128i lo_10_22 = _mm_unpacklo_epi16(in10, in22);
+      const __m128i hi_10_22 = _mm_unpackhi_epi16(in10, in22);
+      const __m128i lo_26_6 = _mm_unpacklo_epi16(in26, in6);
+      const __m128i hi_26_6 = _mm_unpackhi_epi16(in26, in6);
+
+      MULTIPLICATION_AND_ADD(lo_2_30, hi_2_30, lo_18_14, hi_18_14, stg2_0,
+                             stg2_1, stg2_2, stg2_3, stp2_8, stp2_15, stp2_9,
+                             stp2_14)
+      MULTIPLICATION_AND_ADD(lo_10_22, hi_10_22, lo_26_6, hi_26_6, stg2_4,
+                             stg2_5, stg2_6, stg2_7, stp2_10, stp2_13,
+                             stp2_11, stp2_12)
+
+      stp2_16 = _mm_add_epi16(stp1_16, stp1_17);
+      stp2_17 = _mm_sub_epi16(stp1_16, stp1_17);
+      stp2_18 = _mm_sub_epi16(stp1_19, stp1_18);
+      stp2_19 = _mm_add_epi16(stp1_19, stp1_18);
+
+      stp2_20 = _mm_add_epi16(stp1_20, stp1_21);
+      stp2_21 = _mm_sub_epi16(stp1_20, stp1_21);
+      stp2_22 = _mm_sub_epi16(stp1_23, stp1_22);
+      stp2_23 = _mm_add_epi16(stp1_23, stp1_22);
+
+      stp2_24 = _mm_add_epi16(stp1_24, stp1_25);
+      stp2_25 = _mm_sub_epi16(stp1_24, stp1_25);
+      stp2_26 = _mm_sub_epi16(stp1_27, stp1_26);
+      stp2_27 = _mm_add_epi16(stp1_27, stp1_26);
+
+      stp2_28 = _mm_add_epi16(stp1_28, stp1_29);
+      stp2_29 = _mm_sub_epi16(stp1_28, stp1_29);
+      stp2_30 = _mm_sub_epi16(stp1_31, stp1_30);
+      stp2_31 = _mm_add_epi16(stp1_31, stp1_30);
+    }
+
+    // Stage3
+    {
+      const __m128i lo_4_28 = _mm_unpacklo_epi16(in4, in28);
+      const __m128i hi_4_28 = _mm_unpackhi_epi16(in4, in28);
+      const __m128i lo_20_12 = _mm_unpacklo_epi16(in20, in12);
+      const __m128i hi_20_12 = _mm_unpackhi_epi16(in20, in12);
+
+      const __m128i lo_17_30 = _mm_unpacklo_epi16(stp2_17, stp2_30);
+      const __m128i hi_17_30 = _mm_unpackhi_epi16(stp2_17, stp2_30);
+      const __m128i lo_18_29 = _mm_unpacklo_epi16(stp2_18, stp2_29);
+      const __m128i hi_18_29 = _mm_unpackhi_epi16(stp2_18, stp2_29);
+
+      const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26);
+      const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26);
+      const __m128i lo_22_25 = _mm_unpacklo_epi16(stp2_22, stp2_25);
+      const __m128i hi_22_25 = _mm_unpackhi_epi16(stp2_22, stp2_25);
+
+      MULTIPLICATION_AND_ADD(lo_4_28, hi_4_28, lo_20_12, hi_20_12, stg3_0,
+                             stg3_1, stg3_2, stg3_3, stp1_4, stp1_7, stp1_5,
+                             stp1_6)
+
+      stp1_8 = _mm_add_epi16(stp2_8, stp2_9);
+      stp1_9 = _mm_sub_epi16(stp2_8, stp2_9);
+      stp1_10 = _mm_sub_epi16(stp2_11, stp2_10);
+      stp1_11 = _mm_add_epi16(stp2_11, stp2_10);
+      stp1_12 = _mm_add_epi16(stp2_12, stp2_13);
+      stp1_13 = _mm_sub_epi16(stp2_12, stp2_13);
+      stp1_14 = _mm_sub_epi16(stp2_15, stp2_14);
+      stp1_15 = _mm_add_epi16(stp2_15, stp2_14);
+
+      MULTIPLICATION_AND_ADD(lo_17_30, hi_17_30, lo_18_29, hi_18_29, stg3_4,
+                             stg3_5, stg3_6, stg3_4, stp1_17, stp1_30,
+                             stp1_18, stp1_29)
+      MULTIPLICATION_AND_ADD(lo_21_26, hi_21_26, lo_22_25, hi_22_25, stg3_8,
+                             stg3_9, stg3_10, stg3_8, stp1_21, stp1_26,
+                             stp1_22, stp1_25)
+
+      stp1_16 = stp2_16;
+      stp1_31 = stp2_31;
+      stp1_19 = stp2_19;
+      stp1_20 = stp2_20;
+      stp1_23 = stp2_23;
+      stp1_24 = stp2_24;
+      stp1_27 = stp2_27;
+      stp1_28 = stp2_28;
+    }
+
+    // Stage4
+    {
+      const __m128i lo_0_16 = _mm_unpacklo_epi16(in0, in16);
+      const __m128i hi_0_16 = _mm_unpackhi_epi16(in0, in16);
+      const __m128i lo_8_24 = _mm_unpacklo_epi16(in8, in24);
+      const __m128i hi_8_24 = _mm_unpackhi_epi16(in8, in24);
+
+      const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14);
+      const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14);
+      const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
+      const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13);
+
+      MULTIPLICATION_AND_ADD(lo_0_16, hi_0_16, lo_8_24, hi_8_24, stg4_0,
+                             stg4_1, stg4_2, stg4_3, stp2_0, stp2_1,
+                             stp2_2, stp2_3)
+
+      stp2_4 = _mm_add_epi16(stp1_4, stp1_5);
+      stp2_5 = _mm_sub_epi16(stp1_4, stp1_5);
+      stp2_6 = _mm_sub_epi16(stp1_7, stp1_6);
+      stp2_7 = _mm_add_epi16(stp1_7, stp1_6);
+
+      MULTIPLICATION_AND_ADD(lo_9_14, hi_9_14, lo_10_13, hi_10_13, stg4_4,
+                             stg4_5, stg4_6, stg4_4, stp2_9, stp2_14,
+                             stp2_10, stp2_13)
+
+      stp2_8 = stp1_8;
+      stp2_15 = stp1_15;
+      stp2_11 = stp1_11;
+      stp2_12 = stp1_12;
+
+      stp2_16 = _mm_add_epi16(stp1_16, stp1_19);
+      stp2_17 = _mm_add_epi16(stp1_17, stp1_18);
+      stp2_18 = _mm_sub_epi16(stp1_17, stp1_18);
+      stp2_19 = _mm_sub_epi16(stp1_16, stp1_19);
+      stp2_20 = _mm_sub_epi16(stp1_23, stp1_20);
+      stp2_21 = _mm_sub_epi16(stp1_22, stp1_21);
+      stp2_22 = _mm_add_epi16(stp1_22, stp1_21);
+      stp2_23 = _mm_add_epi16(stp1_23, stp1_20);
+
+      stp2_24 = _mm_add_epi16(stp1_24, stp1_27);
+      stp2_25 = _mm_add_epi16(stp1_25, stp1_26);
+      stp2_26 = _mm_sub_epi16(stp1_25, stp1_26);
+      stp2_27 = _mm_sub_epi16(stp1_24, stp1_27);
+      stp2_28 = _mm_sub_epi16(stp1_31, stp1_28);
+      stp2_29 = _mm_sub_epi16(stp1_30, stp1_29);
+      stp2_30 = _mm_add_epi16(stp1_29, stp1_30);
+      stp2_31 = _mm_add_epi16(stp1_28, stp1_31);
+    }
+
+    // Stage5
+    {
+      const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5);
+      const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5);
+      const __m128i lo_18_29 = _mm_unpacklo_epi16(stp2_18, stp2_29);
+      const __m128i hi_18_29 = _mm_unpackhi_epi16(stp2_18, stp2_29);
+
+      const __m128i lo_19_28 = _mm_unpacklo_epi16(stp2_19, stp2_28);
+      const __m128i hi_19_28 = _mm_unpackhi_epi16(stp2_19, stp2_28);
+      const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27);
+      const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27);
+
+      const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26);
+      const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26);
+
+      stp1_0 = _mm_add_epi16(stp2_0, stp2_3);
+      stp1_1 = _mm_add_epi16(stp2_1, stp2_2);
+      stp1_2 = _mm_sub_epi16(stp2_1, stp2_2);
+      stp1_3 = _mm_sub_epi16(stp2_0, stp2_3);
+
+      tmp0 = _mm_madd_epi16(lo_6_5, stg4_1);
+      tmp1 = _mm_madd_epi16(hi_6_5, stg4_1);
+      tmp2 = _mm_madd_epi16(lo_6_5, stg4_0);
+      tmp3 = _mm_madd_epi16(hi_6_5, stg4_0);
+
+      tmp0 = _mm_add_epi32(tmp0, rounding);
+      tmp1 = _mm_add_epi32(tmp1, rounding);
+      tmp2 = _mm_add_epi32(tmp2, rounding);
+      tmp3 = _mm_add_epi32(tmp3, rounding);
+
+      tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
+      tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
+      tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
+      tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
+
+      stp1_5 = _mm_packs_epi32(tmp0, tmp1);
+      stp1_6 = _mm_packs_epi32(tmp2, tmp3);
+
+      stp1_4 = stp2_4;
+      stp1_7 = stp2_7;
+
+      stp1_8 = _mm_add_epi16(stp2_8, stp2_11);
+      stp1_9 = _mm_add_epi16(stp2_9, stp2_10);
+      stp1_10 = _mm_sub_epi16(stp2_9, stp2_10);
+      stp1_11 = _mm_sub_epi16(stp2_8, stp2_11);
+      stp1_12 = _mm_sub_epi16(stp2_15, stp2_12);
+      stp1_13 = _mm_sub_epi16(stp2_14, stp2_13);
+      stp1_14 = _mm_add_epi16(stp2_14, stp2_13);
+      stp1_15 = _mm_add_epi16(stp2_15, stp2_12);
+
+      stp1_16 = stp2_16;
+      stp1_17 = stp2_17;
+
+      MULTIPLICATION_AND_ADD(lo_18_29, hi_18_29, lo_19_28, hi_19_28, stg4_4,
+                             stg4_5, stg4_4, stg4_5, stp1_18, stp1_29,
+                             stp1_19, stp1_28)
+      MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg4_6,
+                             stg4_4, stg4_6, stg4_4, stp1_20, stp1_27,
+                             stp1_21, stp1_26)
+
+      stp1_22 = stp2_22;
+      stp1_23 = stp2_23;
+      stp1_24 = stp2_24;
+      stp1_25 = stp2_25;
+      stp1_30 = stp2_30;
+      stp1_31 = stp2_31;
+    }
+
+    // Stage6
+    {
+      const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
+      const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13);
+      const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12);
+      const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12);
+
+      stp2_0 = _mm_add_epi16(stp1_0, stp1_7);
+      stp2_1 = _mm_add_epi16(stp1_1, stp1_6);
+      stp2_2 = _mm_add_epi16(stp1_2, stp1_5);
+      stp2_3 = _mm_add_epi16(stp1_3, stp1_4);
+      stp2_4 = _mm_sub_epi16(stp1_3, stp1_4);
+      stp2_5 = _mm_sub_epi16(stp1_2, stp1_5);
+      stp2_6 = _mm_sub_epi16(stp1_1, stp1_6);
+      stp2_7 = _mm_sub_epi16(stp1_0, stp1_7);
+
+      stp2_8 = stp1_8;
+      stp2_9 = stp1_9;
+      stp2_14 = stp1_14;
+      stp2_15 = stp1_15;
+
+      MULTIPLICATION_AND_ADD(lo_10_13, hi_10_13, lo_11_12, hi_11_12,
+                             stg6_0, stg4_0, stg6_0, stg4_0, stp2_10,
+                             stp2_13, stp2_11, stp2_12)
+
+      stp2_16 = _mm_add_epi16(stp1_16, stp1_23);
+      stp2_17 = _mm_add_epi16(stp1_17, stp1_22);
+      stp2_18 = _mm_add_epi16(stp1_18, stp1_21);
+      stp2_19 = _mm_add_epi16(stp1_19, stp1_20);
+      stp2_20 = _mm_sub_epi16(stp1_19, stp1_20);
+      stp2_21 = _mm_sub_epi16(stp1_18, stp1_21);
+      stp2_22 = _mm_sub_epi16(stp1_17, stp1_22);
+      stp2_23 = _mm_sub_epi16(stp1_16, stp1_23);
+
+      stp2_24 = _mm_sub_epi16(stp1_31, stp1_24);
+      stp2_25 = _mm_sub_epi16(stp1_30, stp1_25);
+      stp2_26 = _mm_sub_epi16(stp1_29, stp1_26);
+      stp2_27 = _mm_sub_epi16(stp1_28, stp1_27);
+      stp2_28 = _mm_add_epi16(stp1_27, stp1_28);
+      stp2_29 = _mm_add_epi16(stp1_26, stp1_29);
+      stp2_30 = _mm_add_epi16(stp1_25, stp1_30);
+      stp2_31 = _mm_add_epi16(stp1_24, stp1_31);
+    }
+
+    // Stage7
+    {
+      const __m128i lo_20_27 = _mm_unpacklo_epi16(stp2_20, stp2_27);
+      const __m128i hi_20_27 = _mm_unpackhi_epi16(stp2_20, stp2_27);
+      const __m128i lo_21_26 = _mm_unpacklo_epi16(stp2_21, stp2_26);
+      const __m128i hi_21_26 = _mm_unpackhi_epi16(stp2_21, stp2_26);
+
+      const __m128i lo_22_25 = _mm_unpacklo_epi16(stp2_22, stp2_25);
+      const __m128i hi_22_25 = _mm_unpackhi_epi16(stp2_22, stp2_25);
+      const __m128i lo_23_24 = _mm_unpacklo_epi16(stp2_23, stp2_24);
+      const __m128i hi_23_24 = _mm_unpackhi_epi16(stp2_23, stp2_24);
+
+      stp1_0 = _mm_add_epi16(stp2_0, stp2_15);
+      stp1_1 = _mm_add_epi16(stp2_1, stp2_14);
+      stp1_2 = _mm_add_epi16(stp2_2, stp2_13);
+      stp1_3 = _mm_add_epi16(stp2_3, stp2_12);
+      stp1_4 = _mm_add_epi16(stp2_4, stp2_11);
+      stp1_5 = _mm_add_epi16(stp2_5, stp2_10);
+      stp1_6 = _mm_add_epi16(stp2_6, stp2_9);
+      stp1_7 = _mm_add_epi16(stp2_7, stp2_8);
+      stp1_8 = _mm_sub_epi16(stp2_7, stp2_8);
+      stp1_9 = _mm_sub_epi16(stp2_6, stp2_9);
+      stp1_10 = _mm_sub_epi16(stp2_5, stp2_10);
+      stp1_11 = _mm_sub_epi16(stp2_4, stp2_11);
+      stp1_12 = _mm_sub_epi16(stp2_3, stp2_12);
+      stp1_13 = _mm_sub_epi16(stp2_2, stp2_13);
+      stp1_14 = _mm_sub_epi16(stp2_1, stp2_14);
+      stp1_15 = _mm_sub_epi16(stp2_0, stp2_15);
+
+      stp1_16 = stp2_16;
+      stp1_17 = stp2_17;
+      stp1_18 = stp2_18;
+      stp1_19 = stp2_19;
+
+      MULTIPLICATION_AND_ADD(lo_20_27, hi_20_27, lo_21_26, hi_21_26, stg6_0,
+                             stg4_0, stg6_0, stg4_0, stp1_20, stp1_27,
+                             stp1_21, stp1_26)
+      MULTIPLICATION_AND_ADD(lo_22_25, hi_22_25, lo_23_24, hi_23_24, stg6_0,
+                             stg4_0, stg6_0, stg4_0, stp1_22, stp1_25,
+                             stp1_23, stp1_24)
+
+      stp1_28 = stp2_28;
+      stp1_29 = stp2_29;
+      stp1_30 = stp2_30;
+      stp1_31 = stp2_31;
+    }
+
+    // final stage
+    if (i < 4) {
+      // 1_D: Store 32 intermediate results for each 8x32 block.
+      col[i * 32 + 0] = _mm_add_epi16(stp1_0, stp1_31);
+      col[i * 32 + 1] = _mm_add_epi16(stp1_1, stp1_30);
+      col[i * 32 + 2] = _mm_add_epi16(stp1_2, stp1_29);
+      col[i * 32 + 3] = _mm_add_epi16(stp1_3, stp1_28);
+      col[i * 32 + 4] = _mm_add_epi16(stp1_4, stp1_27);
+      col[i * 32 + 5] = _mm_add_epi16(stp1_5, stp1_26);
+      col[i * 32 + 6] = _mm_add_epi16(stp1_6, stp1_25);
+      col[i * 32 + 7] = _mm_add_epi16(stp1_7, stp1_24);
+      col[i * 32 + 8] = _mm_add_epi16(stp1_8, stp1_23);
+      col[i * 32 + 9] = _mm_add_epi16(stp1_9, stp1_22);
+      col[i * 32 + 10] = _mm_add_epi16(stp1_10, stp1_21);
+      col[i * 32 + 11] = _mm_add_epi16(stp1_11, stp1_20);
+      col[i * 32 + 12] = _mm_add_epi16(stp1_12, stp1_19);
+      col[i * 32 + 13] = _mm_add_epi16(stp1_13, stp1_18);
+      col[i * 32 + 14] = _mm_add_epi16(stp1_14, stp1_17);
+      col[i * 32 + 15] = _mm_add_epi16(stp1_15, stp1_16);
+      col[i * 32 + 16] = _mm_sub_epi16(stp1_15, stp1_16);
+      col[i * 32 + 17] = _mm_sub_epi16(stp1_14, stp1_17);
+      col[i * 32 + 18] = _mm_sub_epi16(stp1_13, stp1_18);
+      col[i * 32 + 19] = _mm_sub_epi16(stp1_12, stp1_19);
+      col[i * 32 + 20] = _mm_sub_epi16(stp1_11, stp1_20);
+      col[i * 32 + 21] = _mm_sub_epi16(stp1_10, stp1_21);
+      col[i * 32 + 22] = _mm_sub_epi16(stp1_9, stp1_22);
+      col[i * 32 + 23] = _mm_sub_epi16(stp1_8, stp1_23);
+      col[i * 32 + 24] = _mm_sub_epi16(stp1_7, stp1_24);
+      col[i * 32 + 25] = _mm_sub_epi16(stp1_6, stp1_25);
+      col[i * 32 + 26] = _mm_sub_epi16(stp1_5, stp1_26);
+      col[i * 32 + 27] = _mm_sub_epi16(stp1_4, stp1_27);
+      col[i * 32 + 28] = _mm_sub_epi16(stp1_3, stp1_28);
+      col[i * 32 + 29] = _mm_sub_epi16(stp1_2, stp1_29);
+      col[i * 32 + 30] = _mm_sub_epi16(stp1_1, stp1_30);
+      col[i * 32 + 31] = _mm_sub_epi16(stp1_0, stp1_31);
+    } else {
+      const __m128i zero = _mm_setzero_si128();
+
+      // 2_D: Calculate the results and store them to destination.
+      in0 = _mm_add_epi16(stp1_0, stp1_31);
+      in1 = _mm_add_epi16(stp1_1, stp1_30);
+      in2 = _mm_add_epi16(stp1_2, stp1_29);
+      in3 = _mm_add_epi16(stp1_3, stp1_28);
+      in4 = _mm_add_epi16(stp1_4, stp1_27);
+      in5 = _mm_add_epi16(stp1_5, stp1_26);
+      in6 = _mm_add_epi16(stp1_6, stp1_25);
+      in7 = _mm_add_epi16(stp1_7, stp1_24);
+      in8 = _mm_add_epi16(stp1_8, stp1_23);
+      in9 = _mm_add_epi16(stp1_9, stp1_22);
+      in10 = _mm_add_epi16(stp1_10, stp1_21);
+      in11 = _mm_add_epi16(stp1_11, stp1_20);
+      in12 = _mm_add_epi16(stp1_12, stp1_19);
+      in13 = _mm_add_epi16(stp1_13, stp1_18);
+      in14 = _mm_add_epi16(stp1_14, stp1_17);
+      in15 = _mm_add_epi16(stp1_15, stp1_16);
+      in16 = _mm_sub_epi16(stp1_15, stp1_16);
+      in17 = _mm_sub_epi16(stp1_14, stp1_17);
+      in18 = _mm_sub_epi16(stp1_13, stp1_18);
+      in19 = _mm_sub_epi16(stp1_12, stp1_19);
+      in20 = _mm_sub_epi16(stp1_11, stp1_20);
+      in21 = _mm_sub_epi16(stp1_10, stp1_21);
+      in22 = _mm_sub_epi16(stp1_9, stp1_22);
+      in23 = _mm_sub_epi16(stp1_8, stp1_23);
+      in24 = _mm_sub_epi16(stp1_7, stp1_24);
+      in25 = _mm_sub_epi16(stp1_6, stp1_25);
+      in26 = _mm_sub_epi16(stp1_5, stp1_26);
+      in27 = _mm_sub_epi16(stp1_4, stp1_27);
+      in28 = _mm_sub_epi16(stp1_3, stp1_28);
+      in29 = _mm_sub_epi16(stp1_2, stp1_29);
+      in30 = _mm_sub_epi16(stp1_1, stp1_30);
+      in31 = _mm_sub_epi16(stp1_0, stp1_31);
+
+      // Final rounding and shift
+      in0 = _mm_adds_epi16(in0, final_rounding);
+      in1 = _mm_adds_epi16(in1, final_rounding);
+      in2 = _mm_adds_epi16(in2, final_rounding);
+      in3 = _mm_adds_epi16(in3, final_rounding);
+      in4 = _mm_adds_epi16(in4, final_rounding);
+      in5 = _mm_adds_epi16(in5, final_rounding);
+      in6 = _mm_adds_epi16(in6, final_rounding);
+      in7 = _mm_adds_epi16(in7, final_rounding);
+      in8 = _mm_adds_epi16(in8, final_rounding);
+      in9 = _mm_adds_epi16(in9, final_rounding);
+      in10 = _mm_adds_epi16(in10, final_rounding);
+      in11 = _mm_adds_epi16(in11, final_rounding);
+      in12 = _mm_adds_epi16(in12, final_rounding);
+      in13 = _mm_adds_epi16(in13, final_rounding);
+      in14 = _mm_adds_epi16(in14, final_rounding);
+      in15 = _mm_adds_epi16(in15, final_rounding);
+      in16 = _mm_adds_epi16(in16, final_rounding);
+      in17 = _mm_adds_epi16(in17, final_rounding);
+      in18 = _mm_adds_epi16(in18, final_rounding);
+      in19 = _mm_adds_epi16(in19, final_rounding);
+      in20 = _mm_adds_epi16(in20, final_rounding);
+      in21 = _mm_adds_epi16(in21, final_rounding);
+      in22 = _mm_adds_epi16(in22, final_rounding);
+      in23 = _mm_adds_epi16(in23, final_rounding);
+      in24 = _mm_adds_epi16(in24, final_rounding);
+      in25 = _mm_adds_epi16(in25, final_rounding);
+      in26 = _mm_adds_epi16(in26, final_rounding);
+      in27 = _mm_adds_epi16(in27, final_rounding);
+      in28 = _mm_adds_epi16(in28, final_rounding);
+      in29 = _mm_adds_epi16(in29, final_rounding);
+      in30 = _mm_adds_epi16(in30, final_rounding);
+      in31 = _mm_adds_epi16(in31, final_rounding);
+
+      in0 = _mm_srai_epi16(in0, 6);
+      in1 = _mm_srai_epi16(in1, 6);
+      in2 = _mm_srai_epi16(in2, 6);
+      in3 = _mm_srai_epi16(in3, 6);
+      in4 = _mm_srai_epi16(in4, 6);
+      in5 = _mm_srai_epi16(in5, 6);
+      in6 = _mm_srai_epi16(in6, 6);
+      in7 = _mm_srai_epi16(in7, 6);
+      in8 = _mm_srai_epi16(in8, 6);
+      in9 = _mm_srai_epi16(in9, 6);
+      in10 = _mm_srai_epi16(in10, 6);
+      in11 = _mm_srai_epi16(in11, 6);
+      in12 = _mm_srai_epi16(in12, 6);
+      in13 = _mm_srai_epi16(in13, 6);
+      in14 = _mm_srai_epi16(in14, 6);
+      in15 = _mm_srai_epi16(in15, 6);
+      in16 = _mm_srai_epi16(in16, 6);
+      in17 = _mm_srai_epi16(in17, 6);
+      in18 = _mm_srai_epi16(in18, 6);
+      in19 = _mm_srai_epi16(in19, 6);
+      in20 = _mm_srai_epi16(in20, 6);
+      in21 = _mm_srai_epi16(in21, 6);
+      in22 = _mm_srai_epi16(in22, 6);
+      in23 = _mm_srai_epi16(in23, 6);
+      in24 = _mm_srai_epi16(in24, 6);
+      in25 = _mm_srai_epi16(in25, 6);
+      in26 = _mm_srai_epi16(in26, 6);
+      in27 = _mm_srai_epi16(in27, 6);
+      in28 = _mm_srai_epi16(in28, 6);
+      in29 = _mm_srai_epi16(in29, 6);
+      in30 = _mm_srai_epi16(in30, 6);
+      in31 = _mm_srai_epi16(in31, 6);
+
+      RECON_AND_STORE(dest, in0);
+      RECON_AND_STORE(dest, in1);
+      RECON_AND_STORE(dest, in2);
+      RECON_AND_STORE(dest, in3);
+      RECON_AND_STORE(dest, in4);
+      RECON_AND_STORE(dest, in5);
+      RECON_AND_STORE(dest, in6);
+      RECON_AND_STORE(dest, in7);
+      RECON_AND_STORE(dest, in8);
+      RECON_AND_STORE(dest, in9);
+      RECON_AND_STORE(dest, in10);
+      RECON_AND_STORE(dest, in11);
+      RECON_AND_STORE(dest, in12);
+      RECON_AND_STORE(dest, in13);
+      RECON_AND_STORE(dest, in14);
+      RECON_AND_STORE(dest, in15);
+      RECON_AND_STORE(dest, in16);
+      RECON_AND_STORE(dest, in17);
+      RECON_AND_STORE(dest, in18);
+      RECON_AND_STORE(dest, in19);
+      RECON_AND_STORE(dest, in20);
+      RECON_AND_STORE(dest, in21);
+      RECON_AND_STORE(dest, in22);
+      RECON_AND_STORE(dest, in23);
+      RECON_AND_STORE(dest, in24);
+      RECON_AND_STORE(dest, in25);
+      RECON_AND_STORE(dest, in26);
+      RECON_AND_STORE(dest, in27);
+      RECON_AND_STORE(dest, in28);
+      RECON_AND_STORE(dest, in29);
+      RECON_AND_STORE(dest, in30);
+      RECON_AND_STORE(dest, in31);
+
+      dest += 8 - (stride * 32);
+    }
+  }
+}
diff --git a/libvpx/vp9/common/x86/vp9_iwalsh_mmx.asm b/libvpx/vp9/common/x86/vp9_iwalsh_mmx.asm
new file mode 100644
index 0000000..1af2521
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_iwalsh_mmx.asm
@@ -0,0 +1,173 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_short_inv_walsh4x4_1_mmx(short *input, short *output)
+global sym(vp9_short_inv_walsh4x4_1_mmx) PRIVATE
+sym(vp9_short_inv_walsh4x4_1_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov     rsi, arg(0)
+    mov     rax, 3
+
+    mov     rdi, arg(1)
+    add     rax, [rsi]          ;input[0] + 3
+
+    movd    mm0, eax
+
+    punpcklwd mm0, mm0          ;x x val val
+
+    punpckldq mm0, mm0          ;val val val val
+
+    psraw   mm0, 3            ;(input[0] + 3) >> 3
+
+    movq  [rdi + 0], mm0
+    movq  [rdi + 8], mm0
+    movq  [rdi + 16], mm0
+    movq  [rdi + 24], mm0
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_short_inv_walsh4x4_mmx(short *input, short *output)
+global sym(vp9_short_inv_walsh4x4_mmx) PRIVATE
+sym(vp9_short_inv_walsh4x4_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov     rax, 3
+    mov     rsi, arg(0)
+    mov     rdi, arg(1)
+    shl     rax, 16
+
+    movq    mm0, [rsi + 0]        ;ip[0]
+    movq    mm1, [rsi + 8]        ;ip[4]
+    or      rax, 3            ;00030003h
+
+    movq    mm2, [rsi + 16]       ;ip[8]
+    movq    mm3, [rsi + 24]       ;ip[12]
+
+    movq    mm7, rax
+    movq    mm4, mm0
+
+    punpcklwd mm7, mm7          ;0003000300030003h
+    movq    mm5, mm1
+
+    paddw   mm4, mm3          ;ip[0] + ip[12] aka al
+    paddw   mm5, mm2          ;ip[4] + ip[8] aka bl
+
+    movq    mm6, mm4          ;temp al
+
+    paddw   mm4, mm5          ;al + bl
+    psubw   mm6, mm5          ;al - bl
+
+    psubw   mm0, mm3          ;ip[0] - ip[12] aka d1
+    psubw   mm1, mm2          ;ip[4] - ip[8] aka c1
+
+    movq    mm5, mm0          ;temp dl
+
+    paddw   mm0, mm1          ;dl + cl
+    psubw   mm5, mm1          ;dl - cl
+
+    ; 03 02 01 00
+    ; 13 12 11 10
+    ; 23 22 21 20
+    ; 33 32 31 30
+
+    movq    mm3, mm4          ; 03 02 01 00
+    punpcklwd mm4, mm0          ; 11 01 10 00
+    punpckhwd mm3, mm0          ; 13 03 12 02
+
+    movq    mm1, mm6          ; 23 22 21 20
+    punpcklwd mm6, mm5          ; 31 21 30 20
+    punpckhwd mm1, mm5          ; 33 23 32 22
+
+    movq    mm0, mm4          ; 11 01 10 00
+    movq    mm2, mm3          ; 13 03 12 02
+
+    punpckldq mm0, mm6          ; 30 20 10 00 aka ip[0]
+    punpckhdq mm4, mm6          ; 31 21 11 01 aka ip[4]
+
+    punpckldq mm2, mm1          ; 32 22 12 02 aka ip[8]
+    punpckhdq mm3, mm1          ; 33 23 13 03 aka ip[12]
+;~~~~~~~~~~~~~~~~~~~~~
+    movq    mm1, mm0
+    movq    mm5, mm4
+
+    paddw   mm1, mm3          ;ip[0] + ip[12] aka al
+    paddw   mm5, mm2          ;ip[4] + ip[8] aka bl
+
+    movq    mm6, mm1          ;temp al
+
+    paddw   mm1, mm5          ;al + bl
+    psubw   mm6, mm5          ;al - bl
+
+    psubw   mm0, mm3          ;ip[0] - ip[12] aka d1
+    psubw   mm4, mm2          ;ip[4] - ip[8] aka c1
+
+    movq    mm5, mm0          ;temp dl
+
+    paddw   mm0, mm4          ;dl + cl
+    psubw   mm5, mm4          ;dl - cl
+;~~~~~~~~~~~~~~~~~~~~~
+    movq    mm3, mm1          ; 03 02 01 00
+    punpcklwd mm1, mm0          ; 11 01 10 00
+    punpckhwd mm3, mm0          ; 13 03 12 02
+
+    movq    mm4, mm6          ; 23 22 21 20
+    punpcklwd mm6, mm5          ; 31 21 30 20
+    punpckhwd mm4, mm5          ; 33 23 32 22
+
+    movq    mm0, mm1          ; 11 01 10 00
+    movq    mm2, mm3          ; 13 03 12 02
+
+    punpckldq mm0, mm6          ; 30 20 10 00 aka ip[0]
+    punpckhdq mm1, mm6          ; 31 21 11 01 aka ip[4]
+
+    punpckldq mm2, mm4          ; 32 22 12 02 aka ip[8]
+    punpckhdq mm3, mm4          ; 33 23 13 03 aka ip[12]
+
+    paddw   mm0, mm7
+    paddw   mm1, mm7
+    paddw   mm2, mm7
+    paddw   mm3, mm7
+
+    psraw   mm0, 3
+    psraw   mm1, 3
+    psraw   mm2, 3
+    psraw   mm3, 3
+
+    movq  [rdi + 0], mm0
+    movq  [rdi + 8], mm1
+    movq  [rdi + 16], mm2
+    movq  [rdi + 24], mm3
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
diff --git a/libvpx/vp9/common/x86/vp9_iwalsh_sse2.asm b/libvpx/vp9/common/x86/vp9_iwalsh_sse2.asm
new file mode 100644
index 0000000..84fa2fe
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_iwalsh_sse2.asm
@@ -0,0 +1,119 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_short_inv_walsh4x4_sse2(short *input, short *output)
+global sym(vp9_short_inv_walsh4x4_sse2) PRIVATE
+sym(vp9_short_inv_walsh4x4_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    SAVE_XMM 6
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov     rsi, arg(0)
+    mov     rdi, arg(1)
+    mov     rax, 3
+
+    movdqa    xmm0, [rsi + 0]       ;ip[4] ip[0]
+    movdqa    xmm1, [rsi + 16]      ;ip[12] ip[8]
+
+    shl     rax, 16
+    or      rax, 3            ;00030003h
+
+    pshufd    xmm2, xmm1, 4eh       ;ip[8] ip[12]
+    movdqa    xmm3, xmm0          ;ip[4] ip[0]
+
+    paddw   xmm0, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
+    psubw   xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
+
+    movdqa    xmm4, xmm0
+    punpcklqdq  xmm0, xmm3          ;d1 a1
+    punpckhqdq  xmm4, xmm3          ;c1 b1
+    movd    xmm6, eax
+
+    movdqa    xmm1, xmm4          ;c1 b1
+    paddw   xmm4, xmm0          ;dl+cl a1+b1 aka op[4] op[0]
+    psubw   xmm0, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
+
+;;;temp output
+;;  movdqu  [rdi + 0], xmm4
+;;  movdqu  [rdi + 16], xmm3
+
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ; 13 12 11 10 03 02 01 00
+    ;
+    ; 33 32 31 30 23 22 21 20
+    ;
+    movdqa    xmm3, xmm4          ; 13 12 11 10 03 02 01 00
+    punpcklwd xmm4, xmm0          ; 23 03 22 02 21 01 20 00
+    punpckhwd xmm3, xmm0          ; 33 13 32 12 31 11 30 10
+    movdqa    xmm1, xmm4          ; 23 03 22 02 21 01 20 00
+    punpcklwd xmm4, xmm3          ; 31 21 11 01 30 20 10 00
+    punpckhwd xmm1, xmm3          ; 33 23 13 03 32 22 12 02
+    ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    pshufd    xmm2, xmm1, 4eh       ;ip[8] ip[12]
+    movdqa    xmm3, xmm4          ;ip[4] ip[0]
+
+    pshufd    xmm6, xmm6, 0       ;03 03 03 03 03 03 03 03
+
+    paddw   xmm4, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
+    psubw   xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
+
+    movdqa    xmm5, xmm4
+    punpcklqdq  xmm4, xmm3          ;d1 a1
+    punpckhqdq  xmm5, xmm3          ;c1 b1
+
+    movdqa    xmm1, xmm5          ;c1 b1
+    paddw   xmm5, xmm4          ;dl+cl a1+b1 aka op[4] op[0]
+    psubw   xmm4, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ; 13 12 11 10 03 02 01 00
+    ;
+    ; 33 32 31 30 23 22 21 20
+    ;
+    movdqa    xmm0, xmm5          ; 13 12 11 10 03 02 01 00
+    punpcklwd xmm5, xmm4          ; 23 03 22 02 21 01 20 00
+    punpckhwd xmm0, xmm4          ; 33 13 32 12 31 11 30 10
+    movdqa    xmm1, xmm5          ; 23 03 22 02 21 01 20 00
+    punpcklwd xmm5, xmm0          ; 31 21 11 01 30 20 10 00
+    punpckhwd xmm1, xmm0          ; 33 23 13 03 32 22 12 02
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    paddw   xmm5, xmm6
+    paddw   xmm1, xmm6
+
+    psraw   xmm5, 3
+    psraw   xmm1, 3
+
+    movdqa  [rdi + 0], xmm5
+    movdqa  [rdi + 16], xmm1
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+x_s1sqr2:
+    times 4 dw 0x8A8C
+align 16
+x_c1sqr2less1:
+    times 4 dw 0x4E7B
+align 16
+fours:
+    times 4 dw 0x0004
diff --git a/libvpx/vp9/common/x86/vp9_loopfilter_intrin_sse2.c b/libvpx/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
new file mode 100644
index 0000000..50f890a
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
@@ -0,0 +1,1013 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  /* SSE2 */
+#include "vp9/common/vp9_loopfilter.h"
+#include "vpx_ports/emmintrin_compat.h"
+
+prototype_loopfilter(vp9_loop_filter_vertical_edge_sse2);
+prototype_loopfilter(vp9_loop_filter_horizontal_edge_sse2);
+
+extern loop_filter_uvfunction vp9_loop_filter_horizontal_edge_uv_sse2;
+extern loop_filter_uvfunction vp9_loop_filter_vertical_edge_uv_sse2;
+
+void vp9_mb_lpf_horizontal_edge_w_sse2(unsigned char *s,
+                                       int p,
+                                       const unsigned char *_blimit,
+                                       const unsigned char *_limit,
+                                       const unsigned char *_thresh) {
+  DECLARE_ALIGNED(16, unsigned char, flat2_op[7][8]);
+  DECLARE_ALIGNED(16, unsigned char, flat2_oq[7][8]);
+
+  DECLARE_ALIGNED(16, unsigned char, flat_op[3][8]);
+  DECLARE_ALIGNED(16, unsigned char, flat_oq[3][8]);
+
+  DECLARE_ALIGNED(16, unsigned char, ap[8][8]);
+  DECLARE_ALIGNED(16, unsigned char, aq[8][8]);
+
+
+  __m128i mask, hev, flat, flat2;
+  const __m128i zero = _mm_set1_epi16(0);
+  const __m128i one = _mm_set1_epi8(1);
+  __m128i p7, p6, p5;
+  __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4;
+  __m128i q5, q6, q7;
+  int i = 0;
+  const unsigned int extended_thresh = _thresh[0] * 0x01010101u;
+  const unsigned int extended_limit  = _limit[0]  * 0x01010101u;
+  const unsigned int extended_blimit = _blimit[0] * 0x01010101u;
+  const __m128i thresh =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_thresh), 0);
+  const __m128i limit =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_limit), 0);
+  const __m128i blimit =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_blimit), 0);
+
+  p4 = _mm_loadl_epi64((__m128i *)(s - 5 * p));
+  p3 = _mm_loadl_epi64((__m128i *)(s - 4 * p));
+  p2 = _mm_loadl_epi64((__m128i *)(s - 3 * p));
+  p1 = _mm_loadl_epi64((__m128i *)(s - 2 * p));
+  p0 = _mm_loadl_epi64((__m128i *)(s - 1 * p));
+  q0 = _mm_loadl_epi64((__m128i *)(s - 0 * p));
+  q1 = _mm_loadl_epi64((__m128i *)(s + 1 * p));
+  q2 = _mm_loadl_epi64((__m128i *)(s + 2 * p));
+  q3 = _mm_loadl_epi64((__m128i *)(s + 3 * p));
+  q4 = _mm_loadl_epi64((__m128i *)(s + 4 * p));
+
+  _mm_storel_epi64((__m128i *)ap[4], p4);
+  _mm_storel_epi64((__m128i *)ap[3], p3);
+  _mm_storel_epi64((__m128i *)ap[2], p2);
+  _mm_storel_epi64((__m128i *)ap[1], p1);
+  _mm_storel_epi64((__m128i *)ap[0], p0);
+  _mm_storel_epi64((__m128i *)aq[4], q4);
+  _mm_storel_epi64((__m128i *)aq[3], q3);
+  _mm_storel_epi64((__m128i *)aq[2], q2);
+  _mm_storel_epi64((__m128i *)aq[1], q1);
+  _mm_storel_epi64((__m128i *)aq[0], q0);
+
+
+  {
+    const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
+                                          _mm_subs_epu8(p0, p1));
+    const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
+                                          _mm_subs_epu8(q0, q1));
+    const __m128i fe = _mm_set1_epi8(0xfe);
+    const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
+    __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
+                                    _mm_subs_epu8(q0, p0));
+    __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
+                                    _mm_subs_epu8(q1, p1));
+    __m128i work;
+    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
+    hev = _mm_subs_epu8(flat, thresh);
+    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
+
+    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
+    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
+    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
+    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
+    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+    mask = _mm_max_epu8(flat, mask);
+    // mask |= (abs(p1 - p0) > limit) * -1;
+    // mask |= (abs(q1 - q0) > limit) * -1;
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
+                                     _mm_subs_epu8(p1, p2)),
+                         _mm_or_si128(_mm_subs_epu8(p3, p2),
+                                      _mm_subs_epu8(p2, p3)));
+    mask = _mm_max_epu8(work, mask);
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1),
+                                     _mm_subs_epu8(q1, q2)),
+                         _mm_or_si128(_mm_subs_epu8(q3, q2),
+                                      _mm_subs_epu8(q2, q3)));
+    mask = _mm_max_epu8(work, mask);
+    mask = _mm_subs_epu8(mask, limit);
+    mask = _mm_cmpeq_epi8(mask, zero);
+  }
+
+  // lp filter
+  {
+    const __m128i t4 = _mm_set1_epi8(4);
+    const __m128i t3 = _mm_set1_epi8(3);
+    const __m128i t80 = _mm_set1_epi8(0x80);
+    const __m128i te0 = _mm_set1_epi8(0xe0);
+    const __m128i t1f = _mm_set1_epi8(0x1f);
+    const __m128i t1 = _mm_set1_epi8(0x1);
+    const __m128i t7f = _mm_set1_epi8(0x7f);
+
+    __m128i ps1 = _mm_xor_si128(p1, t80);
+    __m128i ps0 = _mm_xor_si128(p0, t80);
+    __m128i qs0 = _mm_xor_si128(q0, t80);
+    __m128i qs1 = _mm_xor_si128(q1, t80);
+    __m128i filt;
+    __m128i work_a;
+    __m128i filter1, filter2;
+
+    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
+    work_a = _mm_subs_epi8(qs0, ps0);
+    filt = _mm_adds_epi8(filt, work_a);
+    filt = _mm_adds_epi8(filt, work_a);
+    filt = _mm_adds_epi8(filt, work_a);
+    /* (vp9_filter + 3 * (qs0 - ps0)) & mask */
+    filt = _mm_and_si128(filt, mask);
+
+    filter1 = _mm_adds_epi8(filt, t4);
+    filter2 = _mm_adds_epi8(filt, t3);
+
+    /* Filter1 >> 3 */
+    work_a = _mm_cmpgt_epi8(zero, filter1);
+    filter1 = _mm_srli_epi16(filter1, 3);
+    work_a = _mm_and_si128(work_a, te0);
+    filter1 = _mm_and_si128(filter1, t1f);
+    filter1 = _mm_or_si128(filter1, work_a);
+    qs0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
+
+    /* Filter2 >> 3 */
+    work_a = _mm_cmpgt_epi8(zero, filter2);
+    filter2 = _mm_srli_epi16(filter2, 3);
+    work_a = _mm_and_si128(work_a, te0);
+    filter2 = _mm_and_si128(filter2, t1f);
+    filter2 = _mm_or_si128(filter2, work_a);
+    ps0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
+
+    /* filt >> 1 */
+    filt = _mm_adds_epi8(filter1, t1);
+    work_a = _mm_cmpgt_epi8(zero, filt);
+    filt = _mm_srli_epi16(filt, 1);
+    work_a = _mm_and_si128(work_a, t80);
+    filt = _mm_and_si128(filt, t7f);
+    filt = _mm_or_si128(filt, work_a);
+    filt = _mm_andnot_si128(hev, filt);
+    ps1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
+    qs1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
+    // loopfilter done
+
+    {
+      __m128i work;
+      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p0),
+                                       _mm_subs_epu8(p0, p2)),
+                           _mm_or_si128(_mm_subs_epu8(q2, q0),
+                                        _mm_subs_epu8(q0, q2)));
+      flat = _mm_max_epu8(work, flat);
+      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p3, p0),
+                                       _mm_subs_epu8(p0, p3)),
+                           _mm_or_si128(_mm_subs_epu8(q3, q0),
+                                        _mm_subs_epu8(q0, q3)));
+      flat = _mm_max_epu8(work, flat);
+      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p4, p0),
+                                       _mm_subs_epu8(p0, p4)),
+                           _mm_or_si128(_mm_subs_epu8(q4, q0),
+                                        _mm_subs_epu8(q0, q4)));
+      flat = _mm_subs_epu8(flat, one);
+      flat = _mm_cmpeq_epi8(flat, zero);
+      flat = _mm_and_si128(flat, mask);
+
+      p5 = _mm_loadl_epi64((__m128i *)(s - 6 * p));
+      q5 = _mm_loadl_epi64((__m128i *)(s + 5 * p));
+      flat2 = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p5, p0),
+                                       _mm_subs_epu8(p0, p5)),
+                           _mm_or_si128(_mm_subs_epu8(q5, q0),
+                                        _mm_subs_epu8(q0, q5)));
+      _mm_storel_epi64((__m128i *)ap[5], p5);
+      _mm_storel_epi64((__m128i *)aq[5], q5);
+      flat2 = _mm_max_epu8(work, flat2);
+      p6 = _mm_loadl_epi64((__m128i *)(s - 7 * p));
+      q6 = _mm_loadl_epi64((__m128i *)(s + 6 * p));
+      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p6, p0),
+                                       _mm_subs_epu8(p0, p6)),
+                           _mm_or_si128(_mm_subs_epu8(q6, q0),
+                                        _mm_subs_epu8(q0, q6)));
+      _mm_storel_epi64((__m128i *)ap[6], p6);
+      _mm_storel_epi64((__m128i *)aq[6], q6);
+      flat2 = _mm_max_epu8(work, flat2);
+
+      p7 = _mm_loadl_epi64((__m128i *)(s - 8 * p));
+      q7 = _mm_loadl_epi64((__m128i *)(s + 7 * p));
+      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p7, p0),
+                                       _mm_subs_epu8(p0, p7)),
+                           _mm_or_si128(_mm_subs_epu8(q7, q0),
+                                        _mm_subs_epu8(q0, q7)));
+      _mm_storel_epi64((__m128i *)ap[7], p7);
+      _mm_storel_epi64((__m128i *)aq[7], q7);
+      flat2 = _mm_max_epu8(work, flat2);
+      flat2 = _mm_subs_epu8(flat2, one);
+      flat2 = _mm_cmpeq_epi8(flat2, zero);
+      flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
+    }
+
+    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    // flat and wide flat calculations
+    {
+      const __m128i eight = _mm_set1_epi16(8);
+      const __m128i four = _mm_set1_epi16(4);
+      {
+        __m128i workp_shft;
+        __m128i a, b, c;
+
+        p7 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[7])), zero);
+        p6 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[6])), zero);
+        p5 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[5])), zero);
+        p4 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[4])), zero);
+        p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[3])), zero);
+        p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[2])), zero);
+        p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[1])), zero);
+        p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(ap[0])), zero);
+        q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[0])), zero);
+        q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[1])), zero);
+        q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[2])), zero);
+        q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[3])), zero);
+        q4 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[4])), zero);
+        q5 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[5])), zero);
+        q6 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[6])), zero);
+        q7 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(aq[7])), zero);
+
+        c = _mm_sub_epi16(_mm_slli_epi16(p7, 3), p7);  // p7 * 7
+        c = _mm_add_epi16(_mm_slli_epi16(p6, 1), _mm_add_epi16(p4, c));
+
+        b = _mm_add_epi16(_mm_add_epi16(p3, four), _mm_add_epi16(p3, p2));
+        a = _mm_add_epi16(p3, _mm_add_epi16(p2, p1));
+        a = _mm_add_epi16(_mm_add_epi16(p0, q0), a);
+
+        _mm_storel_epi64((__m128i *)&flat_op[2][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+
+        c = _mm_add_epi16(_mm_add_epi16(p5, eight), c);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[6][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q1, a);
+        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p2)), p1);
+        _mm_storel_epi64((__m128i *)&flat_op[1][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p6)), p5);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[5][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q2, a);
+        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p1)), p0);
+        _mm_storel_epi64((__m128i *)&flat_op[0][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p5)), p4);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[4][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q3, a);
+        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p0)), q0);
+        _mm_storel_epi64((__m128i *)&flat_oq[0][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p4)), p3);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[3][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        b = _mm_add_epi16(q3, b);
+        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p2, q0)), q1);
+        _mm_storel_epi64((__m128i *)&flat_oq[1][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+
+        c = _mm_add_epi16(q4, c);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p3)), p2);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[2][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        b = _mm_add_epi16(q3, b);
+        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p1, q1)), q2);
+        _mm_storel_epi64((__m128i *)&flat_oq[2][i*8],
+                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
+                                          , b));
+        a = _mm_add_epi16(q5, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p2)), p1);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[1][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q6, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p1)), p0);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_op[0][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p0)), q0);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[0][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p6, q0)), q1);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[1][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p5, q1)), q2);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[2][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p4, q2)), q3);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[3][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p3, q3)), q4);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[4][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p2, q4)), q5);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[5][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+
+        a = _mm_add_epi16(q7, a);
+        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p1, q5)), q6);
+        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
+        _mm_storel_epi64((__m128i *)&flat2_oq[6][i*8],
+                         _mm_packus_epi16(workp_shft, workp_shft));
+      }
+    }
+    // wide flat
+    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+    work_a = _mm_loadl_epi64((__m128i *)ap[2]);
+    p2 = _mm_loadl_epi64((__m128i *)flat_op[2]);
+    work_a = _mm_andnot_si128(flat, work_a);
+    p2 = _mm_and_si128(flat, p2);
+    p2 = _mm_or_si128(work_a, p2);
+    _mm_storel_epi64((__m128i *)flat_op[2], p2);
+
+    p1 = _mm_loadl_epi64((__m128i *)flat_op[1]);
+    work_a = _mm_andnot_si128(flat, ps1);
+    p1 = _mm_and_si128(flat, p1);
+    p1 = _mm_or_si128(work_a, p1);
+    _mm_storel_epi64((__m128i *)flat_op[1], p1);
+
+    p0 = _mm_loadl_epi64((__m128i *)flat_op[0]);
+    work_a = _mm_andnot_si128(flat, ps0);
+    p0 = _mm_and_si128(flat, p0);
+    p0 = _mm_or_si128(work_a, p0);
+    _mm_storel_epi64((__m128i *)flat_op[0], p0);
+
+    q0 = _mm_loadl_epi64((__m128i *)flat_oq[0]);
+    work_a = _mm_andnot_si128(flat, qs0);
+    q0 = _mm_and_si128(flat, q0);
+    q0 = _mm_or_si128(work_a, q0);
+    _mm_storel_epi64((__m128i *)flat_oq[0], q0);
+
+    q1 = _mm_loadl_epi64((__m128i *)flat_oq[1]);
+    work_a = _mm_andnot_si128(flat, qs1);
+    q1 = _mm_and_si128(flat, q1);
+    q1 = _mm_or_si128(work_a, q1);
+    _mm_storel_epi64((__m128i *)flat_oq[1], q1);
+
+    work_a = _mm_loadl_epi64((__m128i *)aq[2]);
+    q2 = _mm_loadl_epi64((__m128i *)flat_oq[2]);
+    work_a = _mm_andnot_si128(flat, work_a);
+    q2 = _mm_and_si128(flat, q2);
+    q2 = _mm_or_si128(work_a, q2);
+    _mm_storel_epi64((__m128i *)flat_oq[2], q2);
+
+    // write out op6 - op3
+    {
+      unsigned char *dst = (s - 7 * p);
+      for (i = 6; i > 2; i--) {
+        __m128i flat2_output;
+        work_a = _mm_loadl_epi64((__m128i *)ap[i]);
+        flat2_output = _mm_loadl_epi64((__m128i *)flat2_op[i]);
+        work_a = _mm_andnot_si128(flat2, work_a);
+        flat2_output = _mm_and_si128(flat2, flat2_output);
+        work_a = _mm_or_si128(work_a, flat2_output);
+        _mm_storel_epi64((__m128i *)dst, work_a);
+        dst += p;
+      }
+    }
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_op[2]);
+    p2 = _mm_loadl_epi64((__m128i *)flat2_op[2]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    p2 = _mm_and_si128(flat2, p2);
+    p2 = _mm_or_si128(work_a, p2);
+    _mm_storel_epi64((__m128i *)(s - 3 * p), p2);
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_op[1]);
+    p1 = _mm_loadl_epi64((__m128i *)flat2_op[1]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    p1 = _mm_and_si128(flat2, p1);
+    p1 = _mm_or_si128(work_a, p1);
+    _mm_storel_epi64((__m128i *)(s - 2 * p), p1);
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_op[0]);
+    p0 = _mm_loadl_epi64((__m128i *)flat2_op[0]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    p0 = _mm_and_si128(flat2, p0);
+    p0 = _mm_or_si128(work_a, p0);
+    _mm_storel_epi64((__m128i *)(s - 1 * p), p0);
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_oq[0]);
+    q0 = _mm_loadl_epi64((__m128i *)flat2_oq[0]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    q0 = _mm_and_si128(flat2, q0);
+    q0 = _mm_or_si128(work_a, q0);
+    _mm_storel_epi64((__m128i *)(s - 0 * p), q0);
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_oq[1]);
+    q1 = _mm_loadl_epi64((__m128i *)flat2_oq[1]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    q1 = _mm_and_si128(flat2, q1);
+    q1 = _mm_or_si128(work_a, q1);
+    _mm_storel_epi64((__m128i *)(s + 1 * p), q1);
+
+    work_a = _mm_loadl_epi64((__m128i *)flat_oq[2]);
+    q2 = _mm_loadl_epi64((__m128i *)flat2_oq[2]);
+    work_a = _mm_andnot_si128(flat2, work_a);
+    q2 = _mm_and_si128(flat2, q2);
+    q2 = _mm_or_si128(work_a, q2);
+    _mm_storel_epi64((__m128i *)(s + 2 * p), q2);
+
+    // write out oq3 - oq7
+    {
+      unsigned char *dst = (s + 3 * p);
+      for (i = 3; i < 7; i++) {
+        __m128i flat2_output;
+        work_a = _mm_loadl_epi64((__m128i *)aq[i]);
+        flat2_output = _mm_loadl_epi64((__m128i *)flat2_oq[i]);
+        work_a = _mm_andnot_si128(flat2, work_a);
+        flat2_output = _mm_and_si128(flat2, flat2_output);
+        work_a = _mm_or_si128(work_a, flat2_output);
+        _mm_storel_epi64((__m128i *)dst, work_a);
+        dst += p;
+      }
+    }
+  }
+}
+
+void vp9_mbloop_filter_horizontal_edge_sse2(unsigned char *s,
+                                            int p,
+                                            const unsigned char *_blimit,
+                                            const unsigned char *_limit,
+                                            const unsigned char *_thresh,
+                                            int count) {
+  DECLARE_ALIGNED(16, unsigned char, flat_op2[16]);
+  DECLARE_ALIGNED(16, unsigned char, flat_op1[16]);
+  DECLARE_ALIGNED(16, unsigned char, flat_op0[16]);
+  DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]);
+  DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]);
+  DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]);
+  __m128i mask, hev, flat;
+  const __m128i zero = _mm_set1_epi16(0);
+  __m128i p3, p2, p1, p0, q0, q1, q2, q3;
+  const unsigned int extended_thresh = _thresh[0] * 0x01010101u;
+  const unsigned int extended_limit  = _limit[0]  * 0x01010101u;
+  const unsigned int extended_blimit = _blimit[0] * 0x01010101u;
+  const __m128i thresh =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_thresh), 0);
+  const __m128i limit =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_limit), 0);
+  const __m128i blimit =
+      _mm_shuffle_epi32(_mm_cvtsi32_si128((int)extended_blimit), 0);
+
+  (void)count;
+  p3 = _mm_loadl_epi64((__m128i *)(s - 4 * p));
+  p2 = _mm_loadl_epi64((__m128i *)(s - 3 * p));
+  p1 = _mm_loadl_epi64((__m128i *)(s - 2 * p));
+  p0 = _mm_loadl_epi64((__m128i *)(s - 1 * p));
+  q0 = _mm_loadl_epi64((__m128i *)(s - 0 * p));
+  q1 = _mm_loadl_epi64((__m128i *)(s + 1 * p));
+  q2 = _mm_loadl_epi64((__m128i *)(s + 2 * p));
+  q3 = _mm_loadl_epi64((__m128i *)(s + 3 * p));
+  {
+    const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
+                                          _mm_subs_epu8(p0, p1));
+    const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
+                                          _mm_subs_epu8(q0, q1));
+    const __m128i one = _mm_set1_epi8(1);
+    const __m128i fe = _mm_set1_epi8(0xfe);
+    const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
+    __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
+                                    _mm_subs_epu8(q0, p0));
+    __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
+                                    _mm_subs_epu8(q1, p1));
+    __m128i work;
+    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
+    hev = _mm_subs_epu8(flat, thresh);
+    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
+
+    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
+    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
+    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
+    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
+    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+    mask = _mm_max_epu8(flat, mask);
+    // mask |= (abs(p1 - p0) > limit) * -1;
+    // mask |= (abs(q1 - q0) > limit) * -1;
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
+                                     _mm_subs_epu8(p1, p2)),
+                         _mm_or_si128(_mm_subs_epu8(p3, p2),
+                                      _mm_subs_epu8(p2, p3)));
+    mask = _mm_max_epu8(work, mask);
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1),
+                                     _mm_subs_epu8(q1, q2)),
+                         _mm_or_si128(_mm_subs_epu8(q3, q2),
+                                      _mm_subs_epu8(q2, q3)));
+    mask = _mm_max_epu8(work, mask);
+    mask = _mm_subs_epu8(mask, limit);
+    mask = _mm_cmpeq_epi8(mask, zero);
+
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p0),
+                                     _mm_subs_epu8(p0, p2)),
+                         _mm_or_si128(_mm_subs_epu8(q2, q0),
+                                      _mm_subs_epu8(q0, q2)));
+    flat = _mm_max_epu8(work, flat);
+    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p3, p0),
+                                     _mm_subs_epu8(p0, p3)),
+                         _mm_or_si128(_mm_subs_epu8(q3, q0),
+                                      _mm_subs_epu8(q0, q3)));
+    flat = _mm_max_epu8(work, flat);
+    flat = _mm_subs_epu8(flat, one);
+    flat = _mm_cmpeq_epi8(flat, zero);
+    flat = _mm_and_si128(flat, mask);
+  }
+  {
+    const __m128i four = _mm_set1_epi16(4);
+    unsigned char *src = s;
+    {
+      __m128i workp_a, workp_b, workp_shft;
+      p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 4 * p)), zero);
+      p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 3 * p)), zero);
+      p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 2 * p)), zero);
+      p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 1 * p)), zero);
+      q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 0 * p)), zero);
+      q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 1 * p)), zero);
+      q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 2 * p)), zero);
+      q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 3 * p)), zero);
+
+      workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1));
+      workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0);
+      workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_op2[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+
+      workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_op1[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+
+      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2);
+      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_op0[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+
+      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3);
+      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_oq0[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+
+      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3);
+      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_oq1[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+
+      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3);
+      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2);
+      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+      _mm_storel_epi64((__m128i *)&flat_oq2[0],
+                       _mm_packus_epi16(workp_shft, workp_shft));
+    }
+  }
+  // lp filter
+  {
+    const __m128i t4 = _mm_set1_epi8(4);
+    const __m128i t3 = _mm_set1_epi8(3);
+    const __m128i t80 = _mm_set1_epi8(0x80);
+    const __m128i te0 = _mm_set1_epi8(0xe0);
+    const __m128i t1f = _mm_set1_epi8(0x1f);
+    const __m128i t1 = _mm_set1_epi8(0x1);
+    const __m128i t7f = _mm_set1_epi8(0x7f);
+
+    const __m128i ps1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 2 * p)),
+                                      t80);
+    const __m128i ps0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 1 * p)),
+                                      t80);
+    const __m128i qs0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 0 * p)),
+                                      t80);
+    const __m128i qs1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 1 * p)),
+                                      t80);
+    __m128i filt;
+    __m128i work_a;
+    __m128i filter1, filter2;
+
+    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
+    work_a = _mm_subs_epi8(qs0, ps0);
+    filt = _mm_adds_epi8(filt, work_a);
+    filt = _mm_adds_epi8(filt, work_a);
+    filt = _mm_adds_epi8(filt, work_a);
+    /* (vp9_filter + 3 * (qs0 - ps0)) & mask */
+    filt = _mm_and_si128(filt, mask);
+
+    filter1 = _mm_adds_epi8(filt, t4);
+    filter2 = _mm_adds_epi8(filt, t3);
+
+    /* Filter1 >> 3 */
+    work_a = _mm_cmpgt_epi8(zero, filter1);
+    filter1 = _mm_srli_epi16(filter1, 3);
+    work_a = _mm_and_si128(work_a, te0);
+    filter1 = _mm_and_si128(filter1, t1f);
+    filter1 = _mm_or_si128(filter1, work_a);
+
+    /* Filter2 >> 3 */
+    work_a = _mm_cmpgt_epi8(zero, filter2);
+    filter2 = _mm_srli_epi16(filter2, 3);
+    work_a = _mm_and_si128(work_a, te0);
+    filter2 = _mm_and_si128(filter2, t1f);
+    filter2 = _mm_or_si128(filter2, work_a);
+
+    /* filt >> 1 */
+    filt = _mm_adds_epi8(filter1, t1);
+    work_a = _mm_cmpgt_epi8(zero, filt);
+    filt = _mm_srli_epi16(filt, 1);
+    work_a = _mm_and_si128(work_a, t80);
+    filt = _mm_and_si128(filt, t7f);
+    filt = _mm_or_si128(filt, work_a);
+
+    filt = _mm_andnot_si128(hev, filt);
+
+    work_a = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
+    q0 = _mm_loadl_epi64((__m128i *)flat_oq0);
+    work_a = _mm_andnot_si128(flat, work_a);
+    q0 = _mm_and_si128(flat, q0);
+    q0 = _mm_or_si128(work_a, q0);
+
+    work_a = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
+    q1 = _mm_loadl_epi64((__m128i *)flat_oq1);
+    work_a = _mm_andnot_si128(flat, work_a);
+    q1 = _mm_and_si128(flat, q1);
+    q1 = _mm_or_si128(work_a, q1);
+
+    work_a = _mm_loadu_si128((__m128i *)(s + 2 * p));
+    q2 = _mm_loadl_epi64((__m128i *)flat_oq2);
+    work_a = _mm_andnot_si128(flat, work_a);
+    q2 = _mm_and_si128(flat, q2);
+    q2 = _mm_or_si128(work_a, q2);
+
+    work_a = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
+    p0 = _mm_loadl_epi64((__m128i *)flat_op0);
+    work_a = _mm_andnot_si128(flat, work_a);
+    p0 = _mm_and_si128(flat, p0);
+    p0 = _mm_or_si128(work_a, p0);
+
+    work_a = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
+    p1 = _mm_loadl_epi64((__m128i *)flat_op1);
+    work_a = _mm_andnot_si128(flat, work_a);
+    p1 = _mm_and_si128(flat, p1);
+    p1 = _mm_or_si128(work_a, p1);
+
+    work_a = _mm_loadu_si128((__m128i *)(s - 3 * p));
+    p2 = _mm_loadl_epi64((__m128i *)flat_op2);
+    work_a = _mm_andnot_si128(flat, work_a);
+    p2 = _mm_and_si128(flat, p2);
+    p2 = _mm_or_si128(work_a, p2);
+
+    _mm_storel_epi64((__m128i *)(s - 3 * p), p2);
+    _mm_storel_epi64((__m128i *)(s - 2 * p), p1);
+    _mm_storel_epi64((__m128i *)(s - 1 * p), p0);
+    _mm_storel_epi64((__m128i *)(s + 0 * p), q0);
+    _mm_storel_epi64((__m128i *)(s + 1 * p), q1);
+    _mm_storel_epi64((__m128i *)(s + 2 * p), q2);
+  }
+}
+
+void vp9_mbloop_filter_horizontal_edge_uv_sse2(unsigned char *u,
+                                               int p,
+                                               const unsigned char *_blimit,
+                                               const unsigned char *_limit,
+                                               const unsigned char *_thresh,
+                                               unsigned char *v) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, src, 160);
+
+  /* Read source */
+  const __m128i p4 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u - 5 * p)),
+      _mm_loadl_epi64((__m128i *)(v - 5 * p)));
+  const __m128i p3 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u - 4 * p)),
+      _mm_loadl_epi64((__m128i *)(v - 4 * p)));
+  const __m128i p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u - 3 * p)),
+      _mm_loadl_epi64((__m128i *)(v - 3 * p)));
+  const __m128i p1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u - 2 * p)),
+      _mm_loadl_epi64((__m128i *)(v - 2 * p)));
+  const __m128i p0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u - 1 * p)),
+      _mm_loadl_epi64((__m128i *)(v - 1 * p)));
+  const __m128i q0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u)),
+      _mm_loadl_epi64((__m128i *)(v)));
+  const __m128i q1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u + 1 * p)),
+      _mm_loadl_epi64((__m128i *)(v + 1 * p)));
+  const __m128i q2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u + 2 * p)),
+      _mm_loadl_epi64((__m128i *)(v + 2 * p)));
+  const __m128i q3 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u + 3 * p)),
+      _mm_loadl_epi64((__m128i *)(v + 3 * p)));
+  const __m128i q4 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(u + 4 * p)),
+      _mm_loadl_epi64((__m128i *)(v + 4 * p)));
+
+  _mm_store_si128((__m128i *)(src), p4);
+  _mm_store_si128((__m128i *)(src + 16), p3);
+  _mm_store_si128((__m128i *)(src + 32), p2);
+  _mm_store_si128((__m128i *)(src + 48), p1);
+  _mm_store_si128((__m128i *)(src + 64), p0);
+  _mm_store_si128((__m128i *)(src + 80), q0);
+  _mm_store_si128((__m128i *)(src + 96), q1);
+  _mm_store_si128((__m128i *)(src + 112), q2);
+  _mm_store_si128((__m128i *)(src + 128), q3);
+  _mm_store_si128((__m128i *)(src + 144), q4);
+
+  /* Loop filtering */
+  vp9_mbloop_filter_horizontal_edge_sse2(src + 80, 16, _blimit, _limit,
+                                         _thresh, 1);
+
+  /* Store result */
+  _mm_storel_epi64((__m128i *)(u - 3 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 32)));
+  _mm_storel_epi64((__m128i *)(u - 2 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 48)));
+  _mm_storel_epi64((__m128i *)(u - p),
+                   _mm_loadl_epi64((__m128i *)(src + 64)));
+  _mm_storel_epi64((__m128i *)u,
+                   _mm_loadl_epi64((__m128i *)(src + 80)));
+  _mm_storel_epi64((__m128i *)(u + p),
+                   _mm_loadl_epi64((__m128i *)(src + 96)));
+  _mm_storel_epi64((__m128i *)(u + 2 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 112)));
+
+  _mm_storel_epi64((__m128i *)(v - 3 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 40)));
+  _mm_storel_epi64((__m128i *)(v - 2 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 56)));
+  _mm_storel_epi64((__m128i *)(v - p),
+                   _mm_loadl_epi64((__m128i *)(src + 72)));
+  _mm_storel_epi64((__m128i *)v,
+                   _mm_loadl_epi64((__m128i *)(src + 88)));
+  _mm_storel_epi64((__m128i *)(v + p),
+                   _mm_loadl_epi64((__m128i *)(src + 104)));
+  _mm_storel_epi64((__m128i *)(v + 2 * p),
+                   _mm_loadl_epi64((__m128i *)(src + 120)));
+}
+
+static INLINE void transpose8x16(unsigned char *in0, unsigned char *in1,
+                                 int in_p, unsigned char *out, int out_p) {
+  __m128i x0, x1, x2, x3, x4, x5, x6, x7;
+  __m128i x8, x9, x10, x11, x12, x13, x14, x15;
+
+  /* Read in 16 lines */
+  x0 = _mm_loadl_epi64((__m128i *)in0);
+  x8 = _mm_loadl_epi64((__m128i *)in1);
+  x1 = _mm_loadl_epi64((__m128i *)(in0 + in_p));
+  x9 = _mm_loadl_epi64((__m128i *)(in1 + in_p));
+  x2 = _mm_loadl_epi64((__m128i *)(in0 + 2 * in_p));
+  x10 = _mm_loadl_epi64((__m128i *)(in1 + 2 * in_p));
+  x3 = _mm_loadl_epi64((__m128i *)(in0 + 3*in_p));
+  x11 = _mm_loadl_epi64((__m128i *)(in1 + 3*in_p));
+  x4 = _mm_loadl_epi64((__m128i *)(in0 + 4*in_p));
+  x12 = _mm_loadl_epi64((__m128i *)(in1 + 4*in_p));
+  x5 = _mm_loadl_epi64((__m128i *)(in0 + 5*in_p));
+  x13 = _mm_loadl_epi64((__m128i *)(in1 + 5*in_p));
+  x6 = _mm_loadl_epi64((__m128i *)(in0 + 6*in_p));
+  x14 = _mm_loadl_epi64((__m128i *)(in1 + 6*in_p));
+  x7 = _mm_loadl_epi64((__m128i *)(in0 + 7*in_p));
+  x15 = _mm_loadl_epi64((__m128i *)(in1 + 7*in_p));
+
+  x0 = _mm_unpacklo_epi8(x0, x1);
+  x1 = _mm_unpacklo_epi8(x2, x3);
+  x2 = _mm_unpacklo_epi8(x4, x5);
+  x3 = _mm_unpacklo_epi8(x6, x7);
+
+  x8 = _mm_unpacklo_epi8(x8, x9);
+  x9 = _mm_unpacklo_epi8(x10, x11);
+  x10 = _mm_unpacklo_epi8(x12, x13);
+  x11 = _mm_unpacklo_epi8(x14, x15);
+
+  x4 = _mm_unpacklo_epi16(x0, x1);
+  x5 = _mm_unpacklo_epi16(x2, x3);
+  x12 = _mm_unpacklo_epi16(x8, x9);
+  x13 = _mm_unpacklo_epi16(x10, x11);
+
+  x6 = _mm_unpacklo_epi32(x4, x5);
+  x7 = _mm_unpackhi_epi32(x4, x5);
+  x14 = _mm_unpacklo_epi32(x12, x13);
+  x15 = _mm_unpackhi_epi32(x12, x13);
+
+  /* Store first 4-line result */
+  _mm_storeu_si128((__m128i *)out, _mm_unpacklo_epi64(x6, x14));
+  _mm_storeu_si128((__m128i *)(out + out_p), _mm_unpackhi_epi64(x6, x14));
+  _mm_storeu_si128((__m128i *)(out + 2 * out_p), _mm_unpacklo_epi64(x7, x15));
+  _mm_storeu_si128((__m128i *)(out + 3 * out_p), _mm_unpackhi_epi64(x7, x15));
+
+  x4 = _mm_unpackhi_epi16(x0, x1);
+  x5 = _mm_unpackhi_epi16(x2, x3);
+  x12 = _mm_unpackhi_epi16(x8, x9);
+  x13 = _mm_unpackhi_epi16(x10, x11);
+
+  x6 = _mm_unpacklo_epi32(x4, x5);
+  x7 = _mm_unpackhi_epi32(x4, x5);
+  x14 = _mm_unpacklo_epi32(x12, x13);
+  x15 = _mm_unpackhi_epi32(x12, x13);
+
+  /* Store second 4-line result */
+  _mm_storeu_si128((__m128i *)(out + 4 * out_p), _mm_unpacklo_epi64(x6, x14));
+  _mm_storeu_si128((__m128i *)(out + 5 * out_p), _mm_unpackhi_epi64(x6, x14));
+  _mm_storeu_si128((__m128i *)(out + 6 * out_p), _mm_unpacklo_epi64(x7, x15));
+  _mm_storeu_si128((__m128i *)(out + 7 * out_p), _mm_unpackhi_epi64(x7, x15));
+}
+
+static INLINE void transpose(unsigned char *src[], int in_p,
+                             unsigned char *dst[], int out_p,
+                             int num_8x8_to_transpose) {
+  int idx8x8 = 0;
+  __m128i x0, x1, x2, x3, x4, x5, x6, x7;
+  do {
+    unsigned char *in = src[idx8x8];
+    unsigned char *out = dst[idx8x8];
+
+    x0 = _mm_loadl_epi64((__m128i *)(in + 0*in_p));  // 00 01 02 03 04 05 06 07
+    x1 = _mm_loadl_epi64((__m128i *)(in + 1*in_p));  // 10 11 12 13 14 15 16 17
+    x2 = _mm_loadl_epi64((__m128i *)(in + 2*in_p));  // 20 21 22 23 24 25 26 27
+    x3 = _mm_loadl_epi64((__m128i *)(in + 3*in_p));  // 30 31 32 33 34 35 36 37
+    x4 = _mm_loadl_epi64((__m128i *)(in + 4*in_p));  // 40 41 42 43 44 45 46 47
+    x5 = _mm_loadl_epi64((__m128i *)(in + 5*in_p));  // 50 51 52 53 54 55 56 57
+    x6 = _mm_loadl_epi64((__m128i *)(in + 6*in_p));  // 60 61 62 63 64 65 66 67
+    x7 = _mm_loadl_epi64((__m128i *)(in + 7*in_p));  // 70 71 72 73 74 75 76 77
+    // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+    x0 = _mm_unpacklo_epi8(x0, x1);
+    // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+    x1 = _mm_unpacklo_epi8(x2, x3);
+    // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
+    x2 = _mm_unpacklo_epi8(x4, x5);
+    // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
+    x3 = _mm_unpacklo_epi8(x6, x7);
+    // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+    x4 = _mm_unpacklo_epi16(x0, x1);
+    // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+    x5 = _mm_unpacklo_epi16(x2, x3);
+    // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+    x6 = _mm_unpacklo_epi32(x4, x5);
+    // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+    x7 = _mm_unpackhi_epi32(x4, x5);
+
+    _mm_storel_pd((double *)(out + 0*out_p),
+                  _mm_castsi128_pd(x6));  // 00 10 20 30 40 50 60 70
+    _mm_storeh_pd((double *)(out + 1*out_p),
+                  _mm_castsi128_pd(x6));  // 01 11 21 31 41 51 61 71
+    _mm_storel_pd((double *)(out + 2*out_p),
+                  _mm_castsi128_pd(x7));  // 02 12 22 32 42 52 62 72
+    _mm_storeh_pd((double *)(out + 3*out_p),
+                  _mm_castsi128_pd(x7));  // 03 13 23 33 43 53 63 73
+
+    // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+    x4 = _mm_unpackhi_epi16(x0, x1);
+    // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77
+    x5 = _mm_unpackhi_epi16(x2, x3);
+    // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75
+    x6 = _mm_unpacklo_epi32(x4, x5);
+    // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77
+    x7 = _mm_unpackhi_epi32(x4, x5);
+
+    _mm_storel_pd((double *)(out + 4*out_p),
+                  _mm_castsi128_pd(x6));  // 04 14 24 34 44 54 64 74
+    _mm_storeh_pd((double *)(out + 5*out_p),
+                  _mm_castsi128_pd(x6));  // 05 15 25 35 45 55 65 75
+    _mm_storel_pd((double *)(out + 6*out_p),
+                  _mm_castsi128_pd(x7));  // 06 16 26 36 46 56 66 76
+    _mm_storeh_pd((double *)(out + 7*out_p),
+                  _mm_castsi128_pd(x7));  // 07 17 27 37 47 57 67 77
+  } while (++idx8x8 < num_8x8_to_transpose);
+}
+
+void vp9_mbloop_filter_vertical_edge_sse2(unsigned char *s,
+                                          int p,
+                                          const unsigned char *blimit,
+                                          const unsigned char *limit,
+                                          const unsigned char *thresh,
+                                          int count) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, t_dst, 256);
+  unsigned char *src[2];
+  unsigned char *dst[2];
+
+  (void)count;
+  /* Transpose 16x16 */
+  transpose8x16(s - 8, s - 8 + p * 8, p, t_dst, 16);
+  transpose8x16(s, s + p * 8, p, t_dst + 16 * 8, 16);
+
+  /* Loop filtering */
+  vp9_mbloop_filter_horizontal_edge_sse2(t_dst + 8 * 16, 16, blimit, limit,
+                                           thresh, 1);
+  src[0] = t_dst + 3 * 16;
+  src[1] = t_dst + 3 * 16 + 8;
+
+  dst[0] = s - 5;
+  dst[1] = s - 5 + p * 8;
+
+  /* Transpose 16x8 */
+  transpose(src, 16, dst, p, 2);
+}
+
+void vp9_mb_lpf_vertical_edge_w_sse2(unsigned char *s,
+                                          int p,
+                                          const unsigned char *blimit,
+                                          const unsigned char *limit,
+                                          const unsigned char *thresh) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, t_dst, 256);
+  unsigned char *src[4];
+  unsigned char *dst[4];
+
+  dst[0] = t_dst;
+  dst[1] = t_dst + 8 * 16;
+
+  src[0] = s - 8;
+  src[1] = s - 8 + 8;
+
+  /* Transpose 16x16 */
+  transpose(src, p, dst, 16, 2);
+
+  /* Loop filtering */
+  vp9_mb_lpf_horizontal_edge_w_sse2(t_dst + 8 * 16, 16, blimit, limit,
+                                           thresh);
+
+  src[0] = t_dst;
+  src[1] = t_dst + 8 * 16;
+
+  dst[0] = s - 8;
+  dst[1] = s - 8 + 8;
+
+  transpose(src, 16, dst, p, 2);
+}
+
+
+void vp9_mbloop_filter_vertical_edge_uv_sse2(unsigned char *u,
+                                             int p,
+                                             const unsigned char *blimit,
+                                             const unsigned char *limit,
+                                             const unsigned char *thresh,
+                                             unsigned char *v) {
+  DECLARE_ALIGNED_ARRAY(16, unsigned char, t_dst, 256);
+  unsigned char *src[2];
+  unsigned char *dst[2];
+
+  /* Transpose 16x16 */
+  transpose8x16(u - 8, v - 8, p, t_dst, 16);
+  transpose8x16(u, v, p, t_dst + 16 * 8, 16);
+
+  /* Loop filtering */
+  vp9_mbloop_filter_horizontal_edge_sse2(t_dst + 8 * 16, 16, blimit, limit,
+                                           thresh, 1);
+
+  src[0] = t_dst + 3 * 16;
+  src[1] = t_dst + 3 * 16 + 8;
+
+  dst[0] = u - 5;
+  dst[1] = v - 5;
+
+  /* Transpose 16x8 */
+  transpose(src, 16, dst, p, 2);
+}
diff --git a/libvpx/vp9/common/x86/vp9_loopfilter_mmx.asm b/libvpx/vp9/common/x86/vp9_loopfilter_mmx.asm
new file mode 100644
index 0000000..4ebb51b
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_loopfilter_mmx.asm
@@ -0,0 +1,626 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+
+;void vp9_loop_filter_horizontal_edge_mmx
+;(
+;    unsigned char *src_ptr,
+;    int src_pixel_step,
+;    const char *blimit,
+;    const char *limit,
+;    const char *thresh,
+;    int  count
+;)
+global sym(vp9_loop_filter_horizontal_edge_mmx) PRIVATE
+sym(vp9_loop_filter_horizontal_edge_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 32                         ; reserve 32 bytes
+    %define t0 [rsp + 0]    ;__declspec(align(16)) char t0[8];
+    %define t1 [rsp + 16]   ;__declspec(align(16)) char t1[8];
+
+        mov         rsi, arg(0) ;src_ptr
+        movsxd      rax, dword ptr arg(1) ;src_pixel_step     ; destination pitch?
+
+        movsxd      rcx, dword ptr arg(5) ;count
+.next8_h:
+        mov         rdx, arg(3) ;limit
+        movq        mm7, [rdx]
+        mov         rdi, rsi              ; rdi points to row +1 for indirect addressing
+        add         rdi, rax
+
+        ; calculate breakout conditions
+        movq        mm2, [rdi+2*rax]      ; q3
+        movq        mm1, [rsi+2*rax]      ; q2
+        movq        mm6, mm1              ; q2
+        psubusb     mm1, mm2              ; q2-=q3
+        psubusb     mm2, mm6              ; q3-=q2
+        por         mm1, mm2              ; abs(q3-q2)
+        psubusb     mm1, mm7              ;
+
+
+        movq        mm4, [rsi+rax]        ; q1
+        movq        mm3, mm4              ; q1
+        psubusb     mm4, mm6              ; q1-=q2
+        psubusb     mm6, mm3              ; q2-=q1
+        por         mm4, mm6              ; abs(q2-q1)
+
+        psubusb     mm4, mm7
+        por        mm1, mm4
+
+        movq        mm4, [rsi]            ; q0
+        movq        mm0, mm4              ; q0
+        psubusb     mm4, mm3              ; q0-=q1
+        psubusb     mm3, mm0              ; q1-=q0
+        por         mm4, mm3              ; abs(q0-q1)
+        movq        t0, mm4               ; save to t0
+        psubusb     mm4, mm7
+        por        mm1, mm4
+
+
+        neg         rax                   ; negate pitch to deal with above border
+
+        movq        mm2, [rsi+4*rax]      ; p3
+        movq        mm4, [rdi+4*rax]      ; p2
+        movq        mm5, mm4              ; p2
+        psubusb     mm4, mm2              ; p2-=p3
+        psubusb     mm2, mm5              ; p3-=p2
+        por         mm4, mm2              ; abs(p3 - p2)
+        psubusb     mm4, mm7
+        por        mm1, mm4
+
+
+        movq        mm4, [rsi+2*rax]      ; p1
+        movq        mm3, mm4              ; p1
+        psubusb     mm4, mm5              ; p1-=p2
+        psubusb     mm5, mm3              ; p2-=p1
+        por         mm4, mm5              ; abs(p2 - p1)
+        psubusb     mm4, mm7
+        por        mm1, mm4
+
+        movq        mm2, mm3              ; p1
+
+        movq        mm4, [rsi+rax]        ; p0
+        movq        mm5, mm4              ; p0
+        psubusb     mm4, mm3              ; p0-=p1
+        psubusb     mm3, mm5              ; p1-=p0
+        por         mm4, mm3              ; abs(p1 - p0)
+        movq        t1, mm4               ; save to t1
+        psubusb     mm4, mm7
+        por        mm1, mm4
+
+        movq        mm3, [rdi]            ; q1
+        movq        mm4, mm3              ; q1
+        psubusb     mm3, mm2              ; q1-=p1
+        psubusb     mm2, mm4              ; p1-=q1
+        por         mm2, mm3              ; abs(p1-q1)
+        pand        mm2, [GLOBAL(tfe)]    ; set lsb of each byte to zero
+        psrlw       mm2, 1                ; abs(p1-q1)/2
+
+        movq        mm6, mm5              ; p0
+        movq        mm3, [rsi]            ; q0
+        psubusb     mm5, mm3              ; p0-=q0
+        psubusb     mm3, mm6              ; q0-=p0
+        por         mm5, mm3              ; abs(p0 - q0)
+        paddusb     mm5, mm5              ; abs(p0-q0)*2
+        paddusb     mm5, mm2              ; abs (p0 - q0) *2 + abs(p1-q1)/2
+
+        mov         rdx, arg(2) ;blimit           ; get blimit
+        movq        mm7, [rdx]            ; blimit
+
+        psubusb     mm5,    mm7           ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
+        por         mm1,    mm5
+        pxor        mm5,    mm5
+        pcmpeqb     mm1,    mm5           ; mask mm1
+
+        ; calculate high edge variance
+        mov         rdx, arg(4) ;thresh           ; get thresh
+        movq        mm7, [rdx]            ;
+        movq        mm4, t0               ; get abs (q1 - q0)
+        psubusb     mm4, mm7
+        movq        mm3, t1               ; get abs (p1 - p0)
+        psubusb     mm3, mm7
+        paddb       mm4, mm3              ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
+
+        pcmpeqb     mm4,        mm5
+
+        pcmpeqb     mm5,        mm5
+        pxor        mm4,        mm5
+
+
+        ; start work on filters
+        movq        mm2, [rsi+2*rax]      ; p1
+        movq        mm7, [rdi]            ; q1
+        pxor        mm2, [GLOBAL(t80)]    ; p1 offset to convert to signed values
+        pxor        mm7, [GLOBAL(t80)]    ; q1 offset to convert to signed values
+        psubsb      mm2, mm7              ; p1 - q1
+        pand        mm2, mm4              ; high var mask (hvm)(p1 - q1)
+        pxor        mm6, [GLOBAL(t80)]    ; offset to convert to signed values
+        pxor        mm0, [GLOBAL(t80)]    ; offset to convert to signed values
+        movq        mm3, mm0              ; q0
+        psubsb      mm0, mm6              ; q0 - p0
+        paddsb      mm2, mm0              ; 1 * (q0 - p0) + hvm(p1 - q1)
+        paddsb      mm2, mm0              ; 2 * (q0 - p0) + hvm(p1 - q1)
+        paddsb      mm2, mm0              ; 3 * (q0 - p0) + hvm(p1 - q1)
+        pand        mm1, mm2                  ; mask filter values we don't care about
+        movq        mm2, mm1
+        paddsb      mm1, [GLOBAL(t4)]     ; 3* (q0 - p0) + hvm(p1 - q1) + 4
+        paddsb      mm2, [GLOBAL(t3)]     ; 3* (q0 - p0) + hvm(p1 - q1) + 3
+
+        pxor        mm0, mm0             ;
+        pxor        mm5, mm5
+        punpcklbw   mm0, mm2            ;
+        punpckhbw   mm5, mm2            ;
+        psraw       mm0, 11             ;
+        psraw       mm5, 11
+        packsswb    mm0, mm5
+        movq        mm2, mm0            ;  (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
+
+        pxor        mm0, mm0              ; 0
+        movq        mm5, mm1              ; abcdefgh
+        punpcklbw   mm0, mm1              ; e0f0g0h0
+        psraw       mm0, 11               ; sign extended shift right by 3
+        pxor        mm1, mm1              ; 0
+        punpckhbw   mm1, mm5              ; a0b0c0d0
+        psraw       mm1, 11               ; sign extended shift right by 3
+        movq        mm5, mm0              ; save results
+
+        packsswb    mm0, mm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
+        paddsw      mm5, [GLOBAL(ones)]
+        paddsw      mm1, [GLOBAL(ones)]
+        psraw       mm5, 1                ; partial shifted one more time for 2nd tap
+        psraw       mm1, 1                ; partial shifted one more time for 2nd tap
+        packsswb    mm5, mm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
+        pandn       mm4, mm5              ; high edge variance additive
+
+        paddsb      mm6, mm2              ; p0+= p0 add
+        pxor        mm6, [GLOBAL(t80)]    ; unoffset
+        movq        [rsi+rax], mm6        ; write back
+
+        movq        mm6, [rsi+2*rax]      ; p1
+        pxor        mm6, [GLOBAL(t80)]    ; reoffset
+        paddsb      mm6, mm4              ; p1+= p1 add
+        pxor        mm6, [GLOBAL(t80)]    ; unoffset
+        movq        [rsi+2*rax], mm6      ; write back
+
+        psubsb      mm3, mm0              ; q0-= q0 add
+        pxor        mm3, [GLOBAL(t80)]    ; unoffset
+        movq        [rsi], mm3            ; write back
+
+        psubsb      mm7, mm4              ; q1-= q1 add
+        pxor        mm7, [GLOBAL(t80)]    ; unoffset
+        movq        [rdi], mm7            ; write back
+
+        add         rsi,8
+        neg         rax
+        dec         rcx
+        jnz         .next8_h
+
+    add rsp, 32
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_loop_filter_vertical_edge_mmx
+;(
+;    unsigned char *src_ptr,
+;    int  src_pixel_step,
+;    const char *blimit,
+;    const char *limit,
+;    const char *thresh,
+;    int count
+;)
+global sym(vp9_loop_filter_vertical_edge_mmx) PRIVATE
+sym(vp9_loop_filter_vertical_edge_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub          rsp, 64      ; reserve 64 bytes
+    %define t0   [rsp + 0]    ;__declspec(align(16)) char t0[8];
+    %define t1   [rsp + 16]   ;__declspec(align(16)) char t1[8];
+    %define srct [rsp + 32]   ;__declspec(align(16)) char srct[32];
+
+        mov         rsi,        arg(0) ;src_ptr
+        movsxd      rax,        dword ptr arg(1) ;src_pixel_step     ; destination pitch?
+
+        lea         rsi,        [rsi + rax*4 - 4]
+
+        movsxd      rcx,        dword ptr arg(5) ;count
+.next8_v:
+        mov         rdi,        rsi           ; rdi points to row +1 for indirect addressing
+        add         rdi,        rax
+
+
+        ;transpose
+        movq        mm6,        [rsi+2*rax]                 ; 67 66 65 64 63 62 61 60
+        movq        mm7,        mm6                         ; 77 76 75 74 73 72 71 70
+
+        punpckhbw   mm7,        [rdi+2*rax]                 ; 77 67 76 66 75 65 74 64
+        punpcklbw   mm6,        [rdi+2*rax]                 ; 73 63 72 62 71 61 70 60
+
+        movq        mm4,        [rsi]                       ; 47 46 45 44 43 42 41 40
+        movq        mm5,        mm4                         ; 47 46 45 44 43 42 41 40
+
+        punpckhbw   mm5,        [rsi+rax]                   ; 57 47 56 46 55 45 54 44
+        punpcklbw   mm4,        [rsi+rax]                   ; 53 43 52 42 51 41 50 40
+
+        movq        mm3,        mm5                         ; 57 47 56 46 55 45 54 44
+        punpckhwd   mm5,        mm7                         ; 77 67 57 47 76 66 56 46
+
+        punpcklwd   mm3,        mm7                         ; 75 65 55 45 74 64 54 44
+        movq        mm2,        mm4                         ; 53 43 52 42 51 41 50 40
+
+        punpckhwd   mm4,        mm6                         ; 73 63 53 43 72 62 52 42
+        punpcklwd   mm2,        mm6                         ; 71 61 51 41 70 60 50 40
+
+        neg         rax
+        movq        mm6,        [rsi+rax*2]                 ; 27 26 25 24 23 22 21 20
+
+        movq        mm1,        mm6                         ; 27 26 25 24 23 22 21 20
+        punpckhbw   mm6,        [rsi+rax]                   ; 37 27 36 36 35 25 34 24
+
+        punpcklbw   mm1,        [rsi+rax]                   ; 33 23 32 22 31 21 30 20
+        movq        mm7,        [rsi+rax*4];                ; 07 06 05 04 03 02 01 00
+
+        punpckhbw   mm7,        [rdi+rax*4]                 ; 17 07 16 06 15 05 14 04
+        movq        mm0,        mm7                         ; 17 07 16 06 15 05 14 04
+
+        punpckhwd   mm7,        mm6                         ; 37 27 17 07 36 26 16 06
+        punpcklwd   mm0,        mm6                         ; 35 25 15 05 34 24 14 04
+
+        movq        mm6,        mm7                         ; 37 27 17 07 36 26 16 06
+        punpckhdq   mm7,        mm5                         ; 77 67 57 47 37 27 17 07  = q3
+
+        punpckldq   mm6,        mm5                         ; 76 66 56 46 36 26 16 06  = q2
+
+        movq        mm5,        mm6                         ; 76 66 56 46 36 26 16 06
+        psubusb     mm5,        mm7                         ; q2-q3
+
+        psubusb     mm7,        mm6                         ; q3-q2
+        por         mm7,        mm5;                        ; mm7=abs (q3-q2)
+
+        movq        mm5,        mm0                         ; 35 25 15 05 34 24 14 04
+        punpckhdq   mm5,        mm3                         ; 75 65 55 45 35 25 15 05 = q1
+
+        punpckldq   mm0,        mm3                         ; 74 64 54 44 34 24 15 04 = q0
+        movq        mm3,        mm5                         ; 75 65 55 45 35 25 15 05 = q1
+
+        psubusb     mm3,        mm6                         ; q1-q2
+        psubusb     mm6,        mm5                         ; q2-q1
+
+        por         mm6,        mm3                         ; mm6=abs(q2-q1)
+        lea         rdx,        srct
+
+        movq        [rdx+24],   mm5                         ; save q1
+        movq        [rdx+16],   mm0                         ; save q0
+
+        movq        mm3,        [rsi+rax*4]                 ; 07 06 05 04 03 02 01 00
+        punpcklbw   mm3,        [rdi+rax*4]                 ; 13 03 12 02 11 01 10 00
+
+        movq        mm0,        mm3                         ; 13 03 12 02 11 01 10 00
+        punpcklwd   mm0,        mm1                         ; 31 21 11 01 30 20 10 00
+
+        punpckhwd   mm3,        mm1                         ; 33 23 13 03 32 22 12 02
+        movq        mm1,        mm0                         ; 31 21 11 01 30 20 10 00
+
+        punpckldq   mm0,        mm2                         ; 70 60 50 40 30 20 10 00  =p3
+        punpckhdq   mm1,        mm2                         ; 71 61 51 41 31 21 11 01  =p2
+
+        movq        mm2,        mm1                         ; 71 61 51 41 31 21 11 01  =p2
+        psubusb     mm2,        mm0                         ; p2-p3
+
+        psubusb     mm0,        mm1                         ; p3-p2
+        por         mm0,        mm2                         ; mm0=abs(p3-p2)
+
+        movq        mm2,        mm3                         ; 33 23 13 03 32 22 12 02
+        punpckldq   mm2,        mm4                         ; 72 62 52 42 32 22 12 02 = p1
+
+        punpckhdq   mm3,        mm4                         ; 73 63 53 43 33 23 13 03 = p0
+        movq        [rdx+8],    mm3                         ; save p0
+
+        movq        [rdx],      mm2                         ; save p1
+        movq        mm5,        mm2                         ; mm5 = p1
+
+        psubusb     mm2,        mm1                         ; p1-p2
+        psubusb     mm1,        mm5                         ; p2-p1
+
+        por         mm1,        mm2                         ; mm1=abs(p2-p1)
+        mov         rdx,        arg(3) ;limit
+
+        movq        mm4,        [rdx]                       ; mm4 = limit
+        psubusb     mm7,        mm4
+
+        psubusb     mm0,        mm4
+        psubusb     mm1,        mm4
+
+        psubusb     mm6,        mm4
+        por         mm7,        mm6
+
+        por         mm0,        mm1
+        por         mm0,        mm7                         ;   abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit
+
+        movq        mm1,        mm5                         ; p1
+
+        movq        mm7,        mm3                         ; mm3=mm7=p0
+        psubusb     mm7,        mm5                         ; p0 - p1
+
+        psubusb     mm5,        mm3                         ; p1 - p0
+        por         mm5,        mm7                         ; abs(p1-p0)
+
+        movq        t0,         mm5                         ; save abs(p1-p0)
+        lea         rdx,        srct
+
+        psubusb     mm5,        mm4
+        por         mm0,        mm5                         ; mm0=mask
+
+        movq        mm5,        [rdx+16]                    ; mm5=q0
+        movq        mm7,        [rdx+24]                    ; mm7=q1
+
+        movq        mm6,        mm5                         ; mm6=q0
+        movq        mm2,        mm7                         ; q1
+        psubusb     mm5,        mm7                         ; q0-q1
+
+        psubusb     mm7,        mm6                         ; q1-q0
+        por         mm7,        mm5                         ; abs(q1-q0)
+
+        movq        t1,         mm7                         ; save abs(q1-q0)
+        psubusb     mm7,        mm4
+
+        por         mm0,        mm7                         ; mask
+
+        movq        mm5,        mm2                         ; q1
+        psubusb     mm5,        mm1                         ; q1-=p1
+        psubusb     mm1,        mm2                         ; p1-=q1
+        por         mm5,        mm1                         ; abs(p1-q1)
+        pand        mm5,        [GLOBAL(tfe)]               ; set lsb of each byte to zero
+        psrlw       mm5,        1                           ; abs(p1-q1)/2
+
+        mov         rdx,        arg(2) ;blimit                      ;
+
+        movq        mm4,        [rdx]                       ;blimit
+        movq        mm1,        mm3                         ; mm1=mm3=p0
+
+        movq        mm7,        mm6                         ; mm7=mm6=q0
+        psubusb     mm1,        mm7                         ; p0-q0
+
+        psubusb     mm7,        mm3                         ; q0-p0
+        por         mm1,        mm7                         ; abs(q0-p0)
+        paddusb     mm1,        mm1                         ; abs(q0-p0)*2
+        paddusb     mm1,        mm5                         ; abs (p0 - q0) *2 + abs(p1-q1)/2
+
+        psubusb     mm1,        mm4                         ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
+        por         mm1,        mm0;                        ; mask
+
+        pxor        mm0,        mm0
+        pcmpeqb     mm1,        mm0
+
+        ; calculate high edge variance
+        mov         rdx,        arg(4) ;thresh            ; get thresh
+        movq        mm7,        [rdx]
+        ;
+        movq        mm4,        t0              ; get abs (q1 - q0)
+        psubusb     mm4,        mm7
+
+        movq        mm3,        t1              ; get abs (p1 - p0)
+        psubusb     mm3,        mm7
+
+        por         mm4,        mm3             ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
+        pcmpeqb     mm4,        mm0
+
+        pcmpeqb     mm0,        mm0
+        pxor        mm4,        mm0
+
+
+
+        ; start work on filters
+        lea         rdx,        srct
+
+        movq        mm2,        [rdx]           ; p1
+        movq        mm7,        [rdx+24]        ; q1
+
+        movq        mm6,        [rdx+8]         ; p0
+        movq        mm0,        [rdx+16]        ; q0
+
+        pxor        mm2,        [GLOBAL(t80)]   ; p1 offset to convert to signed values
+        pxor        mm7,        [GLOBAL(t80)]   ; q1 offset to convert to signed values
+
+        psubsb      mm2,        mm7             ; p1 - q1
+        pand        mm2,        mm4             ; high var mask (hvm)(p1 - q1)
+
+        pxor        mm6,        [GLOBAL(t80)]   ; offset to convert to signed values
+        pxor        mm0,        [GLOBAL(t80)]   ; offset to convert to signed values
+
+        movq        mm3,        mm0             ; q0
+        psubsb      mm0,        mm6             ; q0 - p0
+
+        paddsb      mm2,        mm0             ; 1 * (q0 - p0) + hvm(p1 - q1)
+        paddsb      mm2,        mm0             ; 2 * (q0 - p0) + hvm(p1 - q1)
+
+        paddsb      mm2,        mm0             ; 3 * (q0 - p0) + hvm(p1 - q1)
+        pand       mm1,        mm2              ; mask filter values we don't care about
+
+        movq        mm2,        mm1
+        paddsb      mm1,        [GLOBAL(t4)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 4
+
+        paddsb      mm2,        [GLOBAL(t3)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 3
+        pxor        mm0,        mm0          ;
+
+        pxor        mm5,        mm5
+        punpcklbw   mm0,        mm2         ;
+
+        punpckhbw   mm5,        mm2         ;
+        psraw       mm0,        11              ;
+
+        psraw       mm5,        11
+        packsswb    mm0,        mm5
+
+        movq        mm2,        mm0         ;  (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
+
+        pxor        mm0,        mm0           ; 0
+        movq        mm5,        mm1           ; abcdefgh
+
+        punpcklbw   mm0,        mm1           ; e0f0g0h0
+        psraw       mm0,        11                ; sign extended shift right by 3
+
+        pxor        mm1,        mm1           ; 0
+        punpckhbw   mm1,        mm5           ; a0b0c0d0
+
+        psraw       mm1,        11                ; sign extended shift right by 3
+        movq        mm5,        mm0              ; save results
+
+        packsswb    mm0,        mm1           ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
+        paddsw      mm5,        [GLOBAL(ones)]
+
+        paddsw      mm1,        [GLOBAL(ones)]
+        psraw       mm5,        1                 ; partial shifted one more time for 2nd tap
+
+        psraw       mm1,        1                 ; partial shifted one more time for 2nd tap
+        packsswb    mm5,        mm1           ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
+
+        pandn       mm4,        mm5             ; high edge variance additive
+
+        paddsb      mm6,        mm2             ; p0+= p0 add
+        pxor        mm6,        [GLOBAL(t80)]   ; unoffset
+
+        ; mm6=p0                               ;
+        movq        mm1,        [rdx]           ; p1
+        pxor        mm1,        [GLOBAL(t80)]   ; reoffset
+
+        paddsb      mm1,        mm4                 ; p1+= p1 add
+        pxor        mm1,        [GLOBAL(t80)]       ; unoffset
+        ; mm6 = p0 mm1 = p1
+
+        psubsb      mm3,        mm0                 ; q0-= q0 add
+        pxor        mm3,        [GLOBAL(t80)]       ; unoffset
+
+        ; mm3 = q0
+        psubsb      mm7,        mm4                 ; q1-= q1 add
+        pxor        mm7,        [GLOBAL(t80)]       ; unoffset
+        ; mm7 = q1
+
+        ; tranpose and write back
+        ; mm1 =    72 62 52 42 32 22 12 02
+        ; mm6 =    73 63 53 43 33 23 13 03
+        ; mm3 =    74 64 54 44 34 24 14 04
+        ; mm7 =    75 65 55 45 35 25 15 05
+
+        movq        mm2,        mm1             ; 72 62 52 42 32 22 12 02
+        punpcklbw   mm2,        mm6             ; 33 32 23 22 13 12 03 02
+
+        movq        mm4,        mm3             ; 74 64 54 44 34 24 14 04
+        punpckhbw   mm1,        mm6             ; 73 72 63 62 53 52 43 42
+
+        punpcklbw   mm4,        mm7             ; 35 34 25 24 15 14 05 04
+        punpckhbw   mm3,        mm7             ; 75 74 65 64 55 54 45 44
+
+        movq        mm6,        mm2             ; 33 32 23 22 13 12 03 02
+        punpcklwd   mm2,        mm4             ; 15 14 13 12 05 04 03 02
+
+        punpckhwd   mm6,        mm4             ; 35 34 33 32 25 24 23 22
+        movq        mm5,        mm1             ; 73 72 63 62 53 52 43 42
+
+        punpcklwd   mm1,        mm3             ; 55 54 53 52 45 44 43 42
+        punpckhwd   mm5,        mm3             ; 75 74 73 72 65 64 63 62
+
+
+        ; mm2 = 15 14 13 12 05 04 03 02
+        ; mm6 = 35 34 33 32 25 24 23 22
+        ; mm5 = 55 54 53 52 45 44 43 42
+        ; mm1 = 75 74 73 72 65 64 63 62
+
+
+
+        movd        [rsi+rax*4+2], mm2
+        psrlq       mm2,        32
+
+        movd        [rdi+rax*4+2], mm2
+        movd        [rsi+rax*2+2], mm6
+
+        psrlq       mm6,        32
+        movd        [rsi+rax+2],mm6
+
+        movd        [rsi+2],    mm1
+        psrlq       mm1,        32
+
+        movd        [rdi+2],    mm1
+        neg         rax
+
+        movd        [rdi+rax+2],mm5
+        psrlq       mm5,        32
+
+        movd        [rdi+rax*2+2], mm5
+
+        lea         rsi,        [rsi+rax*8]
+        dec         rcx
+        jnz         .next8_v
+
+    add rsp, 64
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+tfe:
+    times 8 db 0xfe
+align 16
+t80:
+    times 8 db 0x80
+align 16
+t1s:
+    times 8 db 0x01
+align 16
+t3:
+    times 8 db 0x03
+align 16
+t4:
+    times 8 db 0x04
+align 16
+ones:
+    times 4 dw 0x0001
+align 16
+s27:
+    times 4 dw 0x1b00
+align 16
+s18:
+    times 4 dw 0x1200
+align 16
+s9:
+    times 4 dw 0x0900
+align 16
+s63:
+    times 4 dw 0x003f
diff --git a/libvpx/vp9/common/x86/vp9_loopfilter_sse2.asm b/libvpx/vp9/common/x86/vp9_loopfilter_sse2.asm
new file mode 100644
index 0000000..74236cf
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_loopfilter_sse2.asm
@@ -0,0 +1,872 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+; Use of pmaxub instead of psubusb to compute filter mask was seen
+; in ffvp8
+
+%macro LFH_FILTER_AND_HEV_MASK 1
+%if %1
+        movdqa      xmm2,                   [rdi+2*rax]       ; q3
+        movdqa      xmm1,                   [rsi+2*rax]       ; q2
+        movdqa      xmm4,                   [rsi+rax]         ; q1
+        movdqa      xmm5,                   [rsi]             ; q0
+        neg         rax                     ; negate pitch to deal with above border
+%else
+        movlps      xmm2,                   [rsi + rcx*2]     ; q3
+        movlps      xmm1,                   [rsi + rcx]       ; q2
+        movlps      xmm4,                   [rsi]             ; q1
+        movlps      xmm5,                   [rsi + rax]       ; q0
+
+        movhps      xmm2,                   [rdi + rcx*2]
+        movhps      xmm1,                   [rdi + rcx]
+        movhps      xmm4,                   [rdi]
+        movhps      xmm5,                   [rdi + rax]
+
+        lea         rsi,                    [rsi + rax*4]
+        lea         rdi,                    [rdi + rax*4]
+
+        movdqa      XMMWORD PTR [rsp],      xmm1              ; store q2
+        movdqa      XMMWORD PTR [rsp + 16], xmm4              ; store q1
+%endif
+
+        movdqa      xmm6,                   xmm1              ; q2
+        movdqa      xmm3,                   xmm4              ; q1
+
+        psubusb     xmm1,                   xmm2              ; q2-=q3
+        psubusb     xmm2,                   xmm6              ; q3-=q2
+
+        psubusb     xmm4,                   xmm6              ; q1-=q2
+        psubusb     xmm6,                   xmm3              ; q2-=q1
+
+        por         xmm4,                   xmm6              ; abs(q2-q1)
+        por         xmm1,                   xmm2              ; abs(q3-q2)
+
+        movdqa      xmm0,                   xmm5              ; q0
+        pmaxub      xmm1,                   xmm4
+
+        psubusb     xmm5,                   xmm3              ; q0-=q1
+        psubusb     xmm3,                   xmm0              ; q1-=q0
+
+        por         xmm5,                   xmm3              ; abs(q0-q1)
+        movdqa      t0,                     xmm5              ; save to t0
+
+        pmaxub      xmm1,                   xmm5
+
+%if %1
+        movdqa      xmm2,                   [rsi+4*rax]       ; p3
+        movdqa      xmm4,                   [rdi+4*rax]       ; p2
+        movdqa      xmm6,                   [rsi+2*rax]       ; p1
+%else
+        movlps      xmm2,                   [rsi + rax]       ; p3
+        movlps      xmm4,                   [rsi]             ; p2
+        movlps      xmm6,                   [rsi + rcx]       ; p1
+
+        movhps      xmm2,                   [rdi + rax]
+        movhps      xmm4,                   [rdi]
+        movhps      xmm6,                   [rdi + rcx]
+
+        movdqa      XMMWORD PTR [rsp + 32], xmm4              ; store p2
+        movdqa      XMMWORD PTR [rsp + 48], xmm6              ; store p1
+%endif
+
+        movdqa      xmm5,                   xmm4              ; p2
+        movdqa      xmm3,                   xmm6              ; p1
+
+        psubusb     xmm4,                   xmm2              ; p2-=p3
+        psubusb     xmm2,                   xmm5              ; p3-=p2
+
+        psubusb     xmm3,                   xmm5              ; p1-=p2
+        pmaxub      xmm1,                   xmm4              ; abs(p3 - p2)
+
+        psubusb     xmm5,                   xmm6              ; p2-=p1
+        pmaxub      xmm1,                   xmm2              ; abs(p3 - p2)
+
+        pmaxub      xmm1,                   xmm5              ; abs(p2 - p1)
+        movdqa      xmm2,                   xmm6              ; p1
+
+        pmaxub      xmm1,                   xmm3              ; abs(p2 - p1)
+%if %1
+        movdqa      xmm4,                   [rsi+rax]         ; p0
+        movdqa      xmm3,                   [rdi]             ; q1
+%else
+        movlps      xmm4,                   [rsi + rcx*2]     ; p0
+        movhps      xmm4,                   [rdi + rcx*2]
+        movdqa      xmm3,                   q1                ; q1
+%endif
+
+        movdqa      xmm5,                   xmm4              ; p0
+        psubusb     xmm4,                   xmm6              ; p0-=p1
+
+        psubusb     xmm6,                   xmm5              ; p1-=p0
+
+        por         xmm6,                   xmm4              ; abs(p1 - p0)
+        mov         rdx,                    arg(2)            ; get blimit
+
+        movdqa        t1,                   xmm6              ; save to t1
+
+        movdqa      xmm4,                   xmm3              ; q1
+        pmaxub      xmm1,                   xmm6
+
+        psubusb     xmm3,                   xmm2              ; q1-=p1
+        psubusb     xmm2,                   xmm4              ; p1-=q1
+
+        psubusb     xmm1,                   xmm7
+        por         xmm2,                   xmm3              ; abs(p1-q1)
+
+        movdqa      xmm7,                   XMMWORD PTR [rdx] ; blimit
+
+        movdqa      xmm3,                   xmm0              ; q0
+        pand        xmm2,                   [GLOBAL(tfe)]     ; set lsb of each byte to zero
+
+        mov         rdx,                    arg(4)            ; hev get thresh
+
+        movdqa      xmm6,                   xmm5              ; p0
+        psrlw       xmm2,                   1                 ; abs(p1-q1)/2
+
+        psubusb     xmm5,                   xmm3              ; p0-=q0
+
+        psubusb     xmm3,                   xmm6              ; q0-=p0
+        por         xmm5,                   xmm3              ; abs(p0 - q0)
+
+        paddusb     xmm5,                   xmm5              ; abs(p0-q0)*2
+
+        movdqa      xmm4,                   t0                ; hev get abs (q1 - q0)
+
+        movdqa      xmm3,                   t1                ; get abs (p1 - p0)
+
+        paddusb     xmm5,                   xmm2              ; abs (p0 - q0) *2 + abs(p1-q1)/2
+
+        movdqa      xmm2,                   XMMWORD PTR [rdx] ; hev
+
+        psubusb     xmm5,                   xmm7              ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
+        psubusb     xmm4,                   xmm2              ; hev
+
+        psubusb     xmm3,                   xmm2              ; hev
+        por         xmm1,                   xmm5
+
+        pxor        xmm7,                   xmm7
+        paddb       xmm4,                   xmm3              ; hev abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
+
+        pcmpeqb     xmm4,                   xmm5              ; hev
+        pcmpeqb     xmm3,                   xmm3              ; hev
+
+        pcmpeqb     xmm1,                   xmm7              ; mask xmm1
+        pxor        xmm4,                   xmm3              ; hev
+%endmacro
+
+%macro B_FILTER 1
+%if %1 == 0
+        movdqa      xmm2,                   p1                ; p1
+        movdqa      xmm7,                   q1                ; q1
+%elif %1 == 1
+        movdqa      xmm2,                   [rsi+2*rax]       ; p1
+        movdqa      xmm7,                   [rdi]             ; q1
+%elif %1 == 2
+        lea         rdx,                    srct
+
+        movdqa      xmm2,                   [rdx]             ; p1
+        movdqa      xmm7,                   [rdx+48]          ; q1
+        movdqa      xmm6,                   [rdx+16]          ; p0
+        movdqa      xmm0,                   [rdx+32]          ; q0
+%endif
+
+        pxor        xmm2,                   [GLOBAL(t80)]     ; p1 offset to convert to signed values
+        pxor        xmm7,                   [GLOBAL(t80)]     ; q1 offset to convert to signed values
+
+        psubsb      xmm2,                   xmm7              ; p1 - q1
+        pxor        xmm6,                   [GLOBAL(t80)]     ; offset to convert to signed values
+
+        pand        xmm2,                   xmm4              ; high var mask (hvm)(p1 - q1)
+        pxor        xmm0,                   [GLOBAL(t80)]     ; offset to convert to signed values
+
+        movdqa      xmm3,                   xmm0              ; q0
+        psubsb      xmm0,                   xmm6              ; q0 - p0
+
+        paddsb      xmm2,                   xmm0              ; 1 * (q0 - p0) + hvm(p1 - q1)
+
+        paddsb      xmm2,                   xmm0              ; 2 * (q0 - p0) + hvm(p1 - q1)
+
+        paddsb      xmm2,                   xmm0              ; 3 * (q0 - p0) + hvm(p1 - q1)
+
+        pand        xmm1,                   xmm2              ; mask filter values we don't care about
+
+        movdqa      xmm2,                   xmm1
+
+        paddsb      xmm1,                   [GLOBAL(t4)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 4
+        paddsb      xmm2,                   [GLOBAL(t3)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 3
+
+        punpckhbw   xmm5,                   xmm2              ; axbxcxdx
+        punpcklbw   xmm2,                   xmm2              ; exfxgxhx
+
+        punpcklbw   xmm0,                   xmm1              ; exfxgxhx
+        psraw       xmm5,                   11                ; sign extended shift right by 3
+
+        punpckhbw   xmm1,                   xmm1              ; axbxcxdx
+        psraw       xmm2,                   11                ; sign extended shift right by 3
+
+        packsswb    xmm2,                   xmm5              ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
+        psraw       xmm0,                   11                ; sign extended shift right by 3
+
+        psraw       xmm1,                   11                ; sign extended shift right by 3
+        movdqa      xmm5,                   xmm0              ; save results
+
+        packsswb    xmm0,                   xmm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
+        paddsw      xmm5,                   [GLOBAL(ones)]
+
+        paddsw      xmm1,                   [GLOBAL(ones)]
+        psraw       xmm5,                   1                 ; partial shifted one more time for 2nd tap
+
+        psraw       xmm1,                   1                 ; partial shifted one more time for 2nd tap
+
+        paddsb      xmm6,                   xmm2              ; p0+= p0 add
+        packsswb    xmm5,                   xmm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
+
+%if %1 == 0
+        movdqa      xmm1,                   p1                ; p1
+%elif %1 == 1
+        movdqa      xmm1,                   [rsi+2*rax]       ; p1
+%elif %1 == 2
+        movdqa      xmm1,                   [rdx]             ; p1
+%endif
+        pandn       xmm4,                   xmm5              ; high edge variance additive
+        pxor        xmm6,                   [GLOBAL(t80)]     ; unoffset
+
+        pxor        xmm1,                   [GLOBAL(t80)]     ; reoffset
+        psubsb      xmm3,                   xmm0              ; q0-= q0 add
+
+        paddsb      xmm1,                   xmm4              ; p1+= p1 add
+        pxor        xmm3,                   [GLOBAL(t80)]     ; unoffset
+
+        pxor        xmm1,                   [GLOBAL(t80)]     ; unoffset
+        psubsb      xmm7,                   xmm4              ; q1-= q1 add
+
+        pxor        xmm7,                   [GLOBAL(t80)]     ; unoffset
+%if %1 == 0
+        lea         rsi,                    [rsi + rcx*2]
+        lea         rdi,                    [rdi + rcx*2]
+        movq        MMWORD PTR [rsi],       xmm6              ; p0
+        movhps      MMWORD PTR [rdi],       xmm6
+        movq        MMWORD PTR [rsi + rax], xmm1              ; p1
+        movhps      MMWORD PTR [rdi + rax], xmm1
+        movq        MMWORD PTR [rsi + rcx], xmm3              ; q0
+        movhps      MMWORD PTR [rdi + rcx], xmm3
+        movq        MMWORD PTR [rsi + rcx*2],xmm7             ; q1
+        movhps      MMWORD PTR [rdi + rcx*2],xmm7
+%elif %1 == 1
+        movdqa      [rsi+rax],              xmm6              ; write back
+        movdqa      [rsi+2*rax],            xmm1              ; write back
+        movdqa      [rsi],                  xmm3              ; write back
+        movdqa      [rdi],                  xmm7              ; write back
+%endif
+
+%endmacro
+
+
+;void vp9_loop_filter_horizontal_edge_sse2
+;(
+;    unsigned char *src_ptr,
+;    int            src_pixel_step,
+;    const char    *blimit,
+;    const char    *limit,
+;    const char    *thresh,
+;    int            count
+;)
+global sym(vp9_loop_filter_horizontal_edge_sse2) PRIVATE
+sym(vp9_loop_filter_horizontal_edge_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 32     ; reserve 32 bytes
+    %define t0 [rsp + 0]    ;__declspec(align(16)) char t0[16];
+    %define t1 [rsp + 16]   ;__declspec(align(16)) char t1[16];
+
+        mov         rsi,                    arg(0)           ;src_ptr
+        movsxd      rax,                    dword ptr arg(1) ;src_pixel_step
+
+        mov         rdx,                    arg(3)           ;limit
+        movdqa      xmm7,                   XMMWORD PTR [rdx]
+
+        lea         rdi,                    [rsi+rax]        ; rdi points to row +1 for indirect addressing
+
+        ; calculate breakout conditions and high edge variance
+        LFH_FILTER_AND_HEV_MASK 1
+        ; filter and write back the result
+        B_FILTER 1
+
+    add rsp, 32
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_loop_filter_horizontal_edge_uv_sse2
+;(
+;    unsigned char *src_ptr,
+;    int            src_pixel_step,
+;    const char    *blimit,
+;    const char    *limit,
+;    const char    *thresh,
+;    int            count
+;)
+global sym(vp9_loop_filter_horizontal_edge_uv_sse2) PRIVATE
+sym(vp9_loop_filter_horizontal_edge_uv_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 96       ; reserve 96 bytes
+    %define q2  [rsp + 0]     ;__declspec(align(16)) char q2[16];
+    %define q1  [rsp + 16]    ;__declspec(align(16)) char q1[16];
+    %define p2  [rsp + 32]    ;__declspec(align(16)) char p2[16];
+    %define p1  [rsp + 48]    ;__declspec(align(16)) char p1[16];
+    %define t0  [rsp + 64]    ;__declspec(align(16)) char t0[16];
+    %define t1  [rsp + 80]    ;__declspec(align(16)) char t1[16];
+
+        mov         rsi,                    arg(0)             ; u
+        mov         rdi,                    arg(5)             ; v
+        movsxd      rax,                    dword ptr arg(1)   ; src_pixel_step
+        mov         rcx,                    rax
+        neg         rax                     ; negate pitch to deal with above border
+
+        mov         rdx,                    arg(3)             ;limit
+        movdqa      xmm7,                   XMMWORD PTR [rdx]
+
+        lea         rsi,                    [rsi + rcx]
+        lea         rdi,                    [rdi + rcx]
+
+        ; calculate breakout conditions and high edge variance
+        LFH_FILTER_AND_HEV_MASK 0
+        ; filter and write back the result
+        B_FILTER 0
+
+    add rsp, 96
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+%macro TRANSPOSE_16X8 2
+        movq        xmm4,               QWORD PTR [rsi]        ; xx xx xx xx xx xx xx xx 07 06 05 04 03 02 01 00
+        movq        xmm1,               QWORD PTR [rdi]        ; xx xx xx xx xx xx xx xx 17 16 15 14 13 12 11 10
+        movq        xmm0,               QWORD PTR [rsi+2*rax]  ; xx xx xx xx xx xx xx xx 27 26 25 24 23 22 21 20
+        movq        xmm7,               QWORD PTR [rdi+2*rax]  ; xx xx xx xx xx xx xx xx 37 36 35 34 33 32 31 30
+        movq        xmm5,               QWORD PTR [rsi+4*rax]  ; xx xx xx xx xx xx xx xx 47 46 45 44 43 42 41 40
+        movq        xmm2,               QWORD PTR [rdi+4*rax]  ; xx xx xx xx xx xx xx xx 57 56 55 54 53 52 51 50
+
+        punpcklbw   xmm4,               xmm1            ; 17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00
+
+        movq        xmm1,               QWORD PTR [rdi+2*rcx]  ; xx xx xx xx xx xx xx xx 77 76 75 74 73 72 71 70
+
+        movdqa      xmm3,               xmm4            ; 17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00
+        punpcklbw   xmm0,               xmm7            ; 37 27 36 36 35 25 34 24 33 23 32 22 31 21 30 20
+
+        movq        xmm7,               QWORD PTR [rsi+2*rcx]  ; xx xx xx xx xx xx xx xx 67 66 65 64 63 62 61 60
+
+        punpcklbw   xmm5,               xmm2            ; 57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40
+%if %1
+        lea         rsi,                [rsi+rax*8]
+%else
+        mov         rsi,                arg(5)          ; v_ptr
+%endif
+
+        movdqa      xmm6,               xmm5            ; 57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40
+        punpcklbw   xmm7,               xmm1            ; 77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60
+
+        punpcklwd   xmm5,               xmm7            ; 73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40
+
+        punpckhwd   xmm6,               xmm7            ; 77 67 57 47 76 66 56 46 75 65 55 45 74 64 54 44
+%if %1
+        lea         rdi,                [rdi+rax*8]
+%else
+        lea         rsi,                [rsi - 4]
+%endif
+
+        punpcklwd   xmm3,               xmm0            ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00
+%if %1
+        lea         rdx,                srct
+%else
+        lea         rdi,                [rsi + rax]     ; rdi points to row +1 for indirect addressing
+%endif
+
+        movdqa      xmm2,               xmm3            ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00
+        punpckhwd   xmm4,               xmm0            ; 37 27 17 07 36 26 16 06 35 25 15 05 34 24 14 04
+
+        movdqa      xmm7,               xmm4            ; 37 27 17 07 36 26 16 06 35 25 15 05 34 24 14 04
+        punpckhdq   xmm3,               xmm5            ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
+
+        punpckhdq   xmm7,               xmm6            ; 77 67 57 47 37 27 17 07 76 66 56 46 36 26 16 06
+
+        punpckldq   xmm4,               xmm6            ; 75 65 55 45 35 25 15 05 74 64 54 44 34 24 14 04
+
+        punpckldq   xmm2,               xmm5            ; 71 61 51 41 31 21 11 01 70 60 50 40 30 20 10 00
+
+        movdqa      t0,                 xmm2            ; save to free XMM2
+        movq        xmm2,               QWORD PTR [rsi]       ; xx xx xx xx xx xx xx xx 87 86 85 84 83 82 81 80
+        movq        xmm6,               QWORD PTR [rdi]       ; xx xx xx xx xx xx xx xx 97 96 95 94 93 92 91 90
+        movq        xmm0,               QWORD PTR [rsi+2*rax] ; xx xx xx xx xx xx xx xx a7 a6 a5 a4 a3 a2 a1 a0
+        movq        xmm5,               QWORD PTR [rdi+2*rax] ; xx xx xx xx xx xx xx xx b7 b6 b5 b4 b3 b2 b1 b0
+        movq        xmm1,               QWORD PTR [rsi+4*rax] ; xx xx xx xx xx xx xx xx c7 c6 c5 c4 c3 c2 c1 c0
+
+        punpcklbw   xmm2,               xmm6            ; 97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80
+
+        movq        xmm6,               QWORD PTR [rdi+4*rax] ; xx xx xx xx xx xx xx xx d7 d6 d5 d4 d3 d2 d1 d0
+
+        punpcklbw   xmm0,               xmm5                  ; b7 a7 b6 a6 b5 a5 b4 a4 b3 a3 b2 a2 b1 a1 b0 a0
+
+        movq        xmm5,               QWORD PTR [rsi+2*rcx] ; xx xx xx xx xx xx xx xx e7 e6 e5 e4 e3 e2 e1 e0
+
+        punpcklbw   xmm1,               xmm6            ; d7 c7 d6 c6 d5 c5 d4 c4 d3 c3 d2 c2 d1 e1 d0 c0
+
+        movq        xmm6,               QWORD PTR [rdi+2*rcx] ; xx xx xx xx xx xx xx xx f7 f6 f5 f4 f3 f2 f1 f0
+
+        punpcklbw   xmm5,               xmm6            ; f7 e7 f6 e6 f5 e5 f4 e4 f3 e3 f2 e2 f1 e1 f0 e0
+
+        movdqa      xmm6,               xmm1            ;
+        punpckhwd   xmm6,               xmm5            ; f7 e7 d7 c7 f6 e6 d6 c6 f5 e5 d5 c5 f4 e4 d4 c4
+
+        punpcklwd   xmm1,               xmm5            ; f3 e3 d3 c3 f2 e2 d2 c2 f1 e1 d1 c1 f0 e0 d0 c0
+        movdqa      xmm5,               xmm2            ; 97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80
+
+        punpcklwd   xmm5,               xmm0            ; b3 a3 93 83 b2 a2 92 82 b1 a1 91 81 b0 a0 90 80
+
+        punpckhwd   xmm2,               xmm0            ; b7 a7 97 87 b6 a6 96 86 b5 a5 95 85 b4 a4 94 84
+
+        movdqa      xmm0,               xmm5
+        punpckldq   xmm0,               xmm1            ; f1 e1 d1 c1 b1 a1 91 81 f0 e0 d0 c0 b0 a0 90 80
+
+        punpckhdq   xmm5,               xmm1            ; f3 e3 d3 c3 b3 a3 93 83 f2 e2 d2 c2 b2 a2 92 82
+        movdqa      xmm1,               xmm2            ; b7 a7 97 87 b6 a6 96 86 b5 a5 95 85 b4 a4 94 84
+
+        punpckldq   xmm1,               xmm6            ; f5 e5 d5 c5 b5 a5 95 85 f4 e4 d4 c4 b4 a4 94 84
+
+        punpckhdq   xmm2,               xmm6            ; f7 e7 d7 c7 b7 a7 97 87 f6 e6 d6 c6 b6 a6 96 86
+        movdqa      xmm6,               xmm7            ; 77 67 57 47 37 27 17 07 76 66 56 46 36 26 16 06
+
+        punpcklqdq  xmm6,               xmm2            ; f6 e6 d6 c6 b6 a6 96 86 76 66 56 46 36 26 16 06
+
+        punpckhqdq  xmm7,               xmm2            ; f7 e7 d7 c7 b7 a7 97 87 77 67 57 47 37 27 17 07
+%if %2
+        movdqa      xmm2,               xmm3            ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
+        punpcklqdq  xmm2,               xmm5            ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
+
+        punpckhqdq  xmm3,               xmm5            ; f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03
+
+        movdqa      [rdx],              xmm2            ; save 2
+
+        movdqa      xmm5,               xmm4            ; 75 65 55 45 35 25 15 05 74 64 54 44 34 24 14 04
+        punpcklqdq  xmm4,               xmm1            ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04
+
+        movdqa      [rdx+16],           xmm3            ; save 3
+
+        punpckhqdq  xmm5,               xmm1            ; f5 e5 d5 c5 b5 a5 95 85 75 65 55 45 35 25 15 05
+
+        movdqa      [rdx+32],           xmm4            ; save 4
+        movdqa      [rdx+48],           xmm5            ; save 5
+        movdqa      xmm1,               t0              ; get
+
+        movdqa      xmm2,               xmm1            ;
+        punpckhqdq  xmm1,               xmm0            ; f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01
+
+        punpcklqdq  xmm2,               xmm0            ; f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00
+%else
+        movdqa      [rdx+112],          xmm7            ; save 7
+
+        movdqa      [rdx+96],           xmm6            ; save 6
+
+        movdqa      xmm2,               xmm3            ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
+        punpckhqdq  xmm3,               xmm5            ; f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03
+
+        punpcklqdq  xmm2,               xmm5            ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
+
+        movdqa      [rdx+32],           xmm2            ; save 2
+
+        movdqa      xmm5,               xmm4            ; 75 65 55 45 35 25 15 05 74 64 54 44 34 24 14 04
+        punpcklqdq  xmm4,               xmm1            ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04
+
+        movdqa      [rdx+48],           xmm3            ; save 3
+
+        punpckhqdq  xmm5,               xmm1            ; f5 e5 d5 c5 b5 a5 95 85 75 65 55 45 35 25 15 05
+
+        movdqa      [rdx+64],           xmm4            ; save 4
+        movdqa      [rdx+80],           xmm5            ; save 5
+        movdqa      xmm1,               t0              ; get
+
+        movdqa      xmm2,               xmm1
+        punpckhqdq  xmm1,               xmm0            ; f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01
+
+        punpcklqdq  xmm2,               xmm0            ; f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00
+
+        movdqa      [rdx+16],           xmm1
+
+        movdqa      [rdx],              xmm2
+%endif
+%endmacro
+
+%macro LFV_FILTER_MASK_HEV_MASK 1
+        movdqa      xmm0,               xmm6            ; q2
+        psubusb     xmm0,               xmm7            ; q2-q3
+
+        psubusb     xmm7,               xmm6            ; q3-q2
+        movdqa      xmm4,               xmm5            ; q1
+
+        por         xmm7,               xmm0            ; abs (q3-q2)
+        psubusb     xmm4,               xmm6            ; q1-q2
+
+        movdqa      xmm0,               xmm1
+        psubusb     xmm6,               xmm5            ; q2-q1
+
+        por         xmm6,               xmm4            ; abs (q2-q1)
+        psubusb     xmm0,               xmm2            ; p2 - p3;
+
+        psubusb     xmm2,               xmm1            ; p3 - p2;
+        por         xmm0,               xmm2            ; abs(p2-p3)
+%if %1
+        movdqa      xmm2,               [rdx]           ; p1
+%else
+        movdqa      xmm2,               [rdx+32]        ; p1
+%endif
+        movdqa      xmm5,               xmm2            ; p1
+        pmaxub      xmm0,               xmm7
+
+        psubusb     xmm5,               xmm1            ; p1-p2
+        psubusb     xmm1,               xmm2            ; p2-p1
+
+        movdqa      xmm7,               xmm3            ; p0
+        psubusb     xmm7,               xmm2            ; p0-p1
+
+        por         xmm1,               xmm5            ; abs(p2-p1)
+        pmaxub      xmm0,               xmm6
+
+        pmaxub      xmm0,               xmm1
+        movdqa      xmm1,               xmm2            ; p1
+
+        psubusb     xmm2,               xmm3            ; p1-p0
+        lea         rdx,                srct
+
+        por         xmm2,               xmm7            ; abs(p1-p0)
+
+        movdqa      t0,                 xmm2            ; save abs(p1-p0)
+
+        pmaxub      xmm0,               xmm2
+
+%if %1
+        movdqa      xmm5,               [rdx+32]        ; q0
+        movdqa      xmm7,               [rdx+48]        ; q1
+%else
+        movdqa      xmm5,               [rdx+64]        ; q0
+        movdqa      xmm7,               [rdx+80]        ; q1
+%endif
+        mov         rdx,                arg(3)          ; limit
+
+        movdqa      xmm6,               xmm5            ; q0
+        movdqa      xmm2,               xmm7            ; q1
+
+        psubusb     xmm5,               xmm7            ; q0-q1
+        psubusb     xmm7,               xmm6            ; q1-q0
+
+        por         xmm7,               xmm5            ; abs(q1-q0)
+
+        movdqa      t1,                 xmm7            ; save abs(q1-q0)
+
+        movdqa      xmm4,               XMMWORD PTR [rdx]; limit
+
+        pmaxub      xmm0,               xmm7
+        mov         rdx,                arg(2)          ; blimit
+
+        psubusb     xmm0,               xmm4
+        movdqa      xmm5,               xmm2            ; q1
+
+        psubusb     xmm5,               xmm1            ; q1-=p1
+        psubusb     xmm1,               xmm2            ; p1-=q1
+
+        por         xmm5,               xmm1            ; abs(p1-q1)
+        movdqa      xmm1,               xmm3            ; p0
+
+        pand        xmm5,               [GLOBAL(tfe)]   ; set lsb of each byte to zero
+        psubusb     xmm1,               xmm6            ; p0-q0
+
+        psrlw       xmm5,               1               ; abs(p1-q1)/2
+        psubusb     xmm6,               xmm3            ; q0-p0
+
+        movdqa      xmm4,               XMMWORD PTR [rdx]; blimit
+
+        mov         rdx,                arg(4)          ; get thresh
+
+        por         xmm1,               xmm6            ; abs(q0-p0)
+
+        movdqa      xmm6,               t0              ; get abs (q1 - q0)
+
+        paddusb     xmm1,               xmm1            ; abs(q0-p0)*2
+
+        movdqa      xmm3,               t1              ; get abs (p1 - p0)
+
+        movdqa      xmm7,               XMMWORD PTR [rdx]
+
+        paddusb     xmm1,               xmm5            ; abs (p0 - q0) *2 + abs(p1-q1)/2
+        psubusb     xmm6,               xmm7            ; abs(q1 - q0) > thresh
+
+        psubusb     xmm3,               xmm7            ; abs(p1 - p0)> thresh
+
+        psubusb     xmm1,               xmm4            ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
+        por         xmm6,               xmm3            ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
+
+        por         xmm1,               xmm0            ; mask
+        pcmpeqb     xmm6,               xmm0
+
+        pxor        xmm0,               xmm0
+        pcmpeqb     xmm4,               xmm4
+
+        pcmpeqb     xmm1,               xmm0
+        pxor        xmm4,               xmm6
+%endmacro
+
+%macro BV_TRANSPOSE 0
+        ; xmm1 =    f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
+        ; xmm6 =    f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03
+        ; xmm3 =    f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04
+        ; xmm7 =    f5 e5 d5 c5 b5 a5 95 85 75 65 55 45 35 25 15 05
+        movdqa      xmm2,               xmm1            ; f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
+        punpcklbw   xmm2,               xmm6            ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02
+
+        movdqa      xmm4,               xmm3            ; f4 e4 d4 c4 b4 a4 94 84 74 64 54 44 34 24 14 04
+        punpckhbw   xmm1,               xmm6            ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82
+
+        punpcklbw   xmm4,               xmm7            ; 75 74 65 64 55 54 45 44 35 34 25 24 15 14 05 04
+
+        punpckhbw   xmm3,               xmm7            ; f5 f4 e5 e4 d5 d4 c5 c4 b5 b4 a5 a4 95 94 85 84
+
+        movdqa      xmm6,               xmm2            ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02
+        punpcklwd   xmm2,               xmm4            ; 35 34 33 32 25 24 23 22 15 14 13 12 05 04 03 02
+
+        punpckhwd   xmm6,               xmm4            ; 75 74 73 72 65 64 63 62 55 54 53 52 45 44 43 42
+        movdqa      xmm5,               xmm1            ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82
+
+        punpcklwd   xmm1,               xmm3            ; b5 b4 b3 b2 a5 a4 a3 a2 95 94 93 92 85 84 83 82
+
+        punpckhwd   xmm5,               xmm3            ; f5 f4 f3 f2 e5 e4 e3 e2 d5 d4 d3 d2 c5 c4 c3 c2
+        ; xmm2 = 35 34 33 32 25 24 23 22 15 14 13 12 05 04 03 02
+        ; xmm6 = 75 74 73 72 65 64 63 62 55 54 53 52 45 44 43 42
+        ; xmm1 = b5 b4 b3 b2 a5 a4 a3 a2 95 94 93 92 85 84 83 82
+        ; xmm5 = f5 f4 f3 f2 e5 e4 e3 e2 d5 d4 d3 d2 c5 c4 c3 c2
+%endmacro
+
+%macro BV_WRITEBACK 2
+        movd        [rsi+2],            %1
+        psrldq      %1,                 4
+
+        movd        [rdi+2],            %1
+        psrldq      %1,                 4
+
+        movd        [rsi+2*rax+2],      %1
+        psrldq      %1,                 4
+
+        movd        [rdi+2*rax+2],      %1
+
+        movd        [rsi+4*rax+2],      %2
+        psrldq      %2,                 4
+
+        movd        [rdi+4*rax+2],      %2
+        psrldq      %2,                 4
+
+        movd        [rsi+2*rcx+2],      %2
+        psrldq      %2,                 4
+
+        movd        [rdi+2*rcx+2],      %2
+%endmacro
+
+
+;void vp9_loop_filter_vertical_edge_sse2
+;(
+;    unsigned char *src_ptr,
+;    int            src_pixel_step,
+;    const char    *blimit,
+;    const char    *limit,
+;    const char    *thresh,
+;    int            count
+;)
+global sym(vp9_loop_filter_vertical_edge_sse2) PRIVATE
+sym(vp9_loop_filter_vertical_edge_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub             rsp, 96      ; reserve 96 bytes
+    %define t0      [rsp + 0]    ;__declspec(align(16)) char t0[16];
+    %define t1      [rsp + 16]   ;__declspec(align(16)) char t1[16];
+    %define srct    [rsp + 32]   ;__declspec(align(16)) char srct[64];
+
+        mov         rsi,        arg(0)                  ; src_ptr
+        movsxd      rax,        dword ptr arg(1)        ; src_pixel_step
+
+        lea         rsi,        [rsi - 4]
+        lea         rdi,        [rsi + rax]             ; rdi points to row +1 for indirect addressing
+        lea         rcx,        [rax*2+rax]
+
+        ;transpose 16x8 to 8x16, and store the 8-line result on stack.
+        TRANSPOSE_16X8 1, 1
+
+        ; calculate filter mask and high edge variance
+        LFV_FILTER_MASK_HEV_MASK 1
+
+        ; start work on filters
+        B_FILTER 2
+
+        ; tranpose and write back - only work on q1, q0, p0, p1
+        BV_TRANSPOSE
+        ; store 16-line result
+
+        lea         rdx,        [rax]
+        neg         rdx
+
+        BV_WRITEBACK xmm1, xmm5
+
+        lea         rsi,        [rsi+rdx*8]
+        lea         rdi,        [rdi+rdx*8]
+        BV_WRITEBACK xmm2, xmm6
+
+    add rsp, 96
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_loop_filter_vertical_edge_uv_sse2
+;(
+;    unsigned char *u,
+;    int            src_pixel_step,
+;    const char    *blimit,
+;    const char    *limit,
+;    const char    *thresh,
+;    unsigned char *v
+;)
+global sym(vp9_loop_filter_vertical_edge_uv_sse2) PRIVATE
+sym(vp9_loop_filter_vertical_edge_uv_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub             rsp, 96      ; reserve 96 bytes
+    %define t0      [rsp + 0]    ;__declspec(align(16)) char t0[16];
+    %define t1      [rsp + 16]   ;__declspec(align(16)) char t1[16];
+    %define srct    [rsp + 32]   ;__declspec(align(16)) char srct[64];
+
+        mov         rsi,        arg(0)                  ; u_ptr
+        movsxd      rax,        dword ptr arg(1)        ; src_pixel_step
+
+        lea         rsi,        [rsi - 4]
+        lea         rdi,        [rsi + rax]             ; rdi points to row +1 for indirect addressing
+        lea         rcx,        [rax+2*rax]
+
+        lea         rdx,        srct
+
+        ;transpose 16x8 to 8x16, and store the 8-line result on stack.
+        TRANSPOSE_16X8 0, 1
+
+        ; calculate filter mask and high edge variance
+        LFV_FILTER_MASK_HEV_MASK 1
+
+        ; start work on filters
+        B_FILTER 2
+
+        ; tranpose and write back - only work on q1, q0, p0, p1
+        BV_TRANSPOSE
+
+        lea         rdi,        [rsi + rax]             ; rdi points to row +1 for indirect addressing
+
+        ; store 16-line result
+        BV_WRITEBACK xmm1, xmm5
+
+        mov         rsi,        arg(0)                  ; u_ptr
+        lea         rsi,        [rsi - 4]
+        lea         rdi,        [rsi + rax]             ; rdi points to row +1 for indirect addressing
+        BV_WRITEBACK xmm2, xmm6
+
+    add rsp, 96
+    pop rsp
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+tfe:
+    times 16 db 0xfe
+align 16
+t80:
+    times 16 db 0x80
+align 16
+t1s:
+    times 16 db 0x01
+align 16
+t3:
+    times 16 db 0x03
+align 16
+t4:
+    times 16 db 0x04
+align 16
+ones:
+    times 8 dw 0x0001
+align 16
+s9:
+    times 8 dw 0x0900
+align 16
+s63:
+    times 8 dw 0x003f
diff --git a/libvpx/vp9/common/x86/vp9_loopfilter_x86.h b/libvpx/vp9/common/x86/vp9_loopfilter_x86.h
new file mode 100644
index 0000000..fb5af05
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_loopfilter_x86.h
@@ -0,0 +1,35 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_X86_VP9_LOOPFILTER_X86_H_
+#define VP9_COMMON_X86_VP9_LOOPFILTER_X86_H_
+
+/* Note:
+ *
+ * This platform is commonly built for runtime CPU detection. If you modify
+ * any of the function mappings present in this file, be sure to also update
+ * them in the function pointer initialization code
+ */
+
+#if HAVE_MMX
+extern prototype_loopfilter_block(vp9_loop_filter_mbv_mmx);
+extern prototype_loopfilter_block(vp9_loop_filter_bv_mmx);
+extern prototype_loopfilter_block(vp9_loop_filter_mbh_mmx);
+extern prototype_loopfilter_block(vp9_loop_filter_bh_mmx);
+#endif
+
+#if HAVE_SSE2
+extern prototype_loopfilter_block(vp9_loop_filter_mbv_sse2);
+extern prototype_loopfilter_block(vp9_loop_filter_bv_sse2);
+extern prototype_loopfilter_block(vp9_loop_filter_mbh_sse2);
+extern prototype_loopfilter_block(vp9_loop_filter_bh_sse2);
+#endif
+
+#endif  // LOOPFILTER_X86_H
diff --git a/libvpx/vp9/common/x86/vp9_mask_sse3.asm b/libvpx/vp9/common/x86/vp9_mask_sse3.asm
new file mode 100644
index 0000000..fe46823
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_mask_sse3.asm
@@ -0,0 +1,484 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void int vp8_makemask_sse3(
+;    unsigned char *y,
+;    unsigned char *u,
+;    unsigned char *v,
+;    unsigned char *ym,
+;    unsigned char *uvm,
+;    int yp,
+;    int uvp,
+;    int ys,
+;    int us,
+;    int vs,
+;    int yt,
+;    int ut,
+;    int vt)
+global sym(vp8_makemask_sse3) PRIVATE
+sym(vp8_makemask_sse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 14
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;y
+        mov             rdi,        arg(1) ;u
+        mov             rcx,        arg(2) ;v
+        mov             rax,        arg(3) ;ym
+        movsxd          rbx,        dword arg(4) ;yp
+        movsxd          rdx,        dword arg(5) ;uvp
+
+        pxor            xmm0,xmm0
+
+        ;make 16 copies of the center y value
+        movd            xmm1, arg(6)
+        pshufb          xmm1, xmm0
+
+        ; make 16 copies of the center u value
+        movd            xmm2, arg(7)
+        pshufb          xmm2, xmm0
+
+        ; make 16 copies of the center v value
+        movd            xmm3, arg(8)
+        pshufb          xmm3, xmm0
+        unpcklpd        xmm2, xmm3
+
+        ;make 16 copies of the y tolerance
+        movd            xmm3, arg(9)
+        pshufb          xmm3, xmm0
+
+        ;make 16 copies of the u tolerance
+        movd            xmm4, arg(10)
+        pshufb          xmm4, xmm0
+
+        ;make 16 copies of the v tolerance
+        movd            xmm5, arg(11)
+        pshufb          xmm5, xmm0
+        unpckhpd        xmm4, xmm5
+
+        mov             r8,8
+
+NextPairOfRows:
+
+        ;grab the y source values
+        movdqu          xmm0, [rsi]
+
+        ;compute abs difference between source and y target
+        movdqa          xmm6, xmm1
+        movdqa          xmm7, xmm0
+        psubusb         xmm0, xmm1
+        psubusb         xmm6, xmm7
+        por             xmm0, xmm6
+
+        ;compute abs difference between
+        movdqa          xmm6, xmm3
+        pcmpgtb         xmm6, xmm0
+
+        ;grab the y source values
+        add             rsi, rbx
+        movdqu          xmm0, [rsi]
+
+        ;compute abs difference between source and y target
+        movdqa          xmm11, xmm1
+        movdqa          xmm7, xmm0
+        psubusb         xmm0, xmm1
+        psubusb         xmm11, xmm7
+        por             xmm0, xmm11
+
+        ;compute abs difference between
+        movdqa          xmm11, xmm3
+        pcmpgtb         xmm11, xmm0
+
+
+        ;grab the u and v source values
+        movdqu          xmm7, [rdi]
+        movdqu          xmm8, [rcx]
+        unpcklpd        xmm7, xmm8
+
+        ;compute abs difference between source and uv targets
+        movdqa          xmm9, xmm2
+        movdqa          xmm10, xmm7
+        psubusb         xmm7, xmm2
+        psubusb         xmm9, xmm10
+        por             xmm7, xmm9
+
+        ;check whether the number is < tolerance
+        movdqa          xmm0, xmm4
+        pcmpgtb         xmm0, xmm7
+
+        ;double  u and v masks
+        movdqa          xmm8, xmm0
+        punpckhbw       xmm0, xmm0
+        punpcklbw       xmm8, xmm8
+
+        ;mask row 0 and output
+        pand            xmm6, xmm8
+        pand            xmm6, xmm0
+        movdqa          [rax],xmm6
+
+        ;mask row 1 and output
+        pand            xmm11, xmm8
+        pand            xmm11, xmm0
+        movdqa          [rax+16],xmm11
+
+
+        ; to the next row or set of rows
+        add             rsi, rbx
+        add             rdi, rdx
+        add             rcx, rdx
+        add             rax,32
+        dec r8
+        jnz NextPairOfRows
+
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;GROW_HORIZ (register for result, source register or mem local)
+; takes source and shifts left and ors with source
+; then shifts right and ors with source
+%macro GROW_HORIZ 2
+    movdqa          %1, %2
+    movdqa          xmm14, %1
+    movdqa          xmm15, %1
+    pslldq          xmm14, 1
+    psrldq          xmm15, 1
+    por             %1,xmm14
+    por             %1,xmm15
+%endmacro
+;GROW_VERT (result, center row, above row, below row)
+%macro GROW_VERT 4
+    movdqa          %1,%2
+    por             %1,%3
+    por             %1,%4
+%endmacro
+
+;GROW_NEXTLINE (new line to grow, new source, line to write)
+%macro GROW_NEXTLINE 3
+    GROW_HORIZ %1, %2
+    GROW_VERT xmm3, xmm0, xmm1, xmm2
+    movdqa %3,xmm3
+%endmacro
+
+
+;void int vp8_growmaskmb_sse3(
+;    unsigned char *om,
+;    unsigned char *nm,
+global sym(vp8_growmaskmb_sse3) PRIVATE
+sym(vp8_growmaskmb_sse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0) ;src
+    mov             rdi,        arg(1) ;rst
+
+    GROW_HORIZ xmm0, [rsi]
+    GROW_HORIZ xmm1, [rsi+16]
+    GROW_HORIZ xmm2, [rsi+32]
+
+    GROW_VERT xmm3, xmm0, xmm1, xmm2
+    por xmm0,xmm1
+    movdqa [rdi], xmm0
+    movdqa [rdi+16],xmm3
+
+    GROW_NEXTLINE xmm0,[rsi+48],[rdi+32]
+    GROW_NEXTLINE xmm1,[rsi+64],[rdi+48]
+    GROW_NEXTLINE xmm2,[rsi+80],[rdi+64]
+    GROW_NEXTLINE xmm0,[rsi+96],[rdi+80]
+    GROW_NEXTLINE xmm1,[rsi+112],[rdi+96]
+    GROW_NEXTLINE xmm2,[rsi+128],[rdi+112]
+    GROW_NEXTLINE xmm0,[rsi+144],[rdi+128]
+    GROW_NEXTLINE xmm1,[rsi+160],[rdi+144]
+    GROW_NEXTLINE xmm2,[rsi+176],[rdi+160]
+    GROW_NEXTLINE xmm0,[rsi+192],[rdi+176]
+    GROW_NEXTLINE xmm1,[rsi+208],[rdi+192]
+    GROW_NEXTLINE xmm2,[rsi+224],[rdi+208]
+    GROW_NEXTLINE xmm0,[rsi+240],[rdi+224]
+
+    por xmm0,xmm2
+    movdqa [rdi+240], xmm0
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+;unsigned int vp8_sad16x16_masked_wmt(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned char *mask)
+global sym(vp8_sad16x16_masked_wmt) PRIVATE
+sym(vp8_sad16x16_masked_wmt):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+    mov             rsi,        arg(0) ;src_ptr
+    mov             rdi,        arg(2) ;ref_ptr
+
+    mov             rbx,        arg(4) ;mask
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    mov             rcx,        16
+
+    pxor            xmm3,       xmm3
+
+NextSadRow:
+    movdqu          xmm0,       [rsi]
+    movdqu          xmm1,       [rdi]
+    movdqu          xmm2,       [rbx]
+    pand            xmm0,       xmm2
+    pand            xmm1,       xmm2
+
+    psadbw          xmm0,       xmm1
+    paddw           xmm3,       xmm0
+
+    add             rsi, rax
+    add             rdi, rdx
+    add             rbx,  16
+
+    dec rcx
+    jnz NextSadRow
+
+    movdqa          xmm4 ,     xmm3
+    psrldq          xmm4,       8
+    paddw           xmm3,      xmm4
+    movq            rax,       xmm3
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp8_sad16x16_unmasked_wmt(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned char *mask)
+global sym(vp8_sad16x16_unmasked_wmt) PRIVATE
+sym(vp8_sad16x16_unmasked_wmt):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+    mov             rsi,        arg(0) ;src_ptr
+    mov             rdi,        arg(2) ;ref_ptr
+
+    mov             rbx,        arg(4) ;mask
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    mov             rcx,        16
+
+    pxor            xmm3,       xmm3
+
+next_vp8_sad16x16_unmasked_wmt:
+    movdqu          xmm0,       [rsi]
+    movdqu          xmm1,       [rdi]
+    movdqu          xmm2,       [rbx]
+    por             xmm0,       xmm2
+    por             xmm1,       xmm2
+
+    psadbw          xmm0,       xmm1
+    paddw           xmm3,       xmm0
+
+    add             rsi, rax
+    add             rdi, rdx
+    add             rbx,  16
+
+    dec rcx
+    jnz next_vp8_sad16x16_unmasked_wmt
+
+    movdqa          xmm4 ,     xmm3
+    psrldq          xmm4,       8
+    paddw           xmm3,      xmm4
+    movq            rax,        xmm3
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp8_masked_predictor_wmt(
+;    unsigned char *masked,
+;    unsigned char *unmasked,
+;    int  src_stride,
+;    unsigned char *dst_ptr,
+;    int  dst_stride,
+;    unsigned char *mask)
+global sym(vp8_masked_predictor_wmt) PRIVATE
+sym(vp8_masked_predictor_wmt):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    push        rsi
+    push        rdi
+    ; end prolog
+    mov             rsi,        arg(0) ;src_ptr
+    mov             rdi,        arg(1) ;ref_ptr
+
+    mov             rbx,        arg(5) ;mask
+    movsxd          rax,        dword ptr arg(2) ;src_stride
+    mov             r11,        arg(3) ; destination
+    movsxd          rdx,        dword ptr arg(4) ;dst_stride
+
+    mov             rcx,        16
+
+    pxor            xmm3,       xmm3
+
+next_vp8_masked_predictor_wmt:
+    movdqu          xmm0,       [rsi]
+    movdqu          xmm1,       [rdi]
+    movdqu          xmm2,       [rbx]
+
+    pand            xmm0,       xmm2
+    pandn           xmm2,       xmm1
+    por             xmm0,       xmm2
+    movdqu          [r11],      xmm0
+
+    add             r11, rdx
+    add             rsi, rax
+    add             rdi, rdx
+    add             rbx,  16
+
+    dec rcx
+    jnz next_vp8_masked_predictor_wmt
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;unsigned int vp8_masked_predictor_uv_wmt(
+;    unsigned char *masked,
+;    unsigned char *unmasked,
+;    int  src_stride,
+;    unsigned char *dst_ptr,
+;    int  dst_stride,
+;    unsigned char *mask)
+global sym(vp8_masked_predictor_uv_wmt) PRIVATE
+sym(vp8_masked_predictor_uv_wmt):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    push        rsi
+    push        rdi
+    ; end prolog
+    mov             rsi,        arg(0) ;src_ptr
+    mov             rdi,        arg(1) ;ref_ptr
+
+    mov             rbx,        arg(5) ;mask
+    movsxd          rax,        dword ptr arg(2) ;src_stride
+    mov             r11,        arg(3) ; destination
+    movsxd          rdx,        dword ptr arg(4) ;dst_stride
+
+    mov             rcx,        8
+
+    pxor            xmm3,       xmm3
+
+next_vp8_masked_predictor_uv_wmt:
+    movq            xmm0,       [rsi]
+    movq            xmm1,       [rdi]
+    movq            xmm2,       [rbx]
+
+    pand            xmm0,       xmm2
+    pandn           xmm2,       xmm1
+    por             xmm0,       xmm2
+    movq            [r11],      xmm0
+
+    add             r11, rdx
+    add             rsi, rax
+    add             rdi, rax
+    add             rbx,  8
+
+    dec rcx
+    jnz next_vp8_masked_predictor_uv_wmt
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp8_uv_from_y_mask(
+;    unsigned char *ymask,
+;    unsigned char *uvmask)
+global sym(vp8_uv_from_y_mask) PRIVATE
+sym(vp8_uv_from_y_mask):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    push        rsi
+    push        rdi
+    ; end prolog
+    mov             rsi,        arg(0) ;src_ptr
+    mov             rdi,        arg(1) ;dst_ptr
+
+
+    mov             rcx,        8
+
+    pxor            xmm3,       xmm3
+
+next_p8_uv_from_y_mask:
+    movdqu          xmm0,       [rsi]
+    pshufb          xmm0, [shuf1b] ;[GLOBAL(shuf1b)]
+    movq            [rdi],xmm0
+    add             rdi, 8
+    add             rsi,32
+
+    dec rcx
+    jnz next_p8_uv_from_y_mask
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+shuf1b:
+    db 0, 2, 4, 6, 8, 10, 12, 14, 0, 0, 0, 0, 0, 0, 0, 0
+
diff --git a/libvpx/vp9/common/x86/vp9_postproc_mmx.asm b/libvpx/vp9/common/x86/vp9_postproc_mmx.asm
new file mode 100644
index 0000000..c2118db
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_postproc_mmx.asm
@@ -0,0 +1,534 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%define VP9_FILTER_WEIGHT 128
+%define VP9_FILTER_SHIFT  7
+
+;void vp9_post_proc_down_and_across_mmx
+;(
+;    unsigned char *src_ptr,
+;    unsigned char *dst_ptr,
+;    int src_pixels_per_line,
+;    int dst_pixels_per_line,
+;    int rows,
+;    int cols,
+;    int flimit
+;)
+global sym(vp9_post_proc_down_and_across_mmx) PRIVATE
+sym(vp9_post_proc_down_and_across_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+%if ABI_IS_32BIT=1 && CONFIG_PIC=1
+    ; move the global rd onto the stack, since we don't have enough registers
+    ; to do PIC addressing
+    movq        mm0, [GLOBAL(rd)]
+    sub         rsp, 8
+    movq        [rsp], mm0
+%define RD [rsp]
+%else
+%define RD [GLOBAL(rd)]
+%endif
+
+        push        rbx
+        lea         rbx, [GLOBAL(Blur)]
+        movd        mm2, dword ptr arg(6) ;flimit
+        punpcklwd   mm2, mm2
+        punpckldq   mm2, mm2
+
+        mov         rsi,        arg(0) ;src_ptr
+        mov         rdi,        arg(1) ;dst_ptr
+
+        movsxd      rcx, DWORD PTR arg(4) ;rows
+        movsxd      rax, DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
+        pxor        mm0, mm0              ; mm0 = 00000000
+
+.nextrow:
+
+        xor         rdx,        rdx       ; clear out rdx for use as loop counter
+.nextcol:
+
+        pxor        mm7, mm7              ; mm7 = 00000000
+        movq        mm6, [rbx + 32 ]      ; mm6 = kernel 2 taps
+        movq        mm3, [rsi]            ; mm4 = r0 p0..p7
+        punpcklbw   mm3, mm0              ; mm3 = p0..p3
+        movq        mm1, mm3              ; mm1 = p0..p3
+        pmullw      mm3, mm6              ; mm3 *= kernel 2 modifiers
+
+        movq        mm6, [rbx + 48]       ; mm6 = kernel 3 taps
+        movq        mm5, [rsi + rax]      ; mm4 = r1 p0..p7
+        punpcklbw   mm5, mm0              ; mm5 = r1 p0..p3
+        pmullw      mm6, mm5              ; mm6 *= p0..p3 * kernel 3 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm6
+
+        ; thresholding
+        movq        mm7, mm1              ; mm7 = r0 p0..p3
+        psubusw     mm7, mm5              ; mm7 = r0 p0..p3 - r1 p0..p3
+        psubusw     mm5, mm1              ; mm5 = r1 p0..p3 - r0 p0..p3
+        paddusw     mm7, mm5              ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
+        pcmpgtw     mm7, mm2
+
+        movq        mm6, [rbx + 64 ]      ; mm6 = kernel 4 modifiers
+        movq        mm5, [rsi + 2*rax]    ; mm4 = r2 p0..p7
+        punpcklbw   mm5, mm0              ; mm5 = r2 p0..p3
+        pmullw      mm6, mm5              ; mm5 *= kernel 4 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = r0 p0..p3
+        psubusw     mm6, mm5              ; mm6 = r0 p0..p3 - r2 p0..p3
+        psubusw     mm5, mm1              ; mm5 = r2 p0..p3 - r2 p0..p3
+        paddusw     mm6, mm5              ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+
+        neg         rax
+        movq        mm6, [rbx ]           ; kernel 0 taps
+        movq        mm5, [rsi+2*rax]      ; mm4 = r-2 p0..p7
+        punpcklbw   mm5, mm0              ; mm5 = r-2 p0..p3
+        pmullw      mm6, mm5              ; mm5 *= kernel 0 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = r0 p0..p3
+        psubusw     mm6, mm5              ; mm6 = p0..p3 - r-2 p0..p3
+        psubusw     mm5, mm1              ; mm5 = r-2 p0..p3 - p0..p3
+        paddusw     mm6, mm5              ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+        movq        mm6, [rbx + 16]       ; kernel 1 taps
+        movq        mm4, [rsi+rax]        ; mm4 = r-1 p0..p7
+        punpcklbw   mm4, mm0              ; mm4 = r-1 p0..p3
+        pmullw      mm6, mm4              ; mm4 *= kernel 1 modifiers.
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = r0 p0..p3
+        psubusw     mm6, mm4              ; mm6 = p0..p3 - r-2 p0..p3
+        psubusw     mm4, mm1              ; mm5 = r-1 p0..p3 - p0..p3
+        paddusw     mm6, mm4              ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+
+        paddusw     mm3, RD               ; mm3 += round value
+        psraw       mm3, VP9_FILTER_SHIFT     ; mm3 /= 128
+
+        pand        mm1, mm7              ; mm1 select vals > thresh from source
+        pandn       mm7, mm3              ; mm7 select vals < thresh from blurred result
+        paddusw     mm1, mm7              ; combination
+
+        packuswb    mm1, mm0              ; pack to bytes
+
+        movd        [rdi], mm1            ;
+        neg         rax                   ; pitch is positive
+
+
+        add         rsi, 4
+        add         rdi, 4
+        add         rdx, 4
+
+        cmp         edx, dword ptr arg(5) ;cols
+        jl          .nextcol
+        ; done with the all cols, start the across filtering in place
+        sub         rsi, rdx
+        sub         rdi, rdx
+
+
+        push        rax
+        xor         rdx,    rdx
+        mov         rax,    [rdi-4];
+
+.acrossnextcol:
+        pxor        mm7, mm7              ; mm7 = 00000000
+        movq        mm6, [rbx + 32 ]      ;
+        movq        mm4, [rdi+rdx]        ; mm4 = p0..p7
+        movq        mm3, mm4              ; mm3 = p0..p7
+        punpcklbw   mm3, mm0              ; mm3 = p0..p3
+        movq        mm1, mm3              ; mm1 = p0..p3
+        pmullw      mm3, mm6              ; mm3 *= kernel 2 modifiers
+
+        movq        mm6, [rbx + 48]
+        psrlq       mm4, 8                ; mm4 = p1..p7
+        movq        mm5, mm4              ; mm5 = p1..p7
+        punpcklbw   mm5, mm0              ; mm5 = p1..p4
+        pmullw      mm6, mm5              ; mm6 *= p1..p4 * kernel 3 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm6
+
+        ; thresholding
+        movq        mm7, mm1              ; mm7 = p0..p3
+        psubusw     mm7, mm5              ; mm7 = p0..p3 - p1..p4
+        psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     mm7, mm5              ; mm7 = abs(p0..p3 - p1..p4)
+        pcmpgtw     mm7, mm2
+
+        movq        mm6, [rbx + 64 ]
+        psrlq       mm4, 8                ; mm4 = p2..p7
+        movq        mm5, mm4              ; mm5 = p2..p7
+        punpcklbw   mm5, mm0              ; mm5 = p2..p5
+        pmullw      mm6, mm5              ; mm5 *= kernel 4 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = p0..p3
+        psubusw     mm6, mm5              ; mm6 = p0..p3 - p1..p4
+        psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     mm6, mm5              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+
+        movq        mm6, [rbx ]
+        movq        mm4, [rdi+rdx-2]      ; mm4 = p-2..p5
+        movq        mm5, mm4              ; mm5 = p-2..p5
+        punpcklbw   mm5, mm0              ; mm5 = p-2..p1
+        pmullw      mm6, mm5              ; mm5 *= kernel 0 modifiers
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = p0..p3
+        psubusw     mm6, mm5              ; mm6 = p0..p3 - p1..p4
+        psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     mm6, mm5              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+        movq        mm6, [rbx + 16]
+        psrlq       mm4, 8                ; mm4 = p-1..p5
+        punpcklbw   mm4, mm0              ; mm4 = p-1..p2
+        pmullw      mm6, mm4              ; mm4 *= kernel 1 modifiers.
+        paddusw     mm3, mm6              ; mm3 += mm5
+
+        ; thresholding
+        movq        mm6, mm1              ; mm6 = p0..p3
+        psubusw     mm6, mm4              ; mm6 = p0..p3 - p1..p4
+        psubusw     mm4, mm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     mm6, mm4              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     mm6, mm2
+        por         mm7, mm6              ; accumulate thresholds
+
+        paddusw     mm3, RD               ; mm3 += round value
+        psraw       mm3, VP9_FILTER_SHIFT     ; mm3 /= 128
+
+        pand        mm1, mm7              ; mm1 select vals > thresh from source
+        pandn       mm7, mm3              ; mm7 select vals < thresh from blurred result
+        paddusw     mm1, mm7              ; combination
+
+        packuswb    mm1, mm0              ; pack to bytes
+        mov         DWORD PTR [rdi+rdx-4],  eax   ; store previous four bytes
+        movd        eax,    mm1
+
+        add         rdx, 4
+        cmp         edx, dword ptr arg(5) ;cols
+        jl          .acrossnextcol;
+
+        mov         DWORD PTR [rdi+rdx-4],  eax
+        pop         rax
+
+        ; done with this rwo
+        add         rsi,rax               ; next line
+        movsxd      rax, dword ptr arg(3) ;dst_pixels_per_line ; destination pitch?
+        add         rdi,rax               ; next destination
+        movsxd      rax, dword ptr arg(2) ;src_pixels_per_line ; destination pitch?
+
+        dec         rcx                   ; decrement count
+        jnz         .nextrow               ; next row
+        pop         rbx
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%undef RD
+
+
+;void vp9_mbpost_proc_down_mmx(unsigned char *dst,
+;                             int pitch, int rows, int cols,int flimit)
+extern sym(vp9_rv)
+global sym(vp9_mbpost_proc_down_mmx) PRIVATE
+sym(vp9_mbpost_proc_down_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 136
+
+    ; unsigned char d[16][8] at [rsp]
+    ; create flimit2 at [rsp+128]
+    mov         eax, dword ptr arg(4) ;flimit
+    mov         [rsp+128], eax
+    mov         [rsp+128+4], eax
+%define flimit2 [rsp+128]
+
+%if ABI_IS_32BIT=0
+    lea         r8,       [GLOBAL(sym(vp9_rv))]
+%endif
+
+    ;rows +=8;
+    add         dword ptr arg(2), 8
+
+    ;for(c=0; c<cols; c+=4)
+.loop_col:
+            mov         rsi,        arg(0)  ;s
+            pxor        mm0,        mm0     ;
+
+            movsxd      rax,        dword ptr arg(1) ;pitch       ;
+            neg         rax                                     ; rax = -pitch
+
+            lea         rsi,        [rsi + rax*8];              ; rdi = s[-pitch*8]
+            neg         rax
+
+
+            pxor        mm5,        mm5
+            pxor        mm6,        mm6     ;
+
+            pxor        mm7,        mm7     ;
+            mov         rdi,        rsi
+
+            mov         rcx,        15          ;
+
+.loop_initvar:
+            movd        mm1,        DWORD PTR [rdi];
+            punpcklbw   mm1,        mm0     ;
+
+            paddw       mm5,        mm1     ;
+            pmullw      mm1,        mm1     ;
+
+            movq        mm2,        mm1     ;
+            punpcklwd   mm1,        mm0     ;
+
+            punpckhwd   mm2,        mm0     ;
+            paddd       mm6,        mm1     ;
+
+            paddd       mm7,        mm2     ;
+            lea         rdi,        [rdi+rax]   ;
+
+            dec         rcx
+            jne         .loop_initvar
+            ;save the var and sum
+            xor         rdx,        rdx
+.loop_row:
+            movd        mm1,        DWORD PTR [rsi]     ; [s-pitch*8]
+            movd        mm2,        DWORD PTR [rdi]     ; [s+pitch*7]
+
+            punpcklbw   mm1,        mm0
+            punpcklbw   mm2,        mm0
+
+            paddw       mm5,        mm2
+            psubw       mm5,        mm1
+
+            pmullw      mm2,        mm2
+            movq        mm4,        mm2
+
+            punpcklwd   mm2,        mm0
+            punpckhwd   mm4,        mm0
+
+            paddd       mm6,        mm2
+            paddd       mm7,        mm4
+
+            pmullw      mm1,        mm1
+            movq        mm2,        mm1
+
+            punpcklwd   mm1,        mm0
+            psubd       mm6,        mm1
+
+            punpckhwd   mm2,        mm0
+            psubd       mm7,        mm2
+
+
+            movq        mm3,        mm6
+            pslld       mm3,        4
+
+            psubd       mm3,        mm6
+            movq        mm1,        mm5
+
+            movq        mm4,        mm5
+            pmullw      mm1,        mm1
+
+            pmulhw      mm4,        mm4
+            movq        mm2,        mm1
+
+            punpcklwd   mm1,        mm4
+            punpckhwd   mm2,        mm4
+
+            movq        mm4,        mm7
+            pslld       mm4,        4
+
+            psubd       mm4,        mm7
+
+            psubd       mm3,        mm1
+            psubd       mm4,        mm2
+
+            psubd       mm3,        flimit2
+            psubd       mm4,        flimit2
+
+            psrad       mm3,        31
+            psrad       mm4,        31
+
+            packssdw    mm3,        mm4
+            packsswb    mm3,        mm0
+
+            movd        mm1,        DWORD PTR [rsi+rax*8]
+
+            movq        mm2,        mm1
+            punpcklbw   mm1,        mm0
+
+            paddw       mm1,        mm5
+            mov         rcx,        rdx
+
+            and         rcx,        127
+%if ABI_IS_32BIT=1 && CONFIG_PIC=1
+            push        rax
+            lea         rax,        [GLOBAL(sym(vp9_rv))]
+            movq        mm4,        [rax + rcx*2] ;vp9_rv[rcx*2]
+            pop         rax
+%elif ABI_IS_32BIT=0
+            movq        mm4,        [r8 + rcx*2] ;vp9_rv[rcx*2]
+%else
+            movq        mm4,        [sym(vp9_rv) + rcx*2]
+%endif
+            paddw       mm1,        mm4
+            ;paddw     xmm1,       eight8s
+            psraw       mm1,        4
+
+            packuswb    mm1,        mm0
+            pand        mm1,        mm3
+
+            pandn       mm3,        mm2
+            por         mm1,        mm3
+
+            and         rcx,        15
+            movd        DWORD PTR   [rsp+rcx*4], mm1 ;d[rcx*4]
+
+            mov         rcx,        rdx
+            sub         rcx,        8
+
+            and         rcx,        15
+            movd        mm1,        DWORD PTR [rsp+rcx*4] ;d[rcx*4]
+
+            movd        [rsi],      mm1
+            lea         rsi,        [rsi+rax]
+
+            lea         rdi,        [rdi+rax]
+            add         rdx,        1
+
+            cmp         edx,        dword arg(2) ;rows
+            jl          .loop_row
+
+
+        add         dword arg(0), 4 ; s += 4
+        sub         dword arg(3), 4 ; cols -= 4
+        cmp         dword arg(3), 0
+        jg          .loop_col
+
+    add         rsp, 136
+    pop         rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%undef flimit2
+
+
+;void vp9_plane_add_noise_mmx (unsigned char *start, unsigned char *noise,
+;                            unsigned char blackclamp[16],
+;                            unsigned char whiteclamp[16],
+;                            unsigned char bothclamp[16],
+;                            unsigned int width, unsigned int height, int pitch)
+extern sym(rand)
+global sym(vp9_plane_add_noise_mmx) PRIVATE
+sym(vp9_plane_add_noise_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 8
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+.addnoise_loop:
+    call sym(rand) WRT_PLT
+    mov     rcx, arg(1) ;noise
+    and     rax, 0xff
+    add     rcx, rax
+
+    ; we rely on the fact that the clamping vectors are stored contiguously
+    ; in black/white/both order. Note that we have to reload this here because
+    ; rdx could be trashed by rand()
+    mov     rdx, arg(2) ; blackclamp
+
+
+            mov     rdi, rcx
+            movsxd  rcx, dword arg(5) ;[Width]
+            mov     rsi, arg(0) ;Pos
+            xor         rax,rax
+
+.addnoise_nextset:
+            movq        mm1,[rsi+rax]         ; get the source
+
+            psubusb     mm1, [rdx]    ;blackclamp        ; clamp both sides so we don't outrange adding noise
+            paddusb     mm1, [rdx+32] ;bothclamp
+            psubusb     mm1, [rdx+16] ;whiteclamp
+
+            movq        mm2,[rdi+rax]         ; get the noise for this line
+            paddb       mm1,mm2              ; add it in
+            movq        [rsi+rax],mm1         ; store the result
+
+            add         rax,8                 ; move to the next line
+
+            cmp         rax, rcx
+            jl          .addnoise_nextset
+
+    movsxd  rax, dword arg(7) ; Pitch
+    add     arg(0), rax ; Start += Pitch
+    sub     dword arg(6), 1   ; Height -= 1
+    jg      .addnoise_loop
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+SECTION_RODATA
+align 16
+Blur:
+    times 16 dw 16
+    times  8 dw 64
+    times 16 dw 16
+    times  8 dw  0
+
+rd:
+    times 4 dw 0x40
diff --git a/libvpx/vp9/common/x86/vp9_postproc_sse2.asm b/libvpx/vp9/common/x86/vp9_postproc_sse2.asm
new file mode 100644
index 0000000..858fc99
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_postproc_sse2.asm
@@ -0,0 +1,695 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_post_proc_down_and_across_xmm
+;(
+;    unsigned char *src_ptr,
+;    unsigned char *dst_ptr,
+;    int src_pixels_per_line,
+;    int dst_pixels_per_line,
+;    int rows,
+;    int cols,
+;    int flimit
+;)
+global sym(vp9_post_proc_down_and_across_xmm) PRIVATE
+sym(vp9_post_proc_down_and_across_xmm):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+%if ABI_IS_32BIT=1 && CONFIG_PIC=1
+    ALIGN_STACK 16, rax
+    ; move the global rd onto the stack, since we don't have enough registers
+    ; to do PIC addressing
+    movdqa      xmm0, [GLOBAL(rd42)]
+    sub         rsp, 16
+    movdqa      [rsp], xmm0
+%define RD42 [rsp]
+%else
+%define RD42 [GLOBAL(rd42)]
+%endif
+
+
+        movd        xmm2,       dword ptr arg(6) ;flimit
+        punpcklwd   xmm2,       xmm2
+        punpckldq   xmm2,       xmm2
+        punpcklqdq  xmm2,       xmm2
+
+        mov         rsi,        arg(0) ;src_ptr
+        mov         rdi,        arg(1) ;dst_ptr
+
+        movsxd      rcx,        DWORD PTR arg(4) ;rows
+        movsxd      rax,        DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
+        pxor        xmm0,       xmm0              ; mm0 = 00000000
+
+.nextrow:
+
+        xor         rdx,        rdx       ; clear out rdx for use as loop counter
+.nextcol:
+        movq        xmm3,       QWORD PTR [rsi]         ; mm4 = r0 p0..p7
+        punpcklbw   xmm3,       xmm0                    ; mm3 = p0..p3
+        movdqa      xmm1,       xmm3                    ; mm1 = p0..p3
+        psllw       xmm3,       2                       ;
+
+        movq        xmm5,       QWORD PTR [rsi + rax]   ; mm4 = r1 p0..p7
+        punpcklbw   xmm5,       xmm0                    ; mm5 = r1 p0..p3
+        paddusw     xmm3,       xmm5                    ; mm3 += mm6
+
+        ; thresholding
+        movdqa      xmm7,       xmm1                    ; mm7 = r0 p0..p3
+        psubusw     xmm7,       xmm5                    ; mm7 = r0 p0..p3 - r1 p0..p3
+        psubusw     xmm5,       xmm1                    ; mm5 = r1 p0..p3 - r0 p0..p3
+        paddusw     xmm7,       xmm5                    ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
+        pcmpgtw     xmm7,       xmm2
+
+        movq        xmm5,       QWORD PTR [rsi + 2*rax] ; mm4 = r2 p0..p7
+        punpcklbw   xmm5,       xmm0                    ; mm5 = r2 p0..p3
+        paddusw     xmm3,       xmm5                    ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
+        psubusw     xmm6,       xmm5                    ; mm6 = r0 p0..p3 - r2 p0..p3
+        psubusw     xmm5,       xmm1                    ; mm5 = r2 p0..p3 - r2 p0..p3
+        paddusw     xmm6,       xmm5                    ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6                    ; accumulate thresholds
+
+
+        neg         rax
+        movq        xmm5,       QWORD PTR [rsi+2*rax]   ; mm4 = r-2 p0..p7
+        punpcklbw   xmm5,       xmm0                    ; mm5 = r-2 p0..p3
+        paddusw     xmm3,       xmm5                    ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
+        psubusw     xmm6,       xmm5                    ; mm6 = p0..p3 - r-2 p0..p3
+        psubusw     xmm5,       xmm1                    ; mm5 = r-2 p0..p3 - p0..p3
+        paddusw     xmm6,       xmm5                    ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6                    ; accumulate thresholds
+
+        movq        xmm4,       QWORD PTR [rsi+rax]     ; mm4 = r-1 p0..p7
+        punpcklbw   xmm4,       xmm0                    ; mm4 = r-1 p0..p3
+        paddusw     xmm3,       xmm4                    ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
+        psubusw     xmm6,       xmm4                    ; mm6 = p0..p3 - r-2 p0..p3
+        psubusw     xmm4,       xmm1                    ; mm5 = r-1 p0..p3 - p0..p3
+        paddusw     xmm6,       xmm4                    ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6                    ; accumulate thresholds
+
+
+        paddusw     xmm3,       RD42                    ; mm3 += round value
+        psraw       xmm3,       3                       ; mm3 /= 8
+
+        pand        xmm1,       xmm7                    ; mm1 select vals > thresh from source
+        pandn       xmm7,       xmm3                    ; mm7 select vals < thresh from blurred result
+        paddusw     xmm1,       xmm7                    ; combination
+
+        packuswb    xmm1,       xmm0                    ; pack to bytes
+        movq        QWORD PTR [rdi], xmm1             ;
+
+        neg         rax                   ; pitch is positive
+        add         rsi,        8
+        add         rdi,        8
+
+        add         rdx,        8
+        cmp         edx,        dword arg(5) ;cols
+
+        jl          .nextcol
+
+        ; done with the all cols, start the across filtering in place
+        sub         rsi,        rdx
+        sub         rdi,        rdx
+
+        xor         rdx,        rdx
+        movq        mm0,        QWORD PTR [rdi-8];
+
+.acrossnextcol:
+        movq        xmm7,       QWORD PTR [rdi +rdx -2]
+        movd        xmm4,       DWORD PTR [rdi +rdx +6]
+
+        pslldq      xmm4,       8
+        por         xmm4,       xmm7
+
+        movdqa      xmm3,       xmm4
+        psrldq      xmm3,       2
+        punpcklbw   xmm3,       xmm0              ; mm3 = p0..p3
+        movdqa      xmm1,       xmm3              ; mm1 = p0..p3
+        psllw       xmm3,       2
+
+
+        movdqa      xmm5,       xmm4
+        psrldq      xmm5,       3
+        punpcklbw   xmm5,       xmm0              ; mm5 = p1..p4
+        paddusw     xmm3,       xmm5              ; mm3 += mm6
+
+        ; thresholding
+        movdqa      xmm7,       xmm1              ; mm7 = p0..p3
+        psubusw     xmm7,       xmm5              ; mm7 = p0..p3 - p1..p4
+        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     xmm7,       xmm5              ; mm7 = abs(p0..p3 - p1..p4)
+        pcmpgtw     xmm7,       xmm2
+
+        movdqa      xmm5,       xmm4
+        psrldq      xmm5,       4
+        punpcklbw   xmm5,       xmm0              ; mm5 = p2..p5
+        paddusw     xmm3,       xmm5              ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
+        psubusw     xmm6,       xmm5              ; mm6 = p0..p3 - p1..p4
+        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     xmm6,       xmm5              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6              ; accumulate thresholds
+
+
+        movdqa      xmm5,       xmm4              ; mm5 = p-2..p5
+        punpcklbw   xmm5,       xmm0              ; mm5 = p-2..p1
+        paddusw     xmm3,       xmm5              ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
+        psubusw     xmm6,       xmm5              ; mm6 = p0..p3 - p1..p4
+        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     xmm6,       xmm5              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6              ; accumulate thresholds
+
+        psrldq      xmm4,       1                   ; mm4 = p-1..p5
+        punpcklbw   xmm4,       xmm0              ; mm4 = p-1..p2
+        paddusw     xmm3,       xmm4              ; mm3 += mm5
+
+        ; thresholding
+        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
+        psubusw     xmm6,       xmm4              ; mm6 = p0..p3 - p1..p4
+        psubusw     xmm4,       xmm1              ; mm5 = p1..p4 - p0..p3
+        paddusw     xmm6,       xmm4              ; mm6 = abs(p0..p3 - p1..p4)
+        pcmpgtw     xmm6,       xmm2
+        por         xmm7,       xmm6              ; accumulate thresholds
+
+        paddusw     xmm3,       RD42              ; mm3 += round value
+        psraw       xmm3,       3                 ; mm3 /= 8
+
+        pand        xmm1,       xmm7              ; mm1 select vals > thresh from source
+        pandn       xmm7,       xmm3              ; mm7 select vals < thresh from blurred result
+        paddusw     xmm1,       xmm7              ; combination
+
+        packuswb    xmm1,       xmm0              ; pack to bytes
+        movq        QWORD PTR [rdi+rdx-8],  mm0   ; store previous four bytes
+        movdq2q     mm0,        xmm1
+
+        add         rdx,        8
+        cmp         edx,        dword arg(5) ;cols
+        jl          .acrossnextcol;
+
+        ; last 8 pixels
+        movq        QWORD PTR [rdi+rdx-8],  mm0
+
+        ; done with this rwo
+        add         rsi,rax               ; next line
+        mov         eax, dword arg(3) ;dst_pixels_per_line ; destination pitch?
+        add         rdi,rax               ; next destination
+        mov         eax, dword arg(2) ;src_pixels_per_line ; destination pitch?
+
+        dec         rcx                   ; decrement count
+        jnz         .nextrow              ; next row
+
+%if ABI_IS_32BIT=1 && CONFIG_PIC=1
+    add rsp,16
+    pop rsp
+%endif
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%undef RD42
+
+
+;void vp9_mbpost_proc_down_xmm(unsigned char *dst,
+;                            int pitch, int rows, int cols,int flimit)
+extern sym(vp9_rv)
+global sym(vp9_mbpost_proc_down_xmm) PRIVATE
+sym(vp9_mbpost_proc_down_xmm):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 128+16
+
+    ; unsigned char d[16][8] at [rsp]
+    ; create flimit2 at [rsp+128]
+    mov         eax, dword ptr arg(4) ;flimit
+    mov         [rsp+128], eax
+    mov         [rsp+128+4], eax
+    mov         [rsp+128+8], eax
+    mov         [rsp+128+12], eax
+%define flimit4 [rsp+128]
+
+%if ABI_IS_32BIT=0
+    lea         r8,       [GLOBAL(sym(vp9_rv))]
+%endif
+
+    ;rows +=8;
+    add         dword arg(2), 8
+
+    ;for(c=0; c<cols; c+=8)
+.loop_col:
+            mov         rsi,        arg(0) ; s
+            pxor        xmm0,       xmm0        ;
+
+            movsxd      rax,        dword ptr arg(1) ;pitch       ;
+            neg         rax                                     ; rax = -pitch
+
+            lea         rsi,        [rsi + rax*8];              ; rdi = s[-pitch*8]
+            neg         rax
+
+
+            pxor        xmm5,       xmm5
+            pxor        xmm6,       xmm6        ;
+
+            pxor        xmm7,       xmm7        ;
+            mov         rdi,        rsi
+
+            mov         rcx,        15          ;
+
+.loop_initvar:
+            movq        xmm1,       QWORD PTR [rdi];
+            punpcklbw   xmm1,       xmm0        ;
+
+            paddw       xmm5,       xmm1        ;
+            pmullw      xmm1,       xmm1        ;
+
+            movdqa      xmm2,       xmm1        ;
+            punpcklwd   xmm1,       xmm0        ;
+
+            punpckhwd   xmm2,       xmm0        ;
+            paddd       xmm6,       xmm1        ;
+
+            paddd       xmm7,       xmm2        ;
+            lea         rdi,        [rdi+rax]   ;
+
+            dec         rcx
+            jne         .loop_initvar
+            ;save the var and sum
+            xor         rdx,        rdx
+.loop_row:
+            movq        xmm1,       QWORD PTR [rsi]     ; [s-pitch*8]
+            movq        xmm2,       QWORD PTR [rdi]     ; [s+pitch*7]
+
+            punpcklbw   xmm1,       xmm0
+            punpcklbw   xmm2,       xmm0
+
+            paddw       xmm5,       xmm2
+            psubw       xmm5,       xmm1
+
+            pmullw      xmm2,       xmm2
+            movdqa      xmm4,       xmm2
+
+            punpcklwd   xmm2,       xmm0
+            punpckhwd   xmm4,       xmm0
+
+            paddd       xmm6,       xmm2
+            paddd       xmm7,       xmm4
+
+            pmullw      xmm1,       xmm1
+            movdqa      xmm2,       xmm1
+
+            punpcklwd   xmm1,       xmm0
+            psubd       xmm6,       xmm1
+
+            punpckhwd   xmm2,       xmm0
+            psubd       xmm7,       xmm2
+
+
+            movdqa      xmm3,       xmm6
+            pslld       xmm3,       4
+
+            psubd       xmm3,       xmm6
+            movdqa      xmm1,       xmm5
+
+            movdqa      xmm4,       xmm5
+            pmullw      xmm1,       xmm1
+
+            pmulhw      xmm4,       xmm4
+            movdqa      xmm2,       xmm1
+
+            punpcklwd   xmm1,       xmm4
+            punpckhwd   xmm2,       xmm4
+
+            movdqa      xmm4,       xmm7
+            pslld       xmm4,       4
+
+            psubd       xmm4,       xmm7
+
+            psubd       xmm3,       xmm1
+            psubd       xmm4,       xmm2
+
+            psubd       xmm3,       flimit4
+            psubd       xmm4,       flimit4
+
+            psrad       xmm3,       31
+            psrad       xmm4,       31
+
+            packssdw    xmm3,       xmm4
+            packsswb    xmm3,       xmm0
+
+            movq        xmm1,       QWORD PTR [rsi+rax*8]
+
+            movq        xmm2,       xmm1
+            punpcklbw   xmm1,       xmm0
+
+            paddw       xmm1,       xmm5
+            mov         rcx,        rdx
+
+            and         rcx,        127
+%if ABI_IS_32BIT=1 && CONFIG_PIC=1
+            push        rax
+            lea         rax,        [GLOBAL(sym(vp9_rv))]
+            movdqu      xmm4,       [rax + rcx*2] ;vp9_rv[rcx*2]
+            pop         rax
+%elif ABI_IS_32BIT=0
+            movdqu      xmm4,       [r8 + rcx*2] ;vp9_rv[rcx*2]
+%else
+            movdqu      xmm4,       [sym(vp9_rv) + rcx*2]
+%endif
+
+            paddw       xmm1,       xmm4
+            ;paddw     xmm1,       eight8s
+            psraw       xmm1,       4
+
+            packuswb    xmm1,       xmm0
+            pand        xmm1,       xmm3
+
+            pandn       xmm3,       xmm2
+            por         xmm1,       xmm3
+
+            and         rcx,        15
+            movq        QWORD PTR   [rsp + rcx*8], xmm1 ;d[rcx*8]
+
+            mov         rcx,        rdx
+            sub         rcx,        8
+
+            and         rcx,        15
+            movq        mm0,        [rsp + rcx*8] ;d[rcx*8]
+
+            movq        [rsi],      mm0
+            lea         rsi,        [rsi+rax]
+
+            lea         rdi,        [rdi+rax]
+            add         rdx,        1
+
+            cmp         edx,        dword arg(2) ;rows
+            jl          .loop_row
+
+        add         dword arg(0), 8 ; s += 8
+        sub         dword arg(3), 8 ; cols -= 8
+        cmp         dword arg(3), 0
+        jg          .loop_col
+
+    add         rsp, 128+16
+    pop         rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%undef flimit4
+
+
+;void vp9_mbpost_proc_across_ip_xmm(unsigned char *src,
+;                                int pitch, int rows, int cols,int flimit)
+global sym(vp9_mbpost_proc_across_ip_xmm) PRIVATE
+sym(vp9_mbpost_proc_across_ip_xmm):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16
+
+    ; create flimit4 at [rsp]
+    mov         eax, dword ptr arg(4) ;flimit
+    mov         [rsp], eax
+    mov         [rsp+4], eax
+    mov         [rsp+8], eax
+    mov         [rsp+12], eax
+%define flimit4 [rsp]
+
+
+    ;for(r=0;r<rows;r++)
+.ip_row_loop:
+
+        xor         rdx,    rdx ;sumsq=0;
+        xor         rcx,    rcx ;sum=0;
+        mov         rsi,    arg(0); s
+        mov         rdi,    -8
+.ip_var_loop:
+        ;for(i=-8;i<=6;i++)
+        ;{
+        ;    sumsq += s[i]*s[i];
+        ;    sum   += s[i];
+        ;}
+        movzx       eax, byte [rsi+rdi]
+        add         ecx, eax
+        mul         al
+        add         edx, eax
+        add         rdi, 1
+        cmp         rdi, 6
+        jle         .ip_var_loop
+
+
+            ;mov         rax,    sumsq
+            ;movd        xmm7,   rax
+            movd        xmm7,   edx
+
+            ;mov         rax,    sum
+            ;movd        xmm6,   rax
+            movd        xmm6,   ecx
+
+            mov         rsi,    arg(0) ;s
+            xor         rcx,    rcx
+
+            movsxd      rdx,    dword arg(3) ;cols
+            add         rdx,    8
+            pxor        mm0,    mm0
+            pxor        mm1,    mm1
+
+            pxor        xmm0,   xmm0
+.nextcol4:
+
+            movd        xmm1,   DWORD PTR [rsi+rcx-8]   ; -8 -7 -6 -5
+            movd        xmm2,   DWORD PTR [rsi+rcx+7]   ; +7 +8 +9 +10
+
+            punpcklbw   xmm1,   xmm0                    ; expanding
+            punpcklbw   xmm2,   xmm0                    ; expanding
+
+            punpcklwd   xmm1,   xmm0                    ; expanding to dwords
+            punpcklwd   xmm2,   xmm0                    ; expanding to dwords
+
+            psubd       xmm2,   xmm1                    ; 7--8   8--7   9--6 10--5
+            paddd       xmm1,   xmm1                    ; -8*2   -7*2   -6*2 -5*2
+
+            paddd       xmm1,   xmm2                    ; 7+-8   8+-7   9+-6 10+-5
+            pmaddwd     xmm1,   xmm2                    ; squared of 7+-8   8+-7   9+-6 10+-5
+
+            paddd       xmm6,   xmm2
+            paddd       xmm7,   xmm1
+
+            pshufd      xmm6,   xmm6,   0               ; duplicate the last ones
+            pshufd      xmm7,   xmm7,   0               ; duplicate the last ones
+
+            psrldq      xmm1,       4                   ; 8--7   9--6 10--5  0000
+            psrldq      xmm2,       4                   ; 8--7   9--6 10--5  0000
+
+            pshufd      xmm3,   xmm1,   3               ; 0000  8--7   8--7   8--7 squared
+            pshufd      xmm4,   xmm2,   3               ; 0000  8--7   8--7   8--7 squared
+
+            paddd       xmm6,   xmm4
+            paddd       xmm7,   xmm3
+
+            pshufd      xmm3,   xmm1,   01011111b       ; 0000  0000   9--6   9--6 squared
+            pshufd      xmm4,   xmm2,   01011111b       ; 0000  0000   9--6   9--6 squared
+
+            paddd       xmm7,   xmm3
+            paddd       xmm6,   xmm4
+
+            pshufd      xmm3,   xmm1,   10111111b       ; 0000  0000   8--7   8--7 squared
+            pshufd      xmm4,   xmm2,   10111111b       ; 0000  0000   8--7   8--7 squared
+
+            paddd       xmm7,   xmm3
+            paddd       xmm6,   xmm4
+
+            movdqa      xmm3,   xmm6
+            pmaddwd     xmm3,   xmm3
+
+            movdqa      xmm5,   xmm7
+            pslld       xmm5,   4
+
+            psubd       xmm5,   xmm7
+            psubd       xmm5,   xmm3
+
+            psubd       xmm5,   flimit4
+            psrad       xmm5,   31
+
+            packssdw    xmm5,   xmm0
+            packsswb    xmm5,   xmm0
+
+            movd        xmm1,   DWORD PTR [rsi+rcx]
+            movq        xmm2,   xmm1
+
+            punpcklbw   xmm1,   xmm0
+            punpcklwd   xmm1,   xmm0
+
+            paddd       xmm1,   xmm6
+            paddd       xmm1,   [GLOBAL(four8s)]
+
+            psrad       xmm1,   4
+            packssdw    xmm1,   xmm0
+
+            packuswb    xmm1,   xmm0
+            pand        xmm1,   xmm5
+
+            pandn       xmm5,   xmm2
+            por         xmm5,   xmm1
+
+            movd        [rsi+rcx-8],  mm0
+            movq        mm0,    mm1
+
+            movdq2q     mm1,    xmm5
+            psrldq      xmm7,   12
+
+            psrldq      xmm6,   12
+            add         rcx,    4
+
+            cmp         rcx,    rdx
+            jl          .nextcol4
+
+        ;s+=pitch;
+        movsxd rax, dword arg(1)
+        add    arg(0), rax
+
+        sub dword arg(2), 1 ;rows-=1
+        cmp dword arg(2), 0
+        jg .ip_row_loop
+
+    add         rsp, 16
+    pop         rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%undef flimit4
+
+
+;void vp9_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
+;                            unsigned char blackclamp[16],
+;                            unsigned char whiteclamp[16],
+;                            unsigned char bothclamp[16],
+;                            unsigned int width, unsigned int height, int pitch)
+extern sym(rand)
+global sym(vp9_plane_add_noise_wmt) PRIVATE
+sym(vp9_plane_add_noise_wmt):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 8
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+.addnoise_loop:
+    call sym(rand) WRT_PLT
+    mov     rcx, arg(1) ;noise
+    and     rax, 0xff
+    add     rcx, rax
+
+    ; we rely on the fact that the clamping vectors are stored contiguously
+    ; in black/white/both order. Note that we have to reload this here because
+    ; rdx could be trashed by rand()
+    mov     rdx, arg(2) ; blackclamp
+
+
+            mov     rdi, rcx
+            movsxd  rcx, dword arg(5) ;[Width]
+            mov     rsi, arg(0) ;Pos
+            xor         rax,rax
+
+.addnoise_nextset:
+            movdqu      xmm1,[rsi+rax]         ; get the source
+
+            psubusb     xmm1, [rdx]    ;blackclamp        ; clamp both sides so we don't outrange adding noise
+            paddusb     xmm1, [rdx+32] ;bothclamp
+            psubusb     xmm1, [rdx+16] ;whiteclamp
+
+            movdqu      xmm2,[rdi+rax]         ; get the noise for this line
+            paddb       xmm1,xmm2              ; add it in
+            movdqu      [rsi+rax],xmm1         ; store the result
+
+            add         rax,16                 ; move to the next line
+
+            cmp         rax, rcx
+            jl          .addnoise_nextset
+
+    movsxd  rax, dword arg(7) ; Pitch
+    add     arg(0), rax ; Start += Pitch
+    sub     dword arg(6), 1   ; Height -= 1
+    jg      .addnoise_loop
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+SECTION_RODATA
+align 16
+rd42:
+    times 8 dw 0x04
+four8s:
+    times 4 dd 8
diff --git a/libvpx/vp9/common/x86/vp9_postproc_x86.h b/libvpx/vp9/common/x86/vp9_postproc_x86.h
new file mode 100644
index 0000000..b0e8b18
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_postproc_x86.h
@@ -0,0 +1,64 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_COMMON_X86_VP9_POSTPROC_X86_H_
+#define VP9_COMMON_X86_VP9_POSTPROC_X86_H_
+
+/* Note:
+ *
+ * This platform is commonly built for runtime CPU detection. If you modify
+ * any of the function mappings present in this file, be sure to also update
+ * them in the function pointer initialization code
+ */
+
+#if HAVE_MMX
+extern prototype_postproc_inplace(vp9_mbpost_proc_down_mmx);
+extern prototype_postproc(vp9_post_proc_down_and_across_mmx);
+extern prototype_postproc_addnoise(vp9_plane_add_noise_mmx);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+#undef  vp9_postproc_down
+#define vp9_postproc_down vp9_mbpost_proc_down_mmx
+
+#undef  vp9_postproc_downacross
+#define vp9_postproc_downacross vp9_post_proc_down_and_across_mmx
+
+#undef  vp9_postproc_addnoise
+#define vp9_postproc_addnoise vp9_plane_add_noise_mmx
+
+#endif
+#endif
+
+
+#if HAVE_SSE2
+extern prototype_postproc_inplace(vp9_mbpost_proc_down_xmm);
+extern prototype_postproc_inplace(vp9_mbpost_proc_across_ip_xmm);
+extern prototype_postproc(vp9_post_proc_down_and_across_xmm);
+extern prototype_postproc_addnoise(vp9_plane_add_noise_wmt);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+#undef  vp9_postproc_down
+#define vp9_postproc_down vp9_mbpost_proc_down_xmm
+
+#undef  vp9_postproc_across
+#define vp9_postproc_across vp9_mbpost_proc_across_ip_xmm
+
+#undef  vp9_postproc_downacross
+#define vp9_postproc_downacross vp9_post_proc_down_and_across_xmm
+
+#undef  vp9_postproc_addnoise
+#define vp9_postproc_addnoise vp9_plane_add_noise_wmt
+
+
+#endif
+#endif
+
+#endif
diff --git a/libvpx/vp9/common/x86/vp9_recon_mmx.asm b/libvpx/vp9/common/x86/vp9_recon_mmx.asm
new file mode 100644
index 0000000..6fbbe48
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_recon_mmx.asm
@@ -0,0 +1,272 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+;void copy_mem8x8_mmx(
+;    unsigned char *src,
+;    int src_stride,
+;    unsigned char *dst,
+;    int dst_stride
+;    )
+global sym(vp9_copy_mem8x8_mmx) PRIVATE
+sym(vp9_copy_mem8x8_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov         rsi,        arg(0) ;src;
+        movq        mm0,        [rsi]
+
+        movsxd      rax,        dword ptr arg(1) ;src_stride;
+        mov         rdi,        arg(2) ;dst;
+
+        movq        mm1,        [rsi+rax]
+        movq        mm2,        [rsi+rax*2]
+
+        movsxd      rcx,        dword ptr arg(3) ;dst_stride
+        lea         rsi,        [rsi+rax*2]
+
+        movq        [rdi],      mm0
+        add         rsi,        rax
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx*2],    mm2
+
+
+        lea         rdi,        [rdi+rcx*2]
+        movq        mm3,        [rsi]
+
+        add         rdi,        rcx
+        movq        mm4,        [rsi+rax]
+
+        movq        mm5,        [rsi+rax*2]
+        movq        [rdi],      mm3
+
+        lea         rsi,        [rsi+rax*2]
+        movq        [rdi+rcx],  mm4
+
+        movq        [rdi+rcx*2],    mm5
+        lea         rdi,        [rdi+rcx*2]
+
+        movq        mm0,        [rsi+rax]
+        movq        mm1,        [rsi+rax*2]
+
+        movq        [rdi+rcx],  mm0
+        movq        [rdi+rcx*2],mm1
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void copy_mem8x4_mmx(
+;    unsigned char *src,
+;    int src_stride,
+;    unsigned char *dst,
+;    int dst_stride
+;    )
+global sym(vp9_copy_mem8x4_mmx) PRIVATE
+sym(vp9_copy_mem8x4_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov         rsi,        arg(0) ;src;
+        movq        mm0,        [rsi]
+
+        movsxd      rax,        dword ptr arg(1) ;src_stride;
+        mov         rdi,        arg(2) ;dst;
+
+        movq        mm1,        [rsi+rax]
+        movq        mm2,        [rsi+rax*2]
+
+        movsxd      rcx,        dword ptr arg(3) ;dst_stride
+        lea         rsi,        [rsi+rax*2]
+
+        movq        [rdi],      mm0
+        movq        [rdi+rcx],      mm1
+
+        movq        [rdi+rcx*2],    mm2
+        lea         rdi,        [rdi+rcx*2]
+
+        movq        mm3,        [rsi+rax]
+        movq        [rdi+rcx],      mm3
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void copy_mem16x16_mmx(
+;    unsigned char *src,
+;    int src_stride,
+;    unsigned char *dst,
+;    int dst_stride
+;    )
+global sym(vp9_copy_mem16x16_mmx) PRIVATE
+sym(vp9_copy_mem16x16_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov         rsi,        arg(0) ;src;
+        movsxd      rax,        dword ptr arg(1) ;src_stride;
+
+        mov         rdi,        arg(2) ;dst;
+        movsxd      rcx,        dword ptr arg(3) ;dst_stride
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        mm1,            [rsi+rax]
+        movq        mm4,            [rsi+rax+8]
+
+        movq        mm2,            [rsi+rax*2]
+        movq        mm5,            [rsi+rax*2+8]
+
+        lea         rsi,            [rsi+rax*2]
+        add         rsi,            rax
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx+8],    mm4
+
+        movq        [rdi+rcx*2],    mm2
+        movq        [rdi+rcx*2+8],  mm5
+
+        lea         rdi,            [rdi+rcx*2]
+        add         rdi,            rcx
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        mm1,            [rsi+rax]
+        movq        mm4,            [rsi+rax+8]
+
+        movq        mm2,            [rsi+rax*2]
+        movq        mm5,            [rsi+rax*2+8]
+
+        lea         rsi,            [rsi+rax*2]
+        add         rsi,            rax
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx+8],    mm4
+
+        movq        [rdi+rcx*2],    mm2
+        movq        [rdi+rcx*2+8],  mm5
+
+        lea         rdi,            [rdi+rcx*2]
+        add         rdi,            rcx
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        mm1,            [rsi+rax]
+        movq        mm4,            [rsi+rax+8]
+
+        movq        mm2,            [rsi+rax*2]
+        movq        mm5,            [rsi+rax*2+8]
+
+        lea         rsi,            [rsi+rax*2]
+        add         rsi,            rax
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx+8],    mm4
+
+        movq        [rdi+rcx*2],    mm2
+        movq        [rdi+rcx*2+8],  mm5
+
+        lea         rdi,            [rdi+rcx*2]
+        add         rdi,            rcx
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        mm1,            [rsi+rax]
+        movq        mm4,            [rsi+rax+8]
+
+        movq        mm2,            [rsi+rax*2]
+        movq        mm5,            [rsi+rax*2+8]
+
+        lea         rsi,            [rsi+rax*2]
+        add         rsi,            rax
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx+8],    mm4
+
+        movq        [rdi+rcx*2],    mm2
+        movq        [rdi+rcx*2+8],  mm5
+
+        lea         rdi,            [rdi+rcx*2]
+        add         rdi,            rcx
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        mm1,            [rsi+rax]
+        movq        mm4,            [rsi+rax+8]
+
+        movq        mm2,            [rsi+rax*2]
+        movq        mm5,            [rsi+rax*2+8]
+
+        lea         rsi,            [rsi+rax*2]
+        add         rsi,            rax
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+        movq        [rdi+rcx],      mm1
+        movq        [rdi+rcx+8],    mm4
+
+        movq        [rdi+rcx*2],    mm2
+        movq        [rdi+rcx*2+8],  mm5
+
+        lea         rdi,            [rdi+rcx*2]
+        add         rdi,            rcx
+
+        movq        mm0,            [rsi]
+        movq        mm3,            [rsi+8];
+
+        movq        [rdi],          mm0
+        movq        [rdi+8],        mm3
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/common/x86/vp9_recon_sse2.asm b/libvpx/vp9/common/x86/vp9_recon_sse2.asm
new file mode 100644
index 0000000..9ee3043
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_recon_sse2.asm
@@ -0,0 +1,572 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+;void copy_mem16x16_sse2(
+;    unsigned char *src,
+;    int src_stride,
+;    unsigned char *dst,
+;    int dst_stride
+;    )
+global sym(vp9_copy_mem16x16_sse2) PRIVATE
+sym(vp9_copy_mem16x16_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov         rsi,        arg(0) ;src;
+        movdqu      xmm0,       [rsi]
+
+        movsxd      rax,        dword ptr arg(1) ;src_stride;
+        mov         rdi,        arg(2) ;dst;
+
+        movdqu      xmm1,       [rsi+rax]
+        movdqu      xmm2,       [rsi+rax*2]
+
+        movsxd      rcx,        dword ptr arg(3) ;dst_stride
+        lea         rsi,        [rsi+rax*2]
+
+        movdqa      [rdi],      xmm0
+        add         rsi,        rax
+
+        movdqa      [rdi+rcx],  xmm1
+        movdqa      [rdi+rcx*2],xmm2
+
+        lea         rdi,        [rdi+rcx*2]
+        movdqu      xmm3,       [rsi]
+
+        add         rdi,        rcx
+        movdqu      xmm4,       [rsi+rax]
+
+        movdqu      xmm5,       [rsi+rax*2]
+        lea         rsi,        [rsi+rax*2]
+
+        movdqa      [rdi],  xmm3
+        add         rsi,        rax
+
+        movdqa      [rdi+rcx],  xmm4
+        movdqa      [rdi+rcx*2],xmm5
+
+        lea         rdi,        [rdi+rcx*2]
+        movdqu      xmm0,       [rsi]
+
+        add         rdi,        rcx
+        movdqu      xmm1,       [rsi+rax]
+
+        movdqu      xmm2,       [rsi+rax*2]
+        lea         rsi,        [rsi+rax*2]
+
+        movdqa      [rdi],      xmm0
+        add         rsi,        rax
+
+        movdqa      [rdi+rcx],  xmm1
+
+        movdqa      [rdi+rcx*2],    xmm2
+        movdqu      xmm3,       [rsi]
+
+        movdqu      xmm4,       [rsi+rax]
+        lea         rdi,        [rdi+rcx*2]
+
+        add         rdi,        rcx
+        movdqu      xmm5,       [rsi+rax*2]
+
+        lea         rsi,        [rsi+rax*2]
+        movdqa      [rdi],  xmm3
+
+        add         rsi,        rax
+        movdqa      [rdi+rcx],  xmm4
+
+        movdqa      [rdi+rcx*2],xmm5
+        movdqu      xmm0,       [rsi]
+
+        lea         rdi,        [rdi+rcx*2]
+        movdqu      xmm1,       [rsi+rax]
+
+        add         rdi,        rcx
+        movdqu      xmm2,       [rsi+rax*2]
+
+        lea         rsi,        [rsi+rax*2]
+        movdqa      [rdi],      xmm0
+
+        movdqa      [rdi+rcx],  xmm1
+        movdqa      [rdi+rcx*2],xmm2
+
+        movdqu      xmm3,       [rsi+rax]
+        lea         rdi,        [rdi+rcx*2]
+
+        movdqa      [rdi+rcx],  xmm3
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_intra_pred_uv_dc_mmx2(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+global sym(vp9_intra_pred_uv_dc_mmx2) PRIVATE
+sym(vp9_intra_pred_uv_dc_mmx2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ; from top
+    mov         rsi,        arg(2) ;src;
+    movsxd      rax,        dword ptr arg(3) ;src_stride;
+    sub         rsi,        rax
+    pxor        mm0,        mm0
+    movq        mm1,        [rsi]
+    psadbw      mm1,        mm0
+
+    ; from left
+    dec         rsi
+    lea         rdi,        [rax*3]
+    movzx       ecx,        byte [rsi+rax]
+    movzx       edx,        byte [rsi+rax*2]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rdi]
+    add         ecx,        edx
+    lea         rsi,        [rsi+rax*4]
+    movzx       edx,        byte [rsi]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax*2]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rdi]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax*4]
+    add         ecx,        edx
+
+    ; add up
+    pextrw      edx,        mm1, 0x0
+    lea         edx,        [edx+ecx+8]
+    sar         edx,        4
+    movd        mm1,        edx
+    pshufw      mm1,        mm1, 0x0
+    packuswb    mm1,        mm1
+
+    ; write out
+    mov         rdi,        arg(0) ;dst;
+    movsxd      rcx,        dword ptr arg(1) ;dst_stride
+    lea         rax,        [rcx*3]
+
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+    lea         rdi,        [rdi+rcx*4]
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_intra_pred_uv_dctop_mmx2(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+global sym(vp9_intra_pred_uv_dctop_mmx2) PRIVATE
+sym(vp9_intra_pred_uv_dctop_mmx2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ; from top
+    mov         rsi,        arg(2) ;src;
+    movsxd      rax,        dword ptr arg(3) ;src_stride;
+    sub         rsi,        rax
+    pxor        mm0,        mm0
+    movq        mm1,        [rsi]
+    psadbw      mm1,        mm0
+
+    ; add up
+    paddw       mm1,        [GLOBAL(dc_4)]
+    psraw       mm1,        3
+    pshufw      mm1,        mm1, 0x0
+    packuswb    mm1,        mm1
+
+    ; write out
+    mov         rdi,        arg(0) ;dst;
+    movsxd      rcx,        dword ptr arg(1) ;dst_stride
+    lea         rax,        [rcx*3]
+
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+    lea         rdi,        [rdi+rcx*4]
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_intra_pred_uv_dcleft_mmx2(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+global sym(vp9_intra_pred_uv_dcleft_mmx2) PRIVATE
+sym(vp9_intra_pred_uv_dcleft_mmx2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ; from left
+    mov         rsi,        arg(2) ;src;
+    movsxd      rax,        dword ptr arg(3) ;src_stride;
+    dec         rsi
+    lea         rdi,        [rax*3]
+    movzx       ecx,        byte [rsi]
+    movzx       edx,        byte [rsi+rax]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax*2]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rdi]
+    add         ecx,        edx
+    lea         rsi,        [rsi+rax*4]
+    movzx       edx,        byte [rsi]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rax*2]
+    add         ecx,        edx
+    movzx       edx,        byte [rsi+rdi]
+    lea         edx,        [ecx+edx+4]
+
+    ; add up
+    shr         edx,        3
+    movd        mm1,        edx
+    pshufw      mm1,        mm1, 0x0
+    packuswb    mm1,        mm1
+
+    ; write out
+    mov         rdi,        arg(0) ;dst;
+    movsxd      rcx,        dword ptr arg(1) ;dst_stride
+    lea         rax,        [rcx*3]
+
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+    lea         rdi,        [rdi+rcx*4]
+    movq [rdi      ],       mm1
+    movq [rdi+rcx  ],       mm1
+    movq [rdi+rcx*2],       mm1
+    movq [rdi+rax  ],       mm1
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_intra_pred_uv_dc128_mmx(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+global sym(vp9_intra_pred_uv_dc128_mmx) PRIVATE
+sym(vp9_intra_pred_uv_dc128_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    GET_GOT     rbx
+    ; end prolog
+
+    ; write out
+    movq        mm1,        [GLOBAL(dc_128)]
+    mov         rax,        arg(0) ;dst;
+    movsxd      rdx,        dword ptr arg(1) ;dst_stride
+    lea         rcx,        [rdx*3]
+
+    movq [rax      ],       mm1
+    movq [rax+rdx  ],       mm1
+    movq [rax+rdx*2],       mm1
+    movq [rax+rcx  ],       mm1
+    lea         rax,        [rax+rdx*4]
+    movq [rax      ],       mm1
+    movq [rax+rdx  ],       mm1
+    movq [rax+rdx*2],       mm1
+    movq [rax+rcx  ],       mm1
+
+    ; begin epilog
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_intra_pred_uv_tm_sse2(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+%macro vp9_intra_pred_uv_tm 1
+global sym(vp9_intra_pred_uv_tm_%1) PRIVATE
+sym(vp9_intra_pred_uv_tm_%1):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ; read top row
+    mov         edx,        4
+    mov         rsi,        arg(2) ;src;
+    movsxd      rax,        dword ptr arg(3) ;src_stride;
+    sub         rsi,        rax
+    pxor        xmm0,       xmm0
+%ifidn %1, ssse3
+    movdqa      xmm2,       [GLOBAL(dc_1024)]
+%endif
+    movq        xmm1,       [rsi]
+    punpcklbw   xmm1,       xmm0
+
+    ; set up left ptrs ans subtract topleft
+    movd        xmm3,       [rsi-1]
+    lea         rsi,        [rsi+rax-1]
+%ifidn %1, sse2
+    punpcklbw   xmm3,       xmm0
+    pshuflw     xmm3,       xmm3, 0x0
+    punpcklqdq  xmm3,       xmm3
+%else
+    pshufb      xmm3,       xmm2
+%endif
+    psubw       xmm1,       xmm3
+
+    ; set up dest ptrs
+    mov         rdi,        arg(0) ;dst;
+    movsxd      rcx,        dword ptr arg(1) ;dst_stride
+
+.vp9_intra_pred_uv_tm_%1_loop:
+    movd        xmm3,       [rsi]
+    movd        xmm5,       [rsi+rax]
+%ifidn %1, sse2
+    punpcklbw   xmm3,       xmm0
+    punpcklbw   xmm5,       xmm0
+    pshuflw     xmm3,       xmm3, 0x0
+    pshuflw     xmm5,       xmm5, 0x0
+    punpcklqdq  xmm3,       xmm3
+    punpcklqdq  xmm5,       xmm5
+%else
+    pshufb      xmm3,       xmm2
+    pshufb      xmm5,       xmm2
+%endif
+    paddw       xmm3,       xmm1
+    paddw       xmm5,       xmm1
+    packuswb    xmm3,       xmm5
+    movq  [rdi    ],        xmm3
+    movhps[rdi+rcx],        xmm3
+    lea         rsi,        [rsi+rax*2]
+    lea         rdi,        [rdi+rcx*2]
+    dec         edx
+    jnz .vp9_intra_pred_uv_tm_%1_loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endmacro
+
+vp9_intra_pred_uv_tm sse2
+vp9_intra_pred_uv_tm ssse3
+
+;void vp9_intra_pred_uv_ve_mmx(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+global sym(vp9_intra_pred_uv_ve_mmx) PRIVATE
+sym(vp9_intra_pred_uv_ve_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    ; end prolog
+
+    ; read from top
+    mov         rax,        arg(2) ;src;
+    movsxd      rdx,        dword ptr arg(3) ;src_stride;
+    sub         rax,        rdx
+    movq        mm1,        [rax]
+
+    ; write out
+    mov         rax,        arg(0) ;dst;
+    movsxd      rdx,        dword ptr arg(1) ;dst_stride
+    lea         rcx,        [rdx*3]
+
+    movq [rax      ],       mm1
+    movq [rax+rdx  ],       mm1
+    movq [rax+rdx*2],       mm1
+    movq [rax+rcx  ],       mm1
+    lea         rax,        [rax+rdx*4]
+    movq [rax      ],       mm1
+    movq [rax+rdx  ],       mm1
+    movq [rax+rdx*2],       mm1
+    movq [rax+rcx  ],       mm1
+
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_intra_pred_uv_ho_mmx2(
+;    unsigned char *dst,
+;    int dst_stride
+;    unsigned char *src,
+;    int src_stride,
+;    )
+%macro vp9_intra_pred_uv_ho 1
+global sym(vp9_intra_pred_uv_ho_%1) PRIVATE
+sym(vp9_intra_pred_uv_ho_%1):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push        rsi
+    push        rdi
+%ifidn %1, ssse3
+%ifndef GET_GOT_SAVE_ARG
+    push        rbx
+%endif
+    GET_GOT     rbx
+%endif
+    ; end prolog
+
+    ; read from left and write out
+%ifidn %1, mmx2
+    mov         edx,        4
+%endif
+    mov         rsi,        arg(2) ;src;
+    movsxd      rax,        dword ptr arg(3) ;src_stride;
+    mov         rdi,        arg(0) ;dst;
+    movsxd      rcx,        dword ptr arg(1) ;dst_stride
+%ifidn %1, ssse3
+    lea         rdx,        [rcx*3]
+    movdqa      xmm2,       [GLOBAL(dc_00001111)]
+    lea         rbx,        [rax*3]
+%endif
+    dec         rsi
+%ifidn %1, mmx2
+.vp9_intra_pred_uv_ho_%1_loop:
+    movd        mm0,        [rsi]
+    movd        mm1,        [rsi+rax]
+    punpcklbw   mm0,        mm0
+    punpcklbw   mm1,        mm1
+    pshufw      mm0,        mm0, 0x0
+    pshufw      mm1,        mm1, 0x0
+    movq  [rdi    ],        mm0
+    movq  [rdi+rcx],        mm1
+    lea         rsi,        [rsi+rax*2]
+    lea         rdi,        [rdi+rcx*2]
+    dec         edx
+    jnz .vp9_intra_pred_uv_ho_%1_loop
+%else
+    movd        xmm0,       [rsi]
+    movd        xmm3,       [rsi+rax]
+    movd        xmm1,       [rsi+rax*2]
+    movd        xmm4,       [rsi+rbx]
+    punpcklbw   xmm0,       xmm3
+    punpcklbw   xmm1,       xmm4
+    pshufb      xmm0,       xmm2
+    pshufb      xmm1,       xmm2
+    movq   [rdi    ],       xmm0
+    movhps [rdi+rcx],       xmm0
+    movq [rdi+rcx*2],       xmm1
+    movhps [rdi+rdx],       xmm1
+    lea         rsi,        [rsi+rax*4]
+    lea         rdi,        [rdi+rcx*4]
+    movd        xmm0,       [rsi]
+    movd        xmm3,       [rsi+rax]
+    movd        xmm1,       [rsi+rax*2]
+    movd        xmm4,       [rsi+rbx]
+    punpcklbw   xmm0,       xmm3
+    punpcklbw   xmm1,       xmm4
+    pshufb      xmm0,       xmm2
+    pshufb      xmm1,       xmm2
+    movq   [rdi    ],       xmm0
+    movhps [rdi+rcx],       xmm0
+    movq [rdi+rcx*2],       xmm1
+    movhps [rdi+rdx],       xmm1
+%endif
+
+    ; begin epilog
+%ifidn %1, ssse3
+    RESTORE_GOT
+%ifndef GET_GOT_SAVE_ARG
+    pop         rbx
+%endif
+%endif
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endmacro
+
+vp9_intra_pred_uv_ho mmx2
+vp9_intra_pred_uv_ho ssse3
+
+SECTION_RODATA
+dc_128:
+    times 8 db 128
+dc_4:
+    times 4 dw 4
+align 16
+dc_1024:
+    times 8 dw 0x400
+align 16
+dc_00001111:
+    times 8 db 0
+    times 8 db 1
diff --git a/libvpx/vp9/common/x86/vp9_recon_wrapper_sse2.c b/libvpx/vp9/common/x86/vp9_recon_wrapper_sse2.c
new file mode 100644
index 0000000..97148fb
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_recon_wrapper_sse2.c
@@ -0,0 +1,101 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_blockd.h"
+
+#define build_intra_predictors_mbuv_prototype(sym) \
+  void sym(unsigned char *dst, int dst_stride, \
+           const unsigned char *src, int src_stride)
+typedef build_intra_predictors_mbuv_prototype((*build_intra_pred_mbuv_fn_t));
+
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dc_mmx2);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dctop_mmx2);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dcleft_mmx2);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_dc128_mmx);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ho_mmx2);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ho_ssse3);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_ve_mmx);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_tm_sse2);
+extern build_intra_predictors_mbuv_prototype(vp9_intra_pred_uv_tm_ssse3);
+
+static void build_intra_predictors_mbuv_x86(MACROBLOCKD *xd,
+                                            unsigned char *dst_u,
+                                            unsigned char *dst_v,
+                                            int dst_stride,
+                                            build_intra_pred_mbuv_fn_t tm_fn,
+                                            build_intra_pred_mbuv_fn_t ho_fn) {
+  int mode = xd->mode_info_context->mbmi.uv_mode;
+  build_intra_pred_mbuv_fn_t fn;
+  int src_stride = xd->plane[1].dst.stride;
+
+  switch (mode) {
+    case  V_PRED:
+      fn = vp9_intra_pred_uv_ve_mmx;
+      break;
+    case  H_PRED:
+      fn = ho_fn;
+      break;
+    case TM_PRED:
+      fn = tm_fn;
+      break;
+    case DC_PRED:
+      if (xd->up_available) {
+        if (xd->left_available) {
+          fn = vp9_intra_pred_uv_dc_mmx2;
+          break;
+        } else {
+          fn = vp9_intra_pred_uv_dctop_mmx2;
+          break;
+        }
+      } else if (xd->left_available) {
+        fn = vp9_intra_pred_uv_dcleft_mmx2;
+        break;
+      } else {
+        fn = vp9_intra_pred_uv_dc128_mmx;
+        break;
+      }
+      break;
+    default:
+      return;
+  }
+
+  fn(dst_u, dst_stride, xd->plane[1].dst.buf, src_stride);
+  fn(dst_v, dst_stride, xd->plane[2].dst.buf, src_stride);
+}
+
+void vp9_build_intra_predictors_mbuv_sse2(MACROBLOCKD *xd) {
+  build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
+                                  xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                                  vp9_intra_pred_uv_tm_sse2,
+                                  vp9_intra_pred_uv_ho_mmx2);
+}
+
+void vp9_build_intra_predictors_mbuv_ssse3(MACROBLOCKD *xd) {
+  build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
+                                  xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                                  vp9_intra_pred_uv_tm_ssse3,
+                                  vp9_intra_pred_uv_ho_ssse3);
+}
+
+void vp9_build_intra_predictors_mbuv_s_sse2(MACROBLOCKD *xd) {
+  build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
+                                  xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                                  vp9_intra_pred_uv_tm_sse2,
+                                  vp9_intra_pred_uv_ho_mmx2);
+}
+
+void vp9_build_intra_predictors_mbuv_s_ssse3(MACROBLOCKD *xd) {
+  build_intra_predictors_mbuv_x86(xd, xd->plane[1].dst.buf,
+                                  xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+                                  vp9_intra_pred_uv_tm_ssse3,
+                                  vp9_intra_pred_uv_ho_ssse3);
+}
diff --git a/libvpx/vp9/common/x86/vp9_sadmxn_sse2.c b/libvpx/vp9/common/x86/vp9_sadmxn_sse2.c
new file mode 100644
index 0000000..ed873a5
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_sadmxn_sse2.c
@@ -0,0 +1,95 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  /* SSE2 */
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/emmintrin_compat.h"
+
+unsigned int vp9_sad16x3_sse2(
+  const unsigned char *src_ptr,
+  int  src_stride,
+  const unsigned char *ref_ptr,
+  int  ref_stride) {
+  __m128i s0, s1, s2;
+  __m128i r0, r1, r2;
+  __m128i sad;
+
+  s0 = _mm_loadu_si128((const __m128i *)(src_ptr + 0 * src_stride));
+  s1 = _mm_loadu_si128((const __m128i *)(src_ptr + 1 * src_stride));
+  s2 = _mm_loadu_si128((const __m128i *)(src_ptr + 2 * src_stride));
+
+  r0 = _mm_loadu_si128((const __m128i *)(ref_ptr + 0 * ref_stride));
+  r1 = _mm_loadu_si128((const __m128i *)(ref_ptr + 1 * ref_stride));
+  r2 = _mm_loadu_si128((const __m128i *)(ref_ptr + 2 * ref_stride));
+
+  sad = _mm_sad_epu8(s0, r0);
+  sad = _mm_add_epi16(sad,  _mm_sad_epu8(s1, r1));
+  sad = _mm_add_epi16(sad,  _mm_sad_epu8(s2, r2));
+  sad = _mm_add_epi16(sad,  _mm_srli_si128(sad, 8));
+
+  return _mm_cvtsi128_si32(sad);
+}
+
+unsigned int vp9_sad3x16_sse2(
+  const unsigned char *src_ptr,
+  int  src_stride,
+  const unsigned char *ref_ptr,
+  int  ref_stride) {
+  int r;
+  __m128i s0, s1, s2, s3;
+  __m128i r0, r1, r2, r3;
+  __m128i sad = _mm_setzero_si128();
+  __m128i mask;
+  const int offset = (uintptr_t)src_ptr & 3;
+
+  /* In current use case, the offset is 1 if CONFIG_SUBPELREFMV is off.
+   * Here, for offset=1, we adjust src_ptr to be 4-byte aligned. Then, movd
+   * takes much less time.
+   */
+  if (offset == 1)
+    src_ptr -= 1;
+
+  /* mask = 0xffffffffffff0000ffffffffffff0000 */
+  mask = _mm_cmpeq_epi32(sad, sad);
+  mask = _mm_slli_epi64(mask, 16);
+
+  for (r = 0; r < 16; r += 4) {
+    s0 = _mm_cvtsi32_si128 (*(const int *)(src_ptr + 0 * src_stride));
+    s1 = _mm_cvtsi32_si128 (*(const int *)(src_ptr + 1 * src_stride));
+    s2 = _mm_cvtsi32_si128 (*(const int *)(src_ptr + 2 * src_stride));
+    s3 = _mm_cvtsi32_si128 (*(const int *)(src_ptr + 3 * src_stride));
+    r0 = _mm_cvtsi32_si128 (*(const int *)(ref_ptr + 0 * ref_stride));
+    r1 = _mm_cvtsi32_si128 (*(const int *)(ref_ptr + 1 * ref_stride));
+    r2 = _mm_cvtsi32_si128 (*(const int *)(ref_ptr + 2 * ref_stride));
+    r3 = _mm_cvtsi32_si128 (*(const int *)(ref_ptr + 3 * ref_stride));
+
+    s0 = _mm_unpacklo_epi8(s0, s1);
+    r0 = _mm_unpacklo_epi8(r0, r1);
+    s2 = _mm_unpacklo_epi8(s2, s3);
+    r2 = _mm_unpacklo_epi8(r2, r3);
+    s0 = _mm_unpacklo_epi64(s0, s2);
+    r0 = _mm_unpacklo_epi64(r0, r2);
+
+    // throw out extra byte
+    if (offset == 1)
+      s0 = _mm_and_si128(s0, mask);
+    else
+      s0 = _mm_slli_epi64(s0, 16);
+    r0 = _mm_slli_epi64(r0, 16);
+
+    sad = _mm_add_epi16(sad, _mm_sad_epu8(s0, r0));
+
+    src_ptr += src_stride*4;
+    ref_ptr += ref_stride*4;
+  }
+
+  sad = _mm_add_epi16(sad,  _mm_srli_si128(sad, 8));
+  return _mm_cvtsi128_si32(sad);
+}
diff --git a/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm b/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm
new file mode 100644
index 0000000..bbf9888
--- /dev/null
+++ b/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm
@@ -0,0 +1,1011 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;/************************************************************************************
+; Notes: filter_block1d_h6 applies a 6 tap filter horizontally to the input pixels. The
+; input pixel array has output_height rows. This routine assumes that output_height is an
+; even number. This function handles 8 pixels in horizontal direction, calculating ONE
+; rows each iteration to take advantage of the 128 bits operations.
+;
+; This is an implementation of some of the SSE optimizations first seen in ffvp8
+;
+;*************************************************************************************/
+
+
+%macro VERTx4 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movd        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rdx, DWORD PTR arg(1)       ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+    movsxd      r8, DWORD PTR arg(3)        ;out_pitch
+%endif
+    mov         rax, rsi
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+    add         rax, rdx
+
+    lea         rbx, [rdx + rdx*4]
+    add         rbx, rdx                    ;pitch * 6
+
+.loop:
+    movd        xmm0, [rsi]                 ;A
+    movd        xmm1, [rsi + rdx]           ;B
+    movd        xmm2, [rsi + rdx * 2]       ;C
+    movd        xmm3, [rax + rdx * 2]       ;D
+    movd        xmm4, [rsi + rdx * 4]       ;E
+    movd        xmm5, [rax + rdx * 4]       ;F
+
+    punpcklbw   xmm0, xmm1                  ;A B
+    punpcklbw   xmm2, xmm3                  ;C D
+    punpcklbw   xmm4, xmm5                  ;E F
+
+    movd        xmm6, [rsi + rbx]           ;G
+    movd        xmm7, [rax + rbx]           ;H
+
+    pmaddubsw   xmm0, k0k1
+    pmaddubsw   xmm2, k2k3
+    punpcklbw   xmm6, xmm7                  ;G H
+    pmaddubsw   xmm4, k4k5
+    pmaddubsw   xmm6, k6k7
+
+    paddsw      xmm0, xmm6
+    paddsw      xmm0, xmm2
+    paddsw      xmm0, xmm4
+    paddsw      xmm0, krd
+
+    psraw       xmm0, 7
+    packuswb    xmm0, xmm0
+
+    add         rsi,  rdx
+    add         rax,  rdx
+%if %1
+    movd        xmm1, [rdi]
+    pavgb       xmm0, xmm1
+%endif
+    movd        [rdi], xmm0
+
+%if ABI_IS_32BIT
+    add         rdi, DWORD PTR arg(3)       ;out_pitch
+%else
+    add         rdi, r8
+%endif
+    dec         rcx
+    jnz         .loop
+%endm
+
+%macro VERTx8 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movq        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rdx, DWORD PTR arg(1)       ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+    movsxd      r8, DWORD PTR arg(3)        ;out_pitch
+%endif
+    mov         rax, rsi
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+    add         rax, rdx
+
+    lea         rbx, [rdx + rdx*4]
+    add         rbx, rdx                    ;pitch * 6
+
+.loop:
+    movq        xmm0, [rsi]                 ;A
+    movq        xmm1, [rsi + rdx]           ;B
+    movq        xmm2, [rsi + rdx * 2]       ;C
+    movq        xmm3, [rax + rdx * 2]       ;D
+    movq        xmm4, [rsi + rdx * 4]       ;E
+    movq        xmm5, [rax + rdx * 4]       ;F
+
+    punpcklbw   xmm0, xmm1                  ;A B
+    punpcklbw   xmm2, xmm3                  ;C D
+    punpcklbw   xmm4, xmm5                  ;E F
+
+    movq        xmm6, [rsi + rbx]           ;G
+    movq        xmm7, [rax + rbx]           ;H
+
+    pmaddubsw   xmm0, k0k1
+    pmaddubsw   xmm2, k2k3
+    punpcklbw   xmm6, xmm7                  ;G H
+    pmaddubsw   xmm4, k4k5
+    pmaddubsw   xmm6, k6k7
+
+    paddsw      xmm0, xmm6
+    paddsw      xmm0, xmm2
+    paddsw      xmm0, xmm4
+    paddsw      xmm0, krd
+
+    psraw       xmm0, 7
+    packuswb    xmm0, xmm0
+
+    add         rsi,  rdx
+    add         rax,  rdx
+%if %1
+    movq        xmm1, [rdi]
+    pavgb       xmm0, xmm1
+%endif
+    movq        [rdi], xmm0
+
+%if ABI_IS_32BIT
+    add         rdi, DWORD PTR arg(3)       ;out_pitch
+%else
+    add         rdi, r8
+%endif
+    dec         rcx
+    jnz         .loop
+%endm
+
+
+%macro VERTx16 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movq        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rdx, DWORD PTR arg(1)       ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+    movsxd      r8, DWORD PTR arg(3)        ;out_pitch
+%endif
+    mov         rax, rsi
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+    add         rax, rdx
+
+    lea         rbx, [rdx + rdx*4]
+    add         rbx, rdx                    ;pitch * 6
+
+.loop:
+    movq        xmm0, [rsi]                 ;A
+    movq        xmm1, [rsi + rdx]           ;B
+    movq        xmm2, [rsi + rdx * 2]       ;C
+    movq        xmm3, [rax + rdx * 2]       ;D
+    movq        xmm4, [rsi + rdx * 4]       ;E
+    movq        xmm5, [rax + rdx * 4]       ;F
+
+    punpcklbw   xmm0, xmm1                  ;A B
+    punpcklbw   xmm2, xmm3                  ;C D
+    punpcklbw   xmm4, xmm5                  ;E F
+
+    movq        xmm6, [rsi + rbx]           ;G
+    movq        xmm7, [rax + rbx]           ;H
+
+    pmaddubsw   xmm0, k0k1
+    pmaddubsw   xmm2, k2k3
+    punpcklbw   xmm6, xmm7                  ;G H
+    pmaddubsw   xmm4, k4k5
+    pmaddubsw   xmm6, k6k7
+
+    paddsw      xmm0, xmm6
+    paddsw      xmm0, xmm2
+    paddsw      xmm0, xmm4
+    paddsw      xmm0, krd
+
+    psraw       xmm0, 7
+    packuswb    xmm0, xmm0
+%if %1
+    movq        xmm1, [rdi]
+    pavgb       xmm0, xmm1
+%endif
+    movq        [rdi], xmm0
+
+    movq        xmm0, [rsi + 8]             ;A
+    movq        xmm1, [rsi + rdx + 8]       ;B
+    movq        xmm2, [rsi + rdx * 2 + 8]   ;C
+    movq        xmm3, [rax + rdx * 2 + 8]   ;D
+    movq        xmm4, [rsi + rdx * 4 + 8]   ;E
+    movq        xmm5, [rax + rdx * 4 + 8]   ;F
+
+    punpcklbw   xmm0, xmm1                  ;A B
+    punpcklbw   xmm2, xmm3                  ;C D
+    punpcklbw   xmm4, xmm5                  ;E F
+
+
+    movq        xmm6, [rsi + rbx + 8]       ;G
+    movq        xmm7, [rax + rbx + 8]       ;H
+    punpcklbw   xmm6, xmm7                  ;G H
+
+
+    pmaddubsw   xmm0, k0k1
+    pmaddubsw   xmm2, k2k3
+    pmaddubsw   xmm4, k4k5
+    pmaddubsw   xmm6, k6k7
+
+    paddsw      xmm0, xmm6
+    paddsw      xmm0, xmm2
+    paddsw      xmm0, xmm4
+    paddsw      xmm0, krd
+
+    psraw       xmm0, 7
+    packuswb    xmm0, xmm0
+
+    add         rsi,  rdx
+    add         rax,  rdx
+%if %1
+    movq    xmm1, [rdi+8]
+    pavgb   xmm0, xmm1
+%endif
+
+    movq        [rdi+8], xmm0
+
+%if ABI_IS_32BIT
+    add         rdi, DWORD PTR arg(3)       ;out_pitch
+%else
+    add         rdi, r8
+%endif
+    dec         rcx
+    jnz         .loop
+%endm
+
+;void vp9_filter_block1d8_v8_ssse3
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d4_v8_ssse3) PRIVATE
+sym(vp9_filter_block1d4_v8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx4 0
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_v8_ssse3
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d8_v8_ssse3) PRIVATE
+sym(vp9_filter_block1d8_v8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx8 0
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_v8_ssse3
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d16_v8_ssse3) PRIVATE
+sym(vp9_filter_block1d16_v8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx16 0
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+
+global sym(vp9_filter_block1d4_v8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d4_v8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx4 1
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_filter_block1d8_v8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d8_v8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx8 1
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_filter_block1d16_v8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d16_v8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    VERTx16 1
+
+    add rsp, 16*5
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+%macro HORIZx4 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movq        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rax, dword ptr arg(1)       ;src_pixels_per_line
+    movsxd      rdx, dword ptr arg(3)       ;output_pitch
+    movsxd      rcx, dword ptr arg(4)       ;output_height
+
+.loop:
+    movq        xmm0,   [rsi - 3]    ; -3 -2 -1  0  1  2  3  4
+
+    movq        xmm3,   [rsi + 5]    ; 5  6  7  8  9 10 11 12
+    punpcklqdq  xmm0,   xmm3
+
+    movdqa      xmm1,   xmm0
+    pshufb      xmm0,   [GLOBAL(shuf_t0t1)]
+    pmaddubsw   xmm0,   k0k1
+
+    movdqa      xmm2,   xmm1
+    pshufb      xmm1,   [GLOBAL(shuf_t2t3)]
+    pmaddubsw   xmm1,   k2k3
+
+    movdqa      xmm4,   xmm2
+    pshufb      xmm2,   [GLOBAL(shuf_t4t5)]
+    pmaddubsw   xmm2,   k4k5
+
+    pshufb      xmm4,   [GLOBAL(shuf_t6t7)]
+    pmaddubsw   xmm4,   k6k7
+
+    paddsw      xmm0,   xmm1
+    paddsw      xmm0,   xmm4
+    paddsw      xmm0,   xmm2
+    paddsw      xmm0,   krd
+    psraw       xmm0,   7
+    packuswb    xmm0,   xmm0
+%if %1
+    movd        xmm1,   [rdi]
+    pavgb       xmm0,   xmm1
+%endif
+    lea         rsi,    [rsi + rax]
+    movd        [rdi],  xmm0
+
+    lea         rdi,    [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+%endm
+
+%macro HORIZx8 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movd        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rax, dword ptr arg(1)       ;src_pixels_per_line
+    movsxd      rdx, dword ptr arg(3)       ;output_pitch
+    movsxd      rcx, dword ptr arg(4)       ;output_height
+
+.loop:
+    movq        xmm0,   [rsi - 3]    ; -3 -2 -1  0  1  2  3  4
+
+    movq        xmm3,   [rsi + 5]    ; 5  6  7  8  9 10 11 12
+    punpcklqdq  xmm0,   xmm3
+
+    movdqa      xmm1,   xmm0
+    pshufb      xmm0,   [GLOBAL(shuf_t0t1)]
+    pmaddubsw   xmm0,   k0k1
+
+    movdqa      xmm2,   xmm1
+    pshufb      xmm1,   [GLOBAL(shuf_t2t3)]
+    pmaddubsw   xmm1,   k2k3
+
+    movdqa      xmm4,   xmm2
+    pshufb      xmm2,   [GLOBAL(shuf_t4t5)]
+    pmaddubsw   xmm2,   k4k5
+
+    pshufb      xmm4,   [GLOBAL(shuf_t6t7)]
+    pmaddubsw   xmm4,   k6k7
+
+    paddsw      xmm0,   xmm1
+    paddsw      xmm0,   xmm4
+    paddsw      xmm0,   xmm2
+    paddsw      xmm0,   krd
+    psraw       xmm0,   7
+    packuswb    xmm0,   xmm0
+%if %1
+    movq        xmm1,   [rdi]
+    pavgb       xmm0,   xmm1
+%endif
+
+    lea         rsi,    [rsi + rax]
+    movq        [rdi],  xmm0
+
+    lea         rdi,    [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+%endm
+
+%macro HORIZx16 1
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x0400040
+
+    movdqa      xmm4, [rdx]                 ;load filters
+    movq        xmm5, rcx
+    packsswb    xmm4, xmm4
+    pshuflw     xmm0, xmm4, 0b              ;k0_k1
+    pshuflw     xmm1, xmm4, 01010101b       ;k2_k3
+    pshuflw     xmm2, xmm4, 10101010b       ;k4_k5
+    pshuflw     xmm3, xmm4, 11111111b       ;k6_k7
+
+    punpcklqdq  xmm0, xmm0
+    punpcklqdq  xmm1, xmm1
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+
+    movdqa      k0k1, xmm0
+    movdqa      k2k3, xmm1
+    pshufd      xmm5, xmm5, 0
+    movdqa      k4k5, xmm2
+    movdqa      k6k7, xmm3
+    movdqa      krd, xmm5
+
+    movsxd      rax, dword ptr arg(1)       ;src_pixels_per_line
+    movsxd      rdx, dword ptr arg(3)       ;output_pitch
+    movsxd      rcx, dword ptr arg(4)       ;output_height
+
+.loop:
+    movq        xmm0,   [rsi - 3]    ; -3 -2 -1  0  1  2  3  4
+
+    movq        xmm3,   [rsi + 5]    ; 5  6  7  8  9 10 11 12
+    punpcklqdq  xmm0,   xmm3
+
+    movdqa      xmm1,   xmm0
+    pshufb      xmm0,   [GLOBAL(shuf_t0t1)]
+    pmaddubsw   xmm0,   k0k1
+
+    movdqa      xmm2,   xmm1
+    pshufb      xmm1,   [GLOBAL(shuf_t2t3)]
+    pmaddubsw   xmm1,   k2k3
+
+    movdqa      xmm4,   xmm2
+    pshufb      xmm2,   [GLOBAL(shuf_t4t5)]
+    pmaddubsw   xmm2,   k4k5
+
+    pshufb      xmm4,   [GLOBAL(shuf_t6t7)]
+    pmaddubsw   xmm4,   k6k7
+
+    paddsw      xmm0,   xmm1
+    paddsw      xmm0,   xmm4
+    paddsw      xmm0,   xmm2
+    paddsw      xmm0,   krd
+    psraw       xmm0,   7
+    packuswb    xmm0,   xmm0
+
+
+    movq        xmm3,   [rsi +  5]
+    movq        xmm7,   [rsi + 13]
+    punpcklqdq  xmm3,   xmm7
+
+    movdqa      xmm1,   xmm3
+    pshufb      xmm3,   [GLOBAL(shuf_t0t1)]
+    pmaddubsw   xmm3,   k0k1
+
+    movdqa      xmm2,   xmm1
+    pshufb      xmm1,   [GLOBAL(shuf_t2t3)]
+    pmaddubsw   xmm1,   k2k3
+
+    movdqa      xmm4,   xmm2
+    pshufb      xmm2,   [GLOBAL(shuf_t4t5)]
+    pmaddubsw   xmm2,   k4k5
+
+    pshufb      xmm4,   [GLOBAL(shuf_t6t7)]
+    pmaddubsw   xmm4,   k6k7
+
+    paddsw      xmm3,   xmm1
+    paddsw      xmm3,   xmm4
+    paddsw      xmm3,   xmm2
+    paddsw      xmm3,   krd
+    psraw       xmm3,   7
+    packuswb    xmm3,   xmm3
+    punpcklqdq  xmm0,   xmm3
+%if %1
+    movdqa      xmm1,   [rdi]
+    pavgb       xmm0,   xmm1
+%endif
+
+    lea         rsi,    [rsi + rax]
+    movdqa      [rdi],  xmm0
+
+    lea         rdi,    [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+%endm
+
+;void vp9_filter_block1d4_h8_ssse3
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d4_h8_ssse3) PRIVATE
+sym(vp9_filter_block1d4_h8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx4 0
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_h8_ssse3
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d8_h8_ssse3) PRIVATE
+sym(vp9_filter_block1d8_h8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx8 0
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_h8_ssse3
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_filter_block1d16_h8_ssse3) PRIVATE
+sym(vp9_filter_block1d16_h8_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx16 0
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_filter_block1d4_h8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d4_h8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx4 1
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_filter_block1d8_h8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d8_h8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx8 1
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_filter_block1d16_h8_avg_ssse3) PRIVATE
+sym(vp9_filter_block1d16_h8_avg_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16*5
+    %define k0k1 [rsp + 16*0]
+    %define k2k3 [rsp + 16*1]
+    %define k4k5 [rsp + 16*2]
+    %define k6k7 [rsp + 16*3]
+    %define krd [rsp + 16*4]
+
+    HORIZx16 1
+
+    add rsp, 16*5
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+SECTION_RODATA
+align 16
+shuf_t0t1:
+    db  0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+align 16
+shuf_t2t3:
+    db  2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
+align 16
+shuf_t4t5:
+    db  4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
+align 16
+shuf_t6t7:
+    db  6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
diff --git a/libvpx/vp9/decoder/vp9_asm_dec_offsets.c b/libvpx/vp9/decoder/vp9_asm_dec_offsets.c
new file mode 100644
index 0000000..e4b9c97
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_asm_dec_offsets.c
@@ -0,0 +1,20 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_ports/asm_offsets.h"
+#include "vp9/decoder/vp9_onyxd_int.h"
+
+BEGIN
+
+END
+
+/* add asserts for any offset that is not supported by assembly code */
+/* add asserts for any size that is not supported by assembly code */
diff --git a/libvpx/vp9/decoder/vp9_dboolhuff.c b/libvpx/vp9/decoder/vp9_dboolhuff.c
new file mode 100644
index 0000000..df77d65
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_dboolhuff.c
@@ -0,0 +1,69 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_ports/mem.h"
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/decoder/vp9_dboolhuff.h"
+
+int vp9_reader_init(vp9_reader *r, const uint8_t *buffer, size_t size) {
+  int marker_bit;
+
+  r->buffer_end = buffer + size;
+  r->buffer = buffer;
+  r->value = 0;
+  r->count = -8;
+  r->range = 255;
+
+  if (size && !buffer)
+    return 1;
+
+  vp9_reader_fill(r);
+  marker_bit = vp9_read_bit(r);
+  return marker_bit != 0;
+}
+
+void vp9_reader_fill(vp9_reader *r) {
+  const uint8_t *const buffer_end = r->buffer_end;
+  const uint8_t *buffer = r->buffer;
+  VP9_BD_VALUE value = r->value;
+  int count = r->count;
+  int shift = VP9_BD_VALUE_SIZE - 8 - (count + 8);
+  int loop_end = 0;
+  const int bits_left = (int)((buffer_end - buffer)*CHAR_BIT);
+  const int x = shift + CHAR_BIT - bits_left;
+
+  if (x >= 0) {
+    count += VP9_LOTS_OF_BITS;
+    loop_end = x;
+  }
+
+  if (x < 0 || bits_left) {
+    while (shift >= loop_end) {
+      count += CHAR_BIT;
+      value |= (VP9_BD_VALUE)*buffer++ << shift;
+      shift -= CHAR_BIT;
+    }
+  }
+
+  r->buffer = buffer;
+  r->value = value;
+  r->count = count;
+}
+
+const uint8_t *vp9_reader_find_end(vp9_reader *r) {
+  // Find the end of the coded buffer
+  while (r->count > CHAR_BIT && r->count < VP9_BD_VALUE_SIZE) {
+    r->count -= CHAR_BIT;
+    r->buffer--;
+  }
+  return r->buffer;
+}
+
diff --git a/libvpx/vp9/decoder/vp9_dboolhuff.h b/libvpx/vp9/decoder/vp9_dboolhuff.h
new file mode 100644
index 0000000..b50aa35
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_dboolhuff.h
@@ -0,0 +1,114 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_DECODER_VP9_DBOOLHUFF_H_
+#define VP9_DECODER_VP9_DBOOLHUFF_H_
+
+#include <stddef.h>
+#include <limits.h>
+
+#include "./vpx_config.h"
+#include "vpx_ports/mem.h"
+#include "vpx/vpx_integer.h"
+
+typedef size_t VP9_BD_VALUE;
+
+#define VP9_BD_VALUE_SIZE ((int)sizeof(VP9_BD_VALUE)*CHAR_BIT)
+
+// This is meant to be a large, positive constant that can still be efficiently
+// loaded as an immediate (on platforms like ARM, for example).
+// Even relatively modest values like 100 would work fine.
+#define VP9_LOTS_OF_BITS 0x40000000
+
+typedef struct {
+  const uint8_t *buffer_end;
+  const uint8_t *buffer;
+  VP9_BD_VALUE value;
+  int count;
+  unsigned int range;
+} vp9_reader;
+
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_norm[256]);
+
+int vp9_reader_init(vp9_reader *r, const uint8_t *buffer, size_t size);
+
+void vp9_reader_fill(vp9_reader *r);
+
+const uint8_t *vp9_reader_find_end(vp9_reader *r);
+
+static int vp9_read(vp9_reader *br, int probability) {
+  unsigned int bit = 0;
+  VP9_BD_VALUE value;
+  VP9_BD_VALUE bigsplit;
+  int count;
+  unsigned int range;
+  unsigned int split = 1 + (((br->range - 1) * probability) >> 8);
+
+  if (br->count < 0)
+    vp9_reader_fill(br);
+
+  value = br->value;
+  count = br->count;
+
+  bigsplit = (VP9_BD_VALUE)split << (VP9_BD_VALUE_SIZE - 8);
+
+  range = split;
+
+  if (value >= bigsplit) {
+    range = br->range - split;
+    value = value - bigsplit;
+    bit = 1;
+  }
+
+  {
+    register unsigned int shift = vp9_norm[range];
+    range <<= shift;
+    value <<= shift;
+    count -= shift;
+  }
+  br->value = value;
+  br->count = count;
+  br->range = range;
+
+  return bit;
+}
+
+static int vp9_read_bit(vp9_reader *r) {
+  return vp9_read(r, 128);  // vp9_prob_half
+}
+
+static int vp9_read_literal(vp9_reader *br, int bits) {
+  int z = 0, bit;
+
+  for (bit = bits - 1; bit >= 0; bit--)
+    z |= vp9_read_bit(br) << bit;
+
+  return z;
+}
+
+static int vp9_reader_has_error(vp9_reader *r) {
+  // Check if we have reached the end of the buffer.
+  //
+  // Variable 'count' stores the number of bits in the 'value' buffer, minus
+  // 8. The top byte is part of the algorithm, and the remainder is buffered
+  // to be shifted into it. So if count == 8, the top 16 bits of 'value' are
+  // occupied, 8 for the algorithm and 8 in the buffer.
+  //
+  // When reading a byte from the user's buffer, count is filled with 8 and
+  // one byte is filled into the value buffer. When we reach the end of the
+  // data, count is additionally filled with VP9_LOTS_OF_BITS. So when
+  // count == VP9_LOTS_OF_BITS - 1, the user's data has been exhausted.
+  //
+  // 1 if we have tried to decode bits after the end of stream was encountered.
+  // 0 No error.
+  return r->count > VP9_BD_VALUE_SIZE && r->count < VP9_LOTS_OF_BITS;
+}
+
+#endif  // VP9_DECODER_VP9_DBOOLHUFF_H_
diff --git a/libvpx/vp9/decoder/vp9_decodemv.c b/libvpx/vp9/decoder/vp9_decodemv.c
new file mode 100644
index 0000000..b3d41be
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_decodemv.c
@@ -0,0 +1,823 @@
+/*
+  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/decoder/vp9_treereader.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/decoder/vp9_decodemv.h"
+#include "vp9/decoder/vp9_decodframe.h"
+#include "vp9/common/vp9_mvref_common.h"
+#if CONFIG_DEBUG
+#include <assert.h>
+#endif
+
+// #define DEBUG_DEC_MV
+#ifdef DEBUG_DEC_MV
+int dec_mvcount = 0;
+#endif
+
+// #define DEC_DEBUG
+#ifdef DEC_DEBUG
+extern int dec_debug;
+#endif
+
+static MB_PREDICTION_MODE read_intra_mode(vp9_reader *r, const vp9_prob *p) {
+  MB_PREDICTION_MODE m = treed_read(r, vp9_intra_mode_tree, p);
+  return m;
+}
+
+static int read_mb_segid(vp9_reader *r, MACROBLOCKD *xd) {
+  return treed_read(r, vp9_segment_tree, xd->mb_segment_tree_probs);
+}
+
+static void set_segment_id(VP9_COMMON *cm, MB_MODE_INFO *mbmi,
+                           int mi_row, int mi_col, int segment_id) {
+  const int mi_index = mi_row * cm->mi_cols + mi_col;
+  const BLOCK_SIZE_TYPE sb_type = mbmi->sb_type;
+  const int bw = 1 << mi_width_log2(sb_type);
+  const int bh = 1 << mi_height_log2(sb_type);
+  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = MIN(cm->mi_cols - mi_col, bw);
+  int x, y;
+
+  for (y = 0; y < ymis; y++) {
+    for (x = 0; x < xmis; x++) {
+      const int index = mi_index + (y * cm->mi_cols + x);
+      cm->last_frame_seg_map[index] = segment_id;
+    }
+  }
+}
+
+static TX_SIZE select_txfm_size(VP9_COMMON *cm, MACROBLOCKD *xd,
+                                vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  const int context = vp9_get_pred_context(cm, xd, PRED_TX_SIZE);
+  const vp9_prob *tx_probs = vp9_get_pred_probs(cm, xd, PRED_TX_SIZE);
+  TX_SIZE txfm_size = vp9_read(r, tx_probs[0]);
+  if (txfm_size != TX_4X4 && bsize >= BLOCK_SIZE_MB16X16) {
+    txfm_size += vp9_read(r, tx_probs[1]);
+    if (txfm_size != TX_8X8 && bsize >= BLOCK_SIZE_SB32X32)
+      txfm_size += vp9_read(r, tx_probs[2]);
+  }
+  if (bsize >= BLOCK_SIZE_SB32X32) {
+    cm->fc.tx_count_32x32p[context][txfm_size]++;
+  } else if (bsize >= BLOCK_SIZE_MB16X16) {
+    cm->fc.tx_count_16x16p[context][txfm_size]++;
+  } else {
+    cm->fc.tx_count_8x8p[context][txfm_size]++;
+  }
+  return txfm_size;
+}
+
+
+static void kfread_modes(VP9D_COMP *pbi, MODE_INFO *m,
+                         int mi_row, int mi_col,
+                         vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  const int mis = cm->mode_info_stride;
+
+  // Read segmentation map if it is being updated explicitly this frame
+  m->mbmi.segment_id = 0;
+  if (xd->segmentation_enabled && xd->update_mb_segmentation_map) {
+    m->mbmi.segment_id = read_mb_segid(r, xd);
+    set_segment_id(cm, &m->mbmi, mi_row, mi_col, m->mbmi.segment_id);
+  }
+
+  m->mbmi.mb_skip_coeff = vp9_segfeature_active(xd, m->mbmi.segment_id,
+                                                SEG_LVL_SKIP);
+  if (!m->mbmi.mb_skip_coeff) {
+    m->mbmi.mb_skip_coeff = vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_MBSKIP));
+    cm->fc.mbskip_count[vp9_get_pred_context(cm, xd, PRED_MBSKIP)]
+                       [m->mbmi.mb_skip_coeff]++;
+  }
+
+  if (cm->txfm_mode == TX_MODE_SELECT &&
+      m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    m->mbmi.txfm_size = select_txfm_size(cm, xd, r, m->mbmi.sb_type);
+  } else if (cm->txfm_mode >= ALLOW_32X32 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_SB32X32) {
+    m->mbmi.txfm_size = TX_32X32;
+  } else if (cm->txfm_mode >= ALLOW_16X16 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_MB16X16) {
+    m->mbmi.txfm_size = TX_16X16;
+  } else if (cm->txfm_mode >= ALLOW_8X8 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    m->mbmi.txfm_size = TX_8X8;
+  } else {
+    m->mbmi.txfm_size = TX_4X4;
+  }
+
+  // luma mode
+  m->mbmi.ref_frame[0] = INTRA_FRAME;
+  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
+    const MB_PREDICTION_MODE L = xd->left_available ?
+                                  left_block_mode(m, 0) : DC_PRED;
+    m->mbmi.mode = read_intra_mode(r, cm->kf_y_mode_prob[A][L]);
+  } else {
+    int idx, idy;
+    int bw = 1 << b_width_log2(m->mbmi.sb_type);
+    int bh = 1 << b_height_log2(m->mbmi.sb_type);
+
+    for (idy = 0; idy < 2; idy += bh) {
+      for (idx = 0; idx < 2; idx += bw) {
+        int ib = idy * 2 + idx;
+        int k;
+        const MB_PREDICTION_MODE A = above_block_mode(m, ib, mis);
+        const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
+                                      left_block_mode(m, ib) : DC_PRED;
+        m->bmi[ib].as_mode.first =
+            read_intra_mode(r, cm->kf_y_mode_prob[A][L]);
+        for (k = 1; k < bh; ++k)
+          m->bmi[ib + k * 2].as_mode.first = m->bmi[ib].as_mode.first;
+        for (k = 1; k < bw; ++k)
+          m->bmi[ib + k].as_mode.first = m->bmi[ib].as_mode.first;
+      }
+    }
+    m->mbmi.mode = m->bmi[3].as_mode.first;
+  }
+
+  m->mbmi.uv_mode = read_intra_mode(r, cm->kf_uv_mode_prob[m->mbmi.mode]);
+}
+
+static int read_mv_component(vp9_reader *r,
+                             const nmv_component *mvcomp, int usehp) {
+
+  int mag, d, fr, hp;
+  const int sign = vp9_read(r, mvcomp->sign);
+  const int mv_class = treed_read(r, vp9_mv_class_tree, mvcomp->classes);
+
+  // Integer part
+  if (mv_class == MV_CLASS_0) {
+    d = treed_read(r, vp9_mv_class0_tree, mvcomp->class0);
+  } else {
+    int i;
+    const int n = mv_class + CLASS0_BITS - 1;  // number of bits
+
+    d = 0;
+    for (i = 0; i < n; ++i)
+      d |= vp9_read(r, mvcomp->bits[i]) << i;
+  }
+
+  // Fractional part
+  fr = treed_read(r, vp9_mv_fp_tree,
+                  mv_class == MV_CLASS_0 ? mvcomp->class0_fp[d] : mvcomp->fp);
+
+
+  // High precision part (if hp is not used, the default value of the hp is 1)
+  hp = usehp ? vp9_read(r,
+                        mv_class == MV_CLASS_0 ? mvcomp->class0_hp : mvcomp->hp)
+             : 1;
+
+  // result
+  mag = vp9_get_mv_mag(mv_class, (d << 3) | (fr << 1) | hp) + 1;
+  return sign ? -mag : mag;
+}
+
+static void update_nmv(vp9_reader *r, vp9_prob *const p,
+                       const vp9_prob upd_p) {
+  if (vp9_read(r, upd_p)) {
+#ifdef LOW_PRECISION_MV_UPDATE
+    *p = (vp9_read_literal(r, 7) << 1) | 1;
+#else
+    *p = (vp9_read_literal(r, 8));
+#endif
+  }
+}
+
+static void read_nmvprobs(vp9_reader *r, nmv_context *mvctx,
+                          int usehp) {
+  int i, j, k;
+
+#ifdef MV_GROUP_UPDATE
+  if (!vp9_read_bit(r))
+    return;
+#endif
+  for (j = 0; j < MV_JOINTS - 1; ++j)
+    update_nmv(r, &mvctx->joints[j], VP9_NMV_UPDATE_PROB);
+
+  for (i = 0; i < 2; ++i) {
+    update_nmv(r, &mvctx->comps[i].sign, VP9_NMV_UPDATE_PROB);
+    for (j = 0; j < MV_CLASSES - 1; ++j)
+      update_nmv(r, &mvctx->comps[i].classes[j], VP9_NMV_UPDATE_PROB);
+
+    for (j = 0; j < CLASS0_SIZE - 1; ++j)
+      update_nmv(r, &mvctx->comps[i].class0[j], VP9_NMV_UPDATE_PROB);
+
+    for (j = 0; j < MV_OFFSET_BITS; ++j)
+      update_nmv(r, &mvctx->comps[i].bits[j], VP9_NMV_UPDATE_PROB);
+  }
+
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j)
+      for (k = 0; k < 3; ++k)
+        update_nmv(r, &mvctx->comps[i].class0_fp[j][k], VP9_NMV_UPDATE_PROB);
+
+    for (j = 0; j < 3; ++j)
+      update_nmv(r, &mvctx->comps[i].fp[j], VP9_NMV_UPDATE_PROB);
+  }
+
+  if (usehp) {
+    for (i = 0; i < 2; ++i) {
+      update_nmv(r, &mvctx->comps[i].class0_hp, VP9_NMV_UPDATE_PROB);
+      update_nmv(r, &mvctx->comps[i].hp, VP9_NMV_UPDATE_PROB);
+    }
+  }
+}
+
+// Read the referncence frame
+static void read_ref_frame(VP9D_COMP *pbi, vp9_reader *r,
+                           int segment_id, MV_REFERENCE_FRAME ref_frame[2]) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  const int seg_ref_active = vp9_segfeature_active(xd, segment_id,
+                                                   SEG_LVL_REF_FRAME);
+
+  // Segment reference frame features not available.
+  if (!seg_ref_active) {
+    int is_comp;
+    int comp_ctx = vp9_get_pred_context(cm, xd, PRED_COMP_INTER_INTER);
+
+    if (cm->comp_pred_mode == HYBRID_PREDICTION) {
+      is_comp = vp9_read(r, cm->fc.comp_inter_prob[comp_ctx]);
+      cm->fc.comp_inter_count[comp_ctx][is_comp]++;
+    } else {
+      is_comp = cm->comp_pred_mode == COMP_PREDICTION_ONLY;
+    }
+
+    // FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding
+    if (is_comp) {
+      int b, fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
+      int ref_ctx = vp9_get_pred_context(cm, xd, PRED_COMP_REF_P);
+
+      ref_frame[fix_ref_idx]  = cm->comp_fixed_ref;
+      b = vp9_read(r, cm->fc.comp_ref_prob[ref_ctx]);
+      cm->fc.comp_ref_count[ref_ctx][b]++;
+      ref_frame[!fix_ref_idx] = cm->comp_var_ref[b];
+    } else {
+      int ref1_ctx = vp9_get_pred_context(cm, xd, PRED_SINGLE_REF_P1);
+      ref_frame[1] = NONE;
+      if (vp9_read(r, cm->fc.single_ref_prob[ref1_ctx][0])) {
+        int ref2_ctx = vp9_get_pred_context(cm, xd, PRED_SINGLE_REF_P2);
+        int b2 = vp9_read(r, cm->fc.single_ref_prob[ref2_ctx][1]);
+        ref_frame[0] = b2 ? ALTREF_FRAME : GOLDEN_FRAME;
+        cm->fc.single_ref_count[ref1_ctx][0][1]++;
+        cm->fc.single_ref_count[ref2_ctx][1][b2]++;
+      } else {
+        ref_frame[0] = LAST_FRAME;
+        cm->fc.single_ref_count[ref1_ctx][0][0]++;
+      }
+    }
+  } else {
+    ref_frame[0] = vp9_get_segdata(xd, segment_id, SEG_LVL_REF_FRAME);
+    ref_frame[1] = NONE;
+  }
+}
+
+static MB_PREDICTION_MODE read_sb_mv_ref(vp9_reader *r, const vp9_prob *p) {
+  return (MB_PREDICTION_MODE) treed_read(r, vp9_sb_mv_ref_tree, p);
+}
+
+#ifdef VPX_MODE_COUNT
+unsigned int vp9_mv_cont_count[5][4] = {
+  { 0, 0, 0, 0 },
+  { 0, 0, 0, 0 },
+  { 0, 0, 0, 0 },
+  { 0, 0, 0, 0 },
+  { 0, 0, 0, 0 }
+};
+#endif
+
+static void read_switchable_interp_probs(VP9_COMMON* const cm, vp9_reader *r) {
+  int i, j;
+  for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j)
+    for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i) {
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
+        cm->fc.switchable_interp_prob[j][i] =
+            // vp9_read_prob(r);
+            vp9_read_prob_diff_update(r, cm->fc.switchable_interp_prob[j][i]);
+      }
+    }
+}
+
+static void read_inter_mode_probs(VP9_COMMON *const cm, vp9_reader *r) {
+  int i, j;
+  for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
+    for (j = 0; j < VP9_INTER_MODES - 1; ++j) {
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
+        // cm->fc.inter_mode_probs[i][j] = vp9_read_prob(r);
+        cm->fc.inter_mode_probs[i][j] =
+            vp9_read_prob_diff_update(r, cm->fc.inter_mode_probs[i][j]);
+      }
+    }
+}
+
+static INLINE COMPPREDMODE_TYPE read_comp_pred_mode(vp9_reader *r) {
+  COMPPREDMODE_TYPE mode = vp9_read_bit(r);
+  if (mode)
+     mode += vp9_read_bit(r);
+  return mode;
+}
+
+static void mb_mode_mv_init(VP9D_COMP *pbi, vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+
+  if ((cm->frame_type != KEY_FRAME) && (!cm->intra_only)) {
+    nmv_context *const nmvc = &pbi->common.fc.nmvc;
+    MACROBLOCKD *const xd = &pbi->mb;
+    int i, j;
+
+    read_inter_mode_probs(cm, r);
+
+    if (cm->mcomp_filter_type == SWITCHABLE)
+      read_switchable_interp_probs(cm, r);
+
+    for (i = 0; i < INTRA_INTER_CONTEXTS; i++) {
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        cm->fc.intra_inter_prob[i] =
+            vp9_read_prob_diff_update(r, cm->fc.intra_inter_prob[i]);
+    }
+
+    if (cm->allow_comp_inter_inter) {
+      cm->comp_pred_mode = read_comp_pred_mode(r);
+      if (cm->comp_pred_mode == HYBRID_PREDICTION)
+        for (i = 0; i < COMP_INTER_CONTEXTS; i++)
+          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+            cm->fc.comp_inter_prob[i] =
+                vp9_read_prob_diff_update(r, cm->fc.comp_inter_prob[i]);
+    } else {
+      cm->comp_pred_mode = SINGLE_PREDICTION_ONLY;
+    }
+
+    if (cm->comp_pred_mode != COMP_PREDICTION_ONLY)
+      for (i = 0; i < REF_CONTEXTS; i++) {
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+          cm->fc.single_ref_prob[i][0] =
+              vp9_read_prob_diff_update(r, cm->fc.single_ref_prob[i][0]);
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+          cm->fc.single_ref_prob[i][1] =
+              vp9_read_prob_diff_update(r, cm->fc.single_ref_prob[i][1]);
+      }
+
+    if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
+      for (i = 0; i < REF_CONTEXTS; i++)
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+          cm->fc.comp_ref_prob[i] =
+              vp9_read_prob_diff_update(r, cm->fc.comp_ref_prob[i]);
+
+    // VP9_INTRA_MODES
+    for (j = 0; j < BLOCK_SIZE_GROUPS; j++) {
+      for (i = 0; i < VP9_INTRA_MODES - 1; ++i) {
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
+          cm->fc.y_mode_prob[j][i] =
+              vp9_read_prob_diff_update(r, cm->fc.y_mode_prob[j][i]);
+        }
+      }
+    }
+    for (j = 0; j < NUM_PARTITION_CONTEXTS; ++j) {
+      for (i = 0; i < PARTITION_TYPES - 1; ++i) {
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
+          cm->fc.partition_prob[INTER_FRAME][j][i] =
+              vp9_read_prob_diff_update(r,
+                  cm->fc.partition_prob[INTER_FRAME][j][i]);
+        }
+      }
+    }
+
+    read_nmvprobs(r, nmvc, xd->allow_high_precision_mv);
+  }
+}
+
+// This function either reads the segment id for the current macroblock from
+// the bitstream or if the value is temporally predicted asserts the predicted
+// value
+static int read_mb_segment_id(VP9D_COMP *pbi, int mi_row, int mi_col,
+                              vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  MODE_INFO *const mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+
+  if (!xd->segmentation_enabled)
+    return 0;  // Default for disabled segmentation
+
+  if (xd->update_mb_segmentation_map) {
+    int segment_id;
+
+    if (cm->temporal_update) {
+      // Temporal coding of the segment id for this mb is enabled.
+      // Get the context based probability for reading the
+      // prediction status flag
+      const vp9_prob pred_prob = vp9_get_pred_prob(cm, xd, PRED_SEG_ID);
+      const int pred_flag = vp9_read(r, pred_prob);
+      vp9_set_pred_flag(xd, PRED_SEG_ID, pred_flag);
+
+      // If the value is flagged as correctly predicted
+      // then use the predicted value, otherwise decode it explicitly
+      segment_id = pred_flag ? vp9_get_pred_mi_segid(cm, mbmi->sb_type,
+                                                     mi_row, mi_col)
+                             : read_mb_segid(r, xd);
+    } else {
+      segment_id = read_mb_segid(r, xd);  // Normal unpredicted coding mode
+    }
+
+    set_segment_id(cm, mbmi, mi_row, mi_col, segment_id);  // Side effect
+    return segment_id;
+  } else {
+    return vp9_get_pred_mi_segid(cm, mbmi->sb_type, mi_row, mi_col);
+  }
+}
+
+
+static INLINE void assign_and_clamp_mv(int_mv *dst, const int_mv *src,
+                                       int mb_to_left_edge,
+                                       int mb_to_right_edge,
+                                       int mb_to_top_edge,
+                                       int mb_to_bottom_edge) {
+  dst->as_int = src->as_int;
+  clamp_mv(dst, mb_to_left_edge, mb_to_right_edge, mb_to_top_edge,
+           mb_to_bottom_edge);
+}
+
+static INLINE void decode_mv(vp9_reader *r, MV *mv, const MV *ref,
+                             const nmv_context *ctx,
+                             nmv_context_counts *counts,
+                             int usehp) {
+  const MV_JOINT_TYPE j = treed_read(r, vp9_mv_joint_tree, ctx->joints);
+  MV diff = {0, 0};
+
+  usehp = usehp && vp9_use_nmv_hp(ref);
+  if (mv_joint_vertical(j))
+    diff.row = read_mv_component(r, &ctx->comps[0], usehp);
+
+  if (mv_joint_horizontal(j))
+    diff.col = read_mv_component(r, &ctx->comps[1], usehp);
+
+  vp9_increment_nmv(&diff, ref, counts, usehp);
+
+  mv->row = diff.row + ref->row;
+  mv->col = diff.col + ref->col;
+}
+
+static INLINE INTERPOLATIONFILTERTYPE read_switchable_filter_type(
+    VP9D_COMP *pbi, vp9_reader *r) {
+  const int index = treed_read(r, vp9_switchable_interp_tree,
+                               vp9_get_pred_probs(&pbi->common, &pbi->mb,
+                                                  PRED_SWITCHABLE_INTERP));
+  ++pbi->common.fc.switchable_interp_count
+                [vp9_get_pred_context(
+                    &pbi->common, &pbi->mb, PRED_SWITCHABLE_INTERP)][index];
+  return vp9_switchable_interp[index];
+}
+
+static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
+                             int mi_row, int mi_col,
+                             vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  nmv_context *const nmvc = &cm->fc.nmvc;
+  MACROBLOCKD *const xd = &pbi->mb;
+
+  int_mv *const mv0 = &mbmi->mv[0];
+  int_mv *const mv1 = &mbmi->mv[1];
+  BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
+  int bw = 1 << b_width_log2(bsize);
+  int bh = 1 << b_height_log2(bsize);
+
+  int mb_to_left_edge, mb_to_right_edge, mb_to_top_edge, mb_to_bottom_edge;
+  int j, idx, idy;
+
+  mbmi->ref_frame[1] = NONE;
+
+  // Make sure the MACROBLOCKD mode info pointer is pointed at the
+  // correct entry for the current macroblock.
+  xd->mode_info_context = mi;
+
+  // Distance of Mb to the various image edges.
+  // These specified to 8th pel as they are always compared to MV values
+  // that are in 1/8th pel units
+  set_mi_row_col(cm, xd, mi_row, 1 << mi_height_log2(bsize),
+                         mi_col, 1 << mi_width_log2(bsize));
+
+  mb_to_top_edge = xd->mb_to_top_edge - LEFT_TOP_MARGIN;
+  mb_to_bottom_edge = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN;
+  mb_to_left_edge = xd->mb_to_left_edge - LEFT_TOP_MARGIN;
+  mb_to_right_edge = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN;
+
+  // Read the macroblock segment id.
+  mbmi->segment_id = read_mb_segment_id(pbi, mi_row, mi_col, r);
+
+  mbmi->mb_skip_coeff = vp9_segfeature_active(xd, mbmi->segment_id,
+                                              SEG_LVL_SKIP);
+  if (!mbmi->mb_skip_coeff) {
+    mbmi->mb_skip_coeff = vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_MBSKIP));
+    cm->fc.mbskip_count[vp9_get_pred_context(cm, xd, PRED_MBSKIP)]
+                       [mbmi->mb_skip_coeff]++;
+  }
+
+  // Read the reference frame
+  if (!vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_REF_FRAME)) {
+    mbmi->ref_frame[0] =
+        vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_INTRA_INTER));
+    cm->fc.intra_inter_count[vp9_get_pred_context(cm, xd, PRED_INTRA_INTER)]
+                            [mbmi->ref_frame[0] != INTRA_FRAME]++;
+  } else {
+    mbmi->ref_frame[0] =
+        vp9_get_segdata(xd, mbmi->segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME;
+  }
+
+  if (cm->txfm_mode == TX_MODE_SELECT &&
+      (mbmi->mb_skip_coeff == 0 || mbmi->ref_frame[0] == INTRA_FRAME) &&
+      bsize >= BLOCK_SIZE_SB8X8) {
+    mbmi->txfm_size = select_txfm_size(cm, xd, r, bsize);
+  } else if (bsize >= BLOCK_SIZE_SB32X32 &&
+             cm->txfm_mode >= ALLOW_32X32) {
+    mbmi->txfm_size = TX_32X32;
+  } else if (cm->txfm_mode >= ALLOW_16X16 &&
+             bsize >= BLOCK_SIZE_MB16X16) {
+    mbmi->txfm_size = TX_16X16;
+  } else if (cm->txfm_mode >= ALLOW_8X8 && (bsize >= BLOCK_SIZE_SB8X8)) {
+    mbmi->txfm_size = TX_8X8;
+  } else {
+    mbmi->txfm_size = TX_4X4;
+  }
+
+  // If reference frame is an Inter frame
+  if (mbmi->ref_frame[0] != INTRA_FRAME) {
+    int_mv nearest, nearby, best_mv;
+    int_mv nearest_second, nearby_second, best_mv_second;
+    vp9_prob *mv_ref_p;
+
+    read_ref_frame(pbi, r, mbmi->segment_id, mbmi->ref_frame);
+
+    {
+#ifdef DEC_DEBUG
+      if (dec_debug)
+        printf("%d %d\n", xd->mode_info_context->mbmi.mv[0].as_mv.row,
+               xd->mode_info_context->mbmi.mv[0].as_mv.col);
+#endif
+      vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
+                       mbmi->ref_frame[0], mbmi->ref_mvs[mbmi->ref_frame[0]],
+                       cm->ref_frame_sign_bias);
+
+      mv_ref_p = cm->fc.inter_mode_probs[
+        mbmi->mb_mode_context[mbmi->ref_frame[0]]];
+
+      // If the segment level skip mode enabled
+      if (vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_SKIP)) {
+        mbmi->mode = ZEROMV;
+      } else if (bsize >= BLOCK_SIZE_SB8X8) {
+        mbmi->mode = read_sb_mv_ref(r, mv_ref_p);
+        vp9_accum_mv_refs(cm, mbmi->mode,
+                          mbmi->mb_mode_context[mbmi->ref_frame[0]]);
+      }
+
+      if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
+        vp9_find_best_ref_mvs(xd,
+                              mbmi->ref_mvs[mbmi->ref_frame[0]],
+                              &nearest, &nearby);
+
+        best_mv.as_int = mbmi->ref_mvs[mbmi->ref_frame[0]][0].as_int;
+      }
+
+#ifdef DEC_DEBUG
+      if (dec_debug)
+        printf("[D %d %d] %d %d %d %d\n", ref_frame,
+               mbmi->mb_mode_context[ref_frame],
+               mv_ref_p[0], mv_ref_p[1], mv_ref_p[2], mv_ref_p[3]);
+#endif
+    }
+
+    mbmi->interp_filter = cm->mcomp_filter_type == SWITCHABLE
+                              ? read_switchable_filter_type(pbi, r)
+                              : cm->mcomp_filter_type;
+
+    if (mbmi->ref_frame[1] > INTRA_FRAME) {
+      vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
+                       mbmi->ref_frame[1],
+                       mbmi->ref_mvs[mbmi->ref_frame[1]],
+                       cm->ref_frame_sign_bias);
+
+      if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
+        vp9_find_best_ref_mvs(xd,
+                              mbmi->ref_mvs[mbmi->ref_frame[1]],
+                              &nearest_second,
+                              &nearby_second);
+        best_mv_second.as_int = mbmi->ref_mvs[mbmi->ref_frame[1]][0].as_int;
+      }
+    }
+
+    mbmi->uv_mode = DC_PRED;
+    if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+      for (idy = 0; idy < 2; idy += bh) {
+        for (idx = 0; idx < 2; idx += bw) {
+          int_mv blockmv, secondmv;
+          int blockmode;
+          int i;
+          j = idy * 2 + idx;
+
+          blockmode = read_sb_mv_ref(r, mv_ref_p);
+          vp9_accum_mv_refs(cm, blockmode,
+                            mbmi->mb_mode_context[mbmi->ref_frame[0]]);
+          if (blockmode == NEARESTMV || blockmode == NEARMV) {
+            MV_REFERENCE_FRAME rf2 = mbmi->ref_frame[1];
+            vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest, &nearby, j, 0);
+            if (rf2 > 0) {
+              vp9_append_sub8x8_mvs_for_idx(cm, xd,  &nearest_second,
+                                            &nearby_second, j, 1);
+            }
+          }
+
+          switch (blockmode) {
+            case NEWMV:
+              decode_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
+                         &cm->fc.NMVcount, xd->allow_high_precision_mv);
+
+              if (mbmi->ref_frame[1] > 0)
+                decode_mv(r, &secondmv.as_mv, &best_mv_second.as_mv, nmvc,
+                          &cm->fc.NMVcount, xd->allow_high_precision_mv);
+
+#ifdef VPX_MODE_COUNT
+              vp9_mv_cont_count[mv_contz][3]++;
+#endif
+              break;
+            case NEARESTMV:
+              blockmv.as_int = nearest.as_int;
+              if (mbmi->ref_frame[1] > 0)
+                secondmv.as_int = nearest_second.as_int;
+#ifdef VPX_MODE_COUNT
+              vp9_mv_cont_count[mv_contz][0]++;
+#endif
+              break;
+            case NEARMV:
+              blockmv.as_int = nearby.as_int;
+              if (mbmi->ref_frame[1] > 0)
+                secondmv.as_int = nearby_second.as_int;
+#ifdef VPX_MODE_COUNT
+              vp9_mv_cont_count[mv_contz][1]++;
+#endif
+              break;
+            case ZEROMV:
+              blockmv.as_int = 0;
+              if (mbmi->ref_frame[1] > 0)
+                secondmv.as_int = 0;
+#ifdef VPX_MODE_COUNT
+              vp9_mv_cont_count[mv_contz][2]++;
+#endif
+              break;
+            default:
+              break;
+          }
+          mi->bmi[j].as_mv[0].as_int = blockmv.as_int;
+          if (mbmi->ref_frame[1] > 0)
+            mi->bmi[j].as_mv[1].as_int = secondmv.as_int;
+
+          for (i = 1; i < bh; ++i)
+            vpx_memcpy(&mi->bmi[j + i * 2], &mi->bmi[j], sizeof(mi->bmi[j]));
+          for (i = 1; i < bw; ++i)
+            vpx_memcpy(&mi->bmi[j + i], &mi->bmi[j], sizeof(mi->bmi[j]));
+          mi->mbmi.mode = blockmode;
+        }
+      }
+
+      mv0->as_int = mi->bmi[3].as_mv[0].as_int;
+      mv1->as_int = mi->bmi[3].as_mv[1].as_int;
+    } else {
+      switch (mbmi->mode) {
+        case NEARMV:
+          // Clip "next_nearest" so that it does not extend to far out of image
+          assign_and_clamp_mv(mv0, &nearby, mb_to_left_edge,
+                                            mb_to_right_edge,
+                                            mb_to_top_edge,
+                                            mb_to_bottom_edge);
+          if (mbmi->ref_frame[1] > 0)
+            assign_and_clamp_mv(mv1, &nearby_second, mb_to_left_edge,
+                                                     mb_to_right_edge,
+                                                     mb_to_top_edge,
+                                                     mb_to_bottom_edge);
+          break;
+
+        case NEARESTMV:
+          // Clip "next_nearest" so that it does not extend to far out of image
+          assign_and_clamp_mv(mv0, &nearest, mb_to_left_edge,
+                                             mb_to_right_edge,
+                                             mb_to_top_edge,
+                                             mb_to_bottom_edge);
+          if (mbmi->ref_frame[1] > 0)
+            assign_and_clamp_mv(mv1, &nearest_second, mb_to_left_edge,
+                                                      mb_to_right_edge,
+                                                      mb_to_top_edge,
+                                                      mb_to_bottom_edge);
+          break;
+
+        case ZEROMV:
+          mv0->as_int = 0;
+          if (mbmi->ref_frame[1] > 0)
+            mv1->as_int = 0;
+          break;
+
+        case NEWMV:
+          decode_mv(r, &mv0->as_mv, &best_mv.as_mv, nmvc, &cm->fc.NMVcount,
+                    xd->allow_high_precision_mv);
+          if (mbmi->ref_frame[1] > 0)
+            decode_mv(r, &mv1->as_mv, &best_mv_second.as_mv, nmvc,
+                      &cm->fc.NMVcount, xd->allow_high_precision_mv);
+          break;
+        default:
+#if CONFIG_DEBUG
+          assert(0);
+#endif
+          break;
+      }
+    }
+  } else {
+    // required for left and above block mv
+    mv0->as_int = 0;
+
+    if (bsize >= BLOCK_SIZE_SB8X8) {
+      const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+      const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+      const int bsl = MIN(bwl, bhl);
+      mbmi->mode = read_intra_mode(r, cm->fc.y_mode_prob[MIN(3, bsl)]);
+      cm->fc.y_mode_counts[MIN(3, bsl)][mbmi->mode]++;
+    } else {
+      int idx, idy;
+      for (idy = 0; idy < 2; idy += bh) {
+        for (idx = 0; idx < 2; idx += bw) {
+          int ib = idy * 2 + idx, k;
+          int m = read_intra_mode(r, cm->fc.y_mode_prob[0]);
+          mi->bmi[ib].as_mode.first = m;
+          cm->fc.y_mode_counts[0][m]++;
+          for (k = 1; k < bh; ++k)
+            mi->bmi[ib + k * 2].as_mode.first = m;
+          for (k = 1; k < bw; ++k)
+            mi->bmi[ib + k].as_mode.first = m;
+        }
+      }
+      mbmi->mode = mi->bmi[3].as_mode.first;
+    }
+
+    mbmi->uv_mode = read_intra_mode(r, cm->fc.uv_mode_prob[mbmi->mode]);
+    cm->fc.uv_mode_counts[mbmi->mode][mbmi->uv_mode]++;
+  }
+}
+
+void vp9_decode_mode_mvs_init(VP9D_COMP* const pbi, vp9_reader *r) {
+  VP9_COMMON *cm = &pbi->common;
+  int k;
+
+  // TODO(jkoleszar): does this clear more than MBSKIP_CONTEXTS? Maybe remove.
+  // vpx_memset(cm->fc.mbskip_probs, 0, sizeof(cm->fc.mbskip_probs));
+  for (k = 0; k < MBSKIP_CONTEXTS; ++k) {
+    if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
+      cm->fc.mbskip_probs[k] =
+          vp9_read_prob_diff_update(r, cm->fc.mbskip_probs[k]);
+    }
+    // cm->fc.mbskip_probs[k] = vp9_read_prob(r);
+  }
+
+  mb_mode_mv_init(pbi, r);
+}
+
+void vp9_decode_mb_mode_mv(VP9D_COMP* const pbi,
+                           MACROBLOCKD* const xd,
+                           int mi_row,
+                           int mi_col,
+                           vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MODE_INFO *mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+
+  if ((cm->frame_type == KEY_FRAME) || cm->intra_only) {
+    kfread_modes(pbi, mi, mi_row, mi_col, r);
+  } else {
+    read_mb_modes_mv(pbi, mi, &mi->mbmi, mi_row, mi_col, r);
+  }
+
+  if (1) {
+    const int bw = 1 << mi_width_log2(mbmi->sb_type);
+    const int bh = 1 << mi_height_log2(mbmi->sb_type);
+    const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+    const int x_mis = MIN(bw, cm->mi_cols - mi_col);
+    const int mis = cm->mode_info_stride;
+    int x, y;
+
+    for (y = 0; y < y_mis; y++)
+      for (x = !y; x < x_mis; x++)
+        mi[y * mis + x] = *mi;
+  }
+}
diff --git a/libvpx/vp9/decoder/vp9_decodemv.h b/libvpx/vp9/decoder/vp9_decodemv.h
new file mode 100644
index 0000000..bf5e83c
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_decodemv.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_DECODER_VP9_DECODEMV_H_
+#define VP9_DECODER_VP9_DECODEMV_H_
+
+#include "vp9/decoder/vp9_onyxd_int.h"
+
+void vp9_decode_mb_mode_mv(VP9D_COMP* const pbi,
+                           MACROBLOCKD* const xd,
+                           int mb_row,
+                           int mb_col,
+                           vp9_reader *r);
+void vp9_decode_mode_mvs_init(VP9D_COMP* const pbi, vp9_reader *r);
+
+#endif  // VP9_DECODER_VP9_DECODEMV_H_
diff --git a/libvpx/vp9/decoder/vp9_decodframe.c b/libvpx/vp9/decoder/vp9_decodframe.c
new file mode 100644
index 0000000..49b181d
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_decodframe.c
@@ -0,0 +1,1199 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+
+#include "./vp9_rtcd.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/vpx_scale.h"
+
+#include "vp9/common/vp9_extend.h"
+#include "vp9/common/vp9_modecont.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_tile_common.h"
+
+#include "vp9/decoder/vp9_dboolhuff.h"
+#include "vp9/decoder/vp9_decodframe.h"
+#include "vp9/decoder/vp9_detokenize.h"
+#include "vp9/decoder/vp9_decodemv.h"
+#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_read_bit_buffer.h"
+
+
+// #define DEC_DEBUG
+#ifdef DEC_DEBUG
+int dec_debug = 0;
+#endif
+
+static int read_be32(const uint8_t *p) {
+  return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
+}
+
+// len == 0 is not allowed
+static int read_is_valid(const uint8_t *start, size_t len,
+                         const uint8_t *end) {
+  return start + len > start && start + len <= end;
+}
+
+static void setup_txfm_mode(VP9_COMMON *pc, int lossless, vp9_reader *r) {
+  if (lossless) {
+    pc->txfm_mode = ONLY_4X4;
+  } else {
+    pc->txfm_mode = vp9_read_literal(r, 2);
+    if (pc->txfm_mode == ALLOW_32X32)
+      pc->txfm_mode += vp9_read_bit(r);
+    if (pc->txfm_mode == TX_MODE_SELECT) {
+      int i, j;
+      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+        for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j) {
+          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+            pc->fc.tx_probs_8x8p[i][j] =
+                vp9_read_prob_diff_update(r, pc->fc.tx_probs_8x8p[i][j]);
+        }
+      }
+      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+        for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j) {
+          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+            pc->fc.tx_probs_16x16p[i][j] =
+                vp9_read_prob_diff_update(r, pc->fc.tx_probs_16x16p[i][j]);
+        }
+      }
+      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
+        for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j) {
+          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+            pc->fc.tx_probs_32x32p[i][j] =
+                vp9_read_prob_diff_update(r, pc->fc.tx_probs_32x32p[i][j]);
+        }
+      }
+    }
+  }
+}
+
+static int get_unsigned_bits(unsigned int num_values) {
+  int cat = 0;
+  if (num_values <= 1)
+    return 0;
+  num_values--;
+  while (num_values > 0) {
+    cat++;
+    num_values >>= 1;
+  }
+  return cat;
+}
+
+static int inv_recenter_nonneg(int v, int m) {
+  if (v > 2 * m)
+    return v;
+
+  return v % 2 ? m - (v + 1) / 2 : m + v / 2;
+}
+
+static int decode_uniform(vp9_reader *r, int n) {
+  int v;
+  const int l = get_unsigned_bits(n);
+  const int m = (1 << l) - n;
+  if (!l)
+    return 0;
+
+  v = vp9_read_literal(r, l - 1);
+  return v < m ?  v : (v << 1) - m + vp9_read_bit(r);
+}
+
+static int decode_term_subexp(vp9_reader *r, int k, int num_syms) {
+  int i = 0, mk = 0, word;
+  while (1) {
+    const int b = i ? k + i - 1 : k;
+    const int a = 1 << b;
+    if (num_syms <= mk + 3 * a) {
+      word = decode_uniform(r, num_syms - mk) + mk;
+      break;
+    } else {
+      if (vp9_read_bit(r)) {
+        i++;
+        mk += a;
+      } else {
+        word = vp9_read_literal(r, b) + mk;
+        break;
+      }
+    }
+  }
+  return word;
+}
+
+static int decode_unsigned_max(struct vp9_read_bit_buffer *rb, int max) {
+  const int data = vp9_rb_read_literal(rb, get_unsigned_bits(max));
+  return data > max ? max : data;
+}
+
+static int merge_index(int v, int n, int modulus) {
+  int max1 = (n - 1 - modulus / 2) / modulus + 1;
+  if (v < max1) {
+    v = v * modulus + modulus / 2;
+  } else {
+    int w;
+    v -= max1;
+    w = v;
+    v += (v + modulus - modulus / 2) / modulus;
+    while (v % modulus == modulus / 2 ||
+           w != v - (v + modulus - modulus / 2) / modulus) v++;
+  }
+  return v;
+}
+
+static int inv_remap_prob(int v, int m) {
+  const int n = 255;
+
+  v = merge_index(v, n - 1, MODULUS_PARAM);
+  m--;
+  if ((m << 1) <= n) {
+    return 1 + inv_recenter_nonneg(v + 1, m);
+  } else {
+    return n - inv_recenter_nonneg(v + 1, n - 1 - m);
+  }
+}
+
+vp9_prob vp9_read_prob_diff_update(vp9_reader *r, int oldp) {
+  int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
+  return (vp9_prob)inv_remap_prob(delp, oldp);
+}
+
+void vp9_init_dequantizer(VP9_COMMON *pc) {
+  int q;
+
+  for (q = 0; q < QINDEX_RANGE; q++) {
+    // DC value
+    pc->y_dequant[q][0] = vp9_dc_quant(q, pc->y_dc_delta_q);
+    pc->uv_dequant[q][0] = vp9_dc_quant(q, pc->uv_dc_delta_q);
+
+    // AC values
+    pc->y_dequant[q][1] = vp9_ac_quant(q, 0);
+    pc->uv_dequant[q][1] = vp9_ac_quant(q, pc->uv_ac_delta_q);
+  }
+}
+
+static void mb_init_dequantizer(VP9_COMMON *pc, MACROBLOCKD *xd) {
+  int i;
+  const int segment_id = xd->mode_info_context->mbmi.segment_id;
+  xd->q_index = vp9_get_qindex(xd, segment_id, pc->base_qindex);
+
+  xd->plane[0].dequant = pc->y_dequant[xd->q_index];
+  for (i = 1; i < MAX_MB_PLANE; i++)
+    xd->plane[i].dequant = pc->uv_dequant[xd->q_index];
+}
+
+static void decode_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                         int ss_txfrm_size, void *arg) {
+  MACROBLOCKD* const xd = arg;
+  struct macroblockd_plane *pd = &xd->plane[plane];
+  int16_t* const qcoeff = BLOCK_OFFSET(pd->qcoeff, block, 16);
+  const int stride = pd->dst.stride;
+  const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
+                                                       block, ss_txfrm_size);
+  uint8_t* const dst = raster_block_offset_uint8(xd, bsize, plane,
+                                                 raster_block,
+                                                 pd->dst.buf, stride);
+
+  TX_TYPE tx_type;
+
+  switch (ss_txfrm_size / 2) {
+    case TX_4X4:
+      tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        xd->itxm_add(qcoeff, dst, stride, pd->eobs[block]);
+      else
+        vp9_iht_add_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+      break;
+    case TX_8X8:
+      tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
+      vp9_iht_add_8x8_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+      break;
+    case TX_16X16:
+      tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
+      vp9_iht_add_16x16_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+      break;
+    case TX_32X32:
+      vp9_idct_add_32x32(qcoeff, dst, stride, pd->eobs[block]);
+      break;
+  }
+}
+
+static void decode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                               int ss_txfrm_size, void *arg) {
+  MACROBLOCKD* const xd = arg;
+  struct macroblockd_plane *pd = &xd->plane[plane];
+
+  const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
+                                                       block, ss_txfrm_size);
+  uint8_t* const dst = raster_block_offset_uint8(xd, bsize, plane,
+                                                 raster_block,
+                                                 pd->dst.buf, pd->dst.stride);
+  const TX_SIZE tx_size = (TX_SIZE)(ss_txfrm_size / 2);
+  int b_mode;
+  int plane_b_size;
+  const int tx_ib = raster_block >> tx_size;
+  const int mode = plane == 0 ? xd->mode_info_context->mbmi.mode
+                              : xd->mode_info_context->mbmi.uv_mode;
+
+
+  if (plane == 0 && xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+    assert(bsize == BLOCK_SIZE_SB8X8);
+    b_mode = xd->mode_info_context->bmi[raster_block].as_mode.first;
+  } else {
+    b_mode = mode;
+  }
+
+  if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0)
+    extend_for_intra(xd, plane, block, bsize, ss_txfrm_size);
+
+  plane_b_size = b_width_log2(bsize) - pd->subsampling_x;
+  vp9_predict_intra_block(xd, tx_ib, plane_b_size, tx_size, b_mode,
+                          dst, pd->dst.stride);
+
+  // Early exit if there are no coefficients
+  if (xd->mode_info_context->mbmi.mb_skip_coeff)
+    return;
+
+  decode_block(plane, block, bsize, ss_txfrm_size, arg);
+}
+
+static void decode_atom(VP9D_COMP *pbi, MACROBLOCKD *xd,
+                        int mi_row, int mi_col,
+                        vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  assert(mbmi->ref_frame[0] != INTRA_FRAME);
+
+  if ((pbi->common.frame_type != KEY_FRAME) && (!pbi->common.intra_only))
+    vp9_setup_interp_filters(xd, mbmi->interp_filter, &pbi->common);
+
+  // prediction
+  vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+
+  if (mbmi->mb_skip_coeff) {
+    vp9_reset_sb_tokens_context(xd, bsize);
+  } else {
+    if (xd->segmentation_enabled)
+      mb_init_dequantizer(&pbi->common, xd);
+
+    if (!vp9_reader_has_error(r))
+      vp9_decode_tokens(pbi, r, bsize);
+
+    foreach_transformed_block(xd, bsize, decode_block, xd);
+  }
+}
+
+static void decode_sb_intra(VP9D_COMP *pbi, MACROBLOCKD *xd,
+                          int mi_row, int mi_col,
+                          vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+  if (mbmi->mb_skip_coeff) {
+    vp9_reset_sb_tokens_context(xd, bsize);
+  } else {
+    if (xd->segmentation_enabled)
+      mb_init_dequantizer(&pbi->common, xd);
+
+    if (!vp9_reader_has_error(r))
+      vp9_decode_tokens(pbi, r, bsize);
+  }
+
+  foreach_transformed_block(xd, bsize, decode_block_intra, xd);
+}
+
+
+static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mi_row, int mi_col,
+                      vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  const int bwl = mi_width_log2(bsize), bhl = mi_height_log2(bsize);
+  const int bw = 1 << bwl, bh = 1 << bhl;
+  int n, eobtotal;
+  VP9_COMMON *const pc = &pbi->common;
+  MODE_INFO *const mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+  const int mis = pc->mode_info_stride;
+
+  assert(mbmi->sb_type == bsize);
+  assert(mbmi->ref_frame[0] != INTRA_FRAME);
+
+  if (pbi->common.frame_type != KEY_FRAME)
+    vp9_setup_interp_filters(xd, mbmi->interp_filter, pc);
+
+  // generate prediction
+  vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+
+  if (mbmi->mb_skip_coeff) {
+    vp9_reset_sb_tokens_context(xd, bsize);
+  } else {
+    // re-initialize macroblock dequantizer before detokenization
+    if (xd->segmentation_enabled)
+      mb_init_dequantizer(pc, xd);
+
+    // dequantization and idct
+    eobtotal = vp9_decode_tokens(pbi, r, bsize);
+    if (eobtotal == 0) {  // skip loopfilter
+      for (n = 0; n < bw * bh; n++) {
+        const int x_idx = n & (bw - 1), y_idx = n >> bwl;
+
+        if (mi_col + x_idx < pc->mi_cols && mi_row + y_idx < pc->mi_rows)
+          mi[y_idx * mis + x_idx].mbmi.mb_skip_coeff = 1;
+      }
+    } else {
+      foreach_transformed_block(xd, bsize, decode_block, xd);
+    }
+  }
+}
+
+static void set_offsets(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize,
+                        int mi_row, int mi_col) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  const int bh = 1 << mi_height_log2(bsize);
+  const int bw = 1 << mi_width_log2(bsize);
+  const int mi_idx = mi_row * cm->mode_info_stride + mi_col;
+  int i;
+
+  xd->mode_info_context = cm->mi + mi_idx;
+  xd->mode_info_context->mbmi.sb_type = bsize;
+  // Special case: if prev_mi is NULL, the previous mode info context
+  // cannot be used.
+  xd->prev_mode_info_context = cm->prev_mi ? cm->prev_mi + mi_idx : NULL;
+
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    struct macroblockd_plane *pd = &xd->plane[i];
+    pd->above_context = cm->above_context[i] +
+                            (mi_col * 2 >> pd->subsampling_x);
+    pd->left_context = cm->left_context[i] +
+                           (((mi_row * 2) & 15) >> pd->subsampling_y);
+  }
+  xd->above_seg_context = cm->above_seg_context + mi_col;
+  xd->left_seg_context  = cm->left_seg_context + (mi_row & MI_MASK);
+
+  // Distance of Mb to the various image edges. These are specified to 8th pel
+  // as they are always compared to values that are in 1/8th pel units
+  set_mi_row_col(cm, xd, mi_row, bh, mi_col, bw);
+
+  setup_dst_planes(xd, &cm->yv12_fb[cm->new_fb_idx], mi_row, mi_col);
+}
+
+static void set_refs(VP9D_COMP *pbi, int mi_row, int mi_col) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  if (mbmi->ref_frame[0] > INTRA_FRAME) {
+    // Select the appropriate reference frame for this MB
+    const int fb_idx = cm->active_ref_idx[mbmi->ref_frame[0] - 1];
+    const YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[fb_idx];
+    xd->scale_factor[0]    = cm->active_ref_scale[mbmi->ref_frame[0] - 1];
+    xd->scale_factor_uv[0] = cm->active_ref_scale[mbmi->ref_frame[0] - 1];
+    setup_pre_planes(xd, cfg, NULL, mi_row, mi_col,
+                     xd->scale_factor, xd->scale_factor_uv);
+    xd->corrupted |= cfg->corrupted;
+
+    if (mbmi->ref_frame[1] > INTRA_FRAME) {
+      // Select the appropriate reference frame for this MB
+      const int second_fb_idx = cm->active_ref_idx[mbmi->ref_frame[1] - 1];
+      const YV12_BUFFER_CONFIG *second_cfg = &cm->yv12_fb[second_fb_idx];
+      xd->scale_factor[1]    = cm->active_ref_scale[mbmi->ref_frame[1] - 1];
+      xd->scale_factor_uv[1] = cm->active_ref_scale[mbmi->ref_frame[1] - 1];
+      setup_pre_planes(xd, NULL, second_cfg, mi_row, mi_col,
+                       xd->scale_factor, xd->scale_factor_uv);
+      xd->corrupted |= second_cfg->corrupted;
+    }
+  }
+}
+
+static void decode_modes_b(VP9D_COMP *pbi, int mi_row, int mi_col,
+                           vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &pbi->mb;
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index > 0)
+      return;
+  set_offsets(pbi, bsize, mi_row, mi_col);
+  vp9_decode_mb_mode_mv(pbi, xd, mi_row, mi_col, r);
+  set_refs(pbi, mi_row, mi_col);
+
+  if (xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME)
+    decode_sb_intra(pbi, xd, mi_row, mi_col, r, (bsize < BLOCK_SIZE_SB8X8) ?
+                                     BLOCK_SIZE_SB8X8 : bsize);
+  else if (bsize < BLOCK_SIZE_SB8X8)
+    decode_atom(pbi, xd, mi_row, mi_col, r, BLOCK_SIZE_SB8X8);
+  else
+    decode_sb(pbi, xd, mi_row, mi_col, r, bsize);
+
+  xd->corrupted |= vp9_reader_has_error(r);
+}
+
+static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
+                            vp9_reader* r, BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const pc = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  int bs = (1 << mi_width_log2(bsize)) / 2, n;
+  PARTITION_TYPE partition = PARTITION_NONE;
+  BLOCK_SIZE_TYPE subsize;
+
+  if (mi_row >= pc->mi_rows || mi_col >= pc->mi_cols)
+    return;
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index != 0)
+      return;
+
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    int pl;
+    int idx = check_bsize_coverage(pc, xd, mi_row, mi_col, bsize);
+    // read the partition information
+    xd->left_seg_context = pc->left_seg_context + (mi_row & MI_MASK);
+    xd->above_seg_context = pc->above_seg_context + mi_col;
+    pl = partition_plane_context(xd, bsize);
+
+    if (idx == 0)
+      partition = treed_read(r, vp9_partition_tree,
+                             pc->fc.partition_prob[pc->frame_type][pl]);
+    else if (idx > 0 &&
+        !vp9_read(r, pc->fc.partition_prob[pc->frame_type][pl][idx]))
+      partition = (idx == 1) ? PARTITION_HORZ : PARTITION_VERT;
+    else
+      partition = PARTITION_SPLIT;
+
+    pc->fc.partition_counts[pl][partition]++;
+  }
+
+  subsize = get_subsize(bsize, partition);
+  *(get_sb_index(xd, subsize)) = 0;
+
+  switch (partition) {
+    case PARTITION_NONE:
+      decode_modes_b(pbi, mi_row, mi_col, r, subsize);
+      break;
+    case PARTITION_HORZ:
+      decode_modes_b(pbi, mi_row, mi_col, r, subsize);
+      *(get_sb_index(xd, subsize)) = 1;
+      if (mi_row + bs < pc->mi_rows)
+        decode_modes_b(pbi, mi_row + bs, mi_col, r, subsize);
+      break;
+    case PARTITION_VERT:
+      decode_modes_b(pbi, mi_row, mi_col, r, subsize);
+      *(get_sb_index(xd, subsize)) = 1;
+      if (mi_col + bs < pc->mi_cols)
+        decode_modes_b(pbi, mi_row, mi_col + bs, r, subsize);
+      break;
+    case PARTITION_SPLIT:
+      for (n = 0; n < 4; n++) {
+        int j = n >> 1, i = n & 0x01;
+        *(get_sb_index(xd, subsize)) = n;
+        decode_modes_sb(pbi, mi_row + j * bs, mi_col + i * bs, r, subsize);
+      }
+      break;
+    default:
+      assert(0);
+  }
+  // update partition context
+  if (bsize >= BLOCK_SIZE_SB8X8 &&
+      (bsize == BLOCK_SIZE_SB8X8 || partition != PARTITION_SPLIT)) {
+    set_partition_seg_context(pc, xd, mi_row, mi_col);
+    update_partition_context(xd, subsize, bsize);
+  }
+}
+
+static void setup_token_decoder(VP9D_COMP *pbi,
+                                const uint8_t *data, size_t read_size,
+                                vp9_reader *r) {
+  VP9_COMMON *pc = &pbi->common;
+  const uint8_t *data_end = pbi->source + pbi->source_sz;
+
+  // Validate the calculated partition length. If the buffer
+  // described by the partition can't be fully read, then restrict
+  // it to the portion that can be (for EC mode) or throw an error.
+  if (!read_is_valid(data, read_size, data_end))
+    vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
+                       "Truncated packet or corrupt tile length");
+
+  if (vp9_reader_init(r, data, read_size))
+    vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate bool decoder %d", 1);
+}
+
+static void read_coef_probs_common(FRAME_CONTEXT *fc, TX_SIZE tx_size,
+                                   vp9_reader *r) {
+  vp9_coeff_probs_model *coef_probs = fc->coef_probs[tx_size];
+
+  if (vp9_read_bit(r)) {
+    int i, j, k, l, m;
+    for (i = 0; i < BLOCK_TYPES; i++) {
+      for (j = 0; j < REF_TYPES; j++) {
+        for (k = 0; k < COEF_BANDS; k++) {
+          for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
+            if (l >= 3 && k == 0)
+              continue;
+
+            for (m = 0; m < UNCONSTRAINED_NODES; m++) {
+              vp9_prob *const p = coef_probs[i][j][k][l] + m;
+
+              if (vp9_read(r, VP9_COEF_UPDATE_PROB))
+                *p = vp9_read_prob_diff_update(r, *p);
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+static void read_coef_probs(VP9D_COMP *pbi, vp9_reader *r) {
+  const TXFM_MODE txfm_mode = pbi->common.txfm_mode;
+  FRAME_CONTEXT *const fc = &pbi->common.fc;
+
+  read_coef_probs_common(fc, TX_4X4, r);
+
+  if (txfm_mode > ONLY_4X4)
+    read_coef_probs_common(fc, TX_8X8, r);
+
+  if (txfm_mode > ALLOW_8X8)
+    read_coef_probs_common(fc, TX_16X16, r);
+
+  if (txfm_mode > ALLOW_16X16)
+    read_coef_probs_common(fc, TX_32X32, r);
+}
+
+static void setup_segmentation(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
+  int i, j;
+
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+
+  xd->update_mb_segmentation_map = 0;
+  xd->update_mb_segmentation_data = 0;
+
+  xd->segmentation_enabled = vp9_rb_read_bit(rb);
+  if (!xd->segmentation_enabled)
+    return;
+
+  // Segmentation map update
+  xd->update_mb_segmentation_map = vp9_rb_read_bit(rb);
+  if (xd->update_mb_segmentation_map) {
+    for (i = 0; i < MB_SEG_TREE_PROBS; i++)
+      xd->mb_segment_tree_probs[i] = vp9_rb_read_bit(rb) ?
+                                         vp9_rb_read_literal(rb, 8) : MAX_PROB;
+
+    cm->temporal_update = vp9_rb_read_bit(rb);
+    if (cm->temporal_update) {
+      for (i = 0; i < PREDICTION_PROBS; i++)
+        cm->segment_pred_probs[i] = vp9_rb_read_bit(rb) ?
+                                        vp9_rb_read_literal(rb, 8) : MAX_PROB;
+    } else {
+      for (i = 0; i < PREDICTION_PROBS; i++)
+        cm->segment_pred_probs[i] = MAX_PROB;
+    }
+  }
+
+  // Segmentation data update
+  xd->update_mb_segmentation_data = vp9_rb_read_bit(rb);
+  if (xd->update_mb_segmentation_data) {
+    xd->mb_segment_abs_delta = vp9_rb_read_bit(rb);
+
+    vp9_clearall_segfeatures(xd);
+
+    for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+      for (j = 0; j < SEG_LVL_MAX; j++) {
+        int data = 0;
+        const int feature_enabled = vp9_rb_read_bit(rb);
+        if (feature_enabled) {
+          vp9_enable_segfeature(xd, i, j);
+          data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j));
+          if (vp9_is_segfeature_signed(j))
+            data = vp9_rb_read_bit(rb) ? -data : data;
+        }
+        vp9_set_segdata(xd, i, j, data);
+      }
+    }
+  }
+}
+
+static void setup_loopfilter(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+
+  cm->filter_level = vp9_rb_read_literal(rb, 6);
+  cm->sharpness_level = vp9_rb_read_literal(rb, 3);
+
+  // Read in loop filter deltas applied at the MB level based on mode or ref
+  // frame.
+  xd->mode_ref_lf_delta_update = 0;
+
+  xd->mode_ref_lf_delta_enabled = vp9_rb_read_bit(rb);
+  if (xd->mode_ref_lf_delta_enabled) {
+    xd->mode_ref_lf_delta_update = vp9_rb_read_bit(rb);
+    if (xd->mode_ref_lf_delta_update) {
+      int i;
+
+      for (i = 0; i < MAX_REF_LF_DELTAS; i++) {
+        if (vp9_rb_read_bit(rb)) {
+          const int value = vp9_rb_read_literal(rb, 6);
+          xd->ref_lf_deltas[i] = vp9_rb_read_bit(rb) ? -value : value;
+        }
+      }
+
+      for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
+        if (vp9_rb_read_bit(rb)) {
+          const int value = vp9_rb_read_literal(rb, 6);
+          xd->mode_lf_deltas[i] = vp9_rb_read_bit(rb) ? -value : value;
+        }
+      }
+    }
+  }
+}
+
+static int read_delta_q(struct vp9_read_bit_buffer *rb, int *delta_q) {
+  const int old = *delta_q;
+  if (vp9_rb_read_bit(rb)) {
+    const int value = vp9_rb_read_literal(rb, 4);
+    *delta_q = vp9_rb_read_bit(rb) ? -value : value;
+  }
+  return old != *delta_q;
+}
+
+static void setup_quantization(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
+  MACROBLOCKD *const xd = &pbi->mb;
+  VP9_COMMON *const cm = &pbi->common;
+  int update = 0;
+
+  cm->base_qindex = vp9_rb_read_literal(rb, QINDEX_BITS);
+  update |= read_delta_q(rb, &cm->y_dc_delta_q);
+  update |= read_delta_q(rb, &cm->uv_dc_delta_q);
+  update |= read_delta_q(rb, &cm->uv_ac_delta_q);
+  if (update)
+    vp9_init_dequantizer(cm);
+
+  xd->lossless = cm->base_qindex == 0 &&
+                 cm->y_dc_delta_q == 0 &&
+                 cm->uv_dc_delta_q == 0 &&
+                 cm->uv_ac_delta_q == 0;
+  if (xd->lossless) {
+    xd->itxm_add          = vp9_idct_add_lossless_c;
+  } else {
+    xd->itxm_add          = vp9_idct_add;
+  }
+}
+
+static INTERPOLATIONFILTERTYPE read_interp_filter_type(
+    struct vp9_read_bit_buffer *rb) {
+  return vp9_rb_read_bit(rb) ? SWITCHABLE
+                             : vp9_rb_read_literal(rb, 2);
+}
+
+static void read_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb,
+                            int *width, int *height) {
+  const int w = vp9_rb_read_literal(rb, 16) + 1;
+  const int h = vp9_rb_read_literal(rb, 16) + 1;
+  *width = w;
+  *height = h;
+}
+
+static void setup_display_size(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
+  VP9_COMMON *const cm = &pbi->common;
+  cm->display_width = cm->width;
+  cm->display_height = cm->height;
+  if (vp9_rb_read_bit(rb))
+    read_frame_size(cm, rb, &cm->display_width, &cm->display_height);
+}
+
+static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
+  VP9_COMMON *cm = &pbi->common;
+
+  if (cm->width != width || cm->height != height) {
+    if (!pbi->initial_width || !pbi->initial_height) {
+      if (vp9_alloc_frame_buffers(cm, width, height))
+        vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
+                           "Failed to allocate frame buffers");
+      pbi->initial_width = width;
+      pbi->initial_height = height;
+    } else {
+      if (width > pbi->initial_width)
+        vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+                           "Frame width too large");
+
+      if (height > pbi->initial_height)
+        vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+                           "Frame height too large");
+    }
+
+    cm->width = width;
+    cm->height = height;
+
+    vp9_update_frame_size(cm);
+  }
+
+  vp9_realloc_frame_buffer(&cm->yv12_fb[cm->new_fb_idx], cm->width, cm->height,
+                           cm->subsampling_x, cm->subsampling_y,
+                           VP9BORDERINPIXELS);
+}
+
+static void setup_frame_size(VP9D_COMP *pbi,
+                             struct vp9_read_bit_buffer *rb) {
+  VP9_COMMON *const cm = &pbi->common;
+  int width, height;
+  read_frame_size(cm, rb, &width, &height);
+  setup_display_size(pbi, rb);
+  apply_frame_size(pbi, width, height);
+}
+
+static void setup_frame_size_with_refs(VP9D_COMP *pbi,
+                                       struct vp9_read_bit_buffer *rb) {
+  VP9_COMMON *const cm = &pbi->common;
+
+  int width, height;
+  int found = 0, i;
+  for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
+    if (vp9_rb_read_bit(rb)) {
+      YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->active_ref_idx[i]];
+      width = cfg->y_crop_width;
+      height = cfg->y_crop_height;
+      found = 1;
+      break;
+    }
+  }
+
+  if (!found)
+    read_frame_size(cm, rb, &width, &height);
+
+  if (!width || !height)
+    vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+                       "Referenced frame with invalid size");
+
+  setup_display_size(pbi, rb);
+  apply_frame_size(pbi, width, height);
+}
+
+static void update_frame_context(FRAME_CONTEXT *fc) {
+  vp9_copy(fc->pre_coef_probs, fc->coef_probs);
+  vp9_copy(fc->pre_y_mode_prob, fc->y_mode_prob);
+  vp9_copy(fc->pre_uv_mode_prob, fc->uv_mode_prob);
+  vp9_copy(fc->pre_partition_prob, fc->partition_prob[1]);
+  vp9_copy(fc->pre_intra_inter_prob, fc->intra_inter_prob);
+  vp9_copy(fc->pre_comp_inter_prob, fc->comp_inter_prob);
+  vp9_copy(fc->pre_single_ref_prob, fc->single_ref_prob);
+  vp9_copy(fc->pre_comp_ref_prob, fc->comp_ref_prob);
+  fc->pre_nmvc = fc->nmvc;
+  vp9_copy(fc->pre_switchable_interp_prob, fc->switchable_interp_prob);
+  vp9_copy(fc->pre_inter_mode_probs, fc->inter_mode_probs);
+  vp9_copy(fc->pre_tx_probs_8x8p, fc->tx_probs_8x8p);
+  vp9_copy(fc->pre_tx_probs_16x16p, fc->tx_probs_16x16p);
+  vp9_copy(fc->pre_tx_probs_32x32p, fc->tx_probs_32x32p);
+  vp9_copy(fc->pre_mbskip_probs, fc->mbskip_probs);
+
+  vp9_zero(fc->coef_counts);
+  vp9_zero(fc->eob_branch_counts);
+  vp9_zero(fc->y_mode_counts);
+  vp9_zero(fc->uv_mode_counts);
+  vp9_zero(fc->NMVcount);
+  vp9_zero(fc->inter_mode_counts);
+  vp9_zero(fc->partition_counts);
+  vp9_zero(fc->switchable_interp_count);
+  vp9_zero(fc->intra_inter_count);
+  vp9_zero(fc->comp_inter_count);
+  vp9_zero(fc->single_ref_count);
+  vp9_zero(fc->comp_ref_count);
+  vp9_zero(fc->tx_count_8x8p);
+  vp9_zero(fc->tx_count_16x16p);
+  vp9_zero(fc->tx_count_32x32p);
+  vp9_zero(fc->mbskip_count);
+}
+
+static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {
+  VP9_COMMON *const pc = &pbi->common;
+  int mi_row, mi_col;
+
+  for (mi_row = pc->cur_tile_mi_row_start;
+       mi_row < pc->cur_tile_mi_row_end; mi_row += 64 / MI_SIZE) {
+    // For a SB there are 2 left contexts, each pertaining to a MB row within
+    vpx_memset(&pc->left_context, 0, sizeof(pc->left_context));
+    vpx_memset(pc->left_seg_context, 0, sizeof(pc->left_seg_context));
+    for (mi_col = pc->cur_tile_mi_col_start;
+         mi_col < pc->cur_tile_mi_col_end; mi_col += 64 / MI_SIZE)
+      decode_modes_sb(pbi, mi_row, mi_col, r, BLOCK_SIZE_SB64X64);
+  }
+}
+
+static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
+  int delta_log2_tiles;
+
+  vp9_get_tile_n_bits(cm, &cm->log2_tile_columns, &delta_log2_tiles);
+  while (delta_log2_tiles--) {
+    if (vp9_rb_read_bit(rb)) {
+      cm->log2_tile_columns++;
+    } else {
+      break;
+    }
+  }
+
+  cm->log2_tile_rows = vp9_rb_read_bit(rb);
+  if (cm->log2_tile_rows)
+    cm->log2_tile_rows += vp9_rb_read_bit(rb);
+
+  cm->tile_columns = 1 << cm->log2_tile_columns;
+  cm->tile_rows    = 1 << cm->log2_tile_rows;
+}
+
+static void decode_tiles(VP9D_COMP *pbi,
+                         const uint8_t *data, size_t first_partition_size,
+                         vp9_reader *residual_bc) {
+  VP9_COMMON *const pc = &pbi->common;
+
+  const uint8_t *data_ptr = data + first_partition_size;
+  const uint8_t* const data_end = pbi->source + pbi->source_sz;
+  int tile_row, tile_col;
+
+  // Note: this memset assumes above_context[0], [1] and [2]
+  // are allocated as part of the same buffer.
+  vpx_memset(pc->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 2 *
+                                      MAX_MB_PLANE * mi_cols_aligned_to_sb(pc));
+
+  vpx_memset(pc->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
+                                       mi_cols_aligned_to_sb(pc));
+
+  if (pbi->oxcf.inv_tile_order) {
+    const int n_cols = pc->tile_columns;
+    const uint8_t *data_ptr2[4][1 << 6];
+    vp9_reader bc_bak = {0};
+
+    // pre-initialize the offsets, we're going to read in inverse order
+    data_ptr2[0][0] = data_ptr;
+    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+      if (tile_row) {
+        const int size = read_be32(data_ptr2[tile_row - 1][n_cols - 1]);
+        data_ptr2[tile_row - 1][n_cols - 1] += 4;
+        data_ptr2[tile_row][0] = data_ptr2[tile_row - 1][n_cols - 1] + size;
+      }
+
+      for (tile_col = 1; tile_col < n_cols; tile_col++) {
+        const int size = read_be32(data_ptr2[tile_row][tile_col - 1]);
+        data_ptr2[tile_row][tile_col - 1] += 4;
+        data_ptr2[tile_row][tile_col] =
+            data_ptr2[tile_row][tile_col - 1] + size;
+      }
+    }
+
+    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+      vp9_get_tile_row_offsets(pc, tile_row);
+      for (tile_col = n_cols - 1; tile_col >= 0; tile_col--) {
+        vp9_get_tile_col_offsets(pc, tile_col);
+        setup_token_decoder(pbi, data_ptr2[tile_row][tile_col],
+                            data_end - data_ptr2[tile_row][tile_col],
+                            residual_bc);
+        decode_tile(pbi, residual_bc);
+        if (tile_row == pc->tile_rows - 1 && tile_col == n_cols - 1)
+          bc_bak = *residual_bc;
+      }
+    }
+    *residual_bc = bc_bak;
+  } else {
+    int has_more;
+
+    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+      vp9_get_tile_row_offsets(pc, tile_row);
+      for (tile_col = 0; tile_col < pc->tile_columns; tile_col++) {
+        size_t size;
+
+        vp9_get_tile_col_offsets(pc, tile_col);
+
+        has_more = tile_col < pc->tile_columns - 1 ||
+                   tile_row < pc->tile_rows - 1;
+        if (has_more) {
+          if (!read_is_valid(data_ptr, 4, data_end))
+            vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
+                         "Truncated packet or corrupt tile length");
+
+          size = read_be32(data_ptr);
+          data_ptr += 4;
+        } else {
+          size = data_end - data_ptr;
+        }
+
+        setup_token_decoder(pbi, data_ptr, size, residual_bc);
+        decode_tile(pbi, residual_bc);
+        data_ptr += size;
+      }
+    }
+  }
+}
+
+static void check_sync_code(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
+  if (vp9_rb_read_literal(rb, 8) != SYNC_CODE_0 ||
+      vp9_rb_read_literal(rb, 8) != SYNC_CODE_1 ||
+      vp9_rb_read_literal(rb, 8) != SYNC_CODE_2) {
+    vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
+                       "Invalid frame sync code");
+  }
+}
+
+static void error_handler(void *data, size_t bit_offset) {
+  VP9_COMMON *const cm = (VP9_COMMON *)data;
+  vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet");
+}
+
+static void setup_inter_inter(VP9_COMMON *cm) {
+  int i;
+
+  cm->allow_comp_inter_inter = 0;
+  for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
+    cm->allow_comp_inter_inter |= i > 0 &&
+        cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1];
+  }
+
+  if (cm->allow_comp_inter_inter) {
+    // which one is always-on in comp inter-inter?
+    if (cm->ref_frame_sign_bias[LAST_FRAME] ==
+        cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
+      cm->comp_fixed_ref = ALTREF_FRAME;
+      cm->comp_var_ref[0] = LAST_FRAME;
+      cm->comp_var_ref[1] = GOLDEN_FRAME;
+    } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
+               cm->ref_frame_sign_bias[ALTREF_FRAME]) {
+      cm->comp_fixed_ref = GOLDEN_FRAME;
+      cm->comp_var_ref[0] = LAST_FRAME;
+      cm->comp_var_ref[1] = ALTREF_FRAME;
+    } else {
+      cm->comp_fixed_ref = LAST_FRAME;
+      cm->comp_var_ref[0] = GOLDEN_FRAME;
+      cm->comp_var_ref[1] = ALTREF_FRAME;
+    }
+  }
+}
+
+#define RESERVED \
+  if (vp9_rb_read_bit(rb)) \
+      vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, \
+                         "Reserved bit must be unset")
+
+static size_t read_uncompressed_header(VP9D_COMP *pbi,
+                                       struct vp9_read_bit_buffer *rb) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  int i;
+
+  cm->last_frame_type = cm->frame_type;
+
+  if (vp9_rb_read_literal(rb, 2) != 0x2)
+      vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
+                         "Invalid frame marker");
+
+  cm->version = vp9_rb_read_bit(rb);
+  RESERVED;
+
+  if (vp9_rb_read_bit(rb)) {
+    // show an existing frame directly
+    int frame_to_show = cm->ref_frame_map[vp9_rb_read_literal(rb, 3)];
+    ref_cnt_fb(cm->fb_idx_ref_cnt, &cm->new_fb_idx, frame_to_show);
+    pbi->refresh_frame_flags = 0;
+    cm->filter_level = 0;
+    return 0;
+  }
+
+  cm->frame_type = (FRAME_TYPE) vp9_rb_read_bit(rb);
+  cm->show_frame = vp9_rb_read_bit(rb);
+  cm->error_resilient_mode = vp9_rb_read_bit(rb);
+
+  if (cm->frame_type == KEY_FRAME) {
+    int csp;
+
+    check_sync_code(cm, rb);
+
+    csp = vp9_rb_read_literal(rb, 3);  // colorspace
+    if (csp != 7) {  // != sRGB
+      vp9_rb_read_bit(rb);  // [16,235] (including xvycc) vs [0,255] range
+      if (cm->version == 1) {
+        cm->subsampling_x = vp9_rb_read_bit(rb);
+        cm->subsampling_y = vp9_rb_read_bit(rb);
+        vp9_rb_read_bit(rb);  // has extra plane
+      } else {
+        cm->subsampling_y = cm->subsampling_x = 1;
+      }
+    } else {
+      if (cm->version == 1) {
+        cm->subsampling_y = cm->subsampling_x = 0;
+        vp9_rb_read_bit(rb);  // has extra plane
+      } else {
+        vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
+                           "RGB not supported in profile 0");
+      }
+    }
+
+    pbi->refresh_frame_flags = (1 << NUM_REF_FRAMES) - 1;
+
+    for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i)
+      cm->active_ref_idx[i] = cm->new_fb_idx;
+
+    setup_frame_size(pbi, rb);
+  } else {
+    cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb);
+
+    cm->reset_frame_context = cm->error_resilient_mode ?
+        0 : vp9_rb_read_literal(rb, 2);
+
+    if (cm->intra_only) {
+      check_sync_code(cm, rb);
+
+      pbi->refresh_frame_flags = vp9_rb_read_literal(rb, NUM_REF_FRAMES);
+      setup_frame_size(pbi, rb);
+    } else {
+       pbi->refresh_frame_flags = vp9_rb_read_literal(rb, NUM_REF_FRAMES);
+
+      for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
+        const int ref = vp9_rb_read_literal(rb, NUM_REF_FRAMES_LG2);
+        cm->active_ref_idx[i] = cm->ref_frame_map[ref];
+        cm->ref_frame_sign_bias[LAST_FRAME + i] = vp9_rb_read_bit(rb);
+      }
+
+      setup_frame_size_with_refs(pbi, rb);
+
+      xd->allow_high_precision_mv = vp9_rb_read_bit(rb);
+      cm->mcomp_filter_type = read_interp_filter_type(rb);
+
+      for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i)
+        vp9_setup_scale_factors(cm, i);
+
+      setup_inter_inter(cm);
+    }
+  }
+
+  if (!cm->error_resilient_mode) {
+    cm->refresh_frame_context = vp9_rb_read_bit(rb);
+    cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb);
+  } else {
+    cm->refresh_frame_context = 0;
+    cm->frame_parallel_decoding_mode = 1;
+  }
+
+  cm->frame_context_idx = vp9_rb_read_literal(rb, NUM_FRAME_CONTEXTS_LG2);
+
+  if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode || cm->intra_only)
+    vp9_setup_past_independence(cm, xd);
+
+  setup_loopfilter(pbi, rb);
+  setup_quantization(pbi, rb);
+  setup_segmentation(pbi, rb);
+
+  setup_tile_info(cm, rb);
+
+  return vp9_rb_read_literal(rb, 16);
+}
+
+int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
+  int i;
+  vp9_reader header_bc, residual_bc;
+  VP9_COMMON *const pc = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+
+  const uint8_t *data = pbi->source;
+  const uint8_t *data_end = pbi->source + pbi->source_sz;
+
+  struct vp9_read_bit_buffer rb = { data, data_end, 0,
+                                    pc, error_handler };
+  const size_t first_partition_size = read_uncompressed_header(pbi, &rb);
+  const int keyframe = pc->frame_type == KEY_FRAME;
+  YV12_BUFFER_CONFIG *new_fb = &pc->yv12_fb[pc->new_fb_idx];
+
+  if (!first_partition_size) {
+    // showing a frame directly
+    *p_data_end = data + 1;
+    return 0;
+  }
+  data += vp9_rb_bytes_read(&rb);
+  xd->corrupted = 0;
+  new_fb->corrupted = 0;
+
+  if (!pbi->decoded_key_frame && !keyframe)
+    return -1;
+
+  if (!read_is_valid(data, first_partition_size, data_end))
+    vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
+                       "Truncated packet or corrupt header length");
+
+  xd->mode_info_context = pc->mi;
+  xd->prev_mode_info_context = pc->prev_mi;
+  xd->frame_type = pc->frame_type;
+  xd->mode_info_stride = pc->mode_info_stride;
+
+  if (vp9_reader_init(&header_bc, data, first_partition_size))
+    vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate bool decoder 0");
+
+  mb_init_dequantizer(pc, &pbi->mb);  // MB level dequantizer setup
+
+  if (!keyframe)
+    vp9_setup_interp_filters(xd, pc->mcomp_filter_type, pc);
+
+  pc->fc = pc->frame_contexts[pc->frame_context_idx];
+
+  update_frame_context(&pc->fc);
+
+  setup_txfm_mode(pc, xd->lossless, &header_bc);
+
+  read_coef_probs(pbi, &header_bc);
+
+  // Initialize xd pointers. Any reference should do for xd->pre, so use 0.
+  setup_pre_planes(xd, &pc->yv12_fb[pc->active_ref_idx[0]], NULL,
+                   0, 0, NULL, NULL);
+  setup_dst_planes(xd, new_fb, 0, 0);
+
+  // Create the segmentation map structure and set to 0
+  if (!pc->last_frame_seg_map)
+    CHECK_MEM_ERROR(pc->last_frame_seg_map,
+                    vpx_calloc((pc->mi_rows * pc->mi_cols), 1));
+
+  vp9_setup_block_dptrs(xd, pc->subsampling_x, pc->subsampling_y);
+
+  // clear out the coeff buffer
+  for (i = 0; i < MAX_MB_PLANE; ++i)
+    vp9_zero(xd->plane[i].qcoeff);
+
+  set_prev_mi(pc);
+
+  vp9_decode_mode_mvs_init(pbi, &header_bc);
+
+  decode_tiles(pbi, data, first_partition_size, &residual_bc);
+
+  pc->last_width = pc->width;
+  pc->last_height = pc->height;
+
+  new_fb->corrupted = vp9_reader_has_error(&header_bc) | xd->corrupted;
+
+  if (!pbi->decoded_key_frame) {
+    if (keyframe && !new_fb->corrupted)
+      pbi->decoded_key_frame = 1;
+    else
+      vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
+                         "A stream must start with a complete key frame");
+  }
+
+  // Adaptation
+  if (!pc->error_resilient_mode && !pc->frame_parallel_decoding_mode) {
+    vp9_adapt_coef_probs(pc);
+
+    if ((!keyframe) && (!pc->intra_only)) {
+      vp9_adapt_mode_probs(pc);
+      vp9_adapt_mode_context(pc);
+      vp9_adapt_nmv_probs(pc, xd->allow_high_precision_mv);
+    }
+  }
+
+  if (pc->refresh_frame_context)
+    pc->frame_contexts[pc->frame_context_idx] = pc->fc;
+
+  *p_data_end = vp9_reader_find_end(&residual_bc);
+  return 0;
+}
diff --git a/libvpx/vp9/decoder/vp9_decodframe.h b/libvpx/vp9/decoder/vp9_decodframe.h
new file mode 100644
index 0000000..66e951d
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_decodframe.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_DECODER_VP9_DECODFRAME_H_
+#define VP9_DECODER_VP9_DECODFRAME_H_
+
+struct VP9Common;
+struct VP9Decompressor;
+
+void vp9_init_dequantizer(struct VP9Common *pc);
+int vp9_decode_frame(struct VP9Decompressor *cpi, const uint8_t **p_data_end);
+vp9_prob vp9_read_prob_diff_update(vp9_reader *r, int oldp);
+
+#endif  // VP9_DECODER_VP9_DECODFRAME_H_
diff --git a/libvpx/vp9/decoder/vp9_detokenize.c b/libvpx/vp9/decoder/vp9_detokenize.c
new file mode 100644
index 0000000..3bbb212
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_detokenize.c
@@ -0,0 +1,347 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/mem.h"
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_seg_common.h"
+
+#include "vp9/decoder/vp9_detokenize.h"
+#include "vp9/decoder/vp9_onyxd_int.h"
+
+#if CONFIG_BALANCED_COEFTREE
+#define ZERO_CONTEXT_NODE           0
+#define EOB_CONTEXT_NODE            1
+#else
+#define EOB_CONTEXT_NODE            0
+#define ZERO_CONTEXT_NODE           1
+#endif
+
+#define ONE_CONTEXT_NODE            2
+#define LOW_VAL_CONTEXT_NODE        3
+#define TWO_CONTEXT_NODE            4
+#define THREE_CONTEXT_NODE          5
+#define HIGH_LOW_CONTEXT_NODE       6
+#define CAT_ONE_CONTEXT_NODE        7
+#define CAT_THREEFOUR_CONTEXT_NODE  8
+#define CAT_THREE_CONTEXT_NODE      9
+#define CAT_FIVE_CONTEXT_NODE       10
+
+#define CAT1_MIN_VAL    5
+#define CAT2_MIN_VAL    7
+#define CAT3_MIN_VAL   11
+#define CAT4_MIN_VAL   19
+#define CAT5_MIN_VAL   35
+#define CAT6_MIN_VAL   67
+#define CAT1_PROB0    159
+#define CAT2_PROB0    145
+#define CAT2_PROB1    165
+
+#define CAT3_PROB0 140
+#define CAT3_PROB1 148
+#define CAT3_PROB2 173
+
+#define CAT4_PROB0 135
+#define CAT4_PROB1 140
+#define CAT4_PROB2 155
+#define CAT4_PROB3 176
+
+#define CAT5_PROB0 130
+#define CAT5_PROB1 134
+#define CAT5_PROB2 141
+#define CAT5_PROB3 157
+#define CAT5_PROB4 180
+
+static const vp9_prob cat6_prob[15] = {
+  254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129, 0
+};
+
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+#define INCREMENT_COUNT(token)               \
+  do {                                       \
+    coef_counts[type][ref][band][pt]         \
+               [token >= TWO_TOKEN ?     \
+                (token == DCT_EOB_TOKEN ? DCT_EOB_MODEL_TOKEN : TWO_TOKEN) : \
+                token]++;     \
+    token_cache[scan[c]] = vp9_pt_energy_class[token]; \
+  } while (0)
+
+#define WRITE_COEF_CONTINUE(val, token)                  \
+  {                                                      \
+    qcoeff_ptr[scan[c]] = vp9_read_and_apply_sign(r, val) * \
+                            dq[c > 0] / (1 + (txfm_size == TX_32X32)); \
+    INCREMENT_COUNT(token);                              \
+    c++;                                                 \
+    continue;                                            \
+  }
+
+#define ADJUST_COEF(prob, bits_count)  \
+  do {                                 \
+    if (vp9_read(r, prob))             \
+      val += 1 << bits_count;          \
+  } while (0);
+
+static int decode_coefs(FRAME_CONTEXT *fc, const MACROBLOCKD *xd,
+                        vp9_reader *r, int block_idx,
+                        PLANE_TYPE type, int seg_eob, int16_t *qcoeff_ptr,
+                        TX_SIZE txfm_size, const int16_t *dq,
+                        ENTROPY_CONTEXT *A, ENTROPY_CONTEXT *L) {
+  ENTROPY_CONTEXT above_ec, left_ec;
+  int pt, c = 0, pad, default_eob;
+  int band;
+  vp9_prob (*coef_probs)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES];
+  vp9_prob coef_probs_full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES];
+  uint8_t load_map[COEF_BANDS][PREV_COEF_CONTEXTS] = {
+    {0, 0, 0, 0, 0, 0},
+    {0, 0, 0, 0, 0, 0},
+    {0, 0, 0, 0, 0, 0},
+    {0, 0, 0, 0, 0, 0},
+    {0, 0, 0, 0, 0, 0},
+    {0, 0, 0, 0, 0, 0},
+  };
+
+  vp9_prob *prob;
+  vp9_coeff_count_model *coef_counts;
+  const int ref = xd->mode_info_context->mbmi.ref_frame[0] != INTRA_FRAME;
+  TX_TYPE tx_type = DCT_DCT;
+  const int *scan, *nb;
+  uint8_t token_cache[1024];
+  const uint8_t * band_translate;
+#if CONFIG_BALANCED_COEFTREE
+  int skip_eob_node = 0;
+#endif
+
+  coef_probs  = fc->coef_probs[txfm_size][type][ref];
+  coef_counts = fc->coef_counts[txfm_size];
+  switch (txfm_size) {
+    default:
+    case TX_4X4: {
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_4x4(xd, block_idx) : DCT_DCT;
+      scan = get_scan_4x4(tx_type);
+      above_ec = A[0] != 0;
+      left_ec = L[0] != 0;
+      default_eob = 16;
+      band_translate = vp9_coefband_trans_4x4;
+      break;
+    }
+    case TX_8X8: {
+      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
+      const int sz = 1 + b_width_log2(sb_type);
+      const int x = block_idx & ((1 << sz) - 1);
+      const int y = block_idx - x;
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
+      scan = get_scan_8x8(tx_type);
+      above_ec = (A[0] + A[1]) != 0;
+      left_ec = (L[0] + L[1]) != 0;
+      default_eob = 64;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_16X16: {
+      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
+      const int sz = 2 + b_width_log2(sb_type);
+      const int x = block_idx & ((1 << sz) - 1);
+      const int y = block_idx - x;
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
+      scan = get_scan_16x16(tx_type);
+      above_ec = (A[0] + A[1] + A[2] + A[3]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3]) != 0;
+      default_eob = 256;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_32X32:
+      scan = vp9_default_scan_32x32;
+      above_ec = (A[0] + A[1] + A[2] + A[3] + A[4] + A[5] + A[6] + A[7]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3] + L[4] + L[5] + L[6] + L[7]) != 0;
+      default_eob = 1024;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+  }
+
+  pt = combine_entropy_contexts(above_ec, left_ec);
+  nb = vp9_get_coef_neighbors_handle(scan, &pad);
+
+  while (1) {
+    int val;
+    const uint8_t *cat6 = cat6_prob;
+    if (c >= seg_eob)
+      break;
+    if (c)
+      pt = vp9_get_coef_context(scan, nb, pad, token_cache,
+                                c, default_eob);
+    band = get_coef_band(band_translate, c);
+    prob = coef_probs[band][pt];
+#if !CONFIG_BALANCED_COEFTREE
+    fc->eob_branch_counts[txfm_size][type][ref][band][pt]++;
+    if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
+      break;
+
+SKIP_START:
+#endif
+    if (c >= seg_eob)
+      break;
+    if (c)
+      pt = vp9_get_coef_context(scan, nb, pad, token_cache,
+                                c, default_eob);
+    band = get_coef_band(band_translate, c);
+    prob = coef_probs[band][pt];
+
+    if (!vp9_read(r, prob[ZERO_CONTEXT_NODE])) {
+      INCREMENT_COUNT(ZERO_TOKEN);
+      ++c;
+#if CONFIG_BALANCED_COEFTREE
+      skip_eob_node = 1;
+      continue;
+#else
+      goto SKIP_START;
+#endif
+    }
+#if CONFIG_BALANCED_COEFTREE
+    if (!skip_eob_node) {
+      fc->eob_branch_counts[txfm_size][type][ref][band][pt]++;
+      if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
+        break;
+    }
+    skip_eob_node = 0;
+#endif
+
+    // ONE_CONTEXT_NODE_0_
+    if (!vp9_read(r, prob[ONE_CONTEXT_NODE])) {
+      WRITE_COEF_CONTINUE(1, ONE_TOKEN);
+    }
+    // Load full probabilities if not already loaded
+    if (!load_map[band][pt]) {
+      vp9_model_to_full_probs(coef_probs[band][pt],
+                              coef_probs_full[band][pt]);
+      load_map[band][pt] = 1;
+    }
+    prob = coef_probs_full[band][pt];
+    // LOW_VAL_CONTEXT_NODE_0_
+    if (!vp9_read(r, prob[LOW_VAL_CONTEXT_NODE])) {
+      if (!vp9_read(r, prob[TWO_CONTEXT_NODE])) {
+        WRITE_COEF_CONTINUE(2, TWO_TOKEN);
+      }
+      if (!vp9_read(r, prob[THREE_CONTEXT_NODE])) {
+        WRITE_COEF_CONTINUE(3, THREE_TOKEN);
+      }
+      WRITE_COEF_CONTINUE(4, FOUR_TOKEN);
+    }
+    // HIGH_LOW_CONTEXT_NODE_0_
+    if (!vp9_read(r, prob[HIGH_LOW_CONTEXT_NODE])) {
+      if (!vp9_read(r, prob[CAT_ONE_CONTEXT_NODE])) {
+        val = CAT1_MIN_VAL;
+        ADJUST_COEF(CAT1_PROB0, 0);
+        WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY1);
+      }
+      val = CAT2_MIN_VAL;
+      ADJUST_COEF(CAT2_PROB1, 1);
+      ADJUST_COEF(CAT2_PROB0, 0);
+      WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY2);
+    }
+    // CAT_THREEFOUR_CONTEXT_NODE_0_
+    if (!vp9_read(r, prob[CAT_THREEFOUR_CONTEXT_NODE])) {
+      if (!vp9_read(r, prob[CAT_THREE_CONTEXT_NODE])) {
+        val = CAT3_MIN_VAL;
+        ADJUST_COEF(CAT3_PROB2, 2);
+        ADJUST_COEF(CAT3_PROB1, 1);
+        ADJUST_COEF(CAT3_PROB0, 0);
+        WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY3);
+      }
+      val = CAT4_MIN_VAL;
+      ADJUST_COEF(CAT4_PROB3, 3);
+      ADJUST_COEF(CAT4_PROB2, 2);
+      ADJUST_COEF(CAT4_PROB1, 1);
+      ADJUST_COEF(CAT4_PROB0, 0);
+      WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY4);
+    }
+    // CAT_FIVE_CONTEXT_NODE_0_:
+    if (!vp9_read(r, prob[CAT_FIVE_CONTEXT_NODE])) {
+      val = CAT5_MIN_VAL;
+      ADJUST_COEF(CAT5_PROB4, 4);
+      ADJUST_COEF(CAT5_PROB3, 3);
+      ADJUST_COEF(CAT5_PROB2, 2);
+      ADJUST_COEF(CAT5_PROB1, 1);
+      ADJUST_COEF(CAT5_PROB0, 0);
+      WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY5);
+    }
+    val = 0;
+    while (*cat6) {
+      val = (val << 1) | vp9_read(r, *cat6++);
+    }
+    val += CAT6_MIN_VAL;
+    WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY6);
+  }
+
+  if (c < seg_eob)
+    coef_counts[type][ref][band][pt][DCT_EOB_MODEL_TOKEN]++;
+
+
+  return c;
+}
+
+static int get_eob(MACROBLOCKD* const xd, int segment_id, int eob_max) {
+  return vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
+}
+
+struct decode_block_args {
+  VP9D_COMP *pbi;
+  vp9_reader *r;
+  int *eobtotal;
+};
+
+static void decode_block(int plane, int block,
+                         BLOCK_SIZE_TYPE bsize,
+                         int ss_txfrm_size,
+                         void *argv) {
+  const struct decode_block_args* const arg = argv;
+  const int bw = b_width_log2(bsize);
+
+  // find the maximum eob for this transform size, adjusted by segment
+  MACROBLOCKD *xd = &arg->pbi->mb;
+  struct macroblockd_plane* pd = &xd->plane[plane];
+  const int segment_id = xd->mode_info_context->mbmi.segment_id;
+  const TX_SIZE ss_tx_size = ss_txfrm_size / 2;
+  const int seg_eob = get_eob(xd, segment_id, 16 << ss_txfrm_size);
+  const int off = block >> ss_txfrm_size;
+  const int mod = bw - ss_tx_size - pd->subsampling_x;
+  const int aoff = (off & ((1 << mod) - 1)) << ss_tx_size;
+  const int loff = (off >> mod) << ss_tx_size;
+
+  ENTROPY_CONTEXT *A = pd->above_context + aoff;
+  ENTROPY_CONTEXT *L = pd->left_context + loff;
+  const int eob = decode_coefs(&arg->pbi->common.fc, xd, arg->r, block,
+                               pd->plane_type, seg_eob,
+                               BLOCK_OFFSET(pd->qcoeff, block, 16),
+                               ss_tx_size, pd->dequant, A, L);
+
+  if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
+    set_contexts_on_border(xd, bsize, plane, ss_tx_size, eob, aoff, loff, A, L);
+  } else {
+    int pt;
+    for (pt = 0; pt < (1 << ss_tx_size); pt++)
+      A[pt] = L[pt] = eob > 0;
+  }
+  pd->eobs[block] = eob;
+  *arg->eobtotal += eob;
+}
+
+int vp9_decode_tokens(VP9D_COMP *pbi, vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
+  int eobtotal = 0;
+  struct decode_block_args args = {pbi, r, &eobtotal};
+  foreach_transformed_block(&pbi->mb, bsize, decode_block, &args);
+  return eobtotal;
+}
diff --git a/libvpx/vp9/decoder/vp9_detokenize.h b/libvpx/vp9/decoder/vp9_detokenize.h
new file mode 100644
index 0000000..d46b596
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_detokenize.h
@@ -0,0 +1,19 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_DECODER_VP9_DETOKENIZE_H_
+#define VP9_DECODER_VP9_DETOKENIZE_H_
+
+#include "vp9/decoder/vp9_onyxd_int.h"
+
+int vp9_decode_tokens(VP9D_COMP* pbi, vp9_reader *r, BLOCK_SIZE_TYPE bsize);
+
+#endif  // VP9_DECODER_VP9_DETOKENIZE_H_
diff --git a/libvpx/vp9/decoder/vp9_idct_blk.c b/libvpx/vp9/decoder/vp9_idct_blk.c
new file mode 100644
index 0000000..c52963c
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_idct_blk.c
@@ -0,0 +1,159 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/decoder/vp9_idct_blk.h"
+
+static void add_constant_residual(const int16_t diff, uint8_t *dest, int stride,
+                                  int width, int height) {
+  int r, c;
+
+  for (r = 0; r < height; r++) {
+    for (c = 0; c < width; c++)
+      dest[c] = clip_pixel(diff + dest[c]);
+
+    dest += stride;
+  }
+}
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest,
+                                     int stride) {
+  add_constant_residual(diff, dest, stride, 8, 8);
+}
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest,
+                                       int stride) {
+  add_constant_residual(diff, dest, stride, 16, 16);
+}
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff,  uint8_t *dest,
+                                       int stride) {
+  add_constant_residual(diff, dest, stride, 32, 32);
+}
+
+void vp9_iht_add_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest, int stride,
+                   int eob) {
+  if (tx_type == DCT_DCT) {
+    vp9_idct_add(input, dest, stride, eob);
+  } else {
+    vp9_short_iht4x4_add(input, dest, stride, tx_type);
+    vpx_memset(input, 0, 32);
+  }
+}
+
+void vp9_iht_add_8x8_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+                       int stride, int eob) {
+  if (tx_type == DCT_DCT) {
+    vp9_idct_add_8x8(input, dest, stride, eob);
+  } else {
+    if (eob > 0) {
+      vp9_short_iht8x8_add(input, dest, stride, tx_type);
+      vpx_memset(input, 0, 128);
+    }
+  }
+}
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob) {
+  if (eob > 1) {
+    vp9_short_idct4x4_add(input, dest, stride);
+    vpx_memset(input, 0, 32);
+  } else {
+    vp9_dc_only_idct_add(input[0], dest, dest, stride, stride);
+    ((int *)input)[0] = 0;
+  }
+}
+
+void vp9_idct_add_lossless_c(int16_t *input, uint8_t *dest, int stride,
+                             int eob) {
+  if (eob > 1) {
+    vp9_short_iwalsh4x4_add(input, dest, stride);
+    vpx_memset(input, 0, 32);
+  } else {
+    vp9_short_iwalsh4x4_1_add_c(input, dest, stride);
+    ((int *)input)[0] = 0;
+  }
+}
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob) {
+  // If dc is 1, then input[0] is the reconstructed value, do not need
+  // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
+
+  // The calculation can be simplified if there are not many non-zero dct
+  // coefficients. Use eobs to decide what to do.
+  // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
+  // Combine that with code here.
+  if (eob) {
+    if (eob == 1) {
+      // DC only DCT coefficient
+      int16_t in = input[0];
+      int16_t out;
+
+      // Note: the idct1 will need to be modified accordingly whenever
+      // vp9_short_idct8x8_c() is modified.
+      vp9_short_idct1_8x8_c(&in, &out);
+      input[0] = 0;
+
+      vp9_add_constant_residual_8x8(out, dest, stride);
+    } else {
+      vp9_short_idct8x8_add(input, dest, stride);
+      vpx_memset(input, 0, 128);
+    }
+  }
+}
+
+void vp9_iht_add_16x16_c(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
+                         int stride, int eob) {
+  if (tx_type == DCT_DCT) {
+    vp9_idct_add_16x16(input, dest, stride, eob);
+  } else {
+    if (eob > 0) {
+      vp9_short_iht16x16_add(input, dest, stride, tx_type);
+      vpx_memset(input, 0, 512);
+    }
+  }
+}
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob) {
+  /* The calculation can be simplified if there are not many non-zero dct
+   * coefficients. Use eobs to separate different cases. */
+  if (eob) {
+    if (eob == 1) {
+      /* DC only DCT coefficient. */
+      int16_t in = input[0];
+      int16_t out;
+      /* Note: the idct1 will need to be modified accordingly whenever
+       * vp9_short_idct16x16() is modified. */
+      vp9_short_idct1_16x16_c(&in, &out);
+      input[0] = 0;
+
+      vp9_add_constant_residual_16x16(out, dest, stride);
+    } else {
+      vp9_short_idct16x16_add(input, dest, stride);
+      vpx_memset(input, 0, 512);
+    }
+  }
+}
+
+void vp9_idct_add_32x32_c(int16_t *input, uint8_t *dest, int stride, int eob) {
+  DECLARE_ALIGNED_ARRAY(16, int16_t, output, 1024);
+
+  if (eob) {
+    if (eob == 1) {
+      vp9_short_idct1_32x32(input, output);
+      vp9_add_constant_residual_32x32(output[0], dest, stride);
+      input[0] = 0;
+    } else {
+      vp9_short_idct32x32_add(input, dest, stride);
+      vpx_memset(input, 0, 2048);
+    }
+  }
+}
+
diff --git a/libvpx/vp9/decoder/vp9_idct_blk.h b/libvpx/vp9/decoder/vp9_idct_blk.h
new file mode 100644
index 0000000..1810bd0
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_idct_blk.h
@@ -0,0 +1,30 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_DECODER_VP9_IDCT_BLK_H_
+#define VP9_DECODER_VP9_IDCT_BLK_H_
+
+#include "vp9/common/vp9_blockd.h"
+
+
+void vp9_idct_add_lossless_c(int16_t *input, unsigned char *dest, int stride,
+                             int eob);
+
+void vp9_iht_add_c(TX_TYPE tx_type, int16_t *input, unsigned char *dest,
+                   int stride, int eob);
+
+void vp9_iht_add_8x8_c(TX_TYPE tx_type, int16_t *input, unsigned char *dest,
+                       int stride, int eob);
+
+void vp9_iht_add_16x16_c(TX_TYPE tx_type, int16_t *input, unsigned char *dest,
+                         int stride, int eob);
+
+#endif  // VP9_DECODER_VP9_IDCT_BLK_H_
diff --git a/libvpx/vp9/decoder/vp9_onyxd.h b/libvpx/vp9/decoder/vp9_onyxd.h
new file mode 100644
index 0000000..cd5b750
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_onyxd.h
@@ -0,0 +1,69 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_COMMON_VP9_ONYXD_H_
+#define VP9_COMMON_VP9_ONYXD_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "vpx_scale/yv12config.h"
+#include "vp9/common/vp9_ppflags.h"
+#include "vpx/vpx_codec.h"
+
+typedef void *VP9D_PTR;
+
+typedef struct {
+  int width;
+  int height;
+  int version;
+  int postprocess;
+  int max_threads;
+  int inv_tile_order;
+  int input_partition;
+} VP9D_CONFIG;
+
+typedef enum {
+  VP9_LAST_FLAG = 1,
+  VP9_GOLD_FLAG = 2,
+  VP9_ALT_FLAG = 4
+} VP9_REFFRAME;
+
+void vp9_initialize_dec();
+
+int vp9_receive_compressed_data(VP9D_PTR comp,
+                                uint64_t size, const uint8_t **dest,
+                                int64_t time_stamp);
+
+int vp9_get_raw_frame(VP9D_PTR comp, YV12_BUFFER_CONFIG *sd,
+                      int64_t *time_stamp, int64_t *time_end_stamp,
+                      vp9_ppflags_t *flags);
+
+vpx_codec_err_t vp9_copy_reference_dec(VP9D_PTR comp,
+                                       VP9_REFFRAME ref_frame_flag,
+                                       YV12_BUFFER_CONFIG *sd);
+
+vpx_codec_err_t vp9_set_reference_dec(VP9D_PTR comp,
+                                      VP9_REFFRAME ref_frame_flag,
+                                      YV12_BUFFER_CONFIG *sd);
+
+int vp9_get_reference_dec(VP9D_PTR ptr, int index, YV12_BUFFER_CONFIG **fb);
+
+
+VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf);
+
+void vp9_remove_decompressor(VP9D_PTR comp);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // VP9_COMMON_VP9_ONYXD_H_
diff --git a/libvpx/vp9/decoder/vp9_onyxd_if.c b/libvpx/vp9/decoder/vp9_onyxd_if.c
new file mode 100644
index 0000000..3cef88b
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_onyxd_if.c
@@ -0,0 +1,433 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdio.h>
+#include <assert.h>
+
+#include "vp9/common/vp9_onyxc_int.h"
+#if CONFIG_POSTPROC
+#include "vp9/common/vp9_postproc.h"
+#endif
+#include "vp9/decoder/vp9_onyxd.h"
+#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vpx_ports/vpx_timer.h"
+#include "vp9/decoder/vp9_decodframe.h"
+#include "vp9/decoder/vp9_detokenize.h"
+#include "./vpx_scale_rtcd.h"
+
+#define WRITE_RECON_BUFFER 0
+#if WRITE_RECON_BUFFER == 1
+static void recon_write_yuv_frame(const char *name,
+                                  const YV12_BUFFER_CONFIG *s,
+                                  int w, int _h) {
+  FILE *yuv_file = fopen(name, "ab");
+  const uint8_t *src = s->y_buffer;
+  int h = _h;
+
+  do {
+    fwrite(src, w, 1,  yuv_file);
+    src += s->y_stride;
+  } while (--h);
+
+  src = s->u_buffer;
+  h = (_h + 1) >> 1;
+  w = (w + 1) >> 1;
+
+  do {
+    fwrite(src, w, 1,  yuv_file);
+    src += s->uv_stride;
+  } while (--h);
+
+  src = s->v_buffer;
+  h = (_h + 1) >> 1;
+
+  do {
+    fwrite(src, w, 1, yuv_file);
+    src += s->uv_stride;
+  } while (--h);
+
+  fclose(yuv_file);
+}
+#endif
+#if WRITE_RECON_BUFFER == 2
+void write_dx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame) {
+
+  // write the frame
+  FILE *yframe;
+  int i;
+  char filename[255];
+
+  sprintf(filename, "dx\\y%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->y_height; i++)
+    fwrite(frame->y_buffer + i * frame->y_stride,
+           frame->y_width, 1, yframe);
+
+  fclose(yframe);
+  sprintf(filename, "dx\\u%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->uv_height; i++)
+    fwrite(frame->u_buffer + i * frame->uv_stride,
+           frame->uv_width, 1, yframe);
+
+  fclose(yframe);
+  sprintf(filename, "dx\\v%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->uv_height; i++)
+    fwrite(frame->v_buffer + i * frame->uv_stride,
+           frame->uv_width, 1, yframe);
+
+  fclose(yframe);
+}
+#endif
+
+void vp9_initialize_dec() {
+  static int init_done = 0;
+
+  if (!init_done) {
+    vp9_initialize_common();
+    vp9_init_quant_tables();
+    init_done = 1;
+  }
+}
+
+VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf) {
+  VP9D_COMP *const pbi = vpx_memalign(32, sizeof(VP9D_COMP));
+
+  if (!pbi)
+    return NULL;
+
+  vpx_memset(pbi, 0, sizeof(VP9D_COMP));
+
+  if (setjmp(pbi->common.error.jmp)) {
+    pbi->common.error.setjmp = 0;
+    vp9_remove_decompressor(pbi);
+    return NULL;
+  }
+
+  pbi->common.error.setjmp = 1;
+  vp9_initialize_dec();
+
+  vp9_create_common(&pbi->common);
+
+  pbi->oxcf = *oxcf;
+  pbi->common.current_video_frame = 0;
+  pbi->ready_for_new_data = 1;
+
+  // vp9_init_dequantizer() is first called here. Add check in
+  // frame_init_dequantizer() to avoid unnecessary calling of
+  // vp9_init_dequantizer() for every frame.
+  vp9_init_dequantizer(&pbi->common);
+
+  vp9_loop_filter_init(&pbi->common);
+
+  pbi->common.error.setjmp = 0;
+  pbi->decoded_key_frame = 0;
+
+  return pbi;
+}
+
+void vp9_remove_decompressor(VP9D_PTR ptr) {
+  VP9D_COMP *const pbi = (VP9D_COMP *)ptr;
+
+  if (!pbi)
+    return;
+
+  if (pbi->common.last_frame_seg_map)
+    vpx_free(pbi->common.last_frame_seg_map);
+
+  vp9_remove_common(&pbi->common);
+  vpx_free(pbi->mbc);
+  vpx_free(pbi);
+}
+
+static int equal_dimensions(YV12_BUFFER_CONFIG *a, YV12_BUFFER_CONFIG *b) {
+    return a->y_height == b->y_height && a->y_width == b->y_width &&
+           a->uv_height == b->uv_height && a->uv_width == b->uv_width;
+}
+
+vpx_codec_err_t vp9_copy_reference_dec(VP9D_PTR ptr,
+                                       VP9_REFFRAME ref_frame_flag,
+                                       YV12_BUFFER_CONFIG *sd) {
+  VP9D_COMP *pbi = (VP9D_COMP *) ptr;
+  VP9_COMMON *cm = &pbi->common;
+  int ref_fb_idx;
+
+  /* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
+   * encoder is using the frame buffers for. This is just a stub to keep the
+   * vpxenc --test-decode functionality working, and will be replaced in a
+   * later commit that adds VP9-specific controls for this functionality.
+   */
+  if (ref_frame_flag == VP9_LAST_FLAG) {
+    ref_fb_idx = pbi->common.ref_frame_map[0];
+  } else {
+    vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
+                       "Invalid reference frame");
+    return pbi->common.error.error_code;
+  }
+
+  if (!equal_dimensions(&cm->yv12_fb[ref_fb_idx], sd)) {
+    vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
+                       "Incorrect buffer dimensions");
+  } else {
+    vp8_yv12_copy_frame(&cm->yv12_fb[ref_fb_idx], sd);
+  }
+
+  return pbi->common.error.error_code;
+}
+
+
+vpx_codec_err_t vp9_set_reference_dec(VP9D_PTR ptr, VP9_REFFRAME ref_frame_flag,
+                                      YV12_BUFFER_CONFIG *sd) {
+  VP9D_COMP *pbi = (VP9D_COMP *) ptr;
+  VP9_COMMON *cm = &pbi->common;
+  int *ref_fb_ptr = NULL;
+
+  /* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
+   * encoder is using the frame buffers for. This is just a stub to keep the
+   * vpxenc --test-decode functionality working, and will be replaced in a
+   * later commit that adds VP9-specific controls for this functionality.
+   */
+  if (ref_frame_flag == VP9_LAST_FLAG)
+    ref_fb_ptr = &pbi->common.active_ref_idx[0];
+  else if (ref_frame_flag == VP9_GOLD_FLAG)
+    ref_fb_ptr = &pbi->common.active_ref_idx[1];
+  else if (ref_frame_flag == VP9_ALT_FLAG)
+    ref_fb_ptr = &pbi->common.active_ref_idx[2];
+  else {
+    vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
+                       "Invalid reference frame");
+    return pbi->common.error.error_code;
+  }
+
+  if (!equal_dimensions(&cm->yv12_fb[*ref_fb_ptr], sd)) {
+    vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
+                       "Incorrect buffer dimensions");
+  } else {
+    // Find an empty frame buffer.
+    const int free_fb = get_free_fb(cm);
+    // Decrease fb_idx_ref_cnt since it will be increased again in
+    // ref_cnt_fb() below.
+    cm->fb_idx_ref_cnt[free_fb]--;
+
+    // Manage the reference counters and copy image.
+    ref_cnt_fb(cm->fb_idx_ref_cnt, ref_fb_ptr, free_fb);
+    vp8_yv12_copy_frame(sd, &cm->yv12_fb[*ref_fb_ptr]);
+  }
+
+  return pbi->common.error.error_code;
+}
+
+
+int vp9_get_reference_dec(VP9D_PTR ptr, int index, YV12_BUFFER_CONFIG **fb) {
+  VP9D_COMP *pbi = (VP9D_COMP *) ptr;
+  VP9_COMMON *cm = &pbi->common;
+
+  if (index < 0 || index >= NUM_REF_FRAMES)
+    return -1;
+
+  *fb = &cm->yv12_fb[cm->ref_frame_map[index]];
+  return 0;
+}
+
+/* If any buffer updating is signaled it should be done here. */
+static void swap_frame_buffers(VP9D_COMP *pbi) {
+  int ref_index = 0, mask;
+
+  for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
+    if (mask & 1) {
+      ref_cnt_fb(pbi->common.fb_idx_ref_cnt,
+                 &pbi->common.ref_frame_map[ref_index],
+                 pbi->common.new_fb_idx);
+    }
+    ++ref_index;
+  }
+
+  pbi->common.frame_to_show = &pbi->common.yv12_fb[pbi->common.new_fb_idx];
+  pbi->common.fb_idx_ref_cnt[pbi->common.new_fb_idx]--;
+
+  /* Invalidate these references until the next frame starts. */
+  for (ref_index = 0; ref_index < 3; ref_index++)
+    pbi->common.active_ref_idx[ref_index] = INT_MAX;
+}
+
+int vp9_receive_compressed_data(VP9D_PTR ptr,
+                                uint64_t size, const uint8_t **psource,
+                                int64_t time_stamp) {
+  VP9D_COMP *pbi = (VP9D_COMP *) ptr;
+  VP9_COMMON *cm = &pbi->common;
+  const uint8_t *source = *psource;
+  int retcode = 0;
+
+  /*if(pbi->ready_for_new_data == 0)
+      return -1;*/
+
+  if (ptr == 0)
+    return -1;
+
+  pbi->common.error.error_code = VPX_CODEC_OK;
+
+  pbi->source = source;
+  pbi->source_sz = size;
+
+  if (pbi->source_sz == 0) {
+    /* This is used to signal that we are missing frames.
+     * We do not know if the missing frame(s) was supposed to update
+     * any of the reference buffers, but we act conservative and
+     * mark only the last buffer as corrupted.
+     *
+     * TODO(jkoleszar): Error concealment is undefined and non-normative
+     * at this point, but if it becomes so, [0] may not always be the correct
+     * thing to do here.
+     */
+    if (cm->active_ref_idx[0] != INT_MAX)
+      cm->yv12_fb[cm->active_ref_idx[0]].corrupted = 1;
+  }
+
+  cm->new_fb_idx = get_free_fb(cm);
+
+  if (setjmp(pbi->common.error.jmp)) {
+    pbi->common.error.setjmp = 0;
+
+    /* We do not know if the missing frame(s) was supposed to update
+     * any of the reference buffers, but we act conservative and
+     * mark only the last buffer as corrupted.
+     *
+     * TODO(jkoleszar): Error concealment is undefined and non-normative
+     * at this point, but if it becomes so, [0] may not always be the correct
+     * thing to do here.
+     */
+    if (cm->active_ref_idx[0] != INT_MAX)
+      cm->yv12_fb[cm->active_ref_idx[0]].corrupted = 1;
+
+    if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0)
+      cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
+
+    return -1;
+  }
+
+  pbi->common.error.setjmp = 1;
+
+  retcode = vp9_decode_frame(pbi, psource);
+
+  if (retcode < 0) {
+    pbi->common.error.error_code = VPX_CODEC_ERROR;
+    pbi->common.error.setjmp = 0;
+    if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0)
+      cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
+    return retcode;
+  }
+
+  {
+    swap_frame_buffers(pbi);
+
+#if WRITE_RECON_BUFFER == 2
+    if (cm->show_frame)
+      write_dx_frame_to_file(cm->frame_to_show,
+                             cm->current_video_frame);
+    else
+      write_dx_frame_to_file(cm->frame_to_show,
+                             cm->current_video_frame + 1000);
+#endif
+
+    if (cm->filter_level) {
+      /* Apply the loop filter if appropriate. */
+      vp9_loop_filter_frame(cm, &pbi->mb, cm->filter_level, 0);
+    }
+
+#if WRITE_RECON_BUFFER == 2
+    if (cm->show_frame)
+      write_dx_frame_to_file(cm->frame_to_show,
+                             cm->current_video_frame + 2000);
+    else
+      write_dx_frame_to_file(cm->frame_to_show,
+                             cm->current_video_frame + 3000);
+#endif
+
+    vp9_extend_frame_borders(cm->frame_to_show,
+                             cm->subsampling_x, cm->subsampling_y);
+  }
+
+#if WRITE_RECON_BUFFER == 1
+  if (cm->show_frame)
+    recon_write_yuv_frame("recon.yuv", cm->frame_to_show,
+                          cm->width, cm->height);
+#endif
+
+  vp9_clear_system_state();
+
+  cm->last_show_frame = cm->show_frame;
+  if (cm->show_frame) {
+    // current mip will be the prev_mip for the next frame
+    MODE_INFO *temp = cm->prev_mip;
+    cm->prev_mip = cm->mip;
+    cm->mip = temp;
+
+    // update the upper left visible macroblock ptrs
+    cm->mi = cm->mip + cm->mode_info_stride + 1;
+    cm->prev_mi = cm->prev_mip + cm->mode_info_stride + 1;
+
+    cm->current_video_frame++;
+  }
+
+  pbi->ready_for_new_data = 0;
+  pbi->last_time_stamp = time_stamp;
+  pbi->source_sz = 0;
+
+  pbi->common.error.setjmp = 0;
+  return retcode;
+}
+
+int vp9_get_raw_frame(VP9D_PTR ptr, YV12_BUFFER_CONFIG *sd,
+                      int64_t *time_stamp, int64_t *time_end_stamp,
+                      vp9_ppflags_t *flags) {
+  int ret = -1;
+  VP9D_COMP *pbi = (VP9D_COMP *) ptr;
+
+  if (pbi->ready_for_new_data == 1)
+    return ret;
+
+  /* ie no raw frame to show!!! */
+  if (pbi->common.show_frame == 0)
+    return ret;
+
+  pbi->ready_for_new_data = 1;
+  *time_stamp = pbi->last_time_stamp;
+  *time_end_stamp = 0;
+
+  sd->clrtype = pbi->common.clr_type;
+#if CONFIG_POSTPROC
+  ret = vp9_post_proc_frame(&pbi->common, sd, flags);
+#else
+
+  if (pbi->common.frame_to_show) {
+    *sd = *pbi->common.frame_to_show;
+    sd->y_width = pbi->common.width;
+    sd->y_height = pbi->common.height;
+    sd->uv_height = pbi->common.height / 2;
+    ret = 0;
+  } else {
+    ret = -1;
+  }
+
+#endif /*!CONFIG_POSTPROC*/
+  vp9_clear_system_state();
+  return ret;
+}
diff --git a/libvpx/vp9/decoder/vp9_onyxd_int.h b/libvpx/vp9/decoder/vp9_onyxd_int.h
new file mode 100644
index 0000000..8698570
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_onyxd_int.h
@@ -0,0 +1,62 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_DECODER_VP9_ONYXD_INT_H_
+#define VP9_DECODER_VP9_ONYXD_INT_H_
+#include "./vpx_config.h"
+#include "vp9/decoder/vp9_onyxd.h"
+#include "vp9/decoder/vp9_treereader.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/decoder/vp9_idct_blk.h"
+
+// #define DEC_DEBUG
+
+typedef struct VP9Decompressor {
+  DECLARE_ALIGNED(16, MACROBLOCKD, mb);
+
+  DECLARE_ALIGNED(16, VP9_COMMON, common);
+
+  VP9D_CONFIG oxcf;
+
+  const uint8_t *source;
+  uint32_t source_sz;
+
+  vp9_reader *mbc;
+  int64_t last_time_stamp;
+  int ready_for_new_data;
+
+  int refresh_frame_flags;
+  vp9_prob prob_skip_false;
+
+  int decoded_key_frame;
+
+  int initial_width;
+  int initial_height;
+} VP9D_COMP;
+
+
+#if CONFIG_DEBUG
+#define CHECK_MEM_ERROR(lval,expr) do {\
+    lval = (expr); \
+    if(!lval) \
+      vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\
+                         "Failed to allocate "#lval" at %s:%d", \
+                         __FILE__,__LINE__);\
+  } while(0)
+#else
+#define CHECK_MEM_ERROR(lval,expr) do {\
+    lval = (expr); \
+    if(!lval) \
+      vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\
+                         "Failed to allocate "#lval);\
+  } while(0)
+#endif
+
+#endif  // VP9_DECODER_VP9_TREEREADER_H_
diff --git a/libvpx/vp9/decoder/vp9_read_bit_buffer.h b/libvpx/vp9/decoder/vp9_read_bit_buffer.h
new file mode 100644
index 0000000..f243cb4
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_read_bit_buffer.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_READ_BIT_BUFFER_
+#define VP9_READ_BIT_BUFFER_
+
+#include <limits.h>
+
+#include "vpx/vpx_integer.h"
+
+typedef void (*vp9_rb_error_handler)(void *data, size_t bit_offset);
+
+struct vp9_read_bit_buffer {
+  const uint8_t *bit_buffer;
+  const uint8_t *bit_buffer_end;
+  size_t bit_offset;
+
+  void *error_handler_data;
+  vp9_rb_error_handler error_handler;
+};
+
+static size_t vp9_rb_bytes_read(struct vp9_read_bit_buffer *rb) {
+  return rb->bit_offset / CHAR_BIT + (rb->bit_offset % CHAR_BIT > 0);
+}
+
+static int vp9_rb_read_bit(struct vp9_read_bit_buffer *rb) {
+  const size_t off = rb->bit_offset;
+  const size_t p = off / CHAR_BIT;
+  const int q = CHAR_BIT - 1 - (int)off % CHAR_BIT;
+  if (rb->bit_buffer + p >= rb->bit_buffer_end) {
+    rb->error_handler(rb->error_handler_data, rb->bit_offset);
+    return 0;
+  } else {
+    const int bit = (rb->bit_buffer[p] & (1 << q)) >> q;
+    rb->bit_offset = off + 1;
+    return bit;
+  }
+}
+
+static int vp9_rb_read_literal(struct vp9_read_bit_buffer *rb, int bits) {
+  int value = 0, bit;
+  for (bit = bits - 1; bit >= 0; bit--)
+    value |= vp9_rb_read_bit(rb) << bit;
+  return value;
+}
+
+#endif  // VP9_READ_BIT_BUFFER_
diff --git a/libvpx/vp9/decoder/vp9_treereader.h b/libvpx/vp9/decoder/vp9_treereader.h
new file mode 100644
index 0000000..4535688
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_treereader.h
@@ -0,0 +1,32 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_DECODER_VP9_TREEREADER_H_
+#define VP9_DECODER_VP9_TREEREADER_H_
+
+#include "vp9/common/vp9_treecoder.h"
+#include "vp9/decoder/vp9_dboolhuff.h"
+
+#define vp9_read_prob(r) ((vp9_prob)vp9_read_literal(r, 8))
+#define vp9_read_and_apply_sign(r, value) (vp9_read_bit(r) ? -(value) : (value))
+
+// Intent of tree data structure is to make decoding trivial.
+static int treed_read(vp9_reader *const r, /* !!! must return a 0 or 1 !!! */
+                      vp9_tree t,
+                      const vp9_prob *const p) {
+  register vp9_tree_index i = 0;
+
+  while ((i = t[ i + vp9_read(r, p[i >> 1])]) > 0);
+
+  return -i;
+}
+
+#endif  // VP9_DECODER_VP9_TREEREADER_H_
diff --git a/libvpx/vp9/decoder/x86/vp9_dequantize_sse2.c b/libvpx/vp9/decoder/x86/vp9_dequantize_sse2.c
new file mode 100644
index 0000000..54ec67f
--- /dev/null
+++ b/libvpx/vp9/decoder/x86/vp9_dequantize_sse2.c
@@ -0,0 +1,220 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <emmintrin.h>  // SSE2
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_idct.h"
+
+void vp9_add_constant_residual_8x8_sse2(const int16_t diff, uint8_t *dest,
+                                        int stride) {
+  uint8_t abs_diff;
+  __m128i d;
+
+  // Prediction data.
+  __m128i p0 = _mm_loadl_epi64((const __m128i *)(dest + 0 * stride));
+  __m128i p1 = _mm_loadl_epi64((const __m128i *)(dest + 1 * stride));
+  __m128i p2 = _mm_loadl_epi64((const __m128i *)(dest + 2 * stride));
+  __m128i p3 = _mm_loadl_epi64((const __m128i *)(dest + 3 * stride));
+  __m128i p4 = _mm_loadl_epi64((const __m128i *)(dest + 4 * stride));
+  __m128i p5 = _mm_loadl_epi64((const __m128i *)(dest + 5 * stride));
+  __m128i p6 = _mm_loadl_epi64((const __m128i *)(dest + 6 * stride));
+  __m128i p7 = _mm_loadl_epi64((const __m128i *)(dest + 7 * stride));
+
+  p0 = _mm_unpacklo_epi64(p0, p1);
+  p2 = _mm_unpacklo_epi64(p2, p3);
+  p4 = _mm_unpacklo_epi64(p4, p5);
+  p6 = _mm_unpacklo_epi64(p6, p7);
+
+  // Clip diff value to [0, 255] range. Then, do addition or subtraction
+  // according to its sign.
+  if (diff >= 0) {
+    abs_diff = (diff > 255) ? 255 : diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+
+    p0 = _mm_adds_epu8(p0, d);
+    p2 = _mm_adds_epu8(p2, d);
+    p4 = _mm_adds_epu8(p4, d);
+    p6 = _mm_adds_epu8(p6, d);
+  } else {
+    abs_diff = (diff < -255) ? 255 : -diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+
+    p0 = _mm_subs_epu8(p0, d);
+    p2 = _mm_subs_epu8(p2, d);
+    p4 = _mm_subs_epu8(p4, d);
+    p6 = _mm_subs_epu8(p6, d);
+  }
+
+  _mm_storel_epi64((__m128i *)(dest + 0 * stride), p0);
+  p0 = _mm_srli_si128(p0, 8);
+  _mm_storel_epi64((__m128i *)(dest + 1 * stride), p0);
+
+  _mm_storel_epi64((__m128i *)(dest + 2 * stride), p2);
+  p2 = _mm_srli_si128(p2, 8);
+  _mm_storel_epi64((__m128i *)(dest + 3 * stride), p2);
+
+  _mm_storel_epi64((__m128i *)(dest + 4 * stride), p4);
+  p4 = _mm_srli_si128(p4, 8);
+  _mm_storel_epi64((__m128i *)(dest + 5 * stride), p4);
+
+  _mm_storel_epi64((__m128i *)(dest + 6 * stride), p6);
+  p6 = _mm_srli_si128(p6, 8);
+  _mm_storel_epi64((__m128i *)(dest + 7 * stride), p6);
+}
+
+void vp9_add_constant_residual_16x16_sse2(const int16_t diff, uint8_t *dest,
+                                          int stride) {
+  uint8_t abs_diff;
+  __m128i d;
+
+  // Prediction data.
+  __m128i p0 = _mm_load_si128((const __m128i *)(dest + 0 * stride));
+  __m128i p1 = _mm_load_si128((const __m128i *)(dest + 1 * stride));
+  __m128i p2 = _mm_load_si128((const __m128i *)(dest + 2 * stride));
+  __m128i p3 = _mm_load_si128((const __m128i *)(dest + 3 * stride));
+  __m128i p4 = _mm_load_si128((const __m128i *)(dest + 4 * stride));
+  __m128i p5 = _mm_load_si128((const __m128i *)(dest + 5 * stride));
+  __m128i p6 = _mm_load_si128((const __m128i *)(dest + 6 * stride));
+  __m128i p7 = _mm_load_si128((const __m128i *)(dest + 7 * stride));
+  __m128i p8 = _mm_load_si128((const __m128i *)(dest + 8 * stride));
+  __m128i p9 = _mm_load_si128((const __m128i *)(dest + 9 * stride));
+  __m128i p10 = _mm_load_si128((const __m128i *)(dest + 10 * stride));
+  __m128i p11 = _mm_load_si128((const __m128i *)(dest + 11 * stride));
+  __m128i p12 = _mm_load_si128((const __m128i *)(dest + 12 * stride));
+  __m128i p13 = _mm_load_si128((const __m128i *)(dest + 13 * stride));
+  __m128i p14 = _mm_load_si128((const __m128i *)(dest + 14 * stride));
+  __m128i p15 = _mm_load_si128((const __m128i *)(dest + 15 * stride));
+
+  // Clip diff value to [0, 255] range. Then, do addition or subtraction
+  // according to its sign.
+  if (diff >= 0) {
+    abs_diff = (diff > 255) ? 255 : diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+
+    p0 = _mm_adds_epu8(p0, d);
+    p1 = _mm_adds_epu8(p1, d);
+    p2 = _mm_adds_epu8(p2, d);
+    p3 = _mm_adds_epu8(p3, d);
+    p4 = _mm_adds_epu8(p4, d);
+    p5 = _mm_adds_epu8(p5, d);
+    p6 = _mm_adds_epu8(p6, d);
+    p7 = _mm_adds_epu8(p7, d);
+    p8 = _mm_adds_epu8(p8, d);
+    p9 = _mm_adds_epu8(p9, d);
+    p10 = _mm_adds_epu8(p10, d);
+    p11 = _mm_adds_epu8(p11, d);
+    p12 = _mm_adds_epu8(p12, d);
+    p13 = _mm_adds_epu8(p13, d);
+    p14 = _mm_adds_epu8(p14, d);
+    p15 = _mm_adds_epu8(p15, d);
+  } else {
+    abs_diff = (diff < -255) ? 255 : -diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+
+    p0 = _mm_subs_epu8(p0, d);
+    p1 = _mm_subs_epu8(p1, d);
+    p2 = _mm_subs_epu8(p2, d);
+    p3 = _mm_subs_epu8(p3, d);
+    p4 = _mm_subs_epu8(p4, d);
+    p5 = _mm_subs_epu8(p5, d);
+    p6 = _mm_subs_epu8(p6, d);
+    p7 = _mm_subs_epu8(p7, d);
+    p8 = _mm_subs_epu8(p8, d);
+    p9 = _mm_subs_epu8(p9, d);
+    p10 = _mm_subs_epu8(p10, d);
+    p11 = _mm_subs_epu8(p11, d);
+    p12 = _mm_subs_epu8(p12, d);
+    p13 = _mm_subs_epu8(p13, d);
+    p14 = _mm_subs_epu8(p14, d);
+    p15 = _mm_subs_epu8(p15, d);
+  }
+
+  // Store results
+  _mm_store_si128((__m128i *)(dest + 0 * stride), p0);
+  _mm_store_si128((__m128i *)(dest + 1 * stride), p1);
+  _mm_store_si128((__m128i *)(dest + 2 * stride), p2);
+  _mm_store_si128((__m128i *)(dest + 3 * stride), p3);
+  _mm_store_si128((__m128i *)(dest + 4 * stride), p4);
+  _mm_store_si128((__m128i *)(dest + 5 * stride), p5);
+  _mm_store_si128((__m128i *)(dest + 6 * stride), p6);
+  _mm_store_si128((__m128i *)(dest + 7 * stride), p7);
+  _mm_store_si128((__m128i *)(dest + 8 * stride), p8);
+  _mm_store_si128((__m128i *)(dest + 9 * stride), p9);
+  _mm_store_si128((__m128i *)(dest + 10 * stride), p10);
+  _mm_store_si128((__m128i *)(dest + 11 * stride), p11);
+  _mm_store_si128((__m128i *)(dest + 12 * stride), p12);
+  _mm_store_si128((__m128i *)(dest + 13 * stride), p13);
+  _mm_store_si128((__m128i *)(dest + 14 * stride), p14);
+  _mm_store_si128((__m128i *)(dest + 15 * stride), p15);
+}
+
+void vp9_add_constant_residual_32x32_sse2(const int16_t diff, uint8_t *dest,
+                                          int stride) {
+  uint8_t abs_diff;
+  __m128i d;
+  int i = 8;
+
+  if (diff >= 0) {
+    abs_diff = (diff > 255) ? 255 : diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+  } else {
+    abs_diff = (diff < -255) ? 255 : -diff;
+    d = _mm_shuffle_epi32(_mm_cvtsi32_si128((int)(abs_diff * 0x01010101u)), 0);
+  }
+
+  do {
+    // Prediction data.
+    __m128i p0 = _mm_load_si128((const __m128i *)(dest + 0 * stride));
+    __m128i p1 = _mm_load_si128((const __m128i *)(dest + 0 * stride + 16));
+    __m128i p2 = _mm_load_si128((const __m128i *)(dest + 1 * stride));
+    __m128i p3 = _mm_load_si128((const __m128i *)(dest + 1 * stride + 16));
+    __m128i p4 = _mm_load_si128((const __m128i *)(dest + 2 * stride));
+    __m128i p5 = _mm_load_si128((const __m128i *)(dest + 2 * stride + 16));
+    __m128i p6 = _mm_load_si128((const __m128i *)(dest + 3 * stride));
+    __m128i p7 = _mm_load_si128((const __m128i *)(dest + 3 * stride + 16));
+
+    // Clip diff value to [0, 255] range. Then, do addition or subtraction
+    // according to its sign.
+    if (diff >= 0) {
+      p0 = _mm_adds_epu8(p0, d);
+      p1 = _mm_adds_epu8(p1, d);
+      p2 = _mm_adds_epu8(p2, d);
+      p3 = _mm_adds_epu8(p3, d);
+      p4 = _mm_adds_epu8(p4, d);
+      p5 = _mm_adds_epu8(p5, d);
+      p6 = _mm_adds_epu8(p6, d);
+      p7 = _mm_adds_epu8(p7, d);
+    } else {
+      p0 = _mm_subs_epu8(p0, d);
+      p1 = _mm_subs_epu8(p1, d);
+      p2 = _mm_subs_epu8(p2, d);
+      p3 = _mm_subs_epu8(p3, d);
+      p4 = _mm_subs_epu8(p4, d);
+      p5 = _mm_subs_epu8(p5, d);
+      p6 = _mm_subs_epu8(p6, d);
+      p7 = _mm_subs_epu8(p7, d);
+    }
+
+    // Store results
+    _mm_store_si128((__m128i *)(dest + 0 * stride), p0);
+    _mm_store_si128((__m128i *)(dest + 0 * stride + 16), p1);
+    _mm_store_si128((__m128i *)(dest + 1 * stride), p2);
+    _mm_store_si128((__m128i *)(dest + 1 * stride + 16), p3);
+    _mm_store_si128((__m128i *)(dest + 2 * stride), p4);
+    _mm_store_si128((__m128i *)(dest + 2 * stride + 16), p5);
+    _mm_store_si128((__m128i *)(dest + 3 * stride), p6);
+    _mm_store_si128((__m128i *)(dest + 3 * stride + 16), p7);
+
+    dest += 4 * stride;
+  } while (--i);
+}
diff --git a/libvpx/vp9/encoder/vp9_asm_enc_offsets.c b/libvpx/vp9/encoder/vp9_asm_enc_offsets.c
new file mode 100644
index 0000000..921e8f0
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_asm_enc_offsets.c
@@ -0,0 +1,17 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_ports/asm_offsets.h"
+
+BEGIN
+
+
+END
diff --git a/libvpx/vp9/encoder/vp9_bitstream.c b/libvpx/vp9/encoder/vp9_bitstream.c
new file mode 100644
index 0000000..09ab2db
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_bitstream.c
@@ -0,0 +1,1821 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+#include <limits.h>
+
+#include "vpx/vpx_encoder.h"
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_tile_common.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_mvref_common.h"
+#include "vp9/common/vp9_treecoder.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/common/vp9_pragmas.h"
+
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "vp9/encoder/vp9_bitstream.h"
+#include "vp9/encoder/vp9_segmentation.h"
+#include "vp9/encoder/vp9_write_bit_buffer.h"
+
+
+#if defined(SECTIONBITS_OUTPUT)
+unsigned __int64 Sectionbits[500];
+#endif
+
+#ifdef ENTROPY_STATS
+int intra_mode_stats[VP9_INTRA_MODES]
+                    [VP9_INTRA_MODES]
+                    [VP9_INTRA_MODES];
+vp9_coeff_stats tree_update_hist[TX_SIZE_MAX_SB][BLOCK_TYPES];
+
+extern unsigned int active_section;
+#endif
+
+#define vp9_cost_upd  ((int)(vp9_cost_one(upd) - vp9_cost_zero(upd)) >> 8)
+#define vp9_cost_upd256  ((int)(vp9_cost_one(upd) - vp9_cost_zero(upd)))
+
+#ifdef MODE_STATS
+int64_t tx_count_32x32p_stats[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB];
+int64_t tx_count_16x16p_stats[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
+int64_t tx_count_8x8p_stats[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
+int64_t switchable_interp_stats[VP9_SWITCHABLE_FILTERS+1]
+                               [VP9_SWITCHABLE_FILTERS];
+
+void init_tx_count_stats() {
+  vp9_zero(tx_count_32x32p_stats);
+  vp9_zero(tx_count_16x16p_stats);
+  vp9_zero(tx_count_8x8p_stats);
+}
+
+void init_switchable_interp_stats() {
+  vp9_zero(switchable_interp_stats);
+}
+
+static void update_tx_count_stats(VP9_COMMON *cm) {
+  int i, j;
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    for (j = 0; j < TX_SIZE_MAX_SB; j++) {
+      tx_count_32x32p_stats[i][j] += cm->fc.tx_count_32x32p[i][j];
+    }
+  }
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    for (j = 0; j < TX_SIZE_MAX_SB - 1; j++) {
+      tx_count_16x16p_stats[i][j] += cm->fc.tx_count_16x16p[i][j];
+    }
+  }
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    for (j = 0; j < TX_SIZE_MAX_SB - 2; j++) {
+      tx_count_8x8p_stats[i][j] += cm->fc.tx_count_8x8p[i][j];
+    }
+  }
+}
+
+static void update_switchable_interp_stats(VP9_COMMON *cm) {
+  int i, j;
+  for (i = 0; i < VP9_SWITCHABLE_FILTERS+1; ++i)
+    for (j = 0; j < VP9_SWITCHABLE_FILTERS; ++j) {
+      switchable_interp_stats[i][j] += cm->fc.switchable_interp_count[i][j];
+    }
+}
+
+void write_tx_count_stats() {
+  int i, j;
+  FILE *fp = fopen("tx_count.bin", "wb");
+  fwrite(tx_count_32x32p_stats, sizeof(tx_count_32x32p_stats), 1, fp);
+  fwrite(tx_count_16x16p_stats, sizeof(tx_count_16x16p_stats), 1, fp);
+  fwrite(tx_count_8x8p_stats, sizeof(tx_count_8x8p_stats), 1, fp);
+  fclose(fp);
+
+  printf(
+      "vp9_default_tx_count_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB] = {\n");
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    printf("  { ");
+    for (j = 0; j < TX_SIZE_MAX_SB; j++) {
+      printf("%"PRId64", ", tx_count_32x32p_stats[i][j]);
+    }
+    printf("},\n");
+  }
+  printf("};\n");
+  printf(
+      "vp9_default_tx_count_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB-1] = {\n");
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    printf("  { ");
+    for (j = 0; j < TX_SIZE_MAX_SB - 1; j++) {
+      printf("%"PRId64", ", tx_count_16x16p_stats[i][j]);
+    }
+    printf("},\n");
+  }
+  printf("};\n");
+  printf(
+      "vp9_default_tx_count_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB-2] = {\n");
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    printf("  { ");
+    for (j = 0; j < TX_SIZE_MAX_SB - 2; j++) {
+      printf("%"PRId64", ", tx_count_8x8p_stats[i][j]);
+    }
+    printf("},\n");
+  }
+  printf("};\n");
+}
+
+void write_switchable_interp_stats() {
+  int i, j;
+  FILE *fp = fopen("switchable_interp.bin", "wb");
+  fwrite(switchable_interp_stats, sizeof(switchable_interp_stats), 1, fp);
+  fclose(fp);
+
+  printf(
+      "vp9_default_switchable_filter_count[VP9_SWITCHABLE_FILTERS+1]"
+      "[VP9_SWITCHABLE_FILTERS] = {\n");
+  for (i = 0; i < VP9_SWITCHABLE_FILTERS+1; i++) {
+    printf("  { ");
+    for (j = 0; j < VP9_SWITCHABLE_FILTERS; j++) {
+      printf("%"PRId64", ", switchable_interp_stats[i][j]);
+    }
+    printf("},\n");
+  }
+  printf("};\n");
+}
+#endif
+
+static int update_bits[255];
+
+static INLINE void write_be32(uint8_t *p, int value) {
+  p[0] = value >> 24;
+  p[1] = value >> 16;
+  p[2] = value >> 8;
+  p[3] = value;
+}
+
+
+
+int recenter_nonneg(int v, int m) {
+  if (v > (m << 1))
+    return v;
+  else if (v >= m)
+    return ((v - m) << 1);
+  else
+    return ((m - v) << 1) - 1;
+}
+
+static int get_unsigned_bits(unsigned num_values) {
+  int cat = 0;
+  if ((num_values--) <= 1) return 0;
+  while (num_values > 0) {
+    cat++;
+    num_values >>= 1;
+  }
+  return cat;
+}
+
+void vp9_encode_unsigned_max(struct vp9_write_bit_buffer *wb,
+                             int data, int max) {
+  vp9_wb_write_literal(wb, data, get_unsigned_bits(max));
+}
+
+void encode_uniform(vp9_writer *w, int v, int n) {
+  int l = get_unsigned_bits(n);
+  int m;
+  if (l == 0)
+    return;
+  m = (1 << l) - n;
+  if (v < m) {
+    vp9_write_literal(w, v, l - 1);
+  } else {
+    vp9_write_literal(w, m + ((v - m) >> 1), l - 1);
+    vp9_write_literal(w, (v - m) & 1, 1);
+  }
+}
+
+int count_uniform(int v, int n) {
+  int l = get_unsigned_bits(n);
+  int m;
+  if (l == 0) return 0;
+  m = (1 << l) - n;
+  if (v < m)
+    return l - 1;
+  else
+    return l;
+}
+
+void encode_term_subexp(vp9_writer *w, int word, int k, int num_syms) {
+  int i = 0;
+  int mk = 0;
+  while (1) {
+    int b = (i ? k + i - 1 : k);
+    int a = (1 << b);
+    if (num_syms <= mk + 3 * a) {
+      encode_uniform(w, word - mk, num_syms - mk);
+      break;
+    } else {
+      int t = (word >= mk + a);
+      vp9_write_literal(w, t, 1);
+      if (t) {
+        i = i + 1;
+        mk += a;
+      } else {
+        vp9_write_literal(w, word - mk, b);
+        break;
+      }
+    }
+  }
+}
+
+int count_term_subexp(int word, int k, int num_syms) {
+  int count = 0;
+  int i = 0;
+  int mk = 0;
+  while (1) {
+    int b = (i ? k + i - 1 : k);
+    int a = (1 << b);
+    if (num_syms <= mk + 3 * a) {
+      count += count_uniform(word - mk, num_syms - mk);
+      break;
+    } else {
+      int t = (word >= mk + a);
+      count++;
+      if (t) {
+        i = i + 1;
+        mk += a;
+      } else {
+        count += b;
+        break;
+      }
+    }
+  }
+  return count;
+}
+
+static void compute_update_table() {
+  int i;
+  for (i = 0; i < 254; i++)
+    update_bits[i] = count_term_subexp(i, SUBEXP_PARAM, 255);
+}
+
+static int split_index(int i, int n, int modulus) {
+  int max1 = (n - 1 - modulus / 2) / modulus + 1;
+  if (i % modulus == modulus / 2) i = i / modulus;
+  else i = max1 + i - (i + modulus - modulus / 2) / modulus;
+  return i;
+}
+
+static int remap_prob(int v, int m) {
+  const int n = 255;
+  const int modulus = MODULUS_PARAM;
+  int i;
+  v--;
+  m--;
+  if ((m << 1) <= n)
+    i = recenter_nonneg(v, m) - 1;
+  else
+    i = recenter_nonneg(n - 1 - v, n - 1 - m) - 1;
+
+  i = split_index(i, n - 1, modulus);
+  return i;
+}
+
+static void write_prob_diff_update(vp9_writer *w,
+                                   vp9_prob newp, vp9_prob oldp) {
+  int delp = remap_prob(newp, oldp);
+  encode_term_subexp(w, delp, SUBEXP_PARAM, 255);
+}
+
+static int prob_diff_update_cost(vp9_prob newp, vp9_prob oldp) {
+  int delp = remap_prob(newp, oldp);
+  return update_bits[delp] * 256;
+}
+
+static int prob_update_savings(const unsigned int *ct,
+                               const vp9_prob oldp, const vp9_prob newp,
+                               const vp9_prob upd) {
+  const int old_b = cost_branch256(ct, oldp);
+  const int new_b = cost_branch256(ct, newp);
+  const int update_b = 2048 + vp9_cost_upd256;
+  return old_b - new_b - update_b;
+}
+
+static int prob_diff_update_savings_search(const unsigned int *ct,
+                                           const vp9_prob oldp, vp9_prob *bestp,
+                                           const vp9_prob upd) {
+  const int old_b = cost_branch256(ct, oldp);
+  int new_b, update_b, savings, bestsavings, step;
+  vp9_prob newp, bestnewp;
+
+  bestsavings = 0;
+  bestnewp = oldp;
+
+  step = (*bestp > oldp ? -1 : 1);
+  for (newp = *bestp; newp != oldp; newp += step) {
+    new_b = cost_branch256(ct, newp);
+    update_b = prob_diff_update_cost(newp, oldp) + vp9_cost_upd256;
+    savings = old_b - new_b - update_b;
+    if (savings > bestsavings) {
+      bestsavings = savings;
+      bestnewp = newp;
+    }
+  }
+  *bestp = bestnewp;
+  return bestsavings;
+}
+
+static int prob_diff_update_savings_search_model(const unsigned int *ct,
+                                                 const vp9_prob *oldp,
+                                                 vp9_prob *bestp,
+                                                 const vp9_prob upd,
+                                                 int b, int r) {
+  int i, old_b, new_b, update_b, savings, bestsavings, step;
+  int newp;
+  vp9_prob bestnewp, newplist[ENTROPY_NODES], oldplist[ENTROPY_NODES];
+  vp9_model_to_full_probs(oldp, oldplist);
+  vpx_memcpy(newplist, oldp, sizeof(vp9_prob) * UNCONSTRAINED_NODES);
+  for (i = UNCONSTRAINED_NODES, old_b = 0; i < ENTROPY_NODES; ++i)
+    old_b += cost_branch256(ct + 2 * i, oldplist[i]);
+  old_b += cost_branch256(ct + 2 * PIVOT_NODE, oldplist[PIVOT_NODE]);
+
+  bestsavings = 0;
+  bestnewp = oldp[PIVOT_NODE];
+
+  step = (*bestp > oldp[PIVOT_NODE] ? -1 : 1);
+  newp = *bestp;
+  for (; newp != oldp[PIVOT_NODE]; newp += step) {
+    if (newp < 1 || newp > 255) continue;
+    newplist[PIVOT_NODE] = newp;
+    vp9_model_to_full_probs(newplist, newplist);
+    for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
+      new_b += cost_branch256(ct + 2 * i, newplist[i]);
+    new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
+    update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
+        vp9_cost_upd256;
+    savings = old_b - new_b - update_b;
+    if (savings > bestsavings) {
+      bestsavings = savings;
+      bestnewp = newp;
+    }
+  }
+  *bestp = bestnewp;
+  return bestsavings;
+}
+
+static void vp9_cond_prob_update(vp9_writer *bc, vp9_prob *oldp, vp9_prob upd,
+                                 unsigned int *ct) {
+  vp9_prob newp;
+  int savings;
+  newp = get_binary_prob(ct[0], ct[1]);
+  assert(newp >= 1);
+  savings = prob_update_savings(ct, *oldp, newp, upd);
+  if (savings > 0) {
+    vp9_write(bc, 1, upd);
+    vp9_write_prob(bc, newp);
+    *oldp = newp;
+  } else {
+    vp9_write(bc, 0, upd);
+  }
+}
+
+static void vp9_cond_prob_diff_update(vp9_writer *bc, vp9_prob *oldp,
+                                      vp9_prob upd,
+                                      unsigned int *ct) {
+  vp9_prob newp;
+  int savings;
+  newp = get_binary_prob(ct[0], ct[1]);
+  assert(newp >= 1);
+  savings = prob_diff_update_savings_search(ct, *oldp, &newp, upd);
+  if (savings > 0) {
+    vp9_write(bc, 1, upd);
+    write_prob_diff_update(bc, newp, *oldp);
+    *oldp = newp;
+  } else {
+    vp9_write(bc, 0, upd);
+  }
+}
+
+static void update_mode(
+  vp9_writer *w,
+  int n,
+  const struct vp9_token tok[/* n */],
+  vp9_tree tree,
+  vp9_prob Pnew[/* n-1 */],
+  vp9_prob Pcur[/* n-1 */],
+  unsigned int bct[/* n-1 */] [2],
+  const unsigned int num_events[/* n */]
+) {
+  int i = 0;
+
+  vp9_tree_probs_from_distribution(tree, Pnew, bct, num_events, 0);
+  n--;
+
+  for (i = 0; i < n; ++i) {
+    vp9_cond_prob_diff_update(w, &Pcur[i], VP9_MODE_UPDATE_PROB, bct[i]);
+  }
+}
+
+static void update_mbintra_mode_probs(VP9_COMP* const cpi,
+                                      vp9_writer* const bc) {
+  VP9_COMMON *const cm = &cpi->common;
+  int j;
+  vp9_prob pnew[VP9_INTRA_MODES - 1];
+  unsigned int bct[VP9_INTRA_MODES - 1][2];
+
+  for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
+    update_mode(bc, VP9_INTRA_MODES, vp9_intra_mode_encodings,
+                vp9_intra_mode_tree, pnew,
+                cm->fc.y_mode_prob[j], bct,
+                (unsigned int *)cpi->y_mode_count[j]);
+}
+
+void vp9_update_skip_probs(VP9_COMP *cpi, vp9_writer *bc) {
+  VP9_COMMON *const pc = &cpi->common;
+  int k;
+
+  for (k = 0; k < MBSKIP_CONTEXTS; ++k) {
+    vp9_cond_prob_diff_update(bc, &pc->fc.mbskip_probs[k],
+                              VP9_MODE_UPDATE_PROB, pc->fc.mbskip_count[k]);
+  }
+}
+
+static void write_intra_mode(vp9_writer *bc, int m, const vp9_prob *p) {
+  write_token(bc, vp9_intra_mode_tree, p, vp9_intra_mode_encodings + m);
+}
+
+static void update_switchable_interp_probs(VP9_COMP *const cpi,
+                                           vp9_writer* const bc) {
+  VP9_COMMON *const pc = &cpi->common;
+  unsigned int branch_ct[VP9_SWITCHABLE_FILTERS + 1]
+                        [VP9_SWITCHABLE_FILTERS - 1][2];
+  vp9_prob new_prob[VP9_SWITCHABLE_FILTERS + 1][VP9_SWITCHABLE_FILTERS - 1];
+  int i, j;
+  for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j) {
+    vp9_tree_probs_from_distribution(
+        vp9_switchable_interp_tree,
+        new_prob[j], branch_ct[j],
+        pc->fc.switchable_interp_count[j], 0);
+  }
+  for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j) {
+    for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i) {
+      vp9_cond_prob_diff_update(bc, &pc->fc.switchable_interp_prob[j][i],
+                                VP9_MODE_UPDATE_PROB, branch_ct[j][i]);
+    }
+  }
+#ifdef MODE_STATS
+  if (!cpi->dummy_packing)
+    update_switchable_interp_stats(pc);
+#endif
+}
+
+static void update_inter_mode_probs(VP9_COMMON *pc, vp9_writer* const bc) {
+  int i, j;
+
+  for (i = 0; i < INTER_MODE_CONTEXTS; i++) {
+    for (j = 0; j < VP9_INTER_MODES - 1; j++) {
+      vp9_cond_prob_diff_update(bc, &pc->fc.inter_mode_probs[i][j],
+                                VP9_MODE_UPDATE_PROB,
+                                pc->fc.inter_mode_counts[i][j]);
+    }
+  }
+}
+
+static void pack_mb_tokens(vp9_writer* const bc,
+                           TOKENEXTRA **tp,
+                           const TOKENEXTRA *const stop) {
+  TOKENEXTRA *p = *tp;
+
+  while (p < stop) {
+    const int t = p->token;
+    const struct vp9_token *const a = vp9_coef_encodings + t;
+    const vp9_extra_bit *const b = vp9_extra_bits + t;
+    int i = 0;
+    const vp9_prob *pp;
+    int v = a->value;
+    int n = a->len;
+    vp9_prob probs[ENTROPY_NODES];
+
+    if (t == EOSB_TOKEN) {
+      ++p;
+      break;
+    }
+    if (t >= TWO_TOKEN) {
+      vp9_model_to_full_probs(p->context_tree, probs);
+      pp = probs;
+    } else {
+      pp = p->context_tree;
+    }
+    assert(pp != 0);
+
+    /* skip one or two nodes */
+#if !CONFIG_BALANCED_COEFTREE
+    if (p->skip_eob_node) {
+      n -= p->skip_eob_node;
+      i = 2 * p->skip_eob_node;
+    }
+#endif
+
+    do {
+      const int bb = (v >> --n) & 1;
+#if CONFIG_BALANCED_COEFTREE
+      if (i == 2 && p->skip_eob_node) {
+        i += 2;
+        assert(bb == 1);
+        continue;
+      }
+#endif
+      vp9_write(bc, bb, pp[i >> 1]);
+      i = vp9_coef_tree[i + bb];
+    } while (n);
+
+    if (b->base_val) {
+      const int e = p->extra, l = b->len;
+
+      if (l) {
+        const unsigned char *pb = b->prob;
+        int v = e >> 1;
+        int n = l;              /* number of bits in v, assumed nonzero */
+        int i = 0;
+
+        do {
+          const int bb = (v >> --n) & 1;
+          vp9_write(bc, bb, pb[i >> 1]);
+          i = b->tree[i + bb];
+        } while (n);
+      }
+
+      vp9_write_bit(bc, e & 1);
+    }
+    ++p;
+  }
+
+  *tp = p;
+}
+
+static void write_sb_mv_ref(vp9_writer *bc, MB_PREDICTION_MODE m,
+                            const vp9_prob *p) {
+#if CONFIG_DEBUG
+  assert(NEARESTMV <= m && m <= NEWMV);
+#endif
+  write_token(bc, vp9_sb_mv_ref_tree, p,
+              vp9_sb_mv_ref_encoding_array - NEARESTMV + m);
+}
+
+// This function writes the current macro block's segnment id to the bitstream
+// It should only be called if a segment map update is indicated.
+static void write_mb_segid(vp9_writer *bc,
+                           const MB_MODE_INFO *mi, const MACROBLOCKD *xd) {
+  if (xd->segmentation_enabled && xd->update_mb_segmentation_map)
+    treed_write(bc, vp9_segment_tree, xd->mb_segment_tree_probs,
+                mi->segment_id, 3);
+}
+
+// This function encodes the reference frame
+static void encode_ref_frame(VP9_COMP *cpi, vp9_writer *bc) {
+  VP9_COMMON *const pc = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *mi = &xd->mode_info_context->mbmi;
+  const int segment_id = mi->segment_id;
+  int seg_ref_active = vp9_segfeature_active(xd, segment_id,
+                                             SEG_LVL_REF_FRAME);
+  // If segment level coding of this signal is disabled...
+  // or the segment allows multiple reference frame options
+  if (!seg_ref_active) {
+    // does the feature use compound prediction or not
+    // (if not specified at the frame/segment level)
+    if (pc->comp_pred_mode == HYBRID_PREDICTION) {
+      vp9_write(bc, mi->ref_frame[1] > INTRA_FRAME,
+                vp9_get_pred_prob(pc, xd, PRED_COMP_INTER_INTER));
+    } else {
+      assert((mi->ref_frame[1] <= INTRA_FRAME) ==
+                 (pc->comp_pred_mode == SINGLE_PREDICTION_ONLY));
+    }
+
+    if (mi->ref_frame[1] > INTRA_FRAME) {
+      vp9_write(bc, mi->ref_frame[0] == GOLDEN_FRAME,
+                vp9_get_pred_prob(pc, xd, PRED_COMP_REF_P));
+    } else {
+      vp9_write(bc, mi->ref_frame[0] != LAST_FRAME,
+                vp9_get_pred_prob(pc, xd, PRED_SINGLE_REF_P1));
+      if (mi->ref_frame[0] != LAST_FRAME)
+        vp9_write(bc, mi->ref_frame[0] != GOLDEN_FRAME,
+                  vp9_get_pred_prob(pc, xd, PRED_SINGLE_REF_P2));
+    }
+  } else {
+    assert(mi->ref_frame[1] <= INTRA_FRAME);
+    assert(vp9_get_segdata(xd, segment_id, SEG_LVL_REF_FRAME) ==
+           mi->ref_frame[0]);
+  }
+
+  // if using the prediction mdoel we have nothing further to do because
+  // the reference frame is fully coded by the segment
+}
+
+static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
+                                vp9_writer *bc, int mi_row, int mi_col) {
+  VP9_COMMON *const pc = &cpi->common;
+  const nmv_context *nmvc = &pc->fc.nmvc;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mi = &m->mbmi;
+  const MV_REFERENCE_FRAME rf = mi->ref_frame[0];
+  const MB_PREDICTION_MODE mode = mi->mode;
+  const int segment_id = mi->segment_id;
+  int skip_coeff;
+
+  xd->prev_mode_info_context = pc->prev_mi + (m - pc->mi);
+  x->partition_info = x->pi + (m - pc->mi);
+
+#ifdef ENTROPY_STATS
+  active_section = 9;
+#endif
+
+  if (cpi->mb.e_mbd.update_mb_segmentation_map) {
+    // Is temporal coding of the segment map enabled
+    if (pc->temporal_update) {
+      unsigned char prediction_flag = vp9_get_pred_flag(xd, PRED_SEG_ID);
+      vp9_prob pred_prob = vp9_get_pred_prob(pc, xd, PRED_SEG_ID);
+
+      // Code the segment id prediction flag for this mb
+      vp9_write(bc, prediction_flag, pred_prob);
+
+      // If the mb segment id wasn't predicted code explicitly
+      if (!prediction_flag)
+        write_mb_segid(bc, mi, &cpi->mb.e_mbd);
+    } else {
+      // Normal unpredicted coding
+      write_mb_segid(bc, mi, &cpi->mb.e_mbd);
+    }
+  }
+
+  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
+    skip_coeff = 1;
+  } else {
+    skip_coeff = m->mbmi.mb_skip_coeff;
+    vp9_write(bc, skip_coeff,
+              vp9_get_pred_prob(pc, xd, PRED_MBSKIP));
+  }
+
+  if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME))
+    vp9_write(bc, rf != INTRA_FRAME,
+              vp9_get_pred_prob(pc, xd, PRED_INTRA_INTER));
+
+  if (mi->sb_type >= BLOCK_SIZE_SB8X8 && pc->txfm_mode == TX_MODE_SELECT &&
+      !(rf != INTRA_FRAME &&
+        (skip_coeff || vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)))) {
+    TX_SIZE sz = mi->txfm_size;
+    const vp9_prob *tx_probs = vp9_get_pred_probs(pc, xd, PRED_TX_SIZE);
+    vp9_write(bc, sz != TX_4X4, tx_probs[0]);
+    if (mi->sb_type >= BLOCK_SIZE_MB16X16 && sz != TX_4X4) {
+      vp9_write(bc, sz != TX_8X8, tx_probs[1]);
+      if (mi->sb_type >= BLOCK_SIZE_SB32X32 && sz != TX_8X8)
+        vp9_write(bc, sz != TX_16X16, tx_probs[2]);
+    }
+  }
+
+  if (rf == INTRA_FRAME) {
+#ifdef ENTROPY_STATS
+    active_section = 6;
+#endif
+
+    if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+      const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+      const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+      const int bsl = MIN(bwl, bhl);
+      write_intra_mode(bc, mode, pc->fc.y_mode_prob[MIN(3, bsl)]);
+    } else {
+      int idx, idy;
+      int bw = 1 << b_width_log2(mi->sb_type);
+      int bh = 1 << b_height_log2(mi->sb_type);
+      for (idy = 0; idy < 2; idy += bh)
+        for (idx = 0; idx < 2; idx += bw) {
+          MB_PREDICTION_MODE bm = m->bmi[idy * 2 + idx].as_mode.first;
+          write_intra_mode(bc, bm, pc->fc.y_mode_prob[0]);
+        }
+    }
+    write_intra_mode(bc, mi->uv_mode,
+                     pc->fc.uv_mode_prob[mode]);
+  } else {
+    vp9_prob *mv_ref_p;
+
+    encode_ref_frame(cpi, bc);
+
+    mv_ref_p = cpi->common.fc.inter_mode_probs[mi->mb_mode_context[rf]];
+
+#ifdef ENTROPY_STATS
+    active_section = 3;
+#endif
+
+    // If segment skip is not enabled code the mode.
+    if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
+      if (mi->sb_type >= BLOCK_SIZE_SB8X8) {
+        write_sb_mv_ref(bc, mode, mv_ref_p);
+        vp9_accum_mv_refs(&cpi->common, mode, mi->mb_mode_context[rf]);
+      }
+    }
+
+    if (cpi->common.mcomp_filter_type == SWITCHABLE) {
+      write_token(bc, vp9_switchable_interp_tree,
+                  vp9_get_pred_probs(&cpi->common, xd,
+                                     PRED_SWITCHABLE_INTERP),
+                  vp9_switchable_interp_encodings +
+                  vp9_switchable_interp_map[mi->interp_filter]);
+    } else {
+      assert(mi->interp_filter == cpi->common.mcomp_filter_type);
+    }
+
+    if (xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+      int j;
+      MB_PREDICTION_MODE blockmode;
+      int_mv blockmv;
+      int bwl = b_width_log2(mi->sb_type), bw = 1 << bwl;
+      int bhl = b_height_log2(mi->sb_type), bh = 1 << bhl;
+      int idx, idy;
+      for (idy = 0; idy < 2; idy += bh) {
+        for (idx = 0; idx < 2; idx += bw) {
+          j = idy * 2 + idx;
+          blockmode = cpi->mb.partition_info->bmi[j].mode;
+          blockmv = cpi->mb.partition_info->bmi[j].mv;
+          write_sb_mv_ref(bc, blockmode, mv_ref_p);
+          vp9_accum_mv_refs(&cpi->common, blockmode, mi->mb_mode_context[rf]);
+          if (blockmode == NEWMV) {
+#ifdef ENTROPY_STATS
+            active_section = 11;
+#endif
+            vp9_encode_mv(bc, &blockmv.as_mv, &mi->best_mv.as_mv,
+                          nmvc, xd->allow_high_precision_mv);
+
+            if (mi->ref_frame[1] > INTRA_FRAME)
+              vp9_encode_mv(bc,
+                            &cpi->mb.partition_info->bmi[j].second_mv.as_mv,
+                            &mi->best_second_mv.as_mv,
+                            nmvc, xd->allow_high_precision_mv);
+          }
+        }
+      }
+    } else if (mode == NEWMV) {
+#ifdef ENTROPY_STATS
+      active_section = 5;
+#endif
+      vp9_encode_mv(bc,
+                    &mi->mv[0].as_mv, &mi->best_mv.as_mv,
+                    nmvc, xd->allow_high_precision_mv);
+
+      if (mi->ref_frame[1] > INTRA_FRAME)
+        vp9_encode_mv(bc,
+                      &mi->mv[1].as_mv, &mi->best_second_mv.as_mv,
+                      nmvc, xd->allow_high_precision_mv);
+    }
+  }
+}
+
+static void write_mb_modes_kf(const VP9_COMP *cpi,
+                              MODE_INFO *m,
+                              vp9_writer *bc, int mi_row, int mi_col) {
+  const VP9_COMMON *const c = &cpi->common;
+  const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  const int ym = m->mbmi.mode;
+  const int mis = c->mode_info_stride;
+  const int segment_id = m->mbmi.segment_id;
+  int skip_coeff;
+
+  if (xd->update_mb_segmentation_map)
+    write_mb_segid(bc, &m->mbmi, xd);
+
+  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
+    skip_coeff = 1;
+  } else {
+    skip_coeff = m->mbmi.mb_skip_coeff;
+    vp9_write(bc, skip_coeff, vp9_get_pred_prob(c, xd, PRED_MBSKIP));
+  }
+
+  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8 && c->txfm_mode == TX_MODE_SELECT) {
+    TX_SIZE sz = m->mbmi.txfm_size;
+    const vp9_prob *tx_probs = vp9_get_pred_probs(c, xd, PRED_TX_SIZE);
+    vp9_write(bc, sz != TX_4X4, tx_probs[0]);
+    if (m->mbmi.sb_type >= BLOCK_SIZE_MB16X16 && sz != TX_4X4) {
+      vp9_write(bc, sz != TX_8X8, tx_probs[1]);
+      if (m->mbmi.sb_type >= BLOCK_SIZE_SB32X32 && sz != TX_8X8)
+        vp9_write(bc, sz != TX_16X16, tx_probs[2]);
+    }
+  }
+
+  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
+    const MB_PREDICTION_MODE L = xd->left_available ?
+                                 left_block_mode(m, 0) : DC_PRED;
+    write_intra_mode(bc, ym, c->kf_y_mode_prob[A][L]);
+  } else {
+    int idx, idy;
+    int bw = 1 << b_width_log2(m->mbmi.sb_type);
+    int bh = 1 << b_height_log2(m->mbmi.sb_type);
+    for (idy = 0; idy < 2; idy += bh) {
+      for (idx = 0; idx < 2; idx += bw) {
+        int i = idy * 2 + idx;
+        const MB_PREDICTION_MODE A = above_block_mode(m, i, mis);
+        const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
+                                     left_block_mode(m, i) : DC_PRED;
+        const int bm = m->bmi[i].as_mode.first;
+#ifdef ENTROPY_STATS
+        ++intra_mode_stats[A][L][bm];
+#endif
+        write_intra_mode(bc, bm, c->kf_y_mode_prob[A][L]);
+      }
+    }
+  }
+
+  write_intra_mode(bc, m->mbmi.uv_mode, c->kf_uv_mode_prob[ym]);
+}
+
+static void write_modes_b(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
+                          TOKENEXTRA **tok, TOKENEXTRA *tok_end,
+                          int mi_row, int mi_col) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+
+  if (m->mbmi.sb_type < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index > 0)
+      return;
+  xd->mode_info_context = m;
+  set_mi_row_col(&cpi->common, xd, mi_row,
+                 1 << mi_height_log2(m->mbmi.sb_type),
+                 mi_col, 1 << mi_width_log2(m->mbmi.sb_type));
+  if ((cm->frame_type == KEY_FRAME) || cm->intra_only) {
+    write_mb_modes_kf(cpi, m, bc, mi_row, mi_col);
+#ifdef ENTROPY_STATS
+    active_section = 8;
+#endif
+  } else {
+    pack_inter_mode_mvs(cpi, m, bc, mi_row, mi_col);
+#ifdef ENTROPY_STATS
+    active_section = 1;
+#endif
+  }
+
+  assert(*tok < tok_end);
+  pack_mb_tokens(bc, tok, tok_end);
+}
+
+static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
+                           TOKENEXTRA **tok, TOKENEXTRA *tok_end,
+                           int mi_row, int mi_col,
+                           BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  const int mis = cm->mode_info_stride;
+  int bwl, bhl;
+  int bsl = b_width_log2(bsize);
+  int bs = (1 << bsl) / 4;  // mode_info step for subsize
+  int n;
+  PARTITION_TYPE partition;
+  BLOCK_SIZE_TYPE subsize;
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  bwl = b_width_log2(m->mbmi.sb_type);
+  bhl = b_height_log2(m->mbmi.sb_type);
+
+  // parse the partition type
+  if ((bwl == bsl) && (bhl == bsl))
+    partition = PARTITION_NONE;
+  else if ((bwl == bsl) && (bhl < bsl))
+    partition = PARTITION_HORZ;
+  else if ((bwl < bsl) && (bhl == bsl))
+    partition = PARTITION_VERT;
+  else if ((bwl < bsl) && (bhl < bsl))
+    partition = PARTITION_SPLIT;
+  else
+    assert(0);
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index > 0)
+      return;
+
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    int pl;
+    int idx = check_bsize_coverage(cm, xd, mi_row, mi_col, bsize);
+    xd->left_seg_context = cm->left_seg_context + (mi_row & MI_MASK);
+    xd->above_seg_context = cm->above_seg_context + mi_col;
+    pl = partition_plane_context(xd, bsize);
+    // encode the partition information
+    if (idx == 0)
+      write_token(bc, vp9_partition_tree,
+                  cm->fc.partition_prob[cm->frame_type][pl],
+                  vp9_partition_encodings + partition);
+    else if (idx > 0)
+      vp9_write(bc, partition == PARTITION_SPLIT,
+                cm->fc.partition_prob[cm->frame_type][pl][idx]);
+  }
+
+  subsize = get_subsize(bsize, partition);
+  *(get_sb_index(xd, subsize)) = 0;
+
+  switch (partition) {
+    case PARTITION_NONE:
+      write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
+      break;
+    case PARTITION_HORZ:
+      write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
+      *(get_sb_index(xd, subsize)) = 1;
+      if ((mi_row + bs) < cm->mi_rows)
+        write_modes_b(cpi, m + bs * mis, bc, tok, tok_end, mi_row + bs, mi_col);
+      break;
+    case PARTITION_VERT:
+      write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
+      *(get_sb_index(xd, subsize)) = 1;
+      if ((mi_col + bs) < cm->mi_cols)
+        write_modes_b(cpi, m + bs, bc, tok, tok_end, mi_row, mi_col + bs);
+      break;
+    case PARTITION_SPLIT:
+      for (n = 0; n < 4; n++) {
+        int j = n >> 1, i = n & 0x01;
+        *(get_sb_index(xd, subsize)) = n;
+        write_modes_sb(cpi, m + j * bs * mis + i * bs, bc, tok, tok_end,
+                       mi_row + j * bs, mi_col + i * bs, subsize);
+      }
+      break;
+    default:
+      assert(0);
+  }
+
+  // update partition context
+  if (bsize >= BLOCK_SIZE_SB8X8 &&
+      (bsize == BLOCK_SIZE_SB8X8 || partition != PARTITION_SPLIT)) {
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    update_partition_context(xd, subsize, bsize);
+  }
+}
+
+static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
+                        TOKENEXTRA **tok, TOKENEXTRA *tok_end) {
+  VP9_COMMON *const c = &cpi->common;
+  const int mis = c->mode_info_stride;
+  MODE_INFO *m, *m_ptr = c->mi;
+  int mi_row, mi_col;
+
+  m_ptr += c->cur_tile_mi_col_start + c->cur_tile_mi_row_start * mis;
+
+  for (mi_row = c->cur_tile_mi_row_start;
+       mi_row < c->cur_tile_mi_row_end;
+       mi_row += 8, m_ptr += 8 * mis) {
+    m = m_ptr;
+    vpx_memset(c->left_seg_context, 0, sizeof(c->left_seg_context));
+    for (mi_col = c->cur_tile_mi_col_start;
+         mi_col < c->cur_tile_mi_col_end;
+         mi_col += 64 / MI_SIZE, m += 64 / MI_SIZE)
+      write_modes_sb(cpi, m, bc, tok, tok_end, mi_row, mi_col,
+                     BLOCK_SIZE_SB64X64);
+  }
+}
+
+/* This function is used for debugging probability trees. */
+static void print_prob_tree(vp9_coeff_probs *coef_probs, int block_types) {
+  /* print coef probability tree */
+  int i, j, k, l, m;
+  FILE *f = fopen("enc_tree_probs.txt", "a");
+  fprintf(f, "{\n");
+  for (i = 0; i < block_types; i++) {
+    fprintf(f, "  {\n");
+    for (j = 0; j < REF_TYPES; ++j) {
+      fprintf(f, "  {\n");
+      for (k = 0; k < COEF_BANDS; k++) {
+        fprintf(f, "    {\n");
+        for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
+          fprintf(f, "      {");
+          for (m = 0; m < ENTROPY_NODES; m++) {
+            fprintf(f, "%3u, ",
+                    (unsigned int)(coef_probs[i][j][k][l][m]));
+          }
+        }
+        fprintf(f, " }\n");
+      }
+      fprintf(f, "    }\n");
+    }
+    fprintf(f, "  }\n");
+  }
+  fprintf(f, "}\n");
+  fclose(f);
+}
+
+static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE txfm_size) {
+  vp9_coeff_probs_model *coef_probs = cpi->frame_coef_probs[txfm_size];
+  vp9_coeff_count *coef_counts = cpi->coef_counts[txfm_size];
+  unsigned int (*eob_branch_ct)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS] =
+      cpi->common.fc.eob_branch_counts[txfm_size];
+  vp9_coeff_stats *coef_branch_ct = cpi->frame_branch_ct[txfm_size];
+  vp9_prob full_probs[ENTROPY_NODES];
+  int i, j, k, l;
+
+  for (i = 0; i < BLOCK_TYPES; ++i) {
+    for (j = 0; j < REF_TYPES; ++j) {
+      for (k = 0; k < COEF_BANDS; ++k) {
+        for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
+          if (l >= 3 && k == 0)
+            continue;
+          vp9_tree_probs_from_distribution(vp9_coef_tree,
+                                           full_probs,
+                                           coef_branch_ct[i][j][k][l],
+                                           coef_counts[i][j][k][l], 0);
+          vpx_memcpy(coef_probs[i][j][k][l], full_probs,
+                     sizeof(vp9_prob) * UNCONSTRAINED_NODES);
+#if CONFIG_BALANCED_COEFTREE
+          coef_branch_ct[i][j][k][l][1][1] = eob_branch_ct[i][j][k][l] -
+                                             coef_branch_ct[i][j][k][l][1][0];
+          coef_probs[i][j][k][l][1] =
+              get_binary_prob(coef_branch_ct[i][j][k][l][1][0],
+                              coef_branch_ct[i][j][k][l][1][1]);
+#else
+          coef_branch_ct[i][j][k][l][0][1] = eob_branch_ct[i][j][k][l] -
+                                             coef_branch_ct[i][j][k][l][0][0];
+          coef_probs[i][j][k][l][0] =
+              get_binary_prob(coef_branch_ct[i][j][k][l][0][0],
+                              coef_branch_ct[i][j][k][l][0][1]);
+#endif
+#ifdef ENTROPY_STATS
+          if (!cpi->dummy_packing) {
+            int t;
+            for (t = 0; t < MAX_ENTROPY_TOKENS; ++t)
+              context_counters[txfm_size][i][j][k][l][t] +=
+                  coef_counts[i][j][k][l][t];
+            context_counters[txfm_size][i][j][k][l][MAX_ENTROPY_TOKENS] +=
+                eob_branch_ct[i][j][k][l];
+          }
+#endif
+        }
+      }
+    }
+  }
+}
+
+static void build_coeff_contexts(VP9_COMP *cpi) {
+  TX_SIZE t;
+  for (t = TX_4X4; t <= TX_32X32; t++)
+    build_tree_distribution(cpi, t);
+}
+
+static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
+                                     TX_SIZE tx_size) {
+  vp9_coeff_probs_model *new_frame_coef_probs = cpi->frame_coef_probs[tx_size];
+  vp9_coeff_probs_model *old_frame_coef_probs =
+      cpi->common.fc.coef_probs[tx_size];
+  vp9_coeff_stats *frame_branch_ct = cpi->frame_branch_ct[tx_size];
+  int i, j, k, l, t;
+  int update[2] = {0, 0};
+  int savings;
+
+  const int entropy_nodes_update = UNCONSTRAINED_NODES;
+
+  const int tstart = 0;
+  /* dry run to see if there is any udpate at all needed */
+  savings = 0;
+  for (i = 0; i < BLOCK_TYPES; ++i) {
+    for (j = 0; j < REF_TYPES; ++j) {
+      for (k = 0; k < COEF_BANDS; ++k) {
+        // int prev_coef_savings[ENTROPY_NODES] = {0};
+        for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
+          for (t = tstart; t < entropy_nodes_update; ++t) {
+            vp9_prob newp = new_frame_coef_probs[i][j][k][l][t];
+            const vp9_prob oldp = old_frame_coef_probs[i][j][k][l][t];
+            const vp9_prob upd = VP9_COEF_UPDATE_PROB;
+            int s;
+            int u = 0;
+
+            if (l >= 3 && k == 0)
+              continue;
+            if (t == PIVOT_NODE)
+              s = prob_diff_update_savings_search_model(
+                  frame_branch_ct[i][j][k][l][0],
+                  old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
+            else
+              s = prob_diff_update_savings_search(
+                  frame_branch_ct[i][j][k][l][t], oldp, &newp, upd);
+            if (s > 0 && newp != oldp)
+              u = 1;
+            if (u)
+              savings += s - (int)(vp9_cost_zero(upd));
+            else
+              savings -= (int)(vp9_cost_zero(upd));
+            update[u]++;
+          }
+        }
+      }
+    }
+  }
+
+  // printf("Update %d %d, savings %d\n", update[0], update[1], savings);
+  /* Is coef updated at all */
+  if (update[1] == 0 || savings < 0) {
+    vp9_write_bit(bc, 0);
+    return;
+  }
+  vp9_write_bit(bc, 1);
+  for (i = 0; i < BLOCK_TYPES; ++i) {
+    for (j = 0; j < REF_TYPES; ++j) {
+      for (k = 0; k < COEF_BANDS; ++k) {
+        // int prev_coef_savings[ENTROPY_NODES] = {0};
+        for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
+          // calc probs and branch cts for this frame only
+          for (t = tstart; t < entropy_nodes_update; ++t) {
+            vp9_prob newp = new_frame_coef_probs[i][j][k][l][t];
+            vp9_prob *oldp = old_frame_coef_probs[i][j][k][l] + t;
+            const vp9_prob upd = VP9_COEF_UPDATE_PROB;
+            int s;
+            int u = 0;
+            if (l >= 3 && k == 0)
+              continue;
+            if (t == PIVOT_NODE)
+              s = prob_diff_update_savings_search_model(
+                  frame_branch_ct[i][j][k][l][0],
+                  old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
+            else
+              s = prob_diff_update_savings_search(
+                  frame_branch_ct[i][j][k][l][t],
+                  *oldp, &newp, upd);
+            if (s > 0 && newp != *oldp)
+              u = 1;
+            vp9_write(bc, u, upd);
+#ifdef ENTROPY_STATS
+            if (!cpi->dummy_packing)
+              ++tree_update_hist[tx_size][i][j][k][l][t][u];
+#endif
+            if (u) {
+              /* send/use new probability */
+              write_prob_diff_update(bc, newp, *oldp);
+              *oldp = newp;
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+static void update_coef_probs(VP9_COMP* const cpi, vp9_writer* const bc) {
+  const TXFM_MODE txfm_mode = cpi->common.txfm_mode;
+
+  vp9_clear_system_state();
+
+  // Build the cofficient contexts based on counts collected in encode loop
+  build_coeff_contexts(cpi);
+
+  update_coef_probs_common(bc, cpi, TX_4X4);
+
+  // do not do this if not even allowed
+  if (txfm_mode > ONLY_4X4)
+    update_coef_probs_common(bc, cpi, TX_8X8);
+
+  if (txfm_mode > ALLOW_8X8)
+    update_coef_probs_common(bc, cpi, TX_16X16);
+
+  if (txfm_mode > ALLOW_16X16)
+    update_coef_probs_common(bc, cpi, TX_32X32);
+}
+
+static void encode_loopfilter(VP9_COMMON *pc, MACROBLOCKD *xd,
+                              struct vp9_write_bit_buffer *wb) {
+  int i;
+
+  // Encode the loop filter level and type
+  vp9_wb_write_literal(wb, pc->filter_level, 6);
+  vp9_wb_write_literal(wb, pc->sharpness_level, 3);
+
+  // Write out loop filter deltas applied at the MB level based on mode or
+  // ref frame (if they are enabled).
+  vp9_wb_write_bit(wb, xd->mode_ref_lf_delta_enabled);
+
+  if (xd->mode_ref_lf_delta_enabled) {
+    // Do the deltas need to be updated
+    vp9_wb_write_bit(wb, xd->mode_ref_lf_delta_update);
+    if (xd->mode_ref_lf_delta_update) {
+      // Send update
+      for (i = 0; i < MAX_REF_LF_DELTAS; i++) {
+        const int delta = xd->ref_lf_deltas[i];
+
+        // Frame level data
+        if (delta != xd->last_ref_lf_deltas[i]) {
+          xd->last_ref_lf_deltas[i] = delta;
+          vp9_wb_write_bit(wb, 1);
+
+          assert(delta != 0);
+          vp9_wb_write_literal(wb, abs(delta) & 0x3F, 6);
+          vp9_wb_write_bit(wb, delta < 0);
+        } else {
+          vp9_wb_write_bit(wb, 0);
+        }
+      }
+
+      // Send update
+      for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
+        const int delta = xd->mode_lf_deltas[i];
+        if (delta != xd->last_mode_lf_deltas[i]) {
+          xd->last_mode_lf_deltas[i] = delta;
+          vp9_wb_write_bit(wb, 1);
+
+          assert(delta != 0);
+          vp9_wb_write_literal(wb, abs(delta) & 0x3F, 6);
+          vp9_wb_write_bit(wb, delta < 0);
+        } else {
+          vp9_wb_write_bit(wb, 0);
+        }
+      }
+    }
+  }
+}
+
+static void write_delta_q(struct vp9_write_bit_buffer *wb, int delta_q) {
+  if (delta_q != 0) {
+    vp9_wb_write_bit(wb, 1);
+    vp9_wb_write_literal(wb, abs(delta_q), 4);
+    vp9_wb_write_bit(wb, delta_q < 0);
+  } else {
+    vp9_wb_write_bit(wb, 0);
+  }
+}
+
+static void encode_quantization(VP9_COMMON *cm,
+                                struct vp9_write_bit_buffer *wb) {
+  vp9_wb_write_literal(wb, cm->base_qindex, QINDEX_BITS);
+  write_delta_q(wb, cm->y_dc_delta_q);
+  write_delta_q(wb, cm->uv_dc_delta_q);
+  write_delta_q(wb, cm->uv_ac_delta_q);
+}
+
+
+static void encode_segmentation(VP9_COMP *cpi,
+                               struct vp9_write_bit_buffer *wb) {
+  int i, j;
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+
+  vp9_wb_write_bit(wb, xd->segmentation_enabled);
+  if (!xd->segmentation_enabled)
+    return;
+
+  // Segmentation map
+  vp9_wb_write_bit(wb, xd->update_mb_segmentation_map);
+  if (xd->update_mb_segmentation_map) {
+    // Select the coding strategy (temporal or spatial)
+    vp9_choose_segmap_coding_method(cpi);
+    // Write out probabilities used to decode unpredicted  macro-block segments
+    for (i = 0; i < MB_SEG_TREE_PROBS; i++) {
+      const int prob = xd->mb_segment_tree_probs[i];
+      const int update = prob != MAX_PROB;
+      vp9_wb_write_bit(wb, update);
+      if (update)
+        vp9_wb_write_literal(wb, prob, 8);
+    }
+
+    // Write out the chosen coding method.
+    vp9_wb_write_bit(wb, cm->temporal_update);
+    if (cm->temporal_update) {
+      for (i = 0; i < PREDICTION_PROBS; i++) {
+        const int prob = cm->segment_pred_probs[i];
+        const int update = prob != MAX_PROB;
+        vp9_wb_write_bit(wb, update);
+        if (update)
+          vp9_wb_write_literal(wb, prob, 8);
+      }
+    }
+  }
+
+  // Segmentation data
+  vp9_wb_write_bit(wb, xd->update_mb_segmentation_data);
+  if (xd->update_mb_segmentation_data) {
+    vp9_wb_write_bit(wb, xd->mb_segment_abs_delta);
+
+    for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+      for (j = 0; j < SEG_LVL_MAX; j++) {
+        const int active = vp9_segfeature_active(xd, i, j);
+        vp9_wb_write_bit(wb, active);
+        if (active) {
+          const int data = vp9_get_segdata(xd, i, j);
+          const int data_max = vp9_seg_feature_data_max(j);
+
+          if (vp9_is_segfeature_signed(j)) {
+            vp9_encode_unsigned_max(wb, abs(data), data_max);
+            vp9_wb_write_bit(wb, data < 0);
+          } else {
+            vp9_encode_unsigned_max(wb, data, data_max);
+          }
+        }
+      }
+    }
+  }
+}
+
+
+static void encode_txfm_probs(VP9_COMP *cpi, vp9_writer *w) {
+  VP9_COMMON *const cm = &cpi->common;
+
+  // Mode
+  vp9_write_literal(w, MIN(cm->txfm_mode, ALLOW_32X32), 2);
+  if (cm->txfm_mode >= ALLOW_32X32)
+    vp9_write_bit(w, cm->txfm_mode == TX_MODE_SELECT);
+
+  // Probabilities
+  if (cm->txfm_mode == TX_MODE_SELECT) {
+    int i, j;
+    unsigned int ct_8x8p[TX_SIZE_MAX_SB - 3][2];
+    unsigned int ct_16x16p[TX_SIZE_MAX_SB - 2][2];
+    unsigned int ct_32x32p[TX_SIZE_MAX_SB - 1][2];
+
+
+    for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+      tx_counts_to_branch_counts_8x8(cm->fc.tx_count_8x8p[i],
+                                     ct_8x8p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 3; j++) {
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_8x8p[i][j],
+                                  VP9_MODE_UPDATE_PROB, ct_8x8p[j]);
+      }
+    }
+    for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+      tx_counts_to_branch_counts_16x16(cm->fc.tx_count_16x16p[i],
+                                       ct_16x16p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 2; j++) {
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_16x16p[i][j],
+                                  VP9_MODE_UPDATE_PROB, ct_16x16p[j]);
+      }
+    }
+    for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+      tx_counts_to_branch_counts_32x32(cm->fc.tx_count_32x32p[i],
+                                       ct_32x32p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 1; j++) {
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_32x32p[i][j],
+                                  VP9_MODE_UPDATE_PROB, ct_32x32p[j]);
+      }
+    }
+#ifdef MODE_STATS
+    if (!cpi->dummy_packing)
+      update_tx_count_stats(cm);
+#endif
+  }
+}
+
+static void write_interp_filter_type(INTERPOLATIONFILTERTYPE type,
+                                     struct vp9_write_bit_buffer *wb) {
+  vp9_wb_write_bit(wb, type == SWITCHABLE);
+  if (type != SWITCHABLE)
+    vp9_wb_write_literal(wb, type, 2);
+}
+
+static void fix_mcomp_filter_type(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+
+  if (cm->mcomp_filter_type == SWITCHABLE) {
+    // Check to see if only one of the filters is actually used
+    int count[VP9_SWITCHABLE_FILTERS];
+    int i, j, c = 0;
+    for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
+      count[i] = 0;
+      for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j)
+        count[i] += cm->fc.switchable_interp_count[j][i];
+      c += (count[i] > 0);
+    }
+    if (c == 1) {
+      // Only one filter is used. So set the filter at frame level
+      for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
+        if (count[i]) {
+          cm->mcomp_filter_type = vp9_switchable_interp[i];
+          break;
+        }
+      }
+    }
+  }
+}
+
+static void write_tile_info(VP9_COMMON *cm, struct vp9_write_bit_buffer *wb) {
+  int min_log2_tiles, delta_log2_tiles, n_tile_bits, n;
+  vp9_get_tile_n_bits(cm, &min_log2_tiles, &delta_log2_tiles);
+  n_tile_bits = cm->log2_tile_columns - min_log2_tiles;
+  for (n = 0; n < delta_log2_tiles; n++) {
+    if (n_tile_bits--) {
+      vp9_wb_write_bit(wb, 1);
+    } else {
+      vp9_wb_write_bit(wb, 0);
+      break;
+    }
+  }
+
+  vp9_wb_write_bit(wb, cm->log2_tile_rows != 0);
+  if (cm->log2_tile_rows != 0)
+    vp9_wb_write_bit(wb, cm->log2_tile_rows != 1);
+}
+
+static int get_refresh_mask(VP9_COMP *cpi) {
+    // Should the GF or ARF be updated using the transmitted frame or buffer
+#if CONFIG_MULTIPLE_ARF
+    if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
+        !cpi->refresh_alt_ref_frame) {
+#else
+    if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame) {
+#endif
+      // Preserve the previously existing golden frame and update the frame in
+      // the alt ref slot instead. This is highly specific to the use of
+      // alt-ref as a forward reference, and this needs to be generalized as
+      // other uses are implemented (like RTC/temporal scaling)
+      //
+      // gld_fb_idx and alt_fb_idx need to be swapped for future frames, but
+      // that happens in vp9_onyx_if.c:update_reference_frames() so that it can
+      // be done outside of the recode loop.
+      return (cpi->refresh_last_frame << cpi->lst_fb_idx) |
+             (cpi->refresh_golden_frame << cpi->alt_fb_idx);
+    } else {
+      int arf_idx = cpi->alt_fb_idx;
+#if CONFIG_MULTIPLE_ARF
+      // Determine which ARF buffer to use to encode this ARF frame.
+      if (cpi->multi_arf_enabled) {
+        int sn = cpi->sequence_number;
+        arf_idx = (cpi->frame_coding_order[sn] < 0) ?
+            cpi->arf_buffer_idx[sn + 1] :
+            cpi->arf_buffer_idx[sn];
+      }
+#endif
+      return (cpi->refresh_last_frame << cpi->lst_fb_idx) |
+             (cpi->refresh_golden_frame << cpi->gld_fb_idx) |
+             (cpi->refresh_alt_ref_frame << arf_idx);
+    }
+}
+
+static void write_display_size(VP9_COMP *cpi, struct vp9_write_bit_buffer *wb) {
+  VP9_COMMON *const cm = &cpi->common;
+
+  const int scaling_active = cm->width != cm->display_width ||
+                             cm->height != cm->display_height;
+  vp9_wb_write_bit(wb, scaling_active);
+  if (scaling_active) {
+    vp9_wb_write_literal(wb, cm->display_width - 1, 16);
+    vp9_wb_write_literal(wb, cm->display_height - 1, 16);
+  }
+}
+
+static void write_frame_size(VP9_COMP *cpi,
+                             struct vp9_write_bit_buffer *wb) {
+  VP9_COMMON *const cm = &cpi->common;
+  vp9_wb_write_literal(wb, cm->width - 1, 16);
+  vp9_wb_write_literal(wb, cm->height - 1, 16);
+
+  write_display_size(cpi, wb);
+}
+
+static void write_frame_size_with_refs(VP9_COMP *cpi,
+                                       struct vp9_write_bit_buffer *wb) {
+  VP9_COMMON *const cm = &cpi->common;
+  int refs[ALLOWED_REFS_PER_FRAME] = {cpi->lst_fb_idx, cpi->gld_fb_idx,
+                                      cpi->alt_fb_idx};
+  int i, found = 0;
+
+  for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
+    YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->ref_frame_map[refs[i]]];
+    found = cm->width == cfg->y_crop_width &&
+            cm->height == cfg->y_crop_height;
+    vp9_wb_write_bit(wb, found);
+    if (found)
+      break;
+  }
+
+  if (!found) {
+    vp9_wb_write_literal(wb, cm->width - 1, 16);
+    vp9_wb_write_literal(wb, cm->height - 1, 16);
+  }
+
+  write_display_size(cpi, wb);
+}
+
+static void write_sync_code(struct vp9_write_bit_buffer *wb) {
+  vp9_wb_write_literal(wb, SYNC_CODE_0, 8);
+  vp9_wb_write_literal(wb, SYNC_CODE_1, 8);
+  vp9_wb_write_literal(wb, SYNC_CODE_2, 8);
+}
+
+static void write_uncompressed_header(VP9_COMP *cpi,
+                                      struct vp9_write_bit_buffer *wb) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+
+  // frame marker bits
+  vp9_wb_write_literal(wb, 0x2, 2);
+
+  // bitstream version.
+  // 00 - profile 0. 4:2:0 only
+  // 10 - profile 1. adds 4:4:4, 4:2:2, alpha
+  vp9_wb_write_bit(wb, cm->version);
+  vp9_wb_write_bit(wb, 0);
+
+  vp9_wb_write_bit(wb, 0);
+  vp9_wb_write_bit(wb, cm->frame_type);
+  vp9_wb_write_bit(wb, cm->show_frame);
+  vp9_wb_write_bit(wb, cm->error_resilient_mode);
+
+  if (cm->frame_type == KEY_FRAME) {
+    write_sync_code(wb);
+    // colorspaces
+    // 000 - Unknown
+    // 001 - BT.601
+    // 010 - BT.709
+    // 011 - SMPTE-170
+    // 100 - SMPTE-240
+    // 101 - Reserved
+    // 110 - Reserved
+    // 111 - sRGB (RGB)
+    vp9_wb_write_literal(wb, 0, 3);
+    if (1 /* colorspace != sRGB */) {
+      vp9_wb_write_bit(wb, 0);  // 0: [16, 235] (i.e. xvYCC), 1: [0, 255]
+      if (cm->version == 1) {
+        vp9_wb_write_bit(wb, cm->subsampling_x);
+        vp9_wb_write_bit(wb, cm->subsampling_y);
+        vp9_wb_write_bit(wb, 0);  // has extra plane
+      }
+    } else {
+      assert(cm->version == 1);
+      vp9_wb_write_bit(wb, 0);  // has extra plane
+    }
+
+    write_frame_size(cpi, wb);
+  } else {
+    const int refs[ALLOWED_REFS_PER_FRAME] = {cpi->lst_fb_idx, cpi->gld_fb_idx,
+                                              cpi->alt_fb_idx};
+    if (!cm->show_frame)
+      vp9_wb_write_bit(wb, cm->intra_only);
+
+    if (!cm->error_resilient_mode)
+      vp9_wb_write_literal(wb, cm->reset_frame_context, 2);
+
+    if (cm->intra_only) {
+      write_sync_code(wb);
+
+      vp9_wb_write_literal(wb, get_refresh_mask(cpi), NUM_REF_FRAMES);
+      write_frame_size(cpi, wb);
+    } else {
+      int i;
+      vp9_wb_write_literal(wb, get_refresh_mask(cpi), NUM_REF_FRAMES);
+      for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
+        vp9_wb_write_literal(wb, refs[i], NUM_REF_FRAMES_LG2);
+        vp9_wb_write_bit(wb, cm->ref_frame_sign_bias[LAST_FRAME + i]);
+      }
+
+      write_frame_size_with_refs(cpi, wb);
+
+      vp9_wb_write_bit(wb, xd->allow_high_precision_mv);
+
+      fix_mcomp_filter_type(cpi);
+      write_interp_filter_type(cm->mcomp_filter_type, wb);
+    }
+  }
+
+  if (!cm->error_resilient_mode) {
+    vp9_wb_write_bit(wb, cm->refresh_frame_context);
+    vp9_wb_write_bit(wb, cm->frame_parallel_decoding_mode);
+  }
+
+  vp9_wb_write_literal(wb, cm->frame_context_idx, NUM_FRAME_CONTEXTS_LG2);
+
+  encode_loopfilter(cm, xd, wb);
+  encode_quantization(cm, wb);
+  encode_segmentation(cpi, wb);
+
+  write_tile_info(cm, wb);
+}
+
+void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, unsigned long *size) {
+  int i, bytes_packed;
+  VP9_COMMON *const pc = &cpi->common;
+  vp9_writer header_bc, residual_bc;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+
+  uint8_t *cx_data = dest;
+  struct vp9_write_bit_buffer wb = {dest, 0};
+  struct vp9_write_bit_buffer first_partition_size_wb;
+
+  write_uncompressed_header(cpi, &wb);
+  first_partition_size_wb = wb;
+  vp9_wb_write_literal(&wb, 0, 16);  // don't know in advance first part. size
+
+  bytes_packed = vp9_rb_bytes_written(&wb);
+  cx_data += bytes_packed;
+
+  compute_update_table();
+
+  vp9_start_encode(&header_bc, cx_data);
+
+#ifdef ENTROPY_STATS
+  if (pc->frame_type == INTER_FRAME)
+    active_section = 0;
+  else
+    active_section = 7;
+#endif
+
+  vp9_clear_system_state();  // __asm emms;
+
+  vp9_copy(pc->fc.pre_coef_probs, pc->fc.coef_probs);
+  vp9_copy(pc->fc.pre_y_mode_prob, pc->fc.y_mode_prob);
+  vp9_copy(pc->fc.pre_uv_mode_prob, pc->fc.uv_mode_prob);
+  vp9_copy(pc->fc.pre_partition_prob, pc->fc.partition_prob[INTER_FRAME]);
+  pc->fc.pre_nmvc = pc->fc.nmvc;
+  vp9_copy(pc->fc.pre_switchable_interp_prob, pc->fc.switchable_interp_prob);
+  vp9_copy(pc->fc.pre_inter_mode_probs, pc->fc.inter_mode_probs);
+  vp9_copy(pc->fc.pre_intra_inter_prob, pc->fc.intra_inter_prob);
+  vp9_copy(pc->fc.pre_comp_inter_prob, pc->fc.comp_inter_prob);
+  vp9_copy(pc->fc.pre_comp_ref_prob, pc->fc.comp_ref_prob);
+  vp9_copy(pc->fc.pre_single_ref_prob, pc->fc.single_ref_prob);
+  vp9_copy(pc->fc.pre_tx_probs_8x8p, pc->fc.tx_probs_8x8p);
+  vp9_copy(pc->fc.pre_tx_probs_16x16p, pc->fc.tx_probs_16x16p);
+  vp9_copy(pc->fc.pre_tx_probs_32x32p, pc->fc.tx_probs_32x32p);
+  vp9_copy(pc->fc.pre_mbskip_probs, pc->fc.mbskip_probs);
+
+  if (xd->lossless) {
+    pc->txfm_mode = ONLY_4X4;
+  } else {
+    encode_txfm_probs(cpi, &header_bc);
+  }
+
+  update_coef_probs(cpi, &header_bc);
+
+#ifdef ENTROPY_STATS
+  active_section = 2;
+#endif
+
+  vp9_update_skip_probs(cpi, &header_bc);
+
+  if (pc->frame_type != KEY_FRAME) {
+#ifdef ENTROPY_STATS
+    active_section = 1;
+#endif
+
+    update_inter_mode_probs(pc, &header_bc);
+    vp9_zero(cpi->common.fc.inter_mode_counts);
+
+    if (pc->mcomp_filter_type == SWITCHABLE)
+      update_switchable_interp_probs(cpi, &header_bc);
+
+    for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
+      vp9_cond_prob_diff_update(&header_bc, &pc->fc.intra_inter_prob[i],
+                                VP9_MODE_UPDATE_PROB,
+                                cpi->intra_inter_count[i]);
+
+    if (pc->allow_comp_inter_inter) {
+      const int comp_pred_mode = cpi->common.comp_pred_mode;
+      const int use_compound_pred = (comp_pred_mode != SINGLE_PREDICTION_ONLY);
+      const int use_hybrid_pred = (comp_pred_mode == HYBRID_PREDICTION);
+
+      vp9_write_bit(&header_bc, use_compound_pred);
+      if (use_compound_pred) {
+        vp9_write_bit(&header_bc, use_hybrid_pred);
+        if (use_hybrid_pred) {
+          for (i = 0; i < COMP_INTER_CONTEXTS; i++)
+            vp9_cond_prob_diff_update(&header_bc, &pc->fc.comp_inter_prob[i],
+                                      VP9_MODE_UPDATE_PROB,
+                                      cpi->comp_inter_count[i]);
+        }
+      }
+    }
+
+    if (pc->comp_pred_mode != COMP_PREDICTION_ONLY) {
+      for (i = 0; i < REF_CONTEXTS; i++) {
+        vp9_cond_prob_diff_update(&header_bc, &pc->fc.single_ref_prob[i][0],
+                                  VP9_MODE_UPDATE_PROB,
+                                  cpi->single_ref_count[i][0]);
+        vp9_cond_prob_diff_update(&header_bc, &pc->fc.single_ref_prob[i][1],
+                                  VP9_MODE_UPDATE_PROB,
+                                  cpi->single_ref_count[i][1]);
+      }
+    }
+
+    if (pc->comp_pred_mode != SINGLE_PREDICTION_ONLY) {
+      for (i = 0; i < REF_CONTEXTS; i++)
+        vp9_cond_prob_diff_update(&header_bc, &pc->fc.comp_ref_prob[i],
+                                  VP9_MODE_UPDATE_PROB,
+                                  cpi->comp_ref_count[i]);
+    }
+
+    update_mbintra_mode_probs(cpi, &header_bc);
+
+    for (i = 0; i < NUM_PARTITION_CONTEXTS; ++i) {
+      vp9_prob Pnew[PARTITION_TYPES - 1];
+      unsigned int bct[PARTITION_TYPES - 1][2];
+      update_mode(&header_bc, PARTITION_TYPES, vp9_partition_encodings,
+                  vp9_partition_tree, Pnew,
+                  pc->fc.partition_prob[pc->frame_type][i], bct,
+                  (unsigned int *)cpi->partition_count[i]);
+    }
+
+    vp9_write_nmv_probs(cpi, xd->allow_high_precision_mv, &header_bc);
+  }
+
+
+  vp9_stop_encode(&header_bc);
+
+
+  // first partition size
+  assert(header_bc.pos <= 0xffff);
+  vp9_wb_write_literal(&first_partition_size_wb, header_bc.pos, 16);
+  *size = bytes_packed + header_bc.pos;
+
+  {
+    int tile_row, tile_col, total_size = 0;
+    unsigned char *data_ptr = cx_data + header_bc.pos;
+    TOKENEXTRA *tok[4][1 << 6], *tok_end;
+
+    vpx_memset(cpi->common.above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
+               mi_cols_aligned_to_sb(&cpi->common));
+    tok[0][0] = cpi->tok;
+    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+      if (tile_row) {
+        tok[tile_row][0] = tok[tile_row - 1][pc->tile_columns - 1] +
+                           cpi->tok_count[tile_row - 1][pc->tile_columns - 1];
+      }
+      for (tile_col = 1; tile_col < pc->tile_columns; tile_col++) {
+        tok[tile_row][tile_col] = tok[tile_row][tile_col - 1] +
+                                  cpi->tok_count[tile_row][tile_col - 1];
+      }
+    }
+
+    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+      vp9_get_tile_row_offsets(pc, tile_row);
+      for (tile_col = 0; tile_col < pc->tile_columns; tile_col++) {
+        vp9_get_tile_col_offsets(pc, tile_col);
+        tok_end = tok[tile_row][tile_col] + cpi->tok_count[tile_row][tile_col];
+
+        if (tile_col < pc->tile_columns - 1 || tile_row < pc->tile_rows - 1)
+          vp9_start_encode(&residual_bc, data_ptr + total_size + 4);
+        else
+          vp9_start_encode(&residual_bc, data_ptr + total_size);
+        write_modes(cpi, &residual_bc, &tok[tile_row][tile_col], tok_end);
+        assert(tok[tile_row][tile_col] == tok_end);
+        vp9_stop_encode(&residual_bc);
+        if (tile_col < pc->tile_columns - 1 || tile_row < pc->tile_rows - 1) {
+          // size of this tile
+          write_be32(data_ptr + total_size, residual_bc.pos);
+          total_size += 4;
+        }
+
+        total_size += residual_bc.pos;
+      }
+    }
+
+    *size += total_size;
+  }
+}
+
+#ifdef ENTROPY_STATS
+static void print_tree_update_for_type(FILE *f,
+                                       vp9_coeff_stats *tree_update_hist,
+                                       int block_types, const char *header) {
+  int i, j, k, l, m;
+
+  fprintf(f, "const vp9_coeff_prob %s = {\n", header);
+  for (i = 0; i < block_types; i++) {
+    fprintf(f, "  { \n");
+    for (j = 0; j < REF_TYPES; j++) {
+      fprintf(f, "  { \n");
+      for (k = 0; k < COEF_BANDS; k++) {
+        fprintf(f, "    {\n");
+        for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
+          fprintf(f, "      {");
+          for (m = 0; m < ENTROPY_NODES; m++) {
+            fprintf(f, "%3d, ",
+                    get_binary_prob(tree_update_hist[i][j][k][l][m][0],
+                                    tree_update_hist[i][j][k][l][m][1]));
+          }
+          fprintf(f, "},\n");
+        }
+        fprintf(f, "},\n");
+      }
+      fprintf(f, "    },\n");
+    }
+    fprintf(f, "  },\n");
+  }
+  fprintf(f, "};\n");
+}
+
+void print_tree_update_probs() {
+  FILE *f = fopen("coefupdprob.h", "w");
+  fprintf(f, "\n/* Update probabilities for token entropy tree. */\n\n");
+
+  print_tree_update_for_type(f, tree_update_hist[TX_4X4],   BLOCK_TYPES,
+                             "vp9_coef_update_probs_4x4[BLOCK_TYPES]");
+  print_tree_update_for_type(f, tree_update_hist[TX_8X8],   BLOCK_TYPES,
+                             "vp9_coef_update_probs_8x8[BLOCK_TYPES]");
+  print_tree_update_for_type(f, tree_update_hist[TX_16X16], BLOCK_TYPES,
+                             "vp9_coef_update_probs_16x16[BLOCK_TYPES]");
+  print_tree_update_for_type(f, tree_update_hist[TX_32X32], BLOCK_TYPES,
+                             "vp9_coef_update_probs_32x32[BLOCK_TYPES]");
+
+  fclose(f);
+  f = fopen("treeupdate.bin", "wb");
+  fwrite(tree_update_hist, sizeof(tree_update_hist), 1, f);
+  fclose(f);
+}
+#endif
diff --git a/libvpx/vp9/encoder/vp9_bitstream.h b/libvpx/vp9/encoder/vp9_bitstream.h
new file mode 100644
index 0000000..b3dbee1
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_bitstream.h
@@ -0,0 +1,17 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_BITSTREAM_H_
+#define VP9_ENCODER_VP9_BITSTREAM_H_
+
+void vp9_update_skip_probs(VP9_COMP *cpi, vp9_writer *bc);
+
+#endif  // VP9_ENCODER_VP9_BITSTREAM_H_
diff --git a/libvpx/vp9/encoder/vp9_block.h b/libvpx/vp9/encoder/vp9_block.h
new file mode 100644
index 0000000..59cc3d9
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_block.h
@@ -0,0 +1,176 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_BLOCK_H_
+#define VP9_ENCODER_VP9_BLOCK_H_
+
+#include "vp9/common/vp9_onyx.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+// motion search site
+typedef struct {
+  MV mv;
+  int offset;
+} search_site;
+
+typedef struct {
+  int count;
+  struct {
+    MB_PREDICTION_MODE mode;
+    int_mv mv;
+    int_mv second_mv;
+  } bmi[4];
+} PARTITION_INFO;
+
+// Structure to hold snapshot of coding context during the mode picking process
+// TODO Do we need all of these?
+typedef struct {
+  MODE_INFO mic;
+  PARTITION_INFO partition_info;
+  int skip;
+  int_mv best_ref_mv;
+  int_mv second_best_ref_mv;
+  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
+  int rate;
+  int distortion;
+  int64_t intra_error;
+  int best_mode_index;
+  int rddiv;
+  int rdmult;
+  int hybrid_pred_diff;
+  int comp_pred_diff;
+  int single_pred_diff;
+  int64_t txfm_rd_diff[NB_TXFM_MODES];
+
+  // Bit flag for each mode whether it has high error in comparison to others.
+  unsigned int modes_with_high_error;
+
+  // Bit flag for each ref frame whether it has high error compared to others.
+  unsigned int frames_with_high_error;
+} PICK_MODE_CONTEXT;
+
+struct macroblock_plane {
+  DECLARE_ALIGNED(16, int16_t, src_diff[64*64]);
+  DECLARE_ALIGNED(16, int16_t, coeff[64*64]);
+  struct buf_2d src;
+
+  // Quantizer setings
+  int16_t *quant;
+  uint8_t *quant_shift;
+  int16_t *zbin;
+  int16_t *zrun_zbin_boost;
+  int16_t *round;
+
+  // Zbin Over Quant value
+  int16_t zbin_extra;
+};
+
+typedef struct macroblock MACROBLOCK;
+struct macroblock {
+  struct macroblock_plane plane[MAX_MB_PLANE];
+
+  MACROBLOCKD e_mbd;
+  int skip_block;
+  PARTITION_INFO *partition_info; /* work pointer */
+  PARTITION_INFO *pi;   /* Corresponds to upper left visible macroblock */
+  PARTITION_INFO *pip;  /* Base of allocated array */
+
+  search_site *ss;
+  int ss_count;
+  int searches_per_step;
+
+  int errorperbit;
+  int sadperbit16;
+  int sadperbit4;
+  int rddiv;
+  int rdmult;
+  unsigned int *mb_activity_ptr;
+  int *mb_norm_activity_ptr;
+  signed int act_zbin_adj;
+
+  int mv_best_ref_index[MAX_REF_FRAMES];
+
+  int nmvjointcost[MV_JOINTS];
+  int nmvcosts[2][MV_VALS];
+  int *nmvcost[2];
+  int nmvcosts_hp[2][MV_VALS];
+  int *nmvcost_hp[2];
+  int **mvcost;
+
+  int nmvjointsadcost[MV_JOINTS];
+  int nmvsadcosts[2][MV_VALS];
+  int *nmvsadcost[2];
+  int nmvsadcosts_hp[2][MV_VALS];
+  int *nmvsadcost_hp[2];
+  int **mvsadcost;
+
+  int mbmode_cost[MB_MODE_COUNT];
+  int intra_uv_mode_cost[2][MB_MODE_COUNT];
+  int y_mode_costs[VP9_INTRA_MODES][VP9_INTRA_MODES][VP9_INTRA_MODES];
+  int switchable_interp_costs[VP9_SWITCHABLE_FILTERS + 1]
+                             [VP9_SWITCHABLE_FILTERS];
+
+  // These define limits to motion vector components to prevent them
+  // from extending outside the UMV borders
+  int mv_col_min;
+  int mv_col_max;
+  int mv_row_min;
+  int mv_row_max;
+
+  int skip;
+
+  int encode_breakout;
+
+  unsigned char *active_ptr;
+
+  // note that token_costs is the cost when eob node is skipped
+  vp9_coeff_count token_costs[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  vp9_coeff_count token_costs_noskip[TX_SIZE_MAX_SB][BLOCK_TYPES];
+
+  int optimize;
+
+  // indicate if it is in the rd search loop or encoding process
+  int rd_search;
+
+  // TODO(jingning): Need to refactor the structure arrays that buffers the
+  // coding mode decisions of each partition type.
+  PICK_MODE_CONTEXT ab4x4_context[4][4][4];
+  PICK_MODE_CONTEXT sb8x4_context[4][4][4];
+  PICK_MODE_CONTEXT sb4x8_context[4][4][4];
+  PICK_MODE_CONTEXT sb8x8_context[4][4][4];
+  PICK_MODE_CONTEXT sb8x16_context[4][4][2];
+  PICK_MODE_CONTEXT sb16x8_context[4][4][2];
+  PICK_MODE_CONTEXT mb_context[4][4];
+  PICK_MODE_CONTEXT sb32x16_context[4][2];
+  PICK_MODE_CONTEXT sb16x32_context[4][2];
+  // when 4 MBs share coding parameters:
+  PICK_MODE_CONTEXT sb32_context[4];
+  PICK_MODE_CONTEXT sb32x64_context[2];
+  PICK_MODE_CONTEXT sb64x32_context[2];
+  PICK_MODE_CONTEXT sb64_context;
+  int partition_cost[NUM_PARTITION_CONTEXTS][PARTITION_TYPES];
+
+  BLOCK_SIZE_TYPE b_partitioning[4][4][4];
+  BLOCK_SIZE_TYPE mb_partitioning[4][4];
+  BLOCK_SIZE_TYPE sb_partitioning[4];
+  BLOCK_SIZE_TYPE sb64_partitioning;
+
+  void (*fwd_txm4x4)(int16_t *input, int16_t *output, int pitch);
+  void (*fwd_txm8x4)(int16_t *input, int16_t *output, int pitch);
+  void (*fwd_txm8x8)(int16_t *input, int16_t *output, int pitch);
+  void (*fwd_txm16x16)(int16_t *input, int16_t *output, int pitch);
+  void (*quantize_b_4x4)(MACROBLOCK *x, int b_idx, TX_TYPE tx_type,
+                         int y_blocks);
+};
+
+#endif  // VP9_ENCODER_VP9_BLOCK_H_
diff --git a/libvpx/vp9/encoder/vp9_boolhuff.c b/libvpx/vp9/encoder/vp9_boolhuff.c
new file mode 100644
index 0000000..0f1aa59
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_boolhuff.c
@@ -0,0 +1,63 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include "vp9/encoder/vp9_boolhuff.h"
+#include "vp9/common/vp9_entropy.h"
+
+#if defined(SECTIONBITS_OUTPUT)
+unsigned __int64 Sectionbits[500];
+
+#endif
+
+#ifdef ENTROPY_STATS
+unsigned int active_section = 0;
+#endif
+
+const unsigned int vp9_prob_cost[256] = {
+  2047, 2047, 1791, 1641, 1535, 1452, 1385, 1328, 1279, 1235, 1196, 1161, 1129, 1099, 1072, 1046,
+  1023, 1000,  979,  959,  940,  922,  905,  889,  873,  858,  843,  829,  816,  803,  790,  778,
+  767,  755,  744,  733,  723,  713,  703,  693,  684,  675,  666,  657,  649,  641,  633,  625,
+  617,  609,  602,  594,  587,  580,  573,  567,  560,  553,  547,  541,  534,  528,  522,  516,
+  511,  505,  499,  494,  488,  483,  477,  472,  467,  462,  457,  452,  447,  442,  437,  433,
+  428,  424,  419,  415,  410,  406,  401,  397,  393,  389,  385,  381,  377,  373,  369,  365,
+  361,  357,  353,  349,  346,  342,  338,  335,  331,  328,  324,  321,  317,  314,  311,  307,
+  304,  301,  297,  294,  291,  288,  285,  281,  278,  275,  272,  269,  266,  263,  260,  257,
+  255,  252,  249,  246,  243,  240,  238,  235,  232,  229,  227,  224,  221,  219,  216,  214,
+  211,  208,  206,  203,  201,  198,  196,  194,  191,  189,  186,  184,  181,  179,  177,  174,
+  172,  170,  168,  165,  163,  161,  159,  156,  154,  152,  150,  148,  145,  143,  141,  139,
+  137,  135,  133,  131,  129,  127,  125,  123,  121,  119,  117,  115,  113,  111,  109,  107,
+  105,  103,  101,   99,   97,   95,   93,   92,   90,   88,   86,   84,   82,   81,   79,   77,
+  75,   73,   72,   70,   68,   66,   65,   63,   61,   60,   58,   56,   55,   53,   51,   50,
+  48,   46,   45,   43,   41,   40,   38,   37,   35,   33,   32,   30,   29,   27,   25,   24,
+  22,   21,   19,   18,   16,   15,   13,   12,   10,    9,    7,    6,    4,    3,    1,   1
+};
+
+void vp9_start_encode(vp9_writer *br, uint8_t *source) {
+  br->lowvalue = 0;
+  br->range    = 255;
+  br->value    = 0;
+  br->count    = -24;
+  br->buffer   = source;
+  br->pos      = 0;
+  vp9_write_bit(br, 0);
+}
+
+void vp9_stop_encode(vp9_writer *br) {
+  int i;
+
+  for (i = 0; i < 32; i++)
+    vp9_write_bit(br, 0);
+
+  // Ensure there's no ambigous collision with any index marker bytes
+  if ((br->buffer[br->pos - 1] & 0xe0) == 0xc0)
+    br->buffer[br->pos++] = 0;
+}
+
diff --git a/libvpx/vp9/encoder/vp9_boolhuff.h b/libvpx/vp9/encoder/vp9_boolhuff.h
new file mode 100644
index 0000000..c3f340d
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_boolhuff.h
@@ -0,0 +1,115 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/****************************************************************************
+*
+*   Module Title :     vp9_boolhuff.h
+*
+*   Description  :     Bool Coder header file.
+*
+****************************************************************************/
+#ifndef VP9_ENCODER_VP9_BOOLHUFF_H_
+#define VP9_ENCODER_VP9_BOOLHUFF_H_
+
+#include "vpx_ports/mem.h"
+
+typedef struct {
+  unsigned int lowvalue;
+  unsigned int range;
+  unsigned int value;
+  int count;
+  unsigned int pos;
+  uint8_t *buffer;
+
+  // Variables used to track bit costs without outputing to the bitstream
+  unsigned int  measure_cost;
+  unsigned long bit_counter;
+} vp9_writer;
+
+extern const unsigned int vp9_prob_cost[256];
+
+void vp9_start_encode(vp9_writer *bc, uint8_t *buffer);
+void vp9_stop_encode(vp9_writer *bc);
+
+DECLARE_ALIGNED(16, extern const unsigned char, vp9_norm[256]);
+
+static void vp9_write(vp9_writer *br, int bit, int probability) {
+  unsigned int split;
+  int count = br->count;
+  unsigned int range = br->range;
+  unsigned int lowvalue = br->lowvalue;
+  register unsigned int shift;
+
+#ifdef ENTROPY_STATS
+#if defined(SECTIONBITS_OUTPUT)
+
+  if (bit)
+    Sectionbits[active_section] += vp9_prob_cost[255 - probability];
+  else
+    Sectionbits[active_section] += vp9_prob_cost[probability];
+
+#endif
+#endif
+
+  split = 1 + (((range - 1) * probability) >> 8);
+
+  range = split;
+
+  if (bit) {
+    lowvalue += split;
+    range = br->range - split;
+  }
+
+  shift = vp9_norm[range];
+
+  range <<= shift;
+  count += shift;
+
+  if (count >= 0) {
+    int offset = shift - count;
+
+    if ((lowvalue << (offset - 1)) & 0x80000000) {
+      int x = br->pos - 1;
+
+      while (x >= 0 && br->buffer[x] == 0xff) {
+        br->buffer[x] = 0;
+        x--;
+      }
+
+      br->buffer[x] += 1;
+    }
+
+    br->buffer[br->pos++] = (lowvalue >> (24 - offset));
+    lowvalue <<= offset;
+    shift = count;
+    lowvalue &= 0xffffff;
+    count -= 8;
+  }
+
+  lowvalue <<= shift;
+  br->count = count;
+  br->lowvalue = lowvalue;
+  br->range = range;
+}
+
+static void vp9_write_bit(vp9_writer *w, int bit) {
+  vp9_write(w, bit, 128);  // vp9_prob_half
+}
+
+static void vp9_write_literal(vp9_writer *w, int data, int bits) {
+  int bit;
+
+  for (bit = bits - 1; bit >= 0; bit--)
+    vp9_write_bit(w, 1 & (data >> bit));
+}
+
+
+#endif  // VP9_ENCODER_VP9_BOOLHUFF_H_
diff --git a/libvpx/vp9/encoder/vp9_dct.c b/libvpx/vp9/encoder/vp9_dct.c
new file mode 100644
index 0000000..a90bcf5
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_dct.c
@@ -0,0 +1,1381 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <assert.h>
+#include <math.h>
+#include "./vpx_config.h"
+#include "vp9/common/vp9_systemdependent.h"
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_idct.h"
+
+static void fdct4_1d(int16_t *input, int16_t *output) {
+  int16_t step[4];
+  int temp1, temp2;
+
+  step[0] = input[0] + input[3];
+  step[1] = input[1] + input[2];
+  step[2] = input[1] - input[2];
+  step[3] = input[0] - input[3];
+
+  temp1 = (step[0] + step[1]) * cospi_16_64;
+  temp2 = (step[0] - step[1]) * cospi_16_64;
+  output[0] = dct_const_round_shift(temp1);
+  output[2] = dct_const_round_shift(temp2);
+  temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
+  temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
+  output[1] = dct_const_round_shift(temp1);
+  output[3] = dct_const_round_shift(temp2);
+}
+
+void vp9_short_fdct4x4_c(int16_t *input, int16_t *output, int pitch) {
+  // The 2D transform is done with two passes which are actually pretty
+  // similar. In the first one, we transform the columns and transpose
+  // the results. In the second one, we transform the rows. To achieve that,
+  // as the first pass results are transposed, we tranpose the columns (that
+  // is the transposed rows) and transpose the results (so that it goes back
+  // in normal/row positions).
+  const int stride = pitch >> 1;
+  int pass;
+  // We need an intermediate buffer between passes.
+  int16_t intermediate[4 * 4];
+  int16_t *in = input;
+  int16_t *out = intermediate;
+  // Do the two transform/transpose passes
+  for (pass = 0; pass < 2; ++pass) {
+    /*canbe16*/ int input[4];
+    /*canbe16*/ int step[4];
+    /*needs32*/ int temp1, temp2;
+    int i;
+    for (i = 0; i < 4; ++i) {
+      // Load inputs.
+      if (0 == pass) {
+        input[0] = in[0 * stride] << 4;
+        input[1] = in[1 * stride] << 4;
+        input[2] = in[2 * stride] << 4;
+        input[3] = in[3 * stride] << 4;
+        if (i == 0 && input[0]) {
+          input[0] += 1;
+        }
+      } else {
+        input[0] = in[0 * 4];
+        input[1] = in[1 * 4];
+        input[2] = in[2 * 4];
+        input[3] = in[3 * 4];
+      }
+      // Transform.
+      step[0] = input[0] + input[3];
+      step[1] = input[1] + input[2];
+      step[2] = input[1] - input[2];
+      step[3] = input[0] - input[3];
+      temp1 = (step[0] + step[1]) * cospi_16_64;
+      temp2 = (step[0] - step[1]) * cospi_16_64;
+      out[0] = dct_const_round_shift(temp1);
+      out[2] = dct_const_round_shift(temp2);
+      temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
+      temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
+      out[1] = dct_const_round_shift(temp1);
+      out[3] = dct_const_round_shift(temp2);
+      // Do next column (which is a transposed row in second/horizontal pass)
+      in++;
+      out += 4;
+    }
+    // Setup in/out for next pass.
+    in = intermediate;
+    out = output;
+  }
+
+  {
+    int i, j;
+    for (i = 0; i < 4; ++i) {
+      for (j = 0; j < 4; ++j)
+        output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
+    }
+  }
+}
+
+static void fadst4_1d(int16_t *input, int16_t *output) {
+  int x0, x1, x2, x3;
+  int s0, s1, s2, s3, s4, s5, s6, s7;
+
+  x0 = input[0];
+  x1 = input[1];
+  x2 = input[2];
+  x3 = input[3];
+
+  if (!(x0 | x1 | x2 | x3)) {
+    output[0] = output[1] = output[2] = output[3] = 0;
+    return;
+  }
+
+  s0 = sinpi_1_9 * x0;
+  s1 = sinpi_4_9 * x0;
+  s2 = sinpi_2_9 * x1;
+  s3 = sinpi_1_9 * x1;
+  s4 = sinpi_3_9 * x2;
+  s5 = sinpi_4_9 * x3;
+  s6 = sinpi_2_9 * x3;
+  s7 = x0 + x1 - x3;
+
+  x0 = s0 + s2 + s5;
+  x1 = sinpi_3_9 * s7;
+  x2 = s1 - s3 + s6;
+  x3 = s4;
+
+  s0 = x0 + x3;
+  s1 = x1;
+  s2 = x2 - x3;
+  s3 = x2 - x0 + x3;
+
+  // 1-D transform scaling factor is sqrt(2).
+  output[0] = dct_const_round_shift(s0);
+  output[1] = dct_const_round_shift(s1);
+  output[2] = dct_const_round_shift(s2);
+  output[3] = dct_const_round_shift(s3);
+}
+
+static const transform_2d FHT_4[] = {
+  { fdct4_1d,  fdct4_1d  },  // DCT_DCT  = 0
+  { fadst4_1d, fdct4_1d  },  // ADST_DCT = 1
+  { fdct4_1d,  fadst4_1d },  // DCT_ADST = 2
+  { fadst4_1d, fadst4_1d }   // ADST_ADST = 3
+};
+
+void vp9_short_fht4x4_c(int16_t *input, int16_t *output,
+                        int pitch, TX_TYPE tx_type) {
+  int16_t out[4 * 4];
+  int16_t *outptr = &out[0];
+  int i, j;
+  int16_t temp_in[4], temp_out[4];
+  const transform_2d ht = FHT_4[tx_type];
+
+  // Columns
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = input[j * pitch + i] << 4;
+    if (i == 0 && temp_in[0])
+      temp_in[0] += 1;
+    ht.cols(temp_in, temp_out);
+    for (j = 0; j < 4; ++j)
+      outptr[j * 4 + i] = temp_out[j];
+  }
+
+  // Rows
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = out[j + i * 4];
+    ht.rows(temp_in, temp_out);
+    for (j = 0; j < 4; ++j)
+      output[j + i * 4] = (temp_out[j] + 1) >> 2;
+  }
+}
+
+void vp9_short_fdct8x4_c(int16_t *input, int16_t *output, int pitch) {
+    vp9_short_fdct4x4_c(input, output, pitch);
+    vp9_short_fdct4x4_c(input + 4, output + 16, pitch);
+}
+
+static void fdct8_1d(int16_t *input, int16_t *output) {
+  /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
+  /*needs32*/ int t0, t1, t2, t3;
+  /*canbe16*/ int x0, x1, x2, x3;
+
+  // stage 1
+  s0 = input[0] + input[7];
+  s1 = input[1] + input[6];
+  s2 = input[2] + input[5];
+  s3 = input[3] + input[4];
+  s4 = input[3] - input[4];
+  s5 = input[2] - input[5];
+  s6 = input[1] - input[6];
+  s7 = input[0] - input[7];
+
+  // fdct4_1d(step, step);
+  x0 = s0 + s3;
+  x1 = s1 + s2;
+  x2 = s1 - s2;
+  x3 = s0 - s3;
+  t0 = (x0 + x1) * cospi_16_64;
+  t1 = (x0 - x1) * cospi_16_64;
+  t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
+  t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
+  output[0] = dct_const_round_shift(t0);
+  output[2] = dct_const_round_shift(t2);
+  output[4] = dct_const_round_shift(t1);
+  output[6] = dct_const_round_shift(t3);
+
+  // Stage 2
+  t0 = (s6 - s5) * cospi_16_64;
+  t1 = (s6 + s5) * cospi_16_64;
+  t2 = dct_const_round_shift(t0);
+  t3 = dct_const_round_shift(t1);
+
+  // Stage 3
+  x0 = s4 + t2;
+  x1 = s4 - t2;
+  x2 = s7 - t3;
+  x3 = s7 + t3;
+
+  // Stage 4
+  t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
+  t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
+  t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
+  t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
+  output[1] = dct_const_round_shift(t0);
+  output[3] = dct_const_round_shift(t2);
+  output[5] = dct_const_round_shift(t1);
+  output[7] = dct_const_round_shift(t3);
+}
+
+void vp9_short_fdct8x8_c(int16_t *input, int16_t *final_output, int pitch) {
+  const int stride = pitch >> 1;
+  int i, j;
+  int16_t intermediate[64];
+
+  // Transform columns
+  {
+    int16_t *output = intermediate;
+    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
+    /*needs32*/ int t0, t1, t2, t3;
+    /*canbe16*/ int x0, x1, x2, x3;
+
+    int i;
+    for (i = 0; i < 8; i++) {
+      // stage 1
+      s0 = (input[0 * stride] + input[7 * stride]) << 2;
+      s1 = (input[1 * stride] + input[6 * stride]) << 2;
+      s2 = (input[2 * stride] + input[5 * stride]) << 2;
+      s3 = (input[3 * stride] + input[4 * stride]) << 2;
+      s4 = (input[3 * stride] - input[4 * stride]) << 2;
+      s5 = (input[2 * stride] - input[5 * stride]) << 2;
+      s6 = (input[1 * stride] - input[6 * stride]) << 2;
+      s7 = (input[0 * stride] - input[7 * stride]) << 2;
+
+      // fdct4_1d(step, step);
+      x0 = s0 + s3;
+      x1 = s1 + s2;
+      x2 = s1 - s2;
+      x3 = s0 - s3;
+      t0 = (x0 + x1) * cospi_16_64;
+      t1 = (x0 - x1) * cospi_16_64;
+      t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
+      t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
+      output[0 * 8] = dct_const_round_shift(t0);
+      output[2 * 8] = dct_const_round_shift(t2);
+      output[4 * 8] = dct_const_round_shift(t1);
+      output[6 * 8] = dct_const_round_shift(t3);
+
+      // Stage 2
+      t0 = (s6 - s5) * cospi_16_64;
+      t1 = (s6 + s5) * cospi_16_64;
+      t2 = dct_const_round_shift(t0);
+      t3 = dct_const_round_shift(t1);
+
+      // Stage 3
+      x0 = s4 + t2;
+      x1 = s4 - t2;
+      x2 = s7 - t3;
+      x3 = s7 + t3;
+
+      // Stage 4
+      t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
+      t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
+      t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
+      t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
+      output[1 * 8] = dct_const_round_shift(t0);
+      output[3 * 8] = dct_const_round_shift(t2);
+      output[5 * 8] = dct_const_round_shift(t1);
+      output[7 * 8] = dct_const_round_shift(t3);
+      input++;
+      output++;
+    }
+  }
+
+  // Rows
+  for (i = 0; i < 8; ++i) {
+    fdct8_1d(&intermediate[i * 8], &final_output[i * 8]);
+    for (j = 0; j < 8; ++j)
+      final_output[j + i * 8] /= 2;
+  }
+}
+
+void vp9_short_fdct16x16_c(int16_t *input, int16_t *output, int pitch) {
+  // The 2D transform is done with two passes which are actually pretty
+  // similar. In the first one, we transform the columns and transpose
+  // the results. In the second one, we transform the rows. To achieve that,
+  // as the first pass results are transposed, we tranpose the columns (that
+  // is the transposed rows) and transpose the results (so that it goes back
+  // in normal/row positions).
+  const int stride = pitch >> 1;
+  int pass;
+  // We need an intermediate buffer between passes.
+  int16_t intermediate[256];
+  int16_t *in = input;
+  int16_t *out = intermediate;
+  // Do the two transform/transpose passes
+  for (pass = 0; pass < 2; ++pass) {
+    /*canbe16*/ int step1[8];
+    /*canbe16*/ int step2[8];
+    /*canbe16*/ int step3[8];
+    /*canbe16*/ int input[8];
+    /*needs32*/ int temp1, temp2;
+    int i;
+    for (i = 0; i < 16; i++) {
+      if (0 == pass) {
+        // Calculate input for the first 8 results.
+        input[0] = (in[0 * stride] + in[15 * stride]) << 2;
+        input[1] = (in[1 * stride] + in[14 * stride]) << 2;
+        input[2] = (in[2 * stride] + in[13 * stride]) << 2;
+        input[3] = (in[3 * stride] + in[12 * stride]) << 2;
+        input[4] = (in[4 * stride] + in[11 * stride]) << 2;
+        input[5] = (in[5 * stride] + in[10 * stride]) << 2;
+        input[6] = (in[6 * stride] + in[ 9 * stride]) << 2;
+        input[7] = (in[7 * stride] + in[ 8 * stride]) << 2;
+        // Calculate input for the next 8 results.
+        step1[0] = (in[7 * stride] - in[ 8 * stride]) << 2;
+        step1[1] = (in[6 * stride] - in[ 9 * stride]) << 2;
+        step1[2] = (in[5 * stride] - in[10 * stride]) << 2;
+        step1[3] = (in[4 * stride] - in[11 * stride]) << 2;
+        step1[4] = (in[3 * stride] - in[12 * stride]) << 2;
+        step1[5] = (in[2 * stride] - in[13 * stride]) << 2;
+        step1[6] = (in[1 * stride] - in[14 * stride]) << 2;
+        step1[7] = (in[0 * stride] - in[15 * stride]) << 2;
+      } else {
+        // Calculate input for the first 8 results.
+        input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
+        input[1] = ((in[1 * 16] + 1) >> 2) + ((in[14 * 16] + 1) >> 2);
+        input[2] = ((in[2 * 16] + 1) >> 2) + ((in[13 * 16] + 1) >> 2);
+        input[3] = ((in[3 * 16] + 1) >> 2) + ((in[12 * 16] + 1) >> 2);
+        input[4] = ((in[4 * 16] + 1) >> 2) + ((in[11 * 16] + 1) >> 2);
+        input[5] = ((in[5 * 16] + 1) >> 2) + ((in[10 * 16] + 1) >> 2);
+        input[6] = ((in[6 * 16] + 1) >> 2) + ((in[ 9 * 16] + 1) >> 2);
+        input[7] = ((in[7 * 16] + 1) >> 2) + ((in[ 8 * 16] + 1) >> 2);
+        // Calculate input for the next 8 results.
+        step1[0] = ((in[7 * 16] + 1) >> 2) - ((in[ 8 * 16] + 1) >> 2);
+        step1[1] = ((in[6 * 16] + 1) >> 2) - ((in[ 9 * 16] + 1) >> 2);
+        step1[2] = ((in[5 * 16] + 1) >> 2) - ((in[10 * 16] + 1) >> 2);
+        step1[3] = ((in[4 * 16] + 1) >> 2) - ((in[11 * 16] + 1) >> 2);
+        step1[4] = ((in[3 * 16] + 1) >> 2) - ((in[12 * 16] + 1) >> 2);
+        step1[5] = ((in[2 * 16] + 1) >> 2) - ((in[13 * 16] + 1) >> 2);
+        step1[6] = ((in[1 * 16] + 1) >> 2) - ((in[14 * 16] + 1) >> 2);
+        step1[7] = ((in[0 * 16] + 1) >> 2) - ((in[15 * 16] + 1) >> 2);
+      }
+      // Work on the first eight values; fdct8_1d(input, even_results);
+      {
+        /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
+        /*needs32*/ int t0, t1, t2, t3;
+        /*canbe16*/ int x0, x1, x2, x3;
+
+        // stage 1
+        s0 = input[0] + input[7];
+        s1 = input[1] + input[6];
+        s2 = input[2] + input[5];
+        s3 = input[3] + input[4];
+        s4 = input[3] - input[4];
+        s5 = input[2] - input[5];
+        s6 = input[1] - input[6];
+        s7 = input[0] - input[7];
+
+        // fdct4_1d(step, step);
+        x0 = s0 + s3;
+        x1 = s1 + s2;
+        x2 = s1 - s2;
+        x3 = s0 - s3;
+        t0 = (x0 + x1) * cospi_16_64;
+        t1 = (x0 - x1) * cospi_16_64;
+        t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
+        t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
+        out[0] = dct_const_round_shift(t0);
+        out[4] = dct_const_round_shift(t2);
+        out[8] = dct_const_round_shift(t1);
+        out[12] = dct_const_round_shift(t3);
+
+        // Stage 2
+        t0 = (s6 - s5) * cospi_16_64;
+        t1 = (s6 + s5) * cospi_16_64;
+        t2 = dct_const_round_shift(t0);
+        t3 = dct_const_round_shift(t1);
+
+        // Stage 3
+        x0 = s4 + t2;
+        x1 = s4 - t2;
+        x2 = s7 - t3;
+        x3 = s7 + t3;
+
+        // Stage 4
+        t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
+        t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
+        t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
+        t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
+        out[2] = dct_const_round_shift(t0);
+        out[6] = dct_const_round_shift(t2);
+        out[10] = dct_const_round_shift(t1);
+        out[14] = dct_const_round_shift(t3);
+      }
+      // Work on the next eight values; step1 -> odd_results
+      {
+        // step 2
+        temp1 = (step1[5] - step1[2]) * cospi_16_64;
+        temp2 = (step1[4] - step1[3]) * cospi_16_64;
+        step2[2] = dct_const_round_shift(temp1);
+        step2[3] = dct_const_round_shift(temp2);
+        temp1 = (step1[4] + step1[3]) * cospi_16_64;
+        temp2 = (step1[5] + step1[2]) * cospi_16_64;
+        step2[4] = dct_const_round_shift(temp1);
+        step2[5] = dct_const_round_shift(temp2);
+        // step 3
+        step3[0] = step1[0] + step2[3];
+        step3[1] = step1[1] + step2[2];
+        step3[2] = step1[1] - step2[2];
+        step3[3] = step1[0] - step2[3];
+        step3[4] = step1[7] - step2[4];
+        step3[5] = step1[6] - step2[5];
+        step3[6] = step1[6] + step2[5];
+        step3[7] = step1[7] + step2[4];
+        // step 4
+        temp1 = step3[1] *  -cospi_8_64 + step3[6] * cospi_24_64;
+        temp2 = step3[2] * -cospi_24_64 - step3[5] *  cospi_8_64;
+        step2[1] = dct_const_round_shift(temp1);
+        step2[2] = dct_const_round_shift(temp2);
+        temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
+        temp2 = step3[1] * cospi_24_64 + step3[6] *  cospi_8_64;
+        step2[5] = dct_const_round_shift(temp1);
+        step2[6] = dct_const_round_shift(temp2);
+        // step 5
+        step1[0] = step3[0] + step2[1];
+        step1[1] = step3[0] - step2[1];
+        step1[2] = step3[3] - step2[2];
+        step1[3] = step3[3] + step2[2];
+        step1[4] = step3[4] + step2[5];
+        step1[5] = step3[4] - step2[5];
+        step1[6] = step3[7] - step2[6];
+        step1[7] = step3[7] + step2[6];
+        // step 6
+        temp1 = step1[0] * cospi_30_64 + step1[7] *  cospi_2_64;
+        temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
+        out[1] = dct_const_round_shift(temp1);
+        out[9] = dct_const_round_shift(temp2);
+        temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
+        temp2 = step1[3] *  cospi_6_64 + step1[4] * cospi_26_64;
+        out[5] = dct_const_round_shift(temp1);
+        out[13] = dct_const_round_shift(temp2);
+        temp1 = step1[3] * -cospi_26_64 + step1[4] *  cospi_6_64;
+        temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
+        out[3] = dct_const_round_shift(temp1);
+        out[11] = dct_const_round_shift(temp2);
+        temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
+        temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
+        out[7] = dct_const_round_shift(temp1);
+        out[15] = dct_const_round_shift(temp2);
+      }
+      // Do next column (which is a transposed row in second/horizontal pass)
+      in++;
+      out += 16;
+    }
+    // Setup in/out for next pass.
+    in = intermediate;
+    out = output;
+  }
+}
+
+static void fadst8_1d(int16_t *input, int16_t *output) {
+  int s0, s1, s2, s3, s4, s5, s6, s7;
+
+  int x0 = input[7];
+  int x1 = input[0];
+  int x2 = input[5];
+  int x3 = input[2];
+  int x4 = input[3];
+  int x5 = input[4];
+  int x6 = input[1];
+  int x7 = input[6];
+
+  // stage 1
+  s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
+  s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
+  s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
+  s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
+  s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
+  s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
+  s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
+  s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
+
+  x0 = dct_const_round_shift(s0 + s4);
+  x1 = dct_const_round_shift(s1 + s5);
+  x2 = dct_const_round_shift(s2 + s6);
+  x3 = dct_const_round_shift(s3 + s7);
+  x4 = dct_const_round_shift(s0 - s4);
+  x5 = dct_const_round_shift(s1 - s5);
+  x6 = dct_const_round_shift(s2 - s6);
+  x7 = dct_const_round_shift(s3 - s7);
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = cospi_8_64  * x4 + cospi_24_64 * x5;
+  s5 = cospi_24_64 * x4 - cospi_8_64  * x5;
+  s6 = - cospi_24_64 * x6 + cospi_8_64  * x7;
+  s7 =   cospi_8_64  * x6 + cospi_24_64 * x7;
+
+  x0 = s0 + s2;
+  x1 = s1 + s3;
+  x2 = s0 - s2;
+  x3 = s1 - s3;
+  x4 = dct_const_round_shift(s4 + s6);
+  x5 = dct_const_round_shift(s5 + s7);
+  x6 = dct_const_round_shift(s4 - s6);
+  x7 = dct_const_round_shift(s5 - s7);
+
+  // stage 3
+  s2 = cospi_16_64 * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (x6 - x7);
+
+  x2 = dct_const_round_shift(s2);
+  x3 = dct_const_round_shift(s3);
+  x6 = dct_const_round_shift(s6);
+  x7 = dct_const_round_shift(s7);
+
+  output[0] =   x0;
+  output[1] = - x4;
+  output[2] =   x6;
+  output[3] = - x2;
+  output[4] =   x3;
+  output[5] = - x7;
+  output[6] =   x5;
+  output[7] = - x1;
+}
+
+static const transform_2d FHT_8[] = {
+  { fdct8_1d,  fdct8_1d  },  // DCT_DCT  = 0
+  { fadst8_1d, fdct8_1d  },  // ADST_DCT = 1
+  { fdct8_1d,  fadst8_1d },  // DCT_ADST = 2
+  { fadst8_1d, fadst8_1d }   // ADST_ADST = 3
+};
+
+void vp9_short_fht8x8_c(int16_t *input, int16_t *output,
+                        int pitch, TX_TYPE tx_type) {
+  int16_t out[64];
+  int16_t *outptr = &out[0];
+  int i, j;
+  int16_t temp_in[8], temp_out[8];
+  const transform_2d ht = FHT_8[tx_type];
+
+  // Columns
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = input[j * pitch + i] << 2;
+    ht.cols(temp_in, temp_out);
+    for (j = 0; j < 8; ++j)
+      outptr[j * 8 + i] = temp_out[j];
+  }
+
+  // Rows
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j + i * 8];
+    ht.rows(temp_in, temp_out);
+    for (j = 0; j < 8; ++j)
+      output[j + i * 8] = temp_out[j] >> 1;
+  }
+}
+
+/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
+   pixel. */
+void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
+  int i;
+  int a1, b1, c1, d1, e1;
+  short *ip = input;
+  short *op = output;
+  int pitch_short = pitch >> 1;
+
+  for (i = 0; i < 4; i++) {
+    a1 = ip[0 * pitch_short];
+    b1 = ip[1 * pitch_short];
+    c1 = ip[2 * pitch_short];
+    d1 = ip[3 * pitch_short];
+
+    a1 += b1;
+    d1 = d1 - c1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= c1;
+    d1 += b1;
+    op[0] = a1;
+    op[4] = c1;
+    op[8] = d1;
+    op[12] = b1;
+
+    ip++;
+    op++;
+  }
+  ip = output;
+  op = output;
+
+  for (i = 0; i < 4; i++) {
+    a1 = ip[0];
+    b1 = ip[1];
+    c1 = ip[2];
+    d1 = ip[3];
+
+    a1 += b1;
+    d1 -= c1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= c1;
+    d1 += b1;
+    op[0] = a1 << WHT_UPSCALE_FACTOR;
+    op[1] = c1 << WHT_UPSCALE_FACTOR;
+    op[2] = d1 << WHT_UPSCALE_FACTOR;
+    op[3] = b1 << WHT_UPSCALE_FACTOR;
+
+    ip += 4;
+    op += 4;
+  }
+}
+
+void vp9_short_walsh8x4_c(short *input, short *output, int pitch) {
+  vp9_short_walsh4x4_c(input,   output,    pitch);
+  vp9_short_walsh4x4_c(input + 4, output + 16, pitch);
+}
+
+
+// Rewrote to use same algorithm as others.
+static void fdct16_1d(int16_t in[16], int16_t out[16]) {
+  /*canbe16*/ int step1[8];
+  /*canbe16*/ int step2[8];
+  /*canbe16*/ int step3[8];
+  /*canbe16*/ int input[8];
+  /*needs32*/ int temp1, temp2;
+
+  // step 1
+  input[0] = in[0] + in[15];
+  input[1] = in[1] + in[14];
+  input[2] = in[2] + in[13];
+  input[3] = in[3] + in[12];
+  input[4] = in[4] + in[11];
+  input[5] = in[5] + in[10];
+  input[6] = in[6] + in[ 9];
+  input[7] = in[7] + in[ 8];
+
+  step1[0] = in[7] - in[ 8];
+  step1[1] = in[6] - in[ 9];
+  step1[2] = in[5] - in[10];
+  step1[3] = in[4] - in[11];
+  step1[4] = in[3] - in[12];
+  step1[5] = in[2] - in[13];
+  step1[6] = in[1] - in[14];
+  step1[7] = in[0] - in[15];
+
+  // fdct8_1d(step, step);
+  {
+    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
+    /*needs32*/ int t0, t1, t2, t3;
+    /*canbe16*/ int x0, x1, x2, x3;
+
+    // stage 1
+    s0 = input[0] + input[7];
+    s1 = input[1] + input[6];
+    s2 = input[2] + input[5];
+    s3 = input[3] + input[4];
+    s4 = input[3] - input[4];
+    s5 = input[2] - input[5];
+    s6 = input[1] - input[6];
+    s7 = input[0] - input[7];
+
+    // fdct4_1d(step, step);
+    x0 = s0 + s3;
+    x1 = s1 + s2;
+    x2 = s1 - s2;
+    x3 = s0 - s3;
+    t0 = (x0 + x1) * cospi_16_64;
+    t1 = (x0 - x1) * cospi_16_64;
+    t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
+    t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
+    out[0] = dct_const_round_shift(t0);
+    out[4] = dct_const_round_shift(t2);
+    out[8] = dct_const_round_shift(t1);
+    out[12] = dct_const_round_shift(t3);
+
+    // Stage 2
+    t0 = (s6 - s5) * cospi_16_64;
+    t1 = (s6 + s5) * cospi_16_64;
+    t2 = dct_const_round_shift(t0);
+    t3 = dct_const_round_shift(t1);
+
+    // Stage 3
+    x0 = s4 + t2;
+    x1 = s4 - t2;
+    x2 = s7 - t3;
+    x3 = s7 + t3;
+
+    // Stage 4
+    t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
+    t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
+    t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
+    t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
+    out[2] = dct_const_round_shift(t0);
+    out[6] = dct_const_round_shift(t2);
+    out[10] = dct_const_round_shift(t1);
+    out[14] = dct_const_round_shift(t3);
+  }
+
+  // step 2
+  temp1 = (step1[5] - step1[2]) * cospi_16_64;
+  temp2 = (step1[4] - step1[3]) * cospi_16_64;
+  step2[2] = dct_const_round_shift(temp1);
+  step2[3] = dct_const_round_shift(temp2);
+  temp1 = (step1[4] + step1[3]) * cospi_16_64;
+  temp2 = (step1[5] + step1[2]) * cospi_16_64;
+  step2[4] = dct_const_round_shift(temp1);
+  step2[5] = dct_const_round_shift(temp2);
+
+  // step 3
+  step3[0] = step1[0] + step2[3];
+  step3[1] = step1[1] + step2[2];
+  step3[2] = step1[1] - step2[2];
+  step3[3] = step1[0] - step2[3];
+  step3[4] = step1[7] - step2[4];
+  step3[5] = step1[6] - step2[5];
+  step3[6] = step1[6] + step2[5];
+  step3[7] = step1[7] + step2[4];
+
+  // step 4
+  temp1 = step3[1] *  -cospi_8_64 + step3[6] * cospi_24_64;
+  temp2 = step3[2] * -cospi_24_64 - step3[5] *  cospi_8_64;
+  step2[1] = dct_const_round_shift(temp1);
+  step2[2] = dct_const_round_shift(temp2);
+  temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
+  temp2 = step3[1] * cospi_24_64 + step3[6] *  cospi_8_64;
+  step2[5] = dct_const_round_shift(temp1);
+  step2[6] = dct_const_round_shift(temp2);
+
+  // step 5
+  step1[0] = step3[0] + step2[1];
+  step1[1] = step3[0] - step2[1];
+  step1[2] = step3[3] - step2[2];
+  step1[3] = step3[3] + step2[2];
+  step1[4] = step3[4] + step2[5];
+  step1[5] = step3[4] - step2[5];
+  step1[6] = step3[7] - step2[6];
+  step1[7] = step3[7] + step2[6];
+
+  // step 6
+  temp1 = step1[0] * cospi_30_64 + step1[7] *  cospi_2_64;
+  temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
+  out[1] = dct_const_round_shift(temp1);
+  out[9] = dct_const_round_shift(temp2);
+
+  temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
+  temp2 = step1[3] *  cospi_6_64 + step1[4] * cospi_26_64;
+  out[5] = dct_const_round_shift(temp1);
+  out[13] = dct_const_round_shift(temp2);
+
+  temp1 = step1[3] * -cospi_26_64 + step1[4] *  cospi_6_64;
+  temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
+  out[3] = dct_const_round_shift(temp1);
+  out[11] = dct_const_round_shift(temp2);
+
+  temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
+  temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
+  out[7] = dct_const_round_shift(temp1);
+  out[15] = dct_const_round_shift(temp2);
+}
+
+void fadst16_1d(int16_t *input, int16_t *output) {
+  int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
+
+  int x0 = input[15];
+  int x1 = input[0];
+  int x2 = input[13];
+  int x3 = input[2];
+  int x4 = input[11];
+  int x5 = input[4];
+  int x6 = input[9];
+  int x7 = input[6];
+  int x8 = input[7];
+  int x9 = input[8];
+  int x10 = input[5];
+  int x11 = input[10];
+  int x12 = input[3];
+  int x13 = input[12];
+  int x14 = input[1];
+  int x15 = input[14];
+
+  // stage 1
+  s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
+  s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
+  s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
+  s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
+  s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
+  s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
+  s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
+  s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
+  s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
+  s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
+  s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
+  s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
+  s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
+  s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
+  s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
+  s15 = x14 * cospi_3_64  - x15 * cospi_29_64;
+
+  x0 = dct_const_round_shift(s0 + s8);
+  x1 = dct_const_round_shift(s1 + s9);
+  x2 = dct_const_round_shift(s2 + s10);
+  x3 = dct_const_round_shift(s3 + s11);
+  x4 = dct_const_round_shift(s4 + s12);
+  x5 = dct_const_round_shift(s5 + s13);
+  x6 = dct_const_round_shift(s6 + s14);
+  x7 = dct_const_round_shift(s7 + s15);
+  x8  = dct_const_round_shift(s0 - s8);
+  x9  = dct_const_round_shift(s1 - s9);
+  x10 = dct_const_round_shift(s2 - s10);
+  x11 = dct_const_round_shift(s3 - s11);
+  x12 = dct_const_round_shift(s4 - s12);
+  x13 = dct_const_round_shift(s5 - s13);
+  x14 = dct_const_round_shift(s6 - s14);
+  x15 = dct_const_round_shift(s7 - s15);
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4;
+  s5 = x5;
+  s6 = x6;
+  s7 = x7;
+  s8 =    x8 * cospi_4_64   + x9 * cospi_28_64;
+  s9 =    x8 * cospi_28_64  - x9 * cospi_4_64;
+  s10 =   x10 * cospi_20_64 + x11 * cospi_12_64;
+  s11 =   x10 * cospi_12_64 - x11 * cospi_20_64;
+  s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
+  s13 =   x12 * cospi_4_64  + x13 * cospi_28_64;
+  s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
+  s15 =   x14 * cospi_20_64 + x15 * cospi_12_64;
+
+  x0 = s0 + s4;
+  x1 = s1 + s5;
+  x2 = s2 + s6;
+  x3 = s3 + s7;
+  x4 = s0 - s4;
+  x5 = s1 - s5;
+  x6 = s2 - s6;
+  x7 = s3 - s7;
+  x8 = dct_const_round_shift(s8 + s12);
+  x9 = dct_const_round_shift(s9 + s13);
+  x10 = dct_const_round_shift(s10 + s14);
+  x11 = dct_const_round_shift(s11 + s15);
+  x12 = dct_const_round_shift(s8 - s12);
+  x13 = dct_const_round_shift(s9 - s13);
+  x14 = dct_const_round_shift(s10 - s14);
+  x15 = dct_const_round_shift(s11 - s15);
+
+  // stage 3
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4 * cospi_8_64  + x5 * cospi_24_64;
+  s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
+  s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
+  s7 =   x6 * cospi_8_64  + x7 * cospi_24_64;
+  s8 = x8;
+  s9 = x9;
+  s10 = x10;
+  s11 = x11;
+  s12 = x12 * cospi_8_64  + x13 * cospi_24_64;
+  s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
+  s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
+  s15 =   x14 * cospi_8_64  + x15 * cospi_24_64;
+
+  x0 = s0 + s2;
+  x1 = s1 + s3;
+  x2 = s0 - s2;
+  x3 = s1 - s3;
+  x4 = dct_const_round_shift(s4 + s6);
+  x5 = dct_const_round_shift(s5 + s7);
+  x6 = dct_const_round_shift(s4 - s6);
+  x7 = dct_const_round_shift(s5 - s7);
+  x8 = s8 + s10;
+  x9 = s9 + s11;
+  x10 = s8 - s10;
+  x11 = s9 - s11;
+  x12 = dct_const_round_shift(s12 + s14);
+  x13 = dct_const_round_shift(s13 + s15);
+  x14 = dct_const_round_shift(s12 - s14);
+  x15 = dct_const_round_shift(s13 - s15);
+
+  // stage 4
+  s2 = (- cospi_16_64) * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (- x6 + x7);
+  s10 = cospi_16_64 * (x10 + x11);
+  s11 = cospi_16_64 * (- x10 + x11);
+  s14 = (- cospi_16_64) * (x14 + x15);
+  s15 = cospi_16_64 * (x14 - x15);
+
+  x2 = dct_const_round_shift(s2);
+  x3 = dct_const_round_shift(s3);
+  x6 = dct_const_round_shift(s6);
+  x7 = dct_const_round_shift(s7);
+  x10 = dct_const_round_shift(s10);
+  x11 = dct_const_round_shift(s11);
+  x14 = dct_const_round_shift(s14);
+  x15 = dct_const_round_shift(s15);
+
+  output[0] = x0;
+  output[1] = - x8;
+  output[2] = x12;
+  output[3] = - x4;
+  output[4] = x6;
+  output[5] = x14;
+  output[6] = x10;
+  output[7] = x2;
+  output[8] = x3;
+  output[9] =  x11;
+  output[10] = x15;
+  output[11] = x7;
+  output[12] = x5;
+  output[13] = - x13;
+  output[14] = x9;
+  output[15] = - x1;
+}
+
+static const transform_2d FHT_16[] = {
+  { fdct16_1d,  fdct16_1d  },  // DCT_DCT  = 0
+  { fadst16_1d, fdct16_1d  },  // ADST_DCT = 1
+  { fdct16_1d,  fadst16_1d },  // DCT_ADST = 2
+  { fadst16_1d, fadst16_1d }   // ADST_ADST = 3
+};
+
+void vp9_short_fht16x16_c(int16_t *input, int16_t *output,
+                          int pitch, TX_TYPE tx_type) {
+  int16_t out[256];
+  int16_t *outptr = &out[0];
+  int i, j;
+  int16_t temp_in[16], temp_out[16];
+  const transform_2d ht = FHT_16[tx_type];
+
+  // Columns
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = input[j * pitch + i] << 2;
+    ht.cols(temp_in, temp_out);
+    for (j = 0; j < 16; ++j)
+      outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
+  }
+
+  // Rows
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j + i * 16];
+    ht.rows(temp_in, temp_out);
+    for (j = 0; j < 16; ++j)
+      output[j + i * 16] = temp_out[j];
+  }
+}
+
+static INLINE int dct_32_round(int input) {
+  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  assert(-131072 <= rv && rv <= 131071);
+  return rv;
+}
+
+static INLINE int half_round_shift(int input) {
+  int rv = (input + 1 + (input < 0)) >> 2;
+  return rv;
+}
+
+static void dct32_1d(int *input, int *output, int round) {
+  int step[32];
+  // Stage 1
+  step[0] = input[0] + input[(32 - 1)];
+  step[1] = input[1] + input[(32 - 2)];
+  step[2] = input[2] + input[(32 - 3)];
+  step[3] = input[3] + input[(32 - 4)];
+  step[4] = input[4] + input[(32 - 5)];
+  step[5] = input[5] + input[(32 - 6)];
+  step[6] = input[6] + input[(32 - 7)];
+  step[7] = input[7] + input[(32 - 8)];
+  step[8] = input[8] + input[(32 - 9)];
+  step[9] = input[9] + input[(32 - 10)];
+  step[10] = input[10] + input[(32 - 11)];
+  step[11] = input[11] + input[(32 - 12)];
+  step[12] = input[12] + input[(32 - 13)];
+  step[13] = input[13] + input[(32 - 14)];
+  step[14] = input[14] + input[(32 - 15)];
+  step[15] = input[15] + input[(32 - 16)];
+  step[16] = -input[16] + input[(32 - 17)];
+  step[17] = -input[17] + input[(32 - 18)];
+  step[18] = -input[18] + input[(32 - 19)];
+  step[19] = -input[19] + input[(32 - 20)];
+  step[20] = -input[20] + input[(32 - 21)];
+  step[21] = -input[21] + input[(32 - 22)];
+  step[22] = -input[22] + input[(32 - 23)];
+  step[23] = -input[23] + input[(32 - 24)];
+  step[24] = -input[24] + input[(32 - 25)];
+  step[25] = -input[25] + input[(32 - 26)];
+  step[26] = -input[26] + input[(32 - 27)];
+  step[27] = -input[27] + input[(32 - 28)];
+  step[28] = -input[28] + input[(32 - 29)];
+  step[29] = -input[29] + input[(32 - 30)];
+  step[30] = -input[30] + input[(32 - 31)];
+  step[31] = -input[31] + input[(32 - 32)];
+
+  // Stage 2
+  output[0] = step[0] + step[16 - 1];
+  output[1] = step[1] + step[16 - 2];
+  output[2] = step[2] + step[16 - 3];
+  output[3] = step[3] + step[16 - 4];
+  output[4] = step[4] + step[16 - 5];
+  output[5] = step[5] + step[16 - 6];
+  output[6] = step[6] + step[16 - 7];
+  output[7] = step[7] + step[16 - 8];
+  output[8] = -step[8] + step[16 - 9];
+  output[9] = -step[9] + step[16 - 10];
+  output[10] = -step[10] + step[16 - 11];
+  output[11] = -step[11] + step[16 - 12];
+  output[12] = -step[12] + step[16 - 13];
+  output[13] = -step[13] + step[16 - 14];
+  output[14] = -step[14] + step[16 - 15];
+  output[15] = -step[15] + step[16 - 16];
+
+  output[16] = step[16];
+  output[17] = step[17];
+  output[18] = step[18];
+  output[19] = step[19];
+
+  output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
+  output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
+  output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
+  output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
+
+  output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
+  output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
+  output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
+  output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
+
+  output[28] = step[28];
+  output[29] = step[29];
+  output[30] = step[30];
+  output[31] = step[31];
+
+  // Stage 3
+  step[0] = output[0] + output[(8 - 1)];
+  step[1] = output[1] + output[(8 - 2)];
+  step[2] = output[2] + output[(8 - 3)];
+  step[3] = output[3] + output[(8 - 4)];
+  step[4] = -output[4] + output[(8 - 5)];
+  step[5] = -output[5] + output[(8 - 6)];
+  step[6] = -output[6] + output[(8 - 7)];
+  step[7] = -output[7] + output[(8 - 8)];
+  step[8] = output[8];
+  step[9] = output[9];
+  step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
+  step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
+  step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
+  step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
+  step[14] = output[14];
+  step[15] = output[15];
+
+  step[16] = output[16] + output[23];
+  step[17] = output[17] + output[22];
+  step[18] = output[18] + output[21];
+  step[19] = output[19] + output[20];
+  step[20] = -output[20] + output[19];
+  step[21] = -output[21] + output[18];
+  step[22] = -output[22] + output[17];
+  step[23] = -output[23] + output[16];
+  step[24] = -output[24] + output[31];
+  step[25] = -output[25] + output[30];
+  step[26] = -output[26] + output[29];
+  step[27] = -output[27] + output[28];
+  step[28] = output[28] + output[27];
+  step[29] = output[29] + output[26];
+  step[30] = output[30] + output[25];
+  step[31] = output[31] + output[24];
+
+  // dump the magnitude by half, hence the intermediate values are within 1108
+  // the range of 16 bits.
+  if (round) {
+    step[0] = half_round_shift(step[0]);
+    step[1] = half_round_shift(step[1]);
+    step[2] = half_round_shift(step[2]);
+    step[3] = half_round_shift(step[3]);
+    step[4] = half_round_shift(step[4]);
+    step[5] = half_round_shift(step[5]);
+    step[6] = half_round_shift(step[6]);
+    step[7] = half_round_shift(step[7]);
+    step[8] = half_round_shift(step[8]);
+    step[9] = half_round_shift(step[9]);
+    step[10] = half_round_shift(step[10]);
+    step[11] = half_round_shift(step[11]);
+    step[12] = half_round_shift(step[12]);
+    step[13] = half_round_shift(step[13]);
+    step[14] = half_round_shift(step[14]);
+    step[15] = half_round_shift(step[15]);
+
+    step[16] = half_round_shift(step[16]);
+    step[17] = half_round_shift(step[17]);
+    step[18] = half_round_shift(step[18]);
+    step[19] = half_round_shift(step[19]);
+    step[20] = half_round_shift(step[20]);
+    step[21] = half_round_shift(step[21]);
+    step[22] = half_round_shift(step[22]);
+    step[23] = half_round_shift(step[23]);
+    step[24] = half_round_shift(step[24]);
+    step[25] = half_round_shift(step[25]);
+    step[26] = half_round_shift(step[26]);
+    step[27] = half_round_shift(step[27]);
+    step[28] = half_round_shift(step[28]);
+    step[29] = half_round_shift(step[29]);
+    step[30] = half_round_shift(step[30]);
+    step[31] = half_round_shift(step[31]);
+  }
+
+  // Stage 4
+  output[0] = step[0] + step[3];
+  output[1] = step[1] + step[2];
+  output[2] = -step[2] + step[1];
+  output[3] = -step[3] + step[0];
+  output[4] = step[4];
+  output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
+  output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
+  output[7] = step[7];
+  output[8] = step[8] + step[11];
+  output[9] = step[9] + step[10];
+  output[10] = -step[10] + step[9];
+  output[11] = -step[11] + step[8];
+  output[12] = -step[12] + step[15];
+  output[13] = -step[13] + step[14];
+  output[14] = step[14] + step[13];
+  output[15] = step[15] + step[12];
+
+  output[16] = step[16];
+  output[17] = step[17];
+  output[18] = dct_32_round(step[18] * -cospi_8_64 + step[29] * cospi_24_64);
+  output[19] = dct_32_round(step[19] * -cospi_8_64 + step[28] * cospi_24_64);
+  output[20] = dct_32_round(step[20] * -cospi_24_64 + step[27] * -cospi_8_64);
+  output[21] = dct_32_round(step[21] * -cospi_24_64 + step[26] * -cospi_8_64);
+  output[22] = step[22];
+  output[23] = step[23];
+  output[24] = step[24];
+  output[25] = step[25];
+  output[26] = dct_32_round(step[26] * cospi_24_64 + step[21] * -cospi_8_64);
+  output[27] = dct_32_round(step[27] * cospi_24_64 + step[20] * -cospi_8_64);
+  output[28] = dct_32_round(step[28] * cospi_8_64 + step[19] * cospi_24_64);
+  output[29] = dct_32_round(step[29] * cospi_8_64 + step[18] * cospi_24_64);
+  output[30] = step[30];
+  output[31] = step[31];
+
+  // Stage 5
+  step[0] = dct_32_round((output[0] + output[1]) * cospi_16_64);
+  step[1] = dct_32_round((-output[1] + output[0]) * cospi_16_64);
+  step[2] = dct_32_round(output[2] * cospi_24_64 + output[3] * cospi_8_64);
+  step[3] = dct_32_round(output[3] * cospi_24_64 - output[2] * cospi_8_64);
+  step[4] = output[4] + output[5];
+  step[5] = -output[5] + output[4];
+  step[6] = -output[6] + output[7];
+  step[7] = output[7] + output[6];
+  step[8] = output[8];
+  step[9] = dct_32_round(output[9] * -cospi_8_64 + output[14] * cospi_24_64);
+  step[10] = dct_32_round(output[10] * -cospi_24_64 + output[13] * -cospi_8_64);
+  step[11] = output[11];
+  step[12] = output[12];
+  step[13] = dct_32_round(output[13] * cospi_24_64 + output[10] * -cospi_8_64);
+  step[14] = dct_32_round(output[14] * cospi_8_64 + output[9] * cospi_24_64);
+  step[15] = output[15];
+
+  step[16] = output[16] + output[19];
+  step[17] = output[17] + output[18];
+  step[18] = -output[18] + output[17];
+  step[19] = -output[19] + output[16];
+  step[20] = -output[20] + output[23];
+  step[21] = -output[21] + output[22];
+  step[22] = output[22] + output[21];
+  step[23] = output[23] + output[20];
+  step[24] = output[24] + output[27];
+  step[25] = output[25] + output[26];
+  step[26] = -output[26] + output[25];
+  step[27] = -output[27] + output[24];
+  step[28] = -output[28] + output[31];
+  step[29] = -output[29] + output[30];
+  step[30] = output[30] + output[29];
+  step[31] = output[31] + output[28];
+
+  // Stage 6
+  output[0] = step[0];
+  output[1] = step[1];
+  output[2] = step[2];
+  output[3] = step[3];
+  output[4] = dct_32_round(step[4] * cospi_28_64 + step[7] * cospi_4_64);
+  output[5] = dct_32_round(step[5] * cospi_12_64 + step[6] * cospi_20_64);
+  output[6] = dct_32_round(step[6] * cospi_12_64 + step[5] * -cospi_20_64);
+  output[7] = dct_32_round(step[7] * cospi_28_64 + step[4] * -cospi_4_64);
+  output[8] = step[8] + step[9];
+  output[9] = -step[9] + step[8];
+  output[10] = -step[10] + step[11];
+  output[11] = step[11] + step[10];
+  output[12] = step[12] + step[13];
+  output[13] = -step[13] + step[12];
+  output[14] = -step[14] + step[15];
+  output[15] = step[15] + step[14];
+
+  output[16] = step[16];
+  output[17] = dct_32_round(step[17] * -cospi_4_64 + step[30] * cospi_28_64);
+  output[18] = dct_32_round(step[18] * -cospi_28_64 + step[29] * -cospi_4_64);
+  output[19] = step[19];
+  output[20] = step[20];
+  output[21] = dct_32_round(step[21] * -cospi_20_64 + step[26] * cospi_12_64);
+  output[22] = dct_32_round(step[22] * -cospi_12_64 + step[25] * -cospi_20_64);
+  output[23] = step[23];
+  output[24] = step[24];
+  output[25] = dct_32_round(step[25] * cospi_12_64 + step[22] * -cospi_20_64);
+  output[26] = dct_32_round(step[26] * cospi_20_64 + step[21] * cospi_12_64);
+  output[27] = step[27];
+  output[28] = step[28];
+  output[29] = dct_32_round(step[29] * cospi_28_64 + step[18] * -cospi_4_64);
+  output[30] = dct_32_round(step[30] * cospi_4_64 + step[17] * cospi_28_64);
+  output[31] = step[31];
+
+  // Stage 7
+  step[0] = output[0];
+  step[1] = output[1];
+  step[2] = output[2];
+  step[3] = output[3];
+  step[4] = output[4];
+  step[5] = output[5];
+  step[6] = output[6];
+  step[7] = output[7];
+  step[8] = dct_32_round(output[8] * cospi_30_64 + output[15] * cospi_2_64);
+  step[9] = dct_32_round(output[9] * cospi_14_64 + output[14] * cospi_18_64);
+  step[10] = dct_32_round(output[10] * cospi_22_64 + output[13] * cospi_10_64);
+  step[11] = dct_32_round(output[11] * cospi_6_64 + output[12] * cospi_26_64);
+  step[12] = dct_32_round(output[12] * cospi_6_64 + output[11] * -cospi_26_64);
+  step[13] = dct_32_round(output[13] * cospi_22_64 + output[10] * -cospi_10_64);
+  step[14] = dct_32_round(output[14] * cospi_14_64 + output[9] * -cospi_18_64);
+  step[15] = dct_32_round(output[15] * cospi_30_64 + output[8] * -cospi_2_64);
+
+  step[16] = output[16] + output[17];
+  step[17] = -output[17] + output[16];
+  step[18] = -output[18] + output[19];
+  step[19] = output[19] + output[18];
+  step[20] = output[20] + output[21];
+  step[21] = -output[21] + output[20];
+  step[22] = -output[22] + output[23];
+  step[23] = output[23] + output[22];
+  step[24] = output[24] + output[25];
+  step[25] = -output[25] + output[24];
+  step[26] = -output[26] + output[27];
+  step[27] = output[27] + output[26];
+  step[28] = output[28] + output[29];
+  step[29] = -output[29] + output[28];
+  step[30] = -output[30] + output[31];
+  step[31] = output[31] + output[30];
+
+  // Final stage --- outputs indices are bit-reversed.
+  output[0]  = step[0];
+  output[16] = step[1];
+  output[8]  = step[2];
+  output[24] = step[3];
+  output[4]  = step[4];
+  output[20] = step[5];
+  output[12] = step[6];
+  output[28] = step[7];
+  output[2]  = step[8];
+  output[18] = step[9];
+  output[10] = step[10];
+  output[26] = step[11];
+  output[6]  = step[12];
+  output[22] = step[13];
+  output[14] = step[14];
+  output[30] = step[15];
+
+  output[1]  = dct_32_round(step[16] * cospi_31_64 + step[31] * cospi_1_64);
+  output[17] = dct_32_round(step[17] * cospi_15_64 + step[30] * cospi_17_64);
+  output[9]  = dct_32_round(step[18] * cospi_23_64 + step[29] * cospi_9_64);
+  output[25] = dct_32_round(step[19] * cospi_7_64 + step[28] * cospi_25_64);
+  output[5]  = dct_32_round(step[20] * cospi_27_64 + step[27] * cospi_5_64);
+  output[21] = dct_32_round(step[21] * cospi_11_64 + step[26] * cospi_21_64);
+  output[13] = dct_32_round(step[22] * cospi_19_64 + step[25] * cospi_13_64);
+  output[29] = dct_32_round(step[23] * cospi_3_64 + step[24] * cospi_29_64);
+  output[3]  = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
+  output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
+  output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
+  output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
+  output[7]  = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
+  output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
+  output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
+  output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
+}
+
+void vp9_short_fdct32x32_c(int16_t *input, int16_t *out, int pitch) {
+  int shortpitch = pitch >> 1;
+  int i, j;
+  int output[32 * 32];
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    int temp_in[32], temp_out[32];
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = input[j * shortpitch + i] << 2;
+    dct32_1d(temp_in, temp_out, 0);
+    for (j = 0; j < 32; ++j)
+      output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
+  }
+
+  // Rows
+  for (i = 0; i < 32; ++i) {
+    int temp_in[32], temp_out[32];
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = output[j + i * 32];
+    dct32_1d(temp_in, temp_out, 0);
+    for (j = 0; j < 32; ++j)
+      out[j + i * 32] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
+  }
+}
+
+// Note that although we use dct_32_round in dct32_1d computation flow,
+// this 2d fdct32x32 for rate-distortion optimization loop is operating
+// within 16 bits precision.
+void vp9_short_fdct32x32_rd_c(int16_t *input, int16_t *out, int pitch) {
+  int shortpitch = pitch >> 1;
+  int i, j;
+  int output[32 * 32];
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    int temp_in[32], temp_out[32];
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = input[j * shortpitch + i] << 2;
+    dct32_1d(temp_in, temp_out, 0);
+    for (j = 0; j < 32; ++j)
+      output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
+  }
+
+  // Rows
+  for (i = 0; i < 32; ++i) {
+    int temp_in[32], temp_out[32];
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = output[j + i * 32];
+    dct32_1d(temp_in, temp_out, 1);
+    for (j = 0; j < 32; ++j)
+      out[j + i * 32] = temp_out[j];
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_encodeframe.c b/libvpx/vp9/encoder/vp9_encodeframe.c
new file mode 100644
index 0000000..54b6e24
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodeframe.c
@@ -0,0 +1,2109 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/encoder/vp9_encodeframe.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/common/vp9_extend.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/encoder/vp9_segmentation.h"
+#include "vp9/encoder/vp9_encodeintra.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_tile_common.h"
+#include "vp9/encoder/vp9_tokenize.h"
+#include "./vp9_rtcd.h"
+#include <stdio.h>
+#include <math.h>
+#include <limits.h>
+#include "vpx_ports/vpx_timer.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_mvref_common.h"
+
+#define DBG_PRNT_SEGMAP 0
+
+// #define ENC_DEBUG
+#ifdef ENC_DEBUG
+int enc_debug = 0;
+#endif
+
+void vp9_select_interp_filter_type(VP9_COMP *cpi);
+
+static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
+                              int output_enabled, int mi_row, int mi_col,
+                              BLOCK_SIZE_TYPE bsize);
+
+static void adjust_act_zbin(VP9_COMP *cpi, MACROBLOCK *x);
+
+/* activity_avg must be positive, or flat regions could get a zero weight
+ *  (infinite lambda), which confounds analysis.
+ * This also avoids the need for divide by zero checks in
+ *  vp9_activity_masking().
+ */
+#define VP9_ACTIVITY_AVG_MIN (64)
+
+/* This is used as a reference when computing the source variance for the
+ *  purposes of activity masking.
+ * Eventually this should be replaced by custom no-reference routines,
+ *  which will be faster.
+ */
+static const uint8_t VP9_VAR_OFFS[16] = {
+  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128
+};
+
+
+// Original activity measure from Tim T's code.
+static unsigned int tt_activity_measure(VP9_COMP *cpi, MACROBLOCK *x) {
+  unsigned int act;
+  unsigned int sse;
+  /* TODO: This could also be done over smaller areas (8x8), but that would
+   *  require extensive changes elsewhere, as lambda is assumed to be fixed
+   *  over an entire MB in most of the code.
+   * Another option is to compute four 8x8 variances, and pick a single
+   *  lambda using a non-linear combination (e.g., the smallest, or second
+   *  smallest, etc.).
+   */
+  act = vp9_variance16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                          VP9_VAR_OFFS, 0, &sse);
+  act <<= 4;
+
+  /* If the region is flat, lower the activity some more. */
+  if (act < 8 << 12)
+    act = act < 5 << 12 ? act : 5 << 12;
+
+  return act;
+}
+
+// Stub for alternative experimental activity measures.
+static unsigned int alt_activity_measure(VP9_COMP *cpi,
+                                         MACROBLOCK *x, int use_dc_pred) {
+  return vp9_encode_intra(cpi, x, use_dc_pred);
+}
+
+DECLARE_ALIGNED(16, static const uint8_t, vp9_64x64_zeros[64*64]) = { 0 };
+
+
+// Measure the activity of the current macroblock
+// What we measure here is TBD so abstracted to this function
+#define ALT_ACT_MEASURE 1
+static unsigned int mb_activity_measure(VP9_COMP *cpi, MACROBLOCK *x,
+                                        int mb_row, int mb_col) {
+  unsigned int mb_activity;
+
+  if (ALT_ACT_MEASURE) {
+    int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row);
+
+    // Or use and alternative.
+    mb_activity = alt_activity_measure(cpi, x, use_dc_pred);
+  } else {
+    // Original activity measure from Tim T's code.
+    mb_activity = tt_activity_measure(cpi, x);
+  }
+
+  if (mb_activity < VP9_ACTIVITY_AVG_MIN)
+    mb_activity = VP9_ACTIVITY_AVG_MIN;
+
+  return mb_activity;
+}
+
+// Calculate an "average" mb activity value for the frame
+#define ACT_MEDIAN 0
+static void calc_av_activity(VP9_COMP *cpi, int64_t activity_sum) {
+#if ACT_MEDIAN
+  // Find median: Simple n^2 algorithm for experimentation
+  {
+    unsigned int median;
+    unsigned int i, j;
+    unsigned int *sortlist;
+    unsigned int tmp;
+
+    // Create a list to sort to
+    CHECK_MEM_ERROR(sortlist,
+    vpx_calloc(sizeof(unsigned int),
+    cpi->common.MBs));
+
+    // Copy map to sort list
+    vpx_memcpy(sortlist, cpi->mb_activity_map,
+    sizeof(unsigned int) * cpi->common.MBs);
+
+
+    // Ripple each value down to its correct position
+    for (i = 1; i < cpi->common.MBs; i ++) {
+      for (j = i; j > 0; j --) {
+        if (sortlist[j] < sortlist[j - 1]) {
+          // Swap values
+          tmp = sortlist[j - 1];
+          sortlist[j - 1] = sortlist[j];
+          sortlist[j] = tmp;
+        } else
+          break;
+      }
+    }
+
+    // Even number MBs so estimate median as mean of two either side.
+    median = (1 + sortlist[cpi->common.MBs >> 1] +
+              sortlist[(cpi->common.MBs >> 1) + 1]) >> 1;
+
+    cpi->activity_avg = median;
+
+    vpx_free(sortlist);
+  }
+#else
+  // Simple mean for now
+  cpi->activity_avg = (unsigned int)(activity_sum / cpi->common.MBs);
+#endif
+
+  if (cpi->activity_avg < VP9_ACTIVITY_AVG_MIN)
+    cpi->activity_avg = VP9_ACTIVITY_AVG_MIN;
+
+  // Experimental code: return fixed value normalized for several clips
+  if (ALT_ACT_MEASURE)
+    cpi->activity_avg = 100000;
+}
+
+#define USE_ACT_INDEX   0
+#define OUTPUT_NORM_ACT_STATS   0
+
+#if USE_ACT_INDEX
+// Calculate an activity index for each mb
+static void calc_activity_index(VP9_COMP *cpi, MACROBLOCK *x) {
+  VP9_COMMON *const cm = &cpi->common;
+  int mb_row, mb_col;
+
+  int64_t act;
+  int64_t a;
+  int64_t b;
+
+#if OUTPUT_NORM_ACT_STATS
+  FILE *f = fopen("norm_act.stt", "a");
+  fprintf(f, "\n%12d\n", cpi->activity_avg);
+#endif
+
+  // Reset pointers to start of activity map
+  x->mb_activity_ptr = cpi->mb_activity_map;
+
+  // Calculate normalized mb activity number.
+  for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
+    // for each macroblock col in image
+    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+      // Read activity from the map
+      act = *(x->mb_activity_ptr);
+
+      // Calculate a normalized activity number
+      a = act + 4 * cpi->activity_avg;
+      b = 4 * act + cpi->activity_avg;
+
+      if (b >= a)
+        *(x->activity_ptr) = (int)((b + (a >> 1)) / a) - 1;
+      else
+        *(x->activity_ptr) = 1 - (int)((a + (b >> 1)) / b);
+
+#if OUTPUT_NORM_ACT_STATS
+      fprintf(f, " %6d", *(x->mb_activity_ptr));
+#endif
+      // Increment activity map pointers
+      x->mb_activity_ptr++;
+    }
+
+#if OUTPUT_NORM_ACT_STATS
+    fprintf(f, "\n");
+#endif
+
+  }
+
+#if OUTPUT_NORM_ACT_STATS
+  fclose(f);
+#endif
+
+}
+#endif
+
+// Loop through all MBs. Note activity of each, average activity and
+// calculate a normalized activity for each
+static void build_activity_map(VP9_COMP *cpi) {
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *xd = &x->e_mbd;
+  VP9_COMMON *const cm = &cpi->common;
+
+#if ALT_ACT_MEASURE
+  YV12_BUFFER_CONFIG *new_yv12 = &cm->yv12_fb[cm->new_fb_idx];
+  int recon_yoffset;
+  int recon_y_stride = new_yv12->y_stride;
+#endif
+
+  int mb_row, mb_col;
+  unsigned int mb_activity;
+  int64_t activity_sum = 0;
+
+  x->mb_activity_ptr = cpi->mb_activity_map;
+
+  // for each macroblock row in image
+  for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
+#if ALT_ACT_MEASURE
+    // reset above block coeffs
+    xd->up_available = (mb_row != 0);
+    recon_yoffset = (mb_row * recon_y_stride * 16);
+#endif
+    // for each macroblock col in image
+    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+#if ALT_ACT_MEASURE
+      xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
+      xd->left_available = (mb_col != 0);
+      recon_yoffset += 16;
+#endif
+
+      // measure activity
+      mb_activity = mb_activity_measure(cpi, x, mb_row, mb_col);
+
+      // Keep frame sum
+      activity_sum += mb_activity;
+
+      // Store MB level activity details.
+      *x->mb_activity_ptr = mb_activity;
+
+      // Increment activity map pointer
+      x->mb_activity_ptr++;
+
+      // adjust to the next column of source macroblocks
+      x->plane[0].src.buf += 16;
+    }
+
+
+    // adjust to the next row of mbs
+    x->plane[0].src.buf += 16 * x->plane[0].src.stride - 16 * cm->mb_cols;
+  }
+
+  // Calculate an "average" MB activity
+  calc_av_activity(cpi, activity_sum);
+
+#if USE_ACT_INDEX
+  // Calculate an activity index number of each mb
+  calc_activity_index(cpi, x);
+#endif
+
+}
+
+// Macroblock activity masking
+void vp9_activity_masking(VP9_COMP *cpi, MACROBLOCK *x) {
+#if USE_ACT_INDEX
+  x->rdmult += *(x->mb_activity_ptr) * (x->rdmult >> 2);
+  x->errorperbit = x->rdmult * 100 / (110 * x->rddiv);
+  x->errorperbit += (x->errorperbit == 0);
+#else
+  int64_t a;
+  int64_t b;
+  int64_t act = *(x->mb_activity_ptr);
+
+  // Apply the masking to the RD multiplier.
+  a = act + (2 * cpi->activity_avg);
+  b = (2 * act) + cpi->activity_avg;
+
+  x->rdmult = (unsigned int)(((int64_t)x->rdmult * b + (a >> 1)) / a);
+  x->errorperbit = x->rdmult * 100 / (110 * x->rddiv);
+  x->errorperbit += (x->errorperbit == 0);
+#endif
+
+  // Activity based Zbin adjustment
+  adjust_act_zbin(cpi, x);
+}
+
+static void update_state(VP9_COMP *cpi,
+                         PICK_MODE_CONTEXT *ctx,
+                         BLOCK_SIZE_TYPE bsize,
+                         int output_enabled) {
+  int i, x_idx, y;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MODE_INFO *mi = &ctx->mic;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+#if CONFIG_DEBUG || CONFIG_INTERNAL_STATS
+  MB_PREDICTION_MODE mb_mode = mi->mbmi.mode;
+#endif
+  int mb_mode_index = ctx->best_mode_index;
+  const int mis = cpi->common.mode_info_stride;
+  const int bh = 1 << mi_height_log2(bsize), bw = 1 << mi_width_log2(bsize);
+
+#if CONFIG_DEBUG
+  assert(mb_mode < MB_MODE_COUNT);
+  assert(mb_mode_index < MAX_MODES);
+  assert(mi->mbmi.ref_frame[0] < MAX_REF_FRAMES);
+  assert(mi->mbmi.ref_frame[1] < MAX_REF_FRAMES);
+#endif
+
+  assert(mi->mbmi.sb_type == bsize);
+  // Restore the coding context of the MB to that that was in place
+  // when the mode was picked for it
+  for (y = 0; y < bh; y++) {
+    for (x_idx = 0; x_idx < bw; x_idx++) {
+      if ((xd->mb_to_right_edge >> (3 + LOG2_MI_SIZE)) + bw > x_idx &&
+          (xd->mb_to_bottom_edge >> (3 + LOG2_MI_SIZE)) + bh > y) {
+        MODE_INFO *mi_addr = xd->mode_info_context + x_idx + y * mis;
+        *mi_addr = *mi;
+      }
+    }
+  }
+  if (bsize < BLOCK_SIZE_SB32X32) {
+    if (bsize < BLOCK_SIZE_MB16X16)
+      ctx->txfm_rd_diff[ALLOW_16X16] = ctx->txfm_rd_diff[ALLOW_8X8];
+    ctx->txfm_rd_diff[ALLOW_32X32] = ctx->txfm_rd_diff[ALLOW_16X16];
+  }
+
+  if (mbmi->ref_frame[0] != INTRA_FRAME && mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+    *x->partition_info = ctx->partition_info;
+    mbmi->mv[0].as_int = x->partition_info->bmi[3].mv.as_int;
+    mbmi->mv[1].as_int = x->partition_info->bmi[3].second_mv.as_int;
+  }
+
+  x->skip = ctx->skip;
+  if (!output_enabled)
+    return;
+
+  if (!vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_SKIP)) {
+    for (i = 0; i < NB_TXFM_MODES; i++) {
+      cpi->rd_tx_select_diff[i] += ctx->txfm_rd_diff[i];
+    }
+  }
+
+  if (cpi->common.frame_type == KEY_FRAME) {
+    // Restore the coding modes to that held in the coding context
+    // if (mb_mode == I4X4_PRED)
+    //    for (i = 0; i < 16; i++)
+    //    {
+    //        xd->block[i].bmi.as_mode =
+    //                          xd->mode_info_context->bmi[i].as_mode;
+    //        assert(xd->mode_info_context->bmi[i].as_mode < MB_MODE_COUNT);
+    //    }
+#if CONFIG_INTERNAL_STATS
+    static const int kf_mode_index[] = {
+      THR_DC /*DC_PRED*/,
+      THR_V_PRED /*V_PRED*/,
+      THR_H_PRED /*H_PRED*/,
+      THR_D45_PRED /*D45_PRED*/,
+      THR_D135_PRED /*D135_PRED*/,
+      THR_D117_PRED /*D117_PRED*/,
+      THR_D153_PRED /*D153_PRED*/,
+      THR_D27_PRED /*D27_PRED*/,
+      THR_D63_PRED /*D63_PRED*/,
+      THR_TM /*TM_PRED*/,
+      THR_B_PRED /*I4X4_PRED*/,
+    };
+    cpi->mode_chosen_counts[kf_mode_index[mb_mode]]++;
+#endif
+  } else {
+    /*
+            // Reduce the activation RD thresholds for the best choice mode
+            if ((cpi->rd_baseline_thresh[mb_mode_index] > 0) &&
+                (cpi->rd_baseline_thresh[mb_mode_index] < (INT_MAX >> 2)))
+            {
+                int best_adjustment = (cpi->rd_thresh_mult[mb_mode_index] >> 2);
+
+                cpi->rd_thresh_mult[mb_mode_index] =
+                        (cpi->rd_thresh_mult[mb_mode_index]
+                         >= (MIN_THRESHMULT + best_adjustment)) ?
+                                cpi->rd_thresh_mult[mb_mode_index] - best_adjustment :
+                                MIN_THRESHMULT;
+                cpi->rd_threshes[mb_mode_index] =
+                        (cpi->rd_baseline_thresh[mb_mode_index] >> 7)
+                        * cpi->rd_thresh_mult[mb_mode_index];
+
+            }
+    */
+    // Note how often each mode chosen as best
+    cpi->mode_chosen_counts[mb_mode_index]++;
+    if (mbmi->ref_frame[0] != INTRA_FRAME &&
+        (mbmi->sb_type < BLOCK_SIZE_SB8X8 || mbmi->mode == NEWMV)) {
+      int_mv best_mv, best_second_mv;
+      const MV_REFERENCE_FRAME rf1 = mbmi->ref_frame[0];
+      const MV_REFERENCE_FRAME rf2 = mbmi->ref_frame[1];
+      best_mv.as_int = ctx->best_ref_mv.as_int;
+      best_second_mv.as_int = ctx->second_best_ref_mv.as_int;
+      if (mbmi->mode == NEWMV) {
+        best_mv.as_int = mbmi->ref_mvs[rf1][0].as_int;
+        best_second_mv.as_int = mbmi->ref_mvs[rf2][0].as_int;
+      }
+      mbmi->best_mv.as_int = best_mv.as_int;
+      mbmi->best_second_mv.as_int = best_second_mv.as_int;
+      vp9_update_nmv_count(cpi, x, &best_mv, &best_second_mv);
+    }
+
+    if (bsize > BLOCK_SIZE_SB8X8 && mbmi->mode == NEWMV) {
+      int i, j;
+      for (j = 0; j < bh; ++j)
+        for (i = 0; i < bw; ++i)
+          if ((xd->mb_to_right_edge >> (3 + LOG2_MI_SIZE)) + bw > i &&
+              (xd->mb_to_bottom_edge >> (3 + LOG2_MI_SIZE)) + bh > j)
+            xd->mode_info_context[mis * j + i].mbmi = *mbmi;
+    }
+
+    if (cpi->common.mcomp_filter_type == SWITCHABLE &&
+        is_inter_mode(mbmi->mode)) {
+      ++cpi->common.fc.switchable_interp_count
+          [vp9_get_pred_context(&cpi->common, xd, PRED_SWITCHABLE_INTERP)]
+          [vp9_switchable_interp_map[mbmi->interp_filter]];
+    }
+
+    cpi->rd_comp_pred_diff[SINGLE_PREDICTION_ONLY] += ctx->single_pred_diff;
+    cpi->rd_comp_pred_diff[COMP_PREDICTION_ONLY]   += ctx->comp_pred_diff;
+    cpi->rd_comp_pred_diff[HYBRID_PREDICTION]      += ctx->hybrid_pred_diff;
+  }
+}
+
+static unsigned find_seg_id(VP9_COMMON *cm, uint8_t *buf, BLOCK_SIZE_TYPE bsize,
+                            int start_y, int height, int start_x, int width) {
+  const int bw = 1 << mi_width_log2(bsize), bh = 1 << mi_height_log2(bsize);
+  const int end_x = MIN(start_x + bw, width);
+  const int end_y = MIN(start_y + bh, height);
+  int x, y;
+  unsigned seg_id = -1;
+
+  buf += width * start_y;
+  assert(start_y < cm->mi_rows && start_x < cm->cur_tile_mi_col_end);
+  for (y = start_y; y < end_y; y++, buf += width) {
+    for (x = start_x; x < end_x; x++) {
+      seg_id = MIN(seg_id, buf[x]);
+    }
+  }
+
+  return seg_id;
+}
+
+void vp9_setup_src_planes(MACROBLOCK *x,
+                          const YV12_BUFFER_CONFIG *src,
+                          int mb_row, int mb_col) {
+  uint8_t *buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
+                         src->alpha_buffer};
+  int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
+                    src->alpha_stride};
+  int i;
+
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    setup_pred_plane(&x->plane[i].src,
+                     buffers[i], strides[i],
+                     mb_row, mb_col, NULL,
+                     x->e_mbd.plane[i].subsampling_x,
+                     x->e_mbd.plane[i].subsampling_y);
+  }
+}
+
+static void set_offsets(VP9_COMP *cpi,
+                        int mi_row, int mi_col, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCK *const x = &cpi->mb;
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *mbmi;
+  const int dst_fb_idx = cm->new_fb_idx;
+  const int idx_str = xd->mode_info_stride * mi_row + mi_col;
+  const int bw = 1 << mi_width_log2(bsize), bh = 1 << mi_height_log2(bsize);
+  const int mb_row = mi_row >> 1;
+  const int mb_col = mi_col >> 1;
+  const int idx_map = mb_row * cm->mb_cols + mb_col;
+  int i;
+
+  // entropy context structures
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    xd->plane[i].above_context = cm->above_context[i] +
+        (mi_col * 2 >>  xd->plane[i].subsampling_x);
+    xd->plane[i].left_context = cm->left_context[i] +
+        (((mi_row * 2) & 15) >> xd->plane[i].subsampling_y);
+  }
+
+  // partition contexts
+  set_partition_seg_context(cm, xd, mi_row, mi_col);
+
+  // Activity map pointer
+  x->mb_activity_ptr = &cpi->mb_activity_map[idx_map];
+  x->active_ptr = cpi->active_map + idx_map;
+
+  /* pointers to mode info contexts */
+  x->partition_info          = x->pi + idx_str;
+  xd->mode_info_context      = cm->mi + idx_str;
+  mbmi = &xd->mode_info_context->mbmi;
+  // Special case: if prev_mi is NULL, the previous mode info context
+  // cannot be used.
+  xd->prev_mode_info_context = cm->prev_mi ?
+                                 cm->prev_mi + idx_str : NULL;
+
+  // Set up destination pointers
+  setup_dst_planes(xd, &cm->yv12_fb[dst_fb_idx], mi_row, mi_col);
+
+  /* Set up limit values for MV components to prevent them from
+   * extending beyond the UMV borders assuming 16x16 block size */
+  x->mv_row_min = -((mi_row * MI_SIZE) + VP9BORDERINPIXELS - VP9_INTERP_EXTEND);
+  x->mv_col_min = -((mi_col * MI_SIZE) + VP9BORDERINPIXELS - VP9_INTERP_EXTEND);
+  x->mv_row_max = ((cm->mi_rows - mi_row) * MI_SIZE +
+                   (VP9BORDERINPIXELS - MI_SIZE * bh - VP9_INTERP_EXTEND));
+  x->mv_col_max = ((cm->mi_cols - mi_col) * MI_SIZE +
+                   (VP9BORDERINPIXELS - MI_SIZE * bw - VP9_INTERP_EXTEND));
+
+  // Set up distance of MB to edge of frame in 1/8th pel units
+  assert(!(mi_col & (bw - 1)) && !(mi_row & (bh - 1)));
+  set_mi_row_col(cm, xd, mi_row, bh, mi_col, bw);
+
+  /* set up source buffers */
+  vp9_setup_src_planes(x, cpi->Source, mi_row, mi_col);
+
+  /* R/D setup */
+  x->rddiv = cpi->RDDIV;
+  x->rdmult = cpi->RDMULT;
+
+  /* segment ID */
+  if (xd->segmentation_enabled) {
+    uint8_t *map = xd->update_mb_segmentation_map ? cpi->segmentation_map
+                                                  : cm->last_frame_seg_map;
+    mbmi->segment_id = find_seg_id(cm, map, bsize, mi_row,
+                                   cm->mi_rows, mi_col, cm->mi_cols);
+
+    assert(mbmi->segment_id <= (MAX_MB_SEGMENTS-1));
+    vp9_mb_init_quantizer(cpi, x);
+
+    if (xd->segmentation_enabled && cpi->seg0_cnt > 0 &&
+        !vp9_segfeature_active(xd, 0, SEG_LVL_REF_FRAME) &&
+        vp9_segfeature_active(xd, 1, SEG_LVL_REF_FRAME)) {
+      cpi->seg0_progress = (cpi->seg0_idx << 16) / cpi->seg0_cnt;
+    } else {
+      const int y = mb_row & ~3;
+      const int x = mb_col & ~3;
+      const int p16 = ((mb_row & 1) << 1) +  (mb_col & 1);
+      const int p32 = ((mb_row & 2) << 2) + ((mb_col & 2) << 1);
+      const int tile_progress =
+          cm->cur_tile_mi_col_start * cm->mb_rows >> 1;
+      const int mb_cols =
+          (cm->cur_tile_mi_col_end - cm->cur_tile_mi_col_start) >> 1;
+
+      cpi->seg0_progress =
+          ((y * mb_cols + x * 4 + p32 + p16 + tile_progress) << 16) / cm->MBs;
+    }
+  } else {
+    mbmi->segment_id = 0;
+  }
+}
+
+static void pick_sb_modes(VP9_COMP *cpi, int mi_row, int mi_col,
+                          TOKENEXTRA **tp, int *totalrate, int *totaldist,
+                          BLOCK_SIZE_TYPE bsize, PICK_MODE_CONTEXT *ctx) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  x->rd_search = 1;
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index != 0)
+      return;
+
+  set_offsets(cpi, mi_row, mi_col, bsize);
+  xd->mode_info_context->mbmi.sb_type = bsize;
+  if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
+    vp9_activity_masking(cpi, x);
+
+  /* Find best coding mode & reconstruct the MB so it is available
+   * as a predictor for MBs that follow in the SB */
+  if (cm->frame_type == KEY_FRAME) {
+    vp9_rd_pick_intra_mode_sb(cpi, x, totalrate, totaldist, bsize, ctx);
+  } else {
+    vp9_rd_pick_inter_mode_sb(cpi, x, mi_row, mi_col, totalrate, totaldist,
+                              bsize, ctx);
+  }
+}
+
+static void update_stats(VP9_COMP *cpi, int mi_row, int mi_col) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MODE_INFO *mi = xd->mode_info_context;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+
+  if (cm->frame_type != KEY_FRAME) {
+    int segment_id, seg_ref_active;
+
+    segment_id = mbmi->segment_id;
+    seg_ref_active = vp9_segfeature_active(xd, segment_id,
+                                           SEG_LVL_REF_FRAME);
+
+    if (!seg_ref_active)
+      cpi->intra_inter_count[vp9_get_pred_context(cm, xd, PRED_INTRA_INTER)]
+                            [mbmi->ref_frame[0] > INTRA_FRAME]++;
+
+    // If the segment reference feature is enabled we have only a single
+    // reference frame allowed for the segment so exclude it from
+    // the reference frame counts used to work out probabilities.
+    if ((mbmi->ref_frame[0] > INTRA_FRAME) && !seg_ref_active) {
+      if (cm->comp_pred_mode == HYBRID_PREDICTION)
+        cpi->comp_inter_count[vp9_get_pred_context(cm, xd,
+                                                   PRED_COMP_INTER_INTER)]
+                             [mbmi->ref_frame[1] > INTRA_FRAME]++;
+
+      if (mbmi->ref_frame[1] > INTRA_FRAME) {
+        cpi->comp_ref_count[vp9_get_pred_context(cm, xd, PRED_COMP_REF_P)]
+                           [mbmi->ref_frame[0] == GOLDEN_FRAME]++;
+      } else {
+        cpi->single_ref_count[vp9_get_pred_context(cm, xd, PRED_SINGLE_REF_P1)]
+                             [0][mbmi->ref_frame[0] != LAST_FRAME]++;
+        if (mbmi->ref_frame[0] != LAST_FRAME)
+          cpi->single_ref_count[vp9_get_pred_context(cm, xd,
+                                                     PRED_SINGLE_REF_P2)]
+                               [1][mbmi->ref_frame[0] != GOLDEN_FRAME]++;
+      }
+    }
+    // Count of last ref frame 0,0 usage
+    if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame[0] == LAST_FRAME))
+      cpi->inter_zz_count++;
+  }
+}
+
+// TODO(jingning): the variables used here are little complicated. need further
+// refactoring on organizing the the temporary buffers, when recursive
+// partition down to 4x4 block size is enabled.
+static PICK_MODE_CONTEXT *get_block_context(MACROBLOCK *x,
+                                            BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  switch (bsize) {
+    case BLOCK_SIZE_SB64X64:
+      return &x->sb64_context;
+    case BLOCK_SIZE_SB64X32:
+      return &x->sb64x32_context[xd->sb_index];
+    case BLOCK_SIZE_SB32X64:
+      return &x->sb32x64_context[xd->sb_index];
+    case BLOCK_SIZE_SB32X32:
+      return &x->sb32_context[xd->sb_index];
+    case BLOCK_SIZE_SB32X16:
+      return &x->sb32x16_context[xd->sb_index][xd->mb_index];
+    case BLOCK_SIZE_SB16X32:
+      return &x->sb16x32_context[xd->sb_index][xd->mb_index];
+    case BLOCK_SIZE_MB16X16:
+      return &x->mb_context[xd->sb_index][xd->mb_index];
+    case BLOCK_SIZE_SB16X8:
+      return &x->sb16x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+    case BLOCK_SIZE_SB8X16:
+      return &x->sb8x16_context[xd->sb_index][xd->mb_index][xd->b_index];
+    case BLOCK_SIZE_SB8X8:
+      return &x->sb8x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+    case BLOCK_SIZE_SB8X4:
+      return &x->sb8x4_context[xd->sb_index][xd->mb_index][xd->b_index];
+    case BLOCK_SIZE_SB4X8:
+      return &x->sb4x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+    case BLOCK_SIZE_AB4X4:
+      return &x->ab4x4_context[xd->sb_index][xd->mb_index][xd->b_index];
+    default:
+      assert(0);
+      return NULL;
+  }
+}
+
+static BLOCK_SIZE_TYPE *get_sb_partitioning(MACROBLOCK *x,
+                                            BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *xd = &x->e_mbd;
+  switch (bsize) {
+    case BLOCK_SIZE_SB64X64:
+      return &x->sb64_partitioning;
+    case BLOCK_SIZE_SB32X32:
+      return &x->sb_partitioning[xd->sb_index];
+    case BLOCK_SIZE_MB16X16:
+      return &x->mb_partitioning[xd->sb_index][xd->mb_index];
+    case BLOCK_SIZE_SB8X8:
+      return &x->b_partitioning[xd->sb_index][xd->mb_index][xd->b_index];
+    default:
+      assert(0);
+      return NULL;
+  }
+}
+
+static void restore_context(VP9_COMP *cpi, int mi_row, int mi_col,
+                            ENTROPY_CONTEXT a[16 * MAX_MB_PLANE],
+                            ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
+                            PARTITION_CONTEXT sa[8],
+                            PARTITION_CONTEXT sl[8],
+                            BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int p;
+  int bwl = b_width_log2(bsize), bw = 1 << bwl;
+  int bhl = b_height_log2(bsize), bh = 1 << bhl;
+  int mwl = mi_width_log2(bsize), mw = 1 << mwl;
+  int mhl = mi_height_log2(bsize), mh = 1 << mhl;
+  for (p = 0; p < MAX_MB_PLANE; p++) {
+    vpx_memcpy(cm->above_context[p] +
+               ((mi_col * 2) >> xd->plane[p].subsampling_x),
+               a + bw * p,
+               sizeof(ENTROPY_CONTEXT) * bw >> xd->plane[p].subsampling_x);
+    vpx_memcpy(cm->left_context[p] +
+               ((mi_row & MI_MASK) * 2 >> xd->plane[p].subsampling_y),
+               l + bh * p,
+               sizeof(ENTROPY_CONTEXT) * bh >> xd->plane[p].subsampling_y);
+  }
+  vpx_memcpy(cm->above_seg_context + mi_col, sa,
+             sizeof(PARTITION_CONTEXT) * mw);
+  vpx_memcpy(cm->left_seg_context + (mi_row & MI_MASK), sl,
+             sizeof(PARTITION_CONTEXT) * mh);
+}
+static void save_context(VP9_COMP *cpi, int mi_row, int mi_col,
+                          ENTROPY_CONTEXT a[16 * MAX_MB_PLANE],
+                          ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
+                          PARTITION_CONTEXT sa[8],
+                          PARTITION_CONTEXT sl[8],
+                          BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int p;
+  int bwl = b_width_log2(bsize), bw = 1 << bwl;
+  int bhl = b_height_log2(bsize), bh = 1 << bhl;
+  int mwl = mi_width_log2(bsize), mw = 1 << mwl;
+  int mhl = mi_height_log2(bsize), mh = 1 << mhl;
+
+  // buffer the above/left context information of the block in search.
+  for (p = 0; p < MAX_MB_PLANE; ++p) {
+    vpx_memcpy(a + bw * p, cm->above_context[p] +
+               (mi_col * 2 >> xd->plane[p].subsampling_x),
+               sizeof(ENTROPY_CONTEXT) * bw >> xd->plane[p].subsampling_x);
+    vpx_memcpy(l + bh * p, cm->left_context[p] +
+               ((mi_row & MI_MASK) * 2 >> xd->plane[p].subsampling_y),
+               sizeof(ENTROPY_CONTEXT) * bh >> xd->plane[p].subsampling_y);
+  }
+  vpx_memcpy(sa, cm->above_seg_context + mi_col,
+             sizeof(PARTITION_CONTEXT) * mw);
+  vpx_memcpy(sl, cm->left_seg_context + (mi_row & MI_MASK),
+             sizeof(PARTITION_CONTEXT) * mh);
+}
+
+static void encode_b(VP9_COMP *cpi, TOKENEXTRA **tp,
+                     int mi_row, int mi_col, int output_enabled,
+                     BLOCK_SIZE_TYPE bsize, int sub_index) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  if (sub_index != -1)
+    *(get_sb_index(xd, bsize)) = sub_index;
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index > 0)
+      return;
+  set_offsets(cpi, mi_row, mi_col, bsize);
+  update_state(cpi, get_block_context(x, bsize), bsize, output_enabled);
+  encode_superblock(cpi, tp, output_enabled, mi_row, mi_col, bsize);
+
+  if (output_enabled) {
+    update_stats(cpi, mi_row, mi_col);
+
+    (*tp)->token = EOSB_TOKEN;
+    (*tp)++;
+  }
+}
+
+static void encode_sb(VP9_COMP *cpi, TOKENEXTRA **tp,
+                      int mi_row, int mi_col, int output_enabled,
+                      BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  BLOCK_SIZE_TYPE c1 = BLOCK_SIZE_SB8X8;
+  const int bsl = b_width_log2(bsize), bs = (1 << bsl) / 4;
+  int bwl, bhl;
+  int UNINITIALIZED_IS_SAFE(pl);
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  c1 = BLOCK_SIZE_AB4X4;
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    pl = partition_plane_context(xd, bsize);
+    c1 = *(get_sb_partitioning(x, bsize));
+  }
+
+  bwl = b_width_log2(c1), bhl = b_height_log2(c1);
+
+  if (bsl == bwl && bsl == bhl) {
+    if (output_enabled && bsize >= BLOCK_SIZE_SB8X8)
+        cpi->partition_count[pl][PARTITION_NONE]++;
+    encode_b(cpi, tp, mi_row, mi_col, output_enabled, c1, -1);
+  } else if (bsl == bhl && bsl > bwl) {
+    if (output_enabled)
+      cpi->partition_count[pl][PARTITION_VERT]++;
+    encode_b(cpi, tp, mi_row, mi_col,      output_enabled, c1, 0);
+    encode_b(cpi, tp, mi_row, mi_col + bs, output_enabled, c1, 1);
+  } else if (bsl == bwl && bsl > bhl) {
+    if (output_enabled)
+      cpi->partition_count[pl][PARTITION_HORZ]++;
+    encode_b(cpi, tp, mi_row,      mi_col, output_enabled, c1, 0);
+    encode_b(cpi, tp, mi_row + bs, mi_col, output_enabled, c1, 1);
+  } else {
+    BLOCK_SIZE_TYPE subsize;
+    int i;
+
+    assert(bwl < bsl && bhl < bsl);
+    subsize = get_subsize(bsize, PARTITION_SPLIT);
+
+    if (output_enabled)
+      cpi->partition_count[pl][PARTITION_SPLIT]++;
+
+    for (i = 0; i < 4; i++) {
+      const int x_idx = i & 1, y_idx = i >> 1;
+
+      *(get_sb_index(xd, subsize)) = i;
+      encode_sb(cpi, tp, mi_row + y_idx * bs, mi_col + x_idx * bs,
+                output_enabled, subsize);
+    }
+  }
+
+  if (bsize >= BLOCK_SIZE_SB8X8 &&
+      (bsize == BLOCK_SIZE_SB8X8 || bsl == bwl || bsl == bhl)) {
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    update_partition_context(xd, c1, bsize);
+  }
+}
+
+static void set_partitioning(VP9_COMP *cpi, MODE_INFO *m,
+                             BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  const int mis = cm->mode_info_stride;
+  int bsl = b_width_log2(bsize);
+  int bs = (1 << bsl) / 2;  //
+  int block_row, block_col;
+  int row, col;
+
+  // this test function sets the entire macroblock to the same bsize
+  for (block_row = 0; block_row < 8; block_row += bs) {
+    for (block_col = 0; block_col < 8; block_col += bs) {
+      for (row = 0; row < bs; row++) {
+        for (col = 0; col < bs; col++) {
+          m[(block_row+row)*mis + block_col+col].mbmi.sb_type = bsize;
+        }
+      }
+    }
+  }
+}
+
+static void set_block_size(VP9_COMMON *const cm,
+                           MODE_INFO *m, BLOCK_SIZE_TYPE bsize, int mis,
+                           int mi_row, int mi_col) {
+  int row, col;
+  int bwl = b_width_log2(bsize);
+  int bhl = b_height_log2(bsize);
+  int bsl = (bwl > bhl ? bwl : bhl);
+
+  int bs = (1 << bsl) / 2;  //
+  MODE_INFO *m2 = m + mi_row * mis + mi_col;
+  for (row = 0; row < bs; row++) {
+    for (col = 0; col < bs; col++) {
+      if (mi_row + row >= cm->mi_rows || mi_col + col >= cm->mi_cols)
+        continue;
+      m2[row*mis+col].mbmi.sb_type = bsize;
+    }
+  }
+}
+typedef struct {
+  int64_t sum_square_error;
+  int64_t sum_error;
+  int count;
+  int variance;
+} var;
+
+#define VT(TYPE, BLOCKSIZE) \
+  typedef struct { \
+    var none; \
+    var horz[2]; \
+    var vert[2]; \
+    BLOCKSIZE split[4]; } TYPE;
+
+VT(v8x8, var)
+VT(v16x16, v8x8)
+VT(v32x32, v16x16)
+VT(v64x64, v32x32)
+
+typedef enum {
+  V16X16,
+  V32X32,
+  V64X64,
+} TREE_LEVEL;
+
+// Set variance values given sum square error, sum error, count.
+static void fill_variance(var *v, int64_t s2, int64_t s, int c) {
+  v->sum_square_error = s2;
+  v->sum_error = s;
+  v->count = c;
+  v->variance = 256
+      * (v->sum_square_error - v->sum_error * v->sum_error / v->count)
+      / v->count;
+}
+
+// Combine 2 variance structures by summing the sum_error, sum_square_error,
+// and counts and then calculating the new variance.
+void sum_2_variances(var *r, var *a, var*b) {
+  fill_variance(r, a->sum_square_error + b->sum_square_error,
+                a->sum_error + b->sum_error, a->count + b->count);
+}
+// Fill one level of our variance tree,  by summing the split sums into each of
+// the horizontal, vertical and none from split and recalculating variance.
+#define fill_variance_tree(VT) \
+  sum_2_variances(VT.horz[0], VT.split[0].none, VT.split[1].none); \
+  sum_2_variances(VT.horz[1], VT.split[2].none, VT.split[3].none); \
+  sum_2_variances(VT.vert[0], VT.split[0].none, VT.split[2].none); \
+  sum_2_variances(VT.vert[1], VT.split[1].none, VT.split[3].none); \
+  sum_2_variances(VT.none, VT.vert[0], VT.vert[1]);
+
+// Set the blocksize in the macroblock info structure if the variance is less
+// than our threshold to one of none, horz, vert.
+#define set_vt_size(VT, BLOCKSIZE, R, C, ACTION) \
+  if (VT.none.variance < threshold) { \
+    set_block_size(cm, m, BLOCKSIZE, mis, R, C); \
+    ACTION; \
+  } \
+  if (VT.horz[0].variance < threshold && VT.horz[1].variance < threshold ) { \
+    set_block_size(cm, m, get_subsize(BLOCKSIZE, PARTITION_HORZ), mis, R, C); \
+    ACTION; \
+  } \
+  if (VT.vert[0].variance < threshold && VT.vert[1].variance < threshold ) { \
+    set_block_size(cm, m, get_subsize(BLOCKSIZE, PARTITION_VERT), mis, R, C); \
+    ACTION; \
+  }
+
+static void choose_partitioning(VP9_COMP *cpi, MODE_INFO *m, int mi_row,
+                                int mi_col) {
+  VP9_COMMON * const cm = &cpi->common;
+  MACROBLOCK *x = &cpi->mb;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  const int mis = cm->mode_info_stride;
+  // TODO(JBB): More experimentation or testing of this threshold;
+  int64_t threshold = 4;
+  int i, j, k;
+  v64x64 vt;
+  unsigned char * s;
+  int sp;
+  const unsigned char * d = xd->plane[0].pre->buf;
+  int dp = xd->plane[0].pre->stride;
+  int pixels_wide = 64, pixels_high = 64;
+
+  vpx_memset(&vt, 0, sizeof(vt));
+
+  set_offsets(cpi, mi_row, mi_col, BLOCK_SIZE_SB64X64);
+
+  if (xd->mb_to_right_edge < 0)
+    pixels_wide += (xd->mb_to_right_edge >> 3);
+
+  if (xd->mb_to_bottom_edge < 0)
+    pixels_high += (xd->mb_to_bottom_edge >> 3);
+
+  s = x->plane[0].src.buf;
+  sp = x->plane[0].src.stride;
+
+  // TODO(JBB): Clearly the higher the quantizer the fewer partitions we want
+  // but this needs more experimentation.
+  threshold = threshold * cpi->common.base_qindex * cpi->common.base_qindex;
+
+  // if ( cm->frame_type == KEY_FRAME ) {
+  d = vp9_64x64_zeros;
+  dp = 64;
+  // }
+
+  // Fill in the entire tree of 8x8 variances for splits.
+  for (i = 0; i < 4; i++) {
+    const int x32_idx = ((i & 1) << 5);
+    const int y32_idx = ((i >> 1) << 5);
+    for (j = 0; j < 4; j++) {
+      const int x_idx = x32_idx + ((j & 1) << 4);
+      const int y_idx = y32_idx + ((j >> 1) << 4);
+      const uint8_t *st = s + y_idx * sp + x_idx;
+      const uint8_t *dt = d + y_idx * dp + x_idx;
+      unsigned int sse = 0;
+      int sum = 0;
+      v16x16 *vst = &vt.split[i].split[j];
+      sse = sum = 0;
+      if (x_idx < pixels_wide && y_idx < pixels_high)
+        vp9_get_sse_sum_8x8(st, sp, dt, dp, &sse, &sum);
+      fill_variance(&vst->split[0].none, sse, sum, 64);
+      sse = sum = 0;
+      if (x_idx + 8 < pixels_wide && y_idx < pixels_high)
+        vp9_get_sse_sum_8x8(st + 8, sp, dt + 8, dp, &sse, &sum);
+      fill_variance(&vst->split[1].none, sse, sum, 64);
+      sse = sum = 0;
+      if (x_idx < pixels_wide && y_idx + 8 < pixels_high)
+        vp9_get_sse_sum_8x8(st + 8 * sp, sp, dt + 8 * dp, dp, &sse, &sum);
+      fill_variance(&vst->split[2].none, sse, sum, 64);
+      sse = sum = 0;
+      if (x_idx + 8 < pixels_wide && y_idx + 8 < pixels_high)
+        vp9_get_sse_sum_8x8(st + 8 * sp + 8, sp, dt + 8 + 8 * dp, dp, &sse,
+                            &sum);
+      fill_variance(&vst->split[3].none, sse, sum, 64);
+    }
+  }
+  // Fill the rest of the variance tree by summing the split partition
+  // values.
+  for (i = 0; i < 4; i++) {
+    for (j = 0; j < 4; j++) {
+      fill_variance_tree(&vt.split[i].split[j])
+    }
+    fill_variance_tree(&vt.split[i])
+  }
+  fill_variance_tree(&vt)
+
+  // Now go through the entire structure,  splitting every blocksize until
+  // we get to one that's got a variance lower than our threshold,  or we
+  // hit 8x8.
+  set_vt_size( vt, BLOCK_SIZE_SB64X64, mi_row, mi_col, return);
+  for (i = 0; i < 4; ++i) {
+    const int x32_idx = ((i & 1) << 2);
+    const int y32_idx = ((i >> 1) << 2);
+    set_vt_size(vt, BLOCK_SIZE_SB32X32, mi_row + y32_idx, mi_col + x32_idx,
+                continue);
+
+    for (j = 0; j < 4; ++j) {
+      const int x16_idx = ((j & 1) << 1);
+      const int y16_idx = ((j >> 1) << 1);
+      set_vt_size(vt, BLOCK_SIZE_MB16X16, mi_row + y32_idx + y16_idx,
+                  mi_col+x32_idx+x16_idx, continue);
+
+      for (k = 0; k < 4; ++k) {
+        const int x8_idx = (k & 1);
+        const int y8_idx = (k >> 1);
+        set_block_size(cm, m, BLOCK_SIZE_SB8X8, mis,
+                       mi_row + y32_idx + y16_idx + y8_idx,
+                       mi_col + x32_idx + x16_idx + x8_idx);
+      }
+    }
+  }
+}
+static void rd_use_partition(VP9_COMP *cpi, MODE_INFO *m, TOKENEXTRA **tp,
+                             int mi_row, int mi_col, BLOCK_SIZE_TYPE bsize,
+                             int *rate, int *dist) {
+  VP9_COMMON * const cm = &cpi->common;
+  MACROBLOCK * const x = &cpi->mb;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  const int mis = cm->mode_info_stride;
+  int bwl = b_width_log2(m->mbmi.sb_type);
+  int bhl = b_height_log2(m->mbmi.sb_type);
+  int bsl = b_width_log2(bsize);
+  int bh = (1 << bhl);
+  int bs = (1 << bsl);
+  int bss = (1 << bsl)/4;
+  int i, pl;
+  PARTITION_TYPE partition;
+  BLOCK_SIZE_TYPE subsize;
+  ENTROPY_CONTEXT l[16 * MAX_MB_PLANE], a[16 * MAX_MB_PLANE];
+  PARTITION_CONTEXT sl[8], sa[8];
+  int r = 0, d = 0;
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+
+  // parse the partition type
+  if ((bwl == bsl) && (bhl == bsl))
+    partition = PARTITION_NONE;
+  else if ((bwl == bsl) && (bhl < bsl))
+    partition = PARTITION_HORZ;
+  else if ((bwl < bsl) && (bhl == bsl))
+    partition = PARTITION_VERT;
+  else if ((bwl < bsl) && (bhl < bsl))
+    partition = PARTITION_SPLIT;
+  else
+    assert(0);
+
+  subsize = get_subsize(bsize, partition);
+
+  // TODO(JBB): this restriction is here because pick_sb_modes can return
+  // r's that are INT_MAX meaning we can't select a mode / mv for this block.
+  // when the code is made to work for less than sb8x8 we need to come up with
+  // a solution to this problem.
+  assert(subsize >= BLOCK_SIZE_SB8X8);
+
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    xd->left_seg_context = cm->left_seg_context + (mi_row & MI_MASK);
+    xd->above_seg_context = cm->above_seg_context + mi_col;
+    *(get_sb_partitioning(x, bsize)) = subsize;
+  }
+
+  pl = partition_plane_context(xd, bsize);
+  save_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+  switch (partition) {
+    case PARTITION_NONE:
+      pick_sb_modes(cpi, mi_row, mi_col, tp, &r, &d, bsize,
+                    get_block_context(x, bsize));
+      r += x->partition_cost[pl][PARTITION_NONE];
+      break;
+    case PARTITION_HORZ:
+      *(get_sb_index(xd, subsize)) = 0;
+      pick_sb_modes(cpi, mi_row, mi_col, tp, &r, &d, subsize,
+                    get_block_context(x, subsize));
+      if (mi_row + (bh >> 1) <= cm->mi_rows) {
+        int rt, dt;
+        update_state(cpi, get_block_context(x, subsize), subsize, 0);
+        encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
+        *(get_sb_index(xd, subsize)) = 1;
+        pick_sb_modes(cpi, mi_row + (bs >> 2), mi_col, tp, &rt, &dt, subsize,
+                      get_block_context(x, subsize));
+        r += rt;
+        d += dt;
+      }
+      set_partition_seg_context(cm, xd, mi_row, mi_col);
+      pl = partition_plane_context(xd, bsize);
+      r += x->partition_cost[pl][PARTITION_HORZ];
+      break;
+    case PARTITION_VERT:
+      *(get_sb_index(xd, subsize)) = 0;
+      pick_sb_modes(cpi, mi_row, mi_col, tp, &r, &d, subsize,
+                    get_block_context(x, subsize));
+      if (mi_col + (bs >> 1) <= cm->mi_cols) {
+        int rt, dt;
+        update_state(cpi, get_block_context(x, subsize), subsize, 0);
+        encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
+        *(get_sb_index(xd, subsize)) = 1;
+        pick_sb_modes(cpi, mi_row, mi_col + (bs >> 2), tp, &rt, &dt, subsize,
+                      get_block_context(x, subsize));
+        r += rt;
+        d += dt;
+      }
+      set_partition_seg_context(cm, xd, mi_row, mi_col);
+      pl = partition_plane_context(xd, bsize);
+      r += x->partition_cost[pl][PARTITION_VERT];
+      restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+      break;
+    case PARTITION_SPLIT:
+      for (i = 0; i < 4; i++) {
+        int x_idx = (i & 1) * (bs >> 2);
+        int y_idx = (i >> 1) * (bs >> 2);
+        int jj = i >> 1, ii = i & 0x01;
+        int rt, dt;
+
+        if ((mi_row + y_idx >= cm->mi_rows) || (mi_col + x_idx >= cm->mi_cols))
+          continue;
+
+        *(get_sb_index(xd, subsize)) = i;
+
+        rd_use_partition(cpi, m + jj * bss * mis + ii * bss, tp, mi_row + y_idx,
+                         mi_col + x_idx, subsize, &rt, &dt);
+        r += rt;
+        d += dt;
+      }
+      set_partition_seg_context(cm, xd, mi_row, mi_col);
+      pl = partition_plane_context(xd, bsize);
+      r += x->partition_cost[pl][PARTITION_SPLIT];
+      break;
+    default:
+      assert(0);
+  }
+
+  // update partition context
+#if CONFIG_AB4X4
+  if (bsize >= BLOCK_SIZE_SB8X8 &&
+      (bsize == BLOCK_SIZE_SB8X8 || partition != PARTITION_SPLIT)) {
+#else
+  if (bsize > BLOCK_SIZE_SB8X8
+      && (bsize == BLOCK_SIZE_MB16X16 || partition != PARTITION_SPLIT)) {
+#endif
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    update_partition_context(xd, subsize, bsize);
+  }
+  restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+
+  if (r < INT_MAX && d < INT_MAX)
+    encode_sb(cpi, tp, mi_row, mi_col, bsize == BLOCK_SIZE_SB64X64, bsize);
+  *rate = r;
+  *dist = d;
+}
+
+
+// TODO(jingning,jimbankoski,rbultje): properly skip partition types that are
+// unlikely to be selected depending on previously rate-distortion optimization
+// results, for encoding speed-up.
+static void rd_pick_partition(VP9_COMP *cpi, TOKENEXTRA **tp,
+                              int mi_row, int mi_col,
+                              BLOCK_SIZE_TYPE bsize,
+                              int *rate, int *dist) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int bsl = b_width_log2(bsize), bs = 1 << bsl;
+  int ms = bs / 2;
+  ENTROPY_CONTEXT   l[16 * MAX_MB_PLANE], a[16 * MAX_MB_PLANE];
+  PARTITION_CONTEXT sl[8], sa[8];
+  TOKENEXTRA *tp_orig = *tp;
+  int i, pl;
+  BLOCK_SIZE_TYPE subsize;
+  int srate = INT_MAX, sdist = INT_MAX;
+
+  if (bsize < BLOCK_SIZE_SB8X8)
+    if (xd->ab_index != 0) {
+      *rate = 0;
+      *dist = 0;
+      return;
+    }
+  assert(mi_height_log2(bsize) == mi_width_log2(bsize));
+
+  save_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+
+  // PARTITION_SPLIT
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    int r4 = 0, d4 = 0;
+    subsize = get_subsize(bsize, PARTITION_SPLIT);
+    *(get_sb_partitioning(x, bsize)) = subsize;
+
+    for (i = 0; i < 4; ++i) {
+      int x_idx = (i & 1) * (ms >> 1);
+      int y_idx = (i >> 1) * (ms >> 1);
+      int r = 0, d = 0;
+
+      if ((mi_row + y_idx >= cm->mi_rows) || (mi_col + x_idx >= cm->mi_cols))
+        continue;
+
+      *(get_sb_index(xd, subsize)) = i;
+      rd_pick_partition(cpi, tp, mi_row + y_idx, mi_col + x_idx, subsize,
+                        &r, &d);
+
+      r4 += r;
+      d4 += d;
+    }
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    pl = partition_plane_context(xd, bsize);
+    if (r4 < INT_MAX)
+      r4 += x->partition_cost[pl][PARTITION_SPLIT];
+    assert(r4 >= 0);
+    assert(d4 >= 0);
+    srate = r4;
+    sdist = d4;
+    restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+  }
+
+  // PARTITION_HORZ
+  if (bsize >= BLOCK_SIZE_SB8X8 && mi_col + (ms >> 1) < cm->mi_cols) {
+    int r2, d2;
+    int r = 0, d = 0;
+    subsize = get_subsize(bsize, PARTITION_HORZ);
+    *(get_sb_index(xd, subsize)) = 0;
+    pick_sb_modes(cpi, mi_row, mi_col, tp, &r2, &d2, subsize,
+                  get_block_context(x, subsize));
+
+    if (mi_row + (ms >> 1) < cm->mi_rows) {
+      update_state(cpi, get_block_context(x, subsize), subsize, 0);
+      encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
+
+      *(get_sb_index(xd, subsize)) = 1;
+      pick_sb_modes(cpi, mi_row + (ms >> 1), mi_col, tp, &r, &d, subsize,
+                    get_block_context(x, subsize));
+      r2 += r;
+      d2 += d;
+    }
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    pl = partition_plane_context(xd, bsize);
+    if (r2 < INT_MAX)
+      r2 += x->partition_cost[pl][PARTITION_HORZ];
+    if (RDCOST(x->rdmult, x->rddiv, r2, d2) <
+        RDCOST(x->rdmult, x->rddiv, srate, sdist)) {
+      srate = r2;
+      sdist = d2;
+      *(get_sb_partitioning(x, bsize)) = subsize;
+    }
+    restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+  }
+
+  // PARTITION_VERT
+  if (bsize >= BLOCK_SIZE_SB8X8 && mi_row + (ms >> 1) < cm->mi_rows) {
+    int r2, d2;
+    subsize = get_subsize(bsize, PARTITION_VERT);
+    *(get_sb_index(xd, subsize)) = 0;
+    pick_sb_modes(cpi, mi_row, mi_col, tp, &r2, &d2, subsize,
+                  get_block_context(x, subsize));
+    if (mi_col + (ms >> 1) < cm->mi_cols) {
+      int r = 0, d = 0;
+      update_state(cpi, get_block_context(x, subsize), subsize, 0);
+      encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
+
+      *(get_sb_index(xd, subsize)) = 1;
+      pick_sb_modes(cpi, mi_row, mi_col + (ms >> 1), tp, &r, &d, subsize,
+                    get_block_context(x, subsize));
+      r2 += r;
+      d2 += d;
+    }
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
+    pl = partition_plane_context(xd, bsize);
+    if (r2 < INT_MAX)
+      r2 += x->partition_cost[pl][PARTITION_VERT];
+    if (RDCOST(x->rdmult, x->rddiv, r2, d2) <
+        RDCOST(x->rdmult, x->rddiv, srate, sdist)) {
+      srate = r2;
+      sdist = d2;
+      *(get_sb_partitioning(x, bsize)) = subsize;
+    }
+    restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+  }
+
+  // PARTITION_NONE
+  if ((mi_row + (ms >> 1) < cm->mi_rows) &&
+      (mi_col + (ms >> 1) < cm->mi_cols)) {
+    int r, d;
+    pick_sb_modes(cpi, mi_row, mi_col, tp, &r, &d, bsize,
+                  get_block_context(x, bsize));
+    if (bsize >= BLOCK_SIZE_SB8X8) {
+      set_partition_seg_context(cm, xd, mi_row, mi_col);
+      pl = partition_plane_context(xd, bsize);
+      r += x->partition_cost[pl][PARTITION_NONE];
+    }
+
+    if (RDCOST(x->rdmult, x->rddiv, r, d) <
+        RDCOST(x->rdmult, x->rddiv, srate, sdist)) {
+      srate = r;
+      sdist = d;
+      if (bsize >= BLOCK_SIZE_SB8X8)
+        *(get_sb_partitioning(x, bsize)) = bsize;
+    }
+  }
+
+  *rate = srate;
+  *dist = sdist;
+
+  restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
+
+  if (srate < INT_MAX && sdist < INT_MAX)
+    encode_sb(cpi, tp, mi_row, mi_col, bsize == BLOCK_SIZE_SB64X64, bsize);
+
+  if (bsize == BLOCK_SIZE_SB64X64) {
+    assert(tp_orig < *tp);
+    assert(srate < INT_MAX);
+    assert(sdist < INT_MAX);
+  } else {
+    assert(tp_orig == *tp);
+  }
+}
+
+static void encode_sb_row(VP9_COMP *cpi, int mi_row,
+                       TOKENEXTRA **tp, int *totalrate) {
+  VP9_COMMON *const cm = &cpi->common;
+  int mi_col;
+
+  // Initialize the left context for the new SB row
+  vpx_memset(&cm->left_context, 0, sizeof(cm->left_context));
+  vpx_memset(cm->left_seg_context, 0, sizeof(cm->left_seg_context));
+
+  // Code each SB in the row
+  for (mi_col = cm->cur_tile_mi_col_start;
+       mi_col < cm->cur_tile_mi_col_end; mi_col += 64 / MI_SIZE) {
+    int dummy_rate, dummy_dist;
+    if (cpi->speed < 5) {
+      rd_pick_partition(cpi, tp, mi_row, mi_col, BLOCK_SIZE_SB64X64,
+                        &dummy_rate, &dummy_dist);
+    } else {
+      const int idx_str = cm->mode_info_stride * mi_row + mi_col;
+      MODE_INFO *m = cm->mi + idx_str;
+      // set_partitioning(cpi, m, BLOCK_SIZE_SB64X64);
+      choose_partitioning(cpi, cm->mi, mi_row, mi_col);
+      rd_use_partition(cpi, m, tp, mi_row, mi_col, BLOCK_SIZE_SB64X64,
+                       &dummy_rate, &dummy_dist);
+    }
+  }
+}
+
+static void init_encode_frame_mb_context(VP9_COMP *cpi) {
+  MACROBLOCK *const x = &cpi->mb;
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  x->act_zbin_adj = 0;
+  cpi->seg0_idx = 0;
+
+  xd->mode_info_stride = cm->mode_info_stride;
+  xd->frame_type = cm->frame_type;
+
+  xd->frames_since_golden = cm->frames_since_golden;
+  xd->frames_till_alt_ref_frame = cm->frames_till_alt_ref_frame;
+
+  // reset intra mode contexts
+  if (cm->frame_type == KEY_FRAME)
+    vp9_init_mbmode_probs(cm);
+
+  // Copy data over into macro block data structures.
+  vp9_setup_src_planes(x, cpi->Source, 0, 0);
+
+  // TODO(jkoleszar): are these initializations required?
+  setup_pre_planes(xd, &cm->yv12_fb[cm->ref_frame_map[cpi->lst_fb_idx]], NULL,
+                   0, 0, NULL, NULL);
+  setup_dst_planes(xd, &cm->yv12_fb[cm->new_fb_idx], 0, 0);
+
+  vp9_build_block_offsets(x);
+
+  vp9_setup_block_dptrs(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
+
+  xd->mode_info_context->mbmi.mode = DC_PRED;
+  xd->mode_info_context->mbmi.uv_mode = DC_PRED;
+
+  vp9_zero(cpi->y_mode_count)
+  vp9_zero(cpi->y_uv_mode_count)
+  vp9_zero(cm->fc.inter_mode_counts)
+  vp9_zero(cpi->partition_count);
+  vp9_zero(cpi->intra_inter_count);
+  vp9_zero(cpi->comp_inter_count);
+  vp9_zero(cpi->single_ref_count);
+  vp9_zero(cpi->comp_ref_count);
+  vp9_zero(cm->fc.tx_count_32x32p);
+  vp9_zero(cm->fc.tx_count_16x16p);
+  vp9_zero(cm->fc.tx_count_8x8p);
+  vp9_zero(cm->fc.mbskip_count);
+
+  // Note: this memset assumes above_context[0], [1] and [2]
+  // are allocated as part of the same buffer.
+  vpx_memset(cm->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 2 *
+                                      MAX_MB_PLANE * mi_cols_aligned_to_sb(cm));
+  vpx_memset(cm->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
+                                       mi_cols_aligned_to_sb(cm));
+}
+
+static void switch_lossless_mode(VP9_COMP *cpi, int lossless) {
+  if (lossless) {
+    cpi->mb.fwd_txm8x4            = vp9_short_walsh8x4;
+    cpi->mb.fwd_txm4x4            = vp9_short_walsh4x4;
+    cpi->mb.e_mbd.inv_txm4x4_1_add    = vp9_short_iwalsh4x4_1_add;
+    cpi->mb.e_mbd.inv_txm4x4_add      = vp9_short_iwalsh4x4_add;
+    cpi->mb.optimize              = 0;
+    cpi->common.filter_level      = 0;
+    cpi->zbin_mode_boost_enabled  = 0;
+    cpi->common.txfm_mode         = ONLY_4X4;
+  } else {
+    cpi->mb.fwd_txm8x4            = vp9_short_fdct8x4;
+    cpi->mb.fwd_txm4x4            = vp9_short_fdct4x4;
+    cpi->mb.e_mbd.inv_txm4x4_1_add    = vp9_short_idct4x4_1_add;
+    cpi->mb.e_mbd.inv_txm4x4_add      = vp9_short_idct4x4_add;
+  }
+}
+
+
+static void encode_frame_internal(VP9_COMP *cpi) {
+  int mi_row;
+  MACROBLOCK *const x = &cpi->mb;
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int totalrate;
+
+//  fprintf(stderr, "encode_frame_internal frame %d (%d) type %d\n",
+//           cpi->common.current_video_frame, cpi->common.show_frame,
+//           cm->frame_type);
+
+// debug output
+#if DBG_PRNT_SEGMAP
+  {
+    FILE *statsfile;
+    statsfile = fopen("segmap2.stt", "a");
+    fprintf(statsfile, "\n");
+    fclose(statsfile);
+  }
+#endif
+
+  totalrate = 0;
+
+  // Reset frame count of inter 0,0 motion vector usage.
+  cpi->inter_zz_count = 0;
+
+  vp9_zero(cm->fc.switchable_interp_count);
+  vp9_zero(cpi->best_switchable_interp_count);
+
+  xd->mode_info_context = cm->mi;
+  xd->prev_mode_info_context = cm->prev_mi;
+
+  vp9_zero(cpi->NMVcount);
+  vp9_zero(cpi->coef_counts);
+  vp9_zero(cm->fc.eob_branch_counts);
+
+  cpi->mb.e_mbd.lossless = cm->base_qindex == 0 &&
+                           cm->y_dc_delta_q == 0 &&
+                           cm->uv_dc_delta_q == 0 &&
+                           cm->uv_ac_delta_q == 0;
+  switch_lossless_mode(cpi, cpi->mb.e_mbd.lossless);
+
+  vp9_frame_init_quantizer(cpi);
+
+  vp9_initialize_rd_consts(cpi, cm->base_qindex + cm->y_dc_delta_q);
+  vp9_initialize_me_consts(cpi, cm->base_qindex);
+
+  if (cpi->oxcf.tuning == VP8_TUNE_SSIM) {
+    // Initialize encode frame context.
+    init_encode_frame_mb_context(cpi);
+
+    // Build a frame level activity map
+    build_activity_map(cpi);
+  }
+
+  // re-initencode frame context.
+  init_encode_frame_mb_context(cpi);
+
+  vpx_memset(cpi->rd_comp_pred_diff, 0, sizeof(cpi->rd_comp_pred_diff));
+  vpx_memset(cpi->rd_tx_select_diff, 0, sizeof(cpi->rd_tx_select_diff));
+  vpx_memset(cpi->rd_tx_select_threshes, 0, sizeof(cpi->rd_tx_select_threshes));
+
+  set_prev_mi(cm);
+
+  {
+    struct vpx_usec_timer  emr_timer;
+    vpx_usec_timer_start(&emr_timer);
+
+    {
+      // Take tiles into account and give start/end MB
+      int tile_col, tile_row;
+      TOKENEXTRA *tp = cpi->tok;
+
+      for (tile_row = 0; tile_row < cm->tile_rows; tile_row++) {
+        vp9_get_tile_row_offsets(cm, tile_row);
+
+        for (tile_col = 0; tile_col < cm->tile_columns; tile_col++) {
+          TOKENEXTRA *tp_old = tp;
+
+          // For each row of SBs in the frame
+          vp9_get_tile_col_offsets(cm, tile_col);
+          for (mi_row = cm->cur_tile_mi_row_start;
+               mi_row < cm->cur_tile_mi_row_end;
+               mi_row += 8)
+            encode_sb_row(cpi, mi_row, &tp, &totalrate);
+          cpi->tok_count[tile_row][tile_col] = (unsigned int)(tp - tp_old);
+          assert(tp - cpi->tok <=
+                 get_token_alloc(cm->mb_rows, cm->mb_cols));
+        }
+      }
+    }
+
+    vpx_usec_timer_mark(&emr_timer);
+    cpi->time_encode_mb_row += vpx_usec_timer_elapsed(&emr_timer);
+  }
+
+  // 256 rate units to the bit,
+  // projected_frame_size in units of BYTES
+  cpi->projected_frame_size = totalrate >> 8;
+
+#if 0
+  // Keep record of the total distortion this time around for future use
+  cpi->last_frame_distortion = cpi->frame_distortion;
+#endif
+
+}
+
+static int check_dual_ref_flags(VP9_COMP *cpi) {
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  int ref_flags = cpi->ref_frame_flags;
+
+  if (vp9_segfeature_active(xd, 1, SEG_LVL_REF_FRAME)) {
+    return 0;
+  } else {
+    return (!!(ref_flags & VP9_GOLD_FLAG) +
+            !!(ref_flags & VP9_LAST_FLAG) +
+            !!(ref_flags & VP9_ALT_FLAG)) >= 2;
+  }
+}
+
+static int get_skip_flag(MODE_INFO *mi, int mis, int ymbs, int xmbs) {
+  int x, y;
+
+  for (y = 0; y < ymbs; y++) {
+    for (x = 0; x < xmbs; x++) {
+      if (!mi[y * mis + x].mbmi.mb_skip_coeff)
+        return 0;
+    }
+  }
+
+  return 1;
+}
+
+static void set_txfm_flag(MODE_INFO *mi, int mis, int ymbs, int xmbs,
+                          TX_SIZE txfm_size) {
+  int x, y;
+
+  for (y = 0; y < ymbs; y++) {
+    for (x = 0; x < xmbs; x++)
+      mi[y * mis + x].mbmi.txfm_size = txfm_size;
+  }
+}
+
+static void reset_skip_txfm_size_b(VP9_COMP *cpi, MODE_INFO *mi,
+                                   int mis, TX_SIZE txfm_max,
+                                   int bw, int bh, int mi_row, int mi_col,
+                                   BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  if (mbmi->txfm_size > txfm_max) {
+    MACROBLOCK *const x = &cpi->mb;
+    MACROBLOCKD *const xd = &x->e_mbd;
+    const int segment_id = mbmi->segment_id;
+    const int ymbs = MIN(bh, cm->mi_rows - mi_row);
+    const int xmbs = MIN(bw, cm->mi_cols - mi_col);
+
+    xd->mode_info_context = mi;
+    assert(vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) ||
+           get_skip_flag(mi, mis, ymbs, xmbs));
+    set_txfm_flag(mi, mis, ymbs, xmbs, txfm_max);
+  }
+}
+
+static void reset_skip_txfm_size_sb(VP9_COMP *cpi, MODE_INFO *mi,
+                                    TX_SIZE txfm_max,
+                                    int mi_row, int mi_col,
+                                    BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  const int mis = cm->mode_info_stride;
+  int bwl, bhl;
+  const int bsl = mi_width_log2(bsize), bs = 1 << (bsl - 1);
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  bwl = mi_width_log2(mi->mbmi.sb_type);
+  bhl = mi_height_log2(mi->mbmi.sb_type);
+
+  if (bwl == bsl && bhl == bsl) {
+    reset_skip_txfm_size_b(cpi, mi, mis, txfm_max, 1 << bsl, 1 << bsl,
+                           mi_row, mi_col, bsize);
+  } else if (bwl == bsl && bhl < bsl) {
+    reset_skip_txfm_size_b(cpi, mi, mis, txfm_max, 1 << bsl, bs,
+                           mi_row, mi_col, bsize);
+    reset_skip_txfm_size_b(cpi, mi + bs * mis, mis, txfm_max, 1 << bsl, bs,
+                           mi_row + bs, mi_col, bsize);
+  } else if (bwl < bsl && bhl == bsl) {
+    reset_skip_txfm_size_b(cpi, mi, mis, txfm_max, bs, 1 << bsl,
+                           mi_row, mi_col, bsize);
+    reset_skip_txfm_size_b(cpi, mi + bs, mis, txfm_max, bs, 1 << bsl,
+                           mi_row, mi_col + bs, bsize);
+  } else {
+    BLOCK_SIZE_TYPE subsize;
+    int n;
+
+    assert(bwl < bsl && bhl < bsl);
+    if (bsize == BLOCK_SIZE_SB64X64) {
+      subsize = BLOCK_SIZE_SB32X32;
+    } else if (bsize == BLOCK_SIZE_SB32X32) {
+      subsize = BLOCK_SIZE_MB16X16;
+    } else {
+      assert(bsize == BLOCK_SIZE_MB16X16);
+      subsize = BLOCK_SIZE_SB8X8;
+    }
+
+    for (n = 0; n < 4; n++) {
+      const int y_idx = n >> 1, x_idx = n & 0x01;
+
+      reset_skip_txfm_size_sb(cpi, mi + y_idx * bs * mis + x_idx * bs,
+                              txfm_max, mi_row + y_idx * bs,
+                              mi_col + x_idx * bs, subsize);
+    }
+  }
+}
+
+static void reset_skip_txfm_size(VP9_COMP *cpi, TX_SIZE txfm_max) {
+  VP9_COMMON *const cm = &cpi->common;
+  int mi_row, mi_col;
+  const int mis = cm->mode_info_stride;
+  MODE_INFO *mi, *mi_ptr = cm->mi;
+
+  for (mi_row = 0; mi_row < cm->mi_rows;
+       mi_row += 8, mi_ptr += 8 * mis) {
+    mi = mi_ptr;
+    for (mi_col = 0; mi_col < cm->mi_cols;
+         mi_col += 8, mi += 8) {
+      reset_skip_txfm_size_sb(cpi, mi, txfm_max,
+                              mi_row, mi_col, BLOCK_SIZE_SB64X64);
+    }
+  }
+}
+
+void vp9_encode_frame(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+
+  // In the longer term the encoder should be generalized to match the
+  // decoder such that we allow compound where one of the 3 buffers has a
+  // differnt sign bias and that buffer is then the fixed ref. However, this
+  // requires further work in the rd loop. For now the only supported encoder
+  // side behaviour is where the ALT ref buffer has oppositie sign bias to
+  // the other two.
+  if ((cm->ref_frame_sign_bias[ALTREF_FRAME] ==
+       cm->ref_frame_sign_bias[GOLDEN_FRAME]) ||
+      (cm->ref_frame_sign_bias[ALTREF_FRAME] ==
+       cm->ref_frame_sign_bias[LAST_FRAME])) {
+    cm->allow_comp_inter_inter = 0;
+  } else {
+    cm->allow_comp_inter_inter = 1;
+    cm->comp_fixed_ref = ALTREF_FRAME;
+    cm->comp_var_ref[0] = LAST_FRAME;
+    cm->comp_var_ref[1] = GOLDEN_FRAME;
+  }
+
+  if (cpi->sf.RD) {
+    int i, frame_type, pred_type;
+    TXFM_MODE txfm_type;
+
+    /*
+     * This code does a single RD pass over the whole frame assuming
+     * either compound, single or hybrid prediction as per whatever has
+     * worked best for that type of frame in the past.
+     * It also predicts whether another coding mode would have worked
+     * better that this coding mode. If that is the case, it remembers
+     * that for subsequent frames.
+     * It does the same analysis for transform size selection also.
+     */
+    if (cpi->common.frame_type == KEY_FRAME)
+      frame_type = 0;
+    else if (cpi->is_src_frame_alt_ref && cpi->refresh_golden_frame)
+      frame_type = 3;
+    else if (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)
+      frame_type = 1;
+    else
+      frame_type = 2;
+
+    /* prediction (compound, single or hybrid) mode selection */
+    if (frame_type == 3 || !cm->allow_comp_inter_inter)
+      pred_type = SINGLE_PREDICTION_ONLY;
+    else if (cpi->rd_prediction_type_threshes[frame_type][1] >
+                 cpi->rd_prediction_type_threshes[frame_type][0] &&
+             cpi->rd_prediction_type_threshes[frame_type][1] >
+                 cpi->rd_prediction_type_threshes[frame_type][2] &&
+             check_dual_ref_flags(cpi) && cpi->static_mb_pct == 100)
+      pred_type = COMP_PREDICTION_ONLY;
+    else if (cpi->rd_prediction_type_threshes[frame_type][0] >
+                 cpi->rd_prediction_type_threshes[frame_type][2])
+      pred_type = SINGLE_PREDICTION_ONLY;
+    else
+      pred_type = HYBRID_PREDICTION;
+
+    /* transform size (4x4, 8x8, 16x16 or select-per-mb) selection */
+
+    cpi->mb.e_mbd.lossless = 0;
+    if (cpi->oxcf.lossless) {
+      txfm_type = ONLY_4X4;
+      cpi->mb.e_mbd.lossless = 1;
+    } else
+#if 0
+    /* FIXME (rbultje): this code is disabled until we support cost updates
+     * while a frame is being encoded; the problem is that each time we
+     * "revert" to 4x4 only (or even 8x8 only), the coefficient probabilities
+     * for 16x16 (and 8x8) start lagging behind, thus leading to them lagging
+     * further behind and not being chosen for subsequent frames either. This
+     * is essentially a local minimum problem that we can probably fix by
+     * estimating real costs more closely within a frame, perhaps by re-
+     * calculating costs on-the-fly as frame encoding progresses. */
+    if (cpi->rd_tx_select_threshes[frame_type][TX_MODE_SELECT] >
+            cpi->rd_tx_select_threshes[frame_type][ONLY_4X4] &&
+        cpi->rd_tx_select_threshes[frame_type][TX_MODE_SELECT] >
+            cpi->rd_tx_select_threshes[frame_type][ALLOW_16X16] &&
+        cpi->rd_tx_select_threshes[frame_type][TX_MODE_SELECT] >
+            cpi->rd_tx_select_threshes[frame_type][ALLOW_8X8]) {
+      txfm_type = TX_MODE_SELECT;
+    } else if (cpi->rd_tx_select_threshes[frame_type][ONLY_4X4] >
+                  cpi->rd_tx_select_threshes[frame_type][ALLOW_8X8]
+            && cpi->rd_tx_select_threshes[frame_type][ONLY_4X4] >
+                  cpi->rd_tx_select_threshes[frame_type][ALLOW_16X16]
+               ) {
+      txfm_type = ONLY_4X4;
+    } else if (cpi->rd_tx_select_threshes[frame_type][ALLOW_16X16] >=
+                  cpi->rd_tx_select_threshes[frame_type][ALLOW_8X8]) {
+      txfm_type = ALLOW_16X16;
+    } else
+      txfm_type = ALLOW_8X8;
+#else
+    txfm_type = cpi->rd_tx_select_threshes[frame_type][ALLOW_32X32] >
+                  cpi->rd_tx_select_threshes[frame_type][TX_MODE_SELECT] ?
+                    ALLOW_32X32 : TX_MODE_SELECT;
+#endif
+    cpi->common.txfm_mode = txfm_type;
+    cpi->common.comp_pred_mode = pred_type;
+    encode_frame_internal(cpi);
+
+    for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
+      const int diff = (int)(cpi->rd_comp_pred_diff[i] / cpi->common.MBs);
+      cpi->rd_prediction_type_threshes[frame_type][i] += diff;
+      cpi->rd_prediction_type_threshes[frame_type][i] >>= 1;
+    }
+
+    for (i = 0; i < NB_TXFM_MODES; ++i) {
+      int64_t pd = cpi->rd_tx_select_diff[i];
+      int diff;
+      if (i == TX_MODE_SELECT)
+        pd -= RDCOST(cpi->mb.rdmult, cpi->mb.rddiv,
+                     2048 * (TX_SIZE_MAX_SB - 1), 0);
+      diff = (int)(pd / cpi->common.MBs);
+      cpi->rd_tx_select_threshes[frame_type][i] += diff;
+      cpi->rd_tx_select_threshes[frame_type][i] /= 2;
+    }
+
+    if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+      int single_count_zero = 0;
+      int comp_count_zero = 0;
+
+      for (i = 0; i < COMP_INTER_CONTEXTS; i++) {
+        single_count_zero += cpi->comp_inter_count[i][0];
+        comp_count_zero += cpi->comp_inter_count[i][1];
+      }
+
+      if (comp_count_zero == 0) {
+        cpi->common.comp_pred_mode = SINGLE_PREDICTION_ONLY;
+        vp9_zero(cpi->comp_inter_count);
+      } else if (single_count_zero == 0) {
+        cpi->common.comp_pred_mode = COMP_PREDICTION_ONLY;
+        vp9_zero(cpi->comp_inter_count);
+      }
+    }
+
+    if (cpi->common.txfm_mode == TX_MODE_SELECT) {
+      int count4x4 = 0;
+      int count8x8_lp = 0, count8x8_8x8p = 0;
+      int count16x16_16x16p = 0, count16x16_lp = 0;
+      int count32x32 = 0;
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count4x4 += cm->fc.tx_count_32x32p[i][TX_4X4];
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count4x4 += cm->fc.tx_count_16x16p[i][TX_4X4];
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count4x4 += cm->fc.tx_count_8x8p[i][TX_4X4];
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count8x8_lp += cm->fc.tx_count_32x32p[i][TX_8X8];
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count8x8_lp += cm->fc.tx_count_16x16p[i][TX_8X8];
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count8x8_8x8p += cm->fc.tx_count_8x8p[i][TX_8X8];
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count16x16_16x16p += cm->fc.tx_count_16x16p[i][TX_16X16];
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count16x16_lp += cm->fc.tx_count_32x32p[i][TX_16X16];
+
+      for (i = 0; i < TX_SIZE_CONTEXTS; i++)
+        count32x32 += cm->fc.tx_count_32x32p[i][TX_32X32];
+
+      if (count4x4 == 0 && count16x16_lp == 0 && count16x16_16x16p == 0 &&
+          count32x32 == 0) {
+        cpi->common.txfm_mode = ALLOW_8X8;
+        reset_skip_txfm_size(cpi, TX_8X8);
+      } else if (count8x8_8x8p == 0 && count16x16_16x16p == 0 &&
+                 count8x8_lp == 0 && count16x16_lp == 0 && count32x32 == 0) {
+        cpi->common.txfm_mode = ONLY_4X4;
+        reset_skip_txfm_size(cpi, TX_4X4);
+      } else if (count8x8_lp == 0 && count16x16_lp == 0 && count4x4 == 0) {
+        cpi->common.txfm_mode = ALLOW_32X32;
+      } else if (count32x32 == 0 && count8x8_lp == 0 && count4x4 == 0) {
+        cpi->common.txfm_mode = ALLOW_16X16;
+        reset_skip_txfm_size(cpi, TX_16X16);
+      }
+    }
+
+    // Update interpolation filter strategy for next frame.
+    if ((cpi->common.frame_type != KEY_FRAME) && (cpi->sf.search_best_filter))
+      vp9_select_interp_filter_type(cpi);
+  } else {
+    encode_frame_internal(cpi);
+  }
+
+}
+
+void vp9_build_block_offsets(MACROBLOCK *x) {
+}
+
+static void sum_intra_stats(VP9_COMP *cpi, MACROBLOCK *x) {
+  const MACROBLOCKD *xd = &x->e_mbd;
+  const MB_PREDICTION_MODE m = xd->mode_info_context->mbmi.mode;
+  const MB_PREDICTION_MODE uvm = xd->mode_info_context->mbmi.uv_mode;
+
+  ++cpi->y_uv_mode_count[m][uvm];
+  if (xd->mode_info_context->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+    const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+    const int bsl = MIN(bwl, bhl);
+    ++cpi->y_mode_count[MIN(bsl, 3)][m];
+  } else {
+    int idx, idy;
+    int bw = 1 << b_width_log2(xd->mode_info_context->mbmi.sb_type);
+    int bh = 1 << b_height_log2(xd->mode_info_context->mbmi.sb_type);
+    for (idy = 0; idy < 2; idy += bh) {
+      for (idx = 0; idx < 2; idx += bw) {
+        int m = xd->mode_info_context->bmi[idy * 2 + idx].as_mode.first;
+        ++cpi->y_mode_count[0][m];
+      }
+    }
+  }
+}
+
+// Experimental stub function to create a per MB zbin adjustment based on
+// some previously calculated measure of MB activity.
+static void adjust_act_zbin(VP9_COMP *cpi, MACROBLOCK *x) {
+#if USE_ACT_INDEX
+  x->act_zbin_adj = *(x->mb_activity_ptr);
+#else
+  int64_t a;
+  int64_t b;
+  int64_t act = *(x->mb_activity_ptr);
+
+  // Apply the masking to the RD multiplier.
+  a = act + 4 * cpi->activity_avg;
+  b = 4 * act + cpi->activity_avg;
+
+  if (act > cpi->activity_avg)
+    x->act_zbin_adj = (int)(((int64_t)b + (a >> 1)) / a) - 1;
+  else
+    x->act_zbin_adj = 1 - (int)(((int64_t)a + (b >> 1)) / b);
+#endif
+}
+
+static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
+                              int output_enabled, int mi_row, int mi_col,
+                              BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int n;
+  MODE_INFO *mi = xd->mode_info_context;
+  MB_MODE_INFO *mbmi = &mi->mbmi;
+  unsigned int segment_id = mbmi->segment_id;
+  const int mis = cm->mode_info_stride;
+  const int bwl = mi_width_log2(bsize);
+  const int bw = 1 << bwl, bh = 1 << mi_height_log2(bsize);
+  x->rd_search = 0;
+
+  if (cm->frame_type == KEY_FRAME) {
+    if (cpi->oxcf.tuning == VP8_TUNE_SSIM) {
+      adjust_act_zbin(cpi, x);
+      vp9_update_zbin_extra(cpi, x);
+    }
+  } else {
+    vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
+
+    if (cpi->oxcf.tuning == VP8_TUNE_SSIM) {
+      // Adjust the zbin based on this MB rate.
+      adjust_act_zbin(cpi, x);
+    }
+
+    // Experimental code. Special case for gf and arf zeromv modes.
+    // Increase zbin size to suppress noise
+    cpi->zbin_mode_boost = 0;
+    if (cpi->zbin_mode_boost_enabled) {
+      if (mbmi->ref_frame[0] != INTRA_FRAME) {
+        if (mbmi->mode == ZEROMV) {
+          if (mbmi->ref_frame[0] != LAST_FRAME)
+            cpi->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
+          else
+            cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
+        } else if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+          cpi->zbin_mode_boost = SPLIT_MV_ZBIN_BOOST;
+        } else {
+          cpi->zbin_mode_boost = MV_ZBIN_BOOST;
+        }
+      } else {
+        cpi->zbin_mode_boost = INTRA_ZBIN_BOOST;
+      }
+    }
+
+    vp9_update_zbin_extra(cpi, x);
+  }
+
+  if (mbmi->ref_frame[0] == INTRA_FRAME) {
+    vp9_encode_intra_block_y(cm, x, (bsize < BLOCK_SIZE_SB8X8) ?
+                                    BLOCK_SIZE_SB8X8 : bsize);
+    vp9_encode_intra_block_uv(cm, x, (bsize < BLOCK_SIZE_SB8X8) ?
+                                     BLOCK_SIZE_SB8X8 : bsize);
+    if (output_enabled)
+      sum_intra_stats(cpi, x);
+  } else {
+    int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, mbmi->ref_frame[0])];
+    YV12_BUFFER_CONFIG *ref_fb = &cm->yv12_fb[idx];
+    YV12_BUFFER_CONFIG *second_ref_fb = NULL;
+    if (mbmi->ref_frame[1] > 0) {
+      idx = cm->ref_frame_map[get_ref_frame_idx(cpi, mbmi->ref_frame[1])];
+      second_ref_fb = &cm->yv12_fb[idx];
+    }
+
+    assert(cm->frame_type != KEY_FRAME);
+
+    setup_pre_planes(xd, ref_fb, second_ref_fb,
+                     mi_row, mi_col, xd->scale_factor, xd->scale_factor_uv);
+
+    vp9_build_inter_predictors_sb(xd, mi_row, mi_col,
+                                  bsize < BLOCK_SIZE_SB8X8 ? BLOCK_SIZE_SB8X8
+                                                           : bsize);
+  }
+
+  if (xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME) {
+    vp9_tokenize_sb(cpi, xd, t, !output_enabled,
+                    (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 : bsize);
+  } else if (!x->skip) {
+    vp9_encode_sb(cm, x, (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 : bsize);
+    vp9_tokenize_sb(cpi, xd, t, !output_enabled,
+                    (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 : bsize);
+  } else {
+    // FIXME(rbultje): not tile-aware (mi - 1)
+    int mb_skip_context =
+        (mi - 1)->mbmi.mb_skip_coeff + (mi - mis)->mbmi.mb_skip_coeff;
+
+    mbmi->mb_skip_coeff = 1;
+    if (output_enabled)
+      cm->fc.mbskip_count[mb_skip_context][1]++;
+    vp9_reset_sb_tokens_context(xd,
+                 (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 : bsize);
+  }
+
+  // copy skip flag on all mb_mode_info contexts in this SB
+  // if this was a skip at this txfm size
+  for (n = 1; n < bw * bh; n++) {
+    const int x_idx = n & (bw - 1), y_idx = n >> bwl;
+    if (mi_col + x_idx < cm->mi_cols && mi_row + y_idx < cm->mi_rows)
+      mi[x_idx + y_idx * mis].mbmi.mb_skip_coeff = mi->mbmi.mb_skip_coeff;
+  }
+
+  if (output_enabled) {
+    if (cm->txfm_mode == TX_MODE_SELECT &&
+        mbmi->sb_type >= BLOCK_SIZE_SB8X8 &&
+        !(mbmi->ref_frame[0] != INTRA_FRAME && (mbmi->mb_skip_coeff ||
+          vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)))) {
+      const int context = vp9_get_pred_context(cm, xd, PRED_TX_SIZE);
+      if (bsize >= BLOCK_SIZE_SB32X32) {
+        cm->fc.tx_count_32x32p[context][mbmi->txfm_size]++;
+      } else if (bsize >= BLOCK_SIZE_MB16X16) {
+        cm->fc.tx_count_16x16p[context][mbmi->txfm_size]++;
+      } else {
+        cm->fc.tx_count_8x8p[context][mbmi->txfm_size]++;
+      }
+    } else {
+      int x, y;
+      TX_SIZE sz = (cm->txfm_mode == TX_MODE_SELECT) ? TX_32X32 : cm->txfm_mode;
+       // The new intra coding scheme requires no change of transform size
+      if (mi->mbmi.ref_frame[0] != INTRA_FRAME) {
+        if (sz == TX_32X32 && bsize < BLOCK_SIZE_SB32X32)
+          sz = TX_16X16;
+        if (sz == TX_16X16 && bsize < BLOCK_SIZE_MB16X16)
+          sz = TX_8X8;
+        if (sz == TX_8X8 && bsize < BLOCK_SIZE_SB8X8)
+          sz = TX_4X4;
+      } else if (bsize >= BLOCK_SIZE_SB8X8) {
+        sz = mbmi->txfm_size;
+      } else {
+        sz = TX_4X4;
+      }
+
+      for (y = 0; y < bh; y++) {
+        for (x = 0; x < bw; x++) {
+          if (mi_col + x < cm->mi_cols && mi_row + y < cm->mi_rows) {
+            mi[mis * y + x].mbmi.txfm_size = sz;
+          }
+        }
+      }
+    }
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_encodeframe.h b/libvpx/vp9/encoder/vp9_encodeframe.h
new file mode 100644
index 0000000..d37bdca
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodeframe.h
@@ -0,0 +1,24 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_ENCODEFRAME_H_
+#define VP9_ENCODER_VP9_ENCODEFRAME_H_
+
+struct macroblock;
+struct yv12_buffer_config;
+
+void vp9_build_block_offsets(struct macroblock *x);
+
+void vp9_setup_src_planes(struct macroblock *x,
+                          const struct yv12_buffer_config *src,
+                          int mb_row, int mb_col);
+
+#endif  // VP9_ENCODER_VP9_ENCODEFRAME_H_
diff --git a/libvpx/vp9/encoder/vp9_encodeintra.c b/libvpx/vp9/encoder/vp9_encodeintra.c
new file mode 100644
index 0000000..f29dba0
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodeintra.c
@@ -0,0 +1,32 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "vp9_rtcd.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/encoder/vp9_encodeintra.h"
+
+int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred) {
+  MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
+  (void) cpi;
+  mbmi->mode = DC_PRED;
+  mbmi->ref_frame[0] = INTRA_FRAME;
+  if (use_16x16_pred) {
+    mbmi->txfm_size = mbmi->sb_type >= BLOCK_SIZE_MB16X16 ? TX_16X16 : TX_8X8;
+    vp9_encode_intra_block_y(&cpi->common, x, mbmi->sb_type);
+  } else {
+    mbmi->txfm_size = TX_4X4;
+    vp9_encode_intra_block_y(&cpi->common, x, mbmi->sb_type);
+  }
+
+  return vp9_get_mb_ss(x->plane[0].src_diff);
+}
diff --git a/libvpx/vp9/encoder/vp9_encodeintra.h b/libvpx/vp9/encoder/vp9_encodeintra.h
new file mode 100644
index 0000000..14d144b
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodeintra.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_ENCODEINTRA_H_
+#define VP9_ENCODER_VP9_ENCODEINTRA_H_
+
+#include "vp9/encoder/vp9_onyx_int.h"
+
+int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred);
+void vp9_encode_intra_block_y(VP9_COMMON *const cm, MACROBLOCK *mb,
+                              BLOCK_SIZE_TYPE bs);
+void vp9_encode_intra_block_uv(VP9_COMMON *const cm, MACROBLOCK *mb,
+                               BLOCK_SIZE_TYPE bs);
+
+#endif  // VP9_ENCODER_VP9_ENCODEINTRA_H_
diff --git a/libvpx/vp9/encoder/vp9_encodemb.c b/libvpx/vp9/encoder/vp9_encodemb.c
new file mode 100644
index 0000000..4f45496
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodemb.c
@@ -0,0 +1,705 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/encoder/vp9_tokenize.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9_rtcd.h"
+
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
+void vp9_subtract_block(int rows, int cols,
+                        int16_t *diff_ptr, int diff_stride,
+                        const uint8_t *src_ptr, int src_stride,
+                        const uint8_t *pred_ptr, int pred_stride) {
+  int r, c;
+
+  for (r = 0; r < rows; r++) {
+    for (c = 0; c < cols; c++)
+      diff_ptr[c] = src_ptr[c] - pred_ptr[c];
+
+    diff_ptr += diff_stride;
+    pred_ptr += pred_stride;
+    src_ptr  += src_stride;
+  }
+}
+
+static void inverse_transform_b_4x4_add(MACROBLOCKD *xd, int eob,
+                                        int16_t *dqcoeff, uint8_t *dest,
+                                        int stride) {
+  if (eob <= 1)
+    xd->inv_txm4x4_1_add(dqcoeff, dest, stride);
+  else
+    xd->inv_txm4x4_add(dqcoeff, dest, stride);
+}
+
+
+static void subtract_plane(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize, int plane) {
+  struct macroblock_plane *const p = &x->plane[plane];
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+  const int bw = plane_block_width(bsize, pd);
+  const int bh = plane_block_height(bsize, pd);
+
+  vp9_subtract_block(bh, bw, p->src_diff, bw,
+                     p->src.buf, p->src.stride,
+                     pd->dst.buf, pd->dst.stride);
+}
+
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  subtract_plane(x, bsize, 0);
+}
+
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  int i;
+
+  for (i = 1; i < MAX_MB_PLANE; i++)
+    subtract_plane(x, bsize, i);
+}
+
+void vp9_subtract_sb(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  vp9_subtract_sby(x, bsize);
+  vp9_subtract_sbuv(x, bsize);
+}
+
+
+#define RDTRUNC(RM,DM,R,D) ( (128+(R)*(RM)) & 0xFF )
+#define RDTRUNC_8x8(RM,DM,R,D) ( (128+(R)*(RM)) & 0xFF )
+typedef struct vp9_token_state vp9_token_state;
+
+struct vp9_token_state {
+  int           rate;
+  int           error;
+  int           next;
+  signed char   token;
+  short         qc;
+};
+
+// TODO: experiments to find optimal multiple numbers
+#define Y1_RD_MULT 4
+#define UV_RD_MULT 2
+
+static const int plane_rd_mult[4] = {
+  Y1_RD_MULT,
+  UV_RD_MULT,
+};
+
+#define UPDATE_RD_COST()\
+{\
+  rd_cost0 = RDCOST(rdmult, rddiv, rate0, error0);\
+  rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1);\
+  if (rd_cost0 == rd_cost1) {\
+    rd_cost0 = RDTRUNC(rdmult, rddiv, rate0, error0);\
+    rd_cost1 = RDTRUNC(rdmult, rddiv, rate1, error1);\
+  }\
+}
+
+// This function is a place holder for now but may ultimately need
+// to scan previous tokens to work out the correct context.
+static int trellis_get_coeff_context(const int *scan,
+                                     const int *nb,
+                                     int idx, int token,
+                                     uint8_t *token_cache,
+                                     int pad, int l) {
+  int bak = token_cache[scan[idx]], pt;
+  token_cache[scan[idx]] = vp9_pt_energy_class[token];
+  pt = vp9_get_coef_context(scan, nb, pad, token_cache, idx + 1, l);
+  token_cache[scan[idx]] = bak;
+  return pt;
+}
+
+static void optimize_b(VP9_COMMON *const cm, MACROBLOCK *mb,
+                       int plane, int block, BLOCK_SIZE_TYPE bsize,
+                       ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
+                       TX_SIZE tx_size) {
+  const int ref = mb->e_mbd.mode_info_context->mbmi.ref_frame[0] != INTRA_FRAME;
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  vp9_token_state tokens[1025][2];
+  unsigned best_index[1025][2];
+  const int16_t *coeff_ptr = BLOCK_OFFSET(mb->plane[plane].coeff,
+                                          block, 16);
+  int16_t *qcoeff_ptr;
+  int16_t *dqcoeff_ptr;
+  int eob = xd->plane[plane].eobs[block], final_eob, sz = 0;
+  const int i0 = 0;
+  int rc, x, next, i;
+  int64_t rdmult, rddiv, rd_cost0, rd_cost1;
+  int rate0, rate1, error0, error1, t0, t1;
+  int best, band, pt;
+  PLANE_TYPE type = xd->plane[plane].plane_type;
+  int err_mult = plane_rd_mult[type];
+  int default_eob, pad;
+  int const *scan, *nb;
+  const int mul = 1 + (tx_size == TX_32X32);
+  uint8_t token_cache[1024];
+  const int ib = txfrm_block_to_raster_block(xd, bsize, plane,
+                                             block, 2 * tx_size);
+  const int16_t *dequant_ptr = xd->plane[plane].dequant;
+  const uint8_t * band_translate;
+
+  assert((!type && !plane) || (type && plane));
+  dqcoeff_ptr = BLOCK_OFFSET(xd->plane[plane].dqcoeff, block, 16);
+  qcoeff_ptr = BLOCK_OFFSET(xd->plane[plane].qcoeff, block, 16);
+  switch (tx_size) {
+    default:
+    case TX_4X4: {
+      const TX_TYPE tx_type = plane == 0 ? get_tx_type_4x4(xd, ib) : DCT_DCT;
+      default_eob = 16;
+      scan = get_scan_4x4(tx_type);
+      band_translate = vp9_coefband_trans_4x4;
+      break;
+    }
+    case TX_8X8: {
+      const TX_TYPE tx_type = plane == 0 ? get_tx_type_8x8(xd, ib) : DCT_DCT;
+      scan = get_scan_8x8(tx_type);
+      default_eob = 64;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_16X16: {
+      const TX_TYPE tx_type = plane == 0 ? get_tx_type_16x16(xd, ib) : DCT_DCT;
+      scan = get_scan_16x16(tx_type);
+      default_eob = 256;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_32X32:
+      scan = vp9_default_scan_32x32;
+      default_eob = 1024;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+  }
+  assert(eob <= default_eob);
+
+  /* Now set up a Viterbi trellis to evaluate alternative roundings. */
+  rdmult = mb->rdmult * err_mult;
+  if (mb->e_mbd.mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME)
+    rdmult = (rdmult * 9) >> 4;
+  rddiv = mb->rddiv;
+  memset(best_index, 0, sizeof(best_index));
+  /* Initialize the sentinel node of the trellis. */
+  tokens[eob][0].rate = 0;
+  tokens[eob][0].error = 0;
+  tokens[eob][0].next = default_eob;
+  tokens[eob][0].token = DCT_EOB_TOKEN;
+  tokens[eob][0].qc = 0;
+  *(tokens[eob] + 1) = *(tokens[eob] + 0);
+  next = eob;
+  for (i = 0; i < eob; i++)
+    token_cache[scan[i]] = vp9_pt_energy_class[vp9_dct_value_tokens_ptr[
+        qcoeff_ptr[scan[i]]].token];
+  nb = vp9_get_coef_neighbors_handle(scan, &pad);
+
+  for (i = eob; i-- > i0;) {
+    int base_bits, d2, dx;
+
+    rc = scan[i];
+    x = qcoeff_ptr[rc];
+    /* Only add a trellis state for non-zero coefficients. */
+    if (x) {
+      int shortcut = 0;
+      error0 = tokens[next][0].error;
+      error1 = tokens[next][1].error;
+      /* Evaluate the first possibility for this state. */
+      rate0 = tokens[next][0].rate;
+      rate1 = tokens[next][1].rate;
+      t0 = (vp9_dct_value_tokens_ptr + x)->token;
+      /* Consider both possible successor states. */
+      if (next < default_eob) {
+        band = get_coef_band(band_translate, i + 1);
+        pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache,
+                                       pad, default_eob);
+        rate0 +=
+          mb->token_costs_noskip[tx_size][type][ref][band][pt]
+                                [tokens[next][0].token];
+        rate1 +=
+          mb->token_costs_noskip[tx_size][type][ref][band][pt]
+                                [tokens[next][1].token];
+      }
+      UPDATE_RD_COST();
+      /* And pick the best. */
+      best = rd_cost1 < rd_cost0;
+      base_bits = *(vp9_dct_value_cost_ptr + x);
+      dx = mul * (dqcoeff_ptr[rc] - coeff_ptr[rc]);
+      d2 = dx * dx;
+      tokens[i][0].rate = base_bits + (best ? rate1 : rate0);
+      tokens[i][0].error = d2 + (best ? error1 : error0);
+      tokens[i][0].next = next;
+      tokens[i][0].token = t0;
+      tokens[i][0].qc = x;
+      best_index[i][0] = best;
+
+      /* Evaluate the second possibility for this state. */
+      rate0 = tokens[next][0].rate;
+      rate1 = tokens[next][1].rate;
+
+      if ((abs(x)*dequant_ptr[rc != 0] > abs(coeff_ptr[rc]) * mul) &&
+          (abs(x)*dequant_ptr[rc != 0] < abs(coeff_ptr[rc]) * mul +
+                                         dequant_ptr[rc != 0]))
+        shortcut = 1;
+      else
+        shortcut = 0;
+
+      if (shortcut) {
+        sz = -(x < 0);
+        x -= 2 * sz + 1;
+      }
+
+      /* Consider both possible successor states. */
+      if (!x) {
+        /* If we reduced this coefficient to zero, check to see if
+         *  we need to move the EOB back here.
+         */
+        t0 = tokens[next][0].token == DCT_EOB_TOKEN ?
+             DCT_EOB_TOKEN : ZERO_TOKEN;
+        t1 = tokens[next][1].token == DCT_EOB_TOKEN ?
+             DCT_EOB_TOKEN : ZERO_TOKEN;
+      } else {
+        t0 = t1 = (vp9_dct_value_tokens_ptr + x)->token;
+      }
+      if (next < default_eob) {
+        band = get_coef_band(band_translate, i + 1);
+        if (t0 != DCT_EOB_TOKEN) {
+          pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache,
+                                         pad, default_eob);
+          if (!x)
+            rate0 += mb->token_costs[tx_size][type][ref][band][pt][
+                tokens[next][0].token];
+          else
+            rate0 += mb->token_costs_noskip[tx_size][type][ref][band][pt][
+                tokens[next][0].token];
+        }
+        if (t1 != DCT_EOB_TOKEN) {
+          pt = trellis_get_coeff_context(scan, nb, i, t1, token_cache,
+                                         pad, default_eob);
+          if (!x)
+            rate1 += mb->token_costs[tx_size][type][ref][band][pt][
+                tokens[next][1].token];
+          else
+            rate1 += mb->token_costs_noskip[tx_size][type][ref][band][pt][
+                tokens[next][1].token];
+        }
+      }
+
+      UPDATE_RD_COST();
+      /* And pick the best. */
+      best = rd_cost1 < rd_cost0;
+      base_bits = *(vp9_dct_value_cost_ptr + x);
+
+      if (shortcut) {
+        dx -= (dequant_ptr[rc != 0] + sz) ^ sz;
+        d2 = dx * dx;
+      }
+      tokens[i][1].rate = base_bits + (best ? rate1 : rate0);
+      tokens[i][1].error = d2 + (best ? error1 : error0);
+      tokens[i][1].next = next;
+      tokens[i][1].token = best ? t1 : t0;
+      tokens[i][1].qc = x;
+      best_index[i][1] = best;
+      /* Finally, make this the new head of the trellis. */
+      next = i;
+    }
+    /* There's no choice to make for a zero coefficient, so we don't
+     *  add a new trellis node, but we do need to update the costs.
+     */
+    else {
+      band = get_coef_band(band_translate, i + 1);
+      t0 = tokens[next][0].token;
+      t1 = tokens[next][1].token;
+      /* Update the cost of each path if we're past the EOB token. */
+      if (t0 != DCT_EOB_TOKEN) {
+        tokens[next][0].rate +=
+            mb->token_costs[tx_size][type][ref][band][0][t0];
+        tokens[next][0].token = ZERO_TOKEN;
+      }
+      if (t1 != DCT_EOB_TOKEN) {
+        tokens[next][1].rate +=
+            mb->token_costs[tx_size][type][ref][band][0][t1];
+        tokens[next][1].token = ZERO_TOKEN;
+      }
+      /* Don't update next, because we didn't add a new node. */
+    }
+  }
+
+  /* Now pick the best path through the whole trellis. */
+  band = get_coef_band(band_translate, i + 1);
+  pt = combine_entropy_contexts(*a, *l);
+  rate0 = tokens[next][0].rate;
+  rate1 = tokens[next][1].rate;
+  error0 = tokens[next][0].error;
+  error1 = tokens[next][1].error;
+  t0 = tokens[next][0].token;
+  t1 = tokens[next][1].token;
+  rate0 += mb->token_costs_noskip[tx_size][type][ref][band][pt][t0];
+  rate1 += mb->token_costs_noskip[tx_size][type][ref][band][pt][t1];
+  UPDATE_RD_COST();
+  best = rd_cost1 < rd_cost0;
+  final_eob = i0 - 1;
+  vpx_memset(qcoeff_ptr, 0, sizeof(*qcoeff_ptr) * (16 << (tx_size * 2)));
+  vpx_memset(dqcoeff_ptr, 0, sizeof(*dqcoeff_ptr) * (16 << (tx_size * 2)));
+  for (i = next; i < eob; i = next) {
+    x = tokens[i][best].qc;
+    if (x) {
+      final_eob = i;
+    }
+    rc = scan[i];
+    qcoeff_ptr[rc] = x;
+    dqcoeff_ptr[rc] = (x * dequant_ptr[rc != 0]) / mul;
+
+    next = tokens[i][best].next;
+    best = best_index[i][best];
+  }
+  final_eob++;
+
+  xd->plane[plane].eobs[block] = final_eob;
+  *a = *l = (final_eob > 0);
+}
+
+struct optimize_block_args {
+  VP9_COMMON *cm;
+  MACROBLOCK *x;
+  struct optimize_ctx *ctx;
+};
+
+void vp9_optimize_b(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                    int ss_txfrm_size, VP9_COMMON *cm, MACROBLOCK *mb,
+                    struct optimize_ctx *ctx) {
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  int x, y;
+
+  // find current entropy context
+  txfrm_block_to_raster_xy(xd, bsize, plane, block, ss_txfrm_size, &x, &y);
+
+  optimize_b(cm, mb, plane, block, bsize,
+             &ctx->ta[plane][x], &ctx->tl[plane][y], ss_txfrm_size / 2);
+}
+
+static void optimize_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                           int ss_txfrm_size, void *arg) {
+  const struct optimize_block_args* const args = arg;
+  vp9_optimize_b(plane, block, bsize, ss_txfrm_size, args->cm, args->x,
+                 args->ctx);
+}
+
+void vp9_optimize_init(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize,
+                       struct optimize_ctx *ctx) {
+  int p;
+
+  for (p = 0; p < MAX_MB_PLANE; p++) {
+    const struct macroblockd_plane* const plane = &xd->plane[p];
+    const int bwl = b_width_log2(bsize) - plane->subsampling_x;
+    const int bhl = b_height_log2(bsize) - plane->subsampling_y;
+    const MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+    const TX_SIZE tx_size = p ? get_uv_tx_size(mbmi)
+                              : mbmi->txfm_size;
+    int i, j;
+
+    for (i = 0; i < 1 << bwl; i += 1 << tx_size) {
+      int c = 0;
+      ctx->ta[p][i] = 0;
+      for (j = 0; j < 1 << tx_size && !c; j++) {
+        c = ctx->ta[p][i] |= plane->above_context[i + j];
+      }
+    }
+    for (i = 0; i < 1 << bhl; i += 1 << tx_size) {
+      int c = 0;
+      ctx->tl[p][i] = 0;
+      for (j = 0; j < 1 << tx_size && !c; j++) {
+        c = ctx->tl[p][i] |= plane->left_context[i + j];
+      }
+    }
+  }
+}
+
+void vp9_optimize_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  struct optimize_ctx ctx;
+  struct optimize_block_args arg = {cm, x, &ctx};
+  vp9_optimize_init(&x->e_mbd, bsize, &ctx);
+  foreach_transformed_block_in_plane(&x->e_mbd, bsize, 0, optimize_block, &arg);
+}
+
+void vp9_optimize_sbuv(VP9_COMMON *const cm, MACROBLOCK *x,
+                       BLOCK_SIZE_TYPE bsize) {
+  struct optimize_ctx ctx;
+  struct optimize_block_args arg = {cm, x, &ctx};
+  vp9_optimize_init(&x->e_mbd, bsize, &ctx);
+  foreach_transformed_block_uv(&x->e_mbd, bsize, optimize_block, &arg);
+}
+
+struct encode_b_args {
+  VP9_COMMON *cm;
+  MACROBLOCK *x;
+  struct optimize_ctx *ctx;
+};
+
+static void xform_quant(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                         int ss_txfrm_size, void *arg) {
+  struct encode_b_args* const args = arg;
+  MACROBLOCK* const x = args->x;
+  MACROBLOCKD* const xd = &x->e_mbd;
+  const int bw = plane_block_width(bsize, &xd->plane[plane]);
+  const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
+                                                       block, ss_txfrm_size);
+  int16_t *const coeff = BLOCK_OFFSET(x->plane[plane].coeff, block, 16);
+  int16_t *const src_diff = raster_block_offset_int16(xd, bsize, plane,
+                                                      raster_block,
+                                                      x->plane[plane].src_diff);
+  TX_TYPE tx_type = DCT_DCT;
+
+  switch (ss_txfrm_size / 2) {
+    case TX_32X32:
+      if (x->rd_search)
+        vp9_short_fdct32x32_rd(src_diff, coeff, bw * 2);
+      else
+        vp9_short_fdct32x32(src_diff, coeff, bw * 2);
+      break;
+    case TX_16X16:
+      tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
+      if (tx_type != DCT_DCT)
+        vp9_short_fht16x16(src_diff, coeff, bw, tx_type);
+      else
+        x->fwd_txm16x16(src_diff, coeff, bw * 2);
+      break;
+    case TX_8X8:
+      tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
+      if (tx_type != DCT_DCT)
+        vp9_short_fht8x8(src_diff, coeff, bw, tx_type);
+      else
+        x->fwd_txm8x8(src_diff, coeff, bw * 2);
+      break;
+    case TX_4X4:
+      tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
+      if (tx_type != DCT_DCT)
+        vp9_short_fht4x4(src_diff, coeff, bw, tx_type);
+      else
+        x->fwd_txm4x4(src_diff, coeff, bw * 2);
+      break;
+    default:
+      assert(0);
+  }
+
+  vp9_quantize(x, plane, block, 16 << ss_txfrm_size, tx_type);
+}
+
+static void encode_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                         int ss_txfrm_size, void *arg) {
+  struct encode_b_args *const args = arg;
+  MACROBLOCK *const x = args->x;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
+                                                       block, ss_txfrm_size);
+  struct macroblockd_plane *const pd = &xd->plane[plane];
+  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block, 16);
+  uint8_t *const dst = raster_block_offset_uint8(xd, bsize, plane,
+                                                 raster_block,
+                                                 pd->dst.buf, pd->dst.stride);
+  TX_TYPE tx_type = DCT_DCT;
+
+  xform_quant(plane, block, bsize, ss_txfrm_size, arg);
+
+  if (x->optimize)
+    vp9_optimize_b(plane, block, bsize, ss_txfrm_size, args->cm, x, args->ctx);
+
+  switch (ss_txfrm_size / 2) {
+    case TX_32X32:
+      vp9_short_idct32x32_add(dqcoeff, dst, pd->dst.stride);
+      break;
+    case TX_16X16:
+      tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        vp9_short_idct16x16_add(dqcoeff, dst, pd->dst.stride);
+      else
+        vp9_short_iht16x16_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+    case TX_8X8:
+      tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        vp9_short_idct8x8_add(dqcoeff, dst, pd->dst.stride);
+      else
+        vp9_short_iht8x8_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+    case TX_4X4:
+      tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        // this is like vp9_short_idct4x4 but has a special case around eob<=1
+        // which is significant (not just an optimization) for the lossless
+        // case.
+        inverse_transform_b_4x4_add(xd, pd->eobs[block], dqcoeff,
+                                    dst, pd->dst.stride);
+      else
+        vp9_short_iht4x4_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+  }
+}
+
+void vp9_xform_quant_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD* const xd = &x->e_mbd;
+  struct encode_b_args arg = {cm, x, NULL};
+
+  foreach_transformed_block_in_plane(xd, bsize, 0, xform_quant, &arg);
+}
+
+void vp9_xform_quant_sbuv(VP9_COMMON *cm, MACROBLOCK *x,
+                          BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD* const xd = &x->e_mbd;
+  struct encode_b_args arg = {cm, x, NULL};
+
+  foreach_transformed_block_uv(xd, bsize, xform_quant, &arg);
+}
+
+void vp9_encode_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  struct encode_b_args arg = {cm, x, &ctx};
+
+  vp9_subtract_sby(x, bsize);
+  if (x->optimize)
+    vp9_optimize_init(xd, bsize, &ctx);
+
+  foreach_transformed_block_in_plane(xd, bsize, 0, encode_block, &arg);
+}
+
+void vp9_encode_sbuv(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  struct encode_b_args arg = {cm, x, &ctx};
+
+  vp9_subtract_sbuv(x, bsize);
+  if (x->optimize)
+    vp9_optimize_init(xd, bsize, &ctx);
+
+  foreach_transformed_block_uv(xd, bsize, encode_block, &arg);
+}
+
+void vp9_encode_sb(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  struct encode_b_args arg = {cm, x, &ctx};
+
+  vp9_subtract_sb(x, bsize);
+  if (x->optimize)
+    vp9_optimize_init(xd, bsize, &ctx);
+
+  foreach_transformed_block(xd, bsize, encode_block, &arg);
+}
+
+static void encode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                               int ss_txfrm_size, void *arg) {
+  struct encode_b_args* const args = arg;
+  MACROBLOCK *const x = args->x;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+  const TX_SIZE tx_size = (TX_SIZE)(ss_txfrm_size / 2);
+  struct macroblock_plane *const p = &x->plane[plane];
+  struct macroblockd_plane *const pd = &xd->plane[plane];
+  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block, 16);
+  const int bw = plane_block_width(bsize, pd);
+  const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
+                                                       block, ss_txfrm_size);
+
+  uint8_t *const src = raster_block_offset_uint8(xd, bsize, plane, raster_block,
+                                                 p->src.buf, p->src.stride);
+  uint8_t *const dst = raster_block_offset_uint8(xd, bsize, plane, raster_block,
+                                                 pd->dst.buf, pd->dst.stride);
+  int16_t *const src_diff = raster_block_offset_int16(xd, bsize, plane,
+                                                      raster_block,
+                                                      p->src_diff);
+
+  const int txfm_b_size = 4 << tx_size;
+  int ib = raster_block;
+  int tx_ib = ib >> tx_size;
+  int plane_b_size;
+
+  TX_TYPE tx_type;
+  int mode, b_mode;
+
+  if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
+    extend_for_intra(xd, plane, block, bsize, ss_txfrm_size);
+  }
+
+  mode = plane == 0? mbmi->mode: mbmi->uv_mode;
+  if (plane == 0 &&
+      mbmi->sb_type < BLOCK_SIZE_SB8X8 &&
+      mbmi->ref_frame[0] == INTRA_FRAME)
+    b_mode = xd->mode_info_context->bmi[ib].as_mode.first;
+  else
+    b_mode = mode;
+
+  assert(b_mode >= DC_PRED && b_mode <= TM_PRED);
+
+  plane_b_size = b_width_log2(bsize) - pd->subsampling_x;
+  vp9_predict_intra_block(xd, tx_ib, plane_b_size, tx_size, b_mode,
+                          dst, pd->dst.stride);
+  vp9_subtract_block(txfm_b_size, txfm_b_size, src_diff, bw,
+                     src, p->src.stride, dst, pd->dst.stride);
+
+  xform_quant(plane, block, bsize, ss_txfrm_size, arg);
+
+
+  // if (x->optimize)
+  // vp9_optimize_b(plane, block, bsize, ss_txfrm_size,
+  //                args->cm, x, args->ctx);
+
+  switch (ss_txfrm_size / 2) {
+    case TX_32X32:
+        vp9_short_idct32x32_add(dqcoeff, dst, pd->dst.stride);
+      break;
+    case TX_16X16:
+      tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        vp9_short_idct16x16_add(dqcoeff, dst, pd->dst.stride);
+      else
+        vp9_short_iht16x16_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+    case TX_8X8:
+      tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        vp9_short_idct8x8_add(dqcoeff, dst, pd->dst.stride);
+      else
+        vp9_short_iht8x8_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+    case TX_4X4:
+      tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
+      if (tx_type == DCT_DCT)
+        // this is like vp9_short_idct4x4 but has a special case around eob<=1
+        // which is significant (not just an optimization) for the lossless
+        // case.
+        inverse_transform_b_4x4_add(xd, pd->eobs[block], dqcoeff,
+                                    dst, pd->dst.stride);
+      else
+        vp9_short_iht4x4_add(dqcoeff, dst, pd->dst.stride, tx_type);
+      break;
+  }
+}
+
+void vp9_encode_intra_block_y(VP9_COMMON *cm, MACROBLOCK *x,
+                              BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD* const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  struct encode_b_args arg = {cm, x, &ctx};
+
+  foreach_transformed_block_in_plane(xd, bsize, 0,
+                                     encode_block_intra, &arg);
+}
+void vp9_encode_intra_block_uv(VP9_COMMON *cm, MACROBLOCK *x,
+                              BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD* const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  struct encode_b_args arg = {cm, x, &ctx};
+  foreach_transformed_block_uv(xd, bsize, encode_block_intra, &arg);
+}
+
diff --git a/libvpx/vp9/encoder/vp9_encodemb.h b/libvpx/vp9/encoder/vp9_encodemb.h
new file mode 100644
index 0000000..5796903
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodemb.h
@@ -0,0 +1,53 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_ENCODEMB_H_
+#define VP9_ENCODER_VP9_ENCODEMB_H_
+
+#include "./vpx_config.h"
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
+typedef struct {
+  MB_PREDICTION_MODE mode;
+  MV_REFERENCE_FRAME ref_frame;
+  MV_REFERENCE_FRAME second_ref_frame;
+} MODE_DEFINITION;
+
+struct optimize_ctx {
+  ENTROPY_CONTEXT ta[MAX_MB_PLANE][16];
+  ENTROPY_CONTEXT tl[MAX_MB_PLANE][16];
+};
+
+void vp9_optimize_init(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize,
+                       struct optimize_ctx *ctx);
+void vp9_optimize_b(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                    int ss_txfrm_size, VP9_COMMON *cm, MACROBLOCK *x,
+                    struct optimize_ctx *ctx);
+void vp9_optimize_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_optimize_sbuv(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+
+void vp9_encode_sb(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_encode_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_encode_sbuv(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+
+void vp9_xform_quant_sby(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_xform_quant_sbuv(VP9_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+
+void vp9_subtract_block(int rows, int cols,
+                        int16_t *diff_ptr, int diff_stride,
+                        const uint8_t *src_ptr, int src_stride,
+                        const uint8_t *pred_ptr, int pred_stride);
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sb(MACROBLOCK *xd, BLOCK_SIZE_TYPE bsize);
+
+#endif  // VP9_ENCODER_VP9_ENCODEMB_H_
diff --git a/libvpx/vp9/encoder/vp9_encodemv.c b/libvpx/vp9/encoder/vp9_encodemv.c
new file mode 100644
index 0000000..a582d18
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodemv.c
@@ -0,0 +1,610 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_common.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_systemdependent.h"
+
+#include <math.h>
+
+#ifdef ENTROPY_STATS
+extern unsigned int active_section;
+#endif
+
+#ifdef NMV_STATS
+nmv_context_counts tnmvcounts;
+#endif
+
+static void encode_mv_component(vp9_writer* w, int comp,
+                                const nmv_component* mvcomp, int usehp) {
+  int offset;
+  const int sign = comp < 0;
+  const int mag = sign ? -comp : comp;
+  const int mv_class = vp9_get_mv_class(mag - 1, &offset);
+  const int d = offset >> 3;                // int mv data
+  const int fr = (offset >> 1) & 3;         // fractional mv data
+  const int hp = offset & 1;                // high precision mv data
+
+  assert(comp != 0);
+
+  // Sign
+  vp9_write(w, sign, mvcomp->sign);
+
+  // Class
+  write_token(w, vp9_mv_class_tree, mvcomp->classes,
+              &vp9_mv_class_encodings[mv_class]);
+
+  // Integer bits
+  if (mv_class == MV_CLASS_0) {
+    write_token(w, vp9_mv_class0_tree, mvcomp->class0,
+                &vp9_mv_class0_encodings[d]);
+  } else {
+    int i;
+    const int n = mv_class + CLASS0_BITS - 1;  // number of bits
+    for (i = 0; i < n; ++i)
+      vp9_write(w, (d >> i) & 1, mvcomp->bits[i]);
+  }
+
+  // Fractional bits
+  write_token(w, vp9_mv_fp_tree,
+              mv_class == MV_CLASS_0 ?  mvcomp->class0_fp[d] : mvcomp->fp,
+              &vp9_mv_fp_encodings[fr]);
+
+  // High precision bit
+  if (usehp)
+    vp9_write(w, hp,
+              mv_class == MV_CLASS_0 ? mvcomp->class0_hp : mvcomp->hp);
+}
+
+
+static void build_nmv_component_cost_table(int *mvcost,
+                                           const nmv_component* const mvcomp,
+                                           int usehp) {
+  int i, v;
+  int sign_cost[2], class_cost[MV_CLASSES], class0_cost[CLASS0_SIZE];
+  int bits_cost[MV_OFFSET_BITS][2];
+  int class0_fp_cost[CLASS0_SIZE][4], fp_cost[4];
+  int class0_hp_cost[2], hp_cost[2];
+
+  sign_cost[0] = vp9_cost_zero(mvcomp->sign);
+  sign_cost[1] = vp9_cost_one(mvcomp->sign);
+  vp9_cost_tokens(class_cost, mvcomp->classes, vp9_mv_class_tree);
+  vp9_cost_tokens(class0_cost, mvcomp->class0, vp9_mv_class0_tree);
+  for (i = 0; i < MV_OFFSET_BITS; ++i) {
+    bits_cost[i][0] = vp9_cost_zero(mvcomp->bits[i]);
+    bits_cost[i][1] = vp9_cost_one(mvcomp->bits[i]);
+  }
+
+  for (i = 0; i < CLASS0_SIZE; ++i)
+    vp9_cost_tokens(class0_fp_cost[i], mvcomp->class0_fp[i], vp9_mv_fp_tree);
+  vp9_cost_tokens(fp_cost, mvcomp->fp, vp9_mv_fp_tree);
+
+  if (usehp) {
+    class0_hp_cost[0] = vp9_cost_zero(mvcomp->class0_hp);
+    class0_hp_cost[1] = vp9_cost_one(mvcomp->class0_hp);
+    hp_cost[0] = vp9_cost_zero(mvcomp->hp);
+    hp_cost[1] = vp9_cost_one(mvcomp->hp);
+  }
+  mvcost[0] = 0;
+  for (v = 1; v <= MV_MAX; ++v) {
+    int z, c, o, d, e, f, cost = 0;
+    z = v - 1;
+    c = vp9_get_mv_class(z, &o);
+    cost += class_cost[c];
+    d = (o >> 3);               /* int mv data */
+    f = (o >> 1) & 3;           /* fractional pel mv data */
+    e = (o & 1);                /* high precision mv data */
+    if (c == MV_CLASS_0) {
+      cost += class0_cost[d];
+    } else {
+      int i, b;
+      b = c + CLASS0_BITS - 1;  /* number of bits */
+      for (i = 0; i < b; ++i)
+        cost += bits_cost[i][((d >> i) & 1)];
+    }
+    if (c == MV_CLASS_0) {
+      cost += class0_fp_cost[d][f];
+    } else {
+      cost += fp_cost[f];
+    }
+    if (usehp) {
+      if (c == MV_CLASS_0) {
+        cost += class0_hp_cost[e];
+      } else {
+        cost += hp_cost[e];
+      }
+    }
+    mvcost[v] = cost + sign_cost[0];
+    mvcost[-v] = cost + sign_cost[1];
+  }
+}
+
+static int update_nmv_savings(const unsigned int ct[2],
+                              const vp9_prob cur_p,
+                              const vp9_prob new_p,
+                              const vp9_prob upd_p) {
+
+#ifdef LOW_PRECISION_MV_UPDATE
+  vp9_prob mod_p = new_p | 1;
+#else
+  vp9_prob mod_p = new_p;
+#endif
+  const int cur_b = cost_branch256(ct, cur_p);
+  const int mod_b = cost_branch256(ct, mod_p);
+  const int cost = 7 * 256 +
+#ifndef LOW_PRECISION_MV_UPDATE
+      256 +
+#endif
+      (vp9_cost_one(upd_p) - vp9_cost_zero(upd_p));
+  if (cur_b - mod_b - cost > 0) {
+    return cur_b - mod_b - cost;
+  } else {
+    return 0 - vp9_cost_zero(upd_p);
+  }
+}
+
+static int update_nmv(
+  vp9_writer *const bc,
+  const unsigned int ct[2],
+  vp9_prob *const cur_p,
+  const vp9_prob new_p,
+  const vp9_prob upd_p) {
+
+#ifdef LOW_PRECISION_MV_UPDATE
+  vp9_prob mod_p = new_p | 1;
+#else
+  vp9_prob mod_p = new_p;
+#endif
+
+  const int cur_b = cost_branch256(ct, *cur_p);
+  const int mod_b = cost_branch256(ct, mod_p);
+  const int cost = 7 * 256 +
+#ifndef LOW_PRECISION_MV_UPDATE
+      256 +
+#endif
+      (vp9_cost_one(upd_p) - vp9_cost_zero(upd_p));
+
+  if (cur_b - mod_b > cost) {
+    *cur_p = mod_p;
+    vp9_write(bc, 1, upd_p);
+#ifdef LOW_PRECISION_MV_UPDATE
+    vp9_write_literal(bc, mod_p >> 1, 7);
+#else
+    vp9_write_literal(bc, mod_p, 8);
+#endif
+    return 1;
+  } else {
+    vp9_write(bc, 0, upd_p);
+    return 0;
+  }
+}
+
+void print_nmvcounts(nmv_context_counts tnmvcounts) {
+  int i, j, k;
+  printf("\nCounts =\n  { ");
+  for (j = 0; j < MV_JOINTS; ++j)
+    printf("%d, ", tnmvcounts.joints[j]);
+  printf("},\n");
+  for (i = 0; i < 2; ++i) {
+    printf("  {\n");
+    printf("    %d/%d,\n", tnmvcounts.comps[i].sign[0],
+                           tnmvcounts.comps[i].sign[1]);
+    printf("    { ");
+    for (j = 0; j < MV_CLASSES; ++j)
+      printf("%d, ", tnmvcounts.comps[i].classes[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < CLASS0_SIZE; ++j)
+      printf("%d, ", tnmvcounts.comps[i].class0[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < MV_OFFSET_BITS; ++j)
+      printf("%d/%d, ", tnmvcounts.comps[i].bits[j][0],
+                        tnmvcounts.comps[i].bits[j][1]);
+    printf("},\n");
+
+    printf("    {");
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      printf("{");
+      for (k = 0; k < 4; ++k)
+        printf("%d, ", tnmvcounts.comps[i].class0_fp[j][k]);
+      printf("}, ");
+    }
+    printf("},\n");
+
+    printf("    { ");
+    for (j = 0; j < 4; ++j)
+      printf("%d, ", tnmvcounts.comps[i].fp[j]);
+    printf("},\n");
+
+    printf("    %d/%d,\n",
+           tnmvcounts.comps[i].class0_hp[0],
+           tnmvcounts.comps[i].class0_hp[1]);
+    printf("    %d/%d,\n",
+           tnmvcounts.comps[i].hp[0],
+           tnmvcounts.comps[i].hp[1]);
+    printf("  },\n");
+  }
+}
+
+#ifdef NMV_STATS
+void init_nmvstats() {
+  vp9_zero(tnmvcounts);
+}
+
+void print_nmvstats() {
+  nmv_context prob;
+  unsigned int branch_ct_joint[MV_JOINTS - 1][2];
+  unsigned int branch_ct_sign[2][2];
+  unsigned int branch_ct_classes[2][MV_CLASSES - 1][2];
+  unsigned int branch_ct_class0[2][CLASS0_SIZE - 1][2];
+  unsigned int branch_ct_bits[2][MV_OFFSET_BITS][2];
+  unsigned int branch_ct_class0_fp[2][CLASS0_SIZE][4 - 1][2];
+  unsigned int branch_ct_fp[2][4 - 1][2];
+  unsigned int branch_ct_class0_hp[2][2];
+  unsigned int branch_ct_hp[2][2];
+  int i, j, k;
+  vp9_counts_to_nmv_context(&tnmvcounts, &prob, 1,
+                            branch_ct_joint, branch_ct_sign, branch_ct_classes,
+                            branch_ct_class0, branch_ct_bits,
+                            branch_ct_class0_fp, branch_ct_fp,
+                            branch_ct_class0_hp, branch_ct_hp);
+
+  printf("\nCounts =\n  { ");
+  for (j = 0; j < MV_JOINTS; ++j)
+    printf("%d, ", tnmvcounts.joints[j]);
+  printf("},\n");
+  for (i = 0; i < 2; ++i) {
+    printf("  {\n");
+    printf("    %d/%d,\n", tnmvcounts.comps[i].sign[0],
+                           tnmvcounts.comps[i].sign[1]);
+    printf("    { ");
+    for (j = 0; j < MV_CLASSES; ++j)
+      printf("%d, ", tnmvcounts.comps[i].classes[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < CLASS0_SIZE; ++j)
+      printf("%d, ", tnmvcounts.comps[i].class0[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < MV_OFFSET_BITS; ++j)
+      printf("%d/%d, ", tnmvcounts.comps[i].bits[j][0],
+                        tnmvcounts.comps[i].bits[j][1]);
+    printf("},\n");
+
+    printf("    {");
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      printf("{");
+      for (k = 0; k < 4; ++k)
+        printf("%d, ", tnmvcounts.comps[i].class0_fp[j][k]);
+      printf("}, ");
+    }
+    printf("},\n");
+
+    printf("    { ");
+    for (j = 0; j < 4; ++j)
+      printf("%d, ", tnmvcounts.comps[i].fp[j]);
+    printf("},\n");
+
+    printf("    %d/%d,\n",
+           tnmvcounts.comps[i].class0_hp[0],
+           tnmvcounts.comps[i].class0_hp[1]);
+    printf("    %d/%d,\n",
+           tnmvcounts.comps[i].hp[0],
+           tnmvcounts.comps[i].hp[1]);
+    printf("  },\n");
+  }
+
+  printf("\nProbs =\n  { ");
+  for (j = 0; j < MV_JOINTS - 1; ++j)
+    printf("%d, ", prob.joints[j]);
+  printf("},\n");
+  for (i=0; i< 2; ++i) {
+    printf("  {\n");
+    printf("    %d,\n", prob.comps[i].sign);
+    printf("    { ");
+    for (j = 0; j < MV_CLASSES - 1; ++j)
+      printf("%d, ", prob.comps[i].classes[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < CLASS0_SIZE - 1; ++j)
+      printf("%d, ", prob.comps[i].class0[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < MV_OFFSET_BITS; ++j)
+      printf("%d, ", prob.comps[i].bits[j]);
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      printf("{");
+      for (k = 0; k < 3; ++k)
+        printf("%d, ", prob.comps[i].class0_fp[j][k]);
+      printf("}, ");
+    }
+    printf("},\n");
+    printf("    { ");
+    for (j = 0; j < 3; ++j)
+      printf("%d, ", prob.comps[i].fp[j]);
+    printf("},\n");
+
+    printf("    %d,\n", prob.comps[i].class0_hp);
+    printf("    %d,\n", prob.comps[i].hp);
+    printf("  },\n");
+  }
+}
+
+static void add_nmvcount(nmv_context_counts* const dst,
+                         const nmv_context_counts* const src) {
+  int i, j, k;
+  for (j = 0; j < MV_JOINTS; ++j) {
+    dst->joints[j] += src->joints[j];
+  }
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < MV_VALS; ++j) {
+      dst->comps[i].mvcount[j] += src->comps[i].mvcount[j];
+    }
+    dst->comps[i].sign[0] += src->comps[i].sign[0];
+    dst->comps[i].sign[1] += src->comps[i].sign[1];
+    for (j = 0; j < MV_CLASSES; ++j) {
+      dst->comps[i].classes[j] += src->comps[i].classes[j];
+    }
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      dst->comps[i].class0[j] += src->comps[i].class0[j];
+    }
+    for (j = 0; j < MV_OFFSET_BITS; ++j) {
+      dst->comps[i].bits[j][0] += src->comps[i].bits[j][0];
+      dst->comps[i].bits[j][1] += src->comps[i].bits[j][1];
+    }
+  }
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      for (k = 0; k < 4; ++k) {
+        dst->comps[i].class0_fp[j][k] += src->comps[i].class0_fp[j][k];
+      }
+    }
+    for (j = 0; j < 4; ++j) {
+      dst->comps[i].fp[j] += src->comps[i].fp[j];
+    }
+    dst->comps[i].class0_hp[0] += src->comps[i].class0_hp[0];
+    dst->comps[i].class0_hp[1] += src->comps[i].class0_hp[1];
+    dst->comps[i].hp[0] += src->comps[i].hp[0];
+    dst->comps[i].hp[1] += src->comps[i].hp[1];
+  }
+}
+#endif
+
+void vp9_write_nmv_probs(VP9_COMP* const cpi, int usehp, vp9_writer* const bc) {
+  int i, j;
+  nmv_context prob;
+  unsigned int branch_ct_joint[MV_JOINTS - 1][2];
+  unsigned int branch_ct_sign[2][2];
+  unsigned int branch_ct_classes[2][MV_CLASSES - 1][2];
+  unsigned int branch_ct_class0[2][CLASS0_SIZE - 1][2];
+  unsigned int branch_ct_bits[2][MV_OFFSET_BITS][2];
+  unsigned int branch_ct_class0_fp[2][CLASS0_SIZE][4 - 1][2];
+  unsigned int branch_ct_fp[2][4 - 1][2];
+  unsigned int branch_ct_class0_hp[2][2];
+  unsigned int branch_ct_hp[2][2];
+#ifdef MV_GROUP_UPDATE
+  int savings = 0;
+#endif
+
+#ifdef NMV_STATS
+  if (!cpi->dummy_packing)
+    add_nmvcount(&tnmvcounts, &cpi->NMVcount);
+#endif
+  vp9_counts_to_nmv_context(&cpi->NMVcount, &prob, usehp,
+                            branch_ct_joint, branch_ct_sign, branch_ct_classes,
+                            branch_ct_class0, branch_ct_bits,
+                            branch_ct_class0_fp, branch_ct_fp,
+                            branch_ct_class0_hp, branch_ct_hp);
+  /* write updates if they help */
+#ifdef MV_GROUP_UPDATE
+  for (j = 0; j < MV_JOINTS - 1; ++j) {
+    savings += update_nmv_savings(branch_ct_joint[j],
+                                  cpi->common.fc.nmvc.joints[j],
+                                  prob.joints[j],
+                                  VP9_NMV_UPDATE_PROB);
+  }
+  for (i = 0; i < 2; ++i) {
+    savings += update_nmv_savings(branch_ct_sign[i],
+                                  cpi->common.fc.nmvc.comps[i].sign,
+                                  prob.comps[i].sign,
+                                  VP9_NMV_UPDATE_PROB);
+    for (j = 0; j < MV_CLASSES - 1; ++j) {
+      savings += update_nmv_savings(branch_ct_classes[i][j],
+                                    cpi->common.fc.nmvc.comps[i].classes[j],
+                                    prob.comps[i].classes[j],
+                                    VP9_NMV_UPDATE_PROB);
+    }
+    for (j = 0; j < CLASS0_SIZE - 1; ++j) {
+      savings += update_nmv_savings(branch_ct_class0[i][j],
+                                    cpi->common.fc.nmvc.comps[i].class0[j],
+                                    prob.comps[i].class0[j],
+                                    VP9_NMV_UPDATE_PROB);
+    }
+    for (j = 0; j < MV_OFFSET_BITS; ++j) {
+      savings += update_nmv_savings(branch_ct_bits[i][j],
+                                    cpi->common.fc.nmvc.comps[i].bits[j],
+                                    prob.comps[i].bits[j],
+                                    VP9_NMV_UPDATE_PROB);
+    }
+  }
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      int k;
+      for (k = 0; k < 3; ++k) {
+        savings += update_nmv_savings(branch_ct_class0_fp[i][j][k],
+                                      cpi->common.fc.nmvc.comps[i].class0_fp[j][k],
+                                      prob.comps[i].class0_fp[j][k],
+                                      VP9_NMV_UPDATE_PROB);
+      }
+    }
+    for (j = 0; j < 3; ++j) {
+      savings += update_nmv_savings(branch_ct_fp[i][j],
+                                    cpi->common.fc.nmvc.comps[i].fp[j],
+                                    prob.comps[i].fp[j],
+                                    VP9_NMV_UPDATE_PROB);
+    }
+  }
+  if (usehp) {
+    for (i = 0; i < 2; ++i) {
+      savings += update_nmv_savings(branch_ct_class0_hp[i],
+                                    cpi->common.fc.nmvc.comps[i].class0_hp,
+                                    prob.comps[i].class0_hp,
+                                    VP9_NMV_UPDATE_PROB);
+      savings += update_nmv_savings(branch_ct_hp[i],
+                                    cpi->common.fc.nmvc.comps[i].hp,
+                                    prob.comps[i].hp,
+                                    VP9_NMV_UPDATE_PROB);
+    }
+  }
+  if (savings <= 0) {
+    vp9_write_bit(bc, 0);
+    return;
+  }
+  vp9_write_bit(bc, 1);
+#endif
+
+  for (j = 0; j < MV_JOINTS - 1; ++j) {
+    update_nmv(bc, branch_ct_joint[j],
+               &cpi->common.fc.nmvc.joints[j],
+               prob.joints[j],
+               VP9_NMV_UPDATE_PROB);
+  }
+  for (i = 0; i < 2; ++i) {
+    update_nmv(bc, branch_ct_sign[i],
+               &cpi->common.fc.nmvc.comps[i].sign,
+               prob.comps[i].sign,
+               VP9_NMV_UPDATE_PROB);
+    for (j = 0; j < MV_CLASSES - 1; ++j) {
+      update_nmv(bc, branch_ct_classes[i][j],
+                 &cpi->common.fc.nmvc.comps[i].classes[j],
+                 prob.comps[i].classes[j],
+                 VP9_NMV_UPDATE_PROB);
+    }
+    for (j = 0; j < CLASS0_SIZE - 1; ++j) {
+      update_nmv(bc, branch_ct_class0[i][j],
+                 &cpi->common.fc.nmvc.comps[i].class0[j],
+                 prob.comps[i].class0[j],
+                 VP9_NMV_UPDATE_PROB);
+    }
+    for (j = 0; j < MV_OFFSET_BITS; ++j) {
+      update_nmv(bc, branch_ct_bits[i][j],
+                 &cpi->common.fc.nmvc.comps[i].bits[j],
+                 prob.comps[i].bits[j],
+                 VP9_NMV_UPDATE_PROB);
+    }
+  }
+  for (i = 0; i < 2; ++i) {
+    for (j = 0; j < CLASS0_SIZE; ++j) {
+      int k;
+      for (k = 0; k < 3; ++k) {
+        update_nmv(bc, branch_ct_class0_fp[i][j][k],
+                   &cpi->common.fc.nmvc.comps[i].class0_fp[j][k],
+                   prob.comps[i].class0_fp[j][k],
+                   VP9_NMV_UPDATE_PROB);
+      }
+    }
+    for (j = 0; j < 3; ++j) {
+      update_nmv(bc, branch_ct_fp[i][j],
+                 &cpi->common.fc.nmvc.comps[i].fp[j],
+                 prob.comps[i].fp[j],
+                 VP9_NMV_UPDATE_PROB);
+    }
+  }
+  if (usehp) {
+    for (i = 0; i < 2; ++i) {
+      update_nmv(bc, branch_ct_class0_hp[i],
+                 &cpi->common.fc.nmvc.comps[i].class0_hp,
+                 prob.comps[i].class0_hp,
+                 VP9_NMV_UPDATE_PROB);
+      update_nmv(bc, branch_ct_hp[i],
+                 &cpi->common.fc.nmvc.comps[i].hp,
+                 prob.comps[i].hp,
+                 VP9_NMV_UPDATE_PROB);
+    }
+  }
+}
+
+void vp9_encode_mv(vp9_writer* w, const MV* mv, const MV* ref,
+                   const nmv_context* mvctx, int usehp) {
+  const MV diff = {mv->row - ref->row,
+                   mv->col - ref->col};
+  const MV_JOINT_TYPE j = vp9_get_mv_joint(&diff);
+  usehp = usehp && vp9_use_nmv_hp(ref);
+
+  write_token(w, vp9_mv_joint_tree, mvctx->joints, &vp9_mv_joint_encodings[j]);
+  if (mv_joint_vertical(j))
+    encode_mv_component(w, diff.row, &mvctx->comps[0], usehp);
+
+  if (mv_joint_horizontal(j))
+    encode_mv_component(w, diff.col, &mvctx->comps[1], usehp);
+}
+
+void vp9_build_nmv_cost_table(int *mvjoint,
+                              int *mvcost[2],
+                              const nmv_context* const mvctx,
+                              int usehp,
+                              int mvc_flag_v,
+                              int mvc_flag_h) {
+  vp9_clear_system_state();
+  vp9_cost_tokens(mvjoint, mvctx->joints, vp9_mv_joint_tree);
+  if (mvc_flag_v)
+    build_nmv_component_cost_table(mvcost[0], &mvctx->comps[0], usehp);
+  if (mvc_flag_h)
+    build_nmv_component_cost_table(mvcost[1], &mvctx->comps[1], usehp);
+}
+
+void vp9_update_nmv_count(VP9_COMP *cpi, MACROBLOCK *x,
+                         int_mv *best_ref_mv, int_mv *second_best_ref_mv) {
+  MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
+  MV mv;
+  int bwl = b_width_log2(mbmi->sb_type), bw = 1 << bwl;
+  int bhl = b_height_log2(mbmi->sb_type), bh = 1 << bhl;
+  int idx, idy;
+
+  if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
+    int i;
+    PARTITION_INFO *pi = x->partition_info;
+    for (idy = 0; idy < 2; idy += bh) {
+      for (idx = 0; idx < 2; idx += bw) {
+        i = idy * 2 + idx;
+        if (pi->bmi[i].mode == NEWMV) {
+          mv.row = (pi->bmi[i].mv.as_mv.row - best_ref_mv->as_mv.row);
+          mv.col = (pi->bmi[i].mv.as_mv.col - best_ref_mv->as_mv.col);
+          vp9_increment_nmv(&mv, &best_ref_mv->as_mv, &cpi->NMVcount,
+                            x->e_mbd.allow_high_precision_mv);
+          if (x->e_mbd.mode_info_context->mbmi.ref_frame[1] > INTRA_FRAME) {
+            mv.row = pi->bmi[i].second_mv.as_mv.row -
+                         second_best_ref_mv->as_mv.row;
+            mv.col = pi->bmi[i].second_mv.as_mv.col -
+                         second_best_ref_mv->as_mv.col;
+            vp9_increment_nmv(&mv, &second_best_ref_mv->as_mv, &cpi->NMVcount,
+                              x->e_mbd.allow_high_precision_mv);
+          }
+        }
+      }
+    }
+  } else if (mbmi->mode == NEWMV) {
+    mv.row = (mbmi->mv[0].as_mv.row - best_ref_mv->as_mv.row);
+    mv.col = (mbmi->mv[0].as_mv.col - best_ref_mv->as_mv.col);
+    vp9_increment_nmv(&mv, &best_ref_mv->as_mv, &cpi->NMVcount,
+                      x->e_mbd.allow_high_precision_mv);
+    if (mbmi->ref_frame[1] > INTRA_FRAME) {
+      mv.row = (mbmi->mv[1].as_mv.row - second_best_ref_mv->as_mv.row);
+      mv.col = (mbmi->mv[1].as_mv.col - second_best_ref_mv->as_mv.col);
+      vp9_increment_nmv(&mv, &second_best_ref_mv->as_mv, &cpi->NMVcount,
+                        x->e_mbd.allow_high_precision_mv);
+    }
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_encodemv.h b/libvpx/vp9/encoder/vp9_encodemv.h
new file mode 100644
index 0000000..cb25d85
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_encodemv.h
@@ -0,0 +1,33 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_ENCODEMV_H_
+#define VP9_ENCODER_VP9_ENCODEMV_H_
+
+#include "vp9/encoder/vp9_onyx_int.h"
+
+void vp9_write_nmv_probs(VP9_COMP* const, int usehp, vp9_writer* const);
+
+void vp9_encode_mv(vp9_writer* w, const MV* mv, const MV* ref,
+                   const nmv_context* mvctx, int usehp);
+
+void vp9_build_nmv_cost_table(int *mvjoint,
+                              int *mvcost[2],
+                              const nmv_context* const mvctx,
+                              int usehp,
+                              int mvc_flag_v,
+                              int mvc_flag_h);
+void vp9_update_nmv_count(VP9_COMP *cpi, MACROBLOCK *x,
+                          int_mv *best_ref_mv, int_mv *second_best_ref_mv);
+
+void print_nmvcounts(nmv_context_counts tnmvcounts);
+
+#endif  // VP9_ENCODER_VP9_ENCODEMV_H_
diff --git a/libvpx/vp9/encoder/vp9_firstpass.c b/libvpx/vp9/encoder/vp9_firstpass.c
new file mode 100644
index 0000000..5e26cd8
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_firstpass.c
@@ -0,0 +1,2648 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "math.h"
+#include "limits.h"
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/encoder/vp9_encodeintra.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/encoder/vp9_firstpass.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/encoder/vp9_encodeframe.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/common/vp9_extend.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/yv12config.h"
+#include <stdio.h>
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/encoder/vp9_ratectrl.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_entropymv.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "./vpx_scale_rtcd.h"
+// TODO(jkoleszar): for setup_dst_planes
+#include "vp9/common/vp9_reconinter.h"
+
+#define OUTPUT_FPF 0
+
+#define IIFACTOR   12.5
+#define IIKFACTOR1 12.5
+#define IIKFACTOR2 15.0
+#define RMAX       512.0
+#define GF_RMAX    96.0
+#define ERR_DIVISOR   150.0
+#define MIN_DECAY_FACTOR 0.1
+
+#define KF_MB_INTRA_MIN 150
+#define GF_MB_INTRA_MIN 100
+
+#define DOUBLE_DIVIDE_CHECK(x) ((x) < 0 ? (x) - 0.000001 : (x) + 0.000001)
+
+#define POW1 (double)cpi->oxcf.two_pass_vbrbias/100.0
+#define POW2 (double)cpi->oxcf.two_pass_vbrbias/100.0
+
+static void swap_yv12(YV12_BUFFER_CONFIG *a, YV12_BUFFER_CONFIG *b) {
+  YV12_BUFFER_CONFIG temp = *a;
+  *a = *b;
+  *b = temp;
+}
+
+static void find_next_key_frame(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame);
+
+static int select_cq_level(int qindex) {
+  int ret_val = QINDEX_RANGE - 1;
+  int i;
+
+  double target_q = (vp9_convert_qindex_to_q(qindex) * 0.5847) + 1.0;
+
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    if (target_q <= vp9_convert_qindex_to_q(i)) {
+      ret_val = i;
+      break;
+    }
+  }
+
+  return ret_val;
+}
+
+
+// Resets the first pass file to the given position using a relative seek from the current position
+static void reset_fpf_position(VP9_COMP *cpi, FIRSTPASS_STATS *position) {
+  cpi->twopass.stats_in = position;
+}
+
+static int lookup_next_frame_stats(VP9_COMP *cpi, FIRSTPASS_STATS *next_frame) {
+  if (cpi->twopass.stats_in >= cpi->twopass.stats_in_end)
+    return EOF;
+
+  *next_frame = *cpi->twopass.stats_in;
+  return 1;
+}
+
+// Read frame stats at an offset from the current position
+static int read_frame_stats(VP9_COMP *cpi,
+                            FIRSTPASS_STATS *frame_stats,
+                            int offset) {
+  FIRSTPASS_STATS *fps_ptr = cpi->twopass.stats_in;
+
+  // Check legality of offset
+  if (offset >= 0) {
+    if (&fps_ptr[offset] >= cpi->twopass.stats_in_end)
+      return EOF;
+  } else if (offset < 0) {
+    if (&fps_ptr[offset] < cpi->twopass.stats_in_start)
+      return EOF;
+  }
+
+  *frame_stats = fps_ptr[offset];
+  return 1;
+}
+
+static int input_stats(VP9_COMP *cpi, FIRSTPASS_STATS *fps) {
+  if (cpi->twopass.stats_in >= cpi->twopass.stats_in_end)
+    return EOF;
+
+  *fps = *cpi->twopass.stats_in;
+  cpi->twopass.stats_in =
+    (void *)((char *)cpi->twopass.stats_in + sizeof(FIRSTPASS_STATS));
+  return 1;
+}
+
+static void output_stats(const VP9_COMP            *cpi,
+                         struct vpx_codec_pkt_list *pktlist,
+                         FIRSTPASS_STATS            *stats) {
+  struct vpx_codec_cx_pkt pkt;
+  pkt.kind = VPX_CODEC_STATS_PKT;
+  pkt.data.twopass_stats.buf = stats;
+  pkt.data.twopass_stats.sz = sizeof(FIRSTPASS_STATS);
+  vpx_codec_pkt_list_add(pktlist, &pkt);
+
+// TEMP debug code
+#if OUTPUT_FPF
+
+  {
+    FILE *fpfile;
+    fpfile = fopen("firstpass.stt", "a");
+
+    fprintf(stdout, "%12.0f %12.0f %12.0f %12.0f %12.0f %12.4f %12.4f"
+            "%12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f"
+            "%12.0f %12.0f %12.4f %12.0f %12.0f %12.4f\n",
+            stats->frame,
+            stats->intra_error,
+            stats->coded_error,
+            stats->sr_coded_error,
+            stats->ssim_weighted_pred_err,
+            stats->pcnt_inter,
+            stats->pcnt_motion,
+            stats->pcnt_second_ref,
+            stats->pcnt_neutral,
+            stats->MVr,
+            stats->mvr_abs,
+            stats->MVc,
+            stats->mvc_abs,
+            stats->MVrv,
+            stats->MVcv,
+            stats->mv_in_out_count,
+            stats->new_mv_count,
+            stats->count,
+            stats->duration);
+    fclose(fpfile);
+  }
+#endif
+}
+
+static void zero_stats(FIRSTPASS_STATS *section) {
+  section->frame      = 0.0;
+  section->intra_error = 0.0;
+  section->coded_error = 0.0;
+  section->sr_coded_error = 0.0;
+  section->ssim_weighted_pred_err = 0.0;
+  section->pcnt_inter  = 0.0;
+  section->pcnt_motion  = 0.0;
+  section->pcnt_second_ref = 0.0;
+  section->pcnt_neutral = 0.0;
+  section->MVr        = 0.0;
+  section->mvr_abs     = 0.0;
+  section->MVc        = 0.0;
+  section->mvc_abs     = 0.0;
+  section->MVrv       = 0.0;
+  section->MVcv       = 0.0;
+  section->mv_in_out_count  = 0.0;
+  section->new_mv_count = 0.0;
+  section->count      = 0.0;
+  section->duration   = 1.0;
+}
+
+static void accumulate_stats(FIRSTPASS_STATS *section, FIRSTPASS_STATS *frame) {
+  section->frame += frame->frame;
+  section->intra_error += frame->intra_error;
+  section->coded_error += frame->coded_error;
+  section->sr_coded_error += frame->sr_coded_error;
+  section->ssim_weighted_pred_err += frame->ssim_weighted_pred_err;
+  section->pcnt_inter  += frame->pcnt_inter;
+  section->pcnt_motion += frame->pcnt_motion;
+  section->pcnt_second_ref += frame->pcnt_second_ref;
+  section->pcnt_neutral += frame->pcnt_neutral;
+  section->MVr        += frame->MVr;
+  section->mvr_abs     += frame->mvr_abs;
+  section->MVc        += frame->MVc;
+  section->mvc_abs     += frame->mvc_abs;
+  section->MVrv       += frame->MVrv;
+  section->MVcv       += frame->MVcv;
+  section->mv_in_out_count  += frame->mv_in_out_count;
+  section->new_mv_count += frame->new_mv_count;
+  section->count      += frame->count;
+  section->duration   += frame->duration;
+}
+
+static void subtract_stats(FIRSTPASS_STATS *section, FIRSTPASS_STATS *frame) {
+  section->frame -= frame->frame;
+  section->intra_error -= frame->intra_error;
+  section->coded_error -= frame->coded_error;
+  section->sr_coded_error -= frame->sr_coded_error;
+  section->ssim_weighted_pred_err -= frame->ssim_weighted_pred_err;
+  section->pcnt_inter  -= frame->pcnt_inter;
+  section->pcnt_motion -= frame->pcnt_motion;
+  section->pcnt_second_ref -= frame->pcnt_second_ref;
+  section->pcnt_neutral -= frame->pcnt_neutral;
+  section->MVr        -= frame->MVr;
+  section->mvr_abs     -= frame->mvr_abs;
+  section->MVc        -= frame->MVc;
+  section->mvc_abs     -= frame->mvc_abs;
+  section->MVrv       -= frame->MVrv;
+  section->MVcv       -= frame->MVcv;
+  section->mv_in_out_count  -= frame->mv_in_out_count;
+  section->new_mv_count -= frame->new_mv_count;
+  section->count      -= frame->count;
+  section->duration   -= frame->duration;
+}
+
+static void avg_stats(FIRSTPASS_STATS *section) {
+  if (section->count < 1.0)
+    return;
+
+  section->intra_error /= section->count;
+  section->coded_error /= section->count;
+  section->sr_coded_error /= section->count;
+  section->ssim_weighted_pred_err /= section->count;
+  section->pcnt_inter  /= section->count;
+  section->pcnt_second_ref /= section->count;
+  section->pcnt_neutral /= section->count;
+  section->pcnt_motion /= section->count;
+  section->MVr        /= section->count;
+  section->mvr_abs     /= section->count;
+  section->MVc        /= section->count;
+  section->mvc_abs     /= section->count;
+  section->MVrv       /= section->count;
+  section->MVcv       /= section->count;
+  section->mv_in_out_count   /= section->count;
+  section->duration   /= section->count;
+}
+
+// Calculate a modified Error used in distributing bits between easier and harder frames
+static double calculate_modified_err(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
+  const FIRSTPASS_STATS *const stats = &cpi->twopass.total_stats;
+  const double av_err = stats->ssim_weighted_pred_err / stats->count;
+  const double this_err = this_frame->ssim_weighted_pred_err;
+  return av_err * pow(this_err / DOUBLE_DIVIDE_CHECK(av_err),
+                      this_err > av_err ? POW1 : POW2);
+}
+
+static const double weight_table[256] = {
+  0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000,
+  0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000,
+  0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000,
+  0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000, 0.020000,
+  0.020000, 0.031250, 0.062500, 0.093750, 0.125000, 0.156250, 0.187500, 0.218750,
+  0.250000, 0.281250, 0.312500, 0.343750, 0.375000, 0.406250, 0.437500, 0.468750,
+  0.500000, 0.531250, 0.562500, 0.593750, 0.625000, 0.656250, 0.687500, 0.718750,
+  0.750000, 0.781250, 0.812500, 0.843750, 0.875000, 0.906250, 0.937500, 0.968750,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000,
+  1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000, 1.000000
+};
+
+static double simple_weight(YV12_BUFFER_CONFIG *source) {
+  int i, j;
+
+  uint8_t *src = source->y_buffer;
+  double sum_weights = 0.0;
+
+  // Loop throught the Y plane raw examining levels and creating a weight for the image
+  i = source->y_height;
+  do {
+    j = source->y_width;
+    do {
+      sum_weights += weight_table[ *src];
+      src++;
+    } while (--j);
+    src -= source->y_width;
+    src += source->y_stride;
+  } while (--i);
+
+  sum_weights /= (source->y_height * source->y_width);
+
+  return sum_weights;
+}
+
+
+// This function returns the current per frame maximum bitrate target.
+static int frame_max_bits(VP9_COMP *cpi) {
+  // Max allocation for a single frame based on the max section guidelines
+  // passed in and how many bits are left.
+  // For VBR base this on the bits and frames left plus the
+  // two_pass_vbrmax_section rate passed in by the user.
+  const double max_bits = (1.0 * cpi->twopass.bits_left /
+      (cpi->twopass.total_stats.count - cpi->common.current_video_frame)) *
+      (cpi->oxcf.two_pass_vbrmax_section / 100.0);
+
+  // Trap case where we are out of bits.
+  return MAX((int)max_bits, 0);
+}
+
+void vp9_init_first_pass(VP9_COMP *cpi) {
+  zero_stats(&cpi->twopass.total_stats);
+}
+
+void vp9_end_first_pass(VP9_COMP *cpi) {
+  output_stats(cpi, cpi->output_pkt_list, &cpi->twopass.total_stats);
+}
+
+static void zz_motion_search(VP9_COMP *cpi, MACROBLOCK *x, YV12_BUFFER_CONFIG *recon_buffer, int *best_motion_err, int recon_yoffset) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  // Set up pointers for this macro block recon buffer
+  xd->plane[0].pre[0].buf = recon_buffer->y_buffer + recon_yoffset;
+
+  switch (xd->mode_info_context->mbmi.sb_type) {
+    case BLOCK_SIZE_SB8X8:
+      vp9_mse8x8(x->plane[0].src.buf, x->plane[0].src.stride,
+                 xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                 (unsigned int *)(best_motion_err));
+      break;
+    case BLOCK_SIZE_SB16X8:
+      vp9_mse16x8(x->plane[0].src.buf, x->plane[0].src.stride,
+                  xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                  (unsigned int *)(best_motion_err));
+      break;
+    case BLOCK_SIZE_SB8X16:
+      vp9_mse8x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                  xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                  (unsigned int *)(best_motion_err));
+      break;
+    default:
+      vp9_mse16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                   xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                   (unsigned int *)(best_motion_err));
+      break;
+  }
+}
+
+static enum BlockSize get_bs(BLOCK_SIZE_TYPE b) {
+  switch (b) {
+    case BLOCK_SIZE_SB8X8:
+      return BLOCK_8X8;
+    case BLOCK_SIZE_SB16X8:
+      return BLOCK_16X8;
+    case BLOCK_SIZE_SB8X16:
+      return BLOCK_8X16;
+    default:
+      return BLOCK_16X16;
+  }
+}
+
+static void first_pass_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+                                     int_mv *ref_mv, MV *best_mv,
+                                     YV12_BUFFER_CONFIG *recon_buffer,
+                                     int *best_motion_err, int recon_yoffset) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int num00;
+
+  int_mv tmp_mv;
+  int_mv ref_mv_full;
+
+  int tmp_err;
+  int step_param = 3;
+  int further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
+  int n;
+  vp9_variance_fn_ptr_t v_fn_ptr =
+      cpi->fn_ptr[get_bs(xd->mode_info_context->mbmi.sb_type)];
+  int new_mv_mode_penalty = 256;
+
+  int sr = 0;
+  int quart_frm = MIN(cpi->common.width, cpi->common.height);
+
+  // refine the motion search range accroding to the frame dimension
+  // for first pass test
+  while ((quart_frm << sr) < MAX_FULL_PEL_VAL)
+    sr++;
+  if (sr)
+    sr--;
+
+  step_param    += sr;
+  further_steps -= sr;
+
+  // override the default variance function to use MSE
+  switch (xd->mode_info_context->mbmi.sb_type) {
+    case BLOCK_SIZE_SB8X8:
+      v_fn_ptr.vf = vp9_mse8x8;
+      break;
+    case BLOCK_SIZE_SB16X8:
+      v_fn_ptr.vf = vp9_mse16x8;
+      break;
+    case BLOCK_SIZE_SB8X16:
+      v_fn_ptr.vf = vp9_mse8x16;
+      break;
+    default:
+      v_fn_ptr.vf = vp9_mse16x16;
+      break;
+  }
+
+  // Set up pointers for this macro block recon buffer
+  xd->plane[0].pre[0].buf = recon_buffer->y_buffer + recon_yoffset;
+
+  // Initial step/diamond search centred on best mv
+  tmp_mv.as_int = 0;
+  ref_mv_full.as_mv.col = ref_mv->as_mv.col >> 3;
+  ref_mv_full.as_mv.row = ref_mv->as_mv.row >> 3;
+  tmp_err = cpi->diamond_search_sad(x, &ref_mv_full, &tmp_mv, step_param,
+                                    x->sadperbit16, &num00, &v_fn_ptr,
+                                    x->nmvjointcost,
+                                    x->mvcost, ref_mv);
+  if (tmp_err < INT_MAX - new_mv_mode_penalty)
+    tmp_err += new_mv_mode_penalty;
+
+  if (tmp_err < *best_motion_err) {
+    *best_motion_err = tmp_err;
+    best_mv->row = tmp_mv.as_mv.row;
+    best_mv->col = tmp_mv.as_mv.col;
+  }
+
+  // Further step/diamond searches as necessary
+  n = num00;
+  num00 = 0;
+
+  while (n < further_steps) {
+    n++;
+
+    if (num00)
+      num00--;
+    else {
+      tmp_err = cpi->diamond_search_sad(x, &ref_mv_full, &tmp_mv,
+                                        step_param + n, x->sadperbit16,
+                                        &num00, &v_fn_ptr,
+                                        x->nmvjointcost,
+                                        x->mvcost, ref_mv);
+      if (tmp_err < INT_MAX - new_mv_mode_penalty)
+        tmp_err += new_mv_mode_penalty;
+
+      if (tmp_err < *best_motion_err) {
+        *best_motion_err = tmp_err;
+        best_mv->row = tmp_mv.as_mv.row;
+        best_mv->col = tmp_mv.as_mv.col;
+      }
+    }
+  }
+}
+
+void vp9_first_pass(VP9_COMP *cpi) {
+  int mb_row, mb_col;
+  MACROBLOCK *const x = &cpi->mb;
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  int recon_yoffset, recon_uvoffset;
+  const int lst_yv12_idx = cm->ref_frame_map[cpi->lst_fb_idx];
+  const int gld_yv12_idx = cm->ref_frame_map[cpi->gld_fb_idx];
+  YV12_BUFFER_CONFIG *const lst_yv12 = &cm->yv12_fb[lst_yv12_idx];
+  YV12_BUFFER_CONFIG *const new_yv12 = &cm->yv12_fb[cm->new_fb_idx];
+  YV12_BUFFER_CONFIG *const gld_yv12 = &cm->yv12_fb[gld_yv12_idx];
+  const int recon_y_stride = lst_yv12->y_stride;
+  const int recon_uv_stride = lst_yv12->uv_stride;
+  int64_t intra_error = 0;
+  int64_t coded_error = 0;
+  int64_t sr_coded_error = 0;
+
+  int sum_mvr = 0, sum_mvc = 0;
+  int sum_mvr_abs = 0, sum_mvc_abs = 0;
+  int sum_mvrs = 0, sum_mvcs = 0;
+  int mvcount = 0;
+  int intercount = 0;
+  int second_ref_count = 0;
+  int intrapenalty = 256;
+  int neutral_count = 0;
+  int new_mv_count = 0;
+  int sum_in_vectors = 0;
+  uint32_t lastmv_as_int = 0;
+
+  int_mv zero_ref_mv;
+
+  zero_ref_mv.as_int = 0;
+
+  vp9_clear_system_state();  // __asm emms;
+
+  vp9_setup_src_planes(x, cpi->Source, 0, 0);
+  setup_pre_planes(xd, lst_yv12, NULL, 0, 0, NULL, NULL);
+  setup_dst_planes(xd, new_yv12, 0, 0);
+
+  x->partition_info = x->pi;
+
+  xd->mode_info_context = cm->mi;
+
+  vp9_build_block_offsets(x);
+
+  vp9_setup_block_dptrs(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
+
+  vp9_frame_init_quantizer(cpi);
+
+  // Initialise the MV cost table to the defaults
+  // if( cm->current_video_frame == 0)
+  // if ( 0 )
+  {
+    vp9_init_mv_probs(cm);
+    vp9_initialize_rd_consts(cpi, cm->base_qindex + cm->y_dc_delta_q);
+  }
+
+  // for each macroblock row in image
+  for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
+    int_mv best_ref_mv;
+
+    best_ref_mv.as_int = 0;
+
+    // reset above block coeffs
+    xd->up_available = (mb_row != 0);
+    recon_yoffset = (mb_row * recon_y_stride * 16);
+    recon_uvoffset = (mb_row * recon_uv_stride * 8);
+
+    // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+    x->mv_row_min = -((mb_row * 16) + (VP9BORDERINPIXELS - 8));
+    x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16)
+                    + (VP9BORDERINPIXELS - 8);
+
+    // for each macroblock col in image
+    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+      int this_error;
+      int gf_motion_error = INT_MAX;
+      int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row);
+
+      xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
+      xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
+      xd->plane[2].dst.buf = new_yv12->v_buffer + recon_uvoffset;
+      xd->left_available = (mb_col != 0);
+
+      if (mb_col * 2 + 1 < cm->mi_cols) {
+        if (mb_row * 2 + 1 < cm->mi_rows) {
+          xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_MB16X16;
+        } else {
+          xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_SB16X8;
+        }
+      } else {
+        if (mb_row * 2 + 1 < cm->mi_rows) {
+          xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_SB8X16;
+        } else {
+          xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_SB8X8;
+        }
+      }
+      xd->mode_info_context->mbmi.ref_frame[0] = INTRA_FRAME;
+      set_mi_row_col(cm, xd,
+                     mb_row << 1,
+                     1 << mi_height_log2(xd->mode_info_context->mbmi.sb_type),
+                     mb_col << 1,
+                     1 << mi_height_log2(xd->mode_info_context->mbmi.sb_type));
+
+      // do intra 16x16 prediction
+      this_error = vp9_encode_intra(cpi, x, use_dc_pred);
+
+      // "intrapenalty" below deals with situations where the intra and inter error scores are very low (eg a plain black frame)
+      // We do not have special cases in first pass for 0,0 and nearest etc so all inter modes carry an overhead cost estimate fot the mv.
+      // When the error score is very low this causes us to pick all or lots of INTRA modes and throw lots of key frames.
+      // This penalty adds a cost matching that of a 0,0 mv to the intra case.
+      this_error += intrapenalty;
+
+      // Cumulative intra error total
+      intra_error += (int64_t)this_error;
+
+      // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+      x->mv_col_min = -((mb_col * 16) + (VP9BORDERINPIXELS - 8));
+      x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16)
+                      + (VP9BORDERINPIXELS - 8);
+
+      // Other than for the first frame do a motion search
+      if (cm->current_video_frame > 0) {
+        int tmp_err;
+        int motion_error = INT_MAX;
+        int_mv mv, tmp_mv;
+
+        // Simple 0,0 motion with no mv overhead
+        zz_motion_search(cpi, x, lst_yv12, &motion_error, recon_yoffset);
+        mv.as_int = tmp_mv.as_int = 0;
+
+        // Test last reference frame using the previous best mv as the
+        // starting point (best reference) for the search
+        first_pass_motion_search(cpi, x, &best_ref_mv,
+                                 &mv.as_mv, lst_yv12,
+                                 &motion_error, recon_yoffset);
+
+        // If the current best reference mv is not centred on 0,0 then do a 0,0 based search as well
+        if (best_ref_mv.as_int) {
+          tmp_err = INT_MAX;
+          first_pass_motion_search(cpi, x, &zero_ref_mv, &tmp_mv.as_mv,
+                                   lst_yv12, &tmp_err, recon_yoffset);
+
+          if (tmp_err < motion_error) {
+            motion_error = tmp_err;
+            mv.as_int = tmp_mv.as_int;
+          }
+        }
+
+        // Experimental search in an older reference frame
+        if (cm->current_video_frame > 1) {
+          // Simple 0,0 motion with no mv overhead
+          zz_motion_search(cpi, x, gld_yv12,
+                           &gf_motion_error, recon_yoffset);
+
+          first_pass_motion_search(cpi, x, &zero_ref_mv,
+                                   &tmp_mv.as_mv, gld_yv12,
+                                   &gf_motion_error, recon_yoffset);
+
+          if ((gf_motion_error < motion_error) &&
+              (gf_motion_error < this_error)) {
+            second_ref_count++;
+          }
+
+          // Reset to last frame as reference buffer
+          xd->plane[0].pre[0].buf = lst_yv12->y_buffer + recon_yoffset;
+          xd->plane[1].pre[0].buf = lst_yv12->u_buffer + recon_uvoffset;
+          xd->plane[2].pre[0].buf = lst_yv12->v_buffer + recon_uvoffset;
+
+          // In accumulating a score for the older reference frame
+          // take the best of the motion predicted score and
+          // the intra coded error (just as will be done for)
+          // accumulation of "coded_error" for the last frame.
+          if (gf_motion_error < this_error)
+            sr_coded_error += gf_motion_error;
+          else
+            sr_coded_error += this_error;
+        } else
+          sr_coded_error += motion_error;
+
+        /* Intra assumed best */
+        best_ref_mv.as_int = 0;
+
+        if (motion_error <= this_error) {
+          // Keep a count of cases where the inter and intra were
+          // very close and very low. This helps with scene cut
+          // detection for example in cropped clips with black bars
+          // at the sides or top and bottom.
+          if ((((this_error - intrapenalty) * 9) <=
+               (motion_error * 10)) &&
+              (this_error < (2 * intrapenalty))) {
+            neutral_count++;
+          }
+
+          mv.as_mv.row <<= 3;
+          mv.as_mv.col <<= 3;
+          this_error = motion_error;
+          vp9_set_mbmode_and_mvs(x, NEWMV, &mv);
+          xd->mode_info_context->mbmi.txfm_size = TX_4X4;
+          xd->mode_info_context->mbmi.ref_frame[0] = LAST_FRAME;
+          xd->mode_info_context->mbmi.ref_frame[1] = NONE;
+          vp9_build_inter_predictors_sby(xd, mb_row << 1,
+                                         mb_col << 1,
+                                         xd->mode_info_context->mbmi.sb_type);
+          vp9_encode_sby(cm, x, xd->mode_info_context->mbmi.sb_type);
+          sum_mvr += mv.as_mv.row;
+          sum_mvr_abs += abs(mv.as_mv.row);
+          sum_mvc += mv.as_mv.col;
+          sum_mvc_abs += abs(mv.as_mv.col);
+          sum_mvrs += mv.as_mv.row * mv.as_mv.row;
+          sum_mvcs += mv.as_mv.col * mv.as_mv.col;
+          intercount++;
+
+          best_ref_mv.as_int = mv.as_int;
+
+          // Was the vector non-zero
+          if (mv.as_int) {
+            mvcount++;
+
+            // Was it different from the last non zero vector
+            if (mv.as_int != lastmv_as_int)
+              new_mv_count++;
+            lastmv_as_int = mv.as_int;
+
+            // Does the Row vector point inwards or outwards
+            if (mb_row < cm->mb_rows / 2) {
+              if (mv.as_mv.row > 0)
+                sum_in_vectors--;
+              else if (mv.as_mv.row < 0)
+                sum_in_vectors++;
+            } else if (mb_row > cm->mb_rows / 2) {
+              if (mv.as_mv.row > 0)
+                sum_in_vectors++;
+              else if (mv.as_mv.row < 0)
+                sum_in_vectors--;
+            }
+
+            // Does the Row vector point inwards or outwards
+            if (mb_col < cm->mb_cols / 2) {
+              if (mv.as_mv.col > 0)
+                sum_in_vectors--;
+              else if (mv.as_mv.col < 0)
+                sum_in_vectors++;
+            } else if (mb_col > cm->mb_cols / 2) {
+              if (mv.as_mv.col > 0)
+                sum_in_vectors++;
+              else if (mv.as_mv.col < 0)
+                sum_in_vectors--;
+            }
+          }
+        }
+      } else
+        sr_coded_error += (int64_t)this_error;
+
+      coded_error += (int64_t)this_error;
+
+      // adjust to the next column of macroblocks
+      x->plane[0].src.buf += 16;
+      x->plane[1].src.buf += 8;
+      x->plane[2].src.buf += 8;
+
+      recon_yoffset += 16;
+      recon_uvoffset += 8;
+    }
+
+    // adjust to the next row of mbs
+    x->plane[0].src.buf += 16 * x->plane[0].src.stride - 16 * cm->mb_cols;
+    x->plane[1].src.buf += 8 * x->plane[1].src.stride - 8 * cm->mb_cols;
+    x->plane[2].src.buf += 8 * x->plane[1].src.stride - 8 * cm->mb_cols;
+
+    vp9_clear_system_state();  // __asm emms;
+  }
+
+  vp9_clear_system_state();  // __asm emms;
+  {
+    double weight = 0.0;
+
+    FIRSTPASS_STATS fps;
+
+    fps.frame      = cm->current_video_frame;
+    fps.intra_error = (double)(intra_error >> 8);
+    fps.coded_error = (double)(coded_error >> 8);
+    fps.sr_coded_error = (double)(sr_coded_error >> 8);
+    weight = simple_weight(cpi->Source);
+
+
+    if (weight < 0.1)
+      weight = 0.1;
+
+    fps.ssim_weighted_pred_err = fps.coded_error * weight;
+
+    fps.pcnt_inter  = 0.0;
+    fps.pcnt_motion = 0.0;
+    fps.MVr        = 0.0;
+    fps.mvr_abs     = 0.0;
+    fps.MVc        = 0.0;
+    fps.mvc_abs     = 0.0;
+    fps.MVrv       = 0.0;
+    fps.MVcv       = 0.0;
+    fps.mv_in_out_count  = 0.0;
+    fps.new_mv_count = 0.0;
+    fps.count      = 1.0;
+
+    fps.pcnt_inter   = 1.0 * (double)intercount / cm->MBs;
+    fps.pcnt_second_ref = 1.0 * (double)second_ref_count / cm->MBs;
+    fps.pcnt_neutral = 1.0 * (double)neutral_count / cm->MBs;
+
+    if (mvcount > 0) {
+      fps.MVr = (double)sum_mvr / (double)mvcount;
+      fps.mvr_abs = (double)sum_mvr_abs / (double)mvcount;
+      fps.MVc = (double)sum_mvc / (double)mvcount;
+      fps.mvc_abs = (double)sum_mvc_abs / (double)mvcount;
+      fps.MVrv = ((double)sum_mvrs - (fps.MVr * fps.MVr / (double)mvcount)) / (double)mvcount;
+      fps.MVcv = ((double)sum_mvcs - (fps.MVc * fps.MVc / (double)mvcount)) / (double)mvcount;
+      fps.mv_in_out_count = (double)sum_in_vectors / (double)(mvcount * 2);
+      fps.new_mv_count = new_mv_count;
+
+      fps.pcnt_motion = 1.0 * (double)mvcount / cpi->common.MBs;
+    }
+
+    // TODO:  handle the case when duration is set to 0, or something less
+    // than the full time between subsequent values of cpi->source_time_stamp.
+    fps.duration = (double)(cpi->source->ts_end
+                            - cpi->source->ts_start);
+
+    // don't want to do output stats with a stack variable!
+    cpi->twopass.this_frame_stats = fps;
+    output_stats(cpi, cpi->output_pkt_list, &cpi->twopass.this_frame_stats);
+    accumulate_stats(&cpi->twopass.total_stats, &fps);
+  }
+
+  // Copy the previous Last Frame back into gf and and arf buffers if
+  // the prediction is good enough... but also dont allow it to lag too far
+  if ((cpi->twopass.sr_update_lag > 3) ||
+      ((cm->current_video_frame > 0) &&
+       (cpi->twopass.this_frame_stats.pcnt_inter > 0.20) &&
+       ((cpi->twopass.this_frame_stats.intra_error /
+         DOUBLE_DIVIDE_CHECK(cpi->twopass.this_frame_stats.coded_error)) >
+        2.0))) {
+    vp8_yv12_copy_frame(lst_yv12, gld_yv12);
+    cpi->twopass.sr_update_lag = 1;
+  } else
+    cpi->twopass.sr_update_lag++;
+
+  // swap frame pointers so last frame refers to the frame we just compressed
+  swap_yv12(lst_yv12, new_yv12);
+
+  vp9_extend_frame_borders(lst_yv12, cm->subsampling_x, cm->subsampling_y);
+
+  // Special case for the first frame. Copy into the GF buffer as a second reference.
+  if (cm->current_video_frame == 0)
+    vp8_yv12_copy_frame(lst_yv12, gld_yv12);
+
+  // use this to see what the first pass reconstruction looks like
+  if (0) {
+    char filename[512];
+    FILE *recon_file;
+    sprintf(filename, "enc%04d.yuv", (int) cm->current_video_frame);
+
+    if (cm->current_video_frame == 0)
+      recon_file = fopen(filename, "wb");
+    else
+      recon_file = fopen(filename, "ab");
+
+    (void)fwrite(lst_yv12->buffer_alloc, lst_yv12->frame_size, 1, recon_file);
+    fclose(recon_file);
+  }
+
+  cm->current_video_frame++;
+
+}
+
+// Estimate a cost per mb attributable to overheads such as the coding of
+// modes and motion vectors.
+// Currently simplistic in its assumptions for testing.
+//
+
+
+static double bitcost(double prob) {
+  return -(log(prob) / log(2.0));
+}
+
+static int64_t estimate_modemvcost(VP9_COMP *cpi,
+                                     FIRSTPASS_STATS *fpstats) {
+#if 0
+  int mv_cost;
+  int mode_cost;
+
+  double av_pct_inter = fpstats->pcnt_inter / fpstats->count;
+  double av_pct_motion = fpstats->pcnt_motion / fpstats->count;
+  double av_intra = (1.0 - av_pct_inter);
+
+  double zz_cost;
+  double motion_cost;
+  double intra_cost;
+
+  zz_cost = bitcost(av_pct_inter - av_pct_motion);
+  motion_cost = bitcost(av_pct_motion);
+  intra_cost = bitcost(av_intra);
+
+  // Estimate of extra bits per mv overhead for mbs
+  // << 9 is the normalization to the (bits * 512) used in vp9_bits_per_mb
+  mv_cost = ((int)(fpstats->new_mv_count / fpstats->count) * 8) << 9;
+
+  // Crude estimate of overhead cost from modes
+  // << 9 is the normalization to (bits * 512) used in vp9_bits_per_mb
+  mode_cost =
+    (int)((((av_pct_inter - av_pct_motion) * zz_cost) +
+           (av_pct_motion * motion_cost) +
+           (av_intra * intra_cost)) * cpi->common.MBs) << 9;
+
+  // return mv_cost + mode_cost;
+  // TODO PGW Fix overhead costs for extended Q range
+#endif
+  return 0;
+}
+
+static double calc_correction_factor(double err_per_mb,
+                                     double err_divisor,
+                                     double pt_low,
+                                     double pt_high,
+                                     int q) {
+  const double error_term = err_per_mb / err_divisor;
+
+  // Adjustment based on actual quantizer to power term.
+  const double power_term = MIN(vp9_convert_qindex_to_q(q) * 0.01 + pt_low,
+                                pt_high);
+
+  // Calculate correction factor
+  if (power_term < 1.0)
+    assert(error_term >= 0.0);
+
+  return fclamp(pow(error_term, power_term), 0.05, 5.0);
+}
+
+// Given a current maxQ value sets a range for future values.
+// PGW TODO..
+// This code removes direct dependency on QIndex to determine the range
+// (now uses the actual quantizer) but has not been tuned.
+static void adjust_maxq_qrange(VP9_COMP *cpi) {
+  int i;
+  // Set the max corresponding to cpi->avg_q * 2.0
+  double q = cpi->avg_q * 2.0;
+  cpi->twopass.maxq_max_limit = cpi->worst_quality;
+  for (i = cpi->best_quality; i <= cpi->worst_quality; i++) {
+    cpi->twopass.maxq_max_limit = i;
+    if (vp9_convert_qindex_to_q(i) >= q)
+      break;
+  }
+
+  // Set the min corresponding to cpi->avg_q * 0.5
+  q = cpi->avg_q * 0.5;
+  cpi->twopass.maxq_min_limit = cpi->best_quality;
+  for (i = cpi->worst_quality; i >= cpi->best_quality; i--) {
+    cpi->twopass.maxq_min_limit = i;
+    if (vp9_convert_qindex_to_q(i) <= q)
+      break;
+  }
+}
+
+static int estimate_max_q(VP9_COMP *cpi,
+                          FIRSTPASS_STATS *fpstats,
+                          int section_target_bandwitdh) {
+  int q;
+  int num_mbs = cpi->common.MBs;
+  int target_norm_bits_per_mb;
+
+  double section_err = fpstats->coded_error / fpstats->count;
+  double sr_correction;
+  double err_per_mb = section_err / num_mbs;
+  double err_correction_factor;
+  double speed_correction = 1.0;
+
+  if (section_target_bandwitdh <= 0)
+    return cpi->twopass.maxq_max_limit;          // Highest value allowed
+
+  target_norm_bits_per_mb = section_target_bandwitdh < (1 << 20)
+                              ? (512 * section_target_bandwitdh) / num_mbs
+                              : 512 * (section_target_bandwitdh / num_mbs);
+
+  // Look at the drop in prediction quality between the last frame
+  // and the GF buffer (which contained an older frame).
+  if (fpstats->sr_coded_error > fpstats->coded_error) {
+    double sr_err_diff = (fpstats->sr_coded_error - fpstats->coded_error) /
+                             (fpstats->count * cpi->common.MBs);
+    sr_correction = fclamp(pow(sr_err_diff / 32.0, 0.25), 0.75, 1.25);
+  } else {
+    sr_correction = 0.75;
+  }
+
+  // Calculate a corrective factor based on a rolling ratio of bits spent
+  // vs target bits
+  if (cpi->rolling_target_bits > 0 &&
+      cpi->active_worst_quality < cpi->worst_quality) {
+    double rolling_ratio = (double)cpi->rolling_actual_bits /
+                               (double)cpi->rolling_target_bits;
+
+    if (rolling_ratio < 0.95)
+      cpi->twopass.est_max_qcorrection_factor -= 0.005;
+    else if (rolling_ratio > 1.05)
+      cpi->twopass.est_max_qcorrection_factor += 0.005;
+
+    cpi->twopass.est_max_qcorrection_factor = fclamp(
+        cpi->twopass.est_max_qcorrection_factor, 0.1, 10.0);
+  }
+
+  // Corrections for higher compression speed settings
+  // (reduced compression expected)
+  if (cpi->compressor_speed == 1)
+    speed_correction = cpi->oxcf.cpu_used <= 5 ?
+                          1.04 + (cpi->oxcf.cpu_used * 0.04) :
+                          1.25;
+
+  // Try and pick a max Q that will be high enough to encode the
+  // content at the given rate.
+  for (q = cpi->twopass.maxq_min_limit; q < cpi->twopass.maxq_max_limit; q++) {
+    int bits_per_mb_at_this_q;
+
+    err_correction_factor = calc_correction_factor(err_per_mb,
+                                                   ERR_DIVISOR, 0.4, 0.90, q) *
+                                sr_correction * speed_correction *
+                                cpi->twopass.est_max_qcorrection_factor;
+
+    bits_per_mb_at_this_q = vp9_bits_per_mb(INTER_FRAME, q,
+                                            err_correction_factor);
+
+    if (bits_per_mb_at_this_q <= target_norm_bits_per_mb)
+      break;
+  }
+
+  // Restriction on active max q for constrained quality mode.
+  if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY &&
+      q < cpi->cq_target_quality)
+    q = cpi->cq_target_quality;
+
+  // Adjust maxq_min_limit and maxq_max_limit limits based on
+  // average q observed in clip for non kf/gf/arf frames
+  // Give average a chance to settle though.
+  // PGW TODO.. This code is broken for the extended Q range
+  if (cpi->ni_frames > ((int)cpi->twopass.total_stats.count >> 8) &&
+      cpi->ni_frames > 25)
+    adjust_maxq_qrange(cpi);
+
+  return q;
+}
+
+// For cq mode estimate a cq level that matches the observed
+// complexity and data rate.
+static int estimate_cq(VP9_COMP *cpi,
+                       FIRSTPASS_STATS *fpstats,
+                       int section_target_bandwitdh) {
+  int q;
+  int num_mbs = cpi->common.MBs;
+  int target_norm_bits_per_mb;
+
+  double section_err = (fpstats->coded_error / fpstats->count);
+  double err_per_mb = section_err / num_mbs;
+  double err_correction_factor;
+  double sr_err_diff;
+  double sr_correction;
+  double speed_correction = 1.0;
+  double clip_iiratio;
+  double clip_iifactor;
+
+  target_norm_bits_per_mb = (section_target_bandwitdh < (1 << 20))
+                            ? (512 * section_target_bandwitdh) / num_mbs
+                            : 512 * (section_target_bandwitdh / num_mbs);
+
+
+  // Corrections for higher compression speed settings
+  // (reduced compression expected)
+  if (cpi->compressor_speed == 1) {
+    if (cpi->oxcf.cpu_used <= 5)
+      speed_correction = 1.04 + (cpi->oxcf.cpu_used * 0.04);
+    else
+      speed_correction = 1.25;
+  }
+
+  // Look at the drop in prediction quality between the last frame
+  // and the GF buffer (which contained an older frame).
+  if (fpstats->sr_coded_error > fpstats->coded_error) {
+    sr_err_diff =
+      (fpstats->sr_coded_error - fpstats->coded_error) /
+      (fpstats->count * cpi->common.MBs);
+    sr_correction = (sr_err_diff / 32.0);
+    sr_correction = pow(sr_correction, 0.25);
+    if (sr_correction < 0.75)
+      sr_correction = 0.75;
+    else if (sr_correction > 1.25)
+      sr_correction = 1.25;
+  } else {
+    sr_correction = 0.75;
+  }
+
+  // II ratio correction factor for clip as a whole
+  clip_iiratio = cpi->twopass.total_stats.intra_error /
+                 DOUBLE_DIVIDE_CHECK(cpi->twopass.total_stats.coded_error);
+  clip_iifactor = 1.0 - ((clip_iiratio - 10.0) * 0.025);
+  if (clip_iifactor < 0.80)
+    clip_iifactor = 0.80;
+
+  // Try and pick a Q that can encode the content at the given rate.
+  for (q = 0; q < MAXQ; q++) {
+    int bits_per_mb_at_this_q;
+
+    // Error per MB based correction factor
+    err_correction_factor =
+      calc_correction_factor(err_per_mb, 100.0, 0.4, 0.90, q) *
+      sr_correction * speed_correction * clip_iifactor;
+
+    bits_per_mb_at_this_q =
+      vp9_bits_per_mb(INTER_FRAME, q, err_correction_factor);
+
+    if (bits_per_mb_at_this_q <= target_norm_bits_per_mb)
+      break;
+  }
+
+  // Clip value to range "best allowed to (worst allowed - 1)"
+  q = select_cq_level(q);
+  if (q >= cpi->worst_quality)
+    q = cpi->worst_quality - 1;
+  if (q < cpi->best_quality)
+    q = cpi->best_quality;
+
+  return q;
+}
+
+
+extern void vp9_new_frame_rate(VP9_COMP *cpi, double framerate);
+
+void vp9_init_second_pass(VP9_COMP *cpi) {
+  FIRSTPASS_STATS this_frame;
+  FIRSTPASS_STATS *start_pos;
+
+  double lower_bounds_min_rate = FRAME_OVERHEAD_BITS * cpi->oxcf.frame_rate;
+  double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth
+                                      * cpi->oxcf.two_pass_vbrmin_section / 100);
+
+  if (two_pass_min_rate < lower_bounds_min_rate)
+    two_pass_min_rate = lower_bounds_min_rate;
+
+  zero_stats(&cpi->twopass.total_stats);
+  zero_stats(&cpi->twopass.total_left_stats);
+
+  if (!cpi->twopass.stats_in_end)
+    return;
+
+  cpi->twopass.total_stats = *cpi->twopass.stats_in_end;
+  cpi->twopass.total_left_stats = cpi->twopass.total_stats;
+
+  // each frame can have a different duration, as the frame rate in the source
+  // isn't guaranteed to be constant.   The frame rate prior to the first frame
+  // encoded in the second pass is a guess.  However the sum duration is not.
+  // Its calculated based on the actual durations of all frames from the first
+  // pass.
+  vp9_new_frame_rate(cpi, 10000000.0 * cpi->twopass.total_stats.count /
+                       cpi->twopass.total_stats.duration);
+
+  cpi->output_frame_rate = cpi->oxcf.frame_rate;
+  cpi->twopass.bits_left = (int64_t)(cpi->twopass.total_stats.duration *
+                                     cpi->oxcf.target_bandwidth / 10000000.0);
+  cpi->twopass.bits_left -= (int64_t)(cpi->twopass.total_stats.duration *
+                                      two_pass_min_rate / 10000000.0);
+
+  // Calculate a minimum intra value to be used in determining the IIratio
+  // scores used in the second pass. We have this minimum to make sure
+  // that clips that are static but "low complexity" in the intra domain
+  // are still boosted appropriately for KF/GF/ARF
+  cpi->twopass.kf_intra_err_min = KF_MB_INTRA_MIN * cpi->common.MBs;
+  cpi->twopass.gf_intra_err_min = GF_MB_INTRA_MIN * cpi->common.MBs;
+
+  // This variable monitors how far behind the second ref update is lagging
+  cpi->twopass.sr_update_lag = 1;
+
+  // Scan the first pass file and calculate an average Intra / Inter error score ratio for the sequence
+  {
+    double sum_iiratio = 0.0;
+    double IIRatio;
+
+    start_pos = cpi->twopass.stats_in;               // Note starting "file" position
+
+    while (input_stats(cpi, &this_frame) != EOF) {
+      IIRatio = this_frame.intra_error / DOUBLE_DIVIDE_CHECK(this_frame.coded_error);
+      IIRatio = (IIRatio < 1.0) ? 1.0 : (IIRatio > 20.0) ? 20.0 : IIRatio;
+      sum_iiratio += IIRatio;
+    }
+
+    cpi->twopass.avg_iiratio = sum_iiratio /
+        DOUBLE_DIVIDE_CHECK((double)cpi->twopass.total_stats.count);
+
+    // Reset file position
+    reset_fpf_position(cpi, start_pos);
+  }
+
+  // Scan the first pass file and calculate a modified total error based upon the bias/power function
+  // used to allocate bits
+  {
+    start_pos = cpi->twopass.stats_in;               // Note starting "file" position
+
+    cpi->twopass.modified_error_total = 0.0;
+    cpi->twopass.modified_error_used = 0.0;
+
+    while (input_stats(cpi, &this_frame) != EOF) {
+      cpi->twopass.modified_error_total += calculate_modified_err(cpi, &this_frame);
+    }
+    cpi->twopass.modified_error_left = cpi->twopass.modified_error_total;
+
+    reset_fpf_position(cpi, start_pos);            // Reset file position
+
+  }
+}
+
+void vp9_end_second_pass(VP9_COMP *cpi) {
+}
+
+// This function gives and estimate of how badly we believe
+// the prediction quality is decaying from frame to frame.
+static double get_prediction_decay_rate(VP9_COMP *cpi,
+                                        FIRSTPASS_STATS *next_frame) {
+  double prediction_decay_rate;
+  double second_ref_decay;
+  double mb_sr_err_diff;
+
+  // Initial basis is the % mbs inter coded
+  prediction_decay_rate = next_frame->pcnt_inter;
+
+  // Look at the observed drop in prediction quality between the last frame
+  // and the GF buffer (which contains an older frame).
+  mb_sr_err_diff = (next_frame->sr_coded_error - next_frame->coded_error) /
+                   cpi->common.MBs;
+  if (mb_sr_err_diff <= 512.0) {
+    second_ref_decay = 1.0 - (mb_sr_err_diff / 512.0);
+    second_ref_decay = pow(second_ref_decay, 0.5);
+    if (second_ref_decay < 0.85)
+      second_ref_decay = 0.85;
+    else if (second_ref_decay > 1.0)
+      second_ref_decay = 1.0;
+  } else {
+    second_ref_decay = 0.85;
+  }
+
+  if (second_ref_decay < prediction_decay_rate)
+    prediction_decay_rate = second_ref_decay;
+
+  return prediction_decay_rate;
+}
+
+// Function to test for a condition where a complex transition is followed
+// by a static section. For example in slide shows where there is a fade
+// between slides. This is to help with more optimal kf and gf positioning.
+static int detect_transition_to_still(
+  VP9_COMP *cpi,
+  int frame_interval,
+  int still_interval,
+  double loop_decay_rate,
+  double last_decay_rate) {
+  int trans_to_still = 0;
+
+  // Break clause to detect very still sections after motion
+  // For example a static image after a fade or other transition
+  // instead of a clean scene cut.
+  if (frame_interval > MIN_GF_INTERVAL &&
+      loop_decay_rate >= 0.999 &&
+      last_decay_rate < 0.9) {
+    int j;
+    FIRSTPASS_STATS *position = cpi->twopass.stats_in;
+    FIRSTPASS_STATS tmp_next_frame;
+    double zz_inter;
+
+    // Look ahead a few frames to see if static condition
+    // persists...
+    for (j = 0; j < still_interval; j++) {
+      if (EOF == input_stats(cpi, &tmp_next_frame))
+        break;
+
+      zz_inter =
+        (tmp_next_frame.pcnt_inter - tmp_next_frame.pcnt_motion);
+      if (zz_inter < 0.999)
+        break;
+    }
+    // Reset file position
+    reset_fpf_position(cpi, position);
+
+    // Only if it does do we signal a transition to still
+    if (j == still_interval)
+      trans_to_still = 1;
+  }
+
+  return trans_to_still;
+}
+
+// This function detects a flash through the high relative pcnt_second_ref
+// score in the frame following a flash frame. The offset passed in should
+// reflect this
+static int detect_flash(VP9_COMP *cpi, int offset) {
+  FIRSTPASS_STATS next_frame;
+
+  int flash_detected = 0;
+
+  // Read the frame data.
+  // The return is FALSE (no flash detected) if not a valid frame
+  if (read_frame_stats(cpi, &next_frame, offset) != EOF) {
+    // What we are looking for here is a situation where there is a
+    // brief break in prediction (such as a flash) but subsequent frames
+    // are reasonably well predicted by an earlier (pre flash) frame.
+    // The recovery after a flash is indicated by a high pcnt_second_ref
+    // comapred to pcnt_inter.
+    if (next_frame.pcnt_second_ref > next_frame.pcnt_inter &&
+        next_frame.pcnt_second_ref >= 0.5)
+      flash_detected = 1;
+  }
+
+  return flash_detected;
+}
+
+// Update the motion related elements to the GF arf boost calculation
+static void accumulate_frame_motion_stats(
+  VP9_COMP *cpi,
+  FIRSTPASS_STATS *this_frame,
+  double *this_frame_mv_in_out,
+  double *mv_in_out_accumulator,
+  double *abs_mv_in_out_accumulator,
+  double *mv_ratio_accumulator) {
+  // double this_frame_mv_in_out;
+  double this_frame_mvr_ratio;
+  double this_frame_mvc_ratio;
+  double motion_pct;
+
+  // Accumulate motion stats.
+  motion_pct = this_frame->pcnt_motion;
+
+  // Accumulate Motion In/Out of frame stats
+  *this_frame_mv_in_out = this_frame->mv_in_out_count * motion_pct;
+  *mv_in_out_accumulator += this_frame->mv_in_out_count * motion_pct;
+  *abs_mv_in_out_accumulator +=
+    fabs(this_frame->mv_in_out_count * motion_pct);
+
+  // Accumulate a measure of how uniform (or conversely how random)
+  // the motion field is. (A ratio of absmv / mv)
+  if (motion_pct > 0.05) {
+    this_frame_mvr_ratio = fabs(this_frame->mvr_abs) /
+                           DOUBLE_DIVIDE_CHECK(fabs(this_frame->MVr));
+
+    this_frame_mvc_ratio = fabs(this_frame->mvc_abs) /
+                           DOUBLE_DIVIDE_CHECK(fabs(this_frame->MVc));
+
+    *mv_ratio_accumulator +=
+      (this_frame_mvr_ratio < this_frame->mvr_abs)
+      ? (this_frame_mvr_ratio * motion_pct)
+      : this_frame->mvr_abs * motion_pct;
+
+    *mv_ratio_accumulator +=
+      (this_frame_mvc_ratio < this_frame->mvc_abs)
+      ? (this_frame_mvc_ratio * motion_pct)
+      : this_frame->mvc_abs * motion_pct;
+
+  }
+}
+
+// Calculate a baseline boost number for the current frame.
+static double calc_frame_boost(
+  VP9_COMP *cpi,
+  FIRSTPASS_STATS *this_frame,
+  double this_frame_mv_in_out) {
+  double frame_boost;
+
+  // Underlying boost factor is based on inter intra error ratio
+  if (this_frame->intra_error > cpi->twopass.gf_intra_err_min)
+    frame_boost = (IIFACTOR * this_frame->intra_error /
+                   DOUBLE_DIVIDE_CHECK(this_frame->coded_error));
+  else
+    frame_boost = (IIFACTOR * cpi->twopass.gf_intra_err_min /
+                   DOUBLE_DIVIDE_CHECK(this_frame->coded_error));
+
+  // Increase boost for frames where new data coming into frame
+  // (eg zoom out). Slightly reduce boost if there is a net balance
+  // of motion out of the frame (zoom in).
+  // The range for this_frame_mv_in_out is -1.0 to +1.0
+  if (this_frame_mv_in_out > 0.0)
+    frame_boost += frame_boost * (this_frame_mv_in_out * 2.0);
+  // In extreme case boost is halved
+  else
+    frame_boost += frame_boost * (this_frame_mv_in_out / 2.0);
+
+  // Clip to maximum
+  if (frame_boost > GF_RMAX)
+    frame_boost = GF_RMAX;
+
+  return frame_boost;
+}
+
+static int calc_arf_boost(VP9_COMP *cpi, int offset,
+                          int f_frames, int b_frames,
+                          int *f_boost, int *b_boost) {
+  FIRSTPASS_STATS this_frame;
+
+  int i;
+  double boost_score = 0.0;
+  double mv_ratio_accumulator = 0.0;
+  double decay_accumulator = 1.0;
+  double this_frame_mv_in_out = 0.0;
+  double mv_in_out_accumulator = 0.0;
+  double abs_mv_in_out_accumulator = 0.0;
+  int arf_boost;
+  int flash_detected = 0;
+
+  // Search forward from the proposed arf/next gf position
+  for (i = 0; i < f_frames; i++) {
+    if (read_frame_stats(cpi, &this_frame, (i + offset)) == EOF)
+      break;
+
+    // Update the motion related elements to the boost calculation
+    accumulate_frame_motion_stats(cpi, &this_frame,
+                                  &this_frame_mv_in_out, &mv_in_out_accumulator,
+                                  &abs_mv_in_out_accumulator, &mv_ratio_accumulator);
+
+    // We want to discount the flash frame itself and the recovery
+    // frame that follows as both will have poor scores.
+    flash_detected = detect_flash(cpi, (i + offset)) ||
+                     detect_flash(cpi, (i + offset + 1));
+
+    // Cumulative effect of prediction quality decay
+    if (!flash_detected) {
+      decay_accumulator *= get_prediction_decay_rate(cpi, &this_frame);
+      decay_accumulator = decay_accumulator < MIN_DECAY_FACTOR
+                          ? MIN_DECAY_FACTOR : decay_accumulator;
+    }
+
+    boost_score += (decay_accumulator *
+                    calc_frame_boost(cpi, &this_frame, this_frame_mv_in_out));
+  }
+
+  *f_boost = (int)boost_score;
+
+  // Reset for backward looking loop
+  boost_score = 0.0;
+  mv_ratio_accumulator = 0.0;
+  decay_accumulator = 1.0;
+  this_frame_mv_in_out = 0.0;
+  mv_in_out_accumulator = 0.0;
+  abs_mv_in_out_accumulator = 0.0;
+
+  // Search backward towards last gf position
+  for (i = -1; i >= -b_frames; i--) {
+    if (read_frame_stats(cpi, &this_frame, (i + offset)) == EOF)
+      break;
+
+    // Update the motion related elements to the boost calculation
+    accumulate_frame_motion_stats(cpi, &this_frame,
+                                  &this_frame_mv_in_out, &mv_in_out_accumulator,
+                                  &abs_mv_in_out_accumulator, &mv_ratio_accumulator);
+
+    // We want to discount the the flash frame itself and the recovery
+    // frame that follows as both will have poor scores.
+    flash_detected = detect_flash(cpi, (i + offset)) ||
+                     detect_flash(cpi, (i + offset + 1));
+
+    // Cumulative effect of prediction quality decay
+    if (!flash_detected) {
+      decay_accumulator *= get_prediction_decay_rate(cpi, &this_frame);
+      decay_accumulator = decay_accumulator < MIN_DECAY_FACTOR
+                              ? MIN_DECAY_FACTOR : decay_accumulator;
+    }
+
+    boost_score += (decay_accumulator *
+                    calc_frame_boost(cpi, &this_frame, this_frame_mv_in_out));
+
+  }
+  *b_boost = (int)boost_score;
+
+  arf_boost = (*f_boost + *b_boost);
+  if (arf_boost < ((b_frames + f_frames) * 20))
+    arf_boost = ((b_frames + f_frames) * 20);
+
+  return arf_boost;
+}
+
+#if CONFIG_MULTIPLE_ARF
+// Work out the frame coding order for a GF or an ARF group.
+// The current implementation codes frames in their natural order for a
+// GF group, and inserts additional ARFs into an ARF group using a
+// binary split approach.
+// NOTE: this function is currently implemented recursively.
+static void schedule_frames(VP9_COMP *cpi, const int start, const int end,
+                            const int arf_idx, const int gf_or_arf_group,
+                            const int level) {
+  int i, abs_end, half_range;
+  int *cfo = cpi->frame_coding_order;
+  int idx = cpi->new_frame_coding_order_period;
+
+  // If (end < 0) an ARF should be coded at position (-end).
+  assert(start >= 0);
+
+  // printf("start:%d end:%d\n", start, end);
+
+  // GF Group: code frames in logical order.
+  if (gf_or_arf_group == 0) {
+    assert(end >= start);
+    for (i = start; i <= end; ++i) {
+      cfo[idx] = i;
+      cpi->arf_buffer_idx[idx] = arf_idx;
+      cpi->arf_weight[idx] = -1;
+      ++idx;
+    }
+    cpi->new_frame_coding_order_period = idx;
+    return;
+  }
+
+  // ARF Group: work out the ARF schedule.
+  // Mark ARF frames as negative.
+  if (end < 0) {
+    // printf("start:%d end:%d\n", -end, -end);
+    // ARF frame is at the end of the range.
+    cfo[idx] = end;
+    // What ARF buffer does this ARF use as predictor.
+    cpi->arf_buffer_idx[idx] = (arf_idx > 2) ? (arf_idx - 1) : 2;
+    cpi->arf_weight[idx] = level;
+    ++idx;
+    abs_end = -end;
+  } else {
+    abs_end = end;
+  }
+
+  half_range = (abs_end - start) >> 1;
+
+  // ARFs may not be adjacent, they must be separated by at least
+  // MIN_GF_INTERVAL non-ARF frames.
+  if ((start + MIN_GF_INTERVAL) >= (abs_end - MIN_GF_INTERVAL)) {
+    // printf("start:%d end:%d\n", start, abs_end);
+    // Update the coding order and active ARF.
+    for (i = start; i <= abs_end; ++i) {
+      cfo[idx] = i;
+      cpi->arf_buffer_idx[idx] = arf_idx;
+      cpi->arf_weight[idx] = -1;
+      ++idx;
+    }
+    cpi->new_frame_coding_order_period = idx;
+  } else {
+    // Place a new ARF at the mid-point of the range.
+    cpi->new_frame_coding_order_period = idx;
+    schedule_frames(cpi, start, -(start + half_range), arf_idx + 1,
+                    gf_or_arf_group, level + 1);
+    schedule_frames(cpi, start + half_range + 1, abs_end, arf_idx,
+                    gf_or_arf_group, level + 1);
+  }
+}
+
+#define FIXED_ARF_GROUP_SIZE 16
+
+void define_fixed_arf_period(VP9_COMP *cpi) {
+  int i;
+  int max_level = INT_MIN;
+
+  assert(cpi->multi_arf_enabled);
+  assert(cpi->oxcf.lag_in_frames >= FIXED_ARF_GROUP_SIZE);
+
+  // Save the weight of the last frame in the sequence before next
+  // sequence pattern overwrites it.
+  cpi->this_frame_weight = cpi->arf_weight[cpi->sequence_number];
+  assert(cpi->this_frame_weight >= 0);
+
+  // Initialize frame coding order variables.
+  cpi->new_frame_coding_order_period = 0;
+  cpi->next_frame_in_order = 0;
+  cpi->arf_buffered = 0;
+  vp9_zero(cpi->frame_coding_order);
+  vp9_zero(cpi->arf_buffer_idx);
+  vpx_memset(cpi->arf_weight, -1, sizeof(cpi->arf_weight));
+
+  if (cpi->twopass.frames_to_key <= (FIXED_ARF_GROUP_SIZE + 8)) {
+    // Setup a GF group close to the keyframe.
+    cpi->source_alt_ref_pending = 0;
+    cpi->baseline_gf_interval = cpi->twopass.frames_to_key;
+    schedule_frames(cpi, 0, (cpi->baseline_gf_interval - 1), 2, 0, 0);
+  } else {
+    // Setup a fixed period ARF group.
+    cpi->source_alt_ref_pending = 1;
+    cpi->baseline_gf_interval = FIXED_ARF_GROUP_SIZE;
+    schedule_frames(cpi, 0, -(cpi->baseline_gf_interval - 1), 2, 1, 0);
+  }
+
+  // Replace level indicator of -1 with correct level.
+  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+    if (cpi->arf_weight[i] > max_level) {
+      max_level = cpi->arf_weight[i];
+    }
+  }
+  ++max_level;
+  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+    if (cpi->arf_weight[i] == -1) {
+      cpi->arf_weight[i] = max_level;
+    }
+  }
+  cpi->max_arf_level = max_level;
+#if 0
+  printf("\nSchedule: ");
+  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+    printf("%4d ", cpi->frame_coding_order[i]);
+  }
+  printf("\n");
+  printf("ARFref:   ");
+  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+    printf("%4d ", cpi->arf_buffer_idx[i]);
+  }
+  printf("\n");
+  printf("Weight:   ");
+  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+    printf("%4d ", cpi->arf_weight[i]);
+  }
+  printf("\n");
+#endif
+}
+#endif
+
+// Analyse and define a gf/arf group.
+static void define_gf_group(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
+  FIRSTPASS_STATS next_frame;
+  FIRSTPASS_STATS *start_pos;
+  int i;
+  double boost_score = 0.0;
+  double old_boost_score = 0.0;
+  double gf_group_err = 0.0;
+  double gf_first_frame_err = 0.0;
+  double mod_frame_err = 0.0;
+
+  double mv_ratio_accumulator = 0.0;
+  double decay_accumulator = 1.0;
+  double zero_motion_accumulator = 1.0;
+
+  double loop_decay_rate = 1.00;          // Starting decay rate
+  double last_loop_decay_rate = 1.00;
+
+  double this_frame_mv_in_out = 0.0;
+  double mv_in_out_accumulator = 0.0;
+  double abs_mv_in_out_accumulator = 0.0;
+  double mv_ratio_accumulator_thresh;
+  int max_bits = frame_max_bits(cpi);     // Max for a single frame
+
+  unsigned int allow_alt_ref =
+    cpi->oxcf.play_alternate && cpi->oxcf.lag_in_frames;
+
+  int f_boost = 0;
+  int b_boost = 0;
+  int flash_detected;
+  int active_max_gf_interval;
+
+  cpi->twopass.gf_group_bits = 0;
+
+  vp9_clear_system_state();  // __asm emms;
+
+  start_pos = cpi->twopass.stats_in;
+
+  vpx_memset(&next_frame, 0, sizeof(next_frame)); // assure clean
+
+  // Load stats for the current frame.
+  mod_frame_err = calculate_modified_err(cpi, this_frame);
+
+  // Note the error of the frame at the start of the group (this will be
+  // the GF frame error if we code a normal gf
+  gf_first_frame_err = mod_frame_err;
+
+  // Special treatment if the current frame is a key frame (which is also
+  // a gf). If it is then its error score (and hence bit allocation) need
+  // to be subtracted out from the calculation for the GF group
+  if (cpi->common.frame_type == KEY_FRAME)
+    gf_group_err -= gf_first_frame_err;
+
+  // Motion breakout threshold for loop below depends on image size.
+  mv_ratio_accumulator_thresh = (cpi->common.width + cpi->common.height) / 10.0;
+
+  // Work out a maximum interval for the GF.
+  // If the image appears completely static we can extend beyond this.
+  // The value chosen depends on the active Q range. At low Q we have
+  // bits to spare and are better with a smaller interval and smaller boost.
+  // At high Q when there are few bits to spare we are better with a longer
+  // interval to spread the cost of the GF.
+  active_max_gf_interval =
+    12 + ((int)vp9_convert_qindex_to_q(cpi->active_worst_quality) >> 5);
+
+  if (active_max_gf_interval > cpi->max_gf_interval)
+    active_max_gf_interval = cpi->max_gf_interval;
+
+  i = 0;
+  while (((i < cpi->twopass.static_scene_max_gf_interval) ||
+          ((cpi->twopass.frames_to_key - i) < MIN_GF_INTERVAL)) &&
+         (i < cpi->twopass.frames_to_key)) {
+    i++;    // Increment the loop counter
+
+    // Accumulate error score of frames in this gf group
+    mod_frame_err = calculate_modified_err(cpi, this_frame);
+    gf_group_err += mod_frame_err;
+
+    if (EOF == input_stats(cpi, &next_frame))
+      break;
+
+    // Test for the case where there is a brief flash but the prediction
+    // quality back to an earlier frame is then restored.
+    flash_detected = detect_flash(cpi, 0);
+
+    // Update the motion related elements to the boost calculation
+    accumulate_frame_motion_stats(cpi, &next_frame,
+                                  &this_frame_mv_in_out, &mv_in_out_accumulator,
+                                  &abs_mv_in_out_accumulator, &mv_ratio_accumulator);
+
+    // Cumulative effect of prediction quality decay
+    if (!flash_detected) {
+      last_loop_decay_rate = loop_decay_rate;
+      loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame);
+      decay_accumulator = decay_accumulator * loop_decay_rate;
+
+      // Monitor for static sections.
+      if ((next_frame.pcnt_inter - next_frame.pcnt_motion) <
+          zero_motion_accumulator) {
+        zero_motion_accumulator =
+          (next_frame.pcnt_inter - next_frame.pcnt_motion);
+      }
+
+      // Break clause to detect very still sections after motion
+      // (for example a static image after a fade or other transition).
+      if (detect_transition_to_still(cpi, i, 5, loop_decay_rate,
+                                     last_loop_decay_rate)) {
+        allow_alt_ref = 0;
+        break;
+      }
+    }
+
+    // Calculate a boost number for this frame
+    boost_score +=
+      (decay_accumulator *
+       calc_frame_boost(cpi, &next_frame, this_frame_mv_in_out));
+
+    // Break out conditions.
+    if (
+      // Break at cpi->max_gf_interval unless almost totally static
+      (i >= active_max_gf_interval && (zero_motion_accumulator < 0.995)) ||
+      (
+        // Don't break out with a very short interval
+        (i > MIN_GF_INTERVAL) &&
+        // Don't break out very close to a key frame
+        ((cpi->twopass.frames_to_key - i) >= MIN_GF_INTERVAL) &&
+        ((boost_score > 125.0) || (next_frame.pcnt_inter < 0.75)) &&
+        (!flash_detected) &&
+        ((mv_ratio_accumulator > mv_ratio_accumulator_thresh) ||
+         (abs_mv_in_out_accumulator > 3.0) ||
+         (mv_in_out_accumulator < -2.0) ||
+         ((boost_score - old_boost_score) < IIFACTOR))
+      )) {
+      boost_score = old_boost_score;
+      break;
+    }
+
+    *this_frame = next_frame;
+
+    old_boost_score = boost_score;
+  }
+
+  // Don't allow a gf too near the next kf
+  if ((cpi->twopass.frames_to_key - i) < MIN_GF_INTERVAL) {
+    while (i < cpi->twopass.frames_to_key) {
+      i++;
+
+      if (EOF == input_stats(cpi, this_frame))
+        break;
+
+      if (i < cpi->twopass.frames_to_key) {
+        mod_frame_err = calculate_modified_err(cpi, this_frame);
+        gf_group_err += mod_frame_err;
+      }
+    }
+  }
+
+  // Set the interval until the next gf or arf.
+  cpi->baseline_gf_interval = i;
+
+#if CONFIG_MULTIPLE_ARF
+  if (cpi->multi_arf_enabled) {
+    // Initialize frame coding order variables.
+    cpi->new_frame_coding_order_period = 0;
+    cpi->next_frame_in_order = 0;
+    cpi->arf_buffered = 0;
+    vp9_zero(cpi->frame_coding_order);
+    vp9_zero(cpi->arf_buffer_idx);
+    vpx_memset(cpi->arf_weight, -1, sizeof(cpi->arf_weight));
+  }
+#endif
+
+  // Should we use the alternate reference frame
+  if (allow_alt_ref &&
+      (i < cpi->oxcf.lag_in_frames) &&
+      (i >= MIN_GF_INTERVAL) &&
+      // dont use ARF very near next kf
+      (i <= (cpi->twopass.frames_to_key - MIN_GF_INTERVAL)) &&
+      ((next_frame.pcnt_inter > 0.75) ||
+       (next_frame.pcnt_second_ref > 0.5)) &&
+      ((mv_in_out_accumulator / (double)i > -0.2) ||
+       (mv_in_out_accumulator > -2.0)) &&
+      (boost_score > 100)) {
+    // Alternative boost calculation for alt ref
+    cpi->gfu_boost = calc_arf_boost(cpi, 0, (i - 1), (i - 1), &f_boost, &b_boost);
+    cpi->source_alt_ref_pending = 1;
+
+#if CONFIG_MULTIPLE_ARF
+    // Set the ARF schedule.
+    if (cpi->multi_arf_enabled) {
+      schedule_frames(cpi, 0, -(cpi->baseline_gf_interval - 1), 2, 1, 0);
+    }
+#endif
+  } else {
+    cpi->gfu_boost = (int)boost_score;
+    cpi->source_alt_ref_pending = 0;
+#if CONFIG_MULTIPLE_ARF
+    // Set the GF schedule.
+    if (cpi->multi_arf_enabled) {
+      schedule_frames(cpi, 0, cpi->baseline_gf_interval - 1, 2, 0, 0);
+      assert(cpi->new_frame_coding_order_period == cpi->baseline_gf_interval);
+    }
+#endif
+  }
+
+#if CONFIG_MULTIPLE_ARF
+  if (cpi->multi_arf_enabled && (cpi->common.frame_type != KEY_FRAME)) {
+    int max_level = INT_MIN;
+    // Replace level indicator of -1 with correct level.
+    for (i = 0; i < cpi->frame_coding_order_period; ++i) {
+      if (cpi->arf_weight[i] > max_level) {
+        max_level = cpi->arf_weight[i];
+      }
+    }
+    ++max_level;
+    for (i = 0; i < cpi->frame_coding_order_period; ++i) {
+      if (cpi->arf_weight[i] == -1) {
+        cpi->arf_weight[i] = max_level;
+      }
+    }
+    cpi->max_arf_level = max_level;
+  }
+#if 0
+  if (cpi->multi_arf_enabled) {
+    printf("\nSchedule: ");
+    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+      printf("%4d ", cpi->frame_coding_order[i]);
+    }
+    printf("\n");
+    printf("ARFref:   ");
+    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+      printf("%4d ", cpi->arf_buffer_idx[i]);
+    }
+    printf("\n");
+    printf("Weight:   ");
+    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
+      printf("%4d ", cpi->arf_weight[i]);
+    }
+    printf("\n");
+  }
+#endif
+#endif
+
+  // Now decide how many bits should be allocated to the GF group as  a
+  // proportion of those remaining in the kf group.
+  // The final key frame group in the clip is treated as a special case
+  // where cpi->twopass.kf_group_bits is tied to cpi->twopass.bits_left.
+  // This is also important for short clips where there may only be one
+  // key frame.
+  if (cpi->twopass.frames_to_key >= (int)(cpi->twopass.total_stats.count -
+                                          cpi->common.current_video_frame)) {
+    cpi->twopass.kf_group_bits =
+      (cpi->twopass.bits_left > 0) ? cpi->twopass.bits_left : 0;
+  }
+
+  // Calculate the bits to be allocated to the group as a whole
+  if ((cpi->twopass.kf_group_bits > 0) &&
+      (cpi->twopass.kf_group_error_left > 0)) {
+    cpi->twopass.gf_group_bits =
+      (int64_t)(cpi->twopass.kf_group_bits *
+                (gf_group_err / cpi->twopass.kf_group_error_left));
+  } else
+    cpi->twopass.gf_group_bits = 0;
+
+  cpi->twopass.gf_group_bits =
+    (cpi->twopass.gf_group_bits < 0)
+    ? 0
+    : (cpi->twopass.gf_group_bits > cpi->twopass.kf_group_bits)
+    ? cpi->twopass.kf_group_bits : cpi->twopass.gf_group_bits;
+
+  // Clip cpi->twopass.gf_group_bits based on user supplied data rate
+  // variability limit (cpi->oxcf.two_pass_vbrmax_section)
+  if (cpi->twopass.gf_group_bits >
+      (int64_t)max_bits * cpi->baseline_gf_interval)
+    cpi->twopass.gf_group_bits = (int64_t)max_bits * cpi->baseline_gf_interval;
+
+  // Reset the file position
+  reset_fpf_position(cpi, start_pos);
+
+  // Update the record of error used so far (only done once per gf group)
+  cpi->twopass.modified_error_used += gf_group_err;
+
+  // Assign  bits to the arf or gf.
+  for (i = 0;
+      i <= (cpi->source_alt_ref_pending && cpi->common.frame_type != KEY_FRAME);
+      ++i) {
+    int allocation_chunks;
+    int q = cpi->oxcf.fixed_q < 0 ? cpi->last_q[INTER_FRAME]
+                                  : cpi->oxcf.fixed_q;
+    int gf_bits;
+
+    int boost = (cpi->gfu_boost * vp9_gfboost_qadjust(q)) / 100;
+
+    // Set max and minimum boost and hence minimum allocation
+    boost = clamp(boost, 125, (cpi->baseline_gf_interval + 1) * 200);
+
+    if (cpi->source_alt_ref_pending && i == 0)
+      allocation_chunks = ((cpi->baseline_gf_interval + 1) * 100) + boost;
+    else
+      allocation_chunks = (cpi->baseline_gf_interval * 100) + (boost - 100);
+
+    // Prevent overflow
+    if (boost > 1023) {
+      int divisor = boost >> 10;
+      boost /= divisor;
+      allocation_chunks /= divisor;
+    }
+
+    // Calculate the number of bits to be spent on the gf or arf based on
+    // the boost number
+    gf_bits = (int)((double)boost * (cpi->twopass.gf_group_bits /
+                                       (double)allocation_chunks));
+
+    // If the frame that is to be boosted is simpler than the average for
+    // the gf/arf group then use an alternative calculation
+    // based on the error score of the frame itself
+    if (mod_frame_err < gf_group_err / (double)cpi->baseline_gf_interval) {
+      double alt_gf_grp_bits =
+        (double)cpi->twopass.kf_group_bits  *
+        (mod_frame_err * (double)cpi->baseline_gf_interval) /
+        DOUBLE_DIVIDE_CHECK(cpi->twopass.kf_group_error_left);
+
+      int alt_gf_bits = (int)((double)boost * (alt_gf_grp_bits /
+                                           (double)allocation_chunks));
+
+      if (gf_bits > alt_gf_bits)
+        gf_bits = alt_gf_bits;
+    }
+    // Else if it is harder than other frames in the group make sure it at
+    // least receives an allocation in keeping with its relative error
+    // score, otherwise it may be worse off than an "un-boosted" frame
+    else {
+      int alt_gf_bits = (int)((double)cpi->twopass.kf_group_bits *
+                        mod_frame_err /
+                        DOUBLE_DIVIDE_CHECK(cpi->twopass.kf_group_error_left));
+
+      if (alt_gf_bits > gf_bits)
+        gf_bits = alt_gf_bits;
+    }
+
+    // Dont allow a negative value for gf_bits
+    if (gf_bits < 0)
+      gf_bits = 0;
+
+    // Add in minimum for a frame
+    gf_bits += cpi->min_frame_bandwidth;
+
+    if (i == 0) {
+      cpi->twopass.gf_bits = gf_bits;
+    }
+    if (i == 1 || (!cpi->source_alt_ref_pending
+        && (cpi->common.frame_type != KEY_FRAME))) {
+      // Per frame bit target for this frame
+      cpi->per_frame_bandwidth = gf_bits;
+    }
+  }
+
+  {
+    // Adjust KF group bits and error remaining
+    cpi->twopass.kf_group_error_left -= (int64_t)gf_group_err;
+    cpi->twopass.kf_group_bits -= cpi->twopass.gf_group_bits;
+
+    if (cpi->twopass.kf_group_bits < 0)
+      cpi->twopass.kf_group_bits = 0;
+
+    // Note the error score left in the remaining frames of the group.
+    // For normal GFs we want to remove the error score for the first frame
+    // of the group (except in Key frame case where this has already
+    // happened)
+    if (!cpi->source_alt_ref_pending && cpi->common.frame_type != KEY_FRAME)
+      cpi->twopass.gf_group_error_left = (int64_t)(gf_group_err
+                                                   - gf_first_frame_err);
+    else
+      cpi->twopass.gf_group_error_left = (int64_t)gf_group_err;
+
+    cpi->twopass.gf_group_bits -= cpi->twopass.gf_bits
+        - cpi->min_frame_bandwidth;
+
+    if (cpi->twopass.gf_group_bits < 0)
+      cpi->twopass.gf_group_bits = 0;
+
+    // This condition could fail if there are two kfs very close together
+    // despite (MIN_GF_INTERVAL) and would cause a divide by 0 in the
+    // calculation of alt_extra_bits.
+    if (cpi->baseline_gf_interval >= 3) {
+      const int boost = cpi->source_alt_ref_pending ? b_boost : cpi->gfu_boost;
+
+      if (boost >= 150) {
+        int alt_extra_bits;
+        int pct_extra = (boost - 100) / 50;
+        pct_extra = (pct_extra > 20) ? 20 : pct_extra;
+
+        alt_extra_bits = (int)((cpi->twopass.gf_group_bits * pct_extra) / 100);
+        cpi->twopass.gf_group_bits -= alt_extra_bits;
+      }
+    }
+  }
+
+  if (cpi->common.frame_type != KEY_FRAME) {
+    FIRSTPASS_STATS sectionstats;
+
+    zero_stats(&sectionstats);
+    reset_fpf_position(cpi, start_pos);
+
+    for (i = 0; i < cpi->baseline_gf_interval; i++) {
+      input_stats(cpi, &next_frame);
+      accumulate_stats(&sectionstats, &next_frame);
+    }
+
+    avg_stats(&sectionstats);
+
+    cpi->twopass.section_intra_rating = (int)
+      (sectionstats.intra_error /
+      DOUBLE_DIVIDE_CHECK(sectionstats.coded_error));
+
+    reset_fpf_position(cpi, start_pos);
+  }
+}
+
+// Allocate bits to a normal frame that is neither a gf an arf or a key frame.
+static void assign_std_frame_bits(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
+  int target_frame_size;
+
+  double modified_err;
+  double err_fraction;
+
+  // Max for a single frame.
+  int max_bits = frame_max_bits(cpi);
+
+  // Calculate modified prediction error used in bit allocation.
+  modified_err = calculate_modified_err(cpi, this_frame);
+
+  if (cpi->twopass.gf_group_error_left > 0)
+    // What portion of the remaining GF group error is used by this frame.
+    err_fraction = modified_err / cpi->twopass.gf_group_error_left;
+  else
+    err_fraction = 0.0;
+
+  // How many of those bits available for allocation should we give it?
+  target_frame_size = (int)((double)cpi->twopass.gf_group_bits * err_fraction);
+
+  // Clip target size to 0 - max_bits (or cpi->twopass.gf_group_bits) at
+  // the top end.
+  if (target_frame_size < 0)
+    target_frame_size = 0;
+  else {
+    if (target_frame_size > max_bits)
+      target_frame_size = max_bits;
+
+    if (target_frame_size > cpi->twopass.gf_group_bits)
+      target_frame_size = (int)cpi->twopass.gf_group_bits;
+  }
+
+  // Adjust error and bits remaining.
+  cpi->twopass.gf_group_error_left -= (int64_t)modified_err;
+  cpi->twopass.gf_group_bits -= target_frame_size;
+
+  if (cpi->twopass.gf_group_bits < 0)
+    cpi->twopass.gf_group_bits = 0;
+
+  // Add in the minimum number of bits that is set aside for every frame.
+  target_frame_size += cpi->min_frame_bandwidth;
+
+  // Per frame bit target for this frame.
+  cpi->per_frame_bandwidth = target_frame_size;
+}
+
+// Make a damped adjustment to the active max q.
+static int adjust_active_maxq(int old_maxqi, int new_maxqi) {
+  int i;
+  const double old_q = vp9_convert_qindex_to_q(old_maxqi);
+  const double new_q = vp9_convert_qindex_to_q(new_maxqi);
+  const double target_q = ((old_q * 7.0) + new_q) / 8.0;
+
+  if (target_q > old_q) {
+    for (i = old_maxqi; i <= new_maxqi; i++)
+      if (vp9_convert_qindex_to_q(i) >= target_q)
+        return i;
+  } else {
+    for (i = old_maxqi; i >= new_maxqi; i--)
+      if (vp9_convert_qindex_to_q(i) <= target_q)
+        return i;
+  }
+
+  return new_maxqi;
+}
+
+void vp9_second_pass(VP9_COMP *cpi) {
+  int tmp_q;
+  int frames_left = (int)(cpi->twopass.total_stats.count -
+                          cpi->common.current_video_frame);
+
+  FIRSTPASS_STATS this_frame;
+  FIRSTPASS_STATS this_frame_copy;
+
+  double this_frame_intra_error;
+  double this_frame_coded_error;
+
+  if (!cpi->twopass.stats_in)
+    return;
+
+  vp9_clear_system_state();
+
+  // Special case code for first frame.
+  if (cpi->common.current_video_frame == 0) {
+    cpi->twopass.est_max_qcorrection_factor = 1.0;
+
+    // Set a cq_level in constrained quality mode.
+    if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
+      int est_cq = estimate_cq(cpi, &cpi->twopass.total_left_stats,
+                               (int)(cpi->twopass.bits_left / frames_left));
+
+      cpi->cq_target_quality = cpi->oxcf.cq_level;
+      if (est_cq > cpi->cq_target_quality)
+        cpi->cq_target_quality = est_cq;
+    }
+
+    // guess at maxq needed in 2nd pass
+    cpi->twopass.maxq_max_limit = cpi->worst_quality;
+    cpi->twopass.maxq_min_limit = cpi->best_quality;
+
+    tmp_q = estimate_max_q(cpi, &cpi->twopass.total_left_stats,
+                           (int)(cpi->twopass.bits_left / frames_left));
+
+    cpi->active_worst_quality = tmp_q;
+    cpi->ni_av_qi = tmp_q;
+    cpi->avg_q = vp9_convert_qindex_to_q(tmp_q);
+
+#ifndef ONE_SHOT_Q_ESTIMATE
+    // Limit the maxq value returned subsequently.
+    // This increases the risk of overspend or underspend if the initial
+    // estimate for the clip is bad, but helps prevent excessive
+    // variation in Q, especially near the end of a clip
+    // where for example a small overspend may cause Q to crash
+    adjust_maxq_qrange(cpi);
+#endif
+  }
+
+#ifndef ONE_SHOT_Q_ESTIMATE
+  // The last few frames of a clip almost always have to few or too many
+  // bits and for the sake of over exact rate control we dont want to make
+  // radical adjustments to the allowed quantizer range just to use up a
+  // few surplus bits or get beneath the target rate.
+  else if ((cpi->common.current_video_frame <
+            (((unsigned int)cpi->twopass.total_stats.count * 255) >> 8)) &&
+           ((cpi->common.current_video_frame + cpi->baseline_gf_interval) <
+            (unsigned int)cpi->twopass.total_stats.count)) {
+    if (frames_left < 1)
+      frames_left = 1;
+
+    tmp_q = estimate_max_q(
+              cpi,
+              &cpi->twopass.total_left_stats,
+              (int)(cpi->twopass.bits_left / frames_left));
+
+    // Make a damped adjustment to active max Q
+    cpi->active_worst_quality =
+      adjust_active_maxq(cpi->active_worst_quality, tmp_q);
+  }
+#endif
+
+  vpx_memset(&this_frame, 0, sizeof(FIRSTPASS_STATS));
+  if (EOF == input_stats(cpi, &this_frame))
+    return;
+
+  this_frame_intra_error = this_frame.intra_error;
+  this_frame_coded_error = this_frame.coded_error;
+
+  // keyframe and section processing !
+  if (cpi->twopass.frames_to_key == 0) {
+    // Define next KF group and assign bits to it
+    this_frame_copy = this_frame;
+    find_next_key_frame(cpi, &this_frame_copy);
+  }
+
+  // Is this a GF / ARF (Note that a KF is always also a GF)
+  if (cpi->frames_till_gf_update_due == 0) {
+    // Define next gf group and assign bits to it
+    this_frame_copy = this_frame;
+
+#if CONFIG_MULTIPLE_ARF
+    if (cpi->multi_arf_enabled) {
+      define_fixed_arf_period(cpi);
+    } else {
+#endif
+      define_gf_group(cpi, &this_frame_copy);
+#if CONFIG_MULTIPLE_ARF
+    }
+#endif
+
+    // If we are going to code an altref frame at the end of the group
+    // and the current frame is not a key frame....
+    // If the previous group used an arf this frame has already benefited
+    // from that arf boost and it should not be given extra bits
+    // If the previous group was NOT coded using arf we may want to apply
+    // some boost to this GF as well
+    if (cpi->source_alt_ref_pending && (cpi->common.frame_type != KEY_FRAME)) {
+      // Assign a standard frames worth of bits from those allocated
+      // to the GF group
+      int bak = cpi->per_frame_bandwidth;
+      this_frame_copy = this_frame;
+      assign_std_frame_bits(cpi, &this_frame_copy);
+      cpi->per_frame_bandwidth = bak;
+    }
+  } else {
+    // Otherwise this is an ordinary frame
+    // Assign bits from those allocated to the GF group
+    this_frame_copy =  this_frame;
+    assign_std_frame_bits(cpi, &this_frame_copy);
+  }
+
+  // Keep a globally available copy of this and the next frame's iiratio.
+  cpi->twopass.this_iiratio = (int)(this_frame_intra_error /
+                              DOUBLE_DIVIDE_CHECK(this_frame_coded_error));
+  {
+    FIRSTPASS_STATS next_frame;
+    if (lookup_next_frame_stats(cpi, &next_frame) != EOF) {
+      cpi->twopass.next_iiratio = (int)(next_frame.intra_error /
+                                  DOUBLE_DIVIDE_CHECK(next_frame.coded_error));
+    }
+  }
+
+  // Set nominal per second bandwidth for this frame
+  cpi->target_bandwidth = (int)(cpi->per_frame_bandwidth
+                                * cpi->output_frame_rate);
+  if (cpi->target_bandwidth < 0)
+    cpi->target_bandwidth = 0;
+
+  cpi->twopass.frames_to_key--;
+
+  // Update the total stats remaining structure
+  subtract_stats(&cpi->twopass.total_left_stats, &this_frame);
+}
+
+static int test_candidate_kf(VP9_COMP *cpi,
+                             FIRSTPASS_STATS *last_frame,
+                             FIRSTPASS_STATS *this_frame,
+                             FIRSTPASS_STATS *next_frame) {
+  int is_viable_kf = 0;
+
+  // Does the frame satisfy the primary criteria of a key frame
+  //      If so, then examine how well it predicts subsequent frames
+  if ((this_frame->pcnt_second_ref < 0.10) &&
+      (next_frame->pcnt_second_ref < 0.10) &&
+      ((this_frame->pcnt_inter < 0.05) ||
+       (
+         ((this_frame->pcnt_inter - this_frame->pcnt_neutral) < .35) &&
+         ((this_frame->intra_error / DOUBLE_DIVIDE_CHECK(this_frame->coded_error)) < 2.5) &&
+         ((fabs(last_frame->coded_error - this_frame->coded_error) / DOUBLE_DIVIDE_CHECK(this_frame->coded_error) > .40) ||
+          (fabs(last_frame->intra_error - this_frame->intra_error) / DOUBLE_DIVIDE_CHECK(this_frame->intra_error) > .40) ||
+          ((next_frame->intra_error / DOUBLE_DIVIDE_CHECK(next_frame->coded_error)) > 3.5)
+         )
+       )
+      )
+     ) {
+    int i;
+    FIRSTPASS_STATS *start_pos;
+
+    FIRSTPASS_STATS local_next_frame;
+
+    double boost_score = 0.0;
+    double old_boost_score = 0.0;
+    double decay_accumulator = 1.0;
+    double next_iiratio;
+
+    local_next_frame = *next_frame;
+
+    // Note the starting file position so we can reset to it
+    start_pos = cpi->twopass.stats_in;
+
+    // Examine how well the key frame predicts subsequent frames
+    for (i = 0; i < 16; i++) {
+      next_iiratio = (IIKFACTOR1 * local_next_frame.intra_error / DOUBLE_DIVIDE_CHECK(local_next_frame.coded_error));
+
+      if (next_iiratio > RMAX)
+        next_iiratio = RMAX;
+
+      // Cumulative effect of decay in prediction quality
+      if (local_next_frame.pcnt_inter > 0.85)
+        decay_accumulator = decay_accumulator * local_next_frame.pcnt_inter;
+      else
+        decay_accumulator = decay_accumulator * ((0.85 + local_next_frame.pcnt_inter) / 2.0);
+
+      // decay_accumulator = decay_accumulator * local_next_frame.pcnt_inter;
+
+      // Keep a running total
+      boost_score += (decay_accumulator * next_iiratio);
+
+      // Test various breakout clauses
+      if ((local_next_frame.pcnt_inter < 0.05) ||
+          (next_iiratio < 1.5) ||
+          (((local_next_frame.pcnt_inter -
+             local_next_frame.pcnt_neutral) < 0.20) &&
+           (next_iiratio < 3.0)) ||
+          ((boost_score - old_boost_score) < 3.0) ||
+          (local_next_frame.intra_error < 200)
+         ) {
+        break;
+      }
+
+      old_boost_score = boost_score;
+
+      // Get the next frame details
+      if (EOF == input_stats(cpi, &local_next_frame))
+        break;
+    }
+
+    // If there is tolerable prediction for at least the next 3 frames then
+    // break out else discard this potential key frame and move on
+    if (boost_score > 30.0 && (i > 3))
+      is_viable_kf = 1;
+    else {
+      // Reset the file position
+      reset_fpf_position(cpi, start_pos);
+
+      is_viable_kf = 0;
+    }
+  }
+
+  return is_viable_kf;
+}
+static void find_next_key_frame(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
+  int i, j;
+  FIRSTPASS_STATS last_frame;
+  FIRSTPASS_STATS first_frame;
+  FIRSTPASS_STATS next_frame;
+  FIRSTPASS_STATS *start_position;
+
+  double decay_accumulator = 1.0;
+  double zero_motion_accumulator = 1.0;
+  double boost_score = 0;
+  double loop_decay_rate;
+
+  double kf_mod_err = 0.0;
+  double kf_group_err = 0.0;
+  double kf_group_intra_err = 0.0;
+  double kf_group_coded_err = 0.0;
+  double recent_loop_decay[8] = {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0};
+
+  vpx_memset(&next_frame, 0, sizeof(next_frame)); // assure clean
+
+  vp9_clear_system_state();  // __asm emms;
+  start_position = cpi->twopass.stats_in;
+
+  cpi->common.frame_type = KEY_FRAME;
+
+  // is this a forced key frame by interval
+  cpi->this_key_frame_forced = cpi->next_key_frame_forced;
+
+  // Clear the alt ref active flag as this can never be active on a key frame
+  cpi->source_alt_ref_active = 0;
+
+  // Kf is always a gf so clear frames till next gf counter
+  cpi->frames_till_gf_update_due = 0;
+
+  cpi->twopass.frames_to_key = 1;
+
+  // Take a copy of the initial frame details
+  first_frame = *this_frame;
+
+  cpi->twopass.kf_group_bits = 0;        // Total bits available to kf group
+  cpi->twopass.kf_group_error_left = 0;  // Group modified error score.
+
+  kf_mod_err = calculate_modified_err(cpi, this_frame);
+
+  // find the next keyframe
+  i = 0;
+  while (cpi->twopass.stats_in < cpi->twopass.stats_in_end) {
+    // Accumulate kf group error
+    kf_group_err += calculate_modified_err(cpi, this_frame);
+
+    // These figures keep intra and coded error counts for all frames including key frames in the group.
+    // The effect of the key frame itself can be subtracted out using the first_frame data collected above
+    kf_group_intra_err += this_frame->intra_error;
+    kf_group_coded_err += this_frame->coded_error;
+
+    // load a the next frame's stats
+    last_frame = *this_frame;
+    input_stats(cpi, this_frame);
+
+    // Provided that we are not at the end of the file...
+    if (cpi->oxcf.auto_key
+        && lookup_next_frame_stats(cpi, &next_frame) != EOF) {
+      // Normal scene cut check
+      if (test_candidate_kf(cpi, &last_frame, this_frame, &next_frame))
+        break;
+
+
+      // How fast is prediction quality decaying
+      loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame);
+
+      // We want to know something about the recent past... rather than
+      // as used elsewhere where we are concened with decay in prediction
+      // quality since the last GF or KF.
+      recent_loop_decay[i % 8] = loop_decay_rate;
+      decay_accumulator = 1.0;
+      for (j = 0; j < 8; j++)
+        decay_accumulator *= recent_loop_decay[j];
+
+      // Special check for transition or high motion followed by a
+      // to a static scene.
+      if (detect_transition_to_still(cpi, i, cpi->key_frame_frequency - i,
+                                     loop_decay_rate, decay_accumulator))
+        break;
+
+      // Step on to the next frame
+      cpi->twopass.frames_to_key++;
+
+      // If we don't have a real key frame within the next two
+      // forcekeyframeevery intervals then break out of the loop.
+      if (cpi->twopass.frames_to_key >= 2 * (int)cpi->key_frame_frequency)
+        break;
+    } else
+      cpi->twopass.frames_to_key++;
+
+    i++;
+  }
+
+  // If there is a max kf interval set by the user we must obey it.
+  // We already breakout of the loop above at 2x max.
+  // This code centers the extra kf if the actual natural
+  // interval is between 1x and 2x
+  if (cpi->oxcf.auto_key
+      && cpi->twopass.frames_to_key > (int)cpi->key_frame_frequency) {
+    FIRSTPASS_STATS *current_pos = cpi->twopass.stats_in;
+    FIRSTPASS_STATS tmp_frame;
+
+    cpi->twopass.frames_to_key /= 2;
+
+    // Copy first frame details
+    tmp_frame = first_frame;
+
+    // Reset to the start of the group
+    reset_fpf_position(cpi, start_position);
+
+    kf_group_err = 0;
+    kf_group_intra_err = 0;
+    kf_group_coded_err = 0;
+
+    // Rescan to get the correct error data for the forced kf group
+    for (i = 0; i < cpi->twopass.frames_to_key; i++) {
+      // Accumulate kf group errors
+      kf_group_err += calculate_modified_err(cpi, &tmp_frame);
+      kf_group_intra_err += tmp_frame.intra_error;
+      kf_group_coded_err += tmp_frame.coded_error;
+
+      // Load a the next frame's stats
+      input_stats(cpi, &tmp_frame);
+    }
+
+    // Reset to the start of the group
+    reset_fpf_position(cpi, current_pos);
+
+    cpi->next_key_frame_forced = 1;
+  } else
+    cpi->next_key_frame_forced = 0;
+
+  // Special case for the last frame of the file
+  if (cpi->twopass.stats_in >= cpi->twopass.stats_in_end) {
+    // Accumulate kf group error
+    kf_group_err += calculate_modified_err(cpi, this_frame);
+
+    // These figures keep intra and coded error counts for all frames including key frames in the group.
+    // The effect of the key frame itself can be subtracted out using the first_frame data collected above
+    kf_group_intra_err += this_frame->intra_error;
+    kf_group_coded_err += this_frame->coded_error;
+  }
+
+  // Calculate the number of bits that should be assigned to the kf group.
+  if ((cpi->twopass.bits_left > 0) && (cpi->twopass.modified_error_left > 0.0)) {
+    // Max for a single normal frame (not key frame)
+    int max_bits = frame_max_bits(cpi);
+
+    // Maximum bits for the kf group
+    int64_t max_grp_bits;
+
+    // Default allocation based on bits left and relative
+    // complexity of the section
+    cpi->twopass.kf_group_bits = (int64_t)(cpi->twopass.bits_left *
+                                           (kf_group_err /
+                                            cpi->twopass.modified_error_left));
+
+    // Clip based on maximum per frame rate defined by the user.
+    max_grp_bits = (int64_t)max_bits * (int64_t)cpi->twopass.frames_to_key;
+    if (cpi->twopass.kf_group_bits > max_grp_bits)
+      cpi->twopass.kf_group_bits = max_grp_bits;
+  } else
+    cpi->twopass.kf_group_bits = 0;
+
+  // Reset the first pass file position
+  reset_fpf_position(cpi, start_position);
+
+  // determine how big to make this keyframe based on how well the subsequent frames use inter blocks
+  decay_accumulator = 1.0;
+  boost_score = 0.0;
+  loop_decay_rate = 1.00;       // Starting decay rate
+
+  // Scan through the kf group collating various stats.
+  for (i = 0; i < cpi->twopass.frames_to_key; i++) {
+    double r;
+
+    if (EOF == input_stats(cpi, &next_frame))
+      break;
+
+    // Monitor for static sections.
+    if ((next_frame.pcnt_inter - next_frame.pcnt_motion) <
+        zero_motion_accumulator) {
+      zero_motion_accumulator =
+        (next_frame.pcnt_inter - next_frame.pcnt_motion);
+    }
+
+    // For the first few frames collect data to decide kf boost.
+    if (i <= (cpi->max_gf_interval * 2)) {
+      if (next_frame.intra_error > cpi->twopass.kf_intra_err_min)
+        r = (IIKFACTOR2 * next_frame.intra_error /
+             DOUBLE_DIVIDE_CHECK(next_frame.coded_error));
+      else
+        r = (IIKFACTOR2 * cpi->twopass.kf_intra_err_min /
+             DOUBLE_DIVIDE_CHECK(next_frame.coded_error));
+
+      if (r > RMAX)
+        r = RMAX;
+
+      // How fast is prediction quality decaying
+      if (!detect_flash(cpi, 0)) {
+        loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame);
+        decay_accumulator = decay_accumulator * loop_decay_rate;
+        decay_accumulator = decay_accumulator < MIN_DECAY_FACTOR
+                              ? MIN_DECAY_FACTOR : decay_accumulator;
+      }
+
+      boost_score += (decay_accumulator * r);
+    }
+  }
+
+  {
+    FIRSTPASS_STATS sectionstats;
+
+    zero_stats(&sectionstats);
+    reset_fpf_position(cpi, start_position);
+
+    for (i = 0; i < cpi->twopass.frames_to_key; i++) {
+      input_stats(cpi, &next_frame);
+      accumulate_stats(&sectionstats, &next_frame);
+    }
+
+    avg_stats(&sectionstats);
+
+    cpi->twopass.section_intra_rating = (int)
+      (sectionstats.intra_error
+      / DOUBLE_DIVIDE_CHECK(sectionstats.coded_error));
+  }
+
+  // Reset the first pass file position
+  reset_fpf_position(cpi, start_position);
+
+  // Work out how many bits to allocate for the key frame itself
+  if (1) {
+    int kf_boost = (int)boost_score;
+    int allocation_chunks;
+    int alt_kf_bits;
+
+    if (kf_boost < (cpi->twopass.frames_to_key * 3))
+      kf_boost = (cpi->twopass.frames_to_key * 3);
+
+    if (kf_boost < 300) // Min KF boost
+      kf_boost = 300;
+
+    // Make a note of baseline boost and the zero motion
+    // accumulator value for use elsewhere.
+    cpi->kf_boost = kf_boost;
+    cpi->kf_zeromotion_pct = (int)(zero_motion_accumulator * 100.0);
+
+    // We do three calculations for kf size.
+    // The first is based on the error score for the whole kf group.
+    // The second (optionaly) on the key frames own error if this is
+    // smaller than the average for the group.
+    // The final one insures that the frame receives at least the
+    // allocation it would have received based on its own error score vs
+    // the error score remaining
+    // Special case if the sequence appears almost totaly static
+    // In this case we want to spend almost all of the bits on the
+    // key frame.
+    // cpi->twopass.frames_to_key-1 because key frame itself is taken
+    // care of by kf_boost.
+    if (zero_motion_accumulator >= 0.99) {
+      allocation_chunks =
+        ((cpi->twopass.frames_to_key - 1) * 10) + kf_boost;
+    } else {
+      allocation_chunks =
+        ((cpi->twopass.frames_to_key - 1) * 100) + kf_boost;
+    }
+
+    // Prevent overflow
+    if (kf_boost > 1028) {
+      int divisor = kf_boost >> 10;
+      kf_boost /= divisor;
+      allocation_chunks /= divisor;
+    }
+
+    cpi->twopass.kf_group_bits = (cpi->twopass.kf_group_bits < 0) ? 0 : cpi->twopass.kf_group_bits;
+
+    // Calculate the number of bits to be spent on the key frame
+    cpi->twopass.kf_bits  = (int)((double)kf_boost * ((double)cpi->twopass.kf_group_bits / (double)allocation_chunks));
+
+    // If the key frame is actually easier than the average for the
+    // kf group (which does sometimes happen... eg a blank intro frame)
+    // Then use an alternate calculation based on the kf error score
+    // which should give a smaller key frame.
+    if (kf_mod_err < kf_group_err / cpi->twopass.frames_to_key) {
+      double  alt_kf_grp_bits =
+        ((double)cpi->twopass.bits_left *
+         (kf_mod_err * (double)cpi->twopass.frames_to_key) /
+         DOUBLE_DIVIDE_CHECK(cpi->twopass.modified_error_left));
+
+      alt_kf_bits = (int)((double)kf_boost *
+                          (alt_kf_grp_bits / (double)allocation_chunks));
+
+      if (cpi->twopass.kf_bits > alt_kf_bits) {
+        cpi->twopass.kf_bits = alt_kf_bits;
+      }
+    }
+    // Else if it is much harder than other frames in the group make sure
+    // it at least receives an allocation in keeping with its relative
+    // error score
+    else {
+      alt_kf_bits =
+        (int)((double)cpi->twopass.bits_left *
+              (kf_mod_err /
+               DOUBLE_DIVIDE_CHECK(cpi->twopass.modified_error_left)));
+
+      if (alt_kf_bits > cpi->twopass.kf_bits) {
+        cpi->twopass.kf_bits = alt_kf_bits;
+      }
+    }
+
+    cpi->twopass.kf_group_bits -= cpi->twopass.kf_bits;
+    // Add in the minimum frame allowance
+    cpi->twopass.kf_bits += cpi->min_frame_bandwidth;
+
+    // Peer frame bit target for this frame
+    cpi->per_frame_bandwidth = cpi->twopass.kf_bits;
+    // Convert to a per second bitrate
+    cpi->target_bandwidth = (int)(cpi->twopass.kf_bits *
+                                  cpi->output_frame_rate);
+  }
+
+  // Note the total error score of the kf group minus the key frame itself
+  cpi->twopass.kf_group_error_left = (int)(kf_group_err - kf_mod_err);
+
+  // Adjust the count of total modified error left.
+  // The count of bits left is adjusted elsewhere based on real coded frame sizes
+  cpi->twopass.modified_error_left -= kf_group_err;
+}
diff --git a/libvpx/vp9/encoder/vp9_firstpass.h b/libvpx/vp9/encoder/vp9_firstpass.h
new file mode 100644
index 0000000..2296a66
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_firstpass.h
@@ -0,0 +1,22 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_FIRSTPASS_H_
+#define VP9_ENCODER_VP9_FIRSTPASS_H_
+
+void vp9_init_first_pass(VP9_COMP *cpi);
+void vp9_first_pass(VP9_COMP *cpi);
+void vp9_end_first_pass(VP9_COMP *cpi);
+
+void vp9_init_second_pass(VP9_COMP *cpi);
+void vp9_second_pass(VP9_COMP *cpi);
+void vp9_end_second_pass(VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_FIRSTPASS_H_
diff --git a/libvpx/vp9/encoder/vp9_lookahead.c b/libvpx/vp9/encoder/vp9_lookahead.c
new file mode 100644
index 0000000..b07d92a
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_lookahead.c
@@ -0,0 +1,190 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include <assert.h>
+#include <stdlib.h>
+
+#include "vpx_config.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/encoder/vp9_lookahead.h"
+#include "vp9/common/vp9_extend.h"
+
+#define MAX_LAG_BUFFERS 25
+
+struct lookahead_ctx {
+  unsigned int max_sz;         /* Absolute size of the queue */
+  unsigned int sz;             /* Number of buffers currently in the queue */
+  unsigned int read_idx;       /* Read index */
+  unsigned int write_idx;      /* Write index */
+  struct lookahead_entry *buf; /* Buffer list */
+};
+
+
+/* Return the buffer at the given absolute index and increment the index */
+static struct lookahead_entry * pop(struct lookahead_ctx *ctx,
+                                    unsigned int *idx) {
+  unsigned int index = *idx;
+  struct lookahead_entry *buf = ctx->buf + index;
+
+  assert(index < ctx->max_sz);
+  if (++index >= ctx->max_sz)
+    index -= ctx->max_sz;
+  *idx = index;
+  return buf;
+}
+
+
+void vp9_lookahead_destroy(struct lookahead_ctx *ctx) {
+  if (ctx) {
+    if (ctx->buf) {
+      unsigned int i;
+
+      for (i = 0; i < ctx->max_sz; i++)
+        vp9_free_frame_buffer(&ctx->buf[i].img);
+      free(ctx->buf);
+    }
+    free(ctx);
+  }
+}
+
+
+struct lookahead_ctx * vp9_lookahead_init(unsigned int width,
+                                          unsigned int height,
+                                          unsigned int subsampling_x,
+                                          unsigned int subsampling_y,
+                                          unsigned int depth) {
+  struct lookahead_ctx *ctx = NULL;
+
+  // Clamp the lookahead queue depth
+  depth = clamp(depth, 1, MAX_LAG_BUFFERS);
+
+  // Allocate the lookahead structures
+  ctx = calloc(1, sizeof(*ctx));
+  if (ctx) {
+    unsigned int i;
+    ctx->max_sz = depth;
+    ctx->buf = calloc(depth, sizeof(*ctx->buf));
+    if (!ctx->buf)
+      goto bail;
+    for (i = 0; i < depth; i++)
+      if (vp9_alloc_frame_buffer(&ctx->buf[i].img,
+                                 width, height, subsampling_x, subsampling_y,
+                                 VP9BORDERINPIXELS))
+        goto bail;
+  }
+  return ctx;
+bail:
+  vp9_lookahead_destroy(ctx);
+  return NULL;
+}
+
+#define USE_PARTIAL_COPY 0
+
+int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG   *src,
+                       int64_t ts_start, int64_t ts_end, unsigned int flags,
+                       unsigned char *active_map) {
+  struct lookahead_entry *buf;
+#if USE_PARTIAL_COPY
+  int row, col, active_end;
+  int mb_rows = (src->y_height + 15) >> 4;
+  int mb_cols = (src->y_width + 15) >> 4;
+#endif
+
+  if (ctx->sz + 1 > ctx->max_sz)
+    return 1;
+  ctx->sz++;
+  buf = pop(ctx, &ctx->write_idx);
+
+#if USE_PARTIAL_COPY
+  // TODO(jkoleszar): This is disabled for now, as
+  // vp9_copy_and_extend_frame_with_rect is not subsampling/alpha aware.
+
+  // Only do this partial copy if the following conditions are all met:
+  // 1. Lookahead queue has has size of 1.
+  // 2. Active map is provided.
+  // 3. This is not a key frame, golden nor altref frame.
+  if (ctx->max_sz == 1 && active_map && !flags) {
+    for (row = 0; row < mb_rows; ++row) {
+      col = 0;
+
+      while (1) {
+        // Find the first active macroblock in this row.
+        for (; col < mb_cols; ++col) {
+          if (active_map[col])
+            break;
+        }
+
+        // No more active macroblock in this row.
+        if (col == mb_cols)
+          break;
+
+        // Find the end of active region in this row.
+        active_end = col;
+
+        for (; active_end < mb_cols; ++active_end) {
+          if (!active_map[active_end])
+            break;
+        }
+
+        // Only copy this active region.
+        vp9_copy_and_extend_frame_with_rect(src, &buf->img,
+                                            row << 4,
+                                            col << 4, 16,
+                                            (active_end - col) << 4);
+
+        // Start again from the end of this active region.
+        col = active_end;
+      }
+
+      active_map += mb_cols;
+    }
+  } else {
+    vp9_copy_and_extend_frame(src, &buf->img);
+  }
+#else
+  // Partial copy not implemented yet
+  vp9_copy_and_extend_frame(src, &buf->img);
+#endif
+
+  buf->ts_start = ts_start;
+  buf->ts_end = ts_end;
+  buf->flags = flags;
+  return 0;
+}
+
+
+struct lookahead_entry * vp9_lookahead_pop(struct lookahead_ctx *ctx,
+                                           int drain) {
+  struct lookahead_entry *buf = NULL;
+
+  if (ctx->sz && (drain || ctx->sz == ctx->max_sz)) {
+    buf = pop(ctx, &ctx->read_idx);
+    ctx->sz--;
+  }
+  return buf;
+}
+
+
+struct lookahead_entry * vp9_lookahead_peek(struct lookahead_ctx *ctx,
+                                            int index) {
+  struct lookahead_entry *buf = NULL;
+
+  assert(index < (int)ctx->max_sz);
+  if (index < (int)ctx->sz) {
+    index += ctx->read_idx;
+    if (index >= (int)ctx->max_sz)
+      index -= ctx->max_sz;
+    buf = ctx->buf + index;
+  }
+  return buf;
+}
+
+unsigned int vp9_lookahead_depth(struct lookahead_ctx *ctx) {
+  return ctx->sz;
+}
diff --git a/libvpx/vp9/encoder/vp9_lookahead.h b/libvpx/vp9/encoder/vp9_lookahead.h
new file mode 100644
index 0000000..81baa2c
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_lookahead.h
@@ -0,0 +1,95 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_LOOKAHEAD_H_
+#define VP9_ENCODER_VP9_LOOKAHEAD_H_
+
+#include "vpx_scale/yv12config.h"
+#include "vpx/vpx_integer.h"
+
+struct lookahead_entry {
+  YV12_BUFFER_CONFIG  img;
+  int64_t             ts_start;
+  int64_t             ts_end;
+  unsigned int        flags;
+};
+
+
+struct lookahead_ctx;
+
+/**\brief Initializes the lookahead stage
+ *
+ * The lookahead stage is a queue of frame buffers on which some analysis
+ * may be done when buffers are enqueued.
+ */
+struct lookahead_ctx *vp9_lookahead_init(unsigned int width,
+                                         unsigned int height,
+                                         unsigned int subsampling_x,
+                                         unsigned int subsampling_y,
+                                         unsigned int depth);
+
+
+/**\brief Destroys the lookahead stage
+ */
+void vp9_lookahead_destroy(struct lookahead_ctx *ctx);
+
+
+/**\brief Enqueue a source buffer
+ *
+ * This function will copy the source image into a new framebuffer with
+ * the expected stride/border.
+ *
+ * If active_map is non-NULL and there is only one frame in the queue, then copy
+ * only active macroblocks.
+ *
+ * \param[in] ctx         Pointer to the lookahead context
+ * \param[in] src         Pointer to the image to enqueue
+ * \param[in] ts_start    Timestamp for the start of this frame
+ * \param[in] ts_end      Timestamp for the end of this frame
+ * \param[in] flags       Flags set on this frame
+ * \param[in] active_map  Map that specifies which macroblock is active
+ */
+int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
+                       int64_t ts_start, int64_t ts_end, unsigned int flags,
+                       unsigned char *active_map);
+
+
+/**\brief Get the next source buffer to encode
+ *
+ *
+ * \param[in] ctx       Pointer to the lookahead context
+ * \param[in] drain     Flag indicating the buffer should be drained
+ *                      (return a buffer regardless of the current queue depth)
+ *
+ * \retval NULL, if drain set and queue is empty
+ * \retval NULL, if drain not set and queue not of the configured depth
+ */
+struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx,
+                                          int drain);
+
+
+/**\brief Get a future source buffer to encode
+ *
+ * \param[in] ctx       Pointer to the lookahead context
+ * \param[in] index     Index of the frame to be returned, 0 == next frame
+ *
+ * \retval NULL, if no buffer exists at the specified index
+ */
+struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
+                                           int index);
+
+
+/**\brief Get the number of frames currently in the lookahead queue
+ *
+ * \param[in] ctx       Pointer to the lookahead context
+ */
+unsigned int vp9_lookahead_depth(struct lookahead_ctx *ctx);
+
+#endif  // VP9_ENCODER_VP9_LOOKAHEAD_H_
diff --git a/libvpx/vp9/encoder/vp9_mbgraph.c b/libvpx/vp9/encoder/vp9_mbgraph.c
new file mode 100644
index 0000000..65fdcbe
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_mbgraph.c
@@ -0,0 +1,440 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <limits.h>
+
+#include <vpx_mem/vpx_mem.h>
+#include <vp9/encoder/vp9_encodeintra.h>
+#include <vp9/encoder/vp9_rdopt.h>
+#include <vp9/common/vp9_blockd.h>
+#include <vp9/common/vp9_reconinter.h>
+#include <vp9/common/vp9_systemdependent.h>
+#include <vp9/encoder/vp9_segmentation.h>
+
+static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
+                                              int_mv *ref_mv,
+                                              int_mv *dst_mv,
+                                              int mb_row,
+                                              int mb_col) {
+  MACROBLOCK   *const x  = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  vp9_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[BLOCK_16X16];
+  unsigned int best_err;
+
+  const int tmp_col_min = x->mv_col_min;
+  const int tmp_col_max = x->mv_col_max;
+  const int tmp_row_min = x->mv_row_min;
+  const int tmp_row_max = x->mv_row_max;
+  int_mv ref_full;
+
+  // Further step/diamond searches as necessary
+  int step_param = cpi->sf.first_step +
+      (cpi->speed < 8 ? (cpi->speed > 5 ? 1 : 0) : 2);
+
+  vp9_clamp_mv_min_max(x, ref_mv);
+
+  ref_full.as_mv.col = ref_mv->as_mv.col >> 3;
+  ref_full.as_mv.row = ref_mv->as_mv.row >> 3;
+
+  /*cpi->sf.search_method == HEX*/
+  best_err = vp9_hex_search(x, &ref_full, dst_mv, step_param, x->errorperbit,
+                            &v_fn_ptr, NULL, NULL, NULL, NULL, ref_mv);
+
+  // Try sub-pixel MC
+  // if (bestsme > error_thresh && bestsme < INT_MAX)
+  {
+    int distortion;
+    unsigned int sse;
+    best_err = cpi->find_fractional_mv_step(
+        x,
+        dst_mv, ref_mv,
+        x->errorperbit, &v_fn_ptr,
+        NULL, NULL,
+        & distortion, &sse);
+  }
+
+  vp9_set_mbmode_and_mvs(x, NEWMV, dst_mv);
+  vp9_build_inter_predictors_sby(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
+  best_err = vp9_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                          xd->plane[0].dst.buf, xd->plane[0].dst.stride,
+                          INT_MAX);
+
+  /* restore UMV window */
+  x->mv_col_min = tmp_col_min;
+  x->mv_col_max = tmp_col_max;
+  x->mv_row_min = tmp_row_min;
+  x->mv_row_max = tmp_row_max;
+
+  return best_err;
+}
+
+static int do_16x16_motion_search(VP9_COMP *cpi,
+                                  int_mv *ref_mv, int_mv *dst_mv,
+                                  int buf_mb_y_offset, int mb_y_offset,
+                                  int mb_row, int mb_col) {
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  unsigned int err, tmp_err;
+  int_mv tmp_mv;
+
+  // Try zero MV first
+  // FIXME should really use something like near/nearest MV and/or MV prediction
+  err = vp9_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                     xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                     INT_MAX);
+  dst_mv->as_int = 0;
+
+  // Test last reference frame using the previous best mv as the
+  // starting point (best reference) for the search
+  tmp_err = do_16x16_motion_iteration(cpi, ref_mv, &tmp_mv, mb_row, mb_col);
+  if (tmp_err < err) {
+    err = tmp_err;
+    dst_mv->as_int = tmp_mv.as_int;
+  }
+
+  // If the current best reference mv is not centred on 0,0 then do a 0,0 based search as well
+  if (ref_mv->as_int) {
+    unsigned int tmp_err;
+    int_mv zero_ref_mv, tmp_mv;
+
+    zero_ref_mv.as_int = 0;
+    tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, &tmp_mv,
+                                        mb_row, mb_col);
+    if (tmp_err < err) {
+      dst_mv->as_int = tmp_mv.as_int;
+      err = tmp_err;
+    }
+  }
+
+  return err;
+}
+
+static int do_16x16_zerozero_search(VP9_COMP *cpi,
+                                    int_mv *dst_mv,
+                                    int buf_mb_y_offset, int mb_y_offset) {
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  unsigned int err;
+
+  // Try zero MV first
+  // FIXME should really use something like near/nearest MV and/or MV prediction
+  err = vp9_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                     xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride,
+                     INT_MAX);
+
+  dst_mv->as_int = 0;
+
+  return err;
+}
+static int find_best_16x16_intra(VP9_COMP *cpi,
+                                 int mb_y_offset,
+                                 MB_PREDICTION_MODE *pbest_mode) {
+  MACROBLOCK   *const x  = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_PREDICTION_MODE best_mode = -1, mode;
+  unsigned int best_err = INT_MAX;
+
+  // calculate SATD for each intra prediction mode;
+  // we're intentionally not doing 4x4, we just want a rough estimate
+  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
+    unsigned int err;
+    const int bwl = b_width_log2(BLOCK_SIZE_MB16X16),  bw = 4 << bwl;
+    const int bhl = b_height_log2(BLOCK_SIZE_MB16X16), bh = 4 << bhl;
+
+    xd->mode_info_context->mbmi.mode = mode;
+    vp9_build_intra_predictors(x->plane[0].src.buf, x->plane[0].src.stride,
+                               xd->plane[0].dst.buf, xd->plane[0].dst.stride,
+                               xd->mode_info_context->mbmi.mode,
+                               bw, bh,
+                               xd->up_available, xd->left_available,
+                               xd->right_available);
+    err = vp9_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
+                       xd->plane[0].dst.buf, xd->plane[0].dst.stride, best_err);
+
+    // find best
+    if (err < best_err) {
+      best_err  = err;
+      best_mode = mode;
+    }
+  }
+
+  if (pbest_mode)
+    *pbest_mode = best_mode;
+
+  return best_err;
+}
+
+static void update_mbgraph_mb_stats
+(
+  VP9_COMP *cpi,
+  MBGRAPH_MB_STATS *stats,
+  YV12_BUFFER_CONFIG *buf,
+  int mb_y_offset,
+  YV12_BUFFER_CONFIG *golden_ref,
+  int_mv *prev_golden_ref_mv,
+  int gld_y_offset,
+  YV12_BUFFER_CONFIG *alt_ref,
+  int_mv *prev_alt_ref_mv,
+  int arf_y_offset,
+  int mb_row,
+  int mb_col
+) {
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int intra_error;
+  VP9_COMMON *cm = &cpi->common;
+
+  // FIXME in practice we're completely ignoring chroma here
+  x->plane[0].src.buf = buf->y_buffer + mb_y_offset;
+  x->plane[0].src.stride = buf->y_stride;
+
+  xd->plane[0].dst.buf = cm->yv12_fb[cm->new_fb_idx].y_buffer + mb_y_offset;
+  xd->plane[0].dst.stride = cm->yv12_fb[cm->new_fb_idx].y_stride;
+
+  // do intra 16x16 prediction
+  intra_error = find_best_16x16_intra(cpi, mb_y_offset,
+                                      &stats->ref[INTRA_FRAME].m.mode);
+  if (intra_error <= 0)
+    intra_error = 1;
+  stats->ref[INTRA_FRAME].err = intra_error;
+
+  // Golden frame MV search, if it exists and is different than last frame
+  if (golden_ref) {
+    int g_motion_error;
+    xd->plane[0].pre[0].buf = golden_ref->y_buffer + mb_y_offset;
+    xd->plane[0].pre[0].stride = golden_ref->y_stride;
+    g_motion_error = do_16x16_motion_search(cpi,
+                                            prev_golden_ref_mv,
+                                            &stats->ref[GOLDEN_FRAME].m.mv,
+                                            mb_y_offset, gld_y_offset,
+                                            mb_row, mb_col);
+    stats->ref[GOLDEN_FRAME].err = g_motion_error;
+  } else {
+    stats->ref[GOLDEN_FRAME].err = INT_MAX;
+    stats->ref[GOLDEN_FRAME].m.mv.as_int = 0;
+  }
+
+  // Alt-ref frame MV search, if it exists and is different than last/golden frame
+  if (alt_ref) {
+    int a_motion_error;
+    xd->plane[0].pre[0].buf = alt_ref->y_buffer + mb_y_offset;
+    xd->plane[0].pre[0].stride = alt_ref->y_stride;
+    a_motion_error = do_16x16_zerozero_search(cpi,
+                                              &stats->ref[ALTREF_FRAME].m.mv,
+                                              mb_y_offset, arf_y_offset);
+
+    stats->ref[ALTREF_FRAME].err = a_motion_error;
+  } else {
+    stats->ref[ALTREF_FRAME].err = INT_MAX;
+    stats->ref[ALTREF_FRAME].m.mv.as_int = 0;
+  }
+}
+
+static void update_mbgraph_frame_stats(VP9_COMP *cpi,
+                                       MBGRAPH_FRAME_STATS *stats,
+                                       YV12_BUFFER_CONFIG *buf,
+                                       YV12_BUFFER_CONFIG *golden_ref,
+                                       YV12_BUFFER_CONFIG *alt_ref) {
+  MACROBLOCK *const x = &cpi->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  VP9_COMMON *const cm = &cpi->common;
+
+  int mb_col, mb_row, offset = 0;
+  int mb_y_offset = 0, arf_y_offset = 0, gld_y_offset = 0;
+  int_mv arf_top_mv, gld_top_mv;
+  MODE_INFO mi_local;
+
+  // Make sure the mi context starts in a consistent state.
+  memset(&mi_local, 0, sizeof(mi_local));
+
+  // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+  arf_top_mv.as_int = 0;
+  gld_top_mv.as_int = 0;
+  x->mv_row_min     = -(VP9BORDERINPIXELS - 8 - VP9_INTERP_EXTEND);
+  x->mv_row_max     = (cm->mb_rows - 1) * 8 + VP9BORDERINPIXELS
+                      - 8 - VP9_INTERP_EXTEND;
+  xd->up_available  = 0;
+  xd->plane[0].dst.stride  = buf->y_stride;
+  xd->plane[0].pre[0].stride  = buf->y_stride;
+  xd->plane[1].dst.stride = buf->uv_stride;
+  xd->mode_info_context = &mi_local;
+  mi_local.mbmi.sb_type = BLOCK_SIZE_MB16X16;
+  mi_local.mbmi.ref_frame[0] = LAST_FRAME;
+  mi_local.mbmi.ref_frame[1] = NONE;
+
+  for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
+    int_mv arf_left_mv, gld_left_mv;
+    int mb_y_in_offset  = mb_y_offset;
+    int arf_y_in_offset = arf_y_offset;
+    int gld_y_in_offset = gld_y_offset;
+
+    // Set up limit values for motion vectors to prevent them extending outside the UMV borders
+    arf_left_mv.as_int = arf_top_mv.as_int;
+    gld_left_mv.as_int = gld_top_mv.as_int;
+    x->mv_col_min      = -(VP9BORDERINPIXELS - 8 - VP9_INTERP_EXTEND);
+    x->mv_col_max      = (cm->mb_cols - 1) * 8 + VP9BORDERINPIXELS
+                         - 8 - VP9_INTERP_EXTEND;
+    xd->left_available = 0;
+
+    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+      MBGRAPH_MB_STATS *mb_stats = &stats->mb_stats[offset + mb_col];
+
+      update_mbgraph_mb_stats(cpi, mb_stats, buf, mb_y_in_offset,
+                              golden_ref, &gld_left_mv, gld_y_in_offset,
+                              alt_ref,    &arf_left_mv, arf_y_in_offset,
+                              mb_row, mb_col);
+      arf_left_mv.as_int = mb_stats->ref[ALTREF_FRAME].m.mv.as_int;
+      gld_left_mv.as_int = mb_stats->ref[GOLDEN_FRAME].m.mv.as_int;
+      if (mb_col == 0) {
+        arf_top_mv.as_int = arf_left_mv.as_int;
+        gld_top_mv.as_int = gld_left_mv.as_int;
+      }
+      xd->left_available = 1;
+      mb_y_in_offset    += 16;
+      gld_y_in_offset   += 16;
+      arf_y_in_offset   += 16;
+      x->mv_col_min     -= 16;
+      x->mv_col_max     -= 16;
+    }
+    xd->up_available = 1;
+    mb_y_offset     += buf->y_stride * 16;
+    gld_y_offset    += golden_ref->y_stride * 16;
+    if (alt_ref)
+      arf_y_offset    += alt_ref->y_stride * 16;
+    x->mv_row_min   -= 16;
+    x->mv_row_max   -= 16;
+    offset          += cm->mb_cols;
+  }
+}
+
+// void separate_arf_mbs_byzz
+static void separate_arf_mbs(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  int mb_col, mb_row, offset, i;
+  int ncnt[4];
+  int n_frames = cpi->mbgraph_n_frames;
+
+  int *arf_not_zz;
+
+  CHECK_MEM_ERROR(arf_not_zz,
+                  vpx_calloc(cm->mb_rows * cm->mb_cols * sizeof(*arf_not_zz), 1));
+
+  // We are not interested in results beyond the alt ref itself.
+  if (n_frames > cpi->frames_till_gf_update_due)
+    n_frames = cpi->frames_till_gf_update_due;
+
+  // defer cost to reference frames
+  for (i = n_frames - 1; i >= 0; i--) {
+    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
+
+    for (offset = 0, mb_row = 0; mb_row < cm->mb_rows;
+         offset += cm->mb_cols, mb_row++) {
+      for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+        MBGRAPH_MB_STATS *mb_stats = &frame_stats->mb_stats[offset + mb_col];
+
+        int altref_err = mb_stats->ref[ALTREF_FRAME].err;
+        int intra_err  = mb_stats->ref[INTRA_FRAME ].err;
+        int golden_err = mb_stats->ref[GOLDEN_FRAME].err;
+
+        // Test for altref vs intra and gf and that its mv was 0,0.
+        if (altref_err > 1000 ||
+            altref_err > intra_err ||
+            altref_err > golden_err) {
+          arf_not_zz[offset + mb_col]++;
+        }
+      }
+    }
+  }
+
+  vpx_memset(ncnt, 0, sizeof(ncnt));
+  for (offset = 0, mb_row = 0; mb_row < cm->mb_rows;
+       offset += cm->mb_cols, mb_row++) {
+    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
+      // If any of the blocks in the sequence failed then the MB
+      // goes in segment 0
+      if (arf_not_zz[offset + mb_col]) {
+        ncnt[0]++;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col] = 0;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + 1] = 0;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + cm->mi_cols] = 0;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + cm->mi_cols + 1] = 0;
+      } else {
+        cpi->segmentation_map[offset * 4 + 2 * mb_col] = 1;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + 1] = 1;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + cm->mi_cols] = 1;
+        cpi->segmentation_map[offset * 4 + 2 * mb_col + cm->mi_cols + 1] = 1;
+        ncnt[1]++;
+      }
+    }
+  }
+
+  // Only bother with segmentation if over 10% of the MBs in static segment
+  // if ( ncnt[1] && (ncnt[0] / ncnt[1] < 10) )
+  if (1) {
+    // Note % of blocks that are marked as static
+    if (cm->MBs)
+      cpi->static_mb_pct = (ncnt[1] * 100) / cm->MBs;
+
+    // This error case should not be reachable as this function should
+    // never be called with the common data structure uninitialized.
+    else
+      cpi->static_mb_pct = 0;
+
+    cpi->seg0_cnt = ncnt[0];
+    vp9_enable_segmentation((VP9_PTR)cpi);
+  } else {
+    cpi->static_mb_pct = 0;
+    vp9_disable_segmentation((VP9_PTR)cpi);
+  }
+
+  // Free localy allocated storage
+  vpx_free(arf_not_zz);
+}
+
+void vp9_update_mbgraph_stats(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  int i, n_frames = vp9_lookahead_depth(cpi->lookahead);
+  YV12_BUFFER_CONFIG *golden_ref =
+      &cm->yv12_fb[cm->ref_frame_map[cpi->gld_fb_idx]];
+
+  // we need to look ahead beyond where the ARF transitions into
+  // being a GF - so exit if we don't look ahead beyond that
+  if (n_frames <= cpi->frames_till_gf_update_due)
+    return;
+  if (n_frames > (int)cpi->common.frames_till_alt_ref_frame)
+    n_frames = cpi->common.frames_till_alt_ref_frame;
+  if (n_frames > MAX_LAG_BUFFERS)
+    n_frames = MAX_LAG_BUFFERS;
+
+  cpi->mbgraph_n_frames = n_frames;
+  for (i = 0; i < n_frames; i++) {
+    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
+    vpx_memset(frame_stats->mb_stats, 0,
+               cm->mb_rows * cm->mb_cols * sizeof(*cpi->mbgraph_stats[i].mb_stats));
+  }
+
+  // do motion search to find contribution of each reference to data
+  // later on in this GF group
+  // FIXME really, the GF/last MC search should be done forward, and
+  // the ARF MC search backwards, to get optimal results for MV caching
+  for (i = 0; i < n_frames; i++) {
+    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
+    struct lookahead_entry *q_cur = vp9_lookahead_peek(cpi->lookahead, i);
+
+    assert(q_cur != NULL);
+
+    update_mbgraph_frame_stats(cpi, frame_stats, &q_cur->img,
+                               golden_ref, cpi->Source);
+  }
+
+  vp9_clear_system_state();  // __asm emms;
+
+  separate_arf_mbs(cpi);
+}
diff --git a/libvpx/vp9/encoder/vp9_mbgraph.h b/libvpx/vp9/encoder/vp9_mbgraph.h
new file mode 100644
index 0000000..c5bca4d
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_mbgraph.h
@@ -0,0 +1,16 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_MBGRAPH_H_
+#define VP9_ENCODER_VP9_MBGRAPH_H_
+
+void vp9_update_mbgraph_stats(VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_MBGRAPH_H_
diff --git a/libvpx/vp9/encoder/vp9_mcomp.c b/libvpx/vp9/encoder/vp9_mcomp.c
new file mode 100644
index 0000000..2e99736
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_mcomp.c
@@ -0,0 +1,2429 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdio.h>
+#include <limits.h>
+#include <math.h>
+
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vpx_mem/vpx_mem.h"
+#include "./vpx_config.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_common.h"
+
+void vp9_clamp_mv_min_max(MACROBLOCK *x, int_mv *ref_mv) {
+  int col_min = (ref_mv->as_mv.col >> 3) - MAX_FULL_PEL_VAL +
+                                 ((ref_mv->as_mv.col & 7) ? 1 : 0);
+  int row_min = (ref_mv->as_mv.row >> 3) - MAX_FULL_PEL_VAL +
+                                 ((ref_mv->as_mv.row & 7) ? 1 : 0);
+  int col_max = (ref_mv->as_mv.col >> 3) + MAX_FULL_PEL_VAL;
+  int row_max = (ref_mv->as_mv.row >> 3) + MAX_FULL_PEL_VAL;
+
+  /* Get intersection of UMV window and valid MV window to reduce # of checks in diamond search. */
+  if (x->mv_col_min < col_min)
+    x->mv_col_min = col_min;
+  if (x->mv_col_max > col_max)
+    x->mv_col_max = col_max;
+  if (x->mv_row_min < row_min)
+    x->mv_row_min = row_min;
+  if (x->mv_row_max > row_max)
+    x->mv_row_max = row_max;
+}
+
+int vp9_init_search_range(int width, int height) {
+  int sr = 0;
+  int frm = MIN(width, height);
+
+  while ((frm << sr) < MAX_FULL_PEL_VAL)
+    sr++;
+
+  if (sr)
+    sr--;
+
+  return sr;
+}
+
+int vp9_mv_bit_cost(int_mv *mv, int_mv *ref, int *mvjcost, int *mvcost[2],
+                    int weight, int ishp) {
+  MV v;
+  v.row = mv->as_mv.row - ref->as_mv.row;
+  v.col = mv->as_mv.col - ref->as_mv.col;
+  return ROUND_POWER_OF_TWO((mvjcost[vp9_get_mv_joint(&v)] +
+                             mvcost[0][v.row] +
+                             mvcost[1][v.col]) * weight, 7);
+}
+
+static int mv_err_cost(int_mv *mv, int_mv *ref, int *mvjcost, int *mvcost[2],
+                       int error_per_bit, int ishp) {
+  if (mvcost) {
+    MV v;
+    v.row = mv->as_mv.row - ref->as_mv.row;
+    v.col = mv->as_mv.col - ref->as_mv.col;
+    return ROUND_POWER_OF_TWO((mvjcost[vp9_get_mv_joint(&v)] +
+                               mvcost[0][v.row] +
+                               mvcost[1][v.col]) * error_per_bit, 13);
+  }
+  return 0;
+}
+
+static int mvsad_err_cost(int_mv *mv, int_mv *ref, int *mvjsadcost,
+                          int *mvsadcost[2], int error_per_bit) {
+  if (mvsadcost) {
+    MV v;
+    v.row = mv->as_mv.row - ref->as_mv.row;
+    v.col = mv->as_mv.col - ref->as_mv.col;
+    return ROUND_POWER_OF_TWO((mvjsadcost[vp9_get_mv_joint(&v)] +
+                               mvsadcost[0][v.row] +
+                               mvsadcost[1][v.col]) * error_per_bit, 8);
+  }
+  return 0;
+}
+
+void vp9_init_dsmotion_compensation(MACROBLOCK *x, int stride) {
+  int len;
+  int search_site_count = 0;
+
+  // Generate offsets for 4 search sites per step.
+  x->ss[search_site_count].mv.col = 0;
+  x->ss[search_site_count].mv.row = 0;
+  x->ss[search_site_count].offset = 0;
+  search_site_count++;
+
+  for (len = MAX_FIRST_STEP; len > 0; len /= 2) {
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = 0;
+    x->ss[search_site_count].mv.row = -len;
+    x->ss[search_site_count].offset = -len * stride;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = 0;
+    x->ss[search_site_count].mv.row = len;
+    x->ss[search_site_count].offset = len * stride;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = -len;
+    x->ss[search_site_count].mv.row = 0;
+    x->ss[search_site_count].offset = -len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = len;
+    x->ss[search_site_count].mv.row = 0;
+    x->ss[search_site_count].offset = len;
+    search_site_count++;
+  }
+
+  x->ss_count = search_site_count;
+  x->searches_per_step = 4;
+}
+
+void vp9_init3smotion_compensation(MACROBLOCK *x, int stride) {
+  int len;
+  int search_site_count = 0;
+
+  // Generate offsets for 8 search sites per step.
+  x->ss[search_site_count].mv.col = 0;
+  x->ss[search_site_count].mv.row = 0;
+  x->ss[search_site_count].offset = 0;
+  search_site_count++;
+
+  for (len = MAX_FIRST_STEP; len > 0; len /= 2) {
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = 0;
+    x->ss[search_site_count].mv.row = -len;
+    x->ss[search_site_count].offset = -len * stride;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = 0;
+    x->ss[search_site_count].mv.row = len;
+    x->ss[search_site_count].offset = len * stride;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = -len;
+    x->ss[search_site_count].mv.row = 0;
+    x->ss[search_site_count].offset = -len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = len;
+    x->ss[search_site_count].mv.row = 0;
+    x->ss[search_site_count].offset = len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = -len;
+    x->ss[search_site_count].mv.row = -len;
+    x->ss[search_site_count].offset = -len * stride - len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = len;
+    x->ss[search_site_count].mv.row = -len;
+    x->ss[search_site_count].offset = -len * stride + len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = -len;
+    x->ss[search_site_count].mv.row = len;
+    x->ss[search_site_count].offset = len * stride - len;
+    search_site_count++;
+
+    // Compute offsets for search sites.
+    x->ss[search_site_count].mv.col = len;
+    x->ss[search_site_count].mv.row = len;
+    x->ss[search_site_count].offset = len * stride + len;
+    search_site_count++;
+  }
+
+  x->ss_count = search_site_count;
+  x->searches_per_step = 8;
+}
+
+/*
+ * To avoid the penalty for crossing cache-line read, preload the reference
+ * area in a small buffer, which is aligned to make sure there won't be crossing
+ * cache-line read while reading from this buffer. This reduced the cpu
+ * cycles spent on reading ref data in sub-pixel filter functions.
+ * TODO: Currently, since sub-pixel search range here is -3 ~ 3, copy 22 rows x
+ * 32 cols area that is enough for 16x16 macroblock. Later, for SPLITMV, we
+ * could reduce the area.
+ */
+
+/* estimated cost of a motion vector (r,c) */
+#define MVC(r, c)                                       \
+    (mvcost ?                                           \
+     ((mvjcost[((r) != rr) * 2 + ((c) != rc)] +         \
+       mvcost[0][((r) - rr)] + mvcost[1][((c) - rc)]) * \
+      error_per_bit + 4096) >> 13 : 0)
+
+
+#define SP(x) (((x) & 7) << 1)  // convert motion vector component to offset
+                                // for svf calc
+
+#define IFMVCV(r, c, s, e)                                \
+    if (c >= minc && c <= maxc && r >= minr && r <= maxr) \
+      s                                                   \
+    else                                                  \
+      e;
+
+/* pointer to predictor base of a motionvector */
+#define PRE(r, c) (y + (((r) >> 3) * y_stride + ((c) >> 3) -(offset)))
+
+/* returns subpixel variance error function */
+#define DIST(r, c) \
+    vfp->svf(PRE(r, c), y_stride, SP(c), SP(r), z, src_stride, &sse)
+
+/* checks if (r, c) has better score than previous best */
+#define CHECK_BETTER(v, r, c) \
+    IFMVCV(r, c, {                                                       \
+      thismse = (DIST(r, c));                                            \
+      if ((v = MVC(r, c) + thismse) < besterr) {                         \
+        besterr = v;                                                     \
+        br = r;                                                          \
+        bc = c;                                                          \
+        *distortion = thismse;                                           \
+        *sse1 = sse;                                                     \
+      }                                                                  \
+    },                                                                   \
+    v = INT_MAX;)
+
+int vp9_find_best_sub_pixel_step_iteratively(MACROBLOCK *x,
+                                             int_mv *bestmv, int_mv *ref_mv,
+                                             int error_per_bit,
+                                             const vp9_variance_fn_ptr_t *vfp,
+                                             int *mvjcost, int *mvcost[2],
+                                             int *distortion,
+                                             unsigned int *sse1) {
+  uint8_t *z = x->plane[0].src.buf;
+  int src_stride = x->plane[0].src.stride;
+  MACROBLOCKD *xd = &x->e_mbd;
+
+  int rr, rc, br, bc, hstep;
+  int tr, tc;
+  unsigned int besterr = INT_MAX;
+  unsigned int left, right, up, down, diag;
+  unsigned int sse;
+  unsigned int whichdir;
+  unsigned int halfiters = 4;
+  unsigned int quarteriters = 4;
+  unsigned int eighthiters = 4;
+  int thismse;
+  int maxc, minc, maxr, minr;
+  int y_stride;
+  int offset;
+  int usehp = xd->allow_high_precision_mv;
+
+  uint8_t *y = xd->plane[0].pre[0].buf +
+               (bestmv->as_mv.row) * xd->plane[0].pre[0].stride +
+               bestmv->as_mv.col;
+
+  y_stride = xd->plane[0].pre[0].stride;
+
+  rr = ref_mv->as_mv.row;
+  rc = ref_mv->as_mv.col;
+  br = bestmv->as_mv.row << 3;
+  bc = bestmv->as_mv.col << 3;
+  hstep = 4;
+  minc = MAX(x->mv_col_min << 3, (ref_mv->as_mv.col) - ((1 << MV_MAX_BITS) - 1));
+  maxc = MIN(x->mv_col_max << 3, (ref_mv->as_mv.col) + ((1 << MV_MAX_BITS) - 1));
+  minr = MAX(x->mv_row_min << 3, (ref_mv->as_mv.row) - ((1 << MV_MAX_BITS) - 1));
+  maxr = MIN(x->mv_row_max << 3, (ref_mv->as_mv.row) + ((1 << MV_MAX_BITS) - 1));
+
+  tr = br;
+  tc = bc;
+
+
+  offset = (bestmv->as_mv.row) * y_stride + bestmv->as_mv.col;
+
+  // central mv
+  bestmv->as_mv.row <<= 3;
+  bestmv->as_mv.col <<= 3;
+
+  // calculate central point error
+  besterr = vfp->vf(y, y_stride, z, src_stride, sse1);
+  *distortion = besterr;
+  besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost,
+                         error_per_bit, xd->allow_high_precision_mv);
+
+  // TODO: Each subsequent iteration checks at least one point in
+  // common with the last iteration could be 2 ( if diag selected)
+  while (--halfiters) {
+    // 1/2 pel
+    CHECK_BETTER(left, tr, tc - hstep);
+    CHECK_BETTER(right, tr, tc + hstep);
+    CHECK_BETTER(up, tr - hstep, tc);
+    CHECK_BETTER(down, tr + hstep, tc);
+
+    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+    switch (whichdir) {
+      case 0:
+        CHECK_BETTER(diag, tr - hstep, tc - hstep);
+        break;
+      case 1:
+        CHECK_BETTER(diag, tr - hstep, tc + hstep);
+        break;
+      case 2:
+        CHECK_BETTER(diag, tr + hstep, tc - hstep);
+        break;
+      case 3:
+        CHECK_BETTER(diag, tr + hstep, tc + hstep);
+        break;
+    }
+
+    // no reason to check the same one again.
+    if (tr == br && tc == bc)
+      break;
+
+    tr = br;
+    tc = bc;
+  }
+
+  // TODO: Each subsequent iteration checks at least one point in common with
+  // the last iteration could be 2 ( if diag selected) 1/4 pel
+  hstep >>= 1;
+  while (--quarteriters) {
+    CHECK_BETTER(left, tr, tc - hstep);
+    CHECK_BETTER(right, tr, tc + hstep);
+    CHECK_BETTER(up, tr - hstep, tc);
+    CHECK_BETTER(down, tr + hstep, tc);
+
+    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+    switch (whichdir) {
+      case 0:
+        CHECK_BETTER(diag, tr - hstep, tc - hstep);
+        break;
+      case 1:
+        CHECK_BETTER(diag, tr - hstep, tc + hstep);
+        break;
+      case 2:
+        CHECK_BETTER(diag, tr + hstep, tc - hstep);
+        break;
+      case 3:
+        CHECK_BETTER(diag, tr + hstep, tc + hstep);
+        break;
+    }
+
+    // no reason to check the same one again.
+    if (tr == br && tc == bc)
+      break;
+
+    tr = br;
+    tc = bc;
+  }
+
+  if (xd->allow_high_precision_mv) {
+    usehp = vp9_use_nmv_hp(&ref_mv->as_mv);
+  } else {
+    usehp = 0;
+  }
+
+  if (usehp) {
+    hstep >>= 1;
+    while (--eighthiters) {
+      CHECK_BETTER(left, tr, tc - hstep);
+      CHECK_BETTER(right, tr, tc + hstep);
+      CHECK_BETTER(up, tr - hstep, tc);
+      CHECK_BETTER(down, tr + hstep, tc);
+
+      whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+      switch (whichdir) {
+        case 0:
+          CHECK_BETTER(diag, tr - hstep, tc - hstep);
+          break;
+        case 1:
+          CHECK_BETTER(diag, tr - hstep, tc + hstep);
+          break;
+        case 2:
+          CHECK_BETTER(diag, tr + hstep, tc - hstep);
+          break;
+        case 3:
+          CHECK_BETTER(diag, tr + hstep, tc + hstep);
+          break;
+      }
+
+      // no reason to check the same one again.
+      if (tr == br && tc == bc)
+        break;
+
+      tr = br;
+      tc = bc;
+    }
+  }
+  bestmv->as_mv.row = br;
+  bestmv->as_mv.col = bc;
+
+  if ((abs(bestmv->as_mv.col - ref_mv->as_mv.col) > (MAX_FULL_PEL_VAL << 3)) ||
+      (abs(bestmv->as_mv.row - ref_mv->as_mv.row) > (MAX_FULL_PEL_VAL << 3)))
+    return INT_MAX;
+
+  return besterr;
+}
+
+#undef DIST
+/* returns subpixel variance error function */
+#define DIST(r, c) \
+    vfp->svaf(PRE(r, c), y_stride, SP(c), SP(r), \
+              z, src_stride, &sse, second_pred)
+
+int vp9_find_best_sub_pixel_comp(MACROBLOCK *x,
+                                 int_mv *bestmv, int_mv *ref_mv,
+                                 int error_per_bit,
+                                 const vp9_variance_fn_ptr_t *vfp,
+                                 int *mvjcost, int *mvcost[2],
+                                 int *distortion,
+                                 unsigned int *sse1,
+                                 const uint8_t *second_pred, int w, int h) {
+  uint8_t *z = x->plane[0].src.buf;
+  int src_stride = x->plane[0].src.stride;
+  MACROBLOCKD *xd = &x->e_mbd;
+
+  int rr, rc, br, bc, hstep;
+  int tr, tc;
+  unsigned int besterr = INT_MAX;
+  unsigned int left, right, up, down, diag;
+  unsigned int sse;
+  unsigned int whichdir;
+  unsigned int halfiters = 4;
+  unsigned int quarteriters = 4;
+  unsigned int eighthiters = 4;
+  int thismse;
+  int maxc, minc, maxr, minr;
+  int y_stride;
+  int offset;
+  int usehp = xd->allow_high_precision_mv;
+
+  uint8_t *comp_pred = vpx_memalign(16, w * h * sizeof(uint8_t));
+  uint8_t *y = xd->plane[0].pre[0].buf +
+               (bestmv->as_mv.row) * xd->plane[0].pre[0].stride +
+               bestmv->as_mv.col;
+
+  y_stride = xd->plane[0].pre[0].stride;
+
+  rr = ref_mv->as_mv.row;
+  rc = ref_mv->as_mv.col;
+  br = bestmv->as_mv.row << 3;
+  bc = bestmv->as_mv.col << 3;
+  hstep = 4;
+  minc = MAX(x->mv_col_min << 3, (ref_mv->as_mv.col) -
+             ((1 << MV_MAX_BITS) - 1));
+  maxc = MIN(x->mv_col_max << 3, (ref_mv->as_mv.col) +
+             ((1 << MV_MAX_BITS) - 1));
+  minr = MAX(x->mv_row_min << 3, (ref_mv->as_mv.row) -
+             ((1 << MV_MAX_BITS) - 1));
+  maxr = MIN(x->mv_row_max << 3, (ref_mv->as_mv.row) +
+             ((1 << MV_MAX_BITS) - 1));
+
+  tr = br;
+  tc = bc;
+
+
+  offset = (bestmv->as_mv.row) * y_stride + bestmv->as_mv.col;
+
+  // central mv
+  bestmv->as_mv.row <<= 3;
+  bestmv->as_mv.col <<= 3;
+
+  // calculate central point error
+  // TODO(yunqingwang): central pointer error was already calculated in full-
+  // pixel search, and can be passed in this function.
+  comp_avg_pred(comp_pred, second_pred, w, h, y, y_stride);
+  besterr = vfp->vf(comp_pred, w, z, src_stride, sse1);
+  *distortion = besterr;
+  besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost,
+                         error_per_bit, xd->allow_high_precision_mv);
+
+  // Each subsequent iteration checks at least one point in
+  // common with the last iteration could be 2 ( if diag selected)
+  while (--halfiters) {
+    // 1/2 pel
+    CHECK_BETTER(left, tr, tc - hstep);
+    CHECK_BETTER(right, tr, tc + hstep);
+    CHECK_BETTER(up, tr - hstep, tc);
+    CHECK_BETTER(down, tr + hstep, tc);
+
+    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+    switch (whichdir) {
+      case 0:
+        CHECK_BETTER(diag, tr - hstep, tc - hstep);
+        break;
+      case 1:
+        CHECK_BETTER(diag, tr - hstep, tc + hstep);
+        break;
+      case 2:
+        CHECK_BETTER(diag, tr + hstep, tc - hstep);
+        break;
+      case 3:
+        CHECK_BETTER(diag, tr + hstep, tc + hstep);
+        break;
+    }
+
+    // no reason to check the same one again.
+    if (tr == br && tc == bc)
+      break;
+
+    tr = br;
+    tc = bc;
+  }
+
+  // Each subsequent iteration checks at least one point in common with
+  // the last iteration could be 2 ( if diag selected) 1/4 pel
+  hstep >>= 1;
+  while (--quarteriters) {
+    CHECK_BETTER(left, tr, tc - hstep);
+    CHECK_BETTER(right, tr, tc + hstep);
+    CHECK_BETTER(up, tr - hstep, tc);
+    CHECK_BETTER(down, tr + hstep, tc);
+
+    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+    switch (whichdir) {
+      case 0:
+        CHECK_BETTER(diag, tr - hstep, tc - hstep);
+        break;
+      case 1:
+        CHECK_BETTER(diag, tr - hstep, tc + hstep);
+        break;
+      case 2:
+        CHECK_BETTER(diag, tr + hstep, tc - hstep);
+        break;
+      case 3:
+        CHECK_BETTER(diag, tr + hstep, tc + hstep);
+        break;
+    }
+
+    // no reason to check the same one again.
+    if (tr == br && tc == bc)
+      break;
+
+    tr = br;
+    tc = bc;
+  }
+
+  if (xd->allow_high_precision_mv) {
+    usehp = vp9_use_nmv_hp(&ref_mv->as_mv);
+  } else {
+    usehp = 0;
+  }
+
+  if (usehp) {
+    hstep >>= 1;
+    while (--eighthiters) {
+      CHECK_BETTER(left, tr, tc - hstep);
+      CHECK_BETTER(right, tr, tc + hstep);
+      CHECK_BETTER(up, tr - hstep, tc);
+      CHECK_BETTER(down, tr + hstep, tc);
+
+      whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+      switch (whichdir) {
+        case 0:
+          CHECK_BETTER(diag, tr - hstep, tc - hstep);
+          break;
+        case 1:
+          CHECK_BETTER(diag, tr - hstep, tc + hstep);
+          break;
+        case 2:
+          CHECK_BETTER(diag, tr + hstep, tc - hstep);
+          break;
+        case 3:
+          CHECK_BETTER(diag, tr + hstep, tc + hstep);
+          break;
+      }
+
+      // no reason to check the same one again.
+      if (tr == br && tc == bc)
+        break;
+
+      tr = br;
+      tc = bc;
+    }
+  }
+  bestmv->as_mv.row = br;
+  bestmv->as_mv.col = bc;
+
+  vpx_free(comp_pred);
+
+  if ((abs(bestmv->as_mv.col - ref_mv->as_mv.col) > (MAX_FULL_PEL_VAL << 3)) ||
+      (abs(bestmv->as_mv.row - ref_mv->as_mv.row) > (MAX_FULL_PEL_VAL << 3)))
+    return INT_MAX;
+
+  return besterr;
+}
+
+
+#undef MVC
+#undef PRE
+#undef DIST
+#undef IFMVCV
+#undef CHECK_BETTER
+#undef MIN
+#undef MAX
+
+int vp9_find_best_sub_pixel_step(MACROBLOCK *x,
+                                 int_mv *bestmv, int_mv *ref_mv,
+                                 int error_per_bit,
+                                 const vp9_variance_fn_ptr_t *vfp,
+                                 int *mvjcost, int *mvcost[2], int *distortion,
+                                 unsigned int *sse1) {
+  int bestmse = INT_MAX;
+  int_mv startmv;
+  int_mv this_mv;
+  int_mv orig_mv;
+  int yrow_movedback = 0, ycol_movedback = 0;
+  uint8_t *z = x->plane[0].src.buf;
+  int src_stride = x->plane[0].src.stride;
+  int left, right, up, down, diag;
+  unsigned int sse;
+  int whichdir;
+  int thismse;
+  int y_stride;
+  MACROBLOCKD *xd = &x->e_mbd;
+  int usehp = xd->allow_high_precision_mv;
+
+  uint8_t *y = xd->plane[0].pre[0].buf +
+               (bestmv->as_mv.row) * xd->plane[0].pre[0].stride +
+               bestmv->as_mv.col;
+  y_stride = xd->plane[0].pre[0].stride;
+
+  // central mv
+  bestmv->as_mv.row <<= 3;
+  bestmv->as_mv.col <<= 3;
+  startmv = *bestmv;
+  orig_mv = *bestmv;
+
+  // calculate central point error
+  bestmse = vfp->vf(y, y_stride, z, src_stride, sse1);
+  *distortion = bestmse;
+  bestmse += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit,
+                         xd->allow_high_precision_mv);
+
+  // go left then right and check error
+  this_mv.as_mv.row = startmv.as_mv.row;
+  this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
+  thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, src_stride, &sse);
+  left = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (left < bestmse) {
+    *bestmv = this_mv;
+    bestmse = left;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.col += 8;
+  thismse = vfp->svf_halfpix_h(y, y_stride, z, src_stride, &sse);
+  right = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost,
+                                error_per_bit, xd->allow_high_precision_mv);
+
+  if (right < bestmse) {
+    *bestmv = this_mv;
+    bestmse = right;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // go up then down and check error
+  this_mv.as_mv.col = startmv.as_mv.col;
+  this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
+  thismse =  vfp->svf_halfpix_v(y - y_stride, y_stride, z, src_stride, &sse);
+  up = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                             xd->allow_high_precision_mv);
+
+  if (up < bestmse) {
+    *bestmv = this_mv;
+    bestmse = up;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.row += 8;
+  thismse = vfp->svf_halfpix_v(y, y_stride, z, src_stride, &sse);
+  down = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (down < bestmse) {
+    *bestmv = this_mv;
+    bestmse = down;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+
+  // now check 1 more diagonal
+  whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+  // for(whichdir =0;whichdir<4;whichdir++)
+  // {
+  this_mv = startmv;
+
+  switch (whichdir) {
+    case 0:
+      this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+      this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+      thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride, z, src_stride,
+                                    &sse);
+      break;
+    case 1:
+      this_mv.as_mv.col += 4;
+      this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+      thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride, z, src_stride,
+                                    &sse);
+      break;
+    case 2:
+      this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+      this_mv.as_mv.row += 4;
+      thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, src_stride, &sse);
+      break;
+    case 3:
+    default:
+      this_mv.as_mv.col += 4;
+      this_mv.as_mv.row += 4;
+      thismse = vfp->svf_halfpix_hv(y, y_stride, z, src_stride, &sse);
+      break;
+  }
+
+  diag = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (diag < bestmse) {
+    *bestmv = this_mv;
+    bestmse = diag;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+//  }
+
+
+  // time to check quarter pels.
+  if (bestmv->as_mv.row < startmv.as_mv.row) {
+    y -= y_stride;
+    yrow_movedback = 1;
+  }
+
+  if (bestmv->as_mv.col < startmv.as_mv.col) {
+    y--;
+    ycol_movedback = 1;
+  }
+
+  startmv = *bestmv;
+
+
+
+  // go left then right and check error
+  this_mv.as_mv.row = startmv.as_mv.row;
+
+  if (startmv.as_mv.col & 7) {
+    this_mv.as_mv.col = startmv.as_mv.col - 2;
+    thismse = vfp->svf(y, y_stride,
+                       SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                       z, src_stride, &sse);
+  } else {
+    this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
+    thismse = vfp->svf(y - 1, y_stride, SP(6), SP(this_mv.as_mv.row), z,
+                       src_stride, &sse);
+  }
+
+  left = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (left < bestmse) {
+    *bestmv = this_mv;
+    bestmse = left;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.col += 4;
+  thismse = vfp->svf(y, y_stride,
+                     SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                     z, src_stride, &sse);
+  right = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost,
+                                error_per_bit, xd->allow_high_precision_mv);
+
+  if (right < bestmse) {
+    *bestmv = this_mv;
+    bestmse = right;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // go up then down and check error
+  this_mv.as_mv.col = startmv.as_mv.col;
+
+  if (startmv.as_mv.row & 7) {
+    this_mv.as_mv.row = startmv.as_mv.row - 2;
+    thismse = vfp->svf(y, y_stride,
+                       SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                       z, src_stride, &sse);
+  } else {
+    this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
+    thismse = vfp->svf(y - y_stride, y_stride, SP(this_mv.as_mv.col), SP(6),
+                       z, src_stride, &sse);
+  }
+
+  up = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                             xd->allow_high_precision_mv);
+
+  if (up < bestmse) {
+    *bestmv = this_mv;
+    bestmse = up;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.row += 4;
+  thismse = vfp->svf(y, y_stride, SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                     z, src_stride, &sse);
+  down = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (down < bestmse) {
+    *bestmv = this_mv;
+    bestmse = down;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+
+  // now check 1 more diagonal
+  whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+//  for(whichdir=0;whichdir<4;whichdir++)
+//  {
+  this_mv = startmv;
+
+  switch (whichdir) {
+    case 0:
+
+      if (startmv.as_mv.row & 7) {
+        this_mv.as_mv.row -= 2;
+
+        if (startmv.as_mv.col & 7) {
+          this_mv.as_mv.col -= 2;
+          thismse = vfp->svf(y, y_stride,
+                             SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                             z, src_stride, &sse);
+        } else {
+          this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
+          thismse = vfp->svf(y - 1, y_stride,
+                             SP(6), SP(this_mv.as_mv.row), z, src_stride, &sse);
+        }
+      } else {
+        this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
+
+        if (startmv.as_mv.col & 7) {
+          this_mv.as_mv.col -= 2;
+          thismse = vfp->svf(y - y_stride, y_stride,
+                             SP(this_mv.as_mv.col), SP(6), z, src_stride, &sse);
+        } else {
+          this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
+          thismse = vfp->svf(y - y_stride - 1, y_stride,
+                             SP(6), SP(6), z, src_stride, &sse);
+        }
+      }
+
+      break;
+    case 1:
+      this_mv.as_mv.col += 2;
+
+      if (startmv.as_mv.row & 7) {
+        this_mv.as_mv.row -= 2;
+        thismse = vfp->svf(y, y_stride,
+                           SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                           z, src_stride, &sse);
+      } else {
+        this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
+        thismse = vfp->svf(y - y_stride, y_stride,
+                           SP(this_mv.as_mv.col), SP(6), z, src_stride, &sse);
+      }
+
+      break;
+    case 2:
+      this_mv.as_mv.row += 2;
+
+      if (startmv.as_mv.col & 7) {
+        this_mv.as_mv.col -= 2;
+        thismse = vfp->svf(y, y_stride,
+                           SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                           z, src_stride, &sse);
+      } else {
+        this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
+        thismse = vfp->svf(y - 1, y_stride, SP(6), SP(this_mv.as_mv.row), z,
+                           src_stride, &sse);
+      }
+
+      break;
+    case 3:
+      this_mv.as_mv.col += 2;
+      this_mv.as_mv.row += 2;
+      thismse = vfp->svf(y, y_stride,
+                         SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                         z, src_stride, &sse);
+      break;
+  }
+
+  diag = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (diag < bestmse) {
+    *bestmv = this_mv;
+    bestmse = diag;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  if (x->e_mbd.allow_high_precision_mv) {
+    usehp = vp9_use_nmv_hp(&ref_mv->as_mv);
+  } else {
+    usehp = 0;
+  }
+  if (!usehp)
+    return bestmse;
+
+  /* Now do 1/8th pixel */
+  if (bestmv->as_mv.row < orig_mv.as_mv.row && !yrow_movedback) {
+    y -= y_stride;
+    yrow_movedback = 1;
+  }
+
+  if (bestmv->as_mv.col < orig_mv.as_mv.col && !ycol_movedback) {
+    y--;
+    ycol_movedback = 1;
+  }
+
+  startmv = *bestmv;
+
+  // go left then right and check error
+  this_mv.as_mv.row = startmv.as_mv.row;
+
+  if (startmv.as_mv.col & 7) {
+    this_mv.as_mv.col = startmv.as_mv.col - 1;
+    thismse = vfp->svf(y, y_stride,
+                       SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                       z, src_stride, &sse);
+  } else {
+    this_mv.as_mv.col = (startmv.as_mv.col - 8) | 7;
+    thismse = vfp->svf(y - 1, y_stride, SP(7), SP(this_mv.as_mv.row),
+                       z, src_stride, &sse);
+  }
+
+  left = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (left < bestmse) {
+    *bestmv = this_mv;
+    bestmse = left;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.col += 2;
+  thismse = vfp->svf(y, y_stride, SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                     z, src_stride, &sse);
+  right = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost,
+                                error_per_bit, xd->allow_high_precision_mv);
+
+  if (right < bestmse) {
+    *bestmv = this_mv;
+    bestmse = right;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // go up then down and check error
+  this_mv.as_mv.col = startmv.as_mv.col;
+
+  if (startmv.as_mv.row & 7) {
+    this_mv.as_mv.row = startmv.as_mv.row - 1;
+    thismse = vfp->svf(y, y_stride,
+                       SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                       z, src_stride, &sse);
+  } else {
+    this_mv.as_mv.row = (startmv.as_mv.row - 8) | 7;
+    thismse = vfp->svf(y - y_stride, y_stride,
+                       SP(this_mv.as_mv.col), SP(7), z, src_stride, &sse);
+  }
+
+  up = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                             xd->allow_high_precision_mv);
+
+  if (up < bestmse) {
+    *bestmv = this_mv;
+    bestmse = up;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.row += 2;
+  thismse = vfp->svf(y, y_stride,
+                     SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                     z, src_stride, &sse);
+  down = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (down < bestmse) {
+    *bestmv = this_mv;
+    bestmse = down;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // now check 1 more diagonal
+  whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+
+//  for(whichdir=0;whichdir<4;whichdir++)
+//  {
+  this_mv = startmv;
+
+  switch (whichdir) {
+    case 0:
+
+      if (startmv.as_mv.row & 7) {
+        this_mv.as_mv.row -= 1;
+
+        if (startmv.as_mv.col & 7) {
+          this_mv.as_mv.col -= 1;
+          thismse = vfp->svf(y, y_stride,
+                             SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                             z, src_stride, &sse);
+        } else {
+          this_mv.as_mv.col = (startmv.as_mv.col - 8) | 7;
+          thismse = vfp->svf(y - 1, y_stride,
+                             SP(7), SP(this_mv.as_mv.row),
+                             z, src_stride, &sse);
+        }
+      } else {
+        this_mv.as_mv.row = (startmv.as_mv.row - 8) | 7;
+
+        if (startmv.as_mv.col & 7) {
+          this_mv.as_mv.col -= 1;
+          thismse = vfp->svf(y - y_stride, y_stride,
+                             SP(this_mv.as_mv.col), SP(7), z, src_stride, &sse);
+        } else {
+          this_mv.as_mv.col = (startmv.as_mv.col - 8) | 7;
+          thismse = vfp->svf(y - y_stride - 1, y_stride,
+                             SP(7), SP(7), z, src_stride, &sse);
+        }
+      }
+
+      break;
+    case 1:
+      this_mv.as_mv.col += 1;
+
+      if (startmv.as_mv.row & 7) {
+        this_mv.as_mv.row -= 1;
+        thismse = vfp->svf(y, y_stride,
+                           SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                           z, src_stride, &sse);
+      } else {
+        this_mv.as_mv.row = (startmv.as_mv.row - 8) | 7;
+        thismse = vfp->svf(y - y_stride, y_stride,
+                           SP(this_mv.as_mv.col), SP(7), z, src_stride, &sse);
+      }
+
+      break;
+    case 2:
+      this_mv.as_mv.row += 1;
+
+      if (startmv.as_mv.col & 7) {
+        this_mv.as_mv.col -= 1;
+        thismse = vfp->svf(y, y_stride,
+                           SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                           z, src_stride, &sse);
+      } else {
+        this_mv.as_mv.col = (startmv.as_mv.col - 8) | 7;
+        thismse = vfp->svf(y - 1, y_stride,
+                           SP(7), SP(this_mv.as_mv.row), z, src_stride, &sse);
+      }
+
+      break;
+    case 3:
+      this_mv.as_mv.col += 1;
+      this_mv.as_mv.row += 1;
+      thismse = vfp->svf(y, y_stride,
+                         SP(this_mv.as_mv.col), SP(this_mv.as_mv.row),
+                         z, src_stride, &sse);
+      break;
+  }
+
+  diag = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (diag < bestmse) {
+    *bestmv = this_mv;
+    bestmse = diag;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  return bestmse;
+}
+
+#undef SP
+
+int vp9_find_best_half_pixel_step(MACROBLOCK *x,
+                                  int_mv *bestmv, int_mv *ref_mv,
+                                  int error_per_bit,
+                                  const vp9_variance_fn_ptr_t *vfp,
+                                  int *mvjcost, int *mvcost[2],
+                                  int *distortion,
+                                  unsigned int *sse1) {
+  int bestmse = INT_MAX;
+  int_mv startmv;
+  int_mv this_mv;
+  uint8_t *z = x->plane[0].src.buf;
+  int src_stride = x->plane[0].src.stride;
+  int left, right, up, down, diag;
+  unsigned int sse;
+  int whichdir;
+  int thismse;
+  int y_stride;
+  MACROBLOCKD *xd = &x->e_mbd;
+
+  uint8_t *y = xd->plane[0].pre[0].buf +
+      (bestmv->as_mv.row) * xd->plane[0].pre[0].stride + bestmv->as_mv.col;
+  y_stride = xd->plane[0].pre[0].stride;
+
+  // central mv
+  bestmv->as_mv.row <<= 3;
+  bestmv->as_mv.col <<= 3;
+  startmv = *bestmv;
+
+  // calculate central point error
+  bestmse = vfp->vf(y, y_stride, z, src_stride, sse1);
+  *distortion = bestmse;
+  bestmse += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit,
+                         xd->allow_high_precision_mv);
+
+  // go left then right and check error
+  this_mv.as_mv.row = startmv.as_mv.row;
+  this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
+  thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, src_stride, &sse);
+  left = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (left < bestmse) {
+    *bestmv = this_mv;
+    bestmse = left;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.col += 8;
+  thismse = vfp->svf_halfpix_h(y, y_stride, z, src_stride, &sse);
+  right = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost,
+                                error_per_bit, xd->allow_high_precision_mv);
+
+  if (right < bestmse) {
+    *bestmv = this_mv;
+    bestmse = right;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // go up then down and check error
+  this_mv.as_mv.col = startmv.as_mv.col;
+  this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
+  thismse = vfp->svf_halfpix_v(y - y_stride, y_stride, z, src_stride, &sse);
+  up = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                             xd->allow_high_precision_mv);
+
+  if (up < bestmse) {
+    *bestmv = this_mv;
+    bestmse = up;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  this_mv.as_mv.row += 8;
+  thismse = vfp->svf_halfpix_v(y, y_stride, z, src_stride, &sse);
+  down = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (down < bestmse) {
+    *bestmv = this_mv;
+    bestmse = down;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  // now check 1 more diagonal -
+  whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
+  this_mv = startmv;
+
+  switch (whichdir) {
+    case 0:
+      this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+      this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+      thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride,
+                                    z, src_stride, &sse);
+      break;
+    case 1:
+      this_mv.as_mv.col += 4;
+      this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+      thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride,
+                                    z, src_stride, &sse);
+      break;
+    case 2:
+      this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+      this_mv.as_mv.row += 4;
+      thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, src_stride, &sse);
+      break;
+    case 3:
+    default:
+      this_mv.as_mv.col += 4;
+      this_mv.as_mv.row += 4;
+      thismse = vfp->svf_halfpix_hv(y, y_stride, z, src_stride, &sse);
+      break;
+  }
+
+  diag = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit,
+                               xd->allow_high_precision_mv);
+
+  if (diag < bestmse) {
+    *bestmv = this_mv;
+    bestmse = diag;
+    *distortion = thismse;
+    *sse1 = sse;
+  }
+
+  return bestmse;
+}
+
+#define CHECK_BOUNDS(range) \
+  {\
+    all_in = 1;\
+    all_in &= ((br-range) >= x->mv_row_min);\
+    all_in &= ((br+range) <= x->mv_row_max);\
+    all_in &= ((bc-range) >= x->mv_col_min);\
+    all_in &= ((bc+range) <= x->mv_col_max);\
+  }
+
+#define CHECK_POINT \
+  {\
+    if (this_mv.as_mv.col < x->mv_col_min) continue;\
+    if (this_mv.as_mv.col > x->mv_col_max) continue;\
+    if (this_mv.as_mv.row < x->mv_row_min) continue;\
+    if (this_mv.as_mv.row > x->mv_row_max) continue;\
+  }
+
+#define CHECK_BETTER \
+  {\
+    if (thissad < bestsad)\
+    {\
+      thissad += mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost, mvsadcost, \
+                                sad_per_bit);\
+      if (thissad < bestsad)\
+      {\
+        bestsad = thissad;\
+        best_site = i;\
+      }\
+    }\
+  }
+
+static const MV next_chkpts[6][3] = {
+  {{ -2, 0}, { -1, -2}, {1, -2}},
+  {{ -1, -2}, {1, -2}, {2, 0}},
+  {{1, -2}, {2, 0}, {1, 2}},
+  {{2, 0}, {1, 2}, { -1, 2}},
+  {{1, 2}, { -1, 2}, { -2, 0}},
+  {{ -1, 2}, { -2, 0}, { -1, -2}}
+};
+
+int vp9_hex_search
+(
+  MACROBLOCK *x,
+  int_mv *ref_mv,
+  int_mv *best_mv,
+  int search_param,
+  int sad_per_bit,
+  const vp9_variance_fn_ptr_t *vfp,
+  int *mvjsadcost, int *mvsadcost[2],
+  int *mvjcost, int *mvcost[2],
+  int_mv *center_mv
+) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  MV hex[6] = { { -1, -2}, {1, -2}, {2, 0}, {1, 2}, { -1, 2}, { -2, 0} };
+  MV neighbors[4] = {{0, -1}, { -1, 0}, {1, 0}, {0, 1}};
+  int i, j;
+
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  int br, bc;
+  int_mv this_mv;
+  unsigned int bestsad = 0x7fffffff;
+  unsigned int thissad;
+  uint8_t *base_offset;
+  uint8_t *this_offset;
+  int k = -1;
+  int all_in;
+  int best_site = -1;
+
+  int_mv fcenter_mv;
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  // adjust ref_mv to make sure it is within MV range
+  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+  br = ref_mv->as_mv.row;
+  bc = ref_mv->as_mv.col;
+
+  // Work out the start point for the search
+  base_offset = (uint8_t *)(xd->plane[0].pre[0].buf);
+  this_offset = base_offset + (br * (xd->plane[0].pre[0].stride)) + bc;
+  this_mv.as_mv.row = br;
+  this_mv.as_mv.col = bc;
+  bestsad = vfp->sdf(what, what_stride, this_offset,
+                     in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // hex search
+  // j=0
+  CHECK_BOUNDS(2)
+
+  if (all_in) {
+    for (i = 0; i < 6; i++) {
+      this_mv.as_mv.row = br + hex[i].row;
+      this_mv.as_mv.col = bc + hex[i].col;
+      this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) + this_mv.as_mv.col;
+      thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+      CHECK_BETTER
+    }
+  } else {
+    for (i = 0; i < 6; i++) {
+      this_mv.as_mv.row = br + hex[i].row;
+      this_mv.as_mv.col = bc + hex[i].col;
+      CHECK_POINT
+      this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) + this_mv.as_mv.col;
+      thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+      CHECK_BETTER
+    }
+  }
+
+  if (best_site == -1)
+    goto cal_neighbors;
+  else {
+    br += hex[best_site].row;
+    bc += hex[best_site].col;
+    k = best_site;
+  }
+
+  for (j = 1; j < 127; j++) {
+    best_site = -1;
+    CHECK_BOUNDS(2)
+
+    if (all_in) {
+      for (i = 0; i < 3; i++) {
+        this_mv.as_mv.row = br + next_chkpts[k][i].row;
+        this_mv.as_mv.col = bc + next_chkpts[k][i].col;
+        this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
+        thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+        CHECK_BETTER
+      }
+    } else {
+      for (i = 0; i < 3; i++) {
+        this_mv.as_mv.row = br + next_chkpts[k][i].row;
+        this_mv.as_mv.col = bc + next_chkpts[k][i].col;
+        CHECK_POINT
+        this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
+        thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+        CHECK_BETTER
+      }
+    }
+
+    if (best_site == -1)
+      break;
+    else {
+      br += next_chkpts[k][best_site].row;
+      bc += next_chkpts[k][best_site].col;
+      k += 5 + best_site;
+      if (k >= 12) k -= 12;
+      else if (k >= 6) k -= 6;
+    }
+  }
+
+  // check 4 1-away neighbors
+cal_neighbors:
+  for (j = 0; j < 32; j++) {
+    best_site = -1;
+    CHECK_BOUNDS(1)
+
+    if (all_in) {
+      for (i = 0; i < 4; i++) {
+        this_mv.as_mv.row = br + neighbors[i].row;
+        this_mv.as_mv.col = bc + neighbors[i].col;
+        this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
+        thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+        CHECK_BETTER
+      }
+    } else {
+      for (i = 0; i < 4; i++) {
+        this_mv.as_mv.row = br + neighbors[i].row;
+        this_mv.as_mv.col = bc + neighbors[i].col;
+        CHECK_POINT
+        this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
+        thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
+        CHECK_BETTER
+      }
+    }
+
+    if (best_site == -1)
+      break;
+    else {
+      br += neighbors[best_site].row;
+      bc += neighbors[best_site].col;
+    }
+  }
+
+  best_mv->as_mv.row = br;
+  best_mv->as_mv.col = bc;
+
+  return bestsad;
+}
+#undef CHECK_BOUNDS
+#undef CHECK_POINT
+#undef CHECK_BETTER
+
+int vp9_diamond_search_sad_c(MACROBLOCK *x,
+                             int_mv *ref_mv, int_mv *best_mv,
+                             int search_param, int sad_per_bit, int *num00,
+                             vp9_variance_fn_ptr_t *fn_ptr, int *mvjcost,
+                             int *mvcost[2], int_mv *center_mv) {
+  int i, j, step;
+
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  uint8_t *in_what;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  uint8_t *best_address;
+
+  int tot_steps;
+  int_mv this_mv;
+
+  int bestsad = INT_MAX;
+  int best_site = 0;
+  int last_site = 0;
+
+  int ref_row, ref_col;
+  int this_row_offset, this_col_offset;
+  search_site *ss;
+
+  uint8_t *check_here;
+  int thissad;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+  ref_row = ref_mv->as_mv.row;
+  ref_col = ref_mv->as_mv.col;
+  *num00 = 0;
+  best_mv->as_mv.row = ref_row;
+  best_mv->as_mv.col = ref_col;
+
+  // Work out the start point for the search
+  in_what = (uint8_t *)(xd->plane[0].pre[0].buf +
+                        (ref_row * (xd->plane[0].pre[0].stride)) + ref_col);
+  best_address = in_what;
+
+  // Check the starting position
+  bestsad = fn_ptr->sdf(what, what_stride, in_what,
+                        in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(best_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // search_param determines the length of the initial step and hence the number of iterations
+  // 0 = initial step (MAX_FIRST_STEP) pel : 1 = (MAX_FIRST_STEP/2) pel, 2 = (MAX_FIRST_STEP/4) pel... etc.
+  ss = &x->ss[search_param * x->searches_per_step];
+  tot_steps = (x->ss_count / x->searches_per_step) - search_param;
+
+  i = 1;
+
+  for (step = 0; step < tot_steps; step++) {
+    for (j = 0; j < x->searches_per_step; j++) {
+      // Trap illegal vectors
+      this_row_offset = best_mv->as_mv.row + ss[i].mv.row;
+      this_col_offset = best_mv->as_mv.col + ss[i].mv.col;
+
+      if ((this_col_offset > x->mv_col_min) && (this_col_offset < x->mv_col_max) &&
+          (this_row_offset > x->mv_row_min) && (this_row_offset < x->mv_row_max))
+
+      {
+        check_here = ss[i].offset + best_address;
+        thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.row = this_row_offset;
+          this_mv.as_mv.col = this_col_offset;
+          thissad += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                    mvjsadcost, mvsadcost, sad_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_site = i;
+          }
+        }
+      }
+
+      i++;
+    }
+
+    if (best_site != last_site) {
+      best_mv->as_mv.row += ss[best_site].mv.row;
+      best_mv->as_mv.col += ss[best_site].mv.col;
+      best_address += ss[best_site].offset;
+      last_site = best_site;
+    } else if (best_address == in_what)
+      (*num00)++;
+  }
+
+  this_mv.as_mv.row = best_mv->as_mv.row << 3;
+  this_mv.as_mv.col = best_mv->as_mv.col << 3;
+
+  if (bestsad == INT_MAX)
+    return INT_MAX;
+
+  return
+      fn_ptr->vf(what, what_stride, best_address, in_what_stride,
+                 (unsigned int *)(&thissad)) +
+      mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                  xd->allow_high_precision_mv);
+}
+
+int vp9_diamond_search_sadx4(MACROBLOCK *x,
+                             int_mv *ref_mv, int_mv *best_mv, int search_param,
+                             int sad_per_bit, int *num00,
+                             vp9_variance_fn_ptr_t *fn_ptr,
+                             int *mvjcost, int *mvcost[2], int_mv *center_mv) {
+  int i, j, step;
+
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  uint8_t *in_what;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  uint8_t *best_address;
+
+  int tot_steps;
+  int_mv this_mv;
+
+  unsigned int bestsad = INT_MAX;
+  int best_site = 0;
+  int last_site = 0;
+
+  int ref_row;
+  int ref_col;
+  int this_row_offset;
+  int this_col_offset;
+  search_site *ss;
+
+  uint8_t *check_here;
+  unsigned int thissad;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+  ref_row = ref_mv->as_mv.row;
+  ref_col = ref_mv->as_mv.col;
+  *num00 = 0;
+  best_mv->as_mv.row = ref_row;
+  best_mv->as_mv.col = ref_col;
+
+  // Work out the start point for the search
+  in_what = (uint8_t *)(xd->plane[0].pre[0].buf +
+                        (ref_row * (xd->plane[0].pre[0].stride)) + ref_col);
+  best_address = in_what;
+
+  // Check the starting position
+  bestsad = fn_ptr->sdf(what, what_stride,
+                        in_what, in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(best_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // search_param determines the length of the initial step and hence the number of iterations
+  // 0 = initial step (MAX_FIRST_STEP) pel : 1 = (MAX_FIRST_STEP/2) pel, 2 = (MAX_FIRST_STEP/4) pel... etc.
+  ss = &x->ss[search_param * x->searches_per_step];
+  tot_steps = (x->ss_count / x->searches_per_step) - search_param;
+
+  i = 1;
+
+  for (step = 0; step < tot_steps; step++) {
+    int all_in = 1, t;
+
+    // To know if all neighbor points are within the bounds, 4 bounds checking are enough instead of
+    // checking 4 bounds for each points.
+    all_in &= ((best_mv->as_mv.row + ss[i].mv.row) > x->mv_row_min);
+    all_in &= ((best_mv->as_mv.row + ss[i + 1].mv.row) < x->mv_row_max);
+    all_in &= ((best_mv->as_mv.col + ss[i + 2].mv.col) > x->mv_col_min);
+    all_in &= ((best_mv->as_mv.col + ss[i + 3].mv.col) < x->mv_col_max);
+
+    if (all_in) {
+      unsigned int sad_array[4];
+
+      for (j = 0; j < x->searches_per_step; j += 4) {
+        unsigned char const *block_offset[4];
+
+        for (t = 0; t < 4; t++)
+          block_offset[t] = ss[i + t].offset + best_address;
+
+        fn_ptr->sdx4df(what, what_stride, block_offset, in_what_stride,
+                       sad_array);
+
+        for (t = 0; t < 4; t++, i++) {
+          if (sad_array[t] < bestsad) {
+            this_mv.as_mv.row = best_mv->as_mv.row + ss[i].mv.row;
+            this_mv.as_mv.col = best_mv->as_mv.col + ss[i].mv.col;
+            sad_array[t] += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                           mvjsadcost, mvsadcost, sad_per_bit);
+
+            if (sad_array[t] < bestsad) {
+              bestsad = sad_array[t];
+              best_site = i;
+            }
+          }
+        }
+      }
+    } else {
+      for (j = 0; j < x->searches_per_step; j++) {
+        // Trap illegal vectors
+        this_row_offset = best_mv->as_mv.row + ss[i].mv.row;
+        this_col_offset = best_mv->as_mv.col + ss[i].mv.col;
+
+        if ((this_col_offset > x->mv_col_min) && (this_col_offset < x->mv_col_max) &&
+            (this_row_offset > x->mv_row_min) && (this_row_offset < x->mv_row_max)) {
+          check_here = ss[i].offset + best_address;
+          thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+          if (thissad < bestsad) {
+            this_mv.as_mv.row = this_row_offset;
+            this_mv.as_mv.col = this_col_offset;
+            thissad += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                      mvjsadcost, mvsadcost, sad_per_bit);
+
+            if (thissad < bestsad) {
+              bestsad = thissad;
+              best_site = i;
+            }
+          }
+        }
+        i++;
+      }
+    }
+
+    if (best_site != last_site) {
+      best_mv->as_mv.row += ss[best_site].mv.row;
+      best_mv->as_mv.col += ss[best_site].mv.col;
+      best_address += ss[best_site].offset;
+      last_site = best_site;
+    } else if (best_address == in_what)
+      (*num00)++;
+  }
+
+  this_mv.as_mv.row = best_mv->as_mv.row << 3;
+  this_mv.as_mv.col = best_mv->as_mv.col << 3;
+
+  if (bestsad == INT_MAX)
+    return INT_MAX;
+
+  return
+      fn_ptr->vf(what, what_stride, best_address, in_what_stride,
+                 (unsigned int *)(&thissad)) +
+      mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                  xd->allow_high_precision_mv);
+}
+
+/* do_refine: If last step (1-away) of n-step search doesn't pick the center
+              point as the best match, we will do a final 1-away diamond
+              refining search  */
+int vp9_full_pixel_diamond(VP9_COMP *cpi, MACROBLOCK *x,
+                           int_mv *mvp_full, int step_param,
+                           int sadpb, int further_steps,
+                           int do_refine, vp9_variance_fn_ptr_t *fn_ptr,
+                           int_mv *ref_mv, int_mv *dst_mv) {
+  int_mv temp_mv;
+  int thissme, n, num00;
+  int bestsme = cpi->diamond_search_sad(x, mvp_full, &temp_mv,
+                                        step_param, sadpb, &num00,
+                                        fn_ptr, x->nmvjointcost,
+                                        x->mvcost, ref_mv);
+  dst_mv->as_int = temp_mv.as_int;
+
+  n = num00;
+  num00 = 0;
+
+  /* If there won't be more n-step search, check to see if refining search is needed. */
+  if (n > further_steps)
+    do_refine = 0;
+
+  while (n < further_steps) {
+    n++;
+
+    if (num00)
+      num00--;
+    else {
+      thissme = cpi->diamond_search_sad(x, mvp_full, &temp_mv,
+                                        step_param + n, sadpb, &num00,
+                                        fn_ptr, x->nmvjointcost, x->mvcost,
+                                        ref_mv);
+
+      /* check to see if refining search is needed. */
+      if (num00 > (further_steps - n))
+        do_refine = 0;
+
+      if (thissme < bestsme) {
+        bestsme = thissme;
+        dst_mv->as_int = temp_mv.as_int;
+      }
+    }
+  }
+
+  /* final 1-away diamond refining search */
+  if (do_refine == 1) {
+    int search_range = 8;
+    int_mv best_mv;
+    best_mv.as_int = dst_mv->as_int;
+    thissme = cpi->refining_search_sad(x, &best_mv, sadpb, search_range,
+                                       fn_ptr, x->nmvjointcost, x->mvcost,
+                                       ref_mv);
+
+    if (thissme < bestsme) {
+      bestsme = thissme;
+      dst_mv->as_int = best_mv.as_int;
+    }
+  }
+  return bestsme;
+}
+
+int vp9_full_search_sad_c(MACROBLOCK *x, int_mv *ref_mv,
+                          int sad_per_bit, int distance,
+                          vp9_variance_fn_ptr_t *fn_ptr, int *mvjcost,
+                          int *mvcost[2],
+                          int_mv *center_mv, int n) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  uint8_t *in_what;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  int mv_stride = xd->plane[0].pre[0].stride;
+  uint8_t *bestaddress;
+  int_mv *best_mv = &x->e_mbd.mode_info_context->bmi[n].as_mv[0];
+  int_mv this_mv;
+  int bestsad = INT_MAX;
+  int r, c;
+
+  uint8_t *check_here;
+  int thissad;
+
+  int ref_row = ref_mv->as_mv.row;
+  int ref_col = ref_mv->as_mv.col;
+
+  int row_min = ref_row - distance;
+  int row_max = ref_row + distance;
+  int col_min = ref_col - distance;
+  int col_max = ref_col + distance;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  // Work out the mid point for the search
+  in_what = xd->plane[0].pre[0].buf;
+  bestaddress = in_what + (ref_row * xd->plane[0].pre[0].stride) + ref_col;
+
+  best_mv->as_mv.row = ref_row;
+  best_mv->as_mv.col = ref_col;
+
+  // Baseline value at the centre
+  bestsad = fn_ptr->sdf(what, what_stride, bestaddress,
+                        in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(best_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // Apply further limits to prevent us looking using vectors that stretch beyiond the UMV border
+  if (col_min < x->mv_col_min)
+    col_min = x->mv_col_min;
+
+  if (col_max > x->mv_col_max)
+    col_max = x->mv_col_max;
+
+  if (row_min < x->mv_row_min)
+    row_min = x->mv_row_min;
+
+  if (row_max > x->mv_row_max)
+    row_max = x->mv_row_max;
+
+  for (r = row_min; r < row_max; r++) {
+    this_mv.as_mv.row = r;
+    check_here = r * mv_stride + in_what + col_min;
+
+    for (c = col_min; c < col_max; c++) {
+      thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+      this_mv.as_mv.col = c;
+      thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                 mvjsadcost, mvsadcost, sad_per_bit);
+
+      if (thissad < bestsad) {
+        bestsad = thissad;
+        best_mv->as_mv.row = r;
+        best_mv->as_mv.col = c;
+        bestaddress = check_here;
+      }
+
+      check_here++;
+    }
+  }
+
+  this_mv.as_mv.row = best_mv->as_mv.row << 3;
+  this_mv.as_mv.col = best_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX)
+    return
+        fn_ptr->vf(what, what_stride, bestaddress, in_what_stride,
+                   (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+  else
+    return INT_MAX;
+}
+
+int vp9_full_search_sadx3(MACROBLOCK *x, int_mv *ref_mv,
+                          int sad_per_bit, int distance,
+                          vp9_variance_fn_ptr_t *fn_ptr, int *mvjcost,
+                          int *mvcost[2], int_mv *center_mv, int n) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  uint8_t *in_what;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  int mv_stride = xd->plane[0].pre[0].stride;
+  uint8_t *bestaddress;
+  int_mv *best_mv = &x->e_mbd.mode_info_context->bmi[n].as_mv[0];
+  int_mv this_mv;
+  unsigned int bestsad = INT_MAX;
+  int r, c;
+
+  uint8_t *check_here;
+  unsigned int thissad;
+
+  int ref_row = ref_mv->as_mv.row;
+  int ref_col = ref_mv->as_mv.col;
+
+  int row_min = ref_row - distance;
+  int row_max = ref_row + distance;
+  int col_min = ref_col - distance;
+  int col_max = ref_col + distance;
+
+  unsigned int sad_array[3];
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  // Work out the mid point for the search
+  in_what = xd->plane[0].pre[0].buf;
+  bestaddress = in_what + (ref_row * xd->plane[0].pre[0].stride) + ref_col;
+
+  best_mv->as_mv.row = ref_row;
+  best_mv->as_mv.col = ref_col;
+
+  // Baseline value at the centre
+  bestsad = fn_ptr->sdf(what, what_stride,
+                        bestaddress, in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(best_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // Apply further limits to prevent us looking using vectors that stretch beyiond the UMV border
+  if (col_min < x->mv_col_min)
+    col_min = x->mv_col_min;
+
+  if (col_max > x->mv_col_max)
+    col_max = x->mv_col_max;
+
+  if (row_min < x->mv_row_min)
+    row_min = x->mv_row_min;
+
+  if (row_max > x->mv_row_max)
+    row_max = x->mv_row_max;
+
+  for (r = row_min; r < row_max; r++) {
+    this_mv.as_mv.row = r;
+    check_here = r * mv_stride + in_what + col_min;
+    c = col_min;
+
+    while ((c + 2) < col_max) {
+      int i;
+
+      fn_ptr->sdx3f(what, what_stride, check_here, in_what_stride, sad_array);
+
+      for (i = 0; i < 3; i++) {
+        thissad = sad_array[i];
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.col = c;
+          thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                     mvjsadcost, mvsadcost, sad_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_mv->as_mv.row = r;
+            best_mv->as_mv.col = c;
+            bestaddress = check_here;
+          }
+        }
+
+        check_here++;
+        c++;
+      }
+    }
+
+    while (c < col_max) {
+      thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+      if (thissad < bestsad) {
+        this_mv.as_mv.col = c;
+        thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                   mvjsadcost, mvsadcost, sad_per_bit);
+
+        if (thissad < bestsad) {
+          bestsad = thissad;
+          best_mv->as_mv.row = r;
+          best_mv->as_mv.col = c;
+          bestaddress = check_here;
+        }
+      }
+
+      check_here++;
+      c++;
+    }
+
+  }
+
+  this_mv.as_mv.row = best_mv->as_mv.row << 3;
+  this_mv.as_mv.col = best_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX)
+    return
+        fn_ptr->vf(what, what_stride, bestaddress, in_what_stride,
+                   (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+  else
+    return INT_MAX;
+}
+
+int vp9_full_search_sadx8(MACROBLOCK *x, int_mv *ref_mv,
+                          int sad_per_bit, int distance,
+                          vp9_variance_fn_ptr_t *fn_ptr,
+                          int *mvjcost, int *mvcost[2],
+                          int_mv *center_mv, int n) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  uint8_t *what = x->plane[0].src.buf;
+  int what_stride = x->plane[0].src.stride;
+  uint8_t *in_what;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  int mv_stride = xd->plane[0].pre[0].stride;
+  uint8_t *bestaddress;
+  int_mv *best_mv = &x->e_mbd.mode_info_context->bmi[n].as_mv[0];
+  int_mv this_mv;
+  unsigned int bestsad = INT_MAX;
+  int r, c;
+
+  uint8_t *check_here;
+  unsigned int thissad;
+
+  int ref_row = ref_mv->as_mv.row;
+  int ref_col = ref_mv->as_mv.col;
+
+  int row_min = ref_row - distance;
+  int row_max = ref_row + distance;
+  int col_min = ref_col - distance;
+  int col_max = ref_col + distance;
+
+  DECLARE_ALIGNED_ARRAY(16, uint32_t, sad_array8, 8);
+  unsigned int sad_array[3];
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  // Work out the mid point for the search
+  in_what = xd->plane[0].pre[0].buf;
+  bestaddress = in_what + (ref_row * xd->plane[0].pre[0].stride) + ref_col;
+
+  best_mv->as_mv.row = ref_row;
+  best_mv->as_mv.col = ref_col;
+
+  // Baseline value at the centre
+  bestsad = fn_ptr->sdf(what, what_stride,
+                        bestaddress, in_what_stride, 0x7fffffff)
+            + mvsad_err_cost(best_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+                             sad_per_bit);
+
+  // Apply further limits to prevent us looking using vectors that stretch beyiond the UMV border
+  if (col_min < x->mv_col_min)
+    col_min = x->mv_col_min;
+
+  if (col_max > x->mv_col_max)
+    col_max = x->mv_col_max;
+
+  if (row_min < x->mv_row_min)
+    row_min = x->mv_row_min;
+
+  if (row_max > x->mv_row_max)
+    row_max = x->mv_row_max;
+
+  for (r = row_min; r < row_max; r++) {
+    this_mv.as_mv.row = r;
+    check_here = r * mv_stride + in_what + col_min;
+    c = col_min;
+
+    while ((c + 7) < col_max) {
+      int i;
+
+      fn_ptr->sdx8f(what, what_stride, check_here, in_what_stride, sad_array8);
+
+      for (i = 0; i < 8; i++) {
+        thissad = (unsigned int)sad_array8[i];
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.col = c;
+          thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                     mvjsadcost, mvsadcost, sad_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_mv->as_mv.row = r;
+            best_mv->as_mv.col = c;
+            bestaddress = check_here;
+          }
+        }
+
+        check_here++;
+        c++;
+      }
+    }
+
+    while ((c + 2) < col_max) {
+      int i;
+
+      fn_ptr->sdx3f(what, what_stride, check_here, in_what_stride, sad_array);
+
+      for (i = 0; i < 3; i++) {
+        thissad = sad_array[i];
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.col = c;
+          thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                     mvjsadcost, mvsadcost, sad_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_mv->as_mv.row = r;
+            best_mv->as_mv.col = c;
+            bestaddress = check_here;
+          }
+        }
+
+        check_here++;
+        c++;
+      }
+    }
+
+    while (c < col_max) {
+      thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+      if (thissad < bestsad) {
+        this_mv.as_mv.col = c;
+        thissad  += mvsad_err_cost(&this_mv, &fcenter_mv,
+                                   mvjsadcost, mvsadcost, sad_per_bit);
+
+        if (thissad < bestsad) {
+          bestsad = thissad;
+          best_mv->as_mv.row = r;
+          best_mv->as_mv.col = c;
+          bestaddress = check_here;
+        }
+      }
+
+      check_here++;
+      c++;
+    }
+  }
+
+  this_mv.as_mv.row = best_mv->as_mv.row << 3;
+  this_mv.as_mv.col = best_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX)
+    return
+        fn_ptr->vf(what, what_stride, bestaddress, in_what_stride,
+                   (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+  else
+    return INT_MAX;
+}
+int vp9_refining_search_sad_c(MACROBLOCK *x,
+                              int_mv *ref_mv, int error_per_bit,
+                              int search_range, vp9_variance_fn_ptr_t *fn_ptr,
+                              int *mvjcost, int *mvcost[2], int_mv *center_mv) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
+  int i, j;
+  int this_row_offset, this_col_offset;
+
+  int what_stride = x->plane[0].src.stride;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  uint8_t *what = x->plane[0].src.buf;
+  uint8_t *best_address = xd->plane[0].pre[0].buf +
+                          (ref_mv->as_mv.row * xd->plane[0].pre[0].stride) +
+                          ref_mv->as_mv.col;
+  uint8_t *check_here;
+  unsigned int thissad;
+  int_mv this_mv;
+  unsigned int bestsad = INT_MAX;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  bestsad = fn_ptr->sdf(what, what_stride, best_address, in_what_stride, 0x7fffffff) +
+      mvsad_err_cost(ref_mv, &fcenter_mv, mvjsadcost, mvsadcost, error_per_bit);
+
+  for (i = 0; i < search_range; i++) {
+    int best_site = -1;
+
+    for (j = 0; j < 4; j++) {
+      this_row_offset = ref_mv->as_mv.row + neighbors[j].row;
+      this_col_offset = ref_mv->as_mv.col + neighbors[j].col;
+
+      if ((this_col_offset > x->mv_col_min) && (this_col_offset < x->mv_col_max) &&
+          (this_row_offset > x->mv_row_min) && (this_row_offset < x->mv_row_max)) {
+        check_here = (neighbors[j].row) * in_what_stride + neighbors[j].col + best_address;
+        thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.row = this_row_offset;
+          this_mv.as_mv.col = this_col_offset;
+          thissad += mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost,
+                                    mvsadcost, error_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_site = j;
+          }
+        }
+      }
+    }
+
+    if (best_site == -1)
+      break;
+    else {
+      ref_mv->as_mv.row += neighbors[best_site].row;
+      ref_mv->as_mv.col += neighbors[best_site].col;
+      best_address += (neighbors[best_site].row) * in_what_stride + neighbors[best_site].col;
+    }
+  }
+
+  this_mv.as_mv.row = ref_mv->as_mv.row << 3;
+  this_mv.as_mv.col = ref_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX)
+    return
+        fn_ptr->vf(what, what_stride, best_address, in_what_stride,
+                   (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+  else
+    return INT_MAX;
+}
+
+int vp9_refining_search_sadx4(MACROBLOCK *x,
+                              int_mv *ref_mv, int error_per_bit,
+                              int search_range, vp9_variance_fn_ptr_t *fn_ptr,
+                              int *mvjcost, int *mvcost[2], int_mv *center_mv) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
+  int i, j;
+  int this_row_offset, this_col_offset;
+
+  int what_stride = x->plane[0].src.stride;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  uint8_t *what = x->plane[0].src.buf;
+  uint8_t *best_address = xd->plane[0].pre[0].buf +
+                          (ref_mv->as_mv.row * xd->plane[0].pre[0].stride) +
+                          ref_mv->as_mv.col;
+  uint8_t *check_here;
+  unsigned int thissad;
+  int_mv this_mv;
+  unsigned int bestsad = INT_MAX;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  bestsad = fn_ptr->sdf(what, what_stride, best_address, in_what_stride, 0x7fffffff) +
+      mvsad_err_cost(ref_mv, &fcenter_mv, mvjsadcost, mvsadcost, error_per_bit);
+
+  for (i = 0; i < search_range; i++) {
+    int best_site = -1;
+    int all_in = ((ref_mv->as_mv.row - 1) > x->mv_row_min) &
+                 ((ref_mv->as_mv.row + 1) < x->mv_row_max) &
+                 ((ref_mv->as_mv.col - 1) > x->mv_col_min) &
+                 ((ref_mv->as_mv.col + 1) < x->mv_col_max);
+
+    if (all_in) {
+      unsigned int sad_array[4];
+      unsigned char const *block_offset[4];
+      block_offset[0] = best_address - in_what_stride;
+      block_offset[1] = best_address - 1;
+      block_offset[2] = best_address + 1;
+      block_offset[3] = best_address + in_what_stride;
+
+      fn_ptr->sdx4df(what, what_stride, block_offset, in_what_stride, sad_array);
+
+      for (j = 0; j < 4; j++) {
+        if (sad_array[j] < bestsad) {
+          this_mv.as_mv.row = ref_mv->as_mv.row + neighbors[j].row;
+          this_mv.as_mv.col = ref_mv->as_mv.col + neighbors[j].col;
+          sad_array[j] += mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost,
+                                         mvsadcost, error_per_bit);
+
+          if (sad_array[j] < bestsad) {
+            bestsad = sad_array[j];
+            best_site = j;
+          }
+        }
+      }
+    } else {
+      for (j = 0; j < 4; j++) {
+        this_row_offset = ref_mv->as_mv.row + neighbors[j].row;
+        this_col_offset = ref_mv->as_mv.col + neighbors[j].col;
+
+        if ((this_col_offset > x->mv_col_min) && (this_col_offset < x->mv_col_max) &&
+            (this_row_offset > x->mv_row_min) && (this_row_offset < x->mv_row_max)) {
+          check_here = (neighbors[j].row) * in_what_stride + neighbors[j].col + best_address;
+          thissad = fn_ptr->sdf(what, what_stride, check_here, in_what_stride, bestsad);
+
+          if (thissad < bestsad) {
+            this_mv.as_mv.row = this_row_offset;
+            this_mv.as_mv.col = this_col_offset;
+            thissad += mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost,
+                                      mvsadcost, error_per_bit);
+
+            if (thissad < bestsad) {
+              bestsad = thissad;
+              best_site = j;
+            }
+          }
+        }
+      }
+    }
+
+    if (best_site == -1)
+      break;
+    else {
+      ref_mv->as_mv.row += neighbors[best_site].row;
+      ref_mv->as_mv.col += neighbors[best_site].col;
+      best_address += (neighbors[best_site].row) * in_what_stride + neighbors[best_site].col;
+    }
+  }
+
+  this_mv.as_mv.row = ref_mv->as_mv.row << 3;
+  this_mv.as_mv.col = ref_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX)
+    return
+        fn_ptr->vf(what, what_stride, best_address, in_what_stride,
+                   (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+  else
+    return INT_MAX;
+}
+
+/* This function is called when we do joint motion search in comp_inter_inter
+ * mode.
+ */
+int vp9_refining_search_8p_c(MACROBLOCK *x,
+                             int_mv *ref_mv, int error_per_bit,
+                             int search_range, vp9_variance_fn_ptr_t *fn_ptr,
+                             int *mvjcost, int *mvcost[2], int_mv *center_mv,
+                             const uint8_t *second_pred, int w, int h) {
+  const MACROBLOCKD* const xd = &x->e_mbd;
+  MV neighbors[8] = {{-1, 0}, {0, -1}, {0, 1}, {1, 0},
+      {-1, -1}, {1, -1}, {-1, 1}, {1, 1}};
+  int i, j;
+  int this_row_offset, this_col_offset;
+
+  int what_stride = x->plane[0].src.stride;
+  int in_what_stride = xd->plane[0].pre[0].stride;
+  uint8_t *what = x->plane[0].src.buf;
+  uint8_t *best_address = xd->plane[0].pre[0].buf +
+                          (ref_mv->as_mv.row * xd->plane[0].pre[0].stride) +
+                          ref_mv->as_mv.col;
+  uint8_t *check_here;
+  unsigned int thissad;
+  int_mv this_mv;
+  unsigned int bestsad = INT_MAX;
+  int_mv fcenter_mv;
+
+  int *mvjsadcost = x->nmvjointsadcost;
+  int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+  /* Compound pred buffer */
+  uint8_t *comp_pred = vpx_memalign(16, w * h * sizeof(uint8_t));
+
+  fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
+  fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+
+  /* Get compound pred by averaging two pred blocks. */
+  comp_avg_pred(comp_pred, second_pred, w, h, best_address, in_what_stride);
+
+  bestsad = fn_ptr->sdf(what, what_stride, comp_pred, w, 0x7fffffff) +
+      mvsad_err_cost(ref_mv, &fcenter_mv, mvjsadcost, mvsadcost, error_per_bit);
+
+  for (i = 0; i < search_range; i++) {
+    int best_site = -1;
+
+    for (j = 0; j < 8; j++) {
+      this_row_offset = ref_mv->as_mv.row + neighbors[j].row;
+      this_col_offset = ref_mv->as_mv.col + neighbors[j].col;
+
+      if ((this_col_offset > x->mv_col_min) &&
+          (this_col_offset < x->mv_col_max) &&
+          (this_row_offset > x->mv_row_min) &&
+          (this_row_offset < x->mv_row_max)) {
+        check_here = (neighbors[j].row) * in_what_stride + neighbors[j].col +
+            best_address;
+
+        /* Get compound block and use it to calculate SAD. */
+        comp_avg_pred(comp_pred, second_pred, w, h, check_here,
+                      in_what_stride);
+        thissad = fn_ptr->sdf(what, what_stride, comp_pred, w, bestsad);
+
+        if (thissad < bestsad) {
+          this_mv.as_mv.row = this_row_offset;
+          this_mv.as_mv.col = this_col_offset;
+          thissad += mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost,
+                                    mvsadcost, error_per_bit);
+
+          if (thissad < bestsad) {
+            bestsad = thissad;
+            best_site = j;
+          }
+        }
+      }
+    }
+
+    if (best_site == -1) {
+      break;
+    } else {
+      ref_mv->as_mv.row += neighbors[best_site].row;
+      ref_mv->as_mv.col += neighbors[best_site].col;
+      best_address += (neighbors[best_site].row) * in_what_stride +
+          neighbors[best_site].col;
+    }
+  }
+
+  this_mv.as_mv.row = ref_mv->as_mv.row << 3;
+  this_mv.as_mv.col = ref_mv->as_mv.col << 3;
+
+  if (bestsad < INT_MAX) {
+    int besterr;
+    comp_avg_pred(comp_pred, second_pred, w, h, best_address, in_what_stride);
+    besterr = fn_ptr->vf(what, what_stride, comp_pred, w,
+        (unsigned int *)(&thissad)) +
+        mv_err_cost(&this_mv, center_mv, mvjcost, mvcost, x->errorperbit,
+                    xd->allow_high_precision_mv);
+    vpx_free(comp_pred);
+    return besterr;
+  } else {
+    vpx_free(comp_pred);
+    return INT_MAX;
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_mcomp.h b/libvpx/vp9/encoder/vp9_mcomp.h
new file mode 100644
index 0000000..28b2efd
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_mcomp.h
@@ -0,0 +1,93 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_MCOMP_H_
+#define VP9_ENCODER_VP9_MCOMP_H_
+
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_variance.h"
+
+// The maximum number of steps in a step search given the largest
+// allowed initial step
+#define MAX_MVSEARCH_STEPS 11
+// Max full pel mv specified in 1 pel units
+#define MAX_FULL_PEL_VAL ((1 << (MAX_MVSEARCH_STEPS)) - 1)
+// Maximum size of the first step in full pel units
+#define MAX_FIRST_STEP (1 << (MAX_MVSEARCH_STEPS-1))
+
+void vp9_clamp_mv_min_max(MACROBLOCK *x, int_mv *ref_mv);
+int vp9_init_search_range(int width, int height);
+
+int vp9_mv_bit_cost(int_mv *mv, int_mv *ref, int *mvjcost,
+                           int *mvcost[2], int weight, int ishp);
+void vp9_init_dsmotion_compensation(MACROBLOCK *x, int stride);
+void vp9_init3smotion_compensation(MACROBLOCK *x,  int stride);
+
+// Runs sequence of diamond searches in smaller steps for RD
+struct VP9_COMP;
+int vp9_full_pixel_diamond(struct VP9_COMP *cpi, MACROBLOCK *x,
+                           int_mv *mvp_full, int step_param,
+                           int sadpb, int further_steps, int do_refine,
+                           vp9_variance_fn_ptr_t *fn_ptr,
+                           int_mv *ref_mv, int_mv *dst_mv);
+
+int vp9_hex_search(MACROBLOCK *x,
+                   int_mv *ref_mv, int_mv *best_mv,
+                   int search_param, int error_per_bit,
+                   const vp9_variance_fn_ptr_t *vf,
+                   int *mvjsadcost, int *mvsadcost[2],
+                   int *mvjcost, int *mvcost[2],
+                   int_mv *center_mv);
+
+typedef int (fractional_mv_step_fp) (MACROBLOCK *x, int_mv
+  *bestmv, int_mv *ref_mv, int error_per_bit, const vp9_variance_fn_ptr_t *vfp,
+  int *mvjcost, int *mvcost[2], int *distortion, unsigned int *sse);
+extern fractional_mv_step_fp vp9_find_best_sub_pixel_step_iteratively;
+extern fractional_mv_step_fp vp9_find_best_sub_pixel_step;
+extern fractional_mv_step_fp vp9_find_best_half_pixel_step;
+
+typedef int (*vp9_full_search_fn_t)(MACROBLOCK *x,
+                                    int_mv *ref_mv, int sad_per_bit,
+                                    int distance, vp9_variance_fn_ptr_t *fn_ptr,
+                                    int *mvjcost, int *mvcost[2],
+                                    int_mv *center_mv, int n);
+
+typedef int (*vp9_refining_search_fn_t)(MACROBLOCK *x,
+                                        int_mv *ref_mv, int sad_per_bit,
+                                        int distance,
+                                        vp9_variance_fn_ptr_t *fn_ptr,
+                                        int *mvjcost, int *mvcost[2],
+                                        int_mv *center_mv);
+
+typedef int (*vp9_diamond_search_fn_t)(MACROBLOCK *x,
+                                       int_mv *ref_mv, int_mv *best_mv,
+                                       int search_param, int sad_per_bit,
+                                       int *num00,
+                                       vp9_variance_fn_ptr_t *fn_ptr,
+                                       int *mvjcost, int *mvcost[2],
+                                       int_mv *center_mv);
+
+int vp9_find_best_sub_pixel_comp(MACROBLOCK *x,
+                                 int_mv *bestmv, int_mv *ref_mv,
+                                 int error_per_bit,
+                                 const vp9_variance_fn_ptr_t *vfp,
+                                 int *mvjcost, int *mvcost[2],
+                                 int *distortion, unsigned int *sse1,
+                                 const uint8_t *second_pred,
+                                 int w, int h);
+
+int vp9_refining_search_8p_c(MACROBLOCK *x,
+                             int_mv *ref_mv, int error_per_bit,
+                             int search_range, vp9_variance_fn_ptr_t *fn_ptr,
+                             int *mvjcost, int *mvcost[2],
+                             int_mv *center_mv, const uint8_t *second_pred,
+                             int w, int h);
+#endif  // VP9_ENCODER_VP9_MCOMP_H_
diff --git a/libvpx/vp9/encoder/vp9_modecosts.c b/libvpx/vp9/encoder/vp9_modecosts.c
new file mode 100644
index 0000000..f2e4ce4
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_modecosts.c
@@ -0,0 +1,42 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_treewriter.h"
+#include "vp9/common/vp9_entropymode.h"
+
+
+void vp9_init_mode_costs(VP9_COMP *c) {
+  VP9_COMMON *x = &c->common;
+  const vp9_tree_p KT = vp9_intra_mode_tree;
+  int i, j;
+
+  for (i = 0; i < VP9_INTRA_MODES; i++) {
+    for (j = 0; j < VP9_INTRA_MODES; j++) {
+      vp9_cost_tokens((int *)c->mb.y_mode_costs[i][j],
+                      x->kf_y_mode_prob[i][j], KT);
+    }
+  }
+
+  // TODO(rbultje) separate tables for superblock costing?
+  vp9_cost_tokens(c->mb.mbmode_cost, x->fc.y_mode_prob[1],
+                  vp9_intra_mode_tree);
+  vp9_cost_tokens(c->mb.intra_uv_mode_cost[1],
+                  x->fc.uv_mode_prob[VP9_INTRA_MODES - 1], vp9_intra_mode_tree);
+  vp9_cost_tokens(c->mb.intra_uv_mode_cost[0],
+                  x->kf_uv_mode_prob[VP9_INTRA_MODES - 1], vp9_intra_mode_tree);
+
+  for (i = 0; i <= VP9_SWITCHABLE_FILTERS; ++i)
+    vp9_cost_tokens((int *)c->mb.switchable_interp_costs[i],
+                    x->fc.switchable_interp_prob[i],
+                    vp9_switchable_interp_tree);
+}
diff --git a/libvpx/vp9/encoder/vp9_modecosts.h b/libvpx/vp9/encoder/vp9_modecosts.h
new file mode 100644
index 0000000..f43033e
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_modecosts.h
@@ -0,0 +1,17 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_MODECOSTS_H_
+#define VP9_ENCODER_VP9_MODECOSTS_H_
+
+void vp9_init_mode_costs(VP9_COMP *x);
+
+#endif  // VP9_ENCODER_VP9_MODECOSTS_H_
diff --git a/libvpx/vp9/encoder/vp9_onyx_if.c b/libvpx/vp9/encoder/vp9_onyx_if.c
new file mode 100644
index 0000000..6a14df4
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_onyx_if.c
@@ -0,0 +1,3932 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_config.h"
+#include "vp9/common/vp9_filter.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/encoder/vp9_firstpass.h"
+#include "vp9/encoder/vp9_psnr.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/common/vp9_extend.h"
+#include "vp9/encoder/vp9_ratectrl.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_tile_common.h"
+#include "vp9/encoder/vp9_segmentation.h"
+#include "./vp9_rtcd.h"
+#include "./vpx_scale_rtcd.h"
+#if CONFIG_POSTPROC
+#include "vp9/common/vp9_postproc.h"
+#endif
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/vpx_timer.h"
+
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/encoder/vp9_mbgraph.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/encoder/vp9_bitstream.h"
+#include "vp9/encoder/vp9_picklpf.h"
+#include "vp9/common/vp9_mvref_common.h"
+#include "vp9/encoder/vp9_temporal_filter.h"
+
+#include <math.h>
+#include <stdio.h>
+#include <limits.h>
+
+extern void print_tree_update_probs();
+
+static void set_default_lf_deltas(VP9_COMP *cpi);
+
+#define DEFAULT_INTERP_FILTER SWITCHABLE
+
+#define SEARCH_BEST_FILTER 0            /* to search exhaustively for
+                                           best filter */
+#define RESET_FOREACH_FILTER 0          /* whether to reset the encoder state
+                                           before trying each new filter */
+#define SHARP_FILTER_QTHRESH 0          /* Q threshold for 8-tap sharp filter */
+
+#define ALTREF_HIGH_PRECISION_MV 1      /* whether to use high precision mv
+                                           for altref computation */
+#define HIGH_PRECISION_MV_QTHRESH 200   /* Q threshold for use of high precision
+                                           mv. Choose a very high value for
+                                           now so that HIGH_PRECISION is always
+                                           chosen */
+
+#if CONFIG_INTERNAL_STATS
+#include "math.h"
+
+extern double vp9_calc_ssim(YV12_BUFFER_CONFIG *source,
+                            YV12_BUFFER_CONFIG *dest, int lumamask,
+                            double *weight);
+
+
+extern double vp9_calc_ssimg(YV12_BUFFER_CONFIG *source,
+                             YV12_BUFFER_CONFIG *dest, double *ssim_y,
+                             double *ssim_u, double *ssim_v);
+
+
+#endif
+
+// #define OUTPUT_YUV_REC
+
+#ifdef OUTPUT_YUV_SRC
+FILE *yuv_file;
+#endif
+#ifdef OUTPUT_YUV_REC
+FILE *yuv_rec_file;
+#endif
+
+#if 0
+FILE *framepsnr;
+FILE *kf_list;
+FILE *keyfile;
+#endif
+
+
+#ifdef ENTROPY_STATS
+extern int intra_mode_stats[VP9_INTRA_MODES]
+                           [VP9_INTRA_MODES]
+                           [VP9_INTRA_MODES];
+#endif
+
+#ifdef NMV_STATS
+extern void init_nmvstats();
+extern void print_nmvstats();
+#endif
+#ifdef MODE_STATS
+extern void init_tx_count_stats();
+extern void write_tx_count_stats();
+extern void init_switchable_interp_stats();
+extern void write_switchable_interp_stats();
+#endif
+
+#ifdef SPEEDSTATS
+unsigned int frames_at_speed[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
+#endif
+
+#if defined(SECTIONBITS_OUTPUT)
+extern unsigned __int64 Sectionbits[500];
+#endif
+
+extern void vp9_init_quantizer(VP9_COMP *cpi);
+
+// Tables relating active max Q to active min Q
+static int kf_low_motion_minq[QINDEX_RANGE];
+static int kf_high_motion_minq[QINDEX_RANGE];
+static int gf_low_motion_minq[QINDEX_RANGE];
+static int gf_high_motion_minq[QINDEX_RANGE];
+static int inter_minq[QINDEX_RANGE];
+
+// Functions to compute the active minq lookup table entries based on a
+// formulaic approach to facilitate easier adjustment of the Q tables.
+// The formulae were derived from computing a 3rd order polynomial best
+// fit to the original data (after plotting real maxq vs minq (not q index))
+static int calculate_minq_index(double maxq,
+                                double x3, double x2, double x1, double c) {
+  int i;
+  const double minqtarget = MIN(((x3 * maxq + x2) * maxq + x1) * maxq + c,
+                                maxq);
+
+  // Special case handling to deal with the step from q2.0
+  // down to lossless mode represented by q 1.0.
+  if (minqtarget <= 2.0)
+    return 0;
+
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    if (minqtarget <= vp9_convert_qindex_to_q(i))
+      return i;
+  }
+
+  return QINDEX_RANGE - 1;
+}
+
+static void init_minq_luts(void) {
+  int i;
+
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    const double maxq = vp9_convert_qindex_to_q(i);
+
+
+    kf_low_motion_minq[i] = calculate_minq_index(maxq,
+                                                 0.000001,
+                                                 -0.0004,
+                                                 0.15,
+                                                 0.0);
+    kf_high_motion_minq[i] = calculate_minq_index(maxq,
+                                                  0.000002,
+                                                  -0.0012,
+                                                  0.5,
+                                                  0.0);
+
+    gf_low_motion_minq[i] = calculate_minq_index(maxq,
+                                                 0.0000015,
+                                                 -0.0009,
+                                                 0.33,
+                                                 0.0);
+    gf_high_motion_minq[i] = calculate_minq_index(maxq,
+                                                  0.0000021,
+                                                  -0.00125,
+                                                  0.45,
+                                                  0.0);
+    inter_minq[i] = calculate_minq_index(maxq,
+                                         0.00000271,
+                                         -0.00113,
+                                         0.697,
+                                         0.0);
+
+  }
+}
+
+static void set_mvcost(MACROBLOCK *mb) {
+  if (mb->e_mbd.allow_high_precision_mv) {
+    mb->mvcost = mb->nmvcost_hp;
+    mb->mvsadcost = mb->nmvsadcost_hp;
+  } else {
+    mb->mvcost = mb->nmvcost;
+    mb->mvsadcost = mb->nmvsadcost;
+  }
+}
+
+void vp9_initialize_enc() {
+  static int init_done = 0;
+
+  if (!init_done) {
+    vp9_initialize_common();
+    vp9_tokenize_initialize();
+    vp9_init_quant_tables();
+    vp9_init_me_luts();
+    init_minq_luts();
+    // init_base_skip_probs();
+    init_done = 1;
+  }
+}
+
+static void setup_features(VP9_COMP *cpi) {
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+
+  // Set up default state for MB feature flags
+  xd->segmentation_enabled = 0;
+
+  xd->update_mb_segmentation_map = 0;
+  xd->update_mb_segmentation_data = 0;
+  vpx_memset(xd->mb_segment_tree_probs, 255, sizeof(xd->mb_segment_tree_probs));
+
+  vp9_clearall_segfeatures(xd);
+
+  xd->mode_ref_lf_delta_enabled = 0;
+  xd->mode_ref_lf_delta_update = 0;
+  vpx_memset(xd->ref_lf_deltas, 0, sizeof(xd->ref_lf_deltas));
+  vpx_memset(xd->mode_lf_deltas, 0, sizeof(xd->mode_lf_deltas));
+  vpx_memset(xd->last_ref_lf_deltas, 0, sizeof(xd->ref_lf_deltas));
+  vpx_memset(xd->last_mode_lf_deltas, 0, sizeof(xd->mode_lf_deltas));
+
+  set_default_lf_deltas(cpi);
+}
+
+static void dealloc_compressor_data(VP9_COMP *cpi) {
+  // Delete sementation map
+  vpx_free(cpi->segmentation_map);
+  cpi->segmentation_map = 0;
+  vpx_free(cpi->common.last_frame_seg_map);
+  cpi->common.last_frame_seg_map = 0;
+  vpx_free(cpi->coding_context.last_frame_seg_map_copy);
+  cpi->coding_context.last_frame_seg_map_copy = 0;
+
+  vpx_free(cpi->active_map);
+  cpi->active_map = 0;
+
+  vp9_free_frame_buffers(&cpi->common);
+
+  vp9_free_frame_buffer(&cpi->last_frame_uf);
+  vp9_free_frame_buffer(&cpi->scaled_source);
+  vp9_free_frame_buffer(&cpi->alt_ref_buffer);
+  vp9_lookahead_destroy(cpi->lookahead);
+
+  vpx_free(cpi->tok);
+  cpi->tok = 0;
+
+  // Activity mask based per mb zbin adjustments
+  vpx_free(cpi->mb_activity_map);
+  cpi->mb_activity_map = 0;
+  vpx_free(cpi->mb_norm_activity_map);
+  cpi->mb_norm_activity_map = 0;
+
+  vpx_free(cpi->mb.pip);
+  cpi->mb.pip = 0;
+}
+
+// Computes a q delta (in "q index" terms) to get from a starting q value
+// to a target value
+// target q value
+static int compute_qdelta(VP9_COMP *cpi, double qstart, double qtarget) {
+  int i;
+  int start_index = cpi->worst_quality;
+  int target_index = cpi->worst_quality;
+
+  // Convert the average q value to an index.
+  for (i = cpi->best_quality; i < cpi->worst_quality; i++) {
+    start_index = i;
+    if (vp9_convert_qindex_to_q(i) >= qstart)
+      break;
+  }
+
+  // Convert the q target to an index
+  for (i = cpi->best_quality; i < cpi->worst_quality; i++) {
+    target_index = i;
+    if (vp9_convert_qindex_to_q(i) >= qtarget)
+      break;
+  }
+
+  return target_index - start_index;
+}
+
+static void configure_static_seg_features(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+
+  int high_q = (int)(cpi->avg_q > 48.0);
+  int qi_delta;
+
+  // Disable and clear down for KF
+  if (cm->frame_type == KEY_FRAME) {
+    // Clear down the global segmentation map
+    vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
+    xd->update_mb_segmentation_map = 0;
+    xd->update_mb_segmentation_data = 0;
+    cpi->static_mb_pct = 0;
+
+    // Disable segmentation
+    vp9_disable_segmentation((VP9_PTR)cpi);
+
+    // Clear down the segment features.
+    vp9_clearall_segfeatures(xd);
+  } else if (cpi->refresh_alt_ref_frame) {
+    // If this is an alt ref frame
+    // Clear down the global segmentation map
+    vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
+    xd->update_mb_segmentation_map = 0;
+    xd->update_mb_segmentation_data = 0;
+    cpi->static_mb_pct = 0;
+
+    // Disable segmentation and individual segment features by default
+    vp9_disable_segmentation((VP9_PTR)cpi);
+    vp9_clearall_segfeatures(xd);
+
+    // Scan frames from current to arf frame.
+    // This function re-enables segmentation if appropriate.
+    vp9_update_mbgraph_stats(cpi);
+
+    // If segmentation was enabled set those features needed for the
+    // arf itself.
+    if (xd->segmentation_enabled) {
+      xd->update_mb_segmentation_map = 1;
+      xd->update_mb_segmentation_data = 1;
+
+      qi_delta = compute_qdelta(cpi, cpi->avg_q, (cpi->avg_q * 0.875));
+      vp9_set_segdata(xd, 1, SEG_LVL_ALT_Q, (qi_delta - 2));
+      vp9_set_segdata(xd, 1, SEG_LVL_ALT_LF, -2);
+
+      vp9_enable_segfeature(xd, 1, SEG_LVL_ALT_Q);
+      vp9_enable_segfeature(xd, 1, SEG_LVL_ALT_LF);
+
+      // Where relevant assume segment data is delta data
+      xd->mb_segment_abs_delta = SEGMENT_DELTADATA;
+
+    }
+  } else if (xd->segmentation_enabled) {
+    // All other frames if segmentation has been enabled
+
+    // First normal frame in a valid gf or alt ref group
+    if (cpi->common.frames_since_golden == 0) {
+      // Set up segment features for normal frames in an arf group
+      if (cpi->source_alt_ref_active) {
+        xd->update_mb_segmentation_map = 0;
+        xd->update_mb_segmentation_data = 1;
+        xd->mb_segment_abs_delta = SEGMENT_DELTADATA;
+
+        qi_delta = compute_qdelta(cpi, cpi->avg_q,
+                                  (cpi->avg_q * 1.125));
+        vp9_set_segdata(xd, 1, SEG_LVL_ALT_Q, (qi_delta + 2));
+        vp9_enable_segfeature(xd, 1, SEG_LVL_ALT_Q);
+
+        vp9_set_segdata(xd, 1, SEG_LVL_ALT_LF, -2);
+        vp9_enable_segfeature(xd, 1, SEG_LVL_ALT_LF);
+
+        // Segment coding disabled for compred testing
+        if (high_q || (cpi->static_mb_pct == 100)) {
+          vp9_set_segdata(xd, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
+          vp9_enable_segfeature(xd, 1, SEG_LVL_REF_FRAME);
+          vp9_enable_segfeature(xd, 1, SEG_LVL_SKIP);
+        }
+      } else {
+        // Disable segmentation and clear down features if alt ref
+        // is not active for this group
+
+        vp9_disable_segmentation((VP9_PTR)cpi);
+
+        vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
+
+        xd->update_mb_segmentation_map = 0;
+        xd->update_mb_segmentation_data = 0;
+
+        vp9_clearall_segfeatures(xd);
+      }
+    } else if (cpi->is_src_frame_alt_ref) {
+      // Special case where we are coding over the top of a previous
+      // alt ref frame.
+      // Segment coding disabled for compred testing
+
+      // Enable ref frame features for segment 0 as well
+      vp9_enable_segfeature(xd, 0, SEG_LVL_REF_FRAME);
+      vp9_enable_segfeature(xd, 1, SEG_LVL_REF_FRAME);
+
+      // All mbs should use ALTREF_FRAME
+      vp9_clear_segdata(xd, 0, SEG_LVL_REF_FRAME);
+      vp9_set_segdata(xd, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
+      vp9_clear_segdata(xd, 1, SEG_LVL_REF_FRAME);
+      vp9_set_segdata(xd, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
+
+      // Skip all MBs if high Q (0,0 mv and skip coeffs)
+      if (high_q) {
+          vp9_enable_segfeature(xd, 0, SEG_LVL_SKIP);
+          vp9_enable_segfeature(xd, 1, SEG_LVL_SKIP);
+      }
+      // Enable data udpate
+      xd->update_mb_segmentation_data = 1;
+    } else {
+      // All other frames.
+
+      // No updates.. leave things as they are.
+      xd->update_mb_segmentation_map = 0;
+      xd->update_mb_segmentation_data = 0;
+    }
+  }
+}
+
+#ifdef ENTROPY_STATS
+void vp9_update_mode_context_stats(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  int i, j;
+  unsigned int (*inter_mode_counts)[VP9_INTER_MODES - 1][2] =
+      cm->fc.inter_mode_counts;
+  int64_t (*mv_ref_stats)[VP9_INTER_MODES - 1][2] = cpi->mv_ref_stats;
+  FILE *f;
+
+  // Read the past stats counters
+  f = fopen("mode_context.bin",  "rb");
+  if (!f) {
+    vpx_memset(cpi->mv_ref_stats, 0, sizeof(cpi->mv_ref_stats));
+  } else {
+    fread(cpi->mv_ref_stats, sizeof(cpi->mv_ref_stats), 1, f);
+    fclose(f);
+  }
+
+  // Add in the values for this frame
+  for (i = 0; i < INTER_MODE_CONTEXTS; i++) {
+    for (j = 0; j < VP9_INTER_MODES - 1; j++) {
+      mv_ref_stats[i][j][0] += (int64_t)inter_mode_counts[i][j][0];
+      mv_ref_stats[i][j][1] += (int64_t)inter_mode_counts[i][j][1];
+    }
+  }
+
+  // Write back the accumulated stats
+  f = fopen("mode_context.bin",  "wb");
+  fwrite(cpi->mv_ref_stats, sizeof(cpi->mv_ref_stats), 1, f);
+  fclose(f);
+}
+
+void print_mode_context(VP9_COMP *cpi) {
+  FILE *f = fopen("vp9_modecont.c", "a");
+  int i, j;
+
+  fprintf(f, "#include \"vp9_entropy.h\"\n");
+  fprintf(
+      f,
+      "const int inter_mode_probs[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1] =");
+  fprintf(f, "{\n");
+  for (j = 0; j < INTER_MODE_CONTEXTS; j++) {
+    fprintf(f, "  {/* %d */ ", j);
+    fprintf(f, "    ");
+    for (i = 0; i < VP9_INTER_MODES - 1; i++) {
+      int this_prob;
+      int64_t count = cpi->mv_ref_stats[j][i][0] + cpi->mv_ref_stats[j][i][1];
+      if (count)
+        this_prob = ((cpi->mv_ref_stats[j][i][0] * 256) + (count >> 1)) / count;
+      else
+        this_prob = 128;
+
+      // context probs
+      fprintf(f, "%5d, ", this_prob);
+    }
+    fprintf(f, "  },\n");
+  }
+
+  fprintf(f, "};\n");
+  fclose(f);
+}
+#endif  // ENTROPY_STATS
+
+// DEBUG: Print out the segment id of each MB in the current frame.
+static void print_seg_map(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  int row, col;
+  int map_index = 0;
+  FILE *statsfile = fopen("segmap.stt", "a");
+
+  fprintf(statsfile, "%10d\n", cm->current_video_frame);
+
+  for (row = 0; row < cpi->common.mi_rows; row++) {
+    for (col = 0; col < cpi->common.mi_cols; col++) {
+      fprintf(statsfile, "%10d", cpi->segmentation_map[map_index]);
+      map_index++;
+    }
+    fprintf(statsfile, "\n");
+  }
+  fprintf(statsfile, "\n");
+
+  fclose(statsfile);
+}
+
+static void update_reference_segmentation_map(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  int row, col;
+  MODE_INFO *mi, *mi_ptr = cm->mi;
+  uint8_t *cache_ptr = cm->last_frame_seg_map, *cache;
+
+  for (row = 0; row < cm->mi_rows; row++) {
+    mi = mi_ptr;
+    cache = cache_ptr;
+    for (col = 0; col < cm->mi_cols; col++, mi++, cache++)
+      cache[0] = mi->mbmi.segment_id;
+    mi_ptr += cm->mode_info_stride;
+    cache_ptr += cm->mi_cols;
+  }
+}
+
+static void set_default_lf_deltas(VP9_COMP *cpi) {
+  cpi->mb.e_mbd.mode_ref_lf_delta_enabled = 1;
+  cpi->mb.e_mbd.mode_ref_lf_delta_update = 1;
+
+  vpx_memset(cpi->mb.e_mbd.ref_lf_deltas, 0, sizeof(cpi->mb.e_mbd.ref_lf_deltas));
+  vpx_memset(cpi->mb.e_mbd.mode_lf_deltas, 0, sizeof(cpi->mb.e_mbd.mode_lf_deltas));
+
+  // Test of ref frame deltas
+  cpi->mb.e_mbd.ref_lf_deltas[INTRA_FRAME] = 2;
+  cpi->mb.e_mbd.ref_lf_deltas[LAST_FRAME] = 0;
+  cpi->mb.e_mbd.ref_lf_deltas[GOLDEN_FRAME] = -2;
+  cpi->mb.e_mbd.ref_lf_deltas[ALTREF_FRAME] = -2;
+
+  cpi->mb.e_mbd.mode_lf_deltas[0] = 0;              // Zero
+  cpi->mb.e_mbd.mode_lf_deltas[1] = 0;               // New mv
+}
+
+static void set_rd_speed_thresholds(VP9_COMP *cpi, int mode, int speed) {
+  SPEED_FEATURES *sf = &cpi->sf;
+  int speed_multiplier = speed + 1;
+  int i;
+
+  // Set baseline threshold values
+  for (i = 0; i < MAX_MODES; ++i)
+    sf->thresh_mult[i] = mode == 0 ? -500 : 0;
+
+  sf->thresh_mult[THR_ZEROMV   ] = 0;
+  sf->thresh_mult[THR_ZEROG    ] = 0;
+  sf->thresh_mult[THR_ZEROA    ] = 0;
+
+  sf->thresh_mult[THR_NEARESTMV] = 0;
+  sf->thresh_mult[THR_NEARESTG ] = 0;
+  sf->thresh_mult[THR_NEARESTA ] = 0;
+
+  sf->thresh_mult[THR_NEARMV   ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_NEARG    ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_NEARA    ] += speed_multiplier * 1000;
+
+  sf->thresh_mult[THR_DC       ] = 0;
+  sf->thresh_mult[THR_TM       ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_V_PRED   ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_H_PRED   ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_D45_PRED ] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_D135_PRED] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_D117_PRED] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_D153_PRED] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_D27_PRED ] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_D63_PRED ] += speed_multiplier * 1500;
+
+  sf->thresh_mult[THR_B_PRED   ] += speed_multiplier * 2500;
+
+  sf->thresh_mult[THR_NEWMV    ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_NEWG     ] += speed_multiplier * 1000;
+  sf->thresh_mult[THR_NEWA     ] += speed_multiplier * 1000;
+
+  sf->thresh_mult[THR_SPLITMV  ] += speed_multiplier * 2500;
+  sf->thresh_mult[THR_SPLITG   ] += speed_multiplier * 2500;
+  sf->thresh_mult[THR_SPLITA   ] += speed_multiplier * 2500;
+
+  sf->thresh_mult[THR_COMP_ZEROLA   ] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_COMP_ZEROGA   ] += speed_multiplier * 1500;
+
+  sf->thresh_mult[THR_COMP_NEARESTLA] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_COMP_NEARESTGA] += speed_multiplier * 1500;
+
+  sf->thresh_mult[THR_COMP_NEARLA   ] += speed_multiplier * 1500;
+  sf->thresh_mult[THR_COMP_NEARGA   ] += speed_multiplier * 1500;
+
+  sf->thresh_mult[THR_COMP_NEWLA    ] += speed_multiplier * 2000;
+  sf->thresh_mult[THR_COMP_NEWGA    ] += speed_multiplier * 2000;
+
+  sf->thresh_mult[THR_COMP_SPLITLA  ] += speed_multiplier * 4500;
+  sf->thresh_mult[THR_COMP_SPLITGA  ] += speed_multiplier * 4500;
+
+  if (speed > 4) {
+    for (i = 0; i < MAX_MODES; ++i)
+      sf->thresh_mult[i] = INT_MAX;
+
+    sf->thresh_mult[THR_DC       ] = 0;
+    sf->thresh_mult[THR_TM       ] = 0;
+    sf->thresh_mult[THR_NEWMV    ] = 4000;
+    sf->thresh_mult[THR_NEWG     ] = 4000;
+    sf->thresh_mult[THR_NEWA     ] = 4000;
+    sf->thresh_mult[THR_NEARESTMV] = 0;
+    sf->thresh_mult[THR_NEARESTG ] = 0;
+    sf->thresh_mult[THR_NEARESTA ] = 0;
+    sf->thresh_mult[THR_NEARMV   ] = 2000;
+    sf->thresh_mult[THR_NEARG    ] = 2000;
+    sf->thresh_mult[THR_NEARA    ] = 2000;
+    sf->thresh_mult[THR_COMP_NEARESTLA] = 2000;
+    sf->recode_loop = 0;
+  }
+
+  /* disable frame modes if flags not set */
+  if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
+    sf->thresh_mult[THR_NEWMV    ] = INT_MAX;
+    sf->thresh_mult[THR_NEARESTMV] = INT_MAX;
+    sf->thresh_mult[THR_ZEROMV   ] = INT_MAX;
+    sf->thresh_mult[THR_NEARMV   ] = INT_MAX;
+    sf->thresh_mult[THR_SPLITMV  ] = INT_MAX;
+  }
+  if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
+    sf->thresh_mult[THR_NEARESTG ] = INT_MAX;
+    sf->thresh_mult[THR_ZEROG    ] = INT_MAX;
+    sf->thresh_mult[THR_NEARG    ] = INT_MAX;
+    sf->thresh_mult[THR_NEWG     ] = INT_MAX;
+    sf->thresh_mult[THR_SPLITG   ] = INT_MAX;
+  }
+  if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
+    sf->thresh_mult[THR_NEARESTA ] = INT_MAX;
+    sf->thresh_mult[THR_ZEROA    ] = INT_MAX;
+    sf->thresh_mult[THR_NEARA    ] = INT_MAX;
+    sf->thresh_mult[THR_NEWA     ] = INT_MAX;
+    sf->thresh_mult[THR_SPLITA   ] = INT_MAX;
+  }
+
+  if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
+      (VP9_LAST_FLAG | VP9_ALT_FLAG)) {
+    sf->thresh_mult[THR_COMP_ZEROLA   ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEARLA   ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEWLA    ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_SPLITLA  ] = INT_MAX;
+  }
+  if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
+      (VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
+    sf->thresh_mult[THR_COMP_ZEROGA   ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEARGA   ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_NEWGA    ] = INT_MAX;
+    sf->thresh_mult[THR_COMP_SPLITGA  ] = INT_MAX;
+  }
+}
+
+void vp9_set_speed_features(VP9_COMP *cpi) {
+  SPEED_FEATURES *sf = &cpi->sf;
+  int mode = cpi->compressor_speed;
+  int speed = cpi->speed;
+  int i;
+
+  // Only modes 0 and 1 supported for now in experimental code basae
+  if (mode > 1)
+    mode = 1;
+
+  // Initialise default mode frequency sampling variables
+  for (i = 0; i < MAX_MODES; i ++) {
+    cpi->mode_check_freq[i] = 0;
+    cpi->mode_test_hit_counts[i] = 0;
+    cpi->mode_chosen_counts[i] = 0;
+  }
+
+  // best quality defaults
+  sf->RD = 1;
+  sf->search_method = NSTEP;
+  sf->auto_filter = 1;
+  sf->recode_loop = 1;
+  sf->quarter_pixel_search = 1;
+  sf->half_pixel_search = 1;
+  sf->iterative_sub_pixel = 1;
+  sf->optimize_coefficients = !cpi->oxcf.lossless;
+  sf->first_step = 0;
+  sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
+  sf->comp_inter_joint_search_thresh = BLOCK_SIZE_AB4X4;
+  sf->adpative_rd_thresh = 0;
+
+#if CONFIG_MULTIPLE_ARF
+  // Switch segmentation off.
+  sf->static_segmentation = 0;
+#else
+  sf->static_segmentation = 0;
+#endif
+
+  switch (mode) {
+    case 0: // best quality mode
+      sf->search_best_filter = SEARCH_BEST_FILTER;
+      break;
+
+    case 1:
+#if CONFIG_MULTIPLE_ARF
+      // Switch segmentation off.
+      sf->static_segmentation = 0;
+#else
+      sf->static_segmentation = 0;
+#endif
+      sf->comp_inter_joint_search_thresh = BLOCK_SIZE_SB8X8;
+      sf->adpative_rd_thresh = 1;
+      if (speed > 0) {
+        sf->comp_inter_joint_search_thresh = BLOCK_SIZE_TYPES;
+        sf->optimize_coefficients = 0;
+        sf->first_step = 1;
+      }
+      break;
+
+  }; /* switch */
+
+  // Set rd thresholds based on mode and speed setting
+  set_rd_speed_thresholds(cpi, mode, speed);
+
+  // Slow quant, dct and trellis not worthwhile for first pass
+  // so make sure they are always turned off.
+  if (cpi->pass == 1) {
+    sf->optimize_coefficients = 0;
+  }
+
+  cpi->mb.fwd_txm16x16  = vp9_short_fdct16x16;
+  cpi->mb.fwd_txm8x8    = vp9_short_fdct8x8;
+  cpi->mb.fwd_txm8x4    = vp9_short_fdct8x4;
+  cpi->mb.fwd_txm4x4    = vp9_short_fdct4x4;
+  if (cpi->oxcf.lossless || cpi->mb.e_mbd.lossless) {
+    cpi->mb.fwd_txm8x4    = vp9_short_walsh8x4;
+    cpi->mb.fwd_txm4x4    = vp9_short_walsh4x4;
+  }
+
+  cpi->mb.quantize_b_4x4      = vp9_regular_quantize_b_4x4;
+
+  vp9_init_quantizer(cpi);
+
+  if (cpi->sf.iterative_sub_pixel == 1) {
+    cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_step_iteratively;
+  } else if (cpi->sf.quarter_pixel_search) {
+    cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_step;
+  } else if (cpi->sf.half_pixel_search) {
+    cpi->find_fractional_mv_step = vp9_find_best_half_pixel_step;
+  }
+
+  cpi->mb.optimize = cpi->sf.optimize_coefficients == 1 && cpi->pass != 1;
+
+#ifdef SPEEDSTATS
+  frames_at_speed[cpi->speed]++;
+#endif
+}
+
+static void alloc_raw_frame_buffers(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  cpi->lookahead = vp9_lookahead_init(cpi->oxcf.width, cpi->oxcf.height,
+                                      cm->subsampling_x, cm->subsampling_y,
+                                      cpi->oxcf.lag_in_frames);
+  if (!cpi->lookahead)
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate lag buffers");
+
+  if (vp9_realloc_frame_buffer(&cpi->alt_ref_buffer,
+                               cpi->oxcf.width, cpi->oxcf.height,
+                               cm->subsampling_x, cm->subsampling_y,
+                               VP9BORDERINPIXELS))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate altref buffer");
+}
+
+static int alloc_partition_data(VP9_COMP *cpi) {
+  vpx_free(cpi->mb.pip);
+
+  cpi->mb.pip = vpx_calloc((cpi->common.mode_info_stride) *
+                           (cpi->common.mi_rows + 64 / MI_SIZE),
+                           sizeof(PARTITION_INFO));
+  if (!cpi->mb.pip)
+    return 1;
+
+  cpi->mb.pi = cpi->mb.pip + cpi->common.mode_info_stride + 1;
+
+  return 0;
+}
+
+void vp9_alloc_compressor_data(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  if (vp9_alloc_frame_buffers(cm, cm->width, cm->height))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate frame buffers");
+
+  if (alloc_partition_data(cpi))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate partition data");
+
+  if (vp9_alloc_frame_buffer(&cpi->last_frame_uf,
+                             cm->width, cm->height,
+                             cm->subsampling_x, cm->subsampling_y,
+                             VP9BORDERINPIXELS))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate last frame buffer");
+
+  if (vp9_alloc_frame_buffer(&cpi->scaled_source,
+                             cm->width, cm->height,
+                             cm->subsampling_x, cm->subsampling_y,
+                             VP9BORDERINPIXELS))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate scaled source buffer");
+
+  vpx_free(cpi->tok);
+
+  {
+    unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
+
+    CHECK_MEM_ERROR(cpi->tok, vpx_calloc(tokens, sizeof(*cpi->tok)));
+  }
+
+  // Data used for real time vc mode to see if gf needs refreshing
+  cpi->inter_zz_count = 0;
+  cpi->gf_bad_count = 0;
+  cpi->gf_update_recommended = 0;
+
+  vpx_free(cpi->mb_activity_map);
+  CHECK_MEM_ERROR(cpi->mb_activity_map,
+                  vpx_calloc(sizeof(unsigned int),
+                             cm->mb_rows * cm->mb_cols));
+
+  vpx_free(cpi->mb_norm_activity_map);
+  CHECK_MEM_ERROR(cpi->mb_norm_activity_map,
+                  vpx_calloc(sizeof(unsigned int),
+                             cm->mb_rows * cm->mb_cols));
+}
+
+
+static void update_frame_size(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  vp9_update_frame_size(cm);
+
+  // Update size of buffers local to this frame
+  if (vp9_realloc_frame_buffer(&cpi->last_frame_uf,
+                               cm->width, cm->height,
+                               cm->subsampling_x, cm->subsampling_y,
+                               VP9BORDERINPIXELS))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to reallocate last frame buffer");
+
+  if (vp9_realloc_frame_buffer(&cpi->scaled_source,
+                               cm->width, cm->height,
+                               cm->subsampling_x, cm->subsampling_y,
+                               VP9BORDERINPIXELS))
+    vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
+                       "Failed to reallocate scaled source buffer");
+
+  {
+    int y_stride = cpi->scaled_source.y_stride;
+
+    if (cpi->sf.search_method == NSTEP) {
+      vp9_init3smotion_compensation(&cpi->mb, y_stride);
+    } else if (cpi->sf.search_method == DIAMOND) {
+      vp9_init_dsmotion_compensation(&cpi->mb, y_stride);
+    }
+  }
+}
+
+
+// TODO perhaps change number of steps expose to outside world when setting
+// max and min limits. Also this will likely want refining for the extended Q
+// range.
+//
+// Table that converts 0-63 Q range values passed in outside to the Qindex
+// range used internally.
+static const int q_trans[] = {
+  0,    4,   8,  12,  16,  20,  24,  28,
+  32,   36,  40,  44,  48,  52,  56,  60,
+  64,   68,  72,  76,  80,  84,  88,  92,
+  96,  100, 104, 108, 112, 116, 120, 124,
+  128, 132, 136, 140, 144, 148, 152, 156,
+  160, 164, 168, 172, 176, 180, 184, 188,
+  192, 196, 200, 204, 208, 212, 216, 220,
+  224, 228, 232, 236, 240, 244, 249, 255,
+};
+
+int vp9_reverse_trans(int x) {
+  int i;
+
+  for (i = 0; i < 64; i++)
+    if (q_trans[i] >= x)
+      return i;
+
+  return 63;
+};
+void vp9_new_frame_rate(VP9_COMP *cpi, double framerate) {
+  if (framerate < 0.1)
+    framerate = 30;
+
+  cpi->oxcf.frame_rate             = framerate;
+  cpi->output_frame_rate            = cpi->oxcf.frame_rate;
+  cpi->per_frame_bandwidth          = (int)(cpi->oxcf.target_bandwidth / cpi->output_frame_rate);
+  cpi->av_per_frame_bandwidth        = (int)(cpi->oxcf.target_bandwidth / cpi->output_frame_rate);
+  cpi->min_frame_bandwidth          = (int)(cpi->av_per_frame_bandwidth * cpi->oxcf.two_pass_vbrmin_section / 100);
+
+
+  cpi->min_frame_bandwidth = MAX(cpi->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
+
+  // Set Maximum gf/arf interval
+  cpi->max_gf_interval = 16;
+
+  // Extended interval for genuinely static scenes
+  cpi->twopass.static_scene_max_gf_interval = cpi->key_frame_frequency >> 1;
+
+  // Special conditions when alt ref frame enabled in lagged compress mode
+  if (cpi->oxcf.play_alternate && cpi->oxcf.lag_in_frames) {
+    if (cpi->max_gf_interval > cpi->oxcf.lag_in_frames - 1)
+      cpi->max_gf_interval = cpi->oxcf.lag_in_frames - 1;
+
+    if (cpi->twopass.static_scene_max_gf_interval > cpi->oxcf.lag_in_frames - 1)
+      cpi->twopass.static_scene_max_gf_interval = cpi->oxcf.lag_in_frames - 1;
+  }
+
+  if (cpi->max_gf_interval > cpi->twopass.static_scene_max_gf_interval)
+    cpi->max_gf_interval = cpi->twopass.static_scene_max_gf_interval;
+}
+
+static int64_t rescale(int val, int64_t num, int denom) {
+  int64_t llnum = num;
+  int64_t llden = denom;
+  int64_t llval = val;
+
+  return (llval * llnum / llden);
+}
+
+static void set_tile_limits(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  int min_log2_tiles, max_log2_tiles;
+
+  cm->log2_tile_columns = cpi->oxcf.tile_columns;
+  cm->log2_tile_rows = cpi->oxcf.tile_rows;
+
+  vp9_get_tile_n_bits(cm, &min_log2_tiles, &max_log2_tiles);
+  max_log2_tiles += min_log2_tiles;
+
+  cm->log2_tile_columns = clamp(cm->log2_tile_columns,
+                                min_log2_tiles, max_log2_tiles);
+
+  cm->tile_columns = 1 << cm->log2_tile_columns;
+  cm->tile_rows = 1 << cm->log2_tile_rows;
+}
+
+static void init_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+  VP9_COMMON *const cm = &cpi->common;
+  int i;
+
+  cpi->oxcf = *oxcf;
+  cpi->goldfreq = 7;
+
+  cm->version = oxcf->version;
+
+  cm->width = oxcf->width;
+  cm->height = oxcf->height;
+  cm->subsampling_x = 0;
+  cm->subsampling_y = 0;
+  vp9_alloc_compressor_data(cpi);
+
+  // change includes all joint functionality
+  vp9_change_config(ptr, oxcf);
+
+  // Initialize active best and worst q and average q values.
+  cpi->active_worst_quality         = cpi->oxcf.worst_allowed_q;
+  cpi->active_best_quality          = cpi->oxcf.best_allowed_q;
+  cpi->avg_frame_qindex             = cpi->oxcf.worst_allowed_q;
+
+  // Initialise the starting buffer levels
+  cpi->buffer_level                 = cpi->oxcf.starting_buffer_level;
+  cpi->bits_off_target              = cpi->oxcf.starting_buffer_level;
+
+  cpi->rolling_target_bits          = cpi->av_per_frame_bandwidth;
+  cpi->rolling_actual_bits          = cpi->av_per_frame_bandwidth;
+  cpi->long_rolling_target_bits     = cpi->av_per_frame_bandwidth;
+  cpi->long_rolling_actual_bits     = cpi->av_per_frame_bandwidth;
+
+  cpi->total_actual_bits            = 0;
+  cpi->total_target_vs_actual       = 0;
+
+  cpi->static_mb_pct = 0;
+
+  cpi->lst_fb_idx = 0;
+  cpi->gld_fb_idx = 1;
+  cpi->alt_fb_idx = 2;
+
+  set_tile_limits(cpi);
+
+  cpi->fixed_divide[0] = 0;
+  for (i = 1; i < 512; i++)
+    cpi->fixed_divide[i] = 0x80000 / i;
+}
+
+
+void vp9_change_config(VP9_PTR ptr, VP9_CONFIG *oxcf) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+  VP9_COMMON *const cm = &cpi->common;
+
+  if (!cpi || !oxcf)
+    return;
+
+  if (cm->version != oxcf->version) {
+    cm->version = oxcf->version;
+  }
+
+  cpi->oxcf = *oxcf;
+
+  switch (cpi->oxcf.Mode) {
+      // Real time and one pass deprecated in test code base
+    case MODE_FIRSTPASS:
+      cpi->pass = 1;
+      cpi->compressor_speed = 1;
+      break;
+
+    case MODE_SECONDPASS:
+      cpi->pass = 2;
+      cpi->compressor_speed = 1;
+      cpi->oxcf.cpu_used = clamp(cpi->oxcf.cpu_used, -5, 5);
+      break;
+
+    case MODE_SECONDPASS_BEST:
+      cpi->pass = 2;
+      cpi->compressor_speed = 0;
+      break;
+  }
+
+  cpi->oxcf.worst_allowed_q = q_trans[oxcf->worst_allowed_q];
+  cpi->oxcf.best_allowed_q = q_trans[oxcf->best_allowed_q];
+  cpi->oxcf.cq_level = q_trans[cpi->oxcf.cq_level];
+
+  cpi->oxcf.lossless = oxcf->lossless;
+  if (cpi->oxcf.lossless) {
+    cpi->mb.e_mbd.inv_txm4x4_1_add    = vp9_short_iwalsh4x4_1_add;
+    cpi->mb.e_mbd.inv_txm4x4_add      = vp9_short_iwalsh4x4_add;
+  } else {
+    cpi->mb.e_mbd.inv_txm4x4_1_add    = vp9_short_idct4x4_1_add;
+    cpi->mb.e_mbd.inv_txm4x4_add      = vp9_short_idct4x4_add;
+  }
+
+  cpi->baseline_gf_interval = DEFAULT_GF_INTERVAL;
+
+  cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
+
+  // cpi->use_golden_frame_only = 0;
+  // cpi->use_last_frame_only = 0;
+  cpi->refresh_golden_frame = 0;
+  cpi->refresh_last_frame = 1;
+  cm->refresh_frame_context = 1;
+  cm->reset_frame_context = 0;
+
+  setup_features(cpi);
+  cpi->mb.e_mbd.allow_high_precision_mv = 0;   // Default mv precision adaptation
+  set_mvcost(&cpi->mb);
+
+  {
+    int i;
+
+    for (i = 0; i < MAX_MB_SEGMENTS; i++)
+      cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
+  }
+
+  // At the moment the first order values may not be > MAXQ
+  cpi->oxcf.fixed_q = MIN(cpi->oxcf.fixed_q, MAXQ);
+
+  // local file playback mode == really big buffer
+  if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK) {
+    cpi->oxcf.starting_buffer_level   = 60000;
+    cpi->oxcf.optimal_buffer_level    = 60000;
+    cpi->oxcf.maximum_buffer_size     = 240000;
+  }
+
+  // Convert target bandwidth from Kbit/s to Bit/s
+  cpi->oxcf.target_bandwidth       *= 1000;
+
+  cpi->oxcf.starting_buffer_level = rescale(cpi->oxcf.starting_buffer_level,
+                                            cpi->oxcf.target_bandwidth, 1000);
+
+  // Set or reset optimal and maximum buffer levels.
+  if (cpi->oxcf.optimal_buffer_level == 0)
+    cpi->oxcf.optimal_buffer_level = cpi->oxcf.target_bandwidth / 8;
+  else
+    cpi->oxcf.optimal_buffer_level = rescale(cpi->oxcf.optimal_buffer_level,
+                                             cpi->oxcf.target_bandwidth, 1000);
+
+  if (cpi->oxcf.maximum_buffer_size == 0)
+    cpi->oxcf.maximum_buffer_size = cpi->oxcf.target_bandwidth / 8;
+  else
+    cpi->oxcf.maximum_buffer_size = rescale(cpi->oxcf.maximum_buffer_size,
+                                            cpi->oxcf.target_bandwidth, 1000);
+
+  // Set up frame rate and related parameters rate control values.
+  vp9_new_frame_rate(cpi, cpi->oxcf.frame_rate);
+
+  // Set absolute upper and lower quality limits
+  cpi->worst_quality = cpi->oxcf.worst_allowed_q;
+  cpi->best_quality = cpi->oxcf.best_allowed_q;
+
+  // active values should only be modified if out of new range
+  cpi->active_worst_quality = clamp(cpi->active_worst_quality,
+                                    cpi->oxcf.best_allowed_q,
+                                    cpi->oxcf.worst_allowed_q);
+
+  cpi->active_best_quality = clamp(cpi->active_best_quality,
+                                   cpi->oxcf.best_allowed_q,
+                                   cpi->oxcf.worst_allowed_q);
+
+  cpi->buffered_mode = cpi->oxcf.optimal_buffer_level > 0;
+
+  cpi->cq_target_quality = cpi->oxcf.cq_level;
+
+  cm->mcomp_filter_type = DEFAULT_INTERP_FILTER;
+
+  cpi->target_bandwidth = cpi->oxcf.target_bandwidth;
+
+  cm->display_width = cpi->oxcf.width;
+  cm->display_height = cpi->oxcf.height;
+
+  // VP8 sharpness level mapping 0-7 (vs 0-10 in general VPx dialogs)
+  cpi->oxcf.Sharpness = MIN(7, cpi->oxcf.Sharpness);
+
+  cm->sharpness_level = cpi->oxcf.Sharpness;
+
+  if (cpi->initial_width) {
+    // Increasing the size of the frame beyond the first seen frame, or some
+    // otherwise signalled maximum size, is not supported.
+    // TODO(jkoleszar): exit gracefully.
+    assert(cm->width <= cpi->initial_width);
+    assert(cm->height <= cpi->initial_height);
+  }
+  update_frame_size(cpi);
+
+  if (cpi->oxcf.fixed_q >= 0) {
+    cpi->last_q[0] = cpi->oxcf.fixed_q;
+    cpi->last_q[1] = cpi->oxcf.fixed_q;
+    cpi->last_boosted_qindex = cpi->oxcf.fixed_q;
+  }
+
+  cpi->speed = cpi->oxcf.cpu_used;
+
+  if (cpi->oxcf.lag_in_frames == 0) {
+    // force to allowlag to 0 if lag_in_frames is 0;
+    cpi->oxcf.allow_lag = 0;
+  } else if (cpi->oxcf.lag_in_frames > MAX_LAG_BUFFERS) {
+     // Limit on lag buffers as these are not currently dynamically allocated
+    cpi->oxcf.lag_in_frames = MAX_LAG_BUFFERS;
+  }
+
+  // YX Temp
+#if CONFIG_MULTIPLE_ARF
+  vp9_zero(cpi->alt_ref_source);
+#else
+  cpi->alt_ref_source = NULL;
+#endif
+  cpi->is_src_frame_alt_ref = 0;
+
+#if 0
+  // Experimental RD Code
+  cpi->frame_distortion = 0;
+  cpi->last_frame_distortion = 0;
+#endif
+
+  set_tile_limits(cpi);
+}
+
+#define M_LOG2_E 0.693147180559945309417
+#define log2f(x) (log (x) / (float) M_LOG2_E)
+
+static void cal_nmvjointsadcost(int *mvjointsadcost) {
+  mvjointsadcost[0] = 600;
+  mvjointsadcost[1] = 300;
+  mvjointsadcost[2] = 300;
+  mvjointsadcost[0] = 300;
+}
+
+static void cal_nmvsadcosts(int *mvsadcost[2]) {
+  int i = 1;
+
+  mvsadcost[0][0] = 0;
+  mvsadcost[1][0] = 0;
+
+  do {
+    double z = 256 * (2 * (log2f(8 * i) + .6));
+    mvsadcost[0][i] = (int)z;
+    mvsadcost[1][i] = (int)z;
+    mvsadcost[0][-i] = (int)z;
+    mvsadcost[1][-i] = (int)z;
+  } while (++i <= MV_MAX);
+}
+
+static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
+  int i = 1;
+
+  mvsadcost[0][0] = 0;
+  mvsadcost[1][0] = 0;
+
+  do {
+    double z = 256 * (2 * (log2f(8 * i) + .6));
+    mvsadcost[0][i] = (int)z;
+    mvsadcost[1][i] = (int)z;
+    mvsadcost[0][-i] = (int)z;
+    mvsadcost[1][-i] = (int)z;
+  } while (++i <= MV_MAX);
+}
+
+VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) {
+  int i;
+  volatile union {
+    VP9_COMP *cpi;
+    VP9_PTR   ptr;
+  } ctx;
+
+  VP9_COMP *cpi;
+  VP9_COMMON *cm;
+
+  cpi = ctx.cpi = vpx_memalign(32, sizeof(VP9_COMP));
+  // Check that the CPI instance is valid
+  if (!cpi)
+    return 0;
+
+  cm = &cpi->common;
+
+  vpx_memset(cpi, 0, sizeof(VP9_COMP));
+
+  if (setjmp(cm->error.jmp)) {
+    VP9_PTR ptr = ctx.ptr;
+
+    ctx.cpi->common.error.setjmp = 0;
+    vp9_remove_compressor(&ptr);
+    return 0;
+  }
+
+  cpi->common.error.setjmp = 1;
+
+  CHECK_MEM_ERROR(cpi->mb.ss, vpx_calloc(sizeof(search_site), (MAX_MVSEARCH_STEPS * 8) + 1));
+
+  vp9_create_common(&cpi->common);
+
+  init_config((VP9_PTR)cpi, oxcf);
+
+  cpi->common.current_video_frame   = 0;
+  cpi->kf_overspend_bits            = 0;
+  cpi->kf_bitrate_adjustment        = 0;
+  cpi->frames_till_gf_update_due    = 0;
+  cpi->gf_overspend_bits            = 0;
+  cpi->non_gf_bitrate_adjustment    = 0;
+
+  // Set reference frame sign bias for ALTREF frame to 1 (for now)
+  cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = 1;
+
+  cpi->baseline_gf_interval = DEFAULT_GF_INTERVAL;
+
+  cpi->gold_is_last = 0;
+  cpi->alt_is_last  = 0;
+  cpi->gold_is_alt  = 0;
+
+  // Create the encoder segmentation map and set all entries to 0
+  CHECK_MEM_ERROR(cpi->segmentation_map,
+                  vpx_calloc(cpi->common.mi_rows * cpi->common.mi_cols, 1));
+
+  // And a copy in common for temporal coding
+  CHECK_MEM_ERROR(cm->last_frame_seg_map,
+                  vpx_calloc(cpi->common.mi_rows * cpi->common.mi_cols, 1));
+
+  // And a place holder structure is the coding context
+  // for use if we want to save and restore it
+  CHECK_MEM_ERROR(cpi->coding_context.last_frame_seg_map_copy,
+                  vpx_calloc(cpi->common.mi_rows * cpi->common.mi_cols, 1));
+
+  CHECK_MEM_ERROR(cpi->active_map, vpx_calloc(cpi->common.mb_rows * cpi->common.mb_cols, 1));
+  vpx_memset(cpi->active_map, 1, (cpi->common.mb_rows * cpi->common.mb_cols));
+  cpi->active_map_enabled = 0;
+
+  for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
+                   sizeof(cpi->mbgraph_stats[0])); i++) {
+    CHECK_MEM_ERROR(cpi->mbgraph_stats[i].mb_stats,
+                    vpx_calloc(cpi->common.mb_rows * cpi->common.mb_cols *
+                               sizeof(*cpi->mbgraph_stats[i].mb_stats),
+                               1));
+  }
+
+#ifdef ENTROPY_STATS
+  if (cpi->pass != 1)
+    init_context_counters();
+#endif
+
+#ifdef NMV_STATS
+  init_nmvstats();
+#endif
+#ifdef MODE_STATS
+  init_tx_count_stats();
+  init_switchable_interp_stats();
+#endif
+
+  /*Initialize the feed-forward activity masking.*/
+  cpi->activity_avg = 90 << 12;
+
+  cpi->frames_since_key = 8;        // Give a sensible default for the first frame.
+  cpi->key_frame_frequency = cpi->oxcf.key_freq;
+  cpi->this_key_frame_forced = 0;
+  cpi->next_key_frame_forced = 0;
+
+  cpi->source_alt_ref_pending = 0;
+  cpi->source_alt_ref_active = 0;
+  cpi->refresh_alt_ref_frame = 0;
+
+#if CONFIG_MULTIPLE_ARF
+  // Turn multiple ARF usage on/off. This is a quick hack for the initial test
+  // version. It should eventually be set via the codec API.
+  cpi->multi_arf_enabled = 1;
+
+  if (cpi->multi_arf_enabled) {
+    cpi->sequence_number = 0;
+    cpi->frame_coding_order_period = 0;
+    vp9_zero(cpi->frame_coding_order);
+    vp9_zero(cpi->arf_buffer_idx);
+  }
+#endif
+
+  cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
+#if CONFIG_INTERNAL_STATS
+  cpi->b_calculate_ssimg = 0;
+
+  cpi->count = 0;
+  cpi->bytes = 0;
+
+  if (cpi->b_calculate_psnr) {
+    cpi->total_sq_error = 0.0;
+    cpi->total_sq_error2 = 0.0;
+    cpi->total_y = 0.0;
+    cpi->total_u = 0.0;
+    cpi->total_v = 0.0;
+    cpi->total = 0.0;
+    cpi->totalp_y = 0.0;
+    cpi->totalp_u = 0.0;
+    cpi->totalp_v = 0.0;
+    cpi->totalp = 0.0;
+    cpi->tot_recode_hits = 0;
+    cpi->summed_quality = 0;
+    cpi->summed_weights = 0;
+    cpi->summedp_quality = 0;
+    cpi->summedp_weights = 0;
+  }
+
+  if (cpi->b_calculate_ssimg) {
+    cpi->total_ssimg_y = 0;
+    cpi->total_ssimg_u = 0;
+    cpi->total_ssimg_v = 0;
+    cpi->total_ssimg_all = 0;
+  }
+
+#endif
+
+  cpi->first_time_stamp_ever = INT64_MAX;
+
+  cpi->frames_till_gf_update_due      = 0;
+  cpi->key_frame_count              = 1;
+
+  cpi->ni_av_qi                     = cpi->oxcf.worst_allowed_q;
+  cpi->ni_tot_qi                    = 0;
+  cpi->ni_frames                   = 0;
+  cpi->tot_q = 0.0;
+  cpi->avg_q = vp9_convert_qindex_to_q(cpi->oxcf.worst_allowed_q);
+  cpi->total_byte_count             = 0;
+
+  cpi->rate_correction_factor         = 1.0;
+  cpi->key_frame_rate_correction_factor = 1.0;
+  cpi->gf_rate_correction_factor  = 1.0;
+  cpi->twopass.est_max_qcorrection_factor  = 1.0;
+
+  cal_nmvjointsadcost(cpi->mb.nmvjointsadcost);
+  cpi->mb.nmvcost[0] = &cpi->mb.nmvcosts[0][MV_MAX];
+  cpi->mb.nmvcost[1] = &cpi->mb.nmvcosts[1][MV_MAX];
+  cpi->mb.nmvsadcost[0] = &cpi->mb.nmvsadcosts[0][MV_MAX];
+  cpi->mb.nmvsadcost[1] = &cpi->mb.nmvsadcosts[1][MV_MAX];
+  cal_nmvsadcosts(cpi->mb.nmvsadcost);
+
+  cpi->mb.nmvcost_hp[0] = &cpi->mb.nmvcosts_hp[0][MV_MAX];
+  cpi->mb.nmvcost_hp[1] = &cpi->mb.nmvcosts_hp[1][MV_MAX];
+  cpi->mb.nmvsadcost_hp[0] = &cpi->mb.nmvsadcosts_hp[0][MV_MAX];
+  cpi->mb.nmvsadcost_hp[1] = &cpi->mb.nmvsadcosts_hp[1][MV_MAX];
+  cal_nmvsadcosts_hp(cpi->mb.nmvsadcost_hp);
+
+  for (i = 0; i < KEY_FRAME_CONTEXT; i++)
+    cpi->prior_key_frame_distance[i] = (int)cpi->output_frame_rate;
+
+#ifdef OUTPUT_YUV_SRC
+  yuv_file = fopen("bd.yuv", "ab");
+#endif
+#ifdef OUTPUT_YUV_REC
+  yuv_rec_file = fopen("rec.yuv", "wb");
+#endif
+
+#if 0
+  framepsnr = fopen("framepsnr.stt", "a");
+  kf_list = fopen("kf_list.stt", "w");
+#endif
+
+  cpi->output_pkt_list = oxcf->output_pkt_list;
+
+  if (cpi->pass == 1) {
+    vp9_init_first_pass(cpi);
+  } else if (cpi->pass == 2) {
+    size_t packet_sz = sizeof(FIRSTPASS_STATS);
+    int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
+
+    cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
+    cpi->twopass.stats_in = cpi->twopass.stats_in_start;
+    cpi->twopass.stats_in_end = (void *)((char *)cpi->twopass.stats_in
+                                         + (packets - 1) * packet_sz);
+    vp9_init_second_pass(cpi);
+  }
+
+  vp9_set_speed_features(cpi);
+
+  // Set starting values of RD threshold multipliers (128 = *1)
+  for (i = 0; i < MAX_MODES; i++)
+    cpi->rd_thresh_mult[i] = 128;
+
+#define BFP(BT, SDF, VF, SVF, SVAF, SVFHH, SVFHV, SVFHHV, SDX3F, SDX8F, SDX4DF)\
+    cpi->fn_ptr[BT].sdf            = SDF; \
+    cpi->fn_ptr[BT].vf             = VF; \
+    cpi->fn_ptr[BT].svf            = SVF; \
+    cpi->fn_ptr[BT].svaf           = SVAF; \
+    cpi->fn_ptr[BT].svf_halfpix_h  = SVFHH; \
+    cpi->fn_ptr[BT].svf_halfpix_v  = SVFHV; \
+    cpi->fn_ptr[BT].svf_halfpix_hv = SVFHHV; \
+    cpi->fn_ptr[BT].sdx3f          = SDX3F; \
+    cpi->fn_ptr[BT].sdx8f          = SDX8F; \
+    cpi->fn_ptr[BT].sdx4df         = SDX4DF;
+
+  BFP(BLOCK_32X16, vp9_sad32x16, vp9_variance32x16, vp9_sub_pixel_variance32x16,
+      vp9_sub_pixel_avg_variance32x16, NULL, NULL,
+      NULL, NULL, NULL,
+      vp9_sad32x16x4d)
+
+  BFP(BLOCK_16X32, vp9_sad16x32, vp9_variance16x32, vp9_sub_pixel_variance16x32,
+      vp9_sub_pixel_avg_variance16x32, NULL, NULL,
+      NULL, NULL, NULL,
+      vp9_sad16x32x4d)
+
+  BFP(BLOCK_64X32, vp9_sad64x32, vp9_variance64x32, vp9_sub_pixel_variance64x32,
+      vp9_sub_pixel_avg_variance64x32, NULL, NULL,
+      NULL, NULL, NULL,
+      vp9_sad64x32x4d)
+
+  BFP(BLOCK_32X64, vp9_sad32x64, vp9_variance32x64, vp9_sub_pixel_variance32x64,
+      vp9_sub_pixel_avg_variance32x64, NULL, NULL,
+      NULL, NULL, NULL,
+      vp9_sad32x64x4d)
+
+  BFP(BLOCK_32X32, vp9_sad32x32, vp9_variance32x32, vp9_sub_pixel_variance32x32,
+      vp9_sub_pixel_avg_variance32x32, vp9_variance_halfpixvar32x32_h,
+      vp9_variance_halfpixvar32x32_v,
+      vp9_variance_halfpixvar32x32_hv, vp9_sad32x32x3, vp9_sad32x32x8,
+      vp9_sad32x32x4d)
+
+  BFP(BLOCK_64X64, vp9_sad64x64, vp9_variance64x64, vp9_sub_pixel_variance64x64,
+      vp9_sub_pixel_avg_variance64x64, vp9_variance_halfpixvar64x64_h,
+      vp9_variance_halfpixvar64x64_v,
+      vp9_variance_halfpixvar64x64_hv, vp9_sad64x64x3, vp9_sad64x64x8,
+      vp9_sad64x64x4d)
+
+  BFP(BLOCK_16X16, vp9_sad16x16, vp9_variance16x16, vp9_sub_pixel_variance16x16,
+      vp9_sub_pixel_avg_variance16x16, vp9_variance_halfpixvar16x16_h,
+      vp9_variance_halfpixvar16x16_v,
+      vp9_variance_halfpixvar16x16_hv, vp9_sad16x16x3, vp9_sad16x16x8,
+      vp9_sad16x16x4d)
+
+  BFP(BLOCK_16X8, vp9_sad16x8, vp9_variance16x8, vp9_sub_pixel_variance16x8,
+      vp9_sub_pixel_avg_variance16x8, NULL, NULL, NULL,
+      vp9_sad16x8x3, vp9_sad16x8x8, vp9_sad16x8x4d)
+
+  BFP(BLOCK_8X16, vp9_sad8x16, vp9_variance8x16, vp9_sub_pixel_variance8x16,
+      vp9_sub_pixel_avg_variance8x16, NULL, NULL, NULL,
+      vp9_sad8x16x3, vp9_sad8x16x8, vp9_sad8x16x4d)
+
+  BFP(BLOCK_8X8, vp9_sad8x8, vp9_variance8x8, vp9_sub_pixel_variance8x8,
+      vp9_sub_pixel_avg_variance8x8, NULL, NULL, NULL,
+      vp9_sad8x8x3, vp9_sad8x8x8, vp9_sad8x8x4d)
+
+  BFP(BLOCK_8X4, vp9_sad8x4, vp9_variance8x4, vp9_sub_pixel_variance8x4,
+      vp9_sub_pixel_avg_variance8x4, NULL, NULL,
+      NULL, NULL, vp9_sad8x4x8,
+      vp9_sad8x4x4d)
+
+  BFP(BLOCK_4X8, vp9_sad4x8, vp9_variance4x8, vp9_sub_pixel_variance4x8,
+      vp9_sub_pixel_avg_variance4x8, NULL, NULL,
+      NULL, NULL, vp9_sad4x8x8,
+      vp9_sad4x8x4d)
+
+  BFP(BLOCK_4X4, vp9_sad4x4, vp9_variance4x4, vp9_sub_pixel_variance4x4,
+      vp9_sub_pixel_avg_variance4x4, NULL, NULL, NULL,
+      vp9_sad4x4x3, vp9_sad4x4x8, vp9_sad4x4x4d)
+
+  cpi->full_search_sad = vp9_full_search_sad;
+  cpi->diamond_search_sad = vp9_diamond_search_sad;
+  cpi->refining_search_sad = vp9_refining_search_sad;
+
+  // make sure frame 1 is okay
+  cpi->error_bins[0] = cpi->common.MBs;
+
+  /* vp9_init_quantizer() is first called here. Add check in
+   * vp9_frame_init_quantizer() so that vp9_init_quantizer is only
+   * called later when needed. This will avoid unnecessary calls of
+   * vp9_init_quantizer() for every frame.
+   */
+  vp9_init_quantizer(cpi);
+
+  vp9_loop_filter_init(cm);
+
+  cpi->common.error.setjmp = 0;
+
+  vp9_zero(cpi->y_uv_mode_count)
+
+  return (VP9_PTR) cpi;
+}
+
+void vp9_remove_compressor(VP9_PTR *ptr) {
+  VP9_COMP *cpi = (VP9_COMP *)(*ptr);
+  int i;
+
+  if (!cpi)
+    return;
+
+  if (cpi && (cpi->common.current_video_frame > 0)) {
+    if (cpi->pass == 2) {
+      vp9_end_second_pass(cpi);
+    }
+
+#ifdef ENTROPY_STATS
+    if (cpi->pass != 1) {
+      print_context_counters();
+      print_tree_update_probs();
+      print_mode_context(cpi);
+    }
+#endif
+#ifdef NMV_STATS
+    if (cpi->pass != 1)
+      print_nmvstats();
+#endif
+#ifdef MODE_STATS
+    if (cpi->pass != 1) {
+      write_tx_count_stats();
+      write_switchable_interp_stats();
+    }
+#endif
+
+#if CONFIG_INTERNAL_STATS
+
+    vp9_clear_system_state();
+
+    // printf("\n8x8-4x4:%d-%d\n", cpi->t8x8_count, cpi->t4x4_count);
+    if (cpi->pass != 1) {
+      FILE *f = fopen("opsnr.stt", "a");
+      double time_encoded = (cpi->last_end_time_stamp_seen
+                             - cpi->first_time_stamp_ever) / 10000000.000;
+      double total_encode_time = (cpi->time_receive_data + cpi->time_compress_data)   / 1000.000;
+      double dr = (double)cpi->bytes * (double) 8 / (double)1000  / time_encoded;
+
+      if (cpi->b_calculate_psnr) {
+        YV12_BUFFER_CONFIG *lst_yv12 =
+            &cpi->common.yv12_fb[cpi->common.ref_frame_map[cpi->lst_fb_idx]];
+        double samples = 3.0 / 2 * cpi->count *
+                         lst_yv12->y_width * lst_yv12->y_height;
+        double total_psnr = vp9_mse2psnr(samples, 255.0, cpi->total_sq_error);
+        double total_psnr2 = vp9_mse2psnr(samples, 255.0, cpi->total_sq_error2);
+        double total_ssim = 100 * pow(cpi->summed_quality /
+                                      cpi->summed_weights, 8.0);
+        double total_ssimp = 100 * pow(cpi->summedp_quality /
+                                       cpi->summedp_weights, 8.0);
+
+        fprintf(f, "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
+                "VPXSSIM\tVPSSIMP\t  Time(ms)\n");
+        fprintf(f, "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%8.0f\n",
+                dr, cpi->total / cpi->count, total_psnr,
+                cpi->totalp / cpi->count, total_psnr2, total_ssim, total_ssimp,
+                total_encode_time);
+//         fprintf(f, "%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%8.0f %10ld\n",
+//                 dr, cpi->total / cpi->count, total_psnr,
+//                 cpi->totalp / cpi->count, total_psnr2, total_ssim,
+//                 total_encode_time, cpi->tot_recode_hits);
+      }
+
+      if (cpi->b_calculate_ssimg) {
+        fprintf(f, "BitRate\tSSIM_Y\tSSIM_U\tSSIM_V\tSSIM_A\t  Time(ms)\n");
+        fprintf(f, "%7.2f\t%6.4f\t%6.4f\t%6.4f\t%6.4f\t%8.0f\n", dr,
+                cpi->total_ssimg_y / cpi->count, cpi->total_ssimg_u / cpi->count,
+                cpi->total_ssimg_v / cpi->count, cpi->total_ssimg_all / cpi->count, total_encode_time);
+//                fprintf(f, "%7.3f\t%6.4f\t%6.4f\t%6.4f\t%6.4f\t%8.0f  %10ld\n", dr,
+//                        cpi->total_ssimg_y / cpi->count, cpi->total_ssimg_u / cpi->count,
+//                        cpi->total_ssimg_v / cpi->count, cpi->total_ssimg_all / cpi->count, total_encode_time, cpi->tot_recode_hits);
+      }
+
+      fclose(f);
+    }
+
+#endif
+
+#ifdef ENTROPY_STATS
+    {
+      int i, j, k;
+      FILE *fmode = fopen("vp9_modecontext.c", "w");
+
+      fprintf(fmode, "\n#include \"vp9_entropymode.h\"\n\n");
+      fprintf(fmode, "const unsigned int vp9_kf_default_bmode_counts ");
+      fprintf(fmode, "[VP9_INTRA_MODES][VP9_INTRA_MODES]"
+                     "[VP9_INTRA_MODES] =\n{\n");
+
+      for (i = 0; i < VP9_INTRA_MODES; i++) {
+
+        fprintf(fmode, "    { // Above Mode :  %d\n", i);
+
+        for (j = 0; j < VP9_INTRA_MODES; j++) {
+
+          fprintf(fmode, "        {");
+
+          for (k = 0; k < VP9_INTRA_MODES; k++) {
+            if (!intra_mode_stats[i][j][k])
+              fprintf(fmode, " %5d, ", 1);
+            else
+              fprintf(fmode, " %5d, ", intra_mode_stats[i][j][k]);
+          }
+
+          fprintf(fmode, "}, // left_mode %d\n", j);
+
+        }
+
+        fprintf(fmode, "    },\n");
+
+      }
+
+      fprintf(fmode, "};\n");
+      fclose(fmode);
+    }
+#endif
+
+
+#if defined(SECTIONBITS_OUTPUT)
+
+    if (0) {
+      int i;
+      FILE *f = fopen("tokenbits.stt", "a");
+
+      for (i = 0; i < 28; i++)
+        fprintf(f, "%8d", (int)(Sectionbits[i] / 256));
+
+      fprintf(f, "\n");
+      fclose(f);
+    }
+
+#endif
+
+#if 0
+    {
+      printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
+      printf("\n_frames recive_data encod_mb_row compress_frame  Total\n");
+      printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame, cpi->time_receive_data / 1000, cpi->time_encode_mb_row / 1000, cpi->time_compress_data / 1000, (cpi->time_receive_data + cpi->time_compress_data) / 1000);
+    }
+#endif
+
+  }
+
+  dealloc_compressor_data(cpi);
+  vpx_free(cpi->mb.ss);
+  vpx_free(cpi->tok);
+
+  for (i = 0; i < sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0]); i++) {
+    vpx_free(cpi->mbgraph_stats[i].mb_stats);
+  }
+
+  vp9_remove_common(&cpi->common);
+  vpx_free(cpi);
+  *ptr = 0;
+
+#ifdef OUTPUT_YUV_SRC
+  fclose(yuv_file);
+#endif
+#ifdef OUTPUT_YUV_REC
+  fclose(yuv_rec_file);
+#endif
+
+#if 0
+
+  if (keyfile)
+    fclose(keyfile);
+
+  if (framepsnr)
+    fclose(framepsnr);
+
+  if (kf_list)
+    fclose(kf_list);
+
+#endif
+
+}
+
+
+static uint64_t calc_plane_error(uint8_t *orig, int orig_stride,
+                                 uint8_t *recon, int recon_stride,
+                                 unsigned int cols, unsigned int rows) {
+  unsigned int row, col;
+  uint64_t total_sse = 0;
+  int diff;
+
+  for (row = 0; row + 16 <= rows; row += 16) {
+    for (col = 0; col + 16 <= cols; col += 16) {
+      unsigned int sse;
+
+      vp9_mse16x16(orig + col, orig_stride, recon + col, recon_stride, &sse);
+      total_sse += sse;
+    }
+
+    /* Handle odd-sized width */
+    if (col < cols) {
+      unsigned int border_row, border_col;
+      uint8_t *border_orig = orig;
+      uint8_t *border_recon = recon;
+
+      for (border_row = 0; border_row < 16; border_row++) {
+        for (border_col = col; border_col < cols; border_col++) {
+          diff = border_orig[border_col] - border_recon[border_col];
+          total_sse += diff * diff;
+        }
+
+        border_orig += orig_stride;
+        border_recon += recon_stride;
+      }
+    }
+
+    orig += orig_stride * 16;
+    recon += recon_stride * 16;
+  }
+
+  /* Handle odd-sized height */
+  for (; row < rows; row++) {
+    for (col = 0; col < cols; col++) {
+      diff = orig[col] - recon[col];
+      total_sse += diff * diff;
+    }
+
+    orig += orig_stride;
+    recon += recon_stride;
+  }
+
+  return total_sse;
+}
+
+
+static void generate_psnr_packet(VP9_COMP *cpi) {
+  YV12_BUFFER_CONFIG      *orig = cpi->Source;
+  YV12_BUFFER_CONFIG      *recon = cpi->common.frame_to_show;
+  struct vpx_codec_cx_pkt  pkt;
+  uint64_t                 sse;
+  int                      i;
+  unsigned int             width = cpi->common.width;
+  unsigned int             height = cpi->common.height;
+
+  pkt.kind = VPX_CODEC_PSNR_PKT;
+  sse = calc_plane_error(orig->y_buffer, orig->y_stride,
+                         recon->y_buffer, recon->y_stride,
+                         width, height);
+  pkt.data.psnr.sse[0] = sse;
+  pkt.data.psnr.sse[1] = sse;
+  pkt.data.psnr.samples[0] = width * height;
+  pkt.data.psnr.samples[1] = width * height;
+
+  width = orig->uv_width;
+  height = orig->uv_height;
+
+  sse = calc_plane_error(orig->u_buffer, orig->uv_stride,
+                         recon->u_buffer, recon->uv_stride,
+                         width, height);
+  pkt.data.psnr.sse[0] += sse;
+  pkt.data.psnr.sse[2] = sse;
+  pkt.data.psnr.samples[0] += width * height;
+  pkt.data.psnr.samples[2] = width * height;
+
+  sse = calc_plane_error(orig->v_buffer, orig->uv_stride,
+                         recon->v_buffer, recon->uv_stride,
+                         width, height);
+  pkt.data.psnr.sse[0] += sse;
+  pkt.data.psnr.sse[3] = sse;
+  pkt.data.psnr.samples[0] += width * height;
+  pkt.data.psnr.samples[3] = width * height;
+
+  for (i = 0; i < 4; i++)
+    pkt.data.psnr.psnr[i] = vp9_mse2psnr(pkt.data.psnr.samples[i], 255.0,
+                                         (double)pkt.data.psnr.sse[i]);
+
+  vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
+}
+
+
+int vp9_use_as_reference(VP9_PTR ptr, int ref_frame_flags) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+
+  if (ref_frame_flags > 7)
+    return -1;
+
+  cpi->ref_frame_flags = ref_frame_flags;
+  return 0;
+}
+int vp9_update_reference(VP9_PTR ptr, int ref_frame_flags) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+
+  if (ref_frame_flags > 7)
+    return -1;
+
+  cpi->refresh_golden_frame = 0;
+  cpi->refresh_alt_ref_frame = 0;
+  cpi->refresh_last_frame   = 0;
+
+  if (ref_frame_flags & VP9_LAST_FLAG)
+    cpi->refresh_last_frame = 1;
+
+  if (ref_frame_flags & VP9_GOLD_FLAG)
+    cpi->refresh_golden_frame = 1;
+
+  if (ref_frame_flags & VP9_ALT_FLAG)
+    cpi->refresh_alt_ref_frame = 1;
+
+  return 0;
+}
+
+int vp9_copy_reference_enc(VP9_PTR ptr, VP9_REFFRAME ref_frame_flag,
+                           YV12_BUFFER_CONFIG *sd) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+  VP9_COMMON *cm = &cpi->common;
+  int ref_fb_idx;
+
+  if (ref_frame_flag == VP9_LAST_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->lst_fb_idx];
+  else if (ref_frame_flag == VP9_GOLD_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->gld_fb_idx];
+  else if (ref_frame_flag == VP9_ALT_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->alt_fb_idx];
+  else
+    return -1;
+
+  vp8_yv12_copy_frame(&cm->yv12_fb[ref_fb_idx], sd);
+
+  return 0;
+}
+
+int vp9_get_reference_enc(VP9_PTR ptr, int index, YV12_BUFFER_CONFIG **fb) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+  VP9_COMMON *cm = &cpi->common;
+
+  if (index < 0 || index >= NUM_REF_FRAMES)
+    return -1;
+
+  *fb = &cm->yv12_fb[cm->ref_frame_map[index]];
+  return 0;
+}
+
+int vp9_set_reference_enc(VP9_PTR ptr, VP9_REFFRAME ref_frame_flag,
+                          YV12_BUFFER_CONFIG *sd) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+  VP9_COMMON *cm = &cpi->common;
+
+  int ref_fb_idx;
+
+  if (ref_frame_flag == VP9_LAST_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->lst_fb_idx];
+  else if (ref_frame_flag == VP9_GOLD_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->gld_fb_idx];
+  else if (ref_frame_flag == VP9_ALT_FLAG)
+    ref_fb_idx = cm->ref_frame_map[cpi->alt_fb_idx];
+  else
+    return -1;
+
+  vp8_yv12_copy_frame(sd, &cm->yv12_fb[ref_fb_idx]);
+
+  return 0;
+}
+int vp9_update_entropy(VP9_PTR comp, int update) {
+  ((VP9_COMP *)comp)->common.refresh_frame_context = update;
+  return 0;
+}
+
+
+#ifdef OUTPUT_YUV_SRC
+void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s) {
+  uint8_t *src = s->y_buffer;
+  int h = s->y_height;
+
+  do {
+    fwrite(src, s->y_width, 1,  yuv_file);
+    src += s->y_stride;
+  } while (--h);
+
+  src = s->u_buffer;
+  h = s->uv_height;
+
+  do {
+    fwrite(src, s->uv_width, 1,  yuv_file);
+    src += s->uv_stride;
+  } while (--h);
+
+  src = s->v_buffer;
+  h = s->uv_height;
+
+  do {
+    fwrite(src, s->uv_width, 1, yuv_file);
+    src += s->uv_stride;
+  } while (--h);
+}
+#endif
+
+#ifdef OUTPUT_YUV_REC
+void vp9_write_yuv_rec_frame(VP9_COMMON *cm) {
+  YV12_BUFFER_CONFIG *s = cm->frame_to_show;
+  uint8_t *src = s->y_buffer;
+  int h = cm->height;
+
+  do {
+    fwrite(src, s->y_width, 1,  yuv_rec_file);
+    src += s->y_stride;
+  } while (--h);
+
+  src = s->u_buffer;
+  h = s->uv_height;
+
+  do {
+    fwrite(src, s->uv_width, 1,  yuv_rec_file);
+    src += s->uv_stride;
+  } while (--h);
+
+  src = s->v_buffer;
+  h = s->uv_height;
+
+  do {
+    fwrite(src, s->uv_width, 1, yuv_rec_file);
+    src += s->uv_stride;
+  } while (--h);
+
+#if CONFIG_ALPHA
+  if (s->alpha_buffer) {
+    src = s->alpha_buffer;
+    h = s->alpha_height;
+    do {
+      fwrite(src, s->alpha_width, 1,  yuv_rec_file);
+      src += s->alpha_stride;
+    } while (--h);
+  }
+#endif
+
+  fflush(yuv_rec_file);
+}
+#endif
+
+static void scale_and_extend_frame(YV12_BUFFER_CONFIG *src_fb,
+                                   YV12_BUFFER_CONFIG *dst_fb) {
+  const int in_w = src_fb->y_crop_width;
+  const int in_h = src_fb->y_crop_height;
+  const int out_w = dst_fb->y_crop_width;
+  const int out_h = dst_fb->y_crop_height;
+  int x, y, i;
+
+  uint8_t *srcs[4] = {src_fb->y_buffer, src_fb->u_buffer, src_fb->v_buffer,
+                      src_fb->alpha_buffer};
+  int src_strides[4] = {src_fb->y_stride, src_fb->uv_stride, src_fb->uv_stride,
+                        src_fb->alpha_stride};
+
+  uint8_t *dsts[4] = {dst_fb->y_buffer, dst_fb->u_buffer, dst_fb->v_buffer,
+                      dst_fb->alpha_buffer};
+  int dst_strides[4] = {dst_fb->y_stride, dst_fb->uv_stride, dst_fb->uv_stride,
+                        dst_fb->alpha_stride};
+
+  for (y = 0; y < out_h; y += 16) {
+    for (x = 0; x < out_w; x += 16) {
+      for (i = 0; i < MAX_MB_PLANE; ++i) {
+        const int factor = i == 0 ? 1 : 2;
+        const int x_q4 = x * (16 / factor) * in_w / out_w;
+        const int y_q4 = y * (16 / factor) * in_h / out_h;
+        const int src_stride = src_strides[i];
+        const int dst_stride = dst_strides[i];
+        uint8_t *src = srcs[i] + y / factor * in_h / out_h * src_stride +
+                                 x / factor * in_w / out_w;
+        uint8_t *dst = dsts[i] + y / factor * dst_stride + x / factor;
+
+        vp9_convolve8(src, src_stride, dst, dst_stride,
+                      vp9_sub_pel_filters_8[x_q4 & 0xf], 16 * in_w / out_w,
+                      vp9_sub_pel_filters_8[y_q4 & 0xf], 16 * in_h / out_h,
+                      16 / factor, 16 / factor);
+      }
+    }
+  }
+
+  vp8_yv12_extend_frame_borders(dst_fb);
+}
+
+
+static void update_alt_ref_frame_stats(VP9_COMP *cpi) {
+  // this frame refreshes means next frames don't unless specified by user
+  cpi->common.frames_since_golden = 0;
+
+#if CONFIG_MULTIPLE_ARF
+  if (!cpi->multi_arf_enabled)
+#endif
+    // Clear the alternate reference update pending flag.
+    cpi->source_alt_ref_pending = 0;
+
+  // Set the alternate reference frame active flag
+  cpi->source_alt_ref_active = 1;
+}
+static void update_golden_frame_stats(VP9_COMP *cpi) {
+  // Update the Golden frame usage counts.
+  if (cpi->refresh_golden_frame) {
+    // this frame refreshes means next frames don't unless specified by user
+    cpi->refresh_golden_frame = 0;
+    cpi->common.frames_since_golden = 0;
+
+    // ******** Fixed Q test code only ************
+    // If we are going to use the ALT reference for the next group of frames set a flag to say so.
+    if (cpi->oxcf.fixed_q >= 0 &&
+        cpi->oxcf.play_alternate && !cpi->refresh_alt_ref_frame) {
+      cpi->source_alt_ref_pending = 1;
+      cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
+    }
+
+    if (!cpi->source_alt_ref_pending)
+      cpi->source_alt_ref_active = 0;
+
+    // Decrement count down till next gf
+    if (cpi->frames_till_gf_update_due > 0)
+      cpi->frames_till_gf_update_due--;
+
+  } else if (!cpi->refresh_alt_ref_frame) {
+    // Decrement count down till next gf
+    if (cpi->frames_till_gf_update_due > 0)
+      cpi->frames_till_gf_update_due--;
+
+    if (cpi->common.frames_till_alt_ref_frame)
+      cpi->common.frames_till_alt_ref_frame--;
+
+    cpi->common.frames_since_golden++;
+  }
+}
+
+static int find_fp_qindex() {
+  int i;
+
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    if (vp9_convert_qindex_to_q(i) >= 30.0) {
+      break;
+    }
+  }
+
+  if (i == QINDEX_RANGE)
+    i--;
+
+  return i;
+}
+
+static void Pass1Encode(VP9_COMP *cpi, unsigned long *size, unsigned char *dest, unsigned int *frame_flags) {
+  (void) size;
+  (void) dest;
+  (void) frame_flags;
+
+
+  vp9_set_quantizer(cpi, find_fp_qindex());
+  vp9_first_pass(cpi);
+}
+
+#define WRITE_RECON_BUFFER 0
+#if WRITE_RECON_BUFFER
+void write_cx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame) {
+
+  // write the frame
+  FILE *yframe;
+  int i;
+  char filename[255];
+
+  sprintf(filename, "cx\\y%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->y_height; i++)
+    fwrite(frame->y_buffer + i * frame->y_stride,
+           frame->y_width, 1, yframe);
+
+  fclose(yframe);
+  sprintf(filename, "cx\\u%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->uv_height; i++)
+    fwrite(frame->u_buffer + i * frame->uv_stride,
+           frame->uv_width, 1, yframe);
+
+  fclose(yframe);
+  sprintf(filename, "cx\\v%04d.raw", this_frame);
+  yframe = fopen(filename, "wb");
+
+  for (i = 0; i < frame->uv_height; i++)
+    fwrite(frame->v_buffer + i * frame->uv_stride,
+           frame->uv_width, 1, yframe);
+
+  fclose(yframe);
+}
+#endif
+
+static double compute_edge_pixel_proportion(YV12_BUFFER_CONFIG *frame) {
+#define EDGE_THRESH 128
+  int i, j;
+  int num_edge_pels = 0;
+  int num_pels = (frame->y_height - 2) * (frame->y_width - 2);
+  uint8_t *prev = frame->y_buffer + 1;
+  uint8_t *curr = frame->y_buffer + 1 + frame->y_stride;
+  uint8_t *next = frame->y_buffer + 1 + 2 * frame->y_stride;
+  for (i = 1; i < frame->y_height - 1; i++) {
+    for (j = 1; j < frame->y_width - 1; j++) {
+      /* Sobel hor and ver gradients */
+      int v = 2 * (curr[1] - curr[-1]) + (prev[1] - prev[-1]) + (next[1] - next[-1]);
+      int h = 2 * (prev[0] - next[0]) + (prev[1] - next[1]) + (prev[-1] - next[-1]);
+      h = (h < 0 ? -h : h);
+      v = (v < 0 ? -v : v);
+      if (h > EDGE_THRESH || v > EDGE_THRESH)
+        num_edge_pels++;
+      curr++;
+      prev++;
+      next++;
+    }
+    curr += frame->y_stride - frame->y_width + 2;
+    prev += frame->y_stride - frame->y_width + 2;
+    next += frame->y_stride - frame->y_width + 2;
+  }
+  return (double)num_edge_pels / num_pels;
+}
+
+// Function to test for conditions that indicate we should loop
+// back and recode a frame.
+static int recode_loop_test(VP9_COMP *cpi,
+                            int high_limit, int low_limit,
+                            int q, int maxq, int minq) {
+  int force_recode = 0;
+  VP9_COMMON *cm = &cpi->common;
+
+  // Is frame recode allowed at all
+  // Yes if either recode mode 1 is selected or mode two is selected
+  // and the frame is a key frame. golden frame or alt_ref_frame
+  if ((cpi->sf.recode_loop == 1) ||
+      ((cpi->sf.recode_loop == 2) &&
+       ((cm->frame_type == KEY_FRAME) ||
+        cpi->refresh_golden_frame ||
+        cpi->refresh_alt_ref_frame))) {
+    // General over and under shoot tests
+    if (((cpi->projected_frame_size > high_limit) && (q < maxq)) ||
+        ((cpi->projected_frame_size < low_limit) && (q > minq))) {
+      force_recode = 1;
+    }
+    // Special Constrained quality tests
+    else if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
+      // Undershoot and below auto cq level
+      if (q > cpi->cq_target_quality &&
+          cpi->projected_frame_size < ((cpi->this_frame_target * 7) >> 3)) {
+        force_recode = 1;
+      } else if (q > cpi->oxcf.cq_level &&
+                 cpi->projected_frame_size < cpi->min_frame_bandwidth &&
+                 cpi->active_best_quality > cpi->oxcf.cq_level) {
+        // Severe undershoot and between auto and user cq level
+        force_recode = 1;
+        cpi->active_best_quality = cpi->oxcf.cq_level;
+      }
+    }
+  }
+
+  return force_recode;
+}
+
+static void update_reference_frames(VP9_COMP * const cpi) {
+  VP9_COMMON * const cm = &cpi->common;
+
+  // At this point the new frame has been encoded.
+  // If any buffer copy / swapping is signaled it should be done here.
+  if (cm->frame_type == KEY_FRAME) {
+    ref_cnt_fb(cm->fb_idx_ref_cnt,
+               &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
+    ref_cnt_fb(cm->fb_idx_ref_cnt,
+               &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
+  }
+#if CONFIG_MULTIPLE_ARF
+  else if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
+      !cpi->refresh_alt_ref_frame) {
+#else
+  else if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame) {
+#endif
+    /* Preserve the previously existing golden frame and update the frame in
+     * the alt ref slot instead. This is highly specific to the current use of
+     * alt-ref as a forward reference, and this needs to be generalized as
+     * other uses are implemented (like RTC/temporal scaling)
+     *
+     * The update to the buffer in the alt ref slot was signaled in
+     * vp9_pack_bitstream(), now swap the buffer pointers so that it's treated
+     * as the golden frame next time.
+     */
+    int tmp;
+
+    ref_cnt_fb(cm->fb_idx_ref_cnt,
+               &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
+
+    tmp = cpi->alt_fb_idx;
+    cpi->alt_fb_idx = cpi->gld_fb_idx;
+    cpi->gld_fb_idx = tmp;
+  }  else { /* For non key/golden frames */
+    if (cpi->refresh_alt_ref_frame) {
+      int arf_idx = cpi->alt_fb_idx;
+#if CONFIG_MULTIPLE_ARF
+      if (cpi->multi_arf_enabled) {
+        arf_idx = cpi->arf_buffer_idx[cpi->sequence_number + 1];
+      }
+#endif
+      ref_cnt_fb(cm->fb_idx_ref_cnt,
+                 &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
+    }
+
+    if (cpi->refresh_golden_frame) {
+      ref_cnt_fb(cm->fb_idx_ref_cnt,
+                 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
+    }
+  }
+
+  if (cpi->refresh_last_frame) {
+    ref_cnt_fb(cm->fb_idx_ref_cnt,
+               &cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
+  }
+}
+
+static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) {
+  if (cpi->mb.e_mbd.lossless) {
+    cm->filter_level = 0;
+  } else {
+    struct vpx_usec_timer timer;
+
+    vp9_clear_system_state();
+
+    vpx_usec_timer_start(&timer);
+
+    vp9_pick_filter_level(cpi->Source, cpi);
+
+    vpx_usec_timer_mark(&timer);
+    cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
+  }
+
+  if (cm->filter_level > 0) {
+    vp9_set_alt_lf_level(cpi, cm->filter_level);
+    vp9_loop_filter_frame(cm, &cpi->mb.e_mbd, cm->filter_level, 0);
+  }
+
+  vp9_extend_frame_borders(cm->frame_to_show,
+                           cm->subsampling_x, cm->subsampling_y);
+
+}
+
+void vp9_select_interp_filter_type(VP9_COMP *cpi) {
+  int i;
+  int high_filter_index = 0;
+  unsigned int thresh;
+  unsigned int high_count = 0;
+  unsigned int count_sum = 0;
+  unsigned int *hist = cpi->best_switchable_interp_count;
+
+  if (DEFAULT_INTERP_FILTER != SWITCHABLE) {
+    cpi->common.mcomp_filter_type = DEFAULT_INTERP_FILTER;
+    return;
+  }
+
+  // TODO(agrange): Look at using RD criteria to select the interpolation
+  // filter to use for the next frame rather than this simpler counting scheme.
+
+  // Select the interpolation filter mode for the next frame
+  // based on the selection frequency seen in the current frame.
+  for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
+    unsigned int count = hist[i];
+    count_sum += count;
+    if (count > high_count) {
+      high_count = count;
+      high_filter_index = i;
+    }
+  }
+
+  thresh = (unsigned int)(0.80 * count_sum);
+
+  if (high_count > thresh) {
+    // One filter accounts for 80+% of cases so force the next
+    // frame to use this filter exclusively using frame-level flag.
+    cpi->common.mcomp_filter_type = vp9_switchable_interp[high_filter_index];
+  } else {
+    // Use a MB-level switchable filter selection strategy.
+    cpi->common.mcomp_filter_type = SWITCHABLE;
+  }
+}
+
+static void scale_references(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  int i;
+
+  for (i = 0; i < 3; i++) {
+    YV12_BUFFER_CONFIG *ref = &cm->yv12_fb[cm->ref_frame_map[i]];
+
+    if (ref->y_crop_width != cm->width ||
+        ref->y_crop_height != cm->height) {
+      int new_fb = get_free_fb(cm);
+
+      vp9_realloc_frame_buffer(&cm->yv12_fb[new_fb],
+                               cm->width, cm->height,
+                               cm->subsampling_x, cm->subsampling_y,
+                               VP9BORDERINPIXELS);
+      scale_and_extend_frame(ref, &cm->yv12_fb[new_fb]);
+      cpi->scaled_ref_idx[i] = new_fb;
+    } else {
+      cpi->scaled_ref_idx[i] = cm->ref_frame_map[i];
+      cm->fb_idx_ref_cnt[cm->ref_frame_map[i]]++;
+    }
+  }
+}
+
+static void release_scaled_references(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  int i;
+
+  for (i = 0; i < 3; i++)
+    cm->fb_idx_ref_cnt[cpi->scaled_ref_idx[i]]--;
+}
+
+static void encode_frame_to_data_rate(VP9_COMP *cpi,
+                                      unsigned long *size,
+                                      unsigned char *dest,
+                                      unsigned int *frame_flags) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  TX_SIZE t;
+  int q;
+  int frame_over_shoot_limit;
+  int frame_under_shoot_limit;
+
+  int loop = 0;
+  int loop_count;
+
+  int q_low;
+  int q_high;
+
+  int top_index;
+  int bottom_index;
+  int active_worst_qchanged = 0;
+
+  int overshoot_seen = 0;
+  int undershoot_seen = 0;
+
+  SPEED_FEATURES *sf = &cpi->sf;
+#if RESET_FOREACH_FILTER
+  int q_low0;
+  int q_high0;
+  int Q0;
+  int active_best_quality0;
+  int active_worst_quality0;
+  double rate_correction_factor0;
+  double gf_rate_correction_factor0;
+#endif
+
+  /* list of filters to search over */
+  int mcomp_filters_to_search[] = {
+    EIGHTTAP, EIGHTTAP_SHARP, EIGHTTAP_SMOOTH, SWITCHABLE
+  };
+  int mcomp_filters = sizeof(mcomp_filters_to_search) /
+      sizeof(*mcomp_filters_to_search);
+  int mcomp_filter_index = 0;
+  int64_t mcomp_filter_cost[4];
+
+  /* Scale the source buffer, if required */
+  if (cm->mi_cols * 8 != cpi->un_scaled_source->y_width ||
+      cm->mi_rows * 8 != cpi->un_scaled_source->y_height) {
+    scale_and_extend_frame(cpi->un_scaled_source, &cpi->scaled_source);
+    cpi->Source = &cpi->scaled_source;
+  } else {
+    cpi->Source = cpi->un_scaled_source;
+  }
+
+  scale_references(cpi);
+
+  // Clear down mmx registers to allow floating point in what follows
+  vp9_clear_system_state();
+
+
+  // For an alt ref frame in 2 pass we skip the call to the second
+  // pass function that sets the target bandwidth so must set it here
+  if (cpi->refresh_alt_ref_frame) {
+    // Per frame bit target for the alt ref frame
+    cpi->per_frame_bandwidth = cpi->twopass.gf_bits;
+    // per second target bitrate
+    cpi->target_bandwidth = (int)(cpi->twopass.gf_bits *
+                                  cpi->output_frame_rate);
+  }
+
+  // Clear zbin over-quant value and mode boost values.
+  cpi->zbin_mode_boost = 0;
+
+  // Enable or disable mode based tweaking of the zbin
+  // For 2 Pass Only used where GF/ARF prediction quality
+  // is above a threshold
+  cpi->zbin_mode_boost = 0;
+
+  // if (cpi->oxcf.lossless)
+    cpi->zbin_mode_boost_enabled = 0;
+  // else
+  //   cpi->zbin_mode_boost_enabled = 1;
+
+  // Current default encoder behaviour for the altref sign bias
+    cpi->common.ref_frame_sign_bias[ALTREF_FRAME] = cpi->source_alt_ref_active;
+
+  // Check to see if a key frame is signaled
+  // For two pass with auto key frame enabled cm->frame_type may already be set, but not for one pass.
+  if ((cm->current_video_frame == 0) ||
+      (cm->frame_flags & FRAMEFLAGS_KEY) ||
+      (cpi->oxcf.auto_key && (cpi->frames_since_key % cpi->key_frame_frequency == 0))) {
+    // Key frame from VFW/auto-keyframe/first frame
+    cm->frame_type = KEY_FRAME;
+  }
+
+  // Set default state for segment based loop filter update flags
+  xd->mode_ref_lf_delta_update = 0;
+
+
+  // Set various flags etc to special state if it is a key frame
+  if (cm->frame_type == KEY_FRAME) {
+    int i;
+
+    // Reset the loop filter deltas and segmentation map
+    setup_features(cpi);
+
+    // If segmentation is enabled force a map update for key frames
+    if (xd->segmentation_enabled) {
+      xd->update_mb_segmentation_map = 1;
+      xd->update_mb_segmentation_data = 1;
+    }
+
+    // The alternate reference frame cannot be active for a key frame
+    cpi->source_alt_ref_active = 0;
+
+    // Reset the RD threshold multipliers to default of * 1 (128)
+    for (i = 0; i < MAX_MODES; i++)
+      cpi->rd_thresh_mult[i] = 128;
+
+    cm->error_resilient_mode = (cpi->oxcf.error_resilient_mode != 0);
+    cm->frame_parallel_decoding_mode =
+      (cpi->oxcf.frame_parallel_decoding_mode != 0);
+    if (cm->error_resilient_mode) {
+      cm->frame_parallel_decoding_mode = 1;
+      cm->reset_frame_context = 0;
+      cm->refresh_frame_context = 0;
+    }
+  }
+
+  // Configure experimental use of segmentation for enhanced coding of
+  // static regions if indicated.
+  // Only allowed for now in second pass of two pass (as requires lagged coding)
+  // and if the relevant speed feature flag is set.
+  if ((cpi->pass == 2) && (cpi->sf.static_segmentation)) {
+    configure_static_seg_features(cpi);
+  }
+
+  // Decide how big to make the frame
+  vp9_pick_frame_size(cpi);
+
+  vp9_clear_system_state();
+
+  // Set an active best quality and if necessary active worst quality
+  q = cpi->active_worst_quality;
+
+  if (cm->frame_type == KEY_FRAME) {
+#if !CONFIG_MULTIPLE_ARF
+    // Special case for key frames forced because we have reached
+    // the maximum key frame interval. Here force the Q to a range
+    // based on the ambient Q to reduce the risk of popping
+    if (cpi->this_key_frame_forced) {
+      int delta_qindex;
+      int qindex = cpi->last_boosted_qindex;
+      double last_boosted_q = vp9_convert_qindex_to_q(qindex);
+
+      delta_qindex = compute_qdelta(cpi, last_boosted_q,
+                                    (last_boosted_q * 0.75));
+
+      cpi->active_best_quality = MAX(qindex + delta_qindex, cpi->best_quality);
+    } else {
+      int high = 5000;
+      int low = 400;
+      double q_adj_factor = 1.0;
+      double q_val;
+
+      // Baseline value derived from cpi->active_worst_quality and kf boost
+      if (cpi->kf_boost > high) {
+        cpi->active_best_quality = kf_low_motion_minq[q];
+      } else if (cpi->kf_boost < low) {
+        cpi->active_best_quality = kf_high_motion_minq[q];
+      } else {
+        const int gap = high - low;
+        const int offset = high - cpi->kf_boost;
+        const int qdiff = kf_high_motion_minq[q] - kf_low_motion_minq[q];
+        const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
+
+        cpi->active_best_quality = kf_low_motion_minq[q] + adjustment;
+      }
+
+
+      // Allow somewhat lower kf minq with small image formats.
+      if ((cm->width * cm->height) <= (352 * 288)) {
+        q_adj_factor -= 0.25;
+      }
+
+      // Make a further adjustment based on the kf zero motion measure.
+      q_adj_factor += 0.05 - (0.001 * (double)cpi->kf_zeromotion_pct);
+
+      // Convert the adjustment factor to a qindex delta on active_best_quality.
+      q_val = vp9_convert_qindex_to_q(cpi->active_best_quality);
+      cpi->active_best_quality +=
+        compute_qdelta(cpi, q_val, (q_val * q_adj_factor));
+    }
+#else
+    double current_q;
+
+    // Force the KF quantizer to be 30% of the active_worst_quality.
+    current_q = vp9_convert_qindex_to_q(cpi->active_worst_quality);
+    cpi->active_best_quality = cpi->active_worst_quality
+        + compute_qdelta(cpi, current_q, current_q * 0.3);
+#endif
+  } else if (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame) {
+    int high = 2000;
+    int low = 400;
+
+    // Use the lower of cpi->active_worst_quality and recent
+    // average Q as basis for GF/ARF Q limit unless last frame was
+    // a key frame.
+    if (cpi->frames_since_key > 1 &&
+        cpi->avg_frame_qindex < cpi->active_worst_quality) {
+      q = cpi->avg_frame_qindex;
+    }
+
+    // For constrained quality dont allow Q less than the cq level
+    if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY &&
+        q < cpi->cq_target_quality) {
+      q = cpi->cq_target_quality;
+    }
+
+    if (cpi->gfu_boost > high) {
+      cpi->active_best_quality = gf_low_motion_minq[q];
+    } else if (cpi->gfu_boost < low) {
+      cpi->active_best_quality = gf_high_motion_minq[q];
+    } else {
+      const int gap = high - low;
+      const int offset = high - cpi->gfu_boost;
+      const int qdiff = gf_high_motion_minq[q] - gf_low_motion_minq[q];
+      const int adjustment = ((offset * qdiff) + (gap >> 1)) / gap;
+
+      cpi->active_best_quality = gf_low_motion_minq[q] + adjustment;
+    }
+
+    // Constrained quality use slightly lower active best.
+    if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY)
+      cpi->active_best_quality = cpi->active_best_quality * 15 / 16;
+  } else {
+#ifdef ONE_SHOT_Q_ESTIMATE
+#ifdef STRICT_ONE_SHOT_Q
+    cpi->active_best_quality = q;
+#else
+    cpi->active_best_quality = inter_minq[q];
+#endif
+#else
+    cpi->active_best_quality = inter_minq[q];
+#endif
+
+    // For the constant/constrained quality mode we don't want
+    // q to fall below the cq level.
+    if ((cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) &&
+        (cpi->active_best_quality < cpi->cq_target_quality)) {
+      // If we are strongly undershooting the target rate in the last
+      // frames then use the user passed in cq value not the auto
+      // cq value.
+      if (cpi->rolling_actual_bits < cpi->min_frame_bandwidth)
+        cpi->active_best_quality = cpi->oxcf.cq_level;
+      else
+        cpi->active_best_quality = cpi->cq_target_quality;
+    }
+  }
+
+  // Clip the active best and worst quality values to limits
+  if (cpi->active_worst_quality > cpi->worst_quality)
+    cpi->active_worst_quality = cpi->worst_quality;
+
+  if (cpi->active_best_quality < cpi->best_quality)
+    cpi->active_best_quality = cpi->best_quality;
+
+  if (cpi->active_best_quality > cpi->worst_quality)
+    cpi->active_best_quality = cpi->worst_quality;
+
+  if (cpi->active_worst_quality < cpi->active_best_quality)
+    cpi->active_worst_quality = cpi->active_best_quality;
+
+  // Special case code to try and match quality with forced key frames
+  if ((cm->frame_type == KEY_FRAME) && cpi->this_key_frame_forced) {
+    q = cpi->last_boosted_qindex;
+  } else {
+    // Determine initial Q to try
+    q = vp9_regulate_q(cpi, cpi->this_frame_target);
+  }
+
+  vp9_compute_frame_size_bounds(cpi, &frame_under_shoot_limit,
+                                &frame_over_shoot_limit);
+
+#if CONFIG_MULTIPLE_ARF
+  // Force the quantizer determined by the coding order pattern.
+  if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME)) {
+    double new_q;
+    double current_q = vp9_convert_qindex_to_q(cpi->active_worst_quality);
+    int level = cpi->this_frame_weight;
+    assert(level >= 0);
+
+    // Set quantizer steps at 10% increments.
+    new_q = current_q * (1.0 - (0.2 * (cpi->max_arf_level - level)));
+    q = cpi->active_worst_quality + compute_qdelta(cpi, current_q, new_q);
+
+    bottom_index = q;
+    top_index    = q;
+    q_low  = q;
+    q_high = q;
+
+    printf("frame:%d q:%d\n", cm->current_video_frame, q);
+  } else {
+#endif
+    // Limit Q range for the adaptive loop.
+    bottom_index = cpi->active_best_quality;
+    top_index    = cpi->active_worst_quality;
+    q_low  = cpi->active_best_quality;
+    q_high = cpi->active_worst_quality;
+#if CONFIG_MULTIPLE_ARF
+  }
+#endif
+  loop_count = 0;
+  vpx_memset(cpi->rd_tx_select_threshes, 0, sizeof(cpi->rd_tx_select_threshes));
+
+  if (cm->frame_type != KEY_FRAME) {
+    /* TODO: Decide this more intelligently */
+    if (sf->search_best_filter) {
+      cm->mcomp_filter_type = mcomp_filters_to_search[0];
+      mcomp_filter_index = 0;
+    } else {
+      cm->mcomp_filter_type = DEFAULT_INTERP_FILTER;
+    }
+    /* TODO: Decide this more intelligently */
+    xd->allow_high_precision_mv = q < HIGH_PRECISION_MV_QTHRESH;
+    set_mvcost(&cpi->mb);
+  }
+
+#if CONFIG_POSTPROC
+
+  if (cpi->oxcf.noise_sensitivity > 0) {
+    int l = 0;
+
+    switch (cpi->oxcf.noise_sensitivity) {
+      case 1:
+        l = 20;
+        break;
+      case 2:
+        l = 40;
+        break;
+      case 3:
+        l = 60;
+        break;
+      case 4:
+      case 5:
+        l = 100;
+        break;
+      case 6:
+        l = 150;
+        break;
+    }
+
+    vp9_denoise(cpi->Source, cpi->Source, l);
+  }
+
+#endif
+
+#ifdef OUTPUT_YUV_SRC
+  vp9_write_yuv_frame(cpi->Source);
+#endif
+
+#if RESET_FOREACH_FILTER
+  if (sf->search_best_filter) {
+    q_low0 = q_low;
+    q_high0 = q_high;
+    Q0 = Q;
+    rate_correction_factor0 = cpi->rate_correction_factor;
+    gf_rate_correction_factor0 = cpi->gf_rate_correction_factor;
+    active_best_quality0 = cpi->active_best_quality;
+    active_worst_quality0 = cpi->active_worst_quality;
+  }
+#endif
+  do {
+    vp9_clear_system_state();  // __asm emms;
+
+    vp9_set_quantizer(cpi, q);
+
+    if (loop_count == 0) {
+
+      // Set up entropy depending on frame type.
+      if (cm->frame_type == KEY_FRAME) {
+        /* Choose which entropy context to use. When using a forward reference
+         * frame, it immediately follows the keyframe, and thus benefits from
+         * using the same entropy context established by the keyframe.
+         *  Otherwise, use the default context 0.
+         */
+        cm->frame_context_idx = cpi->oxcf.play_alternate;
+        vp9_setup_key_frame(cpi);
+      } else {
+        /* Choose which entropy context to use. Currently there are only two
+         * contexts used, one for normal frames and one for alt ref frames.
+         */
+        cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
+        vp9_setup_inter_frame(cpi);
+      }
+    }
+
+    // transform / motion compensation build reconstruction frame
+
+    vp9_encode_frame(cpi);
+
+    // Update the skip mb flag probabilities based on the distribution
+    // seen in the last encoder iteration.
+    // update_base_skip_probs(cpi);
+
+    vp9_clear_system_state();  // __asm emms;
+
+    // Dummy pack of the bitstream using up to date stats to get an
+    // accurate estimate of output frame size to determine if we need
+    // to recode.
+    vp9_save_coding_context(cpi);
+    cpi->dummy_packing = 1;
+    vp9_pack_bitstream(cpi, dest, size);
+    cpi->projected_frame_size = (*size) << 3;
+    vp9_restore_coding_context(cpi);
+
+    if (frame_over_shoot_limit == 0)
+      frame_over_shoot_limit = 1;
+    active_worst_qchanged = 0;
+
+    // Special case handling for forced key frames
+    if ((cm->frame_type == KEY_FRAME) && cpi->this_key_frame_forced) {
+      int last_q = q;
+      int kf_err = vp9_calc_ss_err(cpi->Source,
+                                   &cm->yv12_fb[cm->new_fb_idx]);
+
+      int high_err_target = cpi->ambient_err;
+      int low_err_target = cpi->ambient_err >> 1;
+
+      // Prevent possible divide by zero error below for perfect KF
+      kf_err += !kf_err;
+
+      // The key frame is not good enough or we can afford
+      // to make it better without undue risk of popping.
+      if ((kf_err > high_err_target &&
+           cpi->projected_frame_size <= frame_over_shoot_limit) ||
+          (kf_err > low_err_target &&
+           cpi->projected_frame_size <= frame_under_shoot_limit)) {
+        // Lower q_high
+        q_high = q > q_low ? q - 1 : q_low;
+
+        // Adjust Q
+        q = (q * high_err_target) / kf_err;
+        q = MIN(q, (q_high + q_low) >> 1);
+      } else if (kf_err < low_err_target &&
+                cpi->projected_frame_size >= frame_under_shoot_limit) {
+        // The key frame is much better than the previous frame
+        // Raise q_low
+        q_low = q < q_high ? q + 1 : q_high;
+
+        // Adjust Q
+        q = (q * low_err_target) / kf_err;
+        q = MIN(q, (q_high + q_low + 1) >> 1);
+      }
+
+      // Clamp Q to upper and lower limits:
+      q = clamp(q, q_low, q_high);
+
+      loop = q != last_q;
+    }
+
+    // Is the projected frame size out of range and are we allowed to attempt to recode.
+    else if (recode_loop_test(cpi,
+                              frame_over_shoot_limit, frame_under_shoot_limit,
+                              q, top_index, bottom_index)) {
+      int last_q = q;
+      int retries = 0;
+
+      // Frame size out of permitted range:
+      // Update correction factor & compute new Q to try...
+
+      // Frame is too large
+      if (cpi->projected_frame_size > cpi->this_frame_target) {
+        // Raise Qlow as to at least the current value
+        q_low = q < q_high ? q + 1 : q_high;
+
+        if (undershoot_seen || loop_count > 1) {
+          // Update rate_correction_factor unless cpi->active_worst_quality
+          // has changed.
+          if (!active_worst_qchanged)
+            vp9_update_rate_correction_factors(cpi, 1);
+
+          q = (q_high + q_low + 1) / 2;
+        } else {
+          // Update rate_correction_factor unless cpi->active_worst_quality has changed.
+          if (!active_worst_qchanged)
+            vp9_update_rate_correction_factors(cpi, 0);
+
+          q = vp9_regulate_q(cpi, cpi->this_frame_target);
+
+          while (q < q_low && retries < 10) {
+            vp9_update_rate_correction_factors(cpi, 0);
+            q = vp9_regulate_q(cpi, cpi->this_frame_target);
+            retries++;
+          }
+        }
+
+        overshoot_seen = 1;
+      } else {
+        // Frame is too small
+        q_high = q > q_low ? q - 1 : q_low;
+
+        if (overshoot_seen || loop_count > 1) {
+          // Update rate_correction_factor unless cpi->active_worst_quality has changed.
+          if (!active_worst_qchanged)
+            vp9_update_rate_correction_factors(cpi, 1);
+
+          q = (q_high + q_low) / 2;
+        } else {
+          // Update rate_correction_factor unless cpi->active_worst_quality has changed.
+          if (!active_worst_qchanged)
+            vp9_update_rate_correction_factors(cpi, 0);
+
+          q = vp9_regulate_q(cpi, cpi->this_frame_target);
+
+          // Special case reset for qlow for constrained quality.
+          // This should only trigger where there is very substantial
+          // undershoot on a frame and the auto cq level is above
+          // the user passsed in value.
+          if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY && q < q_low) {
+            q_low = q;
+          }
+
+          while (q > q_high && retries < 10) {
+            vp9_update_rate_correction_factors(cpi, 0);
+            q = vp9_regulate_q(cpi, cpi->this_frame_target);
+            retries++;
+          }
+        }
+
+        undershoot_seen = 1;
+      }
+
+      // Clamp Q to upper and lower limits:
+      q = clamp(q, q_low, q_high);
+
+      loop = q != last_q;
+    } else {
+      loop = 0;
+    }
+
+    if (cpi->is_src_frame_alt_ref)
+      loop = 0;
+
+    if (!loop && cm->frame_type != KEY_FRAME && sf->search_best_filter) {
+      if (mcomp_filter_index < mcomp_filters) {
+        int64_t err = vp9_calc_ss_err(cpi->Source,
+                                    &cm->yv12_fb[cm->new_fb_idx]);
+        int64_t rate = cpi->projected_frame_size << 8;
+        mcomp_filter_cost[mcomp_filter_index] =
+          (RDCOST(cpi->RDMULT, cpi->RDDIV, rate, err));
+        mcomp_filter_index++;
+        if (mcomp_filter_index < mcomp_filters) {
+          cm->mcomp_filter_type = mcomp_filters_to_search[mcomp_filter_index];
+          loop_count = -1;
+          loop = 1;
+        } else {
+          int f;
+          int64_t best_cost = mcomp_filter_cost[0];
+          int mcomp_best_filter = mcomp_filters_to_search[0];
+          for (f = 1; f < mcomp_filters; f++) {
+            if (mcomp_filter_cost[f] < best_cost) {
+              mcomp_best_filter = mcomp_filters_to_search[f];
+              best_cost = mcomp_filter_cost[f];
+            }
+          }
+          if (mcomp_best_filter != mcomp_filters_to_search[mcomp_filters - 1]) {
+            loop_count = -1;
+            loop = 1;
+            cm->mcomp_filter_type = mcomp_best_filter;
+          }
+          /*
+          printf("  best filter = %d, ( ", mcomp_best_filter);
+          for (f=0;f<mcomp_filters; f++) printf("%d ",  mcomp_filter_cost[f]);
+          printf(")\n");
+          */
+        }
+#if RESET_FOREACH_FILTER
+        if (loop) {
+          overshoot_seen = 0;
+          undershoot_seen = 0;
+          q_low = q_low0;
+          q_high = q_high0;
+          q = Q0;
+          cpi->rate_correction_factor = rate_correction_factor0;
+          cpi->gf_rate_correction_factor = gf_rate_correction_factor0;
+          cpi->active_best_quality = active_best_quality0;
+          cpi->active_worst_quality = active_worst_quality0;
+        }
+#endif
+      }
+    }
+
+    if (loop) {
+      loop_count++;
+
+#if CONFIG_INTERNAL_STATS
+      cpi->tot_recode_hits++;
+#endif
+    }
+  } while (loop);
+
+  // Special case code to reduce pulsing when key frames are forced at a
+  // fixed interval. Note the reconstruction error if it is the frame before
+  // the force key frame
+  if (cpi->next_key_frame_forced && (cpi->twopass.frames_to_key == 0)) {
+    cpi->ambient_err = vp9_calc_ss_err(cpi->Source,
+                                       &cm->yv12_fb[cm->new_fb_idx]);
+  }
+
+  if (cm->frame_type == KEY_FRAME)
+    cpi->refresh_last_frame = 1;
+
+  cm->frame_to_show = &cm->yv12_fb[cm->new_fb_idx];
+
+#if WRITE_RECON_BUFFER
+  if (cm->show_frame)
+    write_cx_frame_to_file(cm->frame_to_show,
+                           cm->current_video_frame);
+  else
+    write_cx_frame_to_file(cm->frame_to_show,
+                           cm->current_video_frame + 1000);
+#endif
+
+  // Pick the loop filter level for the frame.
+  loopfilter_frame(cpi, cm);
+
+#if WRITE_RECON_BUFFER
+  if (cm->show_frame)
+    write_cx_frame_to_file(cm->frame_to_show,
+                           cm->current_video_frame + 2000);
+  else
+    write_cx_frame_to_file(cm->frame_to_show,
+                           cm->current_video_frame + 3000);
+#endif
+
+  // build the bitstream
+  cpi->dummy_packing = 0;
+  vp9_pack_bitstream(cpi, dest, size);
+
+  if (xd->update_mb_segmentation_map) {
+    update_reference_segmentation_map(cpi);
+  }
+
+  release_scaled_references(cpi);
+  update_reference_frames(cpi);
+
+  for (t = TX_4X4; t <= TX_32X32; t++)
+    vp9_full_to_model_counts(cpi->common.fc.coef_counts[t],
+                             cpi->coef_counts[t]);
+  if (!cpi->common.error_resilient_mode &&
+      !cpi->common.frame_parallel_decoding_mode) {
+    vp9_adapt_coef_probs(&cpi->common);
+  }
+
+  if (cpi->common.frame_type != KEY_FRAME) {
+    vp9_copy(cpi->common.fc.y_mode_counts, cpi->y_mode_count);
+    vp9_copy(cpi->common.fc.uv_mode_counts, cpi->y_uv_mode_count);
+    vp9_copy(cpi->common.fc.partition_counts, cpi->partition_count);
+    vp9_copy(cm->fc.intra_inter_count, cpi->intra_inter_count);
+    vp9_copy(cm->fc.comp_inter_count, cpi->comp_inter_count);
+    vp9_copy(cm->fc.single_ref_count, cpi->single_ref_count);
+    vp9_copy(cm->fc.comp_ref_count, cpi->comp_ref_count);
+    cpi->common.fc.NMVcount = cpi->NMVcount;
+    if (!cpi->common.error_resilient_mode &&
+        !cpi->common.frame_parallel_decoding_mode) {
+      vp9_adapt_mode_probs(&cpi->common);
+      vp9_adapt_mode_context(&cpi->common);
+      vp9_adapt_nmv_probs(&cpi->common, cpi->mb.e_mbd.allow_high_precision_mv);
+    }
+  }
+
+#ifdef ENTROPY_STATS
+  vp9_update_mode_context_stats(cpi);
+#endif
+
+  /* Move storing frame_type out of the above loop since it is also
+   * needed in motion search besides loopfilter */
+  cm->last_frame_type = cm->frame_type;
+
+  // Update rate control heuristics
+  cpi->total_byte_count += (*size);
+  cpi->projected_frame_size = (*size) << 3;
+
+  if (!active_worst_qchanged)
+    vp9_update_rate_correction_factors(cpi, 2);
+
+  cpi->last_q[cm->frame_type] = cm->base_qindex;
+
+  // Keep record of last boosted (KF/KF/ARF) Q value.
+  // If the current frame is coded at a lower Q then we also update it.
+  // If all mbs in this group are skipped only update if the Q value is
+  // better than that already stored.
+  // This is used to help set quality in forced key frames to reduce popping
+  if ((cm->base_qindex < cpi->last_boosted_qindex) ||
+      ((cpi->static_mb_pct < 100) &&
+       ((cm->frame_type == KEY_FRAME) ||
+        cpi->refresh_alt_ref_frame ||
+        (cpi->refresh_golden_frame && !cpi->is_src_frame_alt_ref)))) {
+    cpi->last_boosted_qindex = cm->base_qindex;
+  }
+
+  if (cm->frame_type == KEY_FRAME) {
+    vp9_adjust_key_frame_context(cpi);
+  }
+
+  // Keep a record of ambient average Q.
+  if (cm->frame_type != KEY_FRAME)
+    cpi->avg_frame_qindex = (2 + 3 * cpi->avg_frame_qindex + cm->base_qindex) >> 2;
+
+  // Keep a record from which we can calculate the average Q excluding GF updates and key frames
+  if (cm->frame_type != KEY_FRAME &&
+      !cpi->refresh_golden_frame &&
+      !cpi->refresh_alt_ref_frame) {
+    cpi->ni_frames++;
+    cpi->tot_q += vp9_convert_qindex_to_q(q);
+    cpi->avg_q = cpi->tot_q / (double)cpi->ni_frames;
+
+    // Calculate the average Q for normal inter frames (not key or GFU frames).
+    cpi->ni_tot_qi += q;
+    cpi->ni_av_qi = cpi->ni_tot_qi / cpi->ni_frames;
+  }
+
+  // Update the buffer level variable.
+  // Non-viewable frames are a special case and are treated as pure overhead.
+  if (!cm->show_frame)
+    cpi->bits_off_target -= cpi->projected_frame_size;
+  else
+    cpi->bits_off_target += cpi->av_per_frame_bandwidth - cpi->projected_frame_size;
+
+  // Clip the buffer level at the maximum buffer size
+  if (cpi->bits_off_target > cpi->oxcf.maximum_buffer_size)
+    cpi->bits_off_target = cpi->oxcf.maximum_buffer_size;
+
+  // Rolling monitors of whether we are over or underspending used to help
+  // regulate min and Max Q in two pass.
+  if (cm->frame_type != KEY_FRAME) {
+    cpi->rolling_target_bits =
+      ((cpi->rolling_target_bits * 3) + cpi->this_frame_target + 2) / 4;
+    cpi->rolling_actual_bits =
+      ((cpi->rolling_actual_bits * 3) + cpi->projected_frame_size + 2) / 4;
+    cpi->long_rolling_target_bits =
+      ((cpi->long_rolling_target_bits * 31) + cpi->this_frame_target + 16) / 32;
+    cpi->long_rolling_actual_bits =
+      ((cpi->long_rolling_actual_bits * 31) +
+       cpi->projected_frame_size + 16) / 32;
+  }
+
+  // Actual bits spent
+  cpi->total_actual_bits += cpi->projected_frame_size;
+
+  // Debug stats
+  cpi->total_target_vs_actual += (cpi->this_frame_target - cpi->projected_frame_size);
+
+  cpi->buffer_level = cpi->bits_off_target;
+
+  // Update bits left to the kf and gf groups to account for overshoot or undershoot on these frames
+  if (cm->frame_type == KEY_FRAME) {
+    cpi->twopass.kf_group_bits += cpi->this_frame_target - cpi->projected_frame_size;
+
+    cpi->twopass.kf_group_bits = MAX(cpi->twopass.kf_group_bits, 0);
+  } else if (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame) {
+    cpi->twopass.gf_group_bits += cpi->this_frame_target - cpi->projected_frame_size;
+
+    cpi->twopass.gf_group_bits = MAX(cpi->twopass.gf_group_bits, 0);
+  }
+
+  // Update the skip mb flag probabilities based on the distribution seen
+  // in this frame.
+  // update_base_skip_probs(cpi);
+
+#if 0 && CONFIG_INTERNAL_STATS
+  {
+    FILE *f = fopen("tmp.stt", "a");
+    int recon_err;
+
+    vp9_clear_system_state();  // __asm emms;
+
+    recon_err = vp9_calc_ss_err(cpi->Source,
+                                &cm->yv12_fb[cm->new_fb_idx]);
+
+    if (cpi->twopass.total_left_stats.coded_error != 0.0)
+      fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d"
+              "%7.2f %7.2f %7.2f %7.2f %7.2f %7.2f %7.2f"
+              "%6d %6d %5d %5d %5d %8.2f %10d %10.3f"
+              "%10.3f %8d %10d %10d %10d\n",
+              cpi->common.current_video_frame, cpi->this_frame_target,
+              cpi->projected_frame_size, 0, //loop_size_estimate,
+              (cpi->projected_frame_size - cpi->this_frame_target),
+              (int)cpi->total_target_vs_actual,
+              (int)(cpi->oxcf.starting_buffer_level - cpi->bits_off_target),
+              (int)cpi->total_actual_bits,
+              vp9_convert_qindex_to_q(cm->base_qindex),
+              (double)vp9_dc_quant(cm->base_qindex, 0) / 4.0,
+              vp9_convert_qindex_to_q(cpi->active_best_quality),
+              vp9_convert_qindex_to_q(cpi->active_worst_quality),
+              cpi->avg_q,
+              vp9_convert_qindex_to_q(cpi->ni_av_qi),
+              vp9_convert_qindex_to_q(cpi->cq_target_quality),
+              cpi->refresh_last_frame,
+              cpi->refresh_golden_frame, cpi->refresh_alt_ref_frame,
+              cm->frame_type, cpi->gfu_boost,
+              cpi->twopass.est_max_qcorrection_factor,
+              (int)cpi->twopass.bits_left,
+              cpi->twopass.total_left_stats.coded_error,
+              (double)cpi->twopass.bits_left /
+              cpi->twopass.total_left_stats.coded_error,
+              cpi->tot_recode_hits, recon_err, cpi->kf_boost,
+              cpi->kf_zeromotion_pct);
+    else
+      fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d"
+              "%7.2f %7.2f %7.2f %7.2f %7.2f %7.2f %7.2f"
+              "%5d %5d %5d %8d %8d %8.2f %10d %10.3f"
+              "%8d %10d %10d %10d\n",
+              cpi->common.current_video_frame,
+              cpi->this_frame_target, cpi->projected_frame_size,
+              0, //loop_size_estimate,
+              (cpi->projected_frame_size - cpi->this_frame_target),
+              (int)cpi->total_target_vs_actual,
+              (int)(cpi->oxcf.starting_buffer_level - cpi->bits_off_target),
+              (int)cpi->total_actual_bits,
+              vp9_convert_qindex_to_q(cm->base_qindex),
+              (double)vp9_dc_quant(cm->base_qindex, 0) / 4.0,
+              vp9_convert_qindex_to_q(cpi->active_best_quality),
+              vp9_convert_qindex_to_q(cpi->active_worst_quality),
+              cpi->avg_q,
+              vp9_convert_qindex_to_q(cpi->ni_av_qi),
+              vp9_convert_qindex_to_q(cpi->cq_target_quality),
+              cpi->refresh_last_frame,
+              cpi->refresh_golden_frame, cpi->refresh_alt_ref_frame,
+              cm->frame_type, cpi->gfu_boost,
+              cpi->twopass.est_max_qcorrection_factor,
+              (int)cpi->twopass.bits_left,
+              cpi->twopass.total_left_stats.coded_error,
+              cpi->tot_recode_hits, recon_err, cpi->kf_boost,
+              cpi->kf_zeromotion_pct);
+
+    fclose(f);
+
+    if (0) {
+      FILE *fmodes = fopen("Modes.stt", "a");
+      int i;
+
+      fprintf(fmodes, "%6d:%1d:%1d:%1d ",
+              cpi->common.current_video_frame,
+              cm->frame_type, cpi->refresh_golden_frame,
+              cpi->refresh_alt_ref_frame);
+
+      for (i = 0; i < MAX_MODES; i++)
+        fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
+
+      fprintf(fmodes, "\n");
+
+      fclose(fmodes);
+    }
+  }
+
+#endif
+
+#if 0
+  // Debug stats for segment feature experiments.
+  print_seg_map(cpi);
+#endif
+
+  // If this was a kf or Gf note the Q
+  if ((cm->frame_type == KEY_FRAME)
+      || cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)
+    cm->last_kf_gf_q = cm->base_qindex;
+
+  if (cpi->refresh_golden_frame == 1)
+    cm->frame_flags = cm->frame_flags | FRAMEFLAGS_GOLDEN;
+  else
+    cm->frame_flags = cm->frame_flags&~FRAMEFLAGS_GOLDEN;
+
+  if (cpi->refresh_alt_ref_frame == 1)
+    cm->frame_flags = cm->frame_flags | FRAMEFLAGS_ALTREF;
+  else
+    cm->frame_flags = cm->frame_flags&~FRAMEFLAGS_ALTREF;
+
+
+  if (cpi->refresh_last_frame & cpi->refresh_golden_frame)
+    cpi->gold_is_last = 1;
+  else if (cpi->refresh_last_frame ^ cpi->refresh_golden_frame)
+    cpi->gold_is_last = 0;
+
+  if (cpi->refresh_last_frame & cpi->refresh_alt_ref_frame)
+    cpi->alt_is_last = 1;
+  else if (cpi->refresh_last_frame ^ cpi->refresh_alt_ref_frame)
+    cpi->alt_is_last = 0;
+
+  if (cpi->refresh_alt_ref_frame & cpi->refresh_golden_frame)
+    cpi->gold_is_alt = 1;
+  else if (cpi->refresh_alt_ref_frame ^ cpi->refresh_golden_frame)
+    cpi->gold_is_alt = 0;
+
+  cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
+
+  if (cpi->gold_is_last)
+    cpi->ref_frame_flags &= ~VP9_GOLD_FLAG;
+
+  if (cpi->alt_is_last)
+    cpi->ref_frame_flags &= ~VP9_ALT_FLAG;
+
+  if (cpi->gold_is_alt)
+    cpi->ref_frame_flags &= ~VP9_ALT_FLAG;
+
+  if (cpi->oxcf.play_alternate && cpi->refresh_alt_ref_frame
+      && (cm->frame_type != KEY_FRAME))
+    // Update the alternate reference frame stats as appropriate.
+    update_alt_ref_frame_stats(cpi);
+  else
+    // Update the Golden frame stats as appropriate.
+    update_golden_frame_stats(cpi);
+
+  if (cm->frame_type == KEY_FRAME) {
+    // Tell the caller that the frame was coded as a key frame
+    *frame_flags = cm->frame_flags | FRAMEFLAGS_KEY;
+
+#if CONFIG_MULTIPLE_ARF
+    // Reset the sequence number.
+    if (cpi->multi_arf_enabled) {
+      cpi->sequence_number = 0;
+      cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
+      cpi->new_frame_coding_order_period = -1;
+    }
+#endif
+
+    // As this frame is a key frame the next defaults to an inter frame.
+    cm->frame_type = INTER_FRAME;
+  } else {
+    *frame_flags = cm->frame_flags&~FRAMEFLAGS_KEY;
+
+#if CONFIG_MULTIPLE_ARF
+    /* Increment position in the coded frame sequence. */
+    if (cpi->multi_arf_enabled) {
+      ++cpi->sequence_number;
+      if (cpi->sequence_number >= cpi->frame_coding_order_period) {
+        cpi->sequence_number = 0;
+        cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
+        cpi->new_frame_coding_order_period = -1;
+      }
+      cpi->this_frame_weight = cpi->arf_weight[cpi->sequence_number];
+      assert(cpi->this_frame_weight >= 0);
+    }
+#endif
+  }
+
+  // Clear the one shot update flags for segmentation map and mode/ref loop filter deltas.
+  xd->update_mb_segmentation_map = 0;
+  xd->update_mb_segmentation_data = 0;
+  xd->mode_ref_lf_delta_update = 0;
+
+  // keep track of the last coded dimensions
+  cm->last_width = cm->width;
+  cm->last_height = cm->height;
+
+  // Don't increment frame counters if this was an altref buffer
+  // update not a real frame
+  cm->last_show_frame = cm->show_frame;
+  if (cm->show_frame) {
+    ++cm->current_video_frame;
+    ++cpi->frames_since_key;
+  }
+
+  // reset to normal state now that we are done.
+
+#if 0
+  {
+    char filename[512];
+    FILE *recon_file;
+    sprintf(filename, "enc%04d.yuv", (int) cm->current_video_frame);
+    recon_file = fopen(filename, "wb");
+    fwrite(cm->yv12_fb[cm->ref_frame_map[cpi->lst_fb_idx]].buffer_alloc,
+           cm->yv12_fb[cm->ref_frame_map[cpi->lst_fb_idx]].frame_size,
+           1, recon_file);
+    fclose(recon_file);
+  }
+#endif
+#ifdef OUTPUT_YUV_REC
+  vp9_write_yuv_rec_frame(cm);
+#endif
+
+  if (cm->show_frame) {
+    vpx_memcpy(cm->prev_mip, cm->mip,
+               cm->mode_info_stride * (cm->mi_rows + 64 / MI_SIZE) *
+               sizeof(MODE_INFO));
+  } else {
+    vpx_memset(cm->prev_mip, 0,
+               cm->mode_info_stride * (cm->mi_rows + 64 / MI_SIZE) *
+               sizeof(MODE_INFO));
+  }
+  // restore prev_mi
+  cm->prev_mi = cm->prev_mip + cm->mode_info_stride + 1;
+}
+
+static void Pass2Encode(VP9_COMP *cpi, unsigned long *size,
+                        unsigned char *dest, unsigned int *frame_flags) {
+
+  if (!cpi->refresh_alt_ref_frame)
+    vp9_second_pass(cpi);
+
+  encode_frame_to_data_rate(cpi, size, dest, frame_flags);
+
+#ifdef DISABLE_RC_LONG_TERM_MEM
+  cpi->twopass.bits_left -=  cpi->this_frame_target;
+#else
+  cpi->twopass.bits_left -= 8 * *size;
+#endif
+
+  if (!cpi->refresh_alt_ref_frame) {
+    double lower_bounds_min_rate = FRAME_OVERHEAD_BITS * cpi->oxcf.frame_rate;
+    double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth
+                                        * cpi->oxcf.two_pass_vbrmin_section / 100);
+
+    if (two_pass_min_rate < lower_bounds_min_rate)
+      two_pass_min_rate = lower_bounds_min_rate;
+
+    cpi->twopass.bits_left += (int64_t)(two_pass_min_rate / cpi->oxcf.frame_rate);
+  }
+}
+
+
+int vp9_receive_raw_frame(VP9_PTR ptr, unsigned int frame_flags,
+                          YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
+                          int64_t end_time) {
+  VP9_COMP              *cpi = (VP9_COMP *) ptr;
+  VP9_COMMON            *cm = &cpi->common;
+  struct vpx_usec_timer  timer;
+  int                    res = 0;
+
+  if (!cpi->initial_width) {
+    // TODO(jkoleszar): Support 1/4 subsampling?
+    cm->subsampling_x = sd->uv_width < sd->y_width;
+    cm->subsampling_y = sd->uv_height < sd->y_height;
+    alloc_raw_frame_buffers(cpi);
+
+    cpi->initial_width = cm->width;
+    cpi->initial_height = cm->height;
+  }
+  vpx_usec_timer_start(&timer);
+  if (vp9_lookahead_push(cpi->lookahead, sd, time_stamp, end_time, frame_flags,
+                         cpi->active_map_enabled ? cpi->active_map : NULL))
+    res = -1;
+  cm->clr_type = sd->clrtype;
+  vpx_usec_timer_mark(&timer);
+  cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
+
+  return res;
+}
+
+
+static int frame_is_reference(const VP9_COMP *cpi) {
+  const VP9_COMMON *cm = &cpi->common;
+  const MACROBLOCKD *mb = &cpi->mb.e_mbd;
+
+  return cm->frame_type == KEY_FRAME ||
+         cpi->refresh_last_frame ||
+         cpi->refresh_golden_frame ||
+         cpi->refresh_alt_ref_frame ||
+         cm->refresh_frame_context ||
+         mb->mode_ref_lf_delta_update ||
+         mb->update_mb_segmentation_map ||
+         mb->update_mb_segmentation_data;
+}
+
+#if CONFIG_MULTIPLE_ARF
+int is_next_frame_arf(VP9_COMP *cpi) {
+  // Negative entry in frame_coding_order indicates an ARF at this position.
+  return cpi->frame_coding_order[cpi->sequence_number + 1] < 0 ? 1 : 0;
+}
+#endif
+
+int vp9_get_compressed_data(VP9_PTR ptr, unsigned int *frame_flags,
+                            unsigned long *size, unsigned char *dest,
+                            int64_t *time_stamp, int64_t *time_end, int flush) {
+  VP9_COMP *cpi = (VP9_COMP *) ptr;
+  VP9_COMMON *cm = &cpi->common;
+  struct vpx_usec_timer  cmptimer;
+  YV12_BUFFER_CONFIG    *force_src_buffer = NULL;
+  int i;
+  // FILE *fp_out = fopen("enc_frame_type.txt", "a");
+
+  if (!cpi)
+    return -1;
+
+  vpx_usec_timer_start(&cmptimer);
+
+  cpi->source = NULL;
+
+  cpi->mb.e_mbd.allow_high_precision_mv = ALTREF_HIGH_PRECISION_MV;
+  set_mvcost(&cpi->mb);
+
+  // Should we code an alternate reference frame.
+  if (cpi->oxcf.play_alternate && cpi->source_alt_ref_pending) {
+    int frames_to_arf;
+
+#if CONFIG_MULTIPLE_ARF
+    assert(!cpi->multi_arf_enabled ||
+           cpi->frame_coding_order[cpi->sequence_number] < 0);
+
+    if (cpi->multi_arf_enabled && (cpi->pass == 2))
+      frames_to_arf = (-cpi->frame_coding_order[cpi->sequence_number])
+        - cpi->next_frame_in_order;
+    else
+#endif
+      frames_to_arf = cpi->frames_till_gf_update_due;
+
+    assert(frames_to_arf < cpi->twopass.frames_to_key);
+
+    if ((cpi->source = vp9_lookahead_peek(cpi->lookahead, frames_to_arf))) {
+#if CONFIG_MULTIPLE_ARF
+      cpi->alt_ref_source[cpi->arf_buffered] = cpi->source;
+#else
+      cpi->alt_ref_source = cpi->source;
+#endif
+
+      if (cpi->oxcf.arnr_max_frames > 0) {
+        // Produce the filtered ARF frame.
+        // TODO(agrange) merge these two functions.
+        configure_arnr_filter(cpi, cm->current_video_frame + frames_to_arf,
+                              cpi->gfu_boost);
+        vp9_temporal_filter_prepare(cpi, frames_to_arf);
+        force_src_buffer = &cpi->alt_ref_buffer;
+      }
+
+      cm->show_frame = 0;
+      cm->intra_only = 0;
+      cpi->refresh_alt_ref_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_last_frame = 0;
+      cpi->is_src_frame_alt_ref = 0;
+
+      // TODO(agrange) This needs to vary depending on where the next ARF is.
+      cm->frames_till_alt_ref_frame = frames_to_arf;
+
+#if CONFIG_MULTIPLE_ARF
+      if (!cpi->multi_arf_enabled)
+#endif
+        cpi->source_alt_ref_pending = 0;   // Clear Pending altf Ref flag.
+    }
+  }
+
+  if (!cpi->source) {
+#if CONFIG_MULTIPLE_ARF
+    int i;
+#endif
+    if ((cpi->source = vp9_lookahead_pop(cpi->lookahead, flush))) {
+      cm->show_frame = 1;
+
+#if CONFIG_MULTIPLE_ARF
+      // Is this frame the ARF overlay.
+      cpi->is_src_frame_alt_ref = 0;
+      for (i = 0; i < cpi->arf_buffered; ++i) {
+        if (cpi->source == cpi->alt_ref_source[i]) {
+          cpi->is_src_frame_alt_ref = 1;
+          cpi->refresh_golden_frame = 1;
+          break;
+        }
+      }
+#else
+      cpi->is_src_frame_alt_ref = cpi->alt_ref_source
+                                  && (cpi->source == cpi->alt_ref_source);
+#endif
+      if (cpi->is_src_frame_alt_ref) {
+        // Current frame is an ARF overlay frame.
+#if CONFIG_MULTIPLE_ARF
+        cpi->alt_ref_source[i] = NULL;
+#else
+        cpi->alt_ref_source = NULL;
+#endif
+        // Don't refresh the last buffer for an ARF overlay frame. It will
+        // become the GF so preserve last as an alternative prediction option.
+        cpi->refresh_last_frame = 0;
+      }
+#if CONFIG_MULTIPLE_ARF
+      ++cpi->next_frame_in_order;
+#endif
+    }
+  }
+
+  if (cpi->source) {
+    cpi->un_scaled_source = cpi->Source = force_src_buffer ? force_src_buffer
+                                                           : &cpi->source->img;
+    *time_stamp = cpi->source->ts_start;
+    *time_end = cpi->source->ts_end;
+    *frame_flags = cpi->source->flags;
+
+    // fprintf(fp_out, "   Frame:%d", cm->current_video_frame);
+#if CONFIG_MULTIPLE_ARF
+    if (cpi->multi_arf_enabled) {
+      // fprintf(fp_out, "   seq_no:%d  this_frame_weight:%d",
+      //         cpi->sequence_number, cpi->this_frame_weight);
+    } else {
+      // fprintf(fp_out, "\n");
+    }
+#else
+    // fprintf(fp_out, "\n");
+#endif
+
+#if CONFIG_MULTIPLE_ARF
+    if ((cm->frame_type != KEY_FRAME) && (cpi->pass == 2))
+      cpi->source_alt_ref_pending = is_next_frame_arf(cpi);
+#endif
+  } else {
+    *size = 0;
+    if (flush && cpi->pass == 1 && !cpi->twopass.first_pass_done) {
+      vp9_end_first_pass(cpi);    /* get last stats packet */
+      cpi->twopass.first_pass_done = 1;
+    }
+
+    // fclose(fp_out);
+    return -1;
+  }
+
+  if (cpi->source->ts_start < cpi->first_time_stamp_ever) {
+    cpi->first_time_stamp_ever = cpi->source->ts_start;
+    cpi->last_end_time_stamp_seen = cpi->source->ts_start;
+  }
+
+  // adjust frame rates based on timestamps given
+  if (!cpi->refresh_alt_ref_frame) {
+    int64_t this_duration;
+    int step = 0;
+
+    if (cpi->source->ts_start == cpi->first_time_stamp_ever) {
+      this_duration = cpi->source->ts_end - cpi->source->ts_start;
+      step = 1;
+    } else {
+      int64_t last_duration = cpi->last_end_time_stamp_seen
+                                - cpi->last_time_stamp_seen;
+
+      this_duration = cpi->source->ts_end - cpi->last_end_time_stamp_seen;
+
+      // do a step update if the duration changes by 10%
+      if (last_duration)
+        step = (int)((this_duration - last_duration) * 10 / last_duration);
+    }
+
+    if (this_duration) {
+      if (step) {
+        vp9_new_frame_rate(cpi, 10000000.0 / this_duration);
+      } else {
+        // Average this frame's rate into the last second's average
+        // frame rate. If we haven't seen 1 second yet, then average
+        // over the whole interval seen.
+        const double interval = MIN((double)(cpi->source->ts_end
+                                     - cpi->first_time_stamp_ever), 10000000.0);
+        double avg_duration = 10000000.0 / cpi->oxcf.frame_rate;
+        avg_duration *= (interval - avg_duration + this_duration);
+        avg_duration /= interval;
+
+        vp9_new_frame_rate(cpi, 10000000.0 / avg_duration);
+      }
+    }
+
+    cpi->last_time_stamp_seen = cpi->source->ts_start;
+    cpi->last_end_time_stamp_seen = cpi->source->ts_end;
+  }
+
+  // start with a 0 size frame
+  *size = 0;
+
+  // Clear down mmx registers
+  vp9_clear_system_state();  // __asm emms;
+
+  /* find a free buffer for the new frame, releasing the reference previously
+   * held.
+   */
+  cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
+  cm->new_fb_idx = get_free_fb(cm);
+
+#if CONFIG_MULTIPLE_ARF
+  /* Set up the correct ARF frame. */
+  if (cpi->refresh_alt_ref_frame) {
+    ++cpi->arf_buffered;
+  }
+  if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME) &&
+      (cpi->pass == 2)) {
+    cpi->alt_fb_idx = cpi->arf_buffer_idx[cpi->sequence_number];
+  }
+#endif
+
+  /* Get the mapping of L/G/A to the reference buffer pool */
+  cm->active_ref_idx[0] = cm->ref_frame_map[cpi->lst_fb_idx];
+  cm->active_ref_idx[1] = cm->ref_frame_map[cpi->gld_fb_idx];
+  cm->active_ref_idx[2] = cm->ref_frame_map[cpi->alt_fb_idx];
+
+#if 0  // CONFIG_MULTIPLE_ARF
+  if (cpi->multi_arf_enabled) {
+    fprintf(fp_out, "      idx(%d, %d, %d, %d) active(%d, %d, %d)",
+        cpi->lst_fb_idx, cpi->gld_fb_idx, cpi->alt_fb_idx, cm->new_fb_idx,
+        cm->active_ref_idx[0], cm->active_ref_idx[1], cm->active_ref_idx[2]);
+    if (cpi->refresh_alt_ref_frame)
+      fprintf(fp_out, "  type:ARF");
+    if (cpi->is_src_frame_alt_ref)
+      fprintf(fp_out, "  type:OVERLAY[%d]", cpi->alt_fb_idx);
+    fprintf(fp_out, "\n");
+  }
+#endif
+
+  cm->frame_type = INTER_FRAME;
+  cm->frame_flags = *frame_flags;
+
+  // Reset the frame pointers to the current frame size
+  vp9_realloc_frame_buffer(&cm->yv12_fb[cm->new_fb_idx],
+                           cm->width, cm->height,
+                           cm->subsampling_x, cm->subsampling_y,
+                           VP9BORDERINPIXELS);
+
+  // Calculate scaling factors for each of the 3 available references
+  for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i)
+    vp9_setup_scale_factors(cm, i);
+
+  vp9_setup_interp_filters(&cpi->mb.e_mbd, DEFAULT_INTERP_FILTER, cm);
+
+  if (cpi->pass == 1) {
+    Pass1Encode(cpi, size, dest, frame_flags);
+  } else if (cpi->pass == 2) {
+    Pass2Encode(cpi, size, dest, frame_flags);
+  } else {
+    encode_frame_to_data_rate(cpi, size, dest, frame_flags);
+  }
+
+  if (cm->refresh_frame_context)
+    cm->frame_contexts[cm->frame_context_idx] = cm->fc;
+
+  if (*size > 0) {
+    // if its a dropped frame honor the requests on subsequent frames
+    cpi->droppable = !frame_is_reference(cpi);
+
+    // return to normal state
+    cm->reset_frame_context = 0;
+    cm->refresh_frame_context = 1;
+    cpi->refresh_alt_ref_frame = 0;
+    cpi->refresh_golden_frame = 0;
+    cpi->refresh_last_frame = 1;
+    cm->frame_type = INTER_FRAME;
+  }
+
+  vpx_usec_timer_mark(&cmptimer);
+  cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
+
+  if (cpi->b_calculate_psnr && cpi->pass != 1 && cm->show_frame)
+    generate_psnr_packet(cpi);
+
+#if CONFIG_INTERNAL_STATS
+
+  if (cpi->pass != 1) {
+    cpi->bytes += *size;
+
+    if (cm->show_frame) {
+
+      cpi->count++;
+
+      if (cpi->b_calculate_psnr) {
+        double ye, ue, ve;
+        double frame_psnr;
+        YV12_BUFFER_CONFIG      *orig = cpi->Source;
+        YV12_BUFFER_CONFIG      *recon = cpi->common.frame_to_show;
+        YV12_BUFFER_CONFIG      *pp = &cm->post_proc_buffer;
+        int y_samples = orig->y_height * orig->y_width;
+        int uv_samples = orig->uv_height * orig->uv_width;
+        int t_samples = y_samples + 2 * uv_samples;
+        double sq_error;
+
+        ye = (double)calc_plane_error(orig->y_buffer, orig->y_stride,
+                              recon->y_buffer, recon->y_stride, orig->y_width,
+                              orig->y_height);
+
+        ue = (double)calc_plane_error(orig->u_buffer, orig->uv_stride,
+                              recon->u_buffer, recon->uv_stride, orig->uv_width,
+                              orig->uv_height);
+
+        ve = (double)calc_plane_error(orig->v_buffer, orig->uv_stride,
+                              recon->v_buffer, recon->uv_stride, orig->uv_width,
+                              orig->uv_height);
+
+        sq_error = ye + ue + ve;
+
+        frame_psnr = vp9_mse2psnr(t_samples, 255.0, sq_error);
+
+        cpi->total_y += vp9_mse2psnr(y_samples, 255.0, ye);
+        cpi->total_u += vp9_mse2psnr(uv_samples, 255.0, ue);
+        cpi->total_v += vp9_mse2psnr(uv_samples, 255.0, ve);
+        cpi->total_sq_error += sq_error;
+        cpi->total  += frame_psnr;
+        {
+          double frame_psnr2, frame_ssim2 = 0;
+          double weight = 0;
+#if CONFIG_POSTPROC
+          vp9_deblock(cm->frame_to_show, &cm->post_proc_buffer,
+                      cm->filter_level * 10 / 6);
+#endif
+          vp9_clear_system_state();
+
+          ye = (double)calc_plane_error(orig->y_buffer, orig->y_stride,
+                                pp->y_buffer, pp->y_stride, orig->y_width,
+                                orig->y_height);
+
+          ue = (double)calc_plane_error(orig->u_buffer, orig->uv_stride,
+                                pp->u_buffer, pp->uv_stride, orig->uv_width,
+                                orig->uv_height);
+
+          ve = (double)calc_plane_error(orig->v_buffer, orig->uv_stride,
+                                pp->v_buffer, pp->uv_stride, orig->uv_width,
+                                orig->uv_height);
+
+          sq_error = ye + ue + ve;
+
+          frame_psnr2 = vp9_mse2psnr(t_samples, 255.0, sq_error);
+
+          cpi->totalp_y += vp9_mse2psnr(y_samples, 255.0, ye);
+          cpi->totalp_u += vp9_mse2psnr(uv_samples, 255.0, ue);
+          cpi->totalp_v += vp9_mse2psnr(uv_samples, 255.0, ve);
+          cpi->total_sq_error2 += sq_error;
+          cpi->totalp  += frame_psnr2;
+
+          frame_ssim2 = vp9_calc_ssim(cpi->Source,
+                                      recon, 1, &weight);
+
+          cpi->summed_quality += frame_ssim2 * weight;
+          cpi->summed_weights += weight;
+
+          frame_ssim2 = vp9_calc_ssim(cpi->Source,
+                                      &cm->post_proc_buffer, 1, &weight);
+
+          cpi->summedp_quality += frame_ssim2 * weight;
+          cpi->summedp_weights += weight;
+#if 0
+          {
+            FILE *f = fopen("q_used.stt", "a");
+            fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
+                    cpi->common.current_video_frame, y2, u2, v2,
+                    frame_psnr2, frame_ssim2);
+            fclose(f);
+          }
+#endif
+        }
+      }
+
+      if (cpi->b_calculate_ssimg) {
+        double y, u, v, frame_all;
+        frame_all =  vp9_calc_ssimg(cpi->Source, cm->frame_to_show,
+                                    &y, &u, &v);
+        cpi->total_ssimg_y += y;
+        cpi->total_ssimg_u += u;
+        cpi->total_ssimg_v += v;
+        cpi->total_ssimg_all += frame_all;
+      }
+    }
+  }
+
+#endif
+  // fclose(fp_out);
+  return 0;
+}
+
+int vp9_get_preview_raw_frame(VP9_PTR comp, YV12_BUFFER_CONFIG *dest,
+                              vp9_ppflags_t *flags) {
+  VP9_COMP *cpi = (VP9_COMP *) comp;
+
+  if (!cpi->common.show_frame)
+    return -1;
+  else {
+    int ret;
+#if CONFIG_POSTPROC
+    ret = vp9_post_proc_frame(&cpi->common, dest, flags);
+#else
+
+    if (cpi->common.frame_to_show) {
+      *dest = *cpi->common.frame_to_show;
+      dest->y_width = cpi->common.width;
+      dest->y_height = cpi->common.height;
+      dest->uv_height = cpi->common.height / 2;
+      ret = 0;
+    } else {
+      ret = -1;
+    }
+
+#endif // !CONFIG_POSTPROC
+    vp9_clear_system_state();
+    return ret;
+  }
+}
+
+int vp9_set_roimap(VP9_PTR comp, unsigned char *map, unsigned int rows,
+                   unsigned int cols, int delta_q[MAX_MB_SEGMENTS],
+                   int delta_lf[MAX_MB_SEGMENTS],
+                   unsigned int threshold[MAX_MB_SEGMENTS]) {
+  VP9_COMP *cpi = (VP9_COMP *) comp;
+  signed char feature_data[SEG_LVL_MAX][MAX_MB_SEGMENTS];
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  int i;
+
+  if (cpi->common.mb_rows != rows || cpi->common.mb_cols != cols)
+    return -1;
+
+  if (!map) {
+    vp9_disable_segmentation((VP9_PTR)cpi);
+    return 0;
+  }
+
+  // Set the segmentation Map
+  vp9_set_segmentation_map((VP9_PTR)cpi, map);
+
+  // Activate segmentation.
+  vp9_enable_segmentation((VP9_PTR)cpi);
+
+  // Set up the quan, LF and breakout threshold segment data
+  for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+    feature_data[SEG_LVL_ALT_Q][i] = delta_q[i];
+    feature_data[SEG_LVL_ALT_LF][i] = delta_lf[i];
+    cpi->segment_encode_breakout[i] = threshold[i];
+  }
+
+  // Enable the loop and quant changes in the feature mask
+  for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+    if (delta_q[i])
+      vp9_enable_segfeature(xd, i, SEG_LVL_ALT_Q);
+    else
+      vp9_disable_segfeature(xd, i, SEG_LVL_ALT_Q);
+
+    if (delta_lf[i])
+      vp9_enable_segfeature(xd, i, SEG_LVL_ALT_LF);
+    else
+      vp9_disable_segfeature(xd, i, SEG_LVL_ALT_LF);
+  }
+
+  // Initialise the feature data structure
+  // SEGMENT_DELTADATA    0, SEGMENT_ABSDATA      1
+  vp9_set_segment_data((VP9_PTR)cpi, &feature_data[0][0], SEGMENT_DELTADATA);
+
+  return 0;
+}
+
+int vp9_set_active_map(VP9_PTR comp, unsigned char *map,
+                       unsigned int rows, unsigned int cols) {
+  VP9_COMP *cpi = (VP9_COMP *) comp;
+
+  if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) {
+    if (map) {
+      vpx_memcpy(cpi->active_map, map, rows * cols);
+      cpi->active_map_enabled = 1;
+    } else {
+      cpi->active_map_enabled = 0;
+    }
+
+    return 0;
+  } else {
+    // cpi->active_map_enabled = 0;
+    return -1;
+  }
+}
+
+int vp9_set_internal_size(VP9_PTR comp,
+                          VPX_SCALING horiz_mode, VPX_SCALING vert_mode) {
+  VP9_COMP *cpi = (VP9_COMP *) comp;
+  VP9_COMMON *cm = &cpi->common;
+  int hr = 0, hs = 0, vr = 0, vs = 0;
+
+  if (horiz_mode > ONETWO || vert_mode > ONETWO)
+    return -1;
+
+  Scale2Ratio(horiz_mode, &hr, &hs);
+  Scale2Ratio(vert_mode, &vr, &vs);
+
+  // always go to the next whole number
+  cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
+  cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
+
+  assert(cm->width <= cpi->initial_width);
+  assert(cm->height <= cpi->initial_height);
+  update_frame_size(cpi);
+  return 0;
+}
+
+
+
+int vp9_calc_ss_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest) {
+  int i, j;
+  int total = 0;
+
+  uint8_t *src = source->y_buffer;
+  uint8_t *dst = dest->y_buffer;
+
+  // Loop through the Y plane raw and reconstruction data summing
+  // (square differences)
+  for (i = 0; i < source->y_height; i += 16) {
+    for (j = 0; j < source->y_width; j += 16) {
+      unsigned int sse;
+      total += vp9_mse16x16(src + j, source->y_stride, dst + j, dest->y_stride,
+                            &sse);
+    }
+
+    src += 16 * source->y_stride;
+    dst += 16 * dest->y_stride;
+  }
+
+  return total;
+}
+
+
+int vp9_get_quantizer(VP9_PTR c) {
+  return ((VP9_COMP *)c)->common.base_qindex;
+}
diff --git a/libvpx/vp9/encoder/vp9_onyx_int.h b/libvpx/vp9/encoder/vp9_onyx_int.h
new file mode 100644
index 0000000..f5f1c07
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_onyx_int.h
@@ -0,0 +1,637 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_ONYX_INT_H_
+#define VP9_ENCODER_VP9_ONYX_INT_H_
+
+#include <stdio.h>
+#include "./vpx_config.h"
+#include "vp9/common/vp9_onyx.h"
+#include "vp9/encoder/vp9_treewriter.h"
+#include "vp9/encoder/vp9_tokenize.h"
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vpx_ports/mem.h"
+#include "vpx/internal/vpx_codec_internal.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/encoder/vp9_lookahead.h"
+
+// Experimental rate control switches
+#if CONFIG_ONESHOTQ
+#define ONE_SHOT_Q_ESTIMATE 0
+#define STRICT_ONE_SHOT_Q 0
+#define DISABLE_RC_LONG_TERM_MEM 0
+#endif
+
+// #define SPEEDSTATS 1
+#if CONFIG_MULTIPLE_ARF
+// Set MIN_GF_INTERVAL to 1 for the full decomposition.
+#define MIN_GF_INTERVAL             2
+#else
+#define MIN_GF_INTERVAL             4
+#endif
+#define DEFAULT_GF_INTERVAL         7
+
+#define KEY_FRAME_CONTEXT 5
+
+#define MAX_MODES 36
+
+#define MIN_THRESHMULT  32
+#define MAX_THRESHMULT  512
+
+#define GF_ZEROMV_ZBIN_BOOST 0
+#define LF_ZEROMV_ZBIN_BOOST 0
+#define MV_ZBIN_BOOST        0
+#define SPLIT_MV_ZBIN_BOOST  0
+#define INTRA_ZBIN_BOOST     0
+
+typedef struct {
+  nmv_context nmvc;
+  int nmvjointcost[MV_JOINTS];
+  int nmvcosts[2][MV_VALS];
+  int nmvcosts_hp[2][MV_VALS];
+
+  vp9_prob segment_pred_probs[PREDICTION_PROBS];
+  vp9_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
+  vp9_prob comp_inter_prob[COMP_INTER_CONTEXTS];
+  vp9_prob single_ref_prob[REF_CONTEXTS][2];
+  vp9_prob comp_ref_prob[REF_CONTEXTS];
+
+  unsigned char *last_frame_seg_map_copy;
+
+  // 0 = Intra, Last, GF, ARF
+  signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
+  // 0 = ZERO_MV, MV
+  signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
+
+  vp9_coeff_probs_model coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
+
+  vp9_prob y_mode_prob[4][VP9_INTRA_MODES - 1];
+  vp9_prob uv_mode_prob[VP9_INTRA_MODES][VP9_INTRA_MODES - 1];
+  vp9_prob partition_prob[2][NUM_PARTITION_CONTEXTS][PARTITION_TYPES - 1];
+
+  vp9_prob switchable_interp_prob[VP9_SWITCHABLE_FILTERS + 1]
+                                 [VP9_SWITCHABLE_FILTERS - 1];
+
+  int inter_mode_counts[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1][2];
+  vp9_prob inter_mode_probs[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1];
+
+  vp9_prob tx_probs_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 3];
+  vp9_prob tx_probs_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
+  vp9_prob tx_probs_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
+  vp9_prob mbskip_probs[MBSKIP_CONTEXTS];
+} CODING_CONTEXT;
+
+typedef struct {
+  double frame;
+  double intra_error;
+  double coded_error;
+  double sr_coded_error;
+  double ssim_weighted_pred_err;
+  double pcnt_inter;
+  double pcnt_motion;
+  double pcnt_second_ref;
+  double pcnt_neutral;
+  double MVr;
+  double mvr_abs;
+  double MVc;
+  double mvc_abs;
+  double MVrv;
+  double MVcv;
+  double mv_in_out_count;
+  double new_mv_count;
+  double duration;
+  double count;
+} FIRSTPASS_STATS;
+
+typedef struct {
+  int frames_so_far;
+  double frame_intra_error;
+  double frame_coded_error;
+  double frame_pcnt_inter;
+  double frame_pcnt_motion;
+  double frame_mvr;
+  double frame_mvr_abs;
+  double frame_mvc;
+  double frame_mvc_abs;
+} ONEPASS_FRAMESTATS;
+
+typedef struct {
+  struct {
+    int err;
+    union {
+      int_mv mv;
+      MB_PREDICTION_MODE mode;
+    } m;
+  } ref[MAX_REF_FRAMES];
+} MBGRAPH_MB_STATS;
+
+typedef struct {
+  MBGRAPH_MB_STATS *mb_stats;
+} MBGRAPH_FRAME_STATS;
+
+typedef enum {
+  THR_ZEROMV,
+  THR_DC,
+
+  THR_NEARESTMV,
+  THR_NEARMV,
+
+  THR_ZEROG,
+  THR_NEARESTG,
+
+  THR_ZEROA,
+  THR_NEARESTA,
+
+  THR_NEARG,
+  THR_NEARA,
+
+  THR_V_PRED,
+  THR_H_PRED,
+  THR_D45_PRED,
+  THR_D135_PRED,
+  THR_D117_PRED,
+  THR_D153_PRED,
+  THR_D27_PRED,
+  THR_D63_PRED,
+  THR_TM,
+
+  THR_NEWMV,
+  THR_NEWG,
+  THR_NEWA,
+
+  THR_SPLITMV,
+  THR_SPLITG,
+  THR_SPLITA,
+
+  THR_B_PRED,
+
+  THR_COMP_ZEROLA,
+  THR_COMP_NEARESTLA,
+  THR_COMP_NEARLA,
+
+  THR_COMP_ZEROGA,
+  THR_COMP_NEARESTGA,
+  THR_COMP_NEARGA,
+
+  THR_COMP_NEWLA,
+  THR_COMP_NEWGA,
+
+  THR_COMP_SPLITLA,
+  THR_COMP_SPLITGA,
+} THR_MODES;
+
+typedef enum {
+  DIAMOND = 0,
+  NSTEP = 1,
+  HEX = 2
+} SEARCH_METHODS;
+
+typedef struct {
+  int RD;
+  SEARCH_METHODS search_method;
+  int auto_filter;
+  int recode_loop;
+  int iterative_sub_pixel;
+  int half_pixel_search;
+  int quarter_pixel_search;
+  int thresh_mult[MAX_MODES];
+  int max_step_search_steps;
+  int first_step;
+  int optimize_coefficients;
+  int search_best_filter;
+  int static_segmentation;
+  int comp_inter_joint_search_thresh;
+  int adpative_rd_thresh;
+} SPEED_FEATURES;
+
+enum BlockSize {
+  BLOCK_4X4,
+  BLOCK_4X8,
+  BLOCK_8X4,
+  BLOCK_8X8,
+  BLOCK_8X16,
+  BLOCK_16X8,
+  BLOCK_16X16,
+  BLOCK_32X32,
+  BLOCK_32X16,
+  BLOCK_16X32,
+  BLOCK_64X32,
+  BLOCK_32X64,
+  BLOCK_64X64,
+  BLOCK_MAX_SB_SEGMENTS,
+};
+
+typedef struct VP9_COMP {
+
+  DECLARE_ALIGNED(16, short, y_quant[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, unsigned char, y_quant_shift[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, y_zbin[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, y_round[QINDEX_RANGE][16]);
+
+  DECLARE_ALIGNED(16, short, uv_quant[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, unsigned char, uv_quant_shift[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, uv_zbin[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, uv_round[QINDEX_RANGE][16]);
+
+#if CONFIG_ALPHA
+  DECLARE_ALIGNED(16, short, a_quant[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, unsigned char, a_quant_shift[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, a_zbin[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, a_round[QINDEX_RANGE][16]);
+
+  DECLARE_ALIGNED(16, short, zrun_zbin_boost_a[QINDEX_RANGE][16]);
+#endif
+  DECLARE_ALIGNED(16, short, zrun_zbin_boost_y[QINDEX_RANGE][16]);
+  DECLARE_ALIGNED(16, short, zrun_zbin_boost_uv[QINDEX_RANGE][16]);
+
+  MACROBLOCK mb;
+  VP9_COMMON common;
+  VP9_CONFIG oxcf;
+
+  struct lookahead_ctx    *lookahead;
+  struct lookahead_entry  *source;
+#if CONFIG_MULTIPLE_ARF
+  struct lookahead_entry  *alt_ref_source[NUM_REF_FRAMES];
+#else
+  struct lookahead_entry  *alt_ref_source;
+#endif
+
+  YV12_BUFFER_CONFIG *Source;
+  YV12_BUFFER_CONFIG *un_scaled_source;
+  YV12_BUFFER_CONFIG scaled_source;
+
+  int source_alt_ref_pending; // frame in src_buffers has been identified to be encoded as an alt ref
+  int source_alt_ref_active;  // an alt ref frame has been encoded and is usable
+
+  int is_src_frame_alt_ref;   // source of frame to encode is an exact copy of an alt ref frame
+
+  int gold_is_last; // golden frame same as last frame ( short circuit gold searches)
+  int alt_is_last;  // Alt reference frame same as last ( short circuit altref search)
+  int gold_is_alt;  // don't do both alt and gold search ( just do gold).
+
+  int scaled_ref_idx[3];
+  int lst_fb_idx;
+  int gld_fb_idx;
+  int alt_fb_idx;
+#if CONFIG_MULTIPLE_ARF
+  int alt_ref_fb_idx[NUM_REF_FRAMES - 3];
+#endif
+  int refresh_last_frame;
+  int refresh_golden_frame;
+  int refresh_alt_ref_frame;
+  YV12_BUFFER_CONFIG last_frame_uf;
+
+  TOKENEXTRA *tok;
+  unsigned int tok_count[4][1 << 6];
+
+
+  unsigned int frames_since_key;
+  unsigned int key_frame_frequency;
+  unsigned int this_key_frame_forced;
+  unsigned int next_key_frame_forced;
+#if CONFIG_MULTIPLE_ARF
+  // Position within a frame coding order (including any additional ARF frames).
+  unsigned int sequence_number;
+  // Next frame in naturally occurring order that has not yet been coded.
+  int next_frame_in_order;
+#endif
+
+  // Ambient reconstruction err target for force key frames
+  int ambient_err;
+
+  unsigned int mode_check_freq[MAX_MODES];
+  unsigned int mode_test_hit_counts[MAX_MODES];
+  unsigned int mode_chosen_counts[MAX_MODES];
+
+  int rd_thresh_mult[MAX_MODES];
+  int rd_baseline_thresh[BLOCK_SIZE_TYPES][MAX_MODES];
+  int rd_threshes[BLOCK_SIZE_TYPES][MAX_MODES];
+  int rd_thresh_freq_fact[BLOCK_SIZE_TYPES][MAX_MODES];
+
+  int64_t rd_comp_pred_diff[NB_PREDICTION_TYPES];
+  int rd_prediction_type_threshes[4][NB_PREDICTION_TYPES];
+  unsigned int intra_inter_count[INTRA_INTER_CONTEXTS][2];
+  unsigned int comp_inter_count[COMP_INTER_CONTEXTS][2];
+  unsigned int single_ref_count[REF_CONTEXTS][2][2];
+  unsigned int comp_ref_count[REF_CONTEXTS][2];
+
+  // FIXME contextualize
+
+  int64_t rd_tx_select_diff[NB_TXFM_MODES];
+  int rd_tx_select_threshes[4][NB_TXFM_MODES];
+
+  int RDMULT;
+  int RDDIV;
+
+  CODING_CONTEXT coding_context;
+
+  // Rate targetting variables
+  int this_frame_target;
+  int projected_frame_size;
+  int last_q[2];                   // Separate values for Intra/Inter
+  int last_boosted_qindex;         // Last boosted GF/KF/ARF q
+
+  double rate_correction_factor;
+  double key_frame_rate_correction_factor;
+  double gf_rate_correction_factor;
+
+  int frames_till_gf_update_due;      // Count down till next GF
+
+  int gf_overspend_bits;            // Total bits overspent becasue of GF boost (cumulative)
+
+  int non_gf_bitrate_adjustment;     // Used in the few frames following a GF to recover the extra bits spent in that GF
+
+  int kf_overspend_bits;            // Extra bits spent on key frames that need to be recovered on inter frames
+  int kf_bitrate_adjustment;        // Current number of bit s to try and recover on each inter frame.
+  int max_gf_interval;
+  int baseline_gf_interval;
+  int active_arnr_frames;           // <= cpi->oxcf.arnr_max_frames
+  int active_arnr_strength;         // <= cpi->oxcf.arnr_max_strength
+
+  int64_t key_frame_count;
+  int prior_key_frame_distance[KEY_FRAME_CONTEXT];
+  int per_frame_bandwidth;          // Current section per frame bandwidth target
+  int av_per_frame_bandwidth;        // Average frame size target for clip
+  int min_frame_bandwidth;          // Minimum allocation that should be used for any frame
+  int inter_frame_target;
+  double output_frame_rate;
+  int64_t last_time_stamp_seen;
+  int64_t last_end_time_stamp_seen;
+  int64_t first_time_stamp_ever;
+
+  int ni_av_qi;
+  int ni_tot_qi;
+  int ni_frames;
+  int avg_frame_qindex;
+  double tot_q;
+  double avg_q;
+
+  int zbin_mode_boost;
+  int zbin_mode_boost_enabled;
+
+  int64_t total_byte_count;
+
+  int buffered_mode;
+
+  int buffer_level;
+  int bits_off_target;
+
+  int rolling_target_bits;
+  int rolling_actual_bits;
+
+  int long_rolling_target_bits;
+  int long_rolling_actual_bits;
+
+  int64_t total_actual_bits;
+  int total_target_vs_actual;        // debug stats
+
+  int worst_quality;
+  int active_worst_quality;
+  int best_quality;
+  int active_best_quality;
+
+  int cq_target_quality;
+
+  int y_mode_count[4][VP9_INTRA_MODES];
+  int y_uv_mode_count[VP9_INTRA_MODES][VP9_INTRA_MODES];
+  unsigned int partition_count[NUM_PARTITION_CONTEXTS][PARTITION_TYPES];
+
+  nmv_context_counts NMVcount;
+
+  vp9_coeff_count coef_counts[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  vp9_coeff_probs_model frame_coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
+  vp9_coeff_stats frame_branch_ct[TX_SIZE_MAX_SB][BLOCK_TYPES];
+
+  int gfu_boost;
+  int last_boost;
+  int kf_boost;
+  int kf_zeromotion_pct;
+
+  int64_t target_bandwidth;
+  struct vpx_codec_pkt_list  *output_pkt_list;
+
+#if 0
+  // Experimental code for lagged and one pass
+  ONEPASS_FRAMESTATS one_pass_frame_stats[MAX_LAG_BUFFERS];
+  int one_pass_frame_index;
+#endif
+  MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
+  int mbgraph_n_frames;             // number of frames filled in the above
+  int static_mb_pct;                // % forced skip mbs by segmentation
+  int seg0_progress, seg0_idx, seg0_cnt;
+
+  int decimation_factor;
+  int decimation_count;
+
+  // for real time encoding
+  int avg_encode_time;              // microsecond
+  int avg_pick_mode_time;            // microsecond
+  int speed;
+  unsigned int cpu_freq;           // Mhz
+  int compressor_speed;
+
+  int interquantizer;
+  int goldfreq;
+  int auto_worst_q;
+  int cpu_used;
+  int pass;
+
+  vp9_prob last_skip_false_probs[3][MBSKIP_CONTEXTS];
+  int last_skip_probs_q[3];
+
+  int ref_frame_flags;
+
+  SPEED_FEATURES sf;
+  int error_bins[1024];
+
+  // Data used for real time conferencing mode to help determine if it would be good to update the gf
+  int inter_zz_count;
+  int gf_bad_count;
+  int gf_update_recommended;
+
+  unsigned char *segmentation_map;
+
+  // segment threashold for encode breakout
+  int  segment_encode_breakout[MAX_MB_SEGMENTS];
+
+  unsigned char *active_map;
+  unsigned int active_map_enabled;
+
+  fractional_mv_step_fp *find_fractional_mv_step;
+  vp9_full_search_fn_t full_search_sad;
+  vp9_refining_search_fn_t refining_search_sad;
+  vp9_diamond_search_fn_t diamond_search_sad;
+  vp9_variance_fn_ptr_t fn_ptr[BLOCK_MAX_SB_SEGMENTS];
+  uint64_t time_receive_data;
+  uint64_t time_compress_data;
+  uint64_t time_pick_lpf;
+  uint64_t time_encode_mb_row;
+
+  struct twopass_rc {
+    unsigned int section_intra_rating;
+    unsigned int next_iiratio;
+    unsigned int this_iiratio;
+    FIRSTPASS_STATS total_stats;
+    FIRSTPASS_STATS this_frame_stats;
+    FIRSTPASS_STATS *stats_in, *stats_in_end, *stats_in_start;
+    FIRSTPASS_STATS total_left_stats;
+    int first_pass_done;
+    int64_t bits_left;
+    int64_t clip_bits_total;
+    double avg_iiratio;
+    double modified_error_total;
+    double modified_error_used;
+    double modified_error_left;
+    double kf_intra_err_min;
+    double gf_intra_err_min;
+    int frames_to_key;
+    int maxq_max_limit;
+    int maxq_min_limit;
+    int static_scene_max_gf_interval;
+    int kf_bits;
+    // Remaining error from uncoded frames in a gf group. Two pass use only
+    int64_t gf_group_error_left;
+
+    // Projected total bits available for a key frame group of frames
+    int64_t kf_group_bits;
+
+    // Error score of frames still to be coded in kf group
+    int64_t kf_group_error_left;
+
+    // Projected Bits available for a group of frames including 1 GF or ARF
+    int64_t gf_group_bits;
+    // Bits for the golden frame or ARF - 2 pass only
+    int gf_bits;
+    int alt_extra_bits;
+
+    int sr_update_lag;
+    double est_max_qcorrection_factor;
+  } twopass;
+
+  YV12_BUFFER_CONFIG alt_ref_buffer;
+  YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
+  int fixed_divide[512];
+
+#if CONFIG_INTERNAL_STATS
+  int    count;
+  double total_y;
+  double total_u;
+  double total_v;
+  double total;
+  double total_sq_error;
+  double totalp_y;
+  double totalp_u;
+  double totalp_v;
+  double totalp;
+  double total_sq_error2;
+  int    bytes;
+  double summed_quality;
+  double summed_weights;
+  double summedp_quality;
+  double summedp_weights;
+  unsigned int tot_recode_hits;
+
+
+  double total_ssimg_y;
+  double total_ssimg_u;
+  double total_ssimg_v;
+  double total_ssimg_all;
+
+  int b_calculate_ssimg;
+#endif
+  int b_calculate_psnr;
+
+  // Per MB activity measurement
+  unsigned int activity_avg;
+  unsigned int *mb_activity_map;
+  int *mb_norm_activity_map;
+  int output_partition;
+
+  /* force next frame to intra when kf_auto says so */
+  int force_next_frame_intra;
+
+  int droppable;
+
+  int dummy_packing;    /* flag to indicate if packing is dummy */
+
+  unsigned int switchable_interp_count[VP9_SWITCHABLE_FILTERS + 1]
+                                      [VP9_SWITCHABLE_FILTERS];
+  unsigned int best_switchable_interp_count[VP9_SWITCHABLE_FILTERS];
+
+  int initial_width;
+  int initial_height;
+
+#if CONFIG_MULTIPLE_ARF
+  // ARF tracking variables.
+  int multi_arf_enabled;
+  unsigned int frame_coding_order_period;
+  unsigned int new_frame_coding_order_period;
+  int frame_coding_order[MAX_LAG_BUFFERS * 2];
+  int arf_buffer_idx[MAX_LAG_BUFFERS * 3 / 2];
+  int arf_weight[MAX_LAG_BUFFERS];
+  int arf_buffered;
+  int this_frame_weight;
+  int max_arf_level;
+#endif
+
+#ifdef ENTROPY_STATS
+  int64_t mv_ref_stats[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1][2];
+#endif
+} VP9_COMP;
+
+static int get_ref_frame_idx(VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
+  if (ref_frame == LAST_FRAME) {
+    return cpi->lst_fb_idx;
+  } else if (ref_frame == GOLDEN_FRAME) {
+    return cpi->gld_fb_idx;
+  } else {
+    return cpi->alt_fb_idx;
+  }
+}
+
+void vp9_encode_frame(VP9_COMP *cpi);
+
+void vp9_pack_bitstream(VP9_COMP *cpi, unsigned char *dest,
+                        unsigned long *size);
+
+void vp9_activity_masking(VP9_COMP *cpi, MACROBLOCK *x);
+
+void vp9_set_speed_features(VP9_COMP *cpi);
+
+extern int vp9_calc_ss_err(YV12_BUFFER_CONFIG *source,
+                           YV12_BUFFER_CONFIG *dest);
+
+extern void vp9_alloc_compressor_data(VP9_COMP *cpi);
+
+#if CONFIG_DEBUG
+#define CHECK_MEM_ERROR(lval,expr) do {\
+    lval = (expr); \
+    if(!lval) \
+      vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,\
+                         "Failed to allocate "#lval" at %s:%d", \
+                         __FILE__,__LINE__);\
+  } while(0)
+#else
+#define CHECK_MEM_ERROR(lval,expr) do {\
+    lval = (expr); \
+    if(!lval) \
+      vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,\
+                         "Failed to allocate "#lval);\
+  } while(0)
+#endif
+
+#endif  // VP9_ENCODER_VP9_ONYX_INT_H_
diff --git a/libvpx/vp9/encoder/vp9_picklpf.c b/libvpx/vp9/encoder/vp9_picklpf.c
new file mode 100644
index 0000000..a87d058
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_picklpf.c
@@ -0,0 +1,236 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <limits.h>
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_picklpf.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "./vpx_scale_rtcd.h"
+
+void vp9_yv12_copy_partial_frame_c(YV12_BUFFER_CONFIG *src_ybc,
+                                   YV12_BUFFER_CONFIG *dst_ybc, int Fraction) {
+  uint8_t *src_y, *dst_y;
+  int yheight;
+  int ystride;
+  int yoffset;
+  int linestocopy;
+
+  assert(src_ybc->y_stride == dst_ybc->y_stride);
+  yheight  = src_ybc->y_height;
+  ystride  = src_ybc->y_stride;
+
+  linestocopy = (yheight >> (Fraction + 4));
+
+  if (linestocopy < 1)
+    linestocopy = 1;
+
+  linestocopy <<= 4;
+
+  yoffset  = ystride * ((yheight >> 5) * 16 - 8);
+  src_y = src_ybc->y_buffer + yoffset;
+  dst_y = dst_ybc->y_buffer + yoffset;
+
+  vpx_memcpy(dst_y, src_y, ystride * (linestocopy + 16));
+}
+
+static int calc_partial_ssl_err(YV12_BUFFER_CONFIG *source,
+                                YV12_BUFFER_CONFIG *dest, int Fraction) {
+  int i, j;
+  int Total = 0;
+  int srcoffset, dstoffset;
+  uint8_t *src = source->y_buffer;
+  uint8_t *dst = dest->y_buffer;
+
+  int linestocopy = (source->y_height >> (Fraction + 4));
+
+  if (linestocopy < 1)
+    linestocopy = 1;
+
+  linestocopy <<= 4;
+
+
+  srcoffset = source->y_stride   * (dest->y_height >> 5) * 16;
+  dstoffset = dest->y_stride     * (dest->y_height >> 5) * 16;
+
+  src += srcoffset;
+  dst += dstoffset;
+
+  // Loop through the Y plane raw and reconstruction data summing (square differences)
+  for (i = 0; i < linestocopy; i += 16) {
+    for (j = 0; j < source->y_width; j += 16) {
+      unsigned int sse;
+      Total += vp9_mse16x16(src + j, source->y_stride, dst + j, dest->y_stride,
+                            &sse);
+    }
+
+    src += 16 * source->y_stride;
+    dst += 16 * dest->y_stride;
+  }
+
+  return Total;
+}
+
+// Enforce a minimum filter level based upon baseline Q
+static int get_min_filter_level(VP9_COMP *cpi, int base_qindex) {
+  int min_filter_level;
+  /*int q = (int) vp9_convert_qindex_to_q(base_qindex);
+
+  if (cpi->source_alt_ref_active && cpi->common.refresh_golden_frame && !cpi->common.refresh_alt_ref_frame)
+      min_filter_level = 0;
+  else
+  {
+      if (q <= 10)
+          min_filter_level = 0;
+      else if (q <= 64)
+          min_filter_level = 1;
+      else
+          min_filter_level = (q >> 6);
+  }
+  */
+  min_filter_level = 0;
+
+  return min_filter_level;
+}
+
+// Enforce a maximum filter level based upon baseline Q
+static int get_max_filter_level(VP9_COMP *cpi, int base_qindex) {
+  // PGW August 2006: Highest filter values almost always a bad idea
+
+  // jbb chg: 20100118 - not so any more with this overquant stuff allow high values
+  // with lots of intra coming in.
+  int max_filter_level = MAX_LOOP_FILTER;// * 3 / 4;
+  (void)base_qindex;
+
+  if (cpi->twopass.section_intra_rating > 8)
+    max_filter_level = MAX_LOOP_FILTER * 3 / 4;
+
+  return max_filter_level;
+}
+
+
+// Stub function for now Alt LF not used
+void vp9_set_alt_lf_level(VP9_COMP *cpi, int filt_val) {
+}
+
+void vp9_pick_filter_level(YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  int best_err = 0;
+  int filt_err = 0;
+  int min_filter_level = get_min_filter_level(cpi, cm->base_qindex);
+  int max_filter_level = get_max_filter_level(cpi, cm->base_qindex);
+
+  int filter_step;
+  int filt_high = 0;
+  int filt_mid = cm->filter_level;      // Start search at previous frame filter level
+  int filt_low = 0;
+  int filt_best;
+  int filt_direction = 0;
+
+  int Bias = 0;                       // Bias against raising loop filter and in favour of lowering it
+
+  //  Make a copy of the unfiltered / processed recon buffer
+  vp8_yv12_copy_y(cm->frame_to_show, &cpi->last_frame_uf);
+
+  if (cm->frame_type == KEY_FRAME)
+    cm->sharpness_level = 0;
+  else
+    cm->sharpness_level = cpi->oxcf.Sharpness;
+
+  // Start the search at the previous frame filter level unless it is now out of range.
+  filt_mid = cm->filter_level;
+
+  if (filt_mid < min_filter_level)
+    filt_mid = min_filter_level;
+  else if (filt_mid > max_filter_level)
+    filt_mid = max_filter_level;
+
+  // Define the initial step size
+  filter_step = (filt_mid < 16) ? 4 : filt_mid / 4;
+
+  // Get baseline error score
+  vp9_set_alt_lf_level(cpi, filt_mid);
+  vp9_loop_filter_frame(cm, &cpi->mb.e_mbd, filt_mid, 1);
+
+  best_err = vp9_calc_ss_err(sd, cm->frame_to_show);
+  filt_best = filt_mid;
+
+  //  Re-instate the unfiltered frame
+  vp8_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
+
+  while (filter_step > 0) {
+    Bias = (best_err >> (15 - (filt_mid / 8))) * filter_step; // PGW change 12/12/06 for small images
+
+    // jbb chg: 20100118 - in sections with lots of new material coming in don't bias as much to a low filter value
+    if (cpi->twopass.section_intra_rating < 20)
+      Bias = Bias * cpi->twopass.section_intra_rating / 20;
+
+    // yx, bias less for large block size
+    if (cpi->common.txfm_mode != ONLY_4X4)
+      Bias >>= 1;
+
+    filt_high = ((filt_mid + filter_step) > max_filter_level) ? max_filter_level : (filt_mid + filter_step);
+    filt_low = ((filt_mid - filter_step) < min_filter_level) ? min_filter_level : (filt_mid - filter_step);
+
+    if ((filt_direction <= 0) && (filt_low != filt_mid)) {
+      // Get Low filter error score
+      vp9_set_alt_lf_level(cpi, filt_low);
+      vp9_loop_filter_frame(cm, &cpi->mb.e_mbd, filt_low, 1);
+
+      filt_err = vp9_calc_ss_err(sd, cm->frame_to_show);
+
+      //  Re-instate the unfiltered frame
+      vp8_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
+
+      // If value is close to the best so far then bias towards a lower loop filter value.
+      if ((filt_err - Bias) < best_err) {
+        // Was it actually better than the previous best?
+        if (filt_err < best_err)
+          best_err = filt_err;
+
+        filt_best = filt_low;
+      }
+    }
+
+    // Now look at filt_high
+    if ((filt_direction >= 0) && (filt_high != filt_mid)) {
+      vp9_set_alt_lf_level(cpi, filt_high);
+      vp9_loop_filter_frame(cm, &cpi->mb.e_mbd, filt_high, 1);
+
+      filt_err = vp9_calc_ss_err(sd, cm->frame_to_show);
+
+      //  Re-instate the unfiltered frame
+      vp8_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
+
+      // Was it better than the previous best?
+      if (filt_err < (best_err - Bias)) {
+        best_err = filt_err;
+        filt_best = filt_high;
+      }
+    }
+
+    // Half the step distance if the best filter value was the same as last time
+    if (filt_best == filt_mid) {
+      filter_step = filter_step / 2;
+      filt_direction = 0;
+    } else {
+      filt_direction = (filt_best < filt_mid) ? -1 : 1;
+      filt_mid = filt_best;
+    }
+  }
+
+  cm->filter_level = filt_best;
+}
diff --git a/libvpx/vp9/encoder/vp9_picklpf.h b/libvpx/vp9/encoder/vp9_picklpf.h
new file mode 100644
index 0000000..698cb8d
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_picklpf.h
@@ -0,0 +1,23 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_PICKLPF_H_
+#define VP9_ENCODER_VP9_PICKLPF_H_
+
+struct yv12_buffer_config;
+struct VP9_COMP;
+
+void vp9_set_alt_lf_level(struct VP9_COMP *cpi, int filt_val);
+
+void vp9_pick_filter_level(struct yv12_buffer_config *sd,
+                           struct VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_PICKLPF_H_
diff --git a/libvpx/vp9/encoder/vp9_psnr.c b/libvpx/vp9/encoder/vp9_psnr.c
new file mode 100644
index 0000000..9439434
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_psnr.c
@@ -0,0 +1,29 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx_scale/yv12config.h"
+#include "math.h"
+
+#define MAX_PSNR 100
+
+double vp9_mse2psnr(double samples, double peak, double mse) {
+  double psnr;
+
+  if (mse > 0.0)
+    psnr = 10.0 * log10(peak * peak * samples / mse);
+  else
+    psnr = MAX_PSNR;  // Limit to prevent / 0
+
+  if (psnr > MAX_PSNR)
+    psnr = MAX_PSNR;
+
+  return psnr;
+}
diff --git a/libvpx/vp9/encoder/vp9_psnr.h b/libvpx/vp9/encoder/vp9_psnr.h
new file mode 100644
index 0000000..15dd836
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_psnr.h
@@ -0,0 +1,17 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_PSNR_H_
+#define VP9_ENCODER_VP9_PSNR_H_
+
+double vp9_mse2psnr(double samples, double peak, double mse);
+
+#endif  // VP9_ENCODER_VP9_PSNR_H_
diff --git a/libvpx/vp9/encoder/vp9_quantize.c b/libvpx/vp9/encoder/vp9_quantize.c
new file mode 100644
index 0000000..53d8be7
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_quantize.c
@@ -0,0 +1,312 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/common/vp9_quant_common.h"
+
+#include "vp9/common/vp9_seg_common.h"
+
+#ifdef ENC_DEBUG
+extern int enc_debug;
+#endif
+
+static INLINE int plane_idx(int plane) {
+  return plane == 0 ? 0 :
+         plane == 1 ? 16 : 20;
+}
+
+static void quantize(int16_t *zbin_boost_orig_ptr,
+                     int16_t *coeff_ptr, int n_coeffs, int skip_block,
+                     int16_t *zbin_ptr, int16_t *round_ptr, int16_t *quant_ptr,
+                     uint8_t *quant_shift_ptr,
+                     int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                     int16_t *dequant_ptr, int zbin_oq_value,
+                     uint16_t *eob_ptr,
+                     const int *scan, int mul) {
+  int i, rc, eob;
+  int zbin;
+  int x, y, z, sz;
+  int zero_run = 0;
+  int16_t *zbin_boost_ptr = zbin_boost_orig_ptr;
+
+  vpx_memset(qcoeff_ptr, 0, n_coeffs*sizeof(int16_t));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs*sizeof(int16_t));
+
+  eob = -1;
+
+  if (!skip_block) {
+    for (i = 0; i < n_coeffs; i++) {
+      rc   = scan[i];
+      z    = coeff_ptr[rc] * mul;
+
+      zbin = (zbin_ptr[rc != 0] + zbin_boost_ptr[zero_run] + zbin_oq_value);
+      zero_run += (zero_run < 15);
+
+      sz = (z >> 31);                               // sign of z
+      x  = (z ^ sz) - sz;                           // x = abs(z)
+
+      if (x >= zbin) {
+        x += (round_ptr[rc != 0]);
+        y  = ((int)(((int)(x * quant_ptr[rc != 0]) >> 16) + x))
+            >> quant_shift_ptr[rc != 0];            // quantize (x)
+        x  = (y ^ sz) - sz;                         // get the sign back
+        qcoeff_ptr[rc]  = x;                        // write to destination
+        dqcoeff_ptr[rc] = x * dequant_ptr[rc != 0] / mul;  // dequantized value
+
+        if (y) {
+          eob = i;                                  // last nonzero coeffs
+          zero_run = 0;
+        }
+      }
+    }
+  }
+
+  *eob_ptr = eob + 1;
+}
+
+void vp9_quantize(MACROBLOCK *mb, int plane, int block, int n_coeffs,
+                  TX_TYPE tx_type) {
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  const int mul = n_coeffs == 1024 ? 2 : 1;
+  const int *scan;
+
+  // These contexts may be available in the caller
+  switch (n_coeffs) {
+    case 4 * 4:
+      scan = get_scan_4x4(tx_type);
+      break;
+    case 8 * 8:
+      scan = get_scan_8x8(tx_type);
+      break;
+    case 16 * 16:
+      scan = get_scan_16x16(tx_type);
+      break;
+    default:
+      scan = vp9_default_scan_32x32;
+      break;
+  }
+
+  quantize(mb->plane[plane].zrun_zbin_boost,
+           BLOCK_OFFSET(mb->plane[plane].coeff, block, 16),
+           n_coeffs, mb->skip_block,
+           mb->plane[plane].zbin,
+           mb->plane[plane].round,
+           mb->plane[plane].quant,
+           mb->plane[plane].quant_shift,
+           BLOCK_OFFSET(xd->plane[plane].qcoeff, block, 16),
+           BLOCK_OFFSET(xd->plane[plane].dqcoeff, block, 16),
+           xd->plane[plane].dequant,
+           mb->plane[plane].zbin_extra,
+           &xd->plane[plane].eobs[block],
+           scan, mul);
+}
+
+void vp9_regular_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type,
+                                int y_blocks) {
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  const struct plane_block_idx pb_idx = plane_block_idx(y_blocks, b_idx);
+  const int *pt_scan = get_scan_4x4(tx_type);
+
+  quantize(mb->plane[pb_idx.plane].zrun_zbin_boost,
+           BLOCK_OFFSET(mb->plane[pb_idx.plane].coeff, pb_idx.block, 16),
+           16, mb->skip_block,
+           mb->plane[pb_idx.plane].zbin,
+           mb->plane[pb_idx.plane].round,
+           mb->plane[pb_idx.plane].quant,
+           mb->plane[pb_idx.plane].quant_shift,
+           BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16),
+           BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, pb_idx.block, 16),
+           xd->plane[pb_idx.plane].dequant,
+           mb->plane[pb_idx.plane].zbin_extra,
+           &xd->plane[pb_idx.plane].eobs[pb_idx.block],
+           pt_scan, 1);
+}
+
+static void invert_quant(int16_t *quant, uint8_t *shift, int d) {
+  unsigned t;
+  int l;
+  t = d;
+  for (l = 0; t > 1; l++)
+    t >>= 1;
+  t = 1 + (1 << (16 + l)) / d;
+  *quant = (int16_t)(t - (1 << 16));
+  *shift = l;
+}
+
+void vp9_init_quantizer(VP9_COMP *cpi) {
+  int i;
+  int quant_val;
+  int quant_uv_val;
+#if CONFIG_ALPHA
+  int quant_alpha_val;
+#endif
+  int q;
+
+  static const int zbin_boost[16] = { 0,  0,  0,  8,  8,  8, 10, 12,
+                                     14, 16, 20, 24, 28, 32, 36, 40 };
+
+  for (q = 0; q < QINDEX_RANGE; q++) {
+    int qzbin_factor = (vp9_dc_quant(q, 0) < 148) ? 84 : 80;
+    int qrounding_factor = 48;
+    if (q == 0) {
+      qzbin_factor = 64;
+      qrounding_factor = 64;
+    }
+    // dc values
+    quant_val = vp9_dc_quant(q, cpi->common.y_dc_delta_q);
+    invert_quant(cpi->y_quant[q] + 0, cpi->y_quant_shift[q] + 0, quant_val);
+    cpi->y_zbin[q][0] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
+    cpi->y_round[q][0] = (qrounding_factor * quant_val) >> 7;
+    cpi->common.y_dequant[q][0] = quant_val;
+    cpi->zrun_zbin_boost_y[q][0] = (quant_val * zbin_boost[0]) >> 7;
+
+    quant_val = vp9_dc_quant(q, cpi->common.uv_dc_delta_q);
+    invert_quant(cpi->uv_quant[q] + 0, cpi->uv_quant_shift[q] + 0, quant_val);
+    cpi->uv_zbin[q][0] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
+    cpi->uv_round[q][0] = (qrounding_factor * quant_val) >> 7;
+    cpi->common.uv_dequant[q][0] = quant_val;
+    cpi->zrun_zbin_boost_uv[q][0] = (quant_val * zbin_boost[0]) >> 7;
+
+#if CONFIG_ALPHA
+    quant_val = vp9_dc_quant(q, cpi->common.a_dc_delta_q);
+    invert_quant(cpi->a_quant[q] + 0, cpi->a_quant_shift[q] + 0, quant_val);
+    cpi->a_zbin[q][0] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
+    cpi->a_round[q][0] = (qrounding_factor * quant_val) >> 7;
+    cpi->common.a_dequant[q][0] = quant_val;
+    cpi->zrun_zbin_boost_a[q][0] = (quant_val * zbin_boost[0]) >> 7;
+#endif
+
+    quant_val = vp9_ac_quant(q, 0);
+    cpi->common.y_dequant[q][1] = quant_val;
+    quant_uv_val = vp9_ac_quant(q, cpi->common.uv_ac_delta_q);
+    cpi->common.uv_dequant[q][1] = quant_uv_val;
+#if CONFIG_ALPHA
+    quant_alpha_val = vp9_ac_quant(q, cpi->common.a_ac_delta_q);
+    cpi->common.a_dequant[q][1] = quant_alpha_val;
+#endif
+    // all the 4x4 ac values =;
+    for (i = 1; i < 16; i++) {
+      int rc = vp9_default_scan_4x4[i];
+
+      invert_quant(cpi->y_quant[q] + rc, cpi->y_quant_shift[q] + rc, quant_val);
+      cpi->y_zbin[q][rc] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
+      cpi->y_round[q][rc] = (qrounding_factor * quant_val) >> 7;
+      cpi->zrun_zbin_boost_y[q][i] =
+          ROUND_POWER_OF_TWO(quant_val * zbin_boost[i], 7);
+
+      invert_quant(cpi->uv_quant[q] + rc, cpi->uv_quant_shift[q] + rc,
+        quant_uv_val);
+      cpi->uv_zbin[q][rc] = ROUND_POWER_OF_TWO(qzbin_factor * quant_uv_val, 7);
+      cpi->uv_round[q][rc] = (qrounding_factor * quant_uv_val) >> 7;
+      cpi->zrun_zbin_boost_uv[q][i] =
+          ROUND_POWER_OF_TWO(quant_uv_val * zbin_boost[i], 7);
+
+#if CONFIG_ALPHA
+      invert_quant(cpi->a_quant[q] + rc, cpi->a_quant_shift[q] + rc,
+          quant_alpha_val);
+      cpi->a_zbin[q][rc] =
+          ROUND_POWER_OF_TWO(qzbin_factor * quant_alpha_val, 7);
+      cpi->a_round[q][rc] = (qrounding_factor * quant_alpha_val) >> 7;
+      cpi->zrun_zbin_boost_a[q][i] =
+          ROUND_POWER_OF_TWO(quant_alpha_val * zbin_boost[i], 7);
+#endif
+    }
+  }
+}
+
+void vp9_mb_init_quantizer(VP9_COMP *cpi, MACROBLOCK *x) {
+  int i;
+  MACROBLOCKD *xd = &x->e_mbd;
+  int zbin_extra;
+  int segment_id = xd->mode_info_context->mbmi.segment_id;
+  const int qindex = vp9_get_qindex(xd, segment_id, cpi->common.base_qindex);
+
+  // Y
+  zbin_extra = (cpi->common.y_dequant[qindex][1] *
+                 (cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
+
+  x->plane[0].quant = cpi->y_quant[qindex];
+  x->plane[0].quant_shift = cpi->y_quant_shift[qindex];
+  x->plane[0].zbin = cpi->y_zbin[qindex];
+  x->plane[0].round = cpi->y_round[qindex];
+  x->plane[0].zrun_zbin_boost = cpi->zrun_zbin_boost_y[qindex];
+  x->plane[0].zbin_extra = (int16_t)zbin_extra;
+  x->e_mbd.plane[0].dequant = cpi->common.y_dequant[qindex];
+
+  // UV
+  zbin_extra = (cpi->common.uv_dequant[qindex][1] *
+                (cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
+
+  for (i = 1; i < 3; i++) {
+    x->plane[i].quant = cpi->uv_quant[qindex];
+    x->plane[i].quant_shift = cpi->uv_quant_shift[qindex];
+    x->plane[i].zbin = cpi->uv_zbin[qindex];
+    x->plane[i].round = cpi->uv_round[qindex];
+    x->plane[i].zrun_zbin_boost = cpi->zrun_zbin_boost_uv[qindex];
+    x->plane[i].zbin_extra = (int16_t)zbin_extra;
+    x->e_mbd.plane[i].dequant = cpi->common.uv_dequant[qindex];
+  }
+
+#if CONFIG_ALPHA
+  x->plane[3].quant = cpi->a_quant[qindex];
+  x->plane[3].quant_shift = cpi->a_quant_shift[qindex];
+  x->plane[3].zbin = cpi->a_zbin[qindex];
+  x->plane[3].round = cpi->a_round[qindex];
+  x->plane[3].zrun_zbin_boost = cpi->zrun_zbin_boost_a[qindex];
+  x->plane[3].zbin_extra = (int16_t)zbin_extra;
+  x->e_mbd.plane[3].dequant = cpi->common.a_dequant[qindex];
+#endif
+
+  x->skip_block = vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
+
+  /* save this macroblock QIndex for vp9_update_zbin_extra() */
+  x->e_mbd.q_index = qindex;
+}
+
+void vp9_update_zbin_extra(VP9_COMP *cpi, MACROBLOCK *x) {
+  const int qindex = x->e_mbd.q_index;
+  const int y_zbin_extra = (cpi->common.y_dequant[qindex][1] *
+                (cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
+  const int uv_zbin_extra = (cpi->common.uv_dequant[qindex][1] *
+                  (cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
+
+  x->plane[0].zbin_extra = (int16_t)y_zbin_extra;
+  x->plane[1].zbin_extra = (int16_t)uv_zbin_extra;
+  x->plane[2].zbin_extra = (int16_t)uv_zbin_extra;
+}
+
+void vp9_frame_init_quantizer(VP9_COMP *cpi) {
+  // Clear Zbin mode boost for default case
+  cpi->zbin_mode_boost = 0;
+
+  // MB level quantizer setup
+  vp9_mb_init_quantizer(cpi, &cpi->mb);
+}
+
+void vp9_set_quantizer(struct VP9_COMP *cpi, int Q) {
+  VP9_COMMON *cm = &cpi->common;
+
+  cm->base_qindex = Q;
+
+  // if any of the delta_q values are changing update flag will
+  // have to be set.
+  cm->y_dc_delta_q = 0;
+  cm->uv_dc_delta_q = 0;
+  cm->uv_ac_delta_q = 0;
+
+  // quantizer has to be reinitialized if any delta_q changes.
+  // As there are not any here for now this is inactive code.
+  // if(update)
+  //    vp9_init_quantizer(cpi);
+}
diff --git a/libvpx/vp9/encoder/vp9_quantize.h b/libvpx/vp9/encoder/vp9_quantize.h
new file mode 100644
index 0000000..2b1eeab
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_quantize.h
@@ -0,0 +1,46 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_QUANTIZE_H_
+#define VP9_ENCODER_VP9_QUANTIZE_H_
+
+#include "vp9/encoder/vp9_block.h"
+
+#define prototype_quantize_block(sym) \
+  void (sym)(MACROBLOCK *mb, int b_idx)
+
+#define prototype_quantize_block_pair(sym) \
+  void (sym)(MACROBLOCK *mb, int b_idx1, int b_idx2)
+
+#define prototype_quantize_mb(sym) \
+  void (sym)(MACROBLOCK *x)
+
+void vp9_quantize(MACROBLOCK *mb, int plane, int block, int n_coefs,
+                  TX_TYPE tx_type);
+
+void vp9_regular_quantize_b_4x4_pair(MACROBLOCK *mb, int b_idx1, int b_idx2,
+                                     int y_blocks);
+void vp9_regular_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type,
+                                int y_blocks);
+void vp9_regular_quantize_b_8x8(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type,
+                                int y_blocks);
+struct VP9_COMP;
+
+extern void vp9_set_quantizer(struct VP9_COMP *cpi, int Q);
+
+extern void vp9_frame_init_quantizer(struct VP9_COMP *cpi);
+
+extern void vp9_update_zbin_extra(struct VP9_COMP *cpi, MACROBLOCK *x);
+
+extern void vp9_mb_init_quantizer(struct VP9_COMP *cpi, MACROBLOCK *x);
+
+extern void vp9_init_quantizer(struct VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_QUANTIZE_H_
diff --git a/libvpx/vp9/encoder/vp9_ratectrl.c b/libvpx/vp9/encoder/vp9_ratectrl.c
new file mode 100644
index 0000000..430d3a8
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_ratectrl.c
@@ -0,0 +1,550 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <limits.h>
+#include <assert.h>
+#include <math.h>
+
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/common/vp9_modecont.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/encoder/vp9_ratectrl.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_seg_common.h"
+
+#define MIN_BPB_FACTOR 0.005
+#define MAX_BPB_FACTOR 50
+
+// Bits Per MB at different Q (Multiplied by 512)
+#define BPER_MB_NORMBITS    9
+
+// % adjustment to target kf size based on seperation from previous frame
+static const int kf_boost_seperation_adjustment[16] = {
+  30,   40,   50,   55,   60,   65,   70,   75,
+  80,   85,   90,   95,  100,  100,  100,  100,
+};
+
+static const int gf_adjust_table[101] = {
+  100,
+  115, 130, 145, 160, 175, 190, 200, 210, 220, 230,
+  240, 260, 270, 280, 290, 300, 310, 320, 330, 340,
+  350, 360, 370, 380, 390, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+  400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
+};
+
+static const int gf_intra_usage_adjustment[20] = {
+  125, 120, 115, 110, 105, 100,  95,  85,  80,  75,
+  70,  65,  60,  55,  50,  50,  50,  50,  50,  50,
+};
+
+static const int gf_interval_table[101] = {
+  7,
+  7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+  7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+  7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+  8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+  8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+  9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+  9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+  10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
+  10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
+  11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
+};
+
+static const unsigned int prior_key_frame_weight[KEY_FRAME_CONTEXT] = { 1, 2, 3, 4, 5 };
+
+// These functions use formulaic calculations to make playing with the
+// quantizer tables easier. If necessary they can be replaced by lookup
+// tables if and when things settle down in the experimental bitstream
+double vp9_convert_qindex_to_q(int qindex) {
+  // Convert the index to a real Q value (scaled down to match old Q values)
+  return vp9_ac_quant(qindex, 0) / 4.0;
+}
+
+int vp9_gfboost_qadjust(int qindex) {
+  const double q = vp9_convert_qindex_to_q(qindex);
+  return (int)((0.00000828 * q * q * q) +
+               (-0.0055 * q * q) +
+               (1.32 * q) + 79.3);
+}
+
+static int kfboost_qadjust(int qindex) {
+  const double q = vp9_convert_qindex_to_q(qindex);
+  return (int)((0.00000973 * q * q * q) +
+               (-0.00613 * q * q) +
+               (1.316 * q) + 121.2);
+}
+
+int vp9_bits_per_mb(FRAME_TYPE frame_type, int qindex,
+                    double correction_factor) {
+
+  const double q = vp9_convert_qindex_to_q(qindex);
+  int enumerator = frame_type == KEY_FRAME ? 4000000 : 2500000;
+
+  // q based adjustment to baseline enumerator
+  enumerator += (int)(enumerator * q) >> 12;
+  return (int)(0.5 + (enumerator * correction_factor / q));
+}
+
+void vp9_save_coding_context(VP9_COMP *cpi) {
+  CODING_CONTEXT *const cc = &cpi->coding_context;
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+
+  // Stores a snapshot of key state variables which can subsequently be
+  // restored with a call to vp9_restore_coding_context. These functions are
+  // intended for use in a re-code loop in vp9_compress_frame where the
+  // quantizer value is adjusted between loop iterations.
+
+  cc->nmvc = cm->fc.nmvc;
+  vp9_copy(cc->nmvjointcost,  cpi->mb.nmvjointcost);
+  vp9_copy(cc->nmvcosts,  cpi->mb.nmvcosts);
+  vp9_copy(cc->nmvcosts_hp,  cpi->mb.nmvcosts_hp);
+
+  vp9_copy(cc->inter_mode_probs, cm->fc.inter_mode_probs);
+
+  vp9_copy(cc->y_mode_prob, cm->fc.y_mode_prob);
+  vp9_copy(cc->uv_mode_prob, cm->fc.uv_mode_prob);
+  vp9_copy(cc->partition_prob, cm->fc.partition_prob);
+
+  vp9_copy(cc->segment_pred_probs, cm->segment_pred_probs);
+
+  vp9_copy(cc->intra_inter_prob, cm->fc.intra_inter_prob);
+  vp9_copy(cc->comp_inter_prob, cm->fc.comp_inter_prob);
+  vp9_copy(cc->single_ref_prob, cm->fc.single_ref_prob);
+  vp9_copy(cc->comp_ref_prob, cm->fc.comp_ref_prob);
+
+  vpx_memcpy(cpi->coding_context.last_frame_seg_map_copy,
+             cm->last_frame_seg_map, (cm->mi_rows * cm->mi_cols));
+
+  vp9_copy(cc->last_ref_lf_deltas, xd->last_ref_lf_deltas);
+  vp9_copy(cc->last_mode_lf_deltas, xd->last_mode_lf_deltas);
+
+  vp9_copy(cc->coef_probs, cm->fc.coef_probs);
+  vp9_copy(cc->switchable_interp_prob, cm->fc.switchable_interp_prob);
+  vp9_copy(cc->tx_probs_8x8p, cm->fc.tx_probs_8x8p);
+  vp9_copy(cc->tx_probs_16x16p, cm->fc.tx_probs_16x16p);
+  vp9_copy(cc->tx_probs_32x32p, cm->fc.tx_probs_32x32p);
+  vp9_copy(cc->mbskip_probs, cm->fc.mbskip_probs);
+}
+
+void vp9_restore_coding_context(VP9_COMP *cpi) {
+  CODING_CONTEXT *const cc = &cpi->coding_context;
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+
+  // Restore key state variables to the snapshot state stored in the
+  // previous call to vp9_save_coding_context.
+
+  cm->fc.nmvc = cc->nmvc;
+  vp9_copy(cpi->mb.nmvjointcost, cc->nmvjointcost);
+  vp9_copy(cpi->mb.nmvcosts, cc->nmvcosts);
+  vp9_copy(cpi->mb.nmvcosts_hp, cc->nmvcosts_hp);
+
+  vp9_copy(cm->fc.inter_mode_probs, cc->inter_mode_probs);
+
+  vp9_copy(cm->fc.y_mode_prob, cc->y_mode_prob);
+  vp9_copy(cm->fc.uv_mode_prob, cc->uv_mode_prob);
+  vp9_copy(cm->fc.partition_prob, cc->partition_prob);
+
+  vp9_copy(cm->segment_pred_probs, cc->segment_pred_probs);
+
+  vp9_copy(cm->fc.intra_inter_prob, cc->intra_inter_prob);
+  vp9_copy(cm->fc.comp_inter_prob, cc->comp_inter_prob);
+  vp9_copy(cm->fc.single_ref_prob, cc->single_ref_prob);
+  vp9_copy(cm->fc.comp_ref_prob, cc->comp_ref_prob);
+
+  vpx_memcpy(cm->last_frame_seg_map,
+             cpi->coding_context.last_frame_seg_map_copy,
+             (cm->mi_rows * cm->mi_cols));
+
+  vp9_copy(xd->last_ref_lf_deltas, cc->last_ref_lf_deltas);
+  vp9_copy(xd->last_mode_lf_deltas, cc->last_mode_lf_deltas);
+
+  vp9_copy(cm->fc.coef_probs, cc->coef_probs);
+  vp9_copy(cm->fc.switchable_interp_prob, cc->switchable_interp_prob);
+  vp9_copy(cm->fc.tx_probs_8x8p, cc->tx_probs_8x8p);
+  vp9_copy(cm->fc.tx_probs_16x16p, cc->tx_probs_16x16p);
+  vp9_copy(cm->fc.tx_probs_32x32p, cc->tx_probs_32x32p);
+  vp9_copy(cm->fc.mbskip_probs, cc->mbskip_probs);
+}
+
+void vp9_setup_key_frame(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+
+  vp9_setup_past_independence(cm, xd);
+
+  // interval before next GF
+  cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
+  /* All buffers are implicitly updated on key frames. */
+  cpi->refresh_golden_frame = 1;
+  cpi->refresh_alt_ref_frame = 1;
+}
+
+void vp9_setup_inter_frame(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  if (cm->error_resilient_mode || cm->intra_only)
+    vp9_setup_past_independence(cm, xd);
+
+  assert(cm->frame_context_idx < NUM_FRAME_CONTEXTS);
+  cm->fc = cm->frame_contexts[cm->frame_context_idx];
+}
+
+static int estimate_bits_at_q(int frame_kind, int q, int mbs,
+                              double correction_factor) {
+  const int bpm = (int)(vp9_bits_per_mb(frame_kind, q, correction_factor));
+
+  // Attempt to retain reasonable accuracy without overflow. The cutoff is
+  // chosen such that the maximum product of Bpm and MBs fits 31 bits. The
+  // largest Bpm takes 20 bits.
+  return (mbs > (1 << 11)) ? (bpm >> BPER_MB_NORMBITS) * mbs
+                           : (bpm * mbs) >> BPER_MB_NORMBITS;
+}
+
+
+static void calc_iframe_target_size(VP9_COMP *cpi) {
+  // boost defaults to half second
+  int target;
+
+  // Clear down mmx registers to allow floating point in what follows
+  vp9_clear_system_state();  // __asm emms;
+
+  // New Two pass RC
+  target = cpi->per_frame_bandwidth;
+
+  if (cpi->oxcf.rc_max_intra_bitrate_pct) {
+    int max_rate = cpi->per_frame_bandwidth
+                 * cpi->oxcf.rc_max_intra_bitrate_pct / 100;
+
+    if (target > max_rate)
+      target = max_rate;
+  }
+
+  cpi->this_frame_target = target;
+}
+
+
+//  Do the best we can to define the parameters for the next GF based
+//  on what information we have available.
+//
+//  In this experimental code only two pass is supported
+//  so we just use the interval determined in the two pass code.
+static void calc_gf_params(VP9_COMP *cpi) {
+  // Set the gf interval
+  cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
+}
+
+
+static void calc_pframe_target_size(VP9_COMP *cpi) {
+  const int min_frame_target = MAX(cpi->min_frame_bandwidth,
+                                   cpi->av_per_frame_bandwidth >> 5);
+  if (cpi->refresh_alt_ref_frame) {
+    // Special alt reference frame case
+    // Per frame bit target for the alt ref frame
+    cpi->per_frame_bandwidth = cpi->twopass.gf_bits;
+    cpi->this_frame_target = cpi->per_frame_bandwidth;
+  } else {
+    // Normal frames (gf,and inter)
+    cpi->this_frame_target = cpi->per_frame_bandwidth;
+  }
+
+  // Sanity check that the total sum of adjustments is not above the maximum allowed
+  // That is that having allowed for KF and GF penalties we have not pushed the
+  // current interframe target to low. If the adjustment we apply here is not capable of recovering
+  // all the extra bits we have spent in the KF or GF then the remainder will have to be recovered over
+  // a longer time span via other buffer / rate control mechanisms.
+  if (cpi->this_frame_target < min_frame_target)
+    cpi->this_frame_target = min_frame_target;
+
+  if (!cpi->refresh_alt_ref_frame)
+    // Note the baseline target data rate for this inter frame.
+    cpi->inter_frame_target = cpi->this_frame_target;
+
+  // Adjust target frame size for Golden Frames:
+  if (cpi->frames_till_gf_update_due == 0) {
+    const int q = (cpi->oxcf.fixed_q < 0) ? cpi->last_q[INTER_FRAME]
+                                          : cpi->oxcf.fixed_q;
+
+    cpi->refresh_golden_frame = 1;
+
+    calc_gf_params(cpi);
+
+    // If we are using alternate ref instead of gf then do not apply the boost
+    // It will instead be applied to the altref update
+    // Jims modified boost
+    if (!cpi->source_alt_ref_active) {
+      if (cpi->oxcf.fixed_q < 0) {
+        // The spend on the GF is defined in the two pass code
+        // for two pass encodes
+        cpi->this_frame_target = cpi->per_frame_bandwidth;
+      } else {
+        cpi->this_frame_target =
+          (estimate_bits_at_q(1, q, cpi->common.MBs, 1.0)
+           * cpi->last_boost) / 100;
+      }
+    } else {
+      // If there is an active ARF at this location use the minimum
+      // bits on this frame even if it is a constructed arf.
+      // The active maximum quantizer insures that an appropriate
+      // number of bits will be spent if needed for constructed ARFs.
+      cpi->this_frame_target = 0;
+    }
+  }
+}
+
+
+void vp9_update_rate_correction_factors(VP9_COMP *cpi, int damp_var) {
+  const int q = cpi->common.base_qindex;
+  int correction_factor = 100;
+  double rate_correction_factor;
+  double adjustment_limit;
+
+  int projected_size_based_on_q = 0;
+
+  // Clear down mmx registers to allow floating point in what follows
+  vp9_clear_system_state();  // __asm emms;
+
+  if (cpi->common.frame_type == KEY_FRAME) {
+    rate_correction_factor = cpi->key_frame_rate_correction_factor;
+  } else {
+    if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
+      rate_correction_factor = cpi->gf_rate_correction_factor;
+    else
+      rate_correction_factor = cpi->rate_correction_factor;
+  }
+
+  // Work out how big we would have expected the frame to be at this Q given
+  // the current correction factor.
+  // Stay in double to avoid int overflow when values are large
+  projected_size_based_on_q = estimate_bits_at_q(cpi->common.frame_type, q,
+                                                 cpi->common.MBs,
+                                                 rate_correction_factor);
+
+  // Work out a size correction factor.
+  // if ( cpi->this_frame_target > 0 )
+  //  correction_factor = (100 * cpi->projected_frame_size) / cpi->this_frame_target;
+  if (projected_size_based_on_q > 0)
+    correction_factor = (100 * cpi->projected_frame_size) / projected_size_based_on_q;
+
+  // More heavily damped adjustment used if we have been oscillating either side of target
+  switch (damp_var) {
+    case 0:
+      adjustment_limit = 0.75;
+      break;
+    case 1:
+      adjustment_limit = 0.375;
+      break;
+    case 2:
+    default:
+      adjustment_limit = 0.25;
+      break;
+  }
+
+  // if ( (correction_factor > 102) && (Q < cpi->active_worst_quality) )
+  if (correction_factor > 102) {
+    // We are not already at the worst allowable quality
+    correction_factor = (int)(100.5 + ((correction_factor - 100) * adjustment_limit));
+    rate_correction_factor = ((rate_correction_factor * correction_factor) / 100);
+
+    // Keep rate_correction_factor within limits
+    if (rate_correction_factor > MAX_BPB_FACTOR)
+      rate_correction_factor = MAX_BPB_FACTOR;
+  }
+  // else if ( (correction_factor < 99) && (Q > cpi->active_best_quality) )
+  else if (correction_factor < 99) {
+    // We are not already at the best allowable quality
+    correction_factor = (int)(100.5 - ((100 - correction_factor) * adjustment_limit));
+    rate_correction_factor = ((rate_correction_factor * correction_factor) / 100);
+
+    // Keep rate_correction_factor within limits
+    if (rate_correction_factor < MIN_BPB_FACTOR)
+      rate_correction_factor = MIN_BPB_FACTOR;
+  }
+
+  if (cpi->common.frame_type == KEY_FRAME)
+    cpi->key_frame_rate_correction_factor = rate_correction_factor;
+  else {
+    if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
+      cpi->gf_rate_correction_factor = rate_correction_factor;
+    else
+      cpi->rate_correction_factor = rate_correction_factor;
+  }
+}
+
+
+int vp9_regulate_q(VP9_COMP *cpi, int target_bits_per_frame) {
+  int q = cpi->active_worst_quality;
+
+  int i;
+  int last_error = INT_MAX;
+  int target_bits_per_mb;
+  int bits_per_mb_at_this_q;
+  double correction_factor;
+
+  // Select the appropriate correction factor based upon type of frame.
+  if (cpi->common.frame_type == KEY_FRAME)
+    correction_factor = cpi->key_frame_rate_correction_factor;
+  else {
+    if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
+      correction_factor = cpi->gf_rate_correction_factor;
+    else
+      correction_factor = cpi->rate_correction_factor;
+  }
+
+  // Calculate required scaling factor based on target frame size and size of frame produced using previous Q
+  if (target_bits_per_frame >= (INT_MAX >> BPER_MB_NORMBITS))
+    target_bits_per_mb = (target_bits_per_frame / cpi->common.MBs) << BPER_MB_NORMBITS;       // Case where we would overflow int
+  else
+    target_bits_per_mb = (target_bits_per_frame << BPER_MB_NORMBITS) / cpi->common.MBs;
+
+  i = cpi->active_best_quality;
+
+  do {
+    bits_per_mb_at_this_q = (int)vp9_bits_per_mb(cpi->common.frame_type, i,
+                                                 correction_factor);
+
+    if (bits_per_mb_at_this_q <= target_bits_per_mb) {
+      if ((target_bits_per_mb - bits_per_mb_at_this_q) <= last_error)
+        q = i;
+      else
+        q = i - 1;
+
+      break;
+    } else {
+      last_error = bits_per_mb_at_this_q - target_bits_per_mb;
+    }
+  } while (++i <= cpi->active_worst_quality);
+
+  return q;
+}
+
+
+static int estimate_keyframe_frequency(VP9_COMP *cpi) {
+  int i;
+
+  // Average key frame frequency
+  int av_key_frame_frequency = 0;
+
+  /* First key frame at start of sequence is a special case. We have no
+   * frequency data.
+   */
+  if (cpi->key_frame_count == 1) {
+    /* Assume a default of 1 kf every 2 seconds, or the max kf interval,
+     * whichever is smaller.
+     */
+    int key_freq = cpi->oxcf.key_freq > 0 ? cpi->oxcf.key_freq : 1;
+    av_key_frame_frequency = (int)cpi->output_frame_rate * 2;
+
+    if (cpi->oxcf.auto_key && av_key_frame_frequency > key_freq)
+      av_key_frame_frequency = cpi->oxcf.key_freq;
+
+    cpi->prior_key_frame_distance[KEY_FRAME_CONTEXT - 1]
+      = av_key_frame_frequency;
+  } else {
+    unsigned int total_weight = 0;
+    int last_kf_interval =
+      (cpi->frames_since_key > 0) ? cpi->frames_since_key : 1;
+
+    /* reset keyframe context and calculate weighted average of last
+     * KEY_FRAME_CONTEXT keyframes
+     */
+    for (i = 0; i < KEY_FRAME_CONTEXT; i++) {
+      if (i < KEY_FRAME_CONTEXT - 1)
+        cpi->prior_key_frame_distance[i]
+          = cpi->prior_key_frame_distance[i + 1];
+      else
+        cpi->prior_key_frame_distance[i] = last_kf_interval;
+
+      av_key_frame_frequency += prior_key_frame_weight[i]
+                                * cpi->prior_key_frame_distance[i];
+      total_weight += prior_key_frame_weight[i];
+    }
+
+    av_key_frame_frequency /= total_weight;
+
+  }
+  return av_key_frame_frequency;
+}
+
+
+void vp9_adjust_key_frame_context(VP9_COMP *cpi) {
+  // Clear down mmx registers to allow floating point in what follows
+  vp9_clear_system_state();
+
+  cpi->frames_since_key = 0;
+  cpi->key_frame_count++;
+}
+
+
+void vp9_compute_frame_size_bounds(VP9_COMP *cpi, int *frame_under_shoot_limit,
+                                   int *frame_over_shoot_limit) {
+  // Set-up bounds on acceptable frame size:
+  if (cpi->oxcf.fixed_q >= 0) {
+    // Fixed Q scenario: frame size never outranges target (there is no target!)
+    *frame_under_shoot_limit = 0;
+    *frame_over_shoot_limit  = INT_MAX;
+  } else {
+    if (cpi->common.frame_type == KEY_FRAME) {
+      *frame_over_shoot_limit  = cpi->this_frame_target * 9 / 8;
+      *frame_under_shoot_limit = cpi->this_frame_target * 7 / 8;
+    } else {
+      if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) {
+        *frame_over_shoot_limit  = cpi->this_frame_target * 9 / 8;
+        *frame_under_shoot_limit = cpi->this_frame_target * 7 / 8;
+      } else {
+        // Stron overshoot limit for constrained quality
+        if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
+          *frame_over_shoot_limit  = cpi->this_frame_target * 11 / 8;
+          *frame_under_shoot_limit = cpi->this_frame_target * 2 / 8;
+        } else {
+          *frame_over_shoot_limit  = cpi->this_frame_target * 11 / 8;
+          *frame_under_shoot_limit = cpi->this_frame_target * 5 / 8;
+        }
+      }
+    }
+
+    // For very small rate targets where the fractional adjustment
+    // (eg * 7/8) may be tiny make sure there is at least a minimum
+    // range.
+    *frame_over_shoot_limit += 200;
+    *frame_under_shoot_limit -= 200;
+    if (*frame_under_shoot_limit < 0)
+      *frame_under_shoot_limit = 0;
+  }
+}
+
+
+// return of 0 means drop frame
+int vp9_pick_frame_size(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  if (cm->frame_type == KEY_FRAME)
+    calc_iframe_target_size(cpi);
+  else
+    calc_pframe_target_size(cpi);
+
+  return 1;
+}
diff --git a/libvpx/vp9/encoder/vp9_ratectrl.h b/libvpx/vp9/encoder/vp9_ratectrl.h
new file mode 100644
index 0000000..4733176
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_ratectrl.h
@@ -0,0 +1,39 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_RATECTRL_H_
+#define VP9_ENCODER_VP9_RATECTRL_H_
+
+#include "vp9/encoder/vp9_onyx_int.h"
+
+#define FRAME_OVERHEAD_BITS 200
+
+void vp9_save_coding_context(VP9_COMP *cpi);
+void vp9_restore_coding_context(VP9_COMP *cpi);
+
+void vp9_setup_key_frame(VP9_COMP *cpi);
+void vp9_update_rate_correction_factors(VP9_COMP *cpi, int damp_var);
+int vp9_regulate_q(VP9_COMP *cpi, int target_bits_per_frame);
+void vp9_adjust_key_frame_context(VP9_COMP *cpi);
+void vp9_compute_frame_size_bounds(VP9_COMP *cpi,
+                                   int *frame_under_shoot_limit,
+                                   int *frame_over_shoot_limit);
+
+// return of 0 means drop frame
+int vp9_pick_frame_size(VP9_COMP *cpi);
+
+double vp9_convert_qindex_to_q(int qindex);
+int vp9_gfboost_qadjust(int qindex);
+extern int vp9_bits_per_mb(FRAME_TYPE frame_type, int qindex,
+                           double correction_factor);
+void vp9_setup_inter_frame(VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_RATECTRL_H_
diff --git a/libvpx/vp9/encoder/vp9_rdopt.c b/libvpx/vp9/encoder/vp9_rdopt.c
new file mode 100644
index 0000000..9cb7ab0
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_rdopt.c
@@ -0,0 +1,3255 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdio.h>
+#include <math.h>
+#include <limits.h>
+#include <assert.h>
+
+#include "vp9/common/vp9_pragmas.h"
+#include "vp9/encoder/vp9_tokenize.h"
+#include "vp9/encoder/vp9_treewriter.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_modecosts.h"
+#include "vp9/encoder/vp9_encodeintra.h"
+#include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/common/vp9_findnearmv.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/encoder/vp9_encodemb.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/encoder/vp9_rdopt.h"
+#include "vp9/encoder/vp9_ratectrl.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_encodemv.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vp9_rtcd.h"
+#include "vp9/common/vp9_mvref_common.h"
+#include "vp9/common/vp9_common.h"
+
+#define INVALID_MV 0x80008000
+
+/* Factor to weigh the rate for switchable interp filters */
+#define SWITCHABLE_INTERP_RATE_FACTOR 1
+
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
+#define I4X4_PRED 0x8000
+#define SPLITMV 0x10000
+
+const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
+  {ZEROMV,    LAST_FRAME,   NONE},
+  {DC_PRED,   INTRA_FRAME,  NONE},
+
+  {NEARESTMV, LAST_FRAME,   NONE},
+  {NEARMV,    LAST_FRAME,   NONE},
+
+  {ZEROMV,    GOLDEN_FRAME, NONE},
+  {NEARESTMV, GOLDEN_FRAME, NONE},
+
+  {ZEROMV,    ALTREF_FRAME, NONE},
+  {NEARESTMV, ALTREF_FRAME, NONE},
+
+  {NEARMV,    GOLDEN_FRAME, NONE},
+  {NEARMV,    ALTREF_FRAME, NONE},
+
+  {V_PRED,    INTRA_FRAME,  NONE},
+  {H_PRED,    INTRA_FRAME,  NONE},
+  {D45_PRED,  INTRA_FRAME,  NONE},
+  {D135_PRED, INTRA_FRAME,  NONE},
+  {D117_PRED, INTRA_FRAME,  NONE},
+  {D153_PRED, INTRA_FRAME,  NONE},
+  {D27_PRED,  INTRA_FRAME,  NONE},
+  {D63_PRED,  INTRA_FRAME,  NONE},
+
+  {TM_PRED,   INTRA_FRAME,  NONE},
+
+  {NEWMV,     LAST_FRAME,   NONE},
+  {NEWMV,     GOLDEN_FRAME, NONE},
+  {NEWMV,     ALTREF_FRAME, NONE},
+
+  {SPLITMV,   LAST_FRAME,   NONE},
+  {SPLITMV,   GOLDEN_FRAME, NONE},
+  {SPLITMV,   ALTREF_FRAME, NONE},
+
+  {I4X4_PRED, INTRA_FRAME,  NONE},
+
+  /* compound prediction modes */
+  {ZEROMV,    LAST_FRAME,   ALTREF_FRAME},
+  {NEARESTMV, LAST_FRAME,   ALTREF_FRAME},
+  {NEARMV,    LAST_FRAME,   ALTREF_FRAME},
+
+  {ZEROMV,    GOLDEN_FRAME, ALTREF_FRAME},
+  {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
+  {NEARMV,    GOLDEN_FRAME, ALTREF_FRAME},
+
+  {NEWMV,     LAST_FRAME,   ALTREF_FRAME},
+  {NEWMV,     GOLDEN_FRAME, ALTREF_FRAME},
+
+  {SPLITMV,   LAST_FRAME,   ALTREF_FRAME},
+  {SPLITMV,   GOLDEN_FRAME, ALTREF_FRAME},
+};
+
+// The baseline rd thresholds for breaking out of the rd loop for
+// certain modes are assumed to be based on 8x8 blocks.
+// This table is used to correct for blocks size.
+// The factors here are << 2 (2 = x0.5, 32 = x8 etc).
+static int rd_thresh_block_size_factor[BLOCK_SIZE_TYPES] =
+  {2, 3, 3, 4, 6, 6, 8, 12, 12, 16, 24, 24, 32};
+
+#define BASE_RD_THRESH_FREQ_FACT 16
+#define MAX_RD_THRESH_FREQ_FACT 32
+#define MAX_RD_THRESH_FREQ_INC 1
+
+static void fill_token_costs(vp9_coeff_count (*c)[BLOCK_TYPES],
+                             vp9_coeff_count (*cnoskip)[BLOCK_TYPES],
+                             vp9_coeff_probs_model (*p)[BLOCK_TYPES]) {
+  int i, j, k, l;
+  TX_SIZE t;
+  for (t = TX_4X4; t <= TX_32X32; t++)
+    for (i = 0; i < BLOCK_TYPES; i++)
+      for (j = 0; j < REF_TYPES; j++)
+        for (k = 0; k < COEF_BANDS; k++)
+          for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
+            vp9_prob probs[ENTROPY_NODES];
+            vp9_model_to_full_probs(p[t][i][j][k][l], probs);
+            vp9_cost_tokens((int *)cnoskip[t][i][j][k][l], probs,
+                            vp9_coef_tree);
+#if CONFIG_BALANCED_COEFTREE
+            // Replace the eob node prob with a very small value so that the
+            // cost approximately equals the cost without the eob node
+            probs[1] = 1;
+            vp9_cost_tokens((int *)c[t][i][j][k][l], probs, vp9_coef_tree);
+#else
+            vp9_cost_tokens_skip((int *)c[t][i][j][k][l], probs,
+                                 vp9_coef_tree);
+            assert(c[t][i][j][k][l][DCT_EOB_TOKEN] ==
+                   cnoskip[t][i][j][k][l][DCT_EOB_TOKEN]);
+#endif
+          }
+}
+
+static int rd_iifactor[32] =  { 4, 4, 3, 2, 1, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 0, 0, 0, 0, 0, 0, 0, };
+
+// 3* dc_qlookup[Q]*dc_qlookup[Q];
+
+/* values are now correlated to quantizer */
+static int sad_per_bit16lut[QINDEX_RANGE];
+static int sad_per_bit4lut[QINDEX_RANGE];
+
+void vp9_init_me_luts() {
+  int i;
+
+  // Initialize the sad lut tables using a formulaic calculation for now
+  // This is to make it easier to resolve the impact of experimental changes
+  // to the quantizer tables.
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    sad_per_bit16lut[i] =
+      (int)((0.0418 * vp9_convert_qindex_to_q(i)) + 2.4107);
+    sad_per_bit4lut[i] = (int)(0.063 * vp9_convert_qindex_to_q(i) + 2.742);
+  }
+}
+
+static int compute_rd_mult(int qindex) {
+  const int q = vp9_dc_quant(qindex, 0);
+  return (11 * q * q) >> 2;
+}
+
+void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex) {
+  cpi->mb.sadperbit16 = sad_per_bit16lut[qindex];
+  cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
+}
+
+
+void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
+  int q, i, bsize;
+
+  vp9_clear_system_state();  // __asm emms;
+
+  // Further tests required to see if optimum is different
+  // for key frames, golden frames and arf frames.
+  // if (cpi->common.refresh_golden_frame ||
+  //     cpi->common.refresh_alt_ref_frame)
+  qindex = clamp(qindex, 0, MAXQ);
+
+  cpi->RDMULT = compute_rd_mult(qindex);
+  if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
+    if (cpi->twopass.next_iiratio > 31)
+      cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
+    else
+      cpi->RDMULT +=
+          (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
+  }
+  cpi->mb.errorperbit = cpi->RDMULT >> 6;
+  cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
+
+  vp9_set_speed_features(cpi);
+
+  q = (int)pow(vp9_dc_quant(qindex, 0) >> 2, 1.25);
+  q <<= 2;
+  if (q < 8)
+    q = 8;
+
+  if (cpi->RDMULT > 1000) {
+    cpi->RDDIV = 1;
+    cpi->RDMULT /= 100;
+
+    for (bsize = 0; bsize < BLOCK_SIZE_TYPES; ++bsize) {
+      for (i = 0; i < MAX_MODES; ++i) {
+        // Threshold here seem unecessarily harsh but fine given actual
+        // range of values used for cpi->sf.thresh_mult[]
+        int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
+
+        // *4 relates to the scaling of rd_thresh_block_size_factor[]
+        if ((int64_t)cpi->sf.thresh_mult[i] < thresh_max) {
+          cpi->rd_threshes[bsize][i] =
+            cpi->sf.thresh_mult[i] * q *
+            rd_thresh_block_size_factor[bsize] / (4 * 100);
+        } else {
+          cpi->rd_threshes[bsize][i] = INT_MAX;
+        }
+        cpi->rd_baseline_thresh[bsize][i] = cpi->rd_threshes[bsize][i];
+        cpi->rd_thresh_freq_fact[bsize][i] = BASE_RD_THRESH_FREQ_FACT;
+      }
+    }
+  } else {
+    cpi->RDDIV = 100;
+
+    for (bsize = 0; bsize < BLOCK_SIZE_TYPES; ++bsize) {
+      for (i = 0; i < MAX_MODES; i++) {
+        // Threshold here seem unecessarily harsh but fine given actual
+        // range of values used for cpi->sf.thresh_mult[]
+        int thresh_max = INT_MAX / (q * rd_thresh_block_size_factor[bsize]);
+
+        if (cpi->sf.thresh_mult[i] < thresh_max) {
+          cpi->rd_threshes[bsize][i] =
+            cpi->sf.thresh_mult[i] * q *
+            rd_thresh_block_size_factor[bsize] / 4;
+        } else {
+          cpi->rd_threshes[bsize][i] = INT_MAX;
+        }
+        cpi->rd_baseline_thresh[bsize][i] = cpi->rd_threshes[bsize][i];
+        cpi->rd_thresh_freq_fact[bsize][i] = BASE_RD_THRESH_FREQ_FACT;
+      }
+    }
+  }
+
+  fill_token_costs(cpi->mb.token_costs,
+                   cpi->mb.token_costs_noskip,
+                   cpi->common.fc.coef_probs);
+
+  for (i = 0; i < NUM_PARTITION_CONTEXTS; i++)
+    vp9_cost_tokens(cpi->mb.partition_cost[i],
+                    cpi->common.fc.partition_prob[cpi->common.frame_type][i],
+                    vp9_partition_tree);
+
+  /*rough estimate for costing*/
+  vp9_init_mode_costs(cpi);
+
+  if (cpi->common.frame_type != KEY_FRAME) {
+    vp9_build_nmv_cost_table(
+        cpi->mb.nmvjointcost,
+        cpi->mb.e_mbd.allow_high_precision_mv ?
+        cpi->mb.nmvcost_hp : cpi->mb.nmvcost,
+        &cpi->common.fc.nmvc,
+        cpi->mb.e_mbd.allow_high_precision_mv, 1, 1);
+  }
+}
+
+int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) {
+  int i, error = 0;
+
+  for (i = 0; i < block_size; i++) {
+    int this_diff = coeff[i] - dqcoeff[i];
+    error += this_diff * this_diff;
+  }
+
+  return error;
+}
+
+static INLINE int cost_coeffs(VP9_COMMON *const cm, MACROBLOCK *mb,
+                              int plane, int block, PLANE_TYPE type,
+                              ENTROPY_CONTEXT *A,
+                              ENTROPY_CONTEXT *L,
+                              TX_SIZE tx_size,
+                              int y_blocks) {
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  int pt;
+  int c = 0;
+  int cost = 0, pad;
+  const int *scan, *nb;
+  const int eob = xd->plane[plane].eobs[block];
+  const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[plane].qcoeff,
+                                           block, 16);
+  const int ref = mbmi->ref_frame[0] != INTRA_FRAME;
+  unsigned int (*token_costs)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
+      mb->token_costs[tx_size][type][ref];
+  ENTROPY_CONTEXT above_ec, left_ec;
+  TX_TYPE tx_type = DCT_DCT;
+
+  const int segment_id = xd->mode_info_context->mbmi.segment_id;
+  unsigned int (*token_costs_noskip)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
+      mb->token_costs_noskip[tx_size][type][ref];
+
+  int seg_eob, default_eob;
+  uint8_t token_cache[1024];
+  const uint8_t * band_translate;
+
+  // Check for consistency of tx_size with mode info
+  assert((!type && !plane) || (type && plane));
+  if (type == PLANE_TYPE_Y_WITH_DC) {
+    assert(xd->mode_info_context->mbmi.txfm_size == tx_size);
+  } else {
+    TX_SIZE tx_size_uv = get_uv_tx_size(mbmi);
+    assert(tx_size == tx_size_uv);
+  }
+
+  switch (tx_size) {
+    case TX_4X4: {
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_4x4(xd, block) : DCT_DCT;
+      above_ec = A[0] != 0;
+      left_ec = L[0] != 0;
+      seg_eob = 16;
+      scan = get_scan_4x4(tx_type);
+      band_translate = vp9_coefband_trans_4x4;
+      break;
+    }
+    case TX_8X8: {
+      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
+      const int sz = 1 + b_width_log2(sb_type);
+      const int x = block & ((1 << sz) - 1), y = block - x;
+      TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
+      above_ec = (A[0] + A[1]) != 0;
+      left_ec = (L[0] + L[1]) != 0;
+      scan = get_scan_8x8(tx_type);
+      seg_eob = 64;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_16X16: {
+      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
+      const int sz = 2 + b_width_log2(sb_type);
+      const int x = block & ((1 << sz) - 1), y = block - x;
+      TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
+      scan = get_scan_16x16(tx_type);
+      seg_eob = 256;
+      above_ec = (A[0] + A[1] + A[2] + A[3]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3]) != 0;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_32X32:
+      scan = vp9_default_scan_32x32;
+      seg_eob = 1024;
+      above_ec = (A[0] + A[1] + A[2] + A[3] + A[4] + A[5] + A[6] + A[7]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3] + L[4] + L[5] + L[6] + L[7]) != 0;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    default:
+      abort();
+      break;
+  }
+  assert(eob <= seg_eob);
+
+  pt = combine_entropy_contexts(above_ec, left_ec);
+  nb = vp9_get_coef_neighbors_handle(scan, &pad);
+  default_eob = seg_eob;
+
+  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP))
+    seg_eob = 0;
+
+  /* sanity check to ensure that we do not have spurious non-zero q values */
+  if (eob < seg_eob)
+    assert(qcoeff_ptr[scan[eob]] == 0);
+
+  {
+    for (c = 0; c < eob; c++) {
+      int v = qcoeff_ptr[scan[c]];
+      int t = vp9_dct_value_tokens_ptr[v].token;
+      int band = get_coef_band(band_translate, c);
+      if (c)
+        pt = vp9_get_coef_context(scan, nb, pad, token_cache, c, default_eob);
+
+      if (!c || token_cache[scan[c - 1]])  // do not skip eob
+        cost += token_costs_noskip[band][pt][t] + vp9_dct_value_cost_ptr[v];
+      else
+        cost += token_costs[band][pt][t] + vp9_dct_value_cost_ptr[v];
+      token_cache[scan[c]] = vp9_pt_energy_class[t];
+    }
+    if (c < seg_eob) {
+      if (c)
+        pt = vp9_get_coef_context(scan, nb, pad, token_cache, c, default_eob);
+      cost += mb->token_costs_noskip[tx_size][type][ref]
+          [get_coef_band(band_translate, c)]
+          [pt][DCT_EOB_TOKEN];
+    }
+  }
+
+  // is eob first coefficient;
+  for (pt = 0; pt < (1 << tx_size); pt++) {
+    A[pt] = L[pt] = c > 0;
+  }
+
+  return cost;
+}
+
+static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
+                                     int (*r)[2], int *rate,
+                                     int *d, int *distortion,
+                                     int *s, int *skip,
+                                     int64_t txfm_cache[NB_TXFM_MODES],
+                                     TX_SIZE max_txfm_size) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+  vp9_prob skip_prob = vp9_get_pred_prob(cm, xd, PRED_MBSKIP);
+  int64_t rd[TX_SIZE_MAX_SB][2];
+  int n, m;
+  int s0, s1;
+
+  const vp9_prob *tx_probs = vp9_get_pred_probs(cm, xd, PRED_TX_SIZE);
+
+  for (n = TX_4X4; n <= max_txfm_size; n++) {
+    r[n][1] = r[n][0];
+    for (m = 0; m <= n - (n == max_txfm_size); m++) {
+      if (m == n)
+        r[n][1] += vp9_cost_zero(tx_probs[m]);
+      else
+        r[n][1] += vp9_cost_one(tx_probs[m]);
+    }
+  }
+
+  assert(skip_prob > 0);
+  s0 = vp9_cost_bit(skip_prob, 0);
+  s1 = vp9_cost_bit(skip_prob, 1);
+
+  for (n = TX_4X4; n <= max_txfm_size; n++) {
+    if (s[n]) {
+      rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, d[n]);
+    } else {
+      rd[n][0] = RDCOST(x->rdmult, x->rddiv, r[n][0] + s0, d[n]);
+      rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1] + s0, d[n]);
+    }
+  }
+
+  if (max_txfm_size == TX_32X32 &&
+      (cm->txfm_mode == ALLOW_32X32 ||
+       (cm->txfm_mode == TX_MODE_SELECT &&
+        rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
+        rd[TX_32X32][1] < rd[TX_4X4][1]))) {
+    mbmi->txfm_size = TX_32X32;
+  } else if (max_txfm_size >= TX_16X16 &&
+             (cm->txfm_mode == ALLOW_16X16 ||
+              cm->txfm_mode == ALLOW_32X32 ||
+              (cm->txfm_mode == TX_MODE_SELECT &&
+               rd[TX_16X16][1] < rd[TX_8X8][1] &&
+               rd[TX_16X16][1] < rd[TX_4X4][1]))) {
+    mbmi->txfm_size = TX_16X16;
+  } else if (cm->txfm_mode == ALLOW_8X8 ||
+             cm->txfm_mode == ALLOW_16X16 ||
+             cm->txfm_mode == ALLOW_32X32 ||
+           (cm->txfm_mode == TX_MODE_SELECT && rd[TX_8X8][1] < rd[TX_4X4][1])) {
+    mbmi->txfm_size = TX_8X8;
+  } else {
+    mbmi->txfm_size = TX_4X4;
+  }
+
+  *distortion = d[mbmi->txfm_size];
+  *rate       = r[mbmi->txfm_size][cm->txfm_mode == TX_MODE_SELECT];
+  *skip       = s[mbmi->txfm_size];
+
+  txfm_cache[ONLY_4X4] = rd[TX_4X4][0];
+  txfm_cache[ALLOW_8X8] = rd[TX_8X8][0];
+  txfm_cache[ALLOW_16X16] = rd[MIN(max_txfm_size, TX_16X16)][0];
+  txfm_cache[ALLOW_32X32] = rd[MIN(max_txfm_size, TX_32X32)][0];
+  if (max_txfm_size == TX_32X32 &&
+      rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
+      rd[TX_32X32][1] < rd[TX_4X4][1])
+    txfm_cache[TX_MODE_SELECT] = rd[TX_32X32][1];
+  else if (max_txfm_size >= TX_16X16 &&
+           rd[TX_16X16][1] < rd[TX_8X8][1] && rd[TX_16X16][1] < rd[TX_4X4][1])
+    txfm_cache[TX_MODE_SELECT] = rd[TX_16X16][1];
+  else
+    txfm_cache[TX_MODE_SELECT] = rd[TX_4X4][1] < rd[TX_8X8][1] ?
+                                 rd[TX_4X4][1] : rd[TX_8X8][1];
+}
+
+static int block_error(int16_t *coeff, int16_t *dqcoeff,
+                       int block_size, int shift) {
+  int i;
+  int64_t error = 0;
+
+  for (i = 0; i < block_size; i++) {
+    int this_diff = coeff[i] - dqcoeff[i];
+    error += (unsigned)this_diff * this_diff;
+  }
+  error >>= shift;
+
+  return error > INT_MAX ? INT_MAX : (int)error;
+}
+
+static int block_error_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize, int shift) {
+  const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+  return block_error(x->plane[0].coeff, x->e_mbd.plane[0].dqcoeff,
+                     16 << (bwl + bhl), shift);
+}
+
+static int block_error_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize, int shift) {
+  const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+  int64_t sum = 0;
+  int plane;
+
+  for (plane = 1; plane < MAX_MB_PLANE; plane++) {
+    const int subsampling = x->e_mbd.plane[plane].subsampling_x +
+                            x->e_mbd.plane[plane].subsampling_y;
+    sum += block_error(x->plane[plane].coeff, x->e_mbd.plane[plane].dqcoeff,
+                       16 << (bwl + bhl - subsampling), 0);
+  }
+  sum >>= shift;
+  return sum > INT_MAX ? INT_MAX : (int)sum;
+}
+
+struct rdcost_block_args {
+  VP9_COMMON *cm;
+  MACROBLOCK *x;
+  ENTROPY_CONTEXT t_above[16];
+  ENTROPY_CONTEXT t_left[16];
+  TX_SIZE tx_size;
+  int bw;
+  int bh;
+  int cost;
+};
+
+static void rdcost_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                         int ss_txfrm_size, void *arg) {
+  struct rdcost_block_args* args = arg;
+  int x_idx, y_idx;
+  MACROBLOCKD * const xd = &args->x->e_mbd;
+
+  txfrm_block_to_raster_xy(xd, bsize, plane, block, args->tx_size * 2, &x_idx,
+                           &y_idx);
+
+  args->cost += cost_coeffs(args->cm, args->x, plane, block,
+                            xd->plane[plane].plane_type, args->t_above + x_idx,
+                            args->t_left + y_idx, args->tx_size,
+                            args->bw * args->bh);
+}
+
+static int rdcost_plane(VP9_COMMON * const cm, MACROBLOCK *x, int plane,
+                        BLOCK_SIZE_TYPE bsize, TX_SIZE tx_size) {
+  MACROBLOCKD * const xd = &x->e_mbd;
+  const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
+  const int bhl = b_height_log2(bsize) - xd->plane[plane].subsampling_y;
+  const int bw = 1 << bwl, bh = 1 << bhl;
+  struct rdcost_block_args args = { cm, x, { 0 }, { 0 }, tx_size, bw, bh, 0 };
+
+  vpx_memcpy(&args.t_above, xd->plane[plane].above_context,
+             sizeof(ENTROPY_CONTEXT) * bw);
+  vpx_memcpy(&args.t_left, xd->plane[plane].left_context,
+             sizeof(ENTROPY_CONTEXT) * bh);
+
+  foreach_transformed_block_in_plane(xd, bsize, plane, rdcost_block, &args);
+
+  return args.cost;
+}
+
+static int rdcost_uv(VP9_COMMON *const cm, MACROBLOCK *x,
+                     BLOCK_SIZE_TYPE bsize, TX_SIZE tx_size) {
+  int cost = 0, plane;
+
+  for (plane = 1; plane < MAX_MB_PLANE; plane++) {
+    cost += rdcost_plane(cm, x, plane, bsize, tx_size);
+  }
+  return cost;
+}
+
+static void super_block_yrd_for_txfm(VP9_COMMON *const cm, MACROBLOCK *x,
+                                     int *rate, int *distortion, int *skippable,
+                                     BLOCK_SIZE_TYPE bsize, TX_SIZE tx_size) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  xd->mode_info_context->mbmi.txfm_size = tx_size;
+
+  if (xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME)
+    vp9_encode_intra_block_y(cm, x, bsize);
+  else
+    vp9_xform_quant_sby(cm, x, bsize);
+
+  *distortion = block_error_sby(x, bsize, tx_size == TX_32X32 ? 0 : 2);
+  *rate       = rdcost_plane(cm, x, 0, bsize, tx_size);
+  *skippable  = vp9_sby_is_skippable(xd, bsize);
+}
+
+static void super_block_yrd(VP9_COMP *cpi,
+                            MACROBLOCK *x, int *rate, int *distortion,
+                            int *skip, BLOCK_SIZE_TYPE bs,
+                            int64_t txfm_cache[NB_TXFM_MODES]) {
+  VP9_COMMON *const cm = &cpi->common;
+  int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB];
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  assert(bs == mbmi->sb_type);
+  if (mbmi->ref_frame[0] > INTRA_FRAME)
+    vp9_subtract_sby(x, bs);
+
+  if (cpi->speed > 4) {
+    if (bs >= BLOCK_SIZE_SB32X32) {
+      mbmi->txfm_size = TX_32X32;
+    } else if (bs >= BLOCK_SIZE_MB16X16) {
+      mbmi->txfm_size = TX_16X16;
+    } else if (bs >= BLOCK_SIZE_SB8X8) {
+      mbmi->txfm_size = TX_8X8;
+    } else {
+      mbmi->txfm_size = TX_4X4;
+    }
+    vpx_memset(txfm_cache, 0, NB_TXFM_MODES * sizeof(int64_t));
+    super_block_yrd_for_txfm(cm, x, rate, distortion, skip, bs,
+                             mbmi->txfm_size);
+    return;
+  }
+  if (bs >= BLOCK_SIZE_SB32X32)
+    super_block_yrd_for_txfm(cm, x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32],
+                             bs, TX_32X32);
+  if (bs >= BLOCK_SIZE_MB16X16)
+    super_block_yrd_for_txfm(cm, x, &r[TX_16X16][0], &d[TX_16X16], &s[TX_16X16],
+                             bs, TX_16X16);
+  super_block_yrd_for_txfm(cm, x, &r[TX_8X8][0], &d[TX_8X8], &s[TX_8X8], bs,
+                           TX_8X8);
+  super_block_yrd_for_txfm(cm, x, &r[TX_4X4][0], &d[TX_4X4], &s[TX_4X4], bs,
+                           TX_4X4);
+
+  choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s,
+                           skip, txfm_cache,
+                           TX_32X32 - (bs < BLOCK_SIZE_SB32X32)
+                           - (bs < BLOCK_SIZE_MB16X16));
+}
+
+static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
+                                     MB_PREDICTION_MODE *best_mode,
+                                     int *bmode_costs,
+                                     ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
+                                     int *bestrate, int *bestratey,
+                                     int *bestdistortion,
+                                     BLOCK_SIZE_TYPE bsize) {
+  MB_PREDICTION_MODE mode;
+  MACROBLOCKD *xd = &x->e_mbd;
+  int64_t best_rd = INT64_MAX;
+  int rate = 0;
+  int distortion;
+  VP9_COMMON *const cm = &cpi->common;
+  const int src_stride = x->plane[0].src.stride;
+  uint8_t *src, *dst;
+  int16_t *src_diff, *coeff;
+
+  ENTROPY_CONTEXT ta[2], tempa[2];
+  ENTROPY_CONTEXT tl[2], templ[2];
+  TX_TYPE tx_type = DCT_DCT;
+  TX_TYPE best_tx_type = DCT_DCT;
+  int bw = 1 << b_width_log2(bsize);
+  int bh = 1 << b_height_log2(bsize);
+  int idx, idy, block;
+  DECLARE_ALIGNED(16, int16_t, best_dqcoeff[4][16]);
+
+  assert(ib < 4);
+
+  vpx_memcpy(ta, a, sizeof(ta));
+  vpx_memcpy(tl, l, sizeof(tl));
+  xd->mode_info_context->mbmi.txfm_size = TX_4X4;
+
+  for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
+    int64_t this_rd;
+    int ratey = 0;
+
+    rate = bmode_costs[mode];
+    distortion = 0;
+
+    vpx_memcpy(tempa, ta, sizeof(ta));
+    vpx_memcpy(templ, tl, sizeof(tl));
+
+    for (idy = 0; idy < bh; ++idy) {
+      for (idx = 0; idx < bw; ++idx) {
+        block = ib + idy * 2 + idx;
+        xd->mode_info_context->bmi[block].as_mode.first = mode;
+        src = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, block,
+                                        x->plane[0].src.buf, src_stride);
+        src_diff = raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, block,
+                                             x->plane[0].src_diff);
+        coeff = BLOCK_OFFSET(x->plane[0].coeff, block, 16);
+        dst = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, block,
+                                        xd->plane[0].dst.buf,
+                                        xd->plane[0].dst.stride);
+        vp9_intra4x4_predict(xd, block, BLOCK_SIZE_SB8X8, mode,
+                             dst, xd->plane[0].dst.stride);
+        vp9_subtract_block(4, 4, src_diff, 8,
+                           src, src_stride,
+                           dst, xd->plane[0].dst.stride);
+
+        tx_type = get_tx_type_4x4(xd, block);
+        if (tx_type != DCT_DCT) {
+          vp9_short_fht4x4(src_diff, coeff, 8, tx_type);
+          x->quantize_b_4x4(x, block, tx_type, 16);
+        } else {
+          x->fwd_txm4x4(src_diff, coeff, 16);
+          x->quantize_b_4x4(x, block, tx_type, 16);
+        }
+
+        ratey += cost_coeffs(cm, x, 0, block, PLANE_TYPE_Y_WITH_DC,
+                             tempa + idx, templ + idy, TX_4X4, 16);
+        distortion += vp9_block_error(coeff, BLOCK_OFFSET(xd->plane[0].dqcoeff,
+                                                         block, 16), 16) >> 2;
+
+        if (best_tx_type != DCT_DCT)
+          vp9_short_iht4x4_add(BLOCK_OFFSET(xd->plane[0].dqcoeff, block, 16),
+                               dst, xd->plane[0].dst.stride, best_tx_type);
+        else
+          xd->inv_txm4x4_add(BLOCK_OFFSET(xd->plane[0].dqcoeff, block, 16),
+                             dst, xd->plane[0].dst.stride);
+      }
+    }
+
+    rate += ratey;
+    this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
+
+    if (this_rd < best_rd) {
+      *bestrate = rate;
+      *bestratey = ratey;
+      *bestdistortion = distortion;
+      best_rd = this_rd;
+      *best_mode = mode;
+      best_tx_type = tx_type;
+      vpx_memcpy(a, tempa, sizeof(tempa));
+      vpx_memcpy(l, templ, sizeof(templ));
+      for (idy = 0; idy < bh; ++idy) {
+        for (idx = 0; idx < bw; ++idx) {
+          block = ib + idy * 2 + idx;
+          vpx_memcpy(best_dqcoeff[idy * 2 + idx],
+                     BLOCK_OFFSET(xd->plane[0].dqcoeff, block, 16),
+                     sizeof(best_dqcoeff[0]));
+        }
+      }
+    }
+  }
+
+  for (idy = 0; idy < bh; ++idy) {
+    for (idx = 0; idx < bw; ++idx) {
+      block = ib + idy * 2 + idx;
+      xd->mode_info_context->bmi[block].as_mode.first = *best_mode;
+      dst = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, block,
+                                      xd->plane[0].dst.buf,
+                                      xd->plane[0].dst.stride);
+
+      vp9_intra4x4_predict(xd, block, BLOCK_SIZE_SB8X8, *best_mode,
+                           dst, xd->plane[0].dst.stride);
+      // inverse transform
+      if (best_tx_type != DCT_DCT)
+        vp9_short_iht4x4_add(best_dqcoeff[idy * 2 + idx], dst,
+                             xd->plane[0].dst.stride, best_tx_type);
+      else
+        xd->inv_txm4x4_add(best_dqcoeff[idy * 2 + idx], dst,
+                           xd->plane[0].dst.stride);
+    }
+  }
+
+  return best_rd;
+}
+
+static int64_t rd_pick_intra4x4mby_modes(VP9_COMP *cpi, MACROBLOCK *mb,
+                                         int *Rate, int *rate_y,
+                                         int *Distortion, int64_t best_rd) {
+  int i, j;
+  MACROBLOCKD *const xd = &mb->e_mbd;
+  BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+  int bw = 1 << b_width_log2(bsize);
+  int bh = 1 << b_height_log2(bsize);
+  int idx, idy;
+  int cost = 0;
+  int distortion = 0;
+  int tot_rate_y = 0;
+  int64_t total_rd = 0;
+  ENTROPY_CONTEXT t_above[4], t_left[4];
+  int *bmode_costs;
+  MODE_INFO *const mic = xd->mode_info_context;
+
+  vpx_memcpy(t_above, xd->plane[0].above_context, sizeof(t_above));
+  vpx_memcpy(t_left, xd->plane[0].left_context, sizeof(t_left));
+
+  bmode_costs = mb->mbmode_cost;
+
+  for (idy = 0; idy < 2; idy += bh) {
+    for (idx = 0; idx < 2; idx += bw) {
+      const int mis = xd->mode_info_stride;
+      MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(best_mode);
+      int UNINITIALIZED_IS_SAFE(r), UNINITIALIZED_IS_SAFE(ry);
+      int UNINITIALIZED_IS_SAFE(d);
+      i = idy * 2 + idx;
+
+      if (xd->frame_type == KEY_FRAME) {
+        const MB_PREDICTION_MODE A = above_block_mode(mic, i, mis);
+        const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
+                                     left_block_mode(mic, i) : DC_PRED;
+
+        bmode_costs  = mb->y_mode_costs[A][L];
+      }
+
+      total_rd += rd_pick_intra4x4block(cpi, mb, i, &best_mode, bmode_costs,
+                                        t_above + idx, t_left + idy,
+                                        &r, &ry, &d, bsize);
+      cost += r;
+      distortion += d;
+      tot_rate_y += ry;
+
+      mic->bmi[i].as_mode.first = best_mode;
+      for (j = 1; j < bh; ++j)
+        mic->bmi[i + j * 2].as_mode.first = best_mode;
+      for (j = 1; j < bw; ++j)
+        mic->bmi[i + j].as_mode.first = best_mode;
+
+      if (total_rd >= best_rd)
+        break;
+    }
+  }
+
+  if (total_rd >= best_rd)
+    return INT64_MAX;
+
+  *Rate = cost;
+  *rate_y = tot_rate_y;
+  *Distortion = distortion;
+  xd->mode_info_context->mbmi.mode = mic->bmi[3].as_mode.first;
+
+  return RDCOST(mb->rdmult, mb->rddiv, cost, distortion);
+}
+
+static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                                      int *rate, int *rate_tokenonly,
+                                      int *distortion, int *skippable,
+                                      BLOCK_SIZE_TYPE bsize,
+                                      int64_t txfm_cache[NB_TXFM_MODES]) {
+  MB_PREDICTION_MODE mode;
+  MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
+  MACROBLOCKD *const xd = &x->e_mbd;
+  int this_rate, this_rate_tokenonly;
+  int this_distortion, s;
+  int64_t best_rd = INT64_MAX, this_rd;
+  TX_SIZE UNINITIALIZED_IS_SAFE(best_tx);
+  int i;
+  int *bmode_costs = x->mbmode_cost;
+
+  if (bsize < BLOCK_SIZE_SB8X8) {
+    x->e_mbd.mode_info_context->mbmi.txfm_size = TX_4X4;
+    return best_rd;
+  }
+
+  for (i = 0; i < NB_TXFM_MODES; i++)
+    txfm_cache[i] = INT64_MAX;
+
+  /* Y Search for 32x32 intra prediction mode */
+  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
+    int64_t local_txfm_cache[NB_TXFM_MODES];
+    MODE_INFO *const mic = xd->mode_info_context;
+    const int mis = xd->mode_info_stride;
+
+    if (cpi->common.frame_type == KEY_FRAME) {
+      const MB_PREDICTION_MODE A = above_block_mode(mic, 0, mis);
+      const MB_PREDICTION_MODE L = xd->left_available ?
+                                   left_block_mode(mic, 0) : DC_PRED;
+
+      bmode_costs = x->y_mode_costs[A][L];
+    }
+    x->e_mbd.mode_info_context->mbmi.mode = mode;
+
+    super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion, &s,
+                    bsize, local_txfm_cache);
+
+    this_rate = this_rate_tokenonly + bmode_costs[mode];
+    this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
+
+    if (this_rd < best_rd) {
+      mode_selected   = mode;
+      best_rd         = this_rd;
+      best_tx         = x->e_mbd.mode_info_context->mbmi.txfm_size;
+      *rate           = this_rate;
+      *rate_tokenonly = this_rate_tokenonly;
+      *distortion     = this_distortion;
+      *skippable      = s;
+    }
+
+    for (i = 0; i < NB_TXFM_MODES; i++) {
+      int64_t adj_rd = this_rd + local_txfm_cache[i] -
+                       local_txfm_cache[cpi->common.txfm_mode];
+      if (adj_rd < txfm_cache[i]) {
+        txfm_cache[i] = adj_rd;
+      }
+    }
+  }
+
+  x->e_mbd.mode_info_context->mbmi.mode = mode_selected;
+  x->e_mbd.mode_info_context->mbmi.txfm_size = best_tx;
+
+  return best_rd;
+}
+
+static void super_block_uvrd_for_txfm(VP9_COMMON *const cm, MACROBLOCK *x,
+                                      int *rate, int *distortion,
+                                      int *skippable, BLOCK_SIZE_TYPE bsize,
+                                      TX_SIZE uv_tx_size) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  if (xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME)
+    vp9_encode_intra_block_uv(cm, x, bsize);
+  else
+    vp9_xform_quant_sbuv(cm, x, bsize);
+
+  *distortion = block_error_sbuv(x, bsize, uv_tx_size == TX_32X32 ? 0 : 2);
+  *rate       = rdcost_uv(cm, x, bsize, uv_tx_size);
+  *skippable  = vp9_sbuv_is_skippable(xd, bsize);
+}
+
+static void super_block_uvrd(VP9_COMMON *const cm, MACROBLOCK *x,
+                             int *rate, int *distortion, int *skippable,
+                             BLOCK_SIZE_TYPE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  if (mbmi->ref_frame[0] > INTRA_FRAME)
+    vp9_subtract_sbuv(x, bsize);
+
+  if (mbmi->txfm_size >= TX_32X32 && bsize >= BLOCK_SIZE_SB64X64) {
+    super_block_uvrd_for_txfm(cm, x, rate, distortion, skippable, bsize,
+                              TX_32X32);
+  } else if (mbmi->txfm_size >= TX_16X16 && bsize >= BLOCK_SIZE_SB32X32) {
+    super_block_uvrd_for_txfm(cm, x, rate, distortion, skippable, bsize,
+                              TX_16X16);
+  } else if (mbmi->txfm_size >= TX_8X8 && bsize >= BLOCK_SIZE_MB16X16) {
+    super_block_uvrd_for_txfm(cm, x, rate, distortion, skippable, bsize,
+                              TX_8X8);
+  } else {
+    super_block_uvrd_for_txfm(cm, x, rate, distortion, skippable, bsize,
+                              TX_4X4);
+  }
+}
+
+static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                                       int *rate, int *rate_tokenonly,
+                                       int *distortion, int *skippable,
+                                       BLOCK_SIZE_TYPE bsize) {
+  MB_PREDICTION_MODE mode;
+  MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
+  int64_t best_rd = INT64_MAX, this_rd;
+  int this_rate_tokenonly, this_rate;
+  int this_distortion, s;
+
+  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
+    x->e_mbd.mode_info_context->mbmi.uv_mode = mode;
+    super_block_uvrd(&cpi->common, x, &this_rate_tokenonly,
+                     &this_distortion, &s, bsize);
+    this_rate = this_rate_tokenonly +
+                x->intra_uv_mode_cost[x->e_mbd.frame_type][mode];
+    this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
+
+    if (this_rd < best_rd) {
+      mode_selected   = mode;
+      best_rd         = this_rd;
+      *rate           = this_rate;
+      *rate_tokenonly = this_rate_tokenonly;
+      *distortion     = this_distortion;
+      *skippable      = s;
+    }
+  }
+
+  x->e_mbd.mode_info_context->mbmi.uv_mode = mode_selected;
+
+  return best_rd;
+}
+
+int vp9_cost_mv_ref(VP9_COMP *cpi,
+                    MB_PREDICTION_MODE m,
+                    const int mode_context) {
+  MACROBLOCKD *xd = &cpi->mb.e_mbd;
+  int segment_id = xd->mode_info_context->mbmi.segment_id;
+
+  // Dont account for mode here if segment skip is enabled.
+  if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
+    VP9_COMMON *pc = &cpi->common;
+    assert(NEARESTMV <= m  &&  m <= NEWMV);
+    return cost_token(vp9_sb_mv_ref_tree,
+                      pc->fc.inter_mode_probs[mode_context],
+                      vp9_sb_mv_ref_encoding_array - NEARESTMV + m);
+  } else
+    return 0;
+}
+
+void vp9_set_mbmode_and_mvs(MACROBLOCK *x, MB_PREDICTION_MODE mb, int_mv *mv) {
+  x->e_mbd.mode_info_context->mbmi.mode = mb;
+  x->e_mbd.mode_info_context->mbmi.mv[0].as_int = mv->as_int;
+}
+
+static void joint_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+                                BLOCK_SIZE_TYPE bsize,
+                                int_mv *frame_mv,
+                                int mi_row, int mi_col,
+                                int_mv single_newmv[MAX_REF_FRAMES],
+                                int *rate_mv);
+static void single_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+                                 BLOCK_SIZE_TYPE bsize,
+                                 int mi_row, int mi_col,
+                                 int_mv *tmp_mv, int *rate_mv);
+
+static int labels2mode(MACROBLOCK *x, int i,
+                       MB_PREDICTION_MODE this_mode,
+                       int_mv *this_mv, int_mv *this_second_mv,
+                       int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES],
+                       int_mv seg_mvs[MAX_REF_FRAMES],
+                       int_mv *best_ref_mv,
+                       int_mv *second_best_ref_mv,
+                       int *mvjcost, int *mvcost[2], VP9_COMP *cpi) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MODE_INFO *const mic = xd->mode_info_context;
+  MB_MODE_INFO * mbmi = &mic->mbmi;
+  int cost = 0, thismvcost = 0;
+  int idx, idy;
+  int bw = 1 << b_width_log2(mbmi->sb_type);
+  int bh = 1 << b_height_log2(mbmi->sb_type);
+
+  /* We have to be careful retrieving previously-encoded motion vectors.
+   Ones from this macroblock have to be pulled from the BLOCKD array
+   as they have not yet made it to the bmi array in our MB_MODE_INFO. */
+  MB_PREDICTION_MODE m;
+
+  // the only time we should do costing for new motion vector or mode
+  // is when we are on a new label  (jbb May 08, 2007)
+  switch (m = this_mode) {
+    case NEWMV:
+      this_mv->as_int = seg_mvs[mbmi->ref_frame[0]].as_int;
+      thismvcost  = vp9_mv_bit_cost(this_mv, best_ref_mv, mvjcost, mvcost,
+                                    102, xd->allow_high_precision_mv);
+      if (mbmi->ref_frame[1] > 0) {
+        this_second_mv->as_int = seg_mvs[mbmi->ref_frame[1]].as_int;
+        thismvcost += vp9_mv_bit_cost(this_second_mv, second_best_ref_mv,
+                                      mvjcost, mvcost, 102,
+                                      xd->allow_high_precision_mv);
+      }
+      break;
+    case NEARESTMV:
+      this_mv->as_int = frame_mv[NEARESTMV][mbmi->ref_frame[0]].as_int;
+      if (mbmi->ref_frame[1] > 0)
+        this_second_mv->as_int =
+            frame_mv[NEARESTMV][mbmi->ref_frame[1]].as_int;
+      break;
+    case NEARMV:
+      this_mv->as_int = frame_mv[NEARMV][mbmi->ref_frame[0]].as_int;
+      if (mbmi->ref_frame[1] > 0)
+        this_second_mv->as_int =
+            frame_mv[NEARMV][mbmi->ref_frame[1]].as_int;
+      break;
+    case ZEROMV:
+      this_mv->as_int = 0;
+      if (mbmi->ref_frame[1] > 0)
+        this_second_mv->as_int = 0;
+      break;
+    default:
+      break;
+  }
+
+  cost = vp9_cost_mv_ref(cpi, this_mode,
+                         mbmi->mb_mode_context[mbmi->ref_frame[0]]);
+
+  mic->bmi[i].as_mv[0].as_int = this_mv->as_int;
+  if (mbmi->ref_frame[1] > 0)
+    mic->bmi[i].as_mv[1].as_int = this_second_mv->as_int;
+
+  x->partition_info->bmi[i].mode = m;
+  x->partition_info->bmi[i].mv.as_int = this_mv->as_int;
+  if (mbmi->ref_frame[1] > 0)
+    x->partition_info->bmi[i].second_mv.as_int = this_second_mv->as_int;
+  for (idy = 0; idy < bh; ++idy) {
+    for (idx = 0; idx < bw; ++idx) {
+      vpx_memcpy(&mic->bmi[i + idy * 2 + idx],
+                 &mic->bmi[i], sizeof(mic->bmi[i]));
+      vpx_memcpy(&x->partition_info->bmi[i + idy * 2 + idx],
+                 &x->partition_info->bmi[i],
+                 sizeof(x->partition_info->bmi[i]));
+    }
+  }
+
+  cost += thismvcost;
+  return cost;
+}
+
+static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
+                                       MACROBLOCK *x,
+                                       int i,
+                                       int *labelyrate,
+                                       int *distortion,
+                                       ENTROPY_CONTEXT *ta,
+                                       ENTROPY_CONTEXT *tl) {
+  int k;
+  MACROBLOCKD *xd = &x->e_mbd;
+  BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+  int bwl = b_width_log2(bsize), bw = 1 << bwl;
+  int bhl = b_height_log2(bsize), bh = 1 << bhl;
+  int idx, idy;
+  const int src_stride = x->plane[0].src.stride;
+  uint8_t* const src =
+  raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+                            x->plane[0].src.buf, src_stride);
+  int16_t* src_diff =
+  raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, i,
+                            x->plane[0].src_diff);
+  int16_t* coeff = BLOCK_OFFSET(x->plane[0].coeff, 16, i);
+  uint8_t* const pre =
+  raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+                            xd->plane[0].pre[0].buf,
+                            xd->plane[0].pre[0].stride);
+  uint8_t* const dst =
+  raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+                            xd->plane[0].dst.buf,
+                            xd->plane[0].dst.stride);
+  int thisdistortion = 0;
+  int thisrate = 0;
+
+  *labelyrate = 0;
+  *distortion = 0;
+
+  vp9_build_inter_predictor(pre,
+                            xd->plane[0].pre[0].stride,
+                            dst,
+                            xd->plane[0].dst.stride,
+                            &xd->mode_info_context->bmi[i].as_mv[0],
+                            &xd->scale_factor[0],
+                            4 * bw, 4 * bh, 0 /* no avg */, &xd->subpix);
+
+  // TODO(debargha): Make this work properly with the
+  // implicit-compoundinter-weight experiment when implicit
+  // weighting for splitmv modes is turned on.
+  if (xd->mode_info_context->mbmi.ref_frame[1] > 0) {
+    uint8_t* const second_pre =
+    raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+                              xd->plane[0].pre[1].buf,
+                              xd->plane[0].pre[1].stride);
+    vp9_build_inter_predictor(second_pre, xd->plane[0].pre[1].stride,
+                              dst, xd->plane[0].dst.stride,
+                              &xd->mode_info_context->bmi[i].as_mv[1],
+                              &xd->scale_factor[1], 4 * bw, 4 * bh, 1,
+                              &xd->subpix);
+  }
+
+  vp9_subtract_block(4 * bh, 4 * bw, src_diff, 8,
+                     src, src_stride,
+                     dst, xd->plane[0].dst.stride);
+
+  k = i;
+  for (idy = 0; idy < bh; ++idy) {
+    for (idx = 0; idx < bw; ++idx) {
+      k += (idy * 2 + idx);
+      src_diff = raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, k,
+                                           x->plane[0].src_diff);
+      coeff = BLOCK_OFFSET(x->plane[0].coeff, 16, k);
+      x->fwd_txm4x4(src_diff, coeff, 16);
+      x->quantize_b_4x4(x, k, DCT_DCT, 16);
+      thisdistortion += vp9_block_error(coeff,
+                                        BLOCK_OFFSET(xd->plane[0].dqcoeff,
+                                                     k, 16), 16);
+      thisrate += cost_coeffs(cm, x, 0, k, PLANE_TYPE_Y_WITH_DC,
+                              ta + (k & 1),
+                              tl + (k >> 1), TX_4X4, 16);
+    }
+  }
+  *distortion += thisdistortion;
+  *labelyrate += thisrate;
+
+  *distortion >>= 2;
+  return RDCOST(x->rdmult, x->rddiv, *labelyrate, *distortion);
+}
+
+typedef struct {
+  int_mv *ref_mv, *second_ref_mv;
+  int_mv mvp;
+
+  int64_t segment_rd;
+  int r;
+  int d;
+  int segment_yrate;
+  MB_PREDICTION_MODE modes[4];
+  int_mv mvs[4], second_mvs[4];
+  int eobs[4];
+  int mvthresh;
+} BEST_SEG_INFO;
+
+static INLINE int mv_check_bounds(MACROBLOCK *x, int_mv *mv) {
+  int r = 0;
+  r |= (mv->as_mv.row >> 3) < x->mv_row_min;
+  r |= (mv->as_mv.row >> 3) > x->mv_row_max;
+  r |= (mv->as_mv.col >> 3) < x->mv_col_min;
+  r |= (mv->as_mv.col >> 3) > x->mv_col_max;
+  return r;
+}
+
+static enum BlockSize get_block_size(int bw, int bh) {
+  if (bw == 4 && bh == 4)
+    return BLOCK_4X4;
+
+  if (bw == 4 && bh == 8)
+    return BLOCK_4X8;
+
+  if (bw == 8 && bh == 4)
+    return BLOCK_8X4;
+
+  if (bw == 8 && bh == 8)
+    return BLOCK_8X8;
+
+  if (bw == 8 && bh == 16)
+    return BLOCK_8X16;
+
+  if (bw == 16 && bh == 8)
+    return BLOCK_16X8;
+
+  if (bw == 16 && bh == 16)
+    return BLOCK_16X16;
+
+  if (bw == 32 && bh == 32)
+    return BLOCK_32X32;
+
+  if (bw == 32 && bh == 16)
+    return BLOCK_32X16;
+
+  if (bw == 16 && bh == 32)
+    return BLOCK_16X32;
+
+  if (bw == 64 && bh == 32)
+    return BLOCK_64X32;
+
+  if (bw == 32 && bh == 64)
+    return BLOCK_32X64;
+
+  if (bw == 64 && bh == 64)
+    return BLOCK_64X64;
+
+  assert(0);
+  return -1;
+}
+
+static INLINE void mi_buf_shift(MACROBLOCK *x, int i) {
+  MB_MODE_INFO *mbmi = &x->e_mbd.mode_info_context->mbmi;
+  x->plane[0].src.buf =
+      raster_block_offset_uint8(&x->e_mbd, BLOCK_SIZE_SB8X8, 0, i,
+                                x->plane[0].src.buf,
+                                x->plane[0].src.stride);
+  assert(((intptr_t)x->e_mbd.plane[0].pre[0].buf & 0x7) == 0);
+  x->e_mbd.plane[0].pre[0].buf =
+      raster_block_offset_uint8(&x->e_mbd, BLOCK_SIZE_SB8X8, 0, i,
+                                x->e_mbd.plane[0].pre[0].buf,
+                                x->e_mbd.plane[0].pre[0].stride);
+  if (mbmi->ref_frame[1])
+    x->e_mbd.plane[0].pre[1].buf =
+        raster_block_offset_uint8(&x->e_mbd, BLOCK_SIZE_SB8X8, 0, i,
+                                  x->e_mbd.plane[0].pre[1].buf,
+                                  x->e_mbd.plane[0].pre[1].stride);
+}
+
+static INLINE void mi_buf_restore(MACROBLOCK *x, struct buf_2d orig_src,
+                                  struct buf_2d orig_pre[2]) {
+  MB_MODE_INFO *mbmi = &x->e_mbd.mode_info_context->mbmi;
+  x->plane[0].src = orig_src;
+  x->e_mbd.plane[0].pre[0] = orig_pre[0];
+  if (mbmi->ref_frame[1])
+    x->e_mbd.plane[0].pre[1] = orig_pre[1];
+}
+
+static void rd_check_segment_txsize(VP9_COMP *cpi, MACROBLOCK *x,
+                                    BEST_SEG_INFO *bsi,
+                                    int_mv seg_mvs[4][MAX_REF_FRAMES],
+                                    int mi_row, int mi_col) {
+  int i, j;
+  int br = 0, bd = 0;
+  MB_PREDICTION_MODE this_mode;
+  MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
+  const int label_count = 4;
+  int64_t this_segment_rd = 0, other_segment_rd;
+  int label_mv_thresh;
+  int rate = 0;
+  int sbr = 0, sbd = 0;
+  int segmentyrate = 0;
+  int best_eobs[4] = { 0 };
+  BLOCK_SIZE_TYPE bsize = mbmi->sb_type;
+  int bwl = b_width_log2(bsize), bw = 1 << bwl;
+  int bhl = b_height_log2(bsize), bh = 1 << bhl;
+  int idx, idy;
+  vp9_variance_fn_ptr_t *v_fn_ptr;
+  ENTROPY_CONTEXT t_above[4], t_left[4];
+  ENTROPY_CONTEXT t_above_b[4], t_left_b[4];
+
+  vpx_memcpy(t_above, x->e_mbd.plane[0].above_context, sizeof(t_above));
+  vpx_memcpy(t_left, x->e_mbd.plane[0].left_context, sizeof(t_left));
+
+  v_fn_ptr = &cpi->fn_ptr[get_block_size(4 << bwl, 4 << bhl)];
+
+  // 64 makes this threshold really big effectively
+  // making it so that we very rarely check mvs on
+  // segments.   setting this to 1 would make mv thresh
+  // roughly equal to what it is for macroblocks
+  label_mv_thresh = 1 * bsi->mvthresh / label_count;
+
+  // Segmentation method overheads
+  other_segment_rd = this_segment_rd;
+
+  for (idy = 0; idy < 2; idy += bh) {
+    for (idx = 0; idx < 2; idx += bw) {
+      // TODO(jingning,rbultje): rewrite the rate-distortion optimization
+      // loop for 4x4/4x8/8x4 block coding. to be replaced with new rd loop
+      int_mv mode_mv[MB_MODE_COUNT], second_mode_mv[MB_MODE_COUNT];
+      int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
+      int64_t best_label_rd = INT64_MAX, best_other_rd = INT64_MAX;
+      MB_PREDICTION_MODE mode_selected = ZEROMV;
+      int bestlabelyrate = 0;
+      i = idy * 2 + idx;
+
+      frame_mv[ZEROMV][mbmi->ref_frame[0]].as_int = 0;
+      frame_mv[ZEROMV][mbmi->ref_frame[1]].as_int = 0;
+      vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd,
+                                    &frame_mv[NEARESTMV][mbmi->ref_frame[0]],
+                                    &frame_mv[NEARMV][mbmi->ref_frame[0]],
+                                    i, 0);
+      if (mbmi->ref_frame[1] > 0)
+        vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd,
+                                   &frame_mv[NEARESTMV][mbmi->ref_frame[1]],
+                                   &frame_mv[NEARMV][mbmi->ref_frame[1]],
+                                   i, 1);
+
+      // search for the best motion vector on this segment
+      for (this_mode = NEARESTMV; this_mode <= NEWMV; ++this_mode) {
+        int64_t this_rd;
+        int distortion;
+        int labelyrate;
+        ENTROPY_CONTEXT t_above_s[4], t_left_s[4];
+        const struct buf_2d orig_src = x->plane[0].src;
+        struct buf_2d orig_pre[2];
+
+        vpx_memcpy(orig_pre, x->e_mbd.plane[0].pre, sizeof(orig_pre));
+
+        vpx_memcpy(t_above_s, t_above, sizeof(t_above_s));
+        vpx_memcpy(t_left_s, t_left, sizeof(t_left_s));
+
+        // motion search for newmv (single predictor case only)
+        if (mbmi->ref_frame[1] <= 0 && this_mode == NEWMV) {
+          int step_param = 0;
+          int further_steps;
+          int thissme, bestsme = INT_MAX;
+          int sadpb = x->sadperbit4;
+          int_mv mvp_full;
+
+          /* Is the best so far sufficiently good that we cant justify doing
+           * and new motion search. */
+          if (best_label_rd < label_mv_thresh)
+            break;
+
+          if (cpi->compressor_speed) {
+            // use previous block's result as next block's MV predictor.
+            if (i > 0) {
+              bsi->mvp.as_int =
+              x->e_mbd.mode_info_context->bmi[i - 1].as_mv[0].as_int;
+              if (i == 2)
+                bsi->mvp.as_int =
+                x->e_mbd.mode_info_context->bmi[i - 2].as_mv[0].as_int;
+              step_param = 2;
+            }
+          }
+
+          further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
+
+          mvp_full.as_mv.row = bsi->mvp.as_mv.row >> 3;
+          mvp_full.as_mv.col = bsi->mvp.as_mv.col >> 3;
+
+          // adjust src pointer for this block
+          mi_buf_shift(x, i);
+          bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
+                                           sadpb, further_steps, 0, v_fn_ptr,
+                                           bsi->ref_mv, &mode_mv[NEWMV]);
+
+          // Should we do a full search (best quality only)
+          if (cpi->compressor_speed == 0) {
+            /* Check if mvp_full is within the range. */
+            clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max,
+                     x->mv_row_min, x->mv_row_max);
+
+            thissme = cpi->full_search_sad(x, &mvp_full,
+                                           sadpb, 16, v_fn_ptr,
+                                           x->nmvjointcost, x->mvcost,
+                                           bsi->ref_mv, i);
+
+            if (thissme < bestsme) {
+              bestsme = thissme;
+              mode_mv[NEWMV].as_int =
+                  x->e_mbd.mode_info_context->bmi[i].as_mv[0].as_int;
+            } else {
+              /* The full search result is actually worse so re-instate the
+               * previous best vector */
+              x->e_mbd.mode_info_context->bmi[i].as_mv[0].as_int =
+                  mode_mv[NEWMV].as_int;
+            }
+          }
+
+          if (bestsme < INT_MAX) {
+            int distortion;
+            unsigned int sse;
+            cpi->find_fractional_mv_step(x, &mode_mv[NEWMV],
+                                         bsi->ref_mv, x->errorperbit, v_fn_ptr,
+                                         x->nmvjointcost, x->mvcost,
+                                         &distortion, &sse);
+
+            // safe motion search result for use in compound prediction
+            seg_mvs[i][mbmi->ref_frame[0]].as_int = mode_mv[NEWMV].as_int;
+          }
+
+          // restore src pointers
+          mi_buf_restore(x, orig_src, orig_pre);
+        } else if (mbmi->ref_frame[1] > 0 && this_mode == NEWMV) {
+          if (seg_mvs[i][mbmi->ref_frame[1]].as_int == INVALID_MV ||
+              seg_mvs[i][mbmi->ref_frame[0]].as_int == INVALID_MV)
+            continue;
+
+          // adjust src pointers
+          mi_buf_shift(x, i);
+          if (cpi->sf.comp_inter_joint_search_thresh < bsize) {
+            int rate_mv;
+            joint_motion_search(cpi, x, bsize, frame_mv[this_mode],
+                                mi_row, mi_col, seg_mvs[i],
+                                &rate_mv);
+            seg_mvs[i][mbmi->ref_frame[0]].as_int =
+                frame_mv[this_mode][mbmi->ref_frame[0]].as_int;
+            seg_mvs[i][mbmi->ref_frame[1]].as_int =
+                frame_mv[this_mode][mbmi->ref_frame[1]].as_int;
+          }
+          // restore src pointers
+          mi_buf_restore(x, orig_src, orig_pre);
+        }
+
+        rate = labels2mode(x, i, this_mode, &mode_mv[this_mode],
+                           &second_mode_mv[this_mode], frame_mv, seg_mvs[i],
+                           bsi->ref_mv, bsi->second_ref_mv, x->nmvjointcost,
+                           x->mvcost, cpi);
+
+        // Trap vectors that reach beyond the UMV borders
+        if (((mode_mv[this_mode].as_mv.row >> 3) < x->mv_row_min) ||
+            ((mode_mv[this_mode].as_mv.row >> 3) > x->mv_row_max) ||
+            ((mode_mv[this_mode].as_mv.col >> 3) < x->mv_col_min) ||
+            ((mode_mv[this_mode].as_mv.col >> 3) > x->mv_col_max)) {
+          continue;
+        }
+        if (mbmi->ref_frame[1] > 0 &&
+            mv_check_bounds(x, &second_mode_mv[this_mode]))
+          continue;
+
+        this_rd = encode_inter_mb_segment(&cpi->common,
+                                          x, i, &labelyrate,
+                                          &distortion, t_above_s, t_left_s);
+        this_rd += RDCOST(x->rdmult, x->rddiv, rate, 0);
+        rate += labelyrate;
+
+        if (this_rd < best_label_rd) {
+          sbr = rate;
+          sbd = distortion;
+          bestlabelyrate = labelyrate;
+          mode_selected = this_mode;
+          best_label_rd = this_rd;
+          best_eobs[i] = x->e_mbd.plane[0].eobs[i];
+          vpx_memcpy(t_above_b, t_above_s, sizeof(t_above_s));
+          vpx_memcpy(t_left_b, t_left_s, sizeof(t_left_s));
+        }
+      } /*for each 4x4 mode*/
+
+      vpx_memcpy(t_above, t_above_b, sizeof(t_above));
+      vpx_memcpy(t_left, t_left_b, sizeof(t_left));
+
+      labels2mode(x, i, mode_selected, &mode_mv[mode_selected],
+                  &second_mode_mv[mode_selected], frame_mv, seg_mvs[i],
+                  bsi->ref_mv, bsi->second_ref_mv, x->nmvjointcost,
+                  x->mvcost, cpi);
+
+      br += sbr;
+      bd += sbd;
+      segmentyrate += bestlabelyrate;
+      this_segment_rd += best_label_rd;
+      other_segment_rd += best_other_rd;
+
+      for (j = 1; j < bh; ++j)
+        vpx_memcpy(&x->partition_info->bmi[i + j * 2],
+                   &x->partition_info->bmi[i],
+                   sizeof(x->partition_info->bmi[i]));
+      for (j = 1; j < bw; ++j)
+        vpx_memcpy(&x->partition_info->bmi[i + j],
+                   &x->partition_info->bmi[i],
+                   sizeof(x->partition_info->bmi[i]));
+    }
+  } /* for each label */
+
+  if (this_segment_rd < bsi->segment_rd) {
+    bsi->r = br;
+    bsi->d = bd;
+    bsi->segment_yrate = segmentyrate;
+    bsi->segment_rd = this_segment_rd;
+
+    // store everything needed to come back to this!!
+    for (i = 0; i < 4; i++) {
+      bsi->mvs[i].as_mv = x->partition_info->bmi[i].mv.as_mv;
+      if (mbmi->ref_frame[1] > 0)
+        bsi->second_mvs[i].as_mv = x->partition_info->bmi[i].second_mv.as_mv;
+      bsi->modes[i] = x->partition_info->bmi[i].mode;
+      bsi->eobs[i] = best_eobs[i];
+    }
+  }
+}
+
+static int rd_pick_best_mbsegmentation(VP9_COMP *cpi, MACROBLOCK *x,
+                                       int_mv *best_ref_mv,
+                                       int_mv *second_best_ref_mv,
+                                       int64_t best_rd,
+                                       int *returntotrate,
+                                       int *returnyrate,
+                                       int *returndistortion,
+                                       int *skippable, int mvthresh,
+                                       int_mv seg_mvs[4][MAX_REF_FRAMES],
+                                       int mi_row, int mi_col) {
+  int i;
+  BEST_SEG_INFO bsi;
+  MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi;
+
+  vpx_memset(&bsi, 0, sizeof(bsi));
+
+  bsi.segment_rd = best_rd;
+  bsi.ref_mv = best_ref_mv;
+  bsi.second_ref_mv = second_best_ref_mv;
+  bsi.mvp.as_int = best_ref_mv->as_int;
+  bsi.mvthresh = mvthresh;
+
+  for (i = 0; i < 4; i++)
+    bsi.modes[i] = ZEROMV;
+
+  rd_check_segment_txsize(cpi, x, &bsi, seg_mvs, mi_row, mi_col);
+
+  /* set it to the best */
+  for (i = 0; i < 4; i++) {
+    x->e_mbd.mode_info_context->bmi[i].as_mv[0].as_int = bsi.mvs[i].as_int;
+    if (mbmi->ref_frame[1] > 0)
+      x->e_mbd.mode_info_context->bmi[i].as_mv[1].as_int =
+      bsi.second_mvs[i].as_int;
+    x->e_mbd.plane[0].eobs[i] = bsi.eobs[i];
+  }
+
+  /* save partitions */
+  x->partition_info->count = 4;
+
+  for (i = 0; i < x->partition_info->count; i++) {
+    x->partition_info->bmi[i].mode = bsi.modes[i];
+    x->partition_info->bmi[i].mv.as_mv = bsi.mvs[i].as_mv;
+    if (mbmi->ref_frame[1] > 0)
+      x->partition_info->bmi[i].second_mv.as_mv = bsi.second_mvs[i].as_mv;
+  }
+  /*
+   * used to set mbmi->mv.as_int
+   */
+  x->partition_info->bmi[3].mv.as_int = bsi.mvs[3].as_int;
+  if (mbmi->ref_frame[1] > 0)
+    x->partition_info->bmi[3].second_mv.as_int = bsi.second_mvs[3].as_int;
+
+  *returntotrate = bsi.r;
+  *returndistortion = bsi.d;
+  *returnyrate = bsi.segment_yrate;
+  *skippable = vp9_sby_is_skippable(&x->e_mbd, BLOCK_SIZE_SB8X8);
+  mbmi->mode = bsi.modes[3];
+
+  return (int)(bsi.segment_rd);
+}
+
+static void mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
+                    uint8_t *ref_y_buffer, int ref_y_stride,
+                    int ref_frame, enum BlockSize block_size ) {
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  int_mv this_mv;
+  int i;
+  int zero_seen = 0;
+  int best_index = 0;
+  int best_sad = INT_MAX;
+  int this_sad = INT_MAX;
+
+  uint8_t *src_y_ptr = x->plane[0].src.buf;
+  uint8_t *ref_y_ptr;
+  int row_offset, col_offset;
+
+  // Get the sad for each candidate reference mv
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; i++) {
+    this_mv.as_int = mbmi->ref_mvs[ref_frame][i].as_int;
+
+    // The list is at an end if we see 0 for a second time.
+    if (!this_mv.as_int && zero_seen)
+      break;
+    zero_seen = zero_seen || !this_mv.as_int;
+
+    row_offset = this_mv.as_mv.row >> 3;
+    col_offset = this_mv.as_mv.col >> 3;
+    ref_y_ptr = ref_y_buffer + (ref_y_stride * row_offset) + col_offset;
+
+    // Find sad for current vector.
+    this_sad = cpi->fn_ptr[block_size].sdf(src_y_ptr, x->plane[0].src.stride,
+                                           ref_y_ptr, ref_y_stride,
+                                           0x7fffffff);
+
+    // Note if it is the best so far.
+    if (this_sad < best_sad) {
+      best_sad = this_sad;
+      best_index = i;
+    }
+  }
+
+  // Note the index of the mv that worked best in the reference list.
+  x->mv_best_ref_index[ref_frame] = best_index;
+}
+
+static void estimate_ref_frame_costs(VP9_COMP *cpi, int segment_id,
+                                     unsigned int *ref_costs_single,
+                                     unsigned int *ref_costs_comp,
+                                     vp9_prob *comp_mode_p) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  int seg_ref_active = vp9_segfeature_active(xd, segment_id,
+                                             SEG_LVL_REF_FRAME);
+  if (seg_ref_active) {
+    vpx_memset(ref_costs_single, 0, MAX_REF_FRAMES * sizeof(*ref_costs_single));
+    vpx_memset(ref_costs_comp,   0, MAX_REF_FRAMES * sizeof(*ref_costs_comp));
+    *comp_mode_p = 128;
+  } else {
+    vp9_prob intra_inter_p = vp9_get_pred_prob(cm, xd, PRED_INTRA_INTER);
+    vp9_prob comp_inter_p = 128;
+
+    if (cm->comp_pred_mode == HYBRID_PREDICTION) {
+      comp_inter_p = vp9_get_pred_prob(cm, xd, PRED_COMP_INTER_INTER);
+      *comp_mode_p = comp_inter_p;
+    } else {
+      *comp_mode_p = 128;
+    }
+
+    ref_costs_single[INTRA_FRAME] = vp9_cost_bit(intra_inter_p, 0);
+
+    if (cm->comp_pred_mode != COMP_PREDICTION_ONLY) {
+      vp9_prob ref_single_p1 = vp9_get_pred_prob(cm, xd, PRED_SINGLE_REF_P1);
+      vp9_prob ref_single_p2 = vp9_get_pred_prob(cm, xd, PRED_SINGLE_REF_P2);
+      unsigned int base_cost = vp9_cost_bit(intra_inter_p, 1);
+
+      if (cm->comp_pred_mode == HYBRID_PREDICTION)
+        base_cost += vp9_cost_bit(comp_inter_p, 0);
+
+      ref_costs_single[LAST_FRAME] = ref_costs_single[GOLDEN_FRAME] =
+          ref_costs_single[ALTREF_FRAME] = base_cost;
+      ref_costs_single[LAST_FRAME]   += vp9_cost_bit(ref_single_p1, 0);
+      ref_costs_single[GOLDEN_FRAME] += vp9_cost_bit(ref_single_p1, 1);
+      ref_costs_single[ALTREF_FRAME] += vp9_cost_bit(ref_single_p1, 1);
+      ref_costs_single[GOLDEN_FRAME] += vp9_cost_bit(ref_single_p2, 0);
+      ref_costs_single[ALTREF_FRAME] += vp9_cost_bit(ref_single_p2, 1);
+    } else {
+      ref_costs_single[LAST_FRAME]   = 512;
+      ref_costs_single[GOLDEN_FRAME] = 512;
+      ref_costs_single[ALTREF_FRAME] = 512;
+    }
+    if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY) {
+      vp9_prob ref_comp_p = vp9_get_pred_prob(cm, xd, PRED_COMP_REF_P);
+      unsigned int base_cost = vp9_cost_bit(intra_inter_p, 1);
+
+      if (cm->comp_pred_mode == HYBRID_PREDICTION)
+        base_cost += vp9_cost_bit(comp_inter_p, 1);
+
+      ref_costs_comp[LAST_FRAME]   = base_cost + vp9_cost_bit(ref_comp_p, 0);
+      ref_costs_comp[GOLDEN_FRAME] = base_cost + vp9_cost_bit(ref_comp_p, 1);
+    } else {
+      ref_costs_comp[LAST_FRAME]   = 512;
+      ref_costs_comp[GOLDEN_FRAME] = 512;
+    }
+  }
+}
+
+static void store_coding_context(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
+                                 int mode_index,
+                                 PARTITION_INFO *partition,
+                                 int_mv *ref_mv,
+                                 int_mv *second_ref_mv,
+                                 int64_t comp_pred_diff[NB_PREDICTION_TYPES],
+                                 int64_t txfm_size_diff[NB_TXFM_MODES]) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+
+  // Take a snapshot of the coding context so it can be
+  // restored if we decide to encode this way
+  ctx->skip = x->skip;
+  ctx->best_mode_index = mode_index;
+  ctx->mic = *xd->mode_info_context;
+
+  if (partition)
+    ctx->partition_info = *partition;
+
+  ctx->best_ref_mv.as_int = ref_mv->as_int;
+  ctx->second_best_ref_mv.as_int = second_ref_mv->as_int;
+
+  ctx->single_pred_diff = (int)comp_pred_diff[SINGLE_PREDICTION_ONLY];
+  ctx->comp_pred_diff   = (int)comp_pred_diff[COMP_PREDICTION_ONLY];
+  ctx->hybrid_pred_diff = (int)comp_pred_diff[HYBRID_PREDICTION];
+
+  memcpy(ctx->txfm_rd_diff, txfm_size_diff, sizeof(ctx->txfm_rd_diff));
+}
+
+static void setup_pred_block(const MACROBLOCKD *xd,
+                             struct buf_2d dst[MAX_MB_PLANE],
+                             const YV12_BUFFER_CONFIG *src,
+                             int mi_row, int mi_col,
+                             const struct scale_factors *scale,
+                             const struct scale_factors *scale_uv) {
+  int i;
+
+  dst[0].buf = src->y_buffer;
+  dst[0].stride = src->y_stride;
+  dst[1].buf = src->u_buffer;
+  dst[2].buf = src->v_buffer;
+  dst[1].stride = dst[2].stride = src->uv_stride;
+#if CONFIG_ALPHA
+  dst[3].buf = src->alpha_buffer;
+  dst[3].stride = src->alpha_stride;
+#endif
+
+  // TODO(jkoleszar): Make scale factors per-plane data
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    setup_pred_plane(dst + i, dst[i].buf, dst[i].stride, mi_row, mi_col,
+                     i ? scale_uv : scale,
+                     xd->plane[i].subsampling_x, xd->plane[i].subsampling_y);
+  }
+}
+
+static void setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
+                               int idx, MV_REFERENCE_FRAME frame_type,
+                               enum BlockSize block_size,
+                               int mi_row, int mi_col,
+                               int_mv frame_nearest_mv[MAX_REF_FRAMES],
+                               int_mv frame_near_mv[MAX_REF_FRAMES],
+                               struct buf_2d yv12_mb[4][MAX_MB_PLANE],
+                               struct scale_factors scale[MAX_REF_FRAMES]) {
+  VP9_COMMON *cm = &cpi->common;
+  YV12_BUFFER_CONFIG *yv12 = &cm->yv12_fb[cpi->common.ref_frame_map[idx]];
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  // set up scaling factors
+  scale[frame_type] = cpi->common.active_ref_scale[frame_type - 1];
+
+  scale[frame_type].x_offset_q4 =
+      ROUND_POWER_OF_TWO(mi_col * MI_SIZE * scale[frame_type].x_scale_fp,
+       VP9_REF_SCALE_SHIFT) & 0xf;
+  scale[frame_type].y_offset_q4 =
+      ROUND_POWER_OF_TWO(mi_row * MI_SIZE * scale[frame_type].y_scale_fp,
+       VP9_REF_SCALE_SHIFT) & 0xf;
+
+  // TODO(jkoleszar): Is the UV buffer ever used here? If so, need to make this
+  // use the UV scaling factors.
+  setup_pred_block(xd, yv12_mb[frame_type], yv12, mi_row, mi_col,
+                   &scale[frame_type], &scale[frame_type]);
+
+  // Gets an initial list of candidate vectors from neighbours and orders them
+  vp9_find_mv_refs(&cpi->common, xd, xd->mode_info_context,
+                   xd->prev_mode_info_context,
+                   frame_type,
+                   mbmi->ref_mvs[frame_type],
+                   cpi->common.ref_frame_sign_bias);
+
+  // Candidate refinement carried out at encoder and decoder
+  vp9_find_best_ref_mvs(xd,
+                        mbmi->ref_mvs[frame_type],
+                        &frame_nearest_mv[frame_type],
+                        &frame_near_mv[frame_type]);
+
+  // Further refinement that is encode side only to test the top few candidates
+  // in full and choose the best as the centre point for subsequent searches.
+  // The current implementation doesn't support scaling.
+  if (scale[frame_type].x_scale_fp == (1 << VP9_REF_SCALE_SHIFT) &&
+      scale[frame_type].y_scale_fp == (1 << VP9_REF_SCALE_SHIFT))
+    mv_pred(cpi, x, yv12_mb[frame_type][0].buf, yv12->y_stride,
+            frame_type, block_size);
+}
+
+static YV12_BUFFER_CONFIG *get_scaled_ref_frame(VP9_COMP *cpi, int ref_frame) {
+  YV12_BUFFER_CONFIG *scaled_ref_frame = NULL;
+  int fb = get_ref_frame_idx(cpi, ref_frame);
+  if (cpi->scaled_ref_idx[fb] != cpi->common.ref_frame_map[fb])
+    scaled_ref_frame = &cpi->common.yv12_fb[cpi->scaled_ref_idx[fb]];
+  return scaled_ref_frame;
+}
+
+static void model_rd_from_var_lapndz(int var, int n, int qstep,
+                                     int *rate, int *dist) {
+  // This function models the rate and distortion for a Laplacian
+  // source with given variance when quantized with a uniform quantizer
+  // with given stepsize. The closed form expressions are in:
+  // Hang and Chen, "Source Model for transform video coder and its
+  // application - Part I: Fundamental Theory", IEEE Trans. Circ.
+  // Sys. for Video Tech., April 1997.
+  // The function is implemented as piecewise approximation to the
+  // exact computation.
+  // TODO(debargha): Implement the functions by interpolating from a
+  // look-up table
+  vp9_clear_system_state();
+  if (var == 0 || n == 0) {
+    *rate = 0;
+    *dist = 0;
+  } else {
+    double D, R;
+    double s2 = (double) var / n;
+    double s = sqrt(s2);
+    double x = qstep / s;
+    if (x > 1.0) {
+      double y = exp(-x / 2);
+      double y2 = y * y;
+      D = 2.069981728764738 * y2 - 2.764286806516079 * y + 1.003956960819275;
+      R = 0.924056758535089 * y2 + 2.738636469814024 * y - 0.005169662030017;
+    } else {
+      double x2 = x * x;
+      D = 0.075303187668830 * x2 + 0.004296954321112 * x - 0.000413209252807;
+      if (x > 0.125)
+        R = 1 / (-0.03459733614226 * x2 + 0.36561675733603 * x +
+                 0.1626989668625);
+      else
+        R = -1.442252874826093 * log(x) + 1.944647760719664;
+    }
+    if (R < 0) {
+      *rate = 0;
+      *dist = var;
+    } else {
+      *rate = (n * R * 256 + 0.5);
+      *dist = (n * D * s2 + 0.5);
+    }
+  }
+  vp9_clear_system_state();
+}
+
+static enum BlockSize get_plane_block_size(BLOCK_SIZE_TYPE bsize,
+                                           struct macroblockd_plane *pd) {
+  return get_block_size(plane_block_width(bsize, pd),
+                        plane_block_height(bsize, pd));
+}
+
+static void model_rd_for_sb(VP9_COMP *cpi, BLOCK_SIZE_TYPE bsize,
+                            MACROBLOCK *x, MACROBLOCKD *xd,
+                            int *out_rate_sum, int *out_dist_sum) {
+  // Note our transform coeffs are 8 times an orthogonal transform.
+  // Hence quantizer step is also 8 times. To get effective quantizer
+  // we need to divide by 8 before sending to modeling function.
+  unsigned int sse, var;
+  int i, rate_sum = 0, dist_sum = 0;
+
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+    struct macroblock_plane *const p = &x->plane[i];
+    struct macroblockd_plane *const pd = &xd->plane[i];
+
+    // TODO(dkovalev) the same code in get_plane_block_size
+    const int bw = plane_block_width(bsize, pd);
+    const int bh = plane_block_height(bsize, pd);
+    const enum BlockSize bs = get_block_size(bw, bh);
+    int rate, dist;
+    var = cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride,
+                             pd->dst.buf, pd->dst.stride, &sse);
+    model_rd_from_var_lapndz(var, bw * bh, pd->dequant[1] >> 3, &rate, &dist);
+
+    rate_sum += rate;
+    dist_sum += dist;
+  }
+
+  *out_rate_sum = rate_sum;
+  *out_dist_sum = dist_sum;
+}
+
+static INLINE int get_switchable_rate(VP9_COMMON *cm, MACROBLOCK *x) {
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+
+  const int c = vp9_get_pred_context(cm, xd, PRED_SWITCHABLE_INTERP);
+  const int m = vp9_switchable_interp_map[mbmi->interp_filter];
+  return SWITCHABLE_INTERP_RATE_FACTOR * x->switchable_interp_costs[c][m];
+}
+
+static void single_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+                                 BLOCK_SIZE_TYPE bsize,
+                                 int mi_row, int mi_col,
+                                 int_mv *tmp_mv, int *rate_mv) {
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0}};
+  int bestsme = INT_MAX;
+  int further_steps, step_param = cpi->sf.first_step;
+  int sadpb = x->sadperbit16;
+  int_mv mvp_full;
+  int ref = mbmi->ref_frame[0];
+  int_mv ref_mv = mbmi->ref_mvs[ref][0];
+  int sr = 0;
+  const enum BlockSize block_size = get_plane_block_size(bsize, &xd->plane[0]);
+
+  int tmp_col_min = x->mv_col_min;
+  int tmp_col_max = x->mv_col_max;
+  int tmp_row_min = x->mv_row_min;
+  int tmp_row_max = x->mv_row_max;
+
+  YV12_BUFFER_CONFIG *scaled_ref_frame = get_scaled_ref_frame(cpi, ref);
+
+  if (scaled_ref_frame) {
+    int i;
+    // Swap out the reference frame for a version that's been scaled to
+    // match the resolution of the current frame, allowing the existing
+    // motion search code to be used without additional modifications.
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      backup_yv12[i] = xd->plane[i].pre[0];
+
+    setup_pre_planes(xd, scaled_ref_frame, NULL, mi_row, mi_col,
+                     NULL, NULL);
+  }
+
+  vp9_clamp_mv_min_max(x, &ref_mv);
+
+  sr = vp9_init_search_range(cpi->common.width, cpi->common.height);
+
+  // mvp_full.as_int = ref_mv[0].as_int;
+  mvp_full.as_int =
+      mbmi->ref_mvs[ref][x->mv_best_ref_index[ref]].as_int;
+
+  mvp_full.as_mv.col >>= 3;
+  mvp_full.as_mv.row >>= 3;
+
+  // adjust search range according to sr from mv prediction
+  step_param = MAX(step_param, sr);
+
+  // Further step/diamond searches as necessary
+  further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
+
+  bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
+                                   sadpb, further_steps, 1,
+                                   &cpi->fn_ptr[block_size],
+                                   &ref_mv, tmp_mv);
+
+  x->mv_col_min = tmp_col_min;
+  x->mv_col_max = tmp_col_max;
+  x->mv_row_min = tmp_row_min;
+  x->mv_row_max = tmp_row_max;
+
+  if (bestsme < INT_MAX) {
+    int dis; /* TODO: use dis in distortion calculation later. */
+    unsigned int sse;
+    cpi->find_fractional_mv_step(x, tmp_mv, &ref_mv,
+                                 x->errorperbit,
+                                 &cpi->fn_ptr[block_size],
+                                 x->nmvjointcost, x->mvcost,
+                                 &dis, &sse);
+  }
+  *rate_mv = vp9_mv_bit_cost(tmp_mv, &ref_mv,
+                             x->nmvjointcost, x->mvcost,
+                             96, xd->allow_high_precision_mv);
+  if (scaled_ref_frame) {
+    int i;
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      xd->plane[i].pre[0] = backup_yv12[i];
+  }
+}
+
+static void joint_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+                                BLOCK_SIZE_TYPE bsize,
+                                int_mv *frame_mv,
+                                int mi_row, int mi_col,
+                                int_mv single_newmv[MAX_REF_FRAMES],
+                                int *rate_mv) {
+  int pw = 4 << b_width_log2(bsize), ph = 4 << b_height_log2(bsize);
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  int refs[2] = { mbmi->ref_frame[0],
+    (mbmi->ref_frame[1] < 0 ? 0 : mbmi->ref_frame[1]) };
+  int_mv ref_mv[2];
+  const enum BlockSize block_size = get_plane_block_size(bsize, &xd->plane[0]);
+  int ite;
+  // Prediction buffer from second frame.
+  uint8_t *second_pred = vpx_memalign(16, pw * ph * sizeof(uint8_t));
+
+  // Do joint motion search in compound mode to get more accurate mv.
+  struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0}};
+  struct buf_2d backup_second_yv12[MAX_MB_PLANE] = {{0}};
+  struct buf_2d scaled_first_yv12;
+  int last_besterr[2] = {INT_MAX, INT_MAX};
+  YV12_BUFFER_CONFIG *scaled_ref_frame[2] = {NULL, NULL};
+  scaled_ref_frame[0] = get_scaled_ref_frame(cpi, mbmi->ref_frame[0]);
+  scaled_ref_frame[1] = get_scaled_ref_frame(cpi, mbmi->ref_frame[1]);
+
+  ref_mv[0] = mbmi->ref_mvs[refs[0]][0];
+  ref_mv[1] = mbmi->ref_mvs[refs[1]][0];
+
+  if (scaled_ref_frame[0]) {
+    int i;
+    // Swap out the reference frame for a version that's been scaled to
+    // match the resolution of the current frame, allowing the existing
+    // motion search code to be used without additional modifications.
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      backup_yv12[i] = xd->plane[i].pre[0];
+    setup_pre_planes(xd, scaled_ref_frame[0], NULL, mi_row, mi_col,
+                     NULL, NULL);
+  }
+
+  if (scaled_ref_frame[1]) {
+    int i;
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      backup_second_yv12[i] = xd->plane[i].pre[1];
+
+    setup_pre_planes(xd, scaled_ref_frame[1], NULL, mi_row, mi_col,
+                     NULL, NULL);
+  }
+
+  xd->scale_factor[0].set_scaled_offsets(&xd->scale_factor[0],
+                                         mi_row, mi_col);
+  xd->scale_factor[1].set_scaled_offsets(&xd->scale_factor[1],
+                                         mi_row, mi_col);
+  scaled_first_yv12 = xd->plane[0].pre[0];
+
+  // Initialize mv using single prediction mode result.
+  frame_mv[refs[0]].as_int = single_newmv[refs[0]].as_int;
+  frame_mv[refs[1]].as_int = single_newmv[refs[1]].as_int;
+
+  // Allow joint search multiple times iteratively for each ref frame
+  // and break out the search loop if it couldn't find better mv.
+  for (ite = 0; ite < 4; ite++) {
+    struct buf_2d ref_yv12[2];
+    int bestsme = INT_MAX;
+    int sadpb = x->sadperbit16;
+    int_mv tmp_mv;
+    int search_range = 3;
+
+    int tmp_col_min = x->mv_col_min;
+    int tmp_col_max = x->mv_col_max;
+    int tmp_row_min = x->mv_row_min;
+    int tmp_row_max = x->mv_row_max;
+    int id = ite % 2;
+
+    // Initialized here because of compiler problem in Visual Studio.
+    ref_yv12[0] = xd->plane[0].pre[0];
+    ref_yv12[1] = xd->plane[0].pre[1];
+
+    // Get pred block from second frame.
+    vp9_build_inter_predictor(ref_yv12[!id].buf,
+                              ref_yv12[!id].stride,
+                              second_pred, pw,
+                              &frame_mv[refs[!id]],
+                              &xd->scale_factor[!id],
+                              pw, ph, 0,
+                              &xd->subpix);
+
+    // Compound motion search on first ref frame.
+    if (id)
+      xd->plane[0].pre[0] = ref_yv12[id];
+    vp9_clamp_mv_min_max(x, &ref_mv[id]);
+
+    // Use mv result from single mode as mvp.
+    tmp_mv.as_int = frame_mv[refs[id]].as_int;
+
+    tmp_mv.as_mv.col >>= 3;
+    tmp_mv.as_mv.row >>= 3;
+
+    // Small-range full-pixel motion search
+    bestsme = vp9_refining_search_8p_c(x, &tmp_mv, sadpb,
+                                       search_range,
+                                       &cpi->fn_ptr[block_size],
+                                       x->nmvjointcost, x->mvcost,
+                                       &ref_mv[id], second_pred,
+                                       pw, ph);
+
+    x->mv_col_min = tmp_col_min;
+    x->mv_col_max = tmp_col_max;
+    x->mv_row_min = tmp_row_min;
+    x->mv_row_max = tmp_row_max;
+
+    if (bestsme < INT_MAX) {
+      int dis; /* TODO: use dis in distortion calculation later. */
+      unsigned int sse;
+
+      bestsme = vp9_find_best_sub_pixel_comp(x, &tmp_mv,
+                                             &ref_mv[id],
+                                             x->errorperbit,
+                                             &cpi->fn_ptr[block_size],
+                                             x->nmvjointcost, x->mvcost,
+                                             &dis, &sse, second_pred,
+                                             pw, ph);
+    }
+
+    if (id)
+      xd->plane[0].pre[0] = scaled_first_yv12;
+
+    if (bestsme < last_besterr[id]) {
+      frame_mv[refs[id]].as_int = tmp_mv.as_int;
+      last_besterr[id] = bestsme;
+    } else {
+      break;
+    }
+  }
+
+  // restore the predictor
+  if (scaled_ref_frame[0]) {
+    int i;
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      xd->plane[i].pre[0] = backup_yv12[i];
+  }
+
+  if (scaled_ref_frame[1]) {
+    int i;
+    for (i = 0; i < MAX_MB_PLANE; i++)
+      xd->plane[i].pre[1] = backup_second_yv12[i];
+  }
+  *rate_mv  = vp9_mv_bit_cost(&frame_mv[refs[0]],
+                              &mbmi->ref_mvs[refs[0]][0],
+                              x->nmvjointcost, x->mvcost, 96,
+                              x->e_mbd.allow_high_precision_mv);
+  *rate_mv += vp9_mv_bit_cost(&frame_mv[refs[1]],
+                              &mbmi->ref_mvs[refs[1]][0],
+                              x->nmvjointcost, x->mvcost, 96,
+                              x->e_mbd.allow_high_precision_mv);
+
+  vpx_free(second_pred);
+}
+
+static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                                 BLOCK_SIZE_TYPE bsize,
+                                 int64_t txfm_cache[],
+                                 int *rate2, int *distortion, int *skippable,
+                                 int *rate_y, int *distortion_y,
+                                 int *rate_uv, int *distortion_uv,
+                                 int *mode_excluded, int *disable_skip,
+                                 INTERPOLATIONFILTERTYPE *best_filter,
+                                 int_mv *frame_mv,
+                                 int mi_row, int mi_col,
+                                 int_mv single_newmv[MAX_REF_FRAMES]) {
+  const int bw = 1 << mi_width_log2(bsize), bh = 1 << mi_height_log2(bsize);
+
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &x->e_mbd;
+  const enum BlockSize block_size = get_plane_block_size(bsize, &xd->plane[0]);
+  const enum BlockSize uv_block_size = get_plane_block_size(bsize,
+                                                            &xd->plane[1]);
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  const int is_comp_pred = (mbmi->ref_frame[1] > 0);
+  const int num_refs = is_comp_pred ? 2 : 1;
+  const int this_mode = mbmi->mode;
+  int i;
+  int refs[2] = { mbmi->ref_frame[0],
+    (mbmi->ref_frame[1] < 0 ? 0 : mbmi->ref_frame[1]) };
+  int_mv cur_mv[2];
+  int64_t this_rd = 0;
+  unsigned char tmp_buf[MAX_MB_PLANE][64 * 64];
+  int pred_exists = 0;
+  int interpolating_intpel_seen = 0;
+  int intpel_mv;
+  int64_t rd, best_rd = INT64_MAX;
+
+  switch (this_mode) {
+    int rate_mv;
+    case NEWMV:
+      if (is_comp_pred) {
+        // Initialize mv using single prediction mode result.
+        frame_mv[refs[0]].as_int = single_newmv[refs[0]].as_int;
+        frame_mv[refs[1]].as_int = single_newmv[refs[1]].as_int;
+
+        if (cpi->sf.comp_inter_joint_search_thresh < bsize) {
+          joint_motion_search(cpi, x, bsize, frame_mv,
+                              mi_row, mi_col, single_newmv, &rate_mv);
+        } else {
+          rate_mv  = vp9_mv_bit_cost(&frame_mv[refs[0]],
+                                     &mbmi->ref_mvs[refs[0]][0],
+                                     x->nmvjointcost, x->mvcost, 96,
+                                     x->e_mbd.allow_high_precision_mv);
+          rate_mv += vp9_mv_bit_cost(&frame_mv[refs[1]],
+                                     &mbmi->ref_mvs[refs[1]][0],
+                                     x->nmvjointcost, x->mvcost, 96,
+                                     x->e_mbd.allow_high_precision_mv);
+        }
+        if (frame_mv[refs[0]].as_int == INVALID_MV ||
+            frame_mv[refs[1]].as_int == INVALID_MV)
+          return INT64_MAX;
+        *rate2 += rate_mv;
+
+      } else {
+        int_mv tmp_mv;
+        single_motion_search(cpi, x, bsize, mi_row, mi_col,
+                             &tmp_mv, &rate_mv);
+        *rate2 += rate_mv;
+        frame_mv[refs[0]].as_int =
+            xd->mode_info_context->bmi[0].as_mv[0].as_int = tmp_mv.as_int;
+        single_newmv[refs[0]].as_int = tmp_mv.as_int;
+      }
+      break;
+    case NEARMV:
+    case NEARESTMV:
+    case ZEROMV:
+    default:
+      break;
+  }
+  for (i = 0; i < num_refs; ++i) {
+    cur_mv[i] = frame_mv[refs[i]];
+    // Clip "next_nearest" so that it does not extend to far out of image
+    if (this_mode == NEWMV)
+      assert(!clamp_mv2(&cur_mv[i], xd));
+    else
+      clamp_mv2(&cur_mv[i], xd);
+
+    if (mv_check_bounds(x, &cur_mv[i]))
+      return INT64_MAX;
+    mbmi->mv[i].as_int = cur_mv[i].as_int;
+  }
+
+  /* We don't include the cost of the second reference here, because there
+   * are only three options: Last/Golden, ARF/Last or Golden/ARF, or in other
+   * words if you present them in that order, the second one is always known
+   * if the first is known */
+  *rate2 += vp9_cost_mv_ref(cpi, this_mode,
+                            mbmi->mb_mode_context[mbmi->ref_frame[0]]);
+
+  pred_exists = 0;
+  interpolating_intpel_seen = 0;
+  // Are all MVs integer pel for Y and UV
+  intpel_mv = (mbmi->mv[0].as_mv.row & 15) == 0 &&
+      (mbmi->mv[0].as_mv.col & 15) == 0;
+  if (is_comp_pred)
+    intpel_mv &= (mbmi->mv[1].as_mv.row & 15) == 0 &&
+        (mbmi->mv[1].as_mv.col & 15) == 0;
+  // Search for best switchable filter by checking the variance of
+  // pred error irrespective of whether the filter will be used
+  if (cpi->speed > 4) {
+    *best_filter = EIGHTTAP;
+  } else {
+    int i, newbest;
+    int tmp_rate_sum = 0, tmp_dist_sum = 0;
+    for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
+      int rs = 0;
+      const INTERPOLATIONFILTERTYPE filter = vp9_switchable_interp[i];
+      const int is_intpel_interp = intpel_mv &&
+          vp9_is_interpolating_filter[filter];
+      mbmi->interp_filter = filter;
+      vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
+
+      if (cm->mcomp_filter_type == SWITCHABLE)
+        rs = get_switchable_rate(cm, x);
+
+      if (interpolating_intpel_seen && is_intpel_interp) {
+        rd = RDCOST(x->rdmult, x->rddiv, rs + tmp_rate_sum, tmp_dist_sum);
+      } else {
+        int rate_sum = 0, dist_sum = 0;
+        vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+        model_rd_for_sb(cpi, bsize, x, xd, &rate_sum, &dist_sum);
+        rd = RDCOST(x->rdmult, x->rddiv, rs + rate_sum, dist_sum);
+        if (!interpolating_intpel_seen && is_intpel_interp) {
+          tmp_rate_sum = rate_sum;
+          tmp_dist_sum = dist_sum;
+        }
+      }
+      newbest = i == 0 || rd < best_rd;
+
+      if (newbest) {
+        best_rd = rd;
+        *best_filter = mbmi->interp_filter;
+      }
+
+      if ((cm->mcomp_filter_type == SWITCHABLE && newbest) ||
+          (cm->mcomp_filter_type != SWITCHABLE &&
+           cm->mcomp_filter_type == mbmi->interp_filter)) {
+        int p;
+
+        for (p = 0; p < MAX_MB_PLANE; p++) {
+          const int y = (MI_SIZE * bh) >> xd->plane[p].subsampling_y;
+          const int x = (MI_SIZE * bw) >> xd->plane[p].subsampling_x;
+          int i;
+
+          for (i = 0; i < y; i++)
+            vpx_memcpy(&tmp_buf[p][64 * i],
+                       xd->plane[p].dst.buf + i * xd->plane[p].dst.stride, x);
+        }
+        pred_exists = 1;
+      }
+      interpolating_intpel_seen |= is_intpel_interp;
+    }
+  }
+
+  // Set the appripriate filter
+  mbmi->interp_filter = cm->mcomp_filter_type != SWITCHABLE ?
+      cm->mcomp_filter_type : *best_filter;
+  vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
+
+
+  if (pred_exists) {
+    int p;
+
+    for (p = 0; p < MAX_MB_PLANE; p++) {
+      const int y = (MI_SIZE * bh) >> xd->plane[p].subsampling_y;
+      const int x = (MI_SIZE * bw) >> xd->plane[p].subsampling_x;
+      int i;
+
+      for (i = 0; i < y; i++)
+        vpx_memcpy(xd->plane[p].dst.buf + i * xd->plane[p].dst.stride,
+                   &tmp_buf[p][64 * i], x);
+    }
+  } else {
+    // Handles the special case when a filter that is not in the
+    // switchable list (ex. bilinear, 6-tap) is indicated at the frame level
+    vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+  }
+
+  if (cpi->common.mcomp_filter_type == SWITCHABLE)
+    *rate2 += get_switchable_rate(cm, x);
+
+  if (cpi->active_map_enabled && x->active_ptr[0] == 0)
+    x->skip = 1;
+  else if (x->encode_breakout) {
+    unsigned int var, sse;
+    int threshold = (xd->plane[0].dequant[1]
+                     * xd->plane[0].dequant[1] >> 4);
+
+    if (threshold < x->encode_breakout)
+      threshold = x->encode_breakout;
+
+    var = cpi->fn_ptr[block_size].vf(x->plane[0].src.buf,
+                                     x->plane[0].src.stride,
+                                     xd->plane[0].dst.buf,
+                                     xd->plane[0].dst.stride,
+                                     &sse);
+
+    if ((int)sse < threshold) {
+      unsigned int q2dc = xd->plane[0].dequant[0];
+      /* If there is no codeable 2nd order dc
+         or a very small uniform pixel change change */
+      if ((sse - var < q2dc * q2dc >> 4) ||
+          (sse / 2 > var && sse - var < 64)) {
+        // Check u and v to make sure skip is ok
+        int sse2;
+        unsigned int sse2u, sse2v;
+        var = cpi->fn_ptr[uv_block_size].vf(x->plane[1].src.buf,
+                                            x->plane[1].src.stride,
+                                            xd->plane[1].dst.buf,
+                                            xd->plane[1].dst.stride, &sse2u);
+        var = cpi->fn_ptr[uv_block_size].vf(x->plane[2].src.buf,
+                                            x->plane[1].src.stride,
+                                            xd->plane[2].dst.buf,
+                                            xd->plane[1].dst.stride, &sse2v);
+        sse2 = sse2u + sse2v;
+
+        if (sse2 * 2 < threshold) {
+          x->skip = 1;
+          *distortion = sse + sse2;
+          *rate2 = 500;
+
+          /* for best_yrd calculation */
+          *rate_uv = 0;
+          *distortion_uv = sse2;
+
+          *disable_skip = 1;
+          this_rd = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
+        }
+      }
+    }
+  }
+
+  if (!x->skip) {
+    int skippable_y, skippable_uv;
+
+    // Y cost and distortion
+    super_block_yrd(cpi, x, rate_y, distortion_y, &skippable_y,
+                    bsize, txfm_cache);
+
+    *rate2 += *rate_y;
+    *distortion += *distortion_y;
+
+    super_block_uvrd(cm, x, rate_uv, distortion_uv,
+                     &skippable_uv, bsize);
+
+    *rate2 += *rate_uv;
+    *distortion += *distortion_uv;
+    *skippable = skippable_y && skippable_uv;
+  }
+
+  if (!(*mode_excluded)) {
+    if (is_comp_pred) {
+      *mode_excluded = (cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY);
+    } else {
+      *mode_excluded = (cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY);
+    }
+  }
+
+  return this_rd;  // if 0, this will be re-calculated by caller
+}
+
+void vp9_rd_pick_intra_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+                               int *returnrate, int *returndist,
+                               BLOCK_SIZE_TYPE bsize,
+                               PICK_MODE_CONTEXT *ctx) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &x->e_mbd;
+  int rate_y = 0, rate_uv;
+  int rate_y_tokenonly = 0, rate_uv_tokenonly;
+  int dist_y = 0, dist_uv;
+  int y_skip = 0, uv_skip;
+  int64_t txfm_cache[NB_TXFM_MODES], err;
+  MB_PREDICTION_MODE mode;
+  TX_SIZE txfm_size;
+  int rate4x4_y, rate4x4_y_tokenonly, dist4x4_y;
+  int64_t err4x4 = INT64_MAX;
+  int i;
+
+  vpx_memset(&txfm_cache,0,sizeof(txfm_cache));
+  ctx->skip = 0;
+  xd->mode_info_context->mbmi.mode = DC_PRED;
+  xd->mode_info_context->mbmi.ref_frame[0] = INTRA_FRAME;
+  err = rd_pick_intra_sby_mode(cpi, x, &rate_y, &rate_y_tokenonly,
+                               &dist_y, &y_skip, bsize, txfm_cache);
+  mode = xd->mode_info_context->mbmi.mode;
+  txfm_size = xd->mode_info_context->mbmi.txfm_size;
+  rd_pick_intra_sbuv_mode(cpi, x, &rate_uv, &rate_uv_tokenonly,
+                          &dist_uv, &uv_skip,
+                          (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 :
+                                                       bsize);
+  if (bsize < BLOCK_SIZE_SB8X8)
+    err4x4 = rd_pick_intra4x4mby_modes(cpi, x, &rate4x4_y,
+                                       &rate4x4_y_tokenonly,
+                                       &dist4x4_y, err);
+
+  if (y_skip && uv_skip) {
+    *returnrate = rate_y + rate_uv - rate_y_tokenonly - rate_uv_tokenonly +
+                  vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 1);
+    *returndist = dist_y + (dist_uv >> 2);
+    memset(ctx->txfm_rd_diff, 0, sizeof(ctx->txfm_rd_diff));
+    xd->mode_info_context->mbmi.mode = mode;
+    xd->mode_info_context->mbmi.txfm_size = txfm_size;
+  } else if (bsize < BLOCK_SIZE_SB8X8 && err4x4 < err) {
+    *returnrate = rate4x4_y + rate_uv +
+        vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 0);
+    *returndist = dist4x4_y + (dist_uv >> 2);
+    vpx_memset(ctx->txfm_rd_diff, 0, sizeof(ctx->txfm_rd_diff));
+    xd->mode_info_context->mbmi.txfm_size = TX_4X4;
+  } else {
+    *returnrate = rate_y + rate_uv +
+        vp9_cost_bit(vp9_get_pred_prob(cm, xd, PRED_MBSKIP), 0);
+    *returndist = dist_y + (dist_uv >> 2);
+    for (i = 0; i < NB_TXFM_MODES; i++) {
+      ctx->txfm_rd_diff[i] = txfm_cache[i] - txfm_cache[cm->txfm_mode];
+    }
+    xd->mode_info_context->mbmi.txfm_size = txfm_size;
+    xd->mode_info_context->mbmi.mode = mode;
+  }
+
+  ctx->mic = *xd->mode_info_context;
+}
+
+int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+                                  int mi_row, int mi_col,
+                                  int *returnrate,
+                                  int *returndistortion,
+                                  BLOCK_SIZE_TYPE bsize,
+                                  PICK_MODE_CONTEXT *ctx) {
+  VP9_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &x->e_mbd;
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  const enum BlockSize block_size = get_plane_block_size(bsize, &xd->plane[0]);
+  MB_PREDICTION_MODE this_mode;
+  MB_PREDICTION_MODE best_mode = DC_PRED;
+  MV_REFERENCE_FRAME ref_frame;
+  unsigned char segment_id = xd->mode_info_context->mbmi.segment_id;
+  int comp_pred, i;
+  int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
+  struct buf_2d yv12_mb[4][MAX_MB_PLANE];
+  int_mv single_newmv[MAX_REF_FRAMES];
+  static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
+                                    VP9_ALT_FLAG };
+  int idx_list[4] = {0,
+                     cpi->lst_fb_idx,
+                     cpi->gld_fb_idx,
+                     cpi->alt_fb_idx};
+  int64_t best_rd = INT64_MAX;
+  int64_t best_txfm_rd[NB_TXFM_MODES];
+  int64_t best_txfm_diff[NB_TXFM_MODES];
+  int64_t best_pred_diff[NB_PREDICTION_TYPES];
+  int64_t best_pred_rd[NB_PREDICTION_TYPES];
+  MB_MODE_INFO best_mbmode;
+  int j;
+  int mode_index, best_mode_index = 0;
+  unsigned int ref_costs_single[MAX_REF_FRAMES], ref_costs_comp[MAX_REF_FRAMES];
+  vp9_prob comp_mode_p;
+  int64_t best_overall_rd = INT64_MAX;
+  INTERPOLATIONFILTERTYPE best_filter = SWITCHABLE;
+  INTERPOLATIONFILTERTYPE tmp_best_filter = SWITCHABLE;
+  int rate_uv_intra[TX_SIZE_MAX_SB], rate_uv_tokenonly[TX_SIZE_MAX_SB];
+  int dist_uv[TX_SIZE_MAX_SB], skip_uv[TX_SIZE_MAX_SB];
+  MB_PREDICTION_MODE mode_uv[TX_SIZE_MAX_SB];
+  struct scale_factors scale_factor[4];
+  unsigned int ref_frame_mask = 0;
+  unsigned int mode_mask = 0;
+  int64_t mode_distortions[MB_MODE_COUNT] = {-1};
+  int64_t frame_distortions[MAX_REF_FRAMES] = {-1};
+  int intra_cost_penalty = 20 * vp9_dc_quant(cpi->common.base_qindex,
+                                             cpi->common.y_dc_delta_q);
+  int_mv seg_mvs[4][MAX_REF_FRAMES];
+  union b_mode_info best_bmodes[4];
+  PARTITION_INFO best_partition;
+  int bwsl = b_width_log2(bsize);
+  int bws = (1 << bwsl) / 4;  // mode_info step for subsize
+  int bhsl = b_height_log2(bsize);
+  int bhs = (1 << bhsl) / 4;  // mode_info step for subsize
+
+  for (i = 0; i < 4; i++) {
+    int j;
+
+    for (j = 0; j < MAX_REF_FRAMES; j++)
+      seg_mvs[i][j].as_int = INVALID_MV;
+  }
+  // Everywhere the flag is set the error is much higher than its neighbors.
+  ctx->frames_with_high_error = 0;
+  ctx->modes_with_high_error = 0;
+
+  xd->mode_info_context->mbmi.segment_id = segment_id;
+  estimate_ref_frame_costs(cpi, segment_id, ref_costs_single, ref_costs_comp,
+                           &comp_mode_p);
+  vpx_memset(&best_mbmode, 0, sizeof(best_mbmode));
+  vpx_memset(&single_newmv, 0, sizeof(single_newmv));
+
+  for (i = 0; i < NB_PREDICTION_TYPES; ++i)
+    best_pred_rd[i] = INT64_MAX;
+  for (i = 0; i < NB_TXFM_MODES; i++)
+    best_txfm_rd[i] = INT64_MAX;
+
+  // Create a mask set to 1 for each frame used by a smaller resolution.
+  if (cpi->speed > 0) {
+    switch (block_size) {
+      case BLOCK_64X64:
+        for (i = 0; i < 4; i++) {
+          for (j = 0; j < 4; j++) {
+            ref_frame_mask |= x->mb_context[i][j].frames_with_high_error;
+            mode_mask |= x->mb_context[i][j].modes_with_high_error;
+          }
+        }
+        for (i = 0; i < 4; i++) {
+          ref_frame_mask |= x->sb32_context[i].frames_with_high_error;
+          mode_mask |= x->sb32_context[i].modes_with_high_error;
+        }
+        break;
+      case BLOCK_32X32:
+        for (i = 0; i < 4; i++) {
+          ref_frame_mask |=
+              x->mb_context[xd->sb_index][i].frames_with_high_error;
+          mode_mask |= x->mb_context[xd->sb_index][i].modes_with_high_error;
+        }
+        break;
+      default:
+        // Until we handle all block sizes set it to present;
+        ref_frame_mask = 0;
+        mode_mask = 0;
+        break;
+    }
+    ref_frame_mask = ~ref_frame_mask;
+    mode_mask = ~mode_mask;
+  }
+
+  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
+    if (cpi->ref_frame_flags & flag_list[ref_frame]) {
+      setup_buffer_inter(cpi, x, idx_list[ref_frame], ref_frame, block_size,
+                         mi_row, mi_col, frame_mv[NEARESTMV], frame_mv[NEARMV],
+                         yv12_mb, scale_factor);
+    }
+    frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
+    frame_mv[ZEROMV][ref_frame].as_int = 0;
+  }
+  if (cpi->speed == 0
+      || (cpi->speed > 0 && (ref_frame_mask & (1 << INTRA_FRAME)))) {
+    mbmi->mode = DC_PRED;
+    mbmi->ref_frame[0] = INTRA_FRAME;
+    for (i = 0; i <= (bsize < BLOCK_SIZE_MB16X16 ? TX_4X4 :
+                      (bsize < BLOCK_SIZE_SB32X32 ? TX_8X8 :
+                       (bsize < BLOCK_SIZE_SB64X64 ? TX_16X16 : TX_32X32)));
+         i++) {
+      mbmi->txfm_size = i;
+      rd_pick_intra_sbuv_mode(cpi, x, &rate_uv_intra[i], &rate_uv_tokenonly[i],
+                              &dist_uv[i], &skip_uv[i],
+                              (bsize < BLOCK_SIZE_SB8X8) ? BLOCK_SIZE_SB8X8 :
+                                                           bsize);
+      mode_uv[i] = mbmi->uv_mode;
+    }
+  }
+
+  for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
+    int mode_excluded = 0;
+    int64_t this_rd = INT64_MAX;
+    int disable_skip = 0;
+    int compmode_cost = 0;
+    int rate2 = 0, rate_y = 0, rate_uv = 0;
+    int distortion2 = 0, distortion_y = 0, distortion_uv = 0;
+    int skippable;
+    int64_t txfm_cache[NB_TXFM_MODES];
+    int i;
+
+    for (i = 0; i < NB_TXFM_MODES; ++i)
+      txfm_cache[i] = INT64_MAX;
+
+    // Test best rd so far against threshold for trying this mode.
+    if ((best_rd < ((cpi->rd_threshes[bsize][mode_index] *
+                     cpi->rd_thresh_freq_fact[bsize][mode_index]) >> 4)) ||
+        cpi->rd_threshes[bsize][mode_index] == INT_MAX)
+      continue;
+
+    // Do not allow compound prediction if the segment level reference
+    // frame feature is in use as in this case there can only be one reference.
+    if ((vp9_mode_order[mode_index].second_ref_frame > INTRA_FRAME) &&
+         vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME))
+      continue;
+
+    x->skip = 0;
+    this_mode = vp9_mode_order[mode_index].mode;
+    ref_frame = vp9_mode_order[mode_index].ref_frame;
+
+    if (cpi->speed > 0 && bsize >= BLOCK_SIZE_SB8X8) {
+      if (!(ref_frame_mask & (1 << ref_frame))) {
+        continue;
+      }
+      if (!(mode_mask & (1 << this_mode))) {
+        continue;
+      }
+      if (vp9_mode_order[mode_index].second_ref_frame != NONE
+          && !(ref_frame_mask
+              & (1 << vp9_mode_order[mode_index].second_ref_frame))) {
+        continue;
+      }
+    }
+
+    mbmi->ref_frame[0] = ref_frame;
+    mbmi->ref_frame[1] = vp9_mode_order[mode_index].second_ref_frame;
+
+    if (!(ref_frame == INTRA_FRAME
+        || (cpi->ref_frame_flags & flag_list[ref_frame]))) {
+      continue;
+    }
+    if (!(mbmi->ref_frame[1] == NONE
+        || (cpi->ref_frame_flags & flag_list[mbmi->ref_frame[1]]))) {
+      continue;
+    }
+
+    // TODO(jingning, jkoleszar): scaling reference frame not supported for
+    // SPLITMV.
+    if (mbmi->ref_frame[0] > 0 &&
+          (scale_factor[mbmi->ref_frame[0]].x_scale_fp !=
+           (1 << VP9_REF_SCALE_SHIFT) ||
+           scale_factor[mbmi->ref_frame[0]].y_scale_fp !=
+           (1 << VP9_REF_SCALE_SHIFT)) &&
+        this_mode == SPLITMV)
+      continue;
+
+    if (mbmi->ref_frame[1] > 0 &&
+          (scale_factor[mbmi->ref_frame[1]].x_scale_fp !=
+           (1 << VP9_REF_SCALE_SHIFT) ||
+           scale_factor[mbmi->ref_frame[1]].y_scale_fp !=
+           (1 << VP9_REF_SCALE_SHIFT)) &&
+        this_mode == SPLITMV)
+      continue;
+
+    set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
+                      scale_factor);
+    comp_pred = mbmi->ref_frame[1] > INTRA_FRAME;
+    mbmi->mode = this_mode;
+    mbmi->uv_mode = DC_PRED;
+
+    // Evaluate all sub-pel filters irrespective of whether we can use
+    // them for this frame.
+    mbmi->interp_filter = cm->mcomp_filter_type;
+    vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+
+    if (bsize >= BLOCK_SIZE_SB8X8 &&
+        (this_mode == I4X4_PRED || this_mode == SPLITMV))
+      continue;
+    if (bsize < BLOCK_SIZE_SB8X8 &&
+        !(this_mode == I4X4_PRED || this_mode == SPLITMV))
+      continue;
+
+    if (comp_pred) {
+      if (!(cpi->ref_frame_flags & flag_list[mbmi->ref_frame[1]]))
+        continue;
+      set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
+                        scale_factor);
+
+      mode_excluded =
+          mode_excluded ?
+              mode_excluded : cm->comp_pred_mode == SINGLE_PREDICTION_ONLY;
+    } else {
+      // mbmi->ref_frame[1] = vp9_mode_order[mode_index].ref_frame[1];
+      if (ref_frame != INTRA_FRAME) {
+        if (mbmi->ref_frame[1] != INTRA_FRAME)
+          mode_excluded =
+              mode_excluded ?
+                  mode_excluded : cm->comp_pred_mode == COMP_PREDICTION_ONLY;
+      }
+    }
+
+    // Select predictors
+    for (i = 0; i < MAX_MB_PLANE; i++) {
+      xd->plane[i].pre[0] = yv12_mb[ref_frame][i];
+      if (comp_pred)
+        xd->plane[i].pre[1] = yv12_mb[mbmi->ref_frame[1]][i];
+    }
+
+    // If the segment reference frame feature is enabled....
+    // then do nothing if the current ref frame is not allowed..
+    if (vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
+        vp9_get_segdata(xd, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame) {
+      continue;
+    // If the segment skip feature is enabled....
+    // then do nothing if the current mode is not allowed..
+    } else if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) &&
+               (this_mode != ZEROMV && ref_frame != INTRA_FRAME)) {
+      continue;
+    // Disable this drop out case if the ref frame
+    // segment level feature is enabled for this segment. This is to
+    // prevent the possibility that we end up unable to pick any mode.
+    } else if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME)) {
+      // Only consider ZEROMV/ALTREF_FRAME for alt ref frame,
+      // unless ARNR filtering is enabled in which case we want
+      // an unfiltered alternative
+      if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
+        if (this_mode != ZEROMV || ref_frame != ALTREF_FRAME) {
+          continue;
+        }
+      }
+    }
+    // TODO(JBB): This is to make up for the fact that we don't have sad
+    // functions that work when the block size reads outside the umv.  We
+    // should fix this either by making the motion search just work on
+    // a representative block in the boundary ( first ) and then implement a
+    // function that does sads when inside the border..
+    if (((mi_row + bhs) > cm->mi_rows || (mi_col + bws) > cm->mi_cols) &&
+        this_mode == NEWMV) {
+      continue;
+    }
+
+    if (this_mode == I4X4_PRED) {
+      int rate;
+
+      mbmi->txfm_size = TX_4X4;
+      rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y,
+                                &distortion_y, INT64_MAX);
+      rate2 += rate;
+      rate2 += intra_cost_penalty;
+      distortion2 += distortion_y;
+
+      rate2 += rate_uv_intra[TX_4X4];
+      rate_uv = rate_uv_intra[TX_4X4];
+      distortion2 += dist_uv[TX_4X4];
+      distortion_uv = dist_uv[TX_4X4];
+      mbmi->uv_mode = mode_uv[TX_4X4];
+      txfm_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+      for (i = 0; i < NB_TXFM_MODES; ++i)
+        txfm_cache[i] = txfm_cache[ONLY_4X4];
+    } else if (ref_frame == INTRA_FRAME) {
+      TX_SIZE uv_tx;
+      super_block_yrd(cpi, x, &rate_y, &distortion_y, &skippable,
+                      bsize, txfm_cache);
+
+      uv_tx = mbmi->txfm_size;
+      if (bsize < BLOCK_SIZE_MB16X16 && uv_tx == TX_8X8)
+        uv_tx = TX_4X4;
+      if (bsize < BLOCK_SIZE_SB32X32 && uv_tx == TX_16X16)
+        uv_tx = TX_8X8;
+      else if (bsize < BLOCK_SIZE_SB64X64 && uv_tx == TX_32X32)
+        uv_tx = TX_16X16;
+
+      rate_uv = rate_uv_intra[uv_tx];
+      distortion_uv = dist_uv[uv_tx];
+      skippable = skippable && skip_uv[uv_tx];
+      mbmi->uv_mode = mode_uv[uv_tx];
+
+      rate2 = rate_y + x->mbmode_cost[mbmi->mode] + rate_uv;
+      if (mbmi->mode != DC_PRED && mbmi->mode != TM_PRED)
+        rate2 += intra_cost_penalty;
+      distortion2 = distortion_y + distortion_uv;
+    } else if (this_mode == SPLITMV) {
+      const int is_comp_pred = mbmi->ref_frame[1] > 0;
+      int rate, distortion;
+      int64_t this_rd_thresh;
+      int64_t tmp_rd, tmp_best_rd = INT64_MAX, tmp_best_rdu = INT64_MAX;
+      int tmp_best_rate = INT_MAX, tmp_best_ratey = INT_MAX;
+      int tmp_best_distortion = INT_MAX, tmp_best_skippable = 0;
+      int switchable_filter_index;
+      int_mv *second_ref = is_comp_pred ?
+          &mbmi->ref_mvs[mbmi->ref_frame[1]][0] : NULL;
+      union b_mode_info tmp_best_bmodes[16];
+      MB_MODE_INFO tmp_best_mbmode;
+      PARTITION_INFO tmp_best_partition;
+      int pred_exists = 0;
+      int uv_skippable;
+
+      this_rd_thresh = (mbmi->ref_frame[0] == LAST_FRAME) ?
+          cpi->rd_threshes[bsize][THR_NEWMV] :
+          cpi->rd_threshes[bsize][THR_NEWA];
+      this_rd_thresh = (mbmi->ref_frame[0] == GOLDEN_FRAME) ?
+          cpi->rd_threshes[bsize][THR_NEWG] : this_rd_thresh;
+      xd->mode_info_context->mbmi.txfm_size = TX_4X4;
+
+      for (switchable_filter_index = 0;
+           switchable_filter_index < VP9_SWITCHABLE_FILTERS;
+           ++switchable_filter_index) {
+        int newbest;
+        mbmi->interp_filter =
+        vp9_switchable_interp[switchable_filter_index];
+        vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+
+        tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+                     &mbmi->ref_mvs[mbmi->ref_frame[0]][0],
+                     second_ref, INT64_MAX,
+                     &rate, &rate_y, &distortion,
+                     &skippable,
+                     (int)this_rd_thresh, seg_mvs,
+                     mi_row, mi_col);
+        if (cpi->common.mcomp_filter_type == SWITCHABLE) {
+          const int rs = get_switchable_rate(cm, x);
+          tmp_rd += RDCOST(x->rdmult, x->rddiv, rs, 0);
+        }
+        newbest = (tmp_rd < tmp_best_rd);
+        if (newbest) {
+          tmp_best_filter = mbmi->interp_filter;
+          tmp_best_rd = tmp_rd;
+        }
+        if ((newbest && cm->mcomp_filter_type == SWITCHABLE) ||
+            (mbmi->interp_filter == cm->mcomp_filter_type &&
+             cm->mcomp_filter_type != SWITCHABLE)) {
+              tmp_best_rdu = tmp_rd;
+              tmp_best_rate = rate;
+              tmp_best_ratey = rate_y;
+              tmp_best_distortion = distortion;
+              tmp_best_skippable = skippable;
+              tmp_best_mbmode = *mbmi;
+              tmp_best_partition = *x->partition_info;
+              for (i = 0; i < 4; i++)
+                tmp_best_bmodes[i] = xd->mode_info_context->bmi[i];
+              pred_exists = 1;
+            }
+      }  // switchable_filter_index loop
+
+      mbmi->interp_filter = (cm->mcomp_filter_type == SWITCHABLE ?
+                             tmp_best_filter : cm->mcomp_filter_type);
+      vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+      if (!pred_exists) {
+        // Handles the special case when a filter that is not in the
+        // switchable list (bilinear, 6-tap) is indicated at the frame level
+        tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+                     &mbmi->ref_mvs[mbmi->ref_frame[0]][0],
+                     second_ref, INT64_MAX,
+                     &rate, &rate_y, &distortion,
+                     &skippable,
+                     (int)this_rd_thresh, seg_mvs,
+                     mi_row, mi_col);
+      } else {
+        if (cpi->common.mcomp_filter_type == SWITCHABLE) {
+          int rs = get_switchable_rate(cm, x);
+          tmp_best_rdu -= RDCOST(x->rdmult, x->rddiv, rs, 0);
+        }
+        tmp_rd = tmp_best_rdu;
+        rate = tmp_best_rate;
+        rate_y = tmp_best_ratey;
+        distortion = tmp_best_distortion;
+        skippable = tmp_best_skippable;
+        *mbmi = tmp_best_mbmode;
+        *x->partition_info = tmp_best_partition;
+        for (i = 0; i < 4; i++)
+          xd->mode_info_context->bmi[i] = tmp_best_bmodes[i];
+      }
+
+      rate2 += rate;
+      distortion2 += distortion;
+
+      if (cpi->common.mcomp_filter_type == SWITCHABLE)
+        rate2 += get_switchable_rate(cm, x);
+
+      // If even the 'Y' rd value of split is higher than best so far
+      // then dont bother looking at UV
+      vp9_build_inter_predictors_sbuv(&x->e_mbd, mi_row, mi_col,
+                                      BLOCK_SIZE_SB8X8);
+      vp9_subtract_sbuv(x, BLOCK_SIZE_SB8X8);
+      super_block_uvrd_for_txfm(cm, x, &rate_uv, &distortion_uv,
+                                &uv_skippable, BLOCK_SIZE_SB8X8, TX_4X4);
+      rate2 += rate_uv;
+      distortion2 += distortion_uv;
+      skippable = skippable && uv_skippable;
+
+      txfm_cache[ONLY_4X4] = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+      for (i = 0; i < NB_TXFM_MODES; ++i)
+        txfm_cache[i] = txfm_cache[ONLY_4X4];
+
+      if (!mode_excluded) {
+        if (is_comp_pred)
+          mode_excluded = cpi->common.comp_pred_mode == SINGLE_PREDICTION_ONLY;
+        else
+          mode_excluded = cpi->common.comp_pred_mode == COMP_PREDICTION_ONLY;
+      }
+
+      compmode_cost = vp9_cost_bit(comp_mode_p, is_comp_pred);
+    } else {
+      compmode_cost = vp9_cost_bit(comp_mode_p,
+                                   mbmi->ref_frame[1] > INTRA_FRAME);
+      this_rd = handle_inter_mode(cpi, x, bsize,
+                                  txfm_cache,
+                                  &rate2, &distortion2, &skippable,
+                                  &rate_y, &distortion_y,
+                                  &rate_uv, &distortion_uv,
+                                  &mode_excluded, &disable_skip,
+                                  &tmp_best_filter, frame_mv[this_mode],
+                                  mi_row, mi_col,
+                                  single_newmv);
+      if (this_rd == INT64_MAX)
+        continue;
+    }
+
+    if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+      rate2 += compmode_cost;
+    }
+
+    // Estimate the reference frame signaling cost and add it
+    // to the rolling cost variable.
+    if (mbmi->ref_frame[1] > INTRA_FRAME) {
+      rate2 += ref_costs_comp[mbmi->ref_frame[0]];
+    } else {
+      rate2 += ref_costs_single[mbmi->ref_frame[0]];
+    }
+
+    if (!disable_skip) {
+      // Test for the condition where skip block will be activated
+      // because there are no non zero coefficients and make any
+      // necessary adjustment for rate. Ignore if skip is coded at
+      // segment level as the cost wont have been added in.
+      int mb_skip_allowed;
+
+      // Is Mb level skip allowed (i.e. not coded at segment level).
+      mb_skip_allowed = !vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
+
+      if (skippable && bsize >= BLOCK_SIZE_SB8X8) {
+        // Back out the coefficient coding costs
+        rate2 -= (rate_y + rate_uv);
+        // for best_yrd calculation
+        rate_uv = 0;
+
+        if (mb_skip_allowed) {
+          int prob_skip_cost;
+
+          // Cost the skip mb case
+          vp9_prob skip_prob =
+            vp9_get_pred_prob(cm, xd, PRED_MBSKIP);
+
+          if (skip_prob) {
+            prob_skip_cost = vp9_cost_bit(skip_prob, 1);
+            rate2 += prob_skip_cost;
+          }
+        }
+      } else if (mb_skip_allowed) {
+        // Add in the cost of the no skip flag.
+        int prob_skip_cost = vp9_cost_bit(vp9_get_pred_prob(cm, xd,
+                                                        PRED_MBSKIP), 0);
+        rate2 += prob_skip_cost;
+      }
+
+      // Calculate the final RD estimate for this mode.
+      this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+    }
+
+#if 0
+    // Keep record of best intra distortion
+    if ((xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME) &&
+        (this_rd < best_intra_rd)) {
+      best_intra_rd = this_rd;
+      *returnintra = distortion2;
+    }
+#endif
+
+    if (!disable_skip && mbmi->ref_frame[0] == INTRA_FRAME)
+      for (i = 0; i < NB_PREDICTION_TYPES; ++i)
+        best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+
+    if (this_rd < best_overall_rd) {
+      best_overall_rd = this_rd;
+      best_filter = tmp_best_filter;
+      best_mode = this_mode;
+    }
+
+    if (this_mode != I4X4_PRED && this_mode != SPLITMV) {
+      // Store the respective mode distortions for later use.
+      if (mode_distortions[this_mode] == -1
+          || distortion2 < mode_distortions[this_mode]) {
+        mode_distortions[this_mode] = distortion2;
+      }
+      if (frame_distortions[mbmi->ref_frame[0]] == -1
+          || distortion2 < frame_distortions[mbmi->ref_frame[0]]) {
+        frame_distortions[mbmi->ref_frame[0]] = distortion2;
+      }
+    }
+
+    // Did this mode help.. i.e. is it the new best mode
+    if (this_rd < best_rd || x->skip) {
+      if (!mode_excluded) {
+        // Note index of best mode so far
+        best_mode_index = mode_index;
+
+        if (ref_frame == INTRA_FRAME) {
+          /* required for left and above block mv */
+          mbmi->mv[0].as_int = 0;
+        }
+
+        *returnrate = rate2;
+        *returndistortion = distortion2;
+        best_rd = this_rd;
+        best_mbmode = *mbmi;
+        best_partition = *x->partition_info;
+
+        if (this_mode == I4X4_PRED || this_mode == SPLITMV)
+          for (i = 0; i < 4; i++)
+            best_bmodes[i] = xd->mode_info_context->bmi[i];
+      }
+#if 0
+      // Testing this mode gave rise to an improvement in best error score.
+      // Lower threshold a bit for next time
+      cpi->rd_thresh_mult[mode_index] =
+          (cpi->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
+              cpi->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
+      cpi->rd_threshes[mode_index] =
+          (cpi->rd_baseline_thresh[mode_index] >> 7)
+              * cpi->rd_thresh_mult[mode_index];
+#endif
+    } else {
+      // If the mode did not help improve the best error case then
+      // raise the threshold for testing that mode next time around.
+#if 0
+      cpi->rd_thresh_mult[mode_index] += 4;
+
+      if (cpi->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
+        cpi->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
+
+      cpi->rd_threshes[mode_index] =
+          (cpi->rd_baseline_thresh[mode_index] >> 7)
+              * cpi->rd_thresh_mult[mode_index];
+#endif
+    }
+
+    /* keep record of best compound/single-only prediction */
+    if (!disable_skip && mbmi->ref_frame[0] != INTRA_FRAME) {
+      int single_rd, hybrid_rd, single_rate, hybrid_rate;
+
+      if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+        single_rate = rate2 - compmode_cost;
+        hybrid_rate = rate2;
+      } else {
+        single_rate = rate2;
+        hybrid_rate = rate2 + compmode_cost;
+      }
+
+      single_rd = RDCOST(x->rdmult, x->rddiv, single_rate, distortion2);
+      hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
+
+      if (mbmi->ref_frame[1] <= INTRA_FRAME &&
+          single_rd < best_pred_rd[SINGLE_PREDICTION_ONLY]) {
+        best_pred_rd[SINGLE_PREDICTION_ONLY] = single_rd;
+      } else if (mbmi->ref_frame[1] > INTRA_FRAME &&
+                 single_rd < best_pred_rd[COMP_PREDICTION_ONLY]) {
+        best_pred_rd[COMP_PREDICTION_ONLY] = single_rd;
+      }
+      if (hybrid_rd < best_pred_rd[HYBRID_PREDICTION])
+        best_pred_rd[HYBRID_PREDICTION] = hybrid_rd;
+    }
+
+    /* keep record of best txfm size */
+    if (bsize < BLOCK_SIZE_SB32X32) {
+      if (bsize < BLOCK_SIZE_MB16X16) {
+        if (this_mode == SPLITMV || this_mode == I4X4_PRED)
+          txfm_cache[ALLOW_8X8] = txfm_cache[ONLY_4X4];
+        txfm_cache[ALLOW_16X16] = txfm_cache[ALLOW_8X8];
+      }
+      txfm_cache[ALLOW_32X32] = txfm_cache[ALLOW_16X16];
+    }
+    if (!mode_excluded && this_rd != INT64_MAX) {
+      for (i = 0; i < NB_TXFM_MODES; i++) {
+        int64_t adj_rd = INT64_MAX;
+        if (this_mode != I4X4_PRED) {
+          adj_rd = this_rd + txfm_cache[i] - txfm_cache[cm->txfm_mode];
+        } else {
+          adj_rd = this_rd;
+        }
+
+        if (adj_rd < best_txfm_rd[i])
+          best_txfm_rd[i] = adj_rd;
+      }
+    }
+
+    if (x->skip && !mode_excluded)
+      break;
+  }
+  // Flag all modes that have a distortion thats > 2x the best we found at
+  // this level.
+  for (mode_index = 0; mode_index < MB_MODE_COUNT; ++mode_index) {
+    if (mode_index == NEARESTMV || mode_index == NEARMV || mode_index == NEWMV)
+      continue;
+
+    if (mode_distortions[mode_index] > 2 * *returndistortion) {
+      ctx->modes_with_high_error |= (1 << mode_index);
+    }
+  }
+
+  // Flag all ref frames that have a distortion thats > 2x the best we found at
+  // this level.
+  for (ref_frame = INTRA_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
+    if (frame_distortions[ref_frame] > 2 * *returndistortion) {
+      ctx->frames_with_high_error |= (1 << ref_frame);
+    }
+  }
+
+  if (best_rd == INT64_MAX && bsize < BLOCK_SIZE_SB8X8) {
+    *returnrate = INT_MAX;
+    *returndistortion = INT_MAX;
+    return best_rd;
+  }
+
+  assert((cm->mcomp_filter_type == SWITCHABLE) ||
+         (cm->mcomp_filter_type == best_mbmode.interp_filter) ||
+         (best_mbmode.ref_frame[0] == INTRA_FRAME));
+
+  // Accumulate filter usage stats
+  // TODO(agrange): Use RD criteria to select interpolation filter mode.
+  if (is_inter_mode(best_mode))
+    ++cpi->best_switchable_interp_count[vp9_switchable_interp_map[best_filter]];
+
+  // Updating rd_thresh_freq_fact[] here means that the differnt
+  // partition/block sizes are handled independently based on the best
+  // choice for the current partition. It may well be better to keep a scaled
+  // best rd so far value and update rd_thresh_freq_fact based on the mode/size
+  // combination that wins out.
+  if (cpi->sf.adpative_rd_thresh) {
+    for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
+      if (mode_index == best_mode_index) {
+        cpi->rd_thresh_freq_fact[bsize][mode_index] = BASE_RD_THRESH_FREQ_FACT;
+      } else {
+        cpi->rd_thresh_freq_fact[bsize][mode_index] += MAX_RD_THRESH_FREQ_INC;
+        if (cpi->rd_thresh_freq_fact[bsize][mode_index] >
+            (cpi->sf.adpative_rd_thresh * MAX_RD_THRESH_FREQ_FACT)) {
+          cpi->rd_thresh_freq_fact[bsize][mode_index] =
+            cpi->sf.adpative_rd_thresh * MAX_RD_THRESH_FREQ_FACT;
+        }
+      }
+    }
+  }
+
+  // TODO(rbultje) integrate with RD trd_thresh_freq_facthresholding
+#if 0
+  // Reduce the activation RD thresholds for the best choice mode
+  if ((cpi->rd_baseline_thresh[best_mode_index] > 0) &&
+      (cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2))) {
+    int best_adjustment = (cpi->rd_thresh_mult[best_mode_index] >> 2);
+
+    cpi->rd_thresh_mult[best_mode_index] =
+      (cpi->rd_thresh_mult[best_mode_index] >= (MIN_THRESHMULT + best_adjustment)) ?
+      cpi->rd_thresh_mult[best_mode_index] - best_adjustment : MIN_THRESHMULT;
+    cpi->rd_threshes[best_mode_index] =
+      (cpi->rd_baseline_thresh[best_mode_index] >> 7) * cpi->rd_thresh_mult[best_mode_index];
+  }
+#endif
+
+  // This code forces Altref,0,0 and skip for the frame that overlays a
+  // an alrtef unless Altref is filtered. However, this is unsafe if
+  // segment level coding of ref frame is enabled for this segment.
+  if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME) &&
+      cpi->is_src_frame_alt_ref &&
+      (cpi->oxcf.arnr_max_frames == 0) &&
+      (best_mbmode.mode != ZEROMV || best_mbmode.ref_frame[0] != ALTREF_FRAME)
+      && bsize >= BLOCK_SIZE_SB8X8) {
+    mbmi->mode = ZEROMV;
+    mbmi->ref_frame[0] = ALTREF_FRAME;
+    mbmi->ref_frame[1] = NONE;
+    mbmi->mv[0].as_int = 0;
+    mbmi->uv_mode = DC_PRED;
+    mbmi->mb_skip_coeff = 1;
+    if (cm->txfm_mode == TX_MODE_SELECT) {
+      if (bsize >= BLOCK_SIZE_SB32X32)
+        mbmi->txfm_size = TX_32X32;
+      else if (bsize >= BLOCK_SIZE_MB16X16)
+        mbmi->txfm_size = TX_16X16;
+      else
+        mbmi->txfm_size = TX_8X8;
+    }
+
+    vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
+    vpx_memset(best_pred_diff, 0, sizeof(best_pred_diff));
+    goto end;
+  }
+
+  // macroblock modes
+  *mbmi = best_mbmode;
+  if (best_mbmode.ref_frame[0] == INTRA_FRAME &&
+      best_mbmode.sb_type < BLOCK_SIZE_SB8X8) {
+    for (i = 0; i < 4; i++)
+      xd->mode_info_context->bmi[i].as_mode = best_bmodes[i].as_mode;
+  }
+
+  if (best_mbmode.ref_frame[0] != INTRA_FRAME &&
+      best_mbmode.sb_type < BLOCK_SIZE_SB8X8) {
+    for (i = 0; i < 4; i++)
+      xd->mode_info_context->bmi[i].as_mv[0].as_int =
+          best_bmodes[i].as_mv[0].as_int;
+
+    if (mbmi->ref_frame[1] > 0)
+      for (i = 0; i < 4; i++)
+        xd->mode_info_context->bmi[i].as_mv[1].as_int =
+            best_bmodes[i].as_mv[1].as_int;
+
+    *x->partition_info = best_partition;
+
+    mbmi->mv[0].as_int = x->partition_info->bmi[3].mv.as_int;
+    mbmi->mv[1].as_int = x->partition_info->bmi[3].second_mv.as_int;
+  }
+
+  for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
+    if (best_pred_rd[i] == INT64_MAX)
+      best_pred_diff[i] = INT_MIN;
+    else
+      best_pred_diff[i] = best_rd - best_pred_rd[i];
+  }
+
+  if (!x->skip) {
+    for (i = 0; i < NB_TXFM_MODES; i++) {
+      if (best_txfm_rd[i] == INT64_MAX)
+        best_txfm_diff[i] = 0;
+      else
+        best_txfm_diff[i] = best_rd - best_txfm_rd[i];
+    }
+  } else {
+    vpx_memset(best_txfm_diff, 0, sizeof(best_txfm_diff));
+  }
+
+ end:
+  set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
+                    scale_factor);
+  store_coding_context(x, ctx, best_mode_index,
+                       &best_partition,
+                       &mbmi->ref_mvs[mbmi->ref_frame[0]][0],
+                       &mbmi->ref_mvs[mbmi->ref_frame[1] < 0 ? 0 :
+                                      mbmi->ref_frame[1]][0],
+                       best_pred_diff, best_txfm_diff);
+
+  return best_rd;
+}
diff --git a/libvpx/vp9/encoder/vp9_rdopt.h b/libvpx/vp9/encoder/vp9_rdopt.h
new file mode 100644
index 0000000..dcf5d00
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_rdopt.h
@@ -0,0 +1,36 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_RDOPT_H_
+#define VP9_ENCODER_VP9_RDOPT_H_
+
+#define RDCOST(RM,DM,R,D) ( ((128+((int64_t)R)*(RM)) >> 8) + ((int64_t)DM)*(D) )
+#define RDCOST_8x8(RM,DM,R,D) ( ((128+((int64_t)R)*(RM)) >> 8) + ((int64_t)DM)*(D) )
+
+void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex);
+
+void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex);
+
+void vp9_rd_pick_intra_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+                               int *r, int *d, BLOCK_SIZE_TYPE bsize,
+                               PICK_MODE_CONTEXT *ctx);
+
+int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+                                  int mi_row, int mi_col,
+                                  int *r, int *d, BLOCK_SIZE_TYPE bsize,
+                                  PICK_MODE_CONTEXT *ctx);
+
+void vp9_init_me_luts();
+
+void vp9_set_mbmode_and_mvs(MACROBLOCK *x,
+                            MB_PREDICTION_MODE mb, int_mv *mv);
+
+#endif  // VP9_ENCODER_VP9_RDOPT_H_
diff --git a/libvpx/vp9/encoder/vp9_sad_c.c b/libvpx/vp9/encoder/vp9_sad_c.c
new file mode 100644
index 0000000..6b1ba49
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_sad_c.c
@@ -0,0 +1,686 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+#include "vp9/common/vp9_sadmxn.h"
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "./vp9_rtcd.h"
+
+unsigned int vp9_sad64x64_c(const uint8_t *src_ptr,
+                            int  src_stride,
+                            const uint8_t *ref_ptr,
+                            int  ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 64, 64);
+}
+
+unsigned int vp9_sad64x32_c(const uint8_t *src_ptr,
+                            int  src_stride,
+                            const uint8_t *ref_ptr,
+                            int  ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 64, 32);
+}
+
+void vp9_sad64x32x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad64x32(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad64x32(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad64x32(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad64x32(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+unsigned int vp9_sad32x64_c(const uint8_t *src_ptr,
+                            int  src_stride,
+                            const uint8_t *ref_ptr,
+                            int  ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 32, 64);
+}
+
+void vp9_sad32x64x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad32x64(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad32x64(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad32x64(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad32x64(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+unsigned int vp9_sad32x32_c(const uint8_t *src_ptr,
+                            int  src_stride,
+                            const uint8_t *ref_ptr,
+                            int  ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 32, 32);
+}
+
+unsigned int vp9_sad32x16_c(const uint8_t *src_ptr,
+                            int   src_stride,
+                            const uint8_t *ref_ptr,
+                            int   ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 32, 16);
+}
+
+void vp9_sad32x16x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad32x16(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad32x16(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad32x16(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad32x16(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+unsigned int vp9_sad16x32_c(const uint8_t *src_ptr,
+                            int   src_stride,
+                            const uint8_t *ref_ptr,
+                            int   ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 16, 32);
+}
+
+void vp9_sad16x32x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad16x32(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad16x32(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad16x32(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad16x32(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+unsigned int vp9_sad16x16_c(const uint8_t *src_ptr,
+                            int  src_stride,
+                            const uint8_t *ref_ptr,
+                            int  ref_stride,
+                            unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 16, 16);
+}
+
+unsigned int vp9_sad8x8_c(const uint8_t *src_ptr,
+                          int  src_stride,
+                          const uint8_t *ref_ptr,
+                          int  ref_stride,
+                          unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 8, 8);
+}
+
+
+unsigned int vp9_sad16x8_c(const uint8_t *src_ptr,
+                           int  src_stride,
+                           const uint8_t *ref_ptr,
+                           int  ref_stride,
+                           unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 16, 8);
+}
+
+unsigned int vp9_sad8x16_c(const uint8_t *src_ptr,
+                           int  src_stride,
+                           const uint8_t *ref_ptr,
+                           int  ref_stride,
+                           unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 8, 16);
+}
+
+unsigned int vp9_sad8x4_c(const uint8_t *src_ptr,
+                          int src_stride,
+                          const uint8_t *ref_ptr,
+                          int ref_stride,
+                          unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 8, 4);
+}
+
+unsigned int vp9_sad4x8_c(const uint8_t *src_ptr,
+                          int src_stride,
+                          const uint8_t *ref_ptr,
+                          int ref_stride,
+                          unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 4, 8);
+}
+
+unsigned int vp9_sad4x4_c(const uint8_t *src_ptr,
+                          int  src_stride,
+                          const uint8_t *ref_ptr,
+                          int  ref_stride,
+                          unsigned int max_sad) {
+  return sad_mx_n_c(src_ptr, src_stride, ref_ptr, ref_stride, 4, 4);
+}
+
+void vp9_sad64x64x3_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad64x64(src_ptr, src_stride, ref_ptr, ref_stride,
+                              0x7fffffff);
+  sad_array[1] = vp9_sad64x64(src_ptr, src_stride, ref_ptr + 1, ref_stride,
+                              0x7fffffff);
+  sad_array[2] = vp9_sad64x64(src_ptr, src_stride, ref_ptr + 2, ref_stride,
+                              0x7fffffff);
+}
+
+void vp9_sad32x32x3_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad64x64x8_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr, ref_stride,
+                              0x7fffffff);
+  sad_array[1] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 1, ref_stride,
+                              0x7fffffff);
+  sad_array[2] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 2, ref_stride,
+                              0x7fffffff);
+  sad_array[3] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 3, ref_stride,
+                              0x7fffffff);
+  sad_array[4] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 4, ref_stride,
+                              0x7fffffff);
+  sad_array[5] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 5, ref_stride,
+                              0x7fffffff);
+  sad_array[6] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 6, ref_stride,
+                              0x7fffffff);
+  sad_array[7] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr + 7, ref_stride,
+                              0x7fffffff);
+}
+
+void vp9_sad32x32x8_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr, ref_stride,
+                              0x7fffffff);
+  sad_array[1] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 1, ref_stride,
+                              0x7fffffff);
+  sad_array[2] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 2, ref_stride,
+                              0x7fffffff);
+  sad_array[3] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 3, ref_stride,
+                              0x7fffffff);
+  sad_array[4] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 4, ref_stride,
+                              0x7fffffff);
+  sad_array[5] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 5, ref_stride,
+                              0x7fffffff);
+  sad_array[6] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 6, ref_stride,
+                              0x7fffffff);
+  sad_array[7] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr + 7, ref_stride,
+                              0x7fffffff);
+}
+
+void vp9_sad16x16x3_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad16x16x8_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t *ref_ptr,
+                      int  ref_stride,
+                      uint32_t *sad_array) {
+  sad_array[0] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr, ref_stride,
+                              0x7fffffff);
+  sad_array[1] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 1, ref_stride,
+                              0x7fffffff);
+  sad_array[2] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 2, ref_stride,
+                              0x7fffffff);
+  sad_array[3] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 3, ref_stride,
+                              0x7fffffff);
+  sad_array[4] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 4, ref_stride,
+                              0x7fffffff);
+  sad_array[5] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 5, ref_stride,
+                              0x7fffffff);
+  sad_array[6] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 6, ref_stride,
+                              0x7fffffff);
+  sad_array[7] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr + 7, ref_stride,
+                              0x7fffffff);
+}
+
+void vp9_sad16x8x3_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad16x8x8_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     uint32_t *sad_array) {
+  sad_array[0] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr, ref_stride,
+                             0x7fffffff);
+  sad_array[1] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride,
+                             0x7fffffff);
+  sad_array[2] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride,
+                             0x7fffffff);
+  sad_array[3] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 3, ref_stride,
+                             0x7fffffff);
+  sad_array[4] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 4, ref_stride,
+                             0x7fffffff);
+  sad_array[5] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 5, ref_stride,
+                             0x7fffffff);
+  sad_array[6] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 6, ref_stride,
+                             0x7fffffff);
+  sad_array[7] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr + 7, ref_stride,
+                             0x7fffffff);
+}
+
+void vp9_sad8x8x3_c(const uint8_t *src_ptr,
+                    int  src_stride,
+                    const uint8_t *ref_ptr,
+                    int  ref_stride,
+                    unsigned int *sad_array) {
+  sad_array[0] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x8x8_c(const uint8_t *src_ptr,
+                    int  src_stride,
+                    const uint8_t *ref_ptr,
+                    int  ref_stride,
+                    uint32_t *sad_array) {
+  sad_array[0] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr, ref_stride,
+                            0x7fffffff);
+  sad_array[1] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 1, ref_stride,
+                            0x7fffffff);
+  sad_array[2] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 2, ref_stride,
+                            0x7fffffff);
+  sad_array[3] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 3, ref_stride,
+                            0x7fffffff);
+  sad_array[4] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 4, ref_stride,
+                            0x7fffffff);
+  sad_array[5] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 5, ref_stride,
+                            0x7fffffff);
+  sad_array[6] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 6, ref_stride,
+                            0x7fffffff);
+  sad_array[7] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr + 7, ref_stride,
+                            0x7fffffff);
+}
+
+void vp9_sad8x16x3_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x16x8_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     uint32_t *sad_array) {
+  sad_array[0] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr, ref_stride,
+                             0x7fffffff);
+  sad_array[1] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride,
+                             0x7fffffff);
+  sad_array[2] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride,
+                             0x7fffffff);
+  sad_array[3] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 3, ref_stride,
+                             0x7fffffff);
+  sad_array[4] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 4, ref_stride,
+                             0x7fffffff);
+  sad_array[5] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 5, ref_stride,
+                             0x7fffffff);
+  sad_array[6] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 6, ref_stride,
+                             0x7fffffff);
+  sad_array[7] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr + 7, ref_stride,
+                             0x7fffffff);
+}
+
+void vp9_sad4x4x3_c(const uint8_t *src_ptr,
+                    int  src_stride,
+                    const uint8_t *ref_ptr,
+                    int  ref_stride,
+                    unsigned int *sad_array) {
+  sad_array[0] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr, ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 1, ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 2, ref_stride, 0x7fffffff);
+}
+
+void vp9_sad4x4x8_c(const uint8_t *src_ptr,
+                    int  src_stride,
+                    const uint8_t *ref_ptr,
+                    int  ref_stride,
+                    uint32_t *sad_array) {
+  sad_array[0] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr, ref_stride,
+                            0x7fffffff);
+  sad_array[1] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 1, ref_stride,
+                            0x7fffffff);
+  sad_array[2] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 2, ref_stride,
+                            0x7fffffff);
+  sad_array[3] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 3, ref_stride,
+                            0x7fffffff);
+  sad_array[4] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 4, ref_stride,
+                            0x7fffffff);
+  sad_array[5] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 5, ref_stride,
+                            0x7fffffff);
+  sad_array[6] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 6, ref_stride,
+                            0x7fffffff);
+  sad_array[7] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr + 7, ref_stride,
+                            0x7fffffff);
+}
+
+void vp9_sad64x64x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad64x64(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad32x32x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad32x32(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad16x16x4d_c(const uint8_t *src_ptr,
+                       int  src_stride,
+                       const uint8_t* const ref_ptr[],
+                       int  ref_stride,
+                       unsigned int *sad_array) {
+  sad_array[0] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad16x16(src_ptr, src_stride,
+                              ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad16x8x4d_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t* const ref_ptr[],
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad16x8(src_ptr, src_stride,
+                             ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x8x4d_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t* const ref_ptr[],
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad8x8(src_ptr, src_stride,
+                            ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x16x4d_c(const uint8_t *src_ptr,
+                      int  src_stride,
+                      const uint8_t* const ref_ptr[],
+                      int  ref_stride,
+                      unsigned int *sad_array) {
+  sad_array[0] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad8x16(src_ptr, src_stride,
+                             ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x4x4d_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t* const ref_ptr[],
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad8x4(src_ptr, src_stride,
+                            ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad8x4(src_ptr, src_stride,
+                            ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad8x4(src_ptr, src_stride,
+                            ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad8x4(src_ptr, src_stride,
+                            ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad8x4x8_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     uint32_t *sad_array) {
+  sad_array[0] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr, ref_stride,
+                             0x7fffffff);
+  sad_array[1] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride,
+                             0x7fffffff);
+  sad_array[2] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride,
+                             0x7fffffff);
+  sad_array[3] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 3, ref_stride,
+                             0x7fffffff);
+  sad_array[4] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 4, ref_stride,
+                             0x7fffffff);
+  sad_array[5] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 5, ref_stride,
+                             0x7fffffff);
+  sad_array[6] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 6, ref_stride,
+                             0x7fffffff);
+  sad_array[7] = vp9_sad8x4(src_ptr, src_stride,
+                             ref_ptr + 7, ref_stride,
+                             0x7fffffff);
+}
+
+void vp9_sad4x8x4d_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t* const ref_ptr[],
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad4x8(src_ptr, src_stride,
+                            ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad4x8(src_ptr, src_stride,
+                            ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad4x8(src_ptr, src_stride,
+                            ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad4x8(src_ptr, src_stride,
+                            ref_ptr[3], ref_stride, 0x7fffffff);
+}
+
+void vp9_sad4x8x8_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t *ref_ptr,
+                     int  ref_stride,
+                     uint32_t *sad_array) {
+  sad_array[0] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr, ref_stride,
+                             0x7fffffff);
+  sad_array[1] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 1, ref_stride,
+                             0x7fffffff);
+  sad_array[2] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 2, ref_stride,
+                             0x7fffffff);
+  sad_array[3] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 3, ref_stride,
+                             0x7fffffff);
+  sad_array[4] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 4, ref_stride,
+                             0x7fffffff);
+  sad_array[5] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 5, ref_stride,
+                             0x7fffffff);
+  sad_array[6] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 6, ref_stride,
+                             0x7fffffff);
+  sad_array[7] = vp9_sad4x8(src_ptr, src_stride,
+                             ref_ptr + 7, ref_stride,
+                             0x7fffffff);
+}
+
+void vp9_sad4x4x4d_c(const uint8_t *src_ptr,
+                     int  src_stride,
+                     const uint8_t* const ref_ptr[],
+                     int  ref_stride,
+                     unsigned int *sad_array) {
+  sad_array[0] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr[0], ref_stride, 0x7fffffff);
+  sad_array[1] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr[1], ref_stride, 0x7fffffff);
+  sad_array[2] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr[2], ref_stride, 0x7fffffff);
+  sad_array[3] = vp9_sad4x4(src_ptr, src_stride,
+                            ref_ptr[3], ref_stride, 0x7fffffff);
+}
diff --git a/libvpx/vp9/encoder/vp9_segmentation.c b/libvpx/vp9/encoder/vp9_segmentation.c
new file mode 100644
index 0000000..fe995ad
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_segmentation.c
@@ -0,0 +1,295 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <limits.h>
+#include "vpx_mem/vpx_mem.h"
+#include "vp9/encoder/vp9_segmentation.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_tile_common.h"
+
+void vp9_enable_segmentation(VP9_PTR ptr) {
+  VP9_COMP *cpi = (VP9_COMP *)ptr;
+
+  cpi->mb.e_mbd.segmentation_enabled = 1;
+  cpi->mb.e_mbd.update_mb_segmentation_map = 1;
+  cpi->mb.e_mbd.update_mb_segmentation_data = 1;
+}
+
+void vp9_disable_segmentation(VP9_PTR ptr) {
+  VP9_COMP *cpi = (VP9_COMP *)ptr;
+  cpi->mb.e_mbd.segmentation_enabled = 0;
+}
+
+void vp9_set_segmentation_map(VP9_PTR ptr,
+                              unsigned char *segmentation_map) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+
+  // Copy in the new segmentation map
+  vpx_memcpy(cpi->segmentation_map, segmentation_map,
+             (cpi->common.mi_rows * cpi->common.mi_cols));
+
+  // Signal that the map should be updated.
+  cpi->mb.e_mbd.update_mb_segmentation_map = 1;
+  cpi->mb.e_mbd.update_mb_segmentation_data = 1;
+}
+
+void vp9_set_segment_data(VP9_PTR ptr,
+                          signed char *feature_data,
+                          unsigned char abs_delta) {
+  VP9_COMP *cpi = (VP9_COMP *)(ptr);
+
+  cpi->mb.e_mbd.mb_segment_abs_delta = abs_delta;
+
+  vpx_memcpy(cpi->mb.e_mbd.segment_feature_data, feature_data,
+             sizeof(cpi->mb.e_mbd.segment_feature_data));
+
+  // TBD ?? Set the feature mask
+  // vpx_memcpy(cpi->mb.e_mbd.segment_feature_mask, 0,
+  //            sizeof(cpi->mb.e_mbd.segment_feature_mask));
+}
+
+// Based on set of segment counts calculate a probability tree
+static void calc_segtree_probs(MACROBLOCKD *xd, int *segcounts,
+                               vp9_prob *segment_tree_probs) {
+  // Work out probabilities of each segment
+  const int c01 = segcounts[0] + segcounts[1];
+  const int c23 = segcounts[2] + segcounts[3];
+  const int c45 = segcounts[4] + segcounts[5];
+  const int c67 = segcounts[6] + segcounts[7];
+
+  segment_tree_probs[0] = get_binary_prob(c01 + c23, c45 + c67);
+  segment_tree_probs[1] = get_binary_prob(c01, c23);
+  segment_tree_probs[2] = get_binary_prob(c45, c67);
+  segment_tree_probs[3] = get_binary_prob(segcounts[0], segcounts[1]);
+  segment_tree_probs[4] = get_binary_prob(segcounts[2], segcounts[3]);
+  segment_tree_probs[5] = get_binary_prob(segcounts[4], segcounts[5]);
+  segment_tree_probs[6] = get_binary_prob(segcounts[6], segcounts[7]);
+}
+
+// Based on set of segment counts and probabilities calculate a cost estimate
+static int cost_segmap(MACROBLOCKD *xd, int *segcounts, vp9_prob *probs) {
+  const int c01 = segcounts[0] + segcounts[1];
+  const int c23 = segcounts[2] + segcounts[3];
+  const int c45 = segcounts[4] + segcounts[5];
+  const int c67 = segcounts[6] + segcounts[7];
+  const int c0123 = c01 + c23;
+  const int c4567 = c45 + c67;
+
+  // Cost the top node of the tree
+  int cost = c0123 * vp9_cost_zero(probs[0]) +
+             c4567 * vp9_cost_one(probs[0]);
+
+  // Cost subsequent levels
+  if (c0123 > 0) {
+    cost += c01 * vp9_cost_zero(probs[1]) +
+            c23 * vp9_cost_one(probs[1]);
+
+    if (c01 > 0)
+      cost += segcounts[0] * vp9_cost_zero(probs[3]) +
+              segcounts[1] * vp9_cost_one(probs[3]);
+    if (c23 > 0)
+      cost += segcounts[2] * vp9_cost_zero(probs[4]) +
+              segcounts[3] * vp9_cost_one(probs[4]);
+  }
+
+  if (c4567 > 0) {
+    cost += c45 * vp9_cost_zero(probs[2]) +
+            c67 * vp9_cost_one(probs[2]);
+
+    if (c45 > 0)
+      cost += segcounts[4] * vp9_cost_zero(probs[5]) +
+              segcounts[5] * vp9_cost_one(probs[5]);
+    if (c67 > 0)
+      cost += segcounts[6] * vp9_cost_zero(probs[6]) +
+              segcounts[7] * vp9_cost_one(probs[6]);
+  }
+
+  return cost;
+}
+
+static void count_segs(VP9_COMP *cpi,
+                       MODE_INFO *mi,
+                       int *no_pred_segcounts,
+                       int (*temporal_predictor_count)[2],
+                       int *t_unpred_seg_counts,
+                       int bw, int bh, int mi_row, int mi_col) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  int segment_id;
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  segment_id = mi->mbmi.segment_id;
+  xd->mode_info_context = mi;
+  set_mi_row_col(cm, xd, mi_row, bh, mi_col, bw);
+
+  // Count the number of hits on each segment with no prediction
+  no_pred_segcounts[segment_id]++;
+
+  // Temporal prediction not allowed on key frames
+  if (cm->frame_type != KEY_FRAME) {
+    // Test to see if the segment id matches the predicted value.
+    const int pred_seg_id = vp9_get_pred_mi_segid(cm, mi->mbmi.sb_type,
+                                                  mi_row, mi_col);
+    const int seg_predicted = (segment_id == pred_seg_id);
+
+    // Get the segment id prediction context
+    const int pred_context = vp9_get_pred_context(cm, xd, PRED_SEG_ID);
+
+    // Store the prediction status for this mb and update counts
+    // as appropriate
+    vp9_set_pred_flag(xd, PRED_SEG_ID, seg_predicted);
+    temporal_predictor_count[pred_context][seg_predicted]++;
+
+    if (!seg_predicted)
+      // Update the "unpredicted" segment count
+      t_unpred_seg_counts[segment_id]++;
+  }
+}
+
+static void count_segs_sb(VP9_COMP *cpi, MODE_INFO *mi,
+                          int *no_pred_segcounts,
+                          int (*temporal_predictor_count)[2],
+                          int *t_unpred_seg_counts,
+                          int mi_row, int mi_col,
+                          BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  const int mis = cm->mode_info_stride;
+  int bwl, bhl;
+  const int bsl = mi_width_log2(bsize), bs = 1 << (bsl - 1);
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  bwl = mi_width_log2(mi->mbmi.sb_type);
+  bhl = mi_height_log2(mi->mbmi.sb_type);
+
+  if (bwl == bsl && bhl == bsl) {
+    count_segs(cpi, mi, no_pred_segcounts, temporal_predictor_count,
+               t_unpred_seg_counts, 1 << bsl, 1 << bsl, mi_row, mi_col);
+  } else if (bwl == bsl && bhl < bsl) {
+    count_segs(cpi, mi, no_pred_segcounts, temporal_predictor_count,
+               t_unpred_seg_counts, 1 << bsl, bs, mi_row, mi_col);
+    count_segs(cpi, mi + bs * mis, no_pred_segcounts, temporal_predictor_count,
+               t_unpred_seg_counts, 1 << bsl, bs, mi_row + bs, mi_col);
+  } else if (bwl < bsl && bhl == bsl) {
+    count_segs(cpi, mi, no_pred_segcounts, temporal_predictor_count,
+               t_unpred_seg_counts, bs, 1 << bsl, mi_row, mi_col);
+    count_segs(cpi, mi + bs, no_pred_segcounts, temporal_predictor_count,
+               t_unpred_seg_counts, bs, 1 << bsl, mi_row, mi_col + bs);
+  } else {
+    BLOCK_SIZE_TYPE subsize;
+    int n;
+
+    assert(bwl < bsl && bhl < bsl);
+    if (bsize == BLOCK_SIZE_SB64X64) {
+      subsize = BLOCK_SIZE_SB32X32;
+    } else if (bsize == BLOCK_SIZE_SB32X32) {
+      subsize = BLOCK_SIZE_MB16X16;
+    } else {
+      assert(bsize == BLOCK_SIZE_MB16X16);
+      subsize = BLOCK_SIZE_SB8X8;
+    }
+
+    for (n = 0; n < 4; n++) {
+      const int y_idx = n >> 1, x_idx = n & 0x01;
+
+      count_segs_sb(cpi, mi + y_idx * bs * mis + x_idx * bs,
+                    no_pred_segcounts, temporal_predictor_count,
+                    t_unpred_seg_counts,
+                    mi_row + y_idx * bs, mi_col + x_idx * bs, subsize);
+    }
+  }
+}
+
+void vp9_choose_segmap_coding_method(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+
+  int no_pred_cost;
+  int t_pred_cost = INT_MAX;
+
+  int i;
+  int tile_col, mi_row, mi_col;
+
+  int temporal_predictor_count[PREDICTION_PROBS][2];
+  int no_pred_segcounts[MAX_MB_SEGMENTS];
+  int t_unpred_seg_counts[MAX_MB_SEGMENTS];
+
+  vp9_prob no_pred_tree[MB_SEG_TREE_PROBS];
+  vp9_prob t_pred_tree[MB_SEG_TREE_PROBS];
+  vp9_prob t_nopred_prob[PREDICTION_PROBS];
+
+  const int mis = cm->mode_info_stride;
+  MODE_INFO *mi_ptr, *mi;
+
+  // Set default state for the segment tree probabilities and the
+  // temporal coding probabilities
+  vpx_memset(xd->mb_segment_tree_probs, 255, sizeof(xd->mb_segment_tree_probs));
+  vpx_memset(cm->segment_pred_probs, 255, sizeof(cm->segment_pred_probs));
+
+  vpx_memset(no_pred_segcounts, 0, sizeof(no_pred_segcounts));
+  vpx_memset(t_unpred_seg_counts, 0, sizeof(t_unpred_seg_counts));
+  vpx_memset(temporal_predictor_count, 0, sizeof(temporal_predictor_count));
+
+  // First of all generate stats regarding how well the last segment map
+  // predicts this one
+  for (tile_col = 0; tile_col < cm->tile_columns; tile_col++) {
+    vp9_get_tile_col_offsets(cm, tile_col);
+    mi_ptr = cm->mi + cm->cur_tile_mi_col_start;
+    for (mi_row = 0; mi_row < cm->mi_rows;
+         mi_row += 8, mi_ptr += 8 * mis) {
+      mi = mi_ptr;
+      for (mi_col = cm->cur_tile_mi_col_start;
+           mi_col < cm->cur_tile_mi_col_end;
+           mi_col += 8, mi += 8) {
+        count_segs_sb(cpi, mi, no_pred_segcounts, temporal_predictor_count,
+                      t_unpred_seg_counts, mi_row, mi_col, BLOCK_SIZE_SB64X64);
+      }
+    }
+  }
+
+  // Work out probability tree for coding segments without prediction
+  // and the cost.
+  calc_segtree_probs(xd, no_pred_segcounts, no_pred_tree);
+  no_pred_cost = cost_segmap(xd, no_pred_segcounts, no_pred_tree);
+
+  // Key frames cannot use temporal prediction
+  if (cm->frame_type != KEY_FRAME) {
+    // Work out probability tree for coding those segments not
+    // predicted using the temporal method and the cost.
+    calc_segtree_probs(xd, t_unpred_seg_counts, t_pred_tree);
+    t_pred_cost = cost_segmap(xd, t_unpred_seg_counts, t_pred_tree);
+
+    // Add in the cost of the signalling for each prediction context
+    for (i = 0; i < PREDICTION_PROBS; i++) {
+      const int count0 = temporal_predictor_count[i][0];
+      const int count1 = temporal_predictor_count[i][1];
+
+      t_nopred_prob[i] = get_binary_prob(count0, count1);
+
+      // Add in the predictor signaling cost
+      t_pred_cost += count0 * vp9_cost_zero(t_nopred_prob[i]) +
+                     count1 * vp9_cost_one(t_nopred_prob[i]);
+    }
+  }
+
+  // Now choose which coding method to use.
+  if (t_pred_cost < no_pred_cost) {
+    cm->temporal_update = 1;
+    vpx_memcpy(xd->mb_segment_tree_probs, t_pred_tree, sizeof(t_pred_tree));
+    vpx_memcpy(cm->segment_pred_probs, t_nopred_prob, sizeof(t_nopred_prob));
+  } else {
+    cm->temporal_update = 0;
+    vpx_memcpy(xd->mb_segment_tree_probs, no_pred_tree, sizeof(no_pred_tree));
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_segmentation.h b/libvpx/vp9/encoder/vp9_segmentation.h
new file mode 100644
index 0000000..2183771
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_segmentation.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_SEGMENTATION_H_
+#define VP9_ENCODER_VP9_SEGMENTATION_H_
+
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+
+void vp9_enable_segmentation(VP9_PTR ptr);
+void vp9_disable_segmentation(VP9_PTR ptr);
+
+// Valid values for a segment are 0 to 3
+// Segmentation map is arrange as [Rows][Columns]
+void vp9_set_segmentation_map(VP9_PTR ptr, unsigned char *segmentation_map);
+
+// The values given for each segment can be either deltas (from the default
+// value chosen for the frame) or absolute values.
+//
+// Valid range for abs values is (0-127 for MB_LVL_ALT_Q), (0-63 for
+// SEGMENT_ALT_LF)
+// Valid range for delta values are (+/-127 for MB_LVL_ALT_Q), (+/-63 for
+// SEGMENT_ALT_LF)
+//
+// abs_delta = SEGMENT_DELTADATA (deltas) abs_delta = SEGMENT_ABSDATA (use
+// the absolute values given).
+void vp9_set_segment_data(VP9_PTR ptr, signed char *feature_data,
+                          unsigned char abs_delta);
+
+void vp9_choose_segmap_coding_method(VP9_COMP *cpi);
+
+#endif  // VP9_ENCODER_VP9_SEGMENTATION_H_
diff --git a/libvpx/vp9/encoder/vp9_ssim.c b/libvpx/vp9/encoder/vp9_ssim.c
new file mode 100644
index 0000000..363ed84
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_ssim.c
@@ -0,0 +1,147 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/encoder/vp9_onyx_int.h"
+
+void vp9_ssim_parms_16x16_c(uint8_t *s, int sp, uint8_t *r,
+                            int rp, unsigned long *sum_s, unsigned long *sum_r,
+                            unsigned long *sum_sq_s, unsigned long *sum_sq_r,
+                            unsigned long *sum_sxr) {
+  int i, j;
+  for (i = 0; i < 16; i++, s += sp, r += rp) {
+    for (j = 0; j < 16; j++) {
+      *sum_s += s[j];
+      *sum_r += r[j];
+      *sum_sq_s += s[j] * s[j];
+      *sum_sq_r += r[j] * r[j];
+      *sum_sxr += s[j] * r[j];
+    }
+  }
+}
+void vp9_ssim_parms_8x8_c(uint8_t *s, int sp, uint8_t *r, int rp,
+                          unsigned long *sum_s, unsigned long *sum_r,
+                          unsigned long *sum_sq_s, unsigned long *sum_sq_r,
+                          unsigned long *sum_sxr) {
+  int i, j;
+  for (i = 0; i < 8; i++, s += sp, r += rp) {
+    for (j = 0; j < 8; j++) {
+      *sum_s += s[j];
+      *sum_r += r[j];
+      *sum_sq_s += s[j] * s[j];
+      *sum_sq_r += r[j] * r[j];
+      *sum_sxr += s[j] * r[j];
+    }
+  }
+}
+
+const static int64_t cc1 =  26634; // (64^2*(.01*255)^2
+const static int64_t cc2 = 239708; // (64^2*(.03*255)^2
+
+static double similarity(unsigned long sum_s, unsigned long sum_r,
+                         unsigned long sum_sq_s, unsigned long sum_sq_r,
+                         unsigned long sum_sxr, int count) {
+  int64_t ssim_n, ssim_d;
+  int64_t c1, c2;
+
+  // scale the constants by number of pixels
+  c1 = (cc1 * count * count) >> 12;
+  c2 = (cc2 * count * count) >> 12;
+
+  ssim_n = (2 * sum_s * sum_r + c1) * ((int64_t) 2 * count * sum_sxr -
+                                       (int64_t) 2 * sum_s * sum_r + c2);
+
+  ssim_d = (sum_s * sum_s + sum_r * sum_r + c1) *
+           ((int64_t)count * sum_sq_s - (int64_t)sum_s * sum_s +
+            (int64_t)count * sum_sq_r - (int64_t) sum_r * sum_r + c2);
+
+  return ssim_n * 1.0 / ssim_d;
+}
+
+static double ssim_16x16(uint8_t *s, int sp, uint8_t *r, int rp) {
+  unsigned long sum_s = 0, sum_r = 0, sum_sq_s = 0, sum_sq_r = 0, sum_sxr = 0;
+  vp9_ssim_parms_16x16(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r,
+                       &sum_sxr);
+  return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 256);
+}
+static double ssim_8x8(uint8_t *s, int sp, uint8_t *r, int rp) {
+  unsigned long sum_s = 0, sum_r = 0, sum_sq_s = 0, sum_sq_r = 0, sum_sxr = 0;
+  vp9_ssim_parms_8x8(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r,
+                     &sum_sxr);
+  return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 64);
+}
+
+// We are using a 8x8 moving window with starting location of each 8x8 window
+// on the 4x4 pixel grid. Such arrangement allows the windows to overlap
+// block boundaries to penalize blocking artifacts.
+double vp9_ssim2(uint8_t *img1, uint8_t *img2, int stride_img1,
+                 int stride_img2, int width, int height) {
+  int i, j;
+  int samples = 0;
+  double ssim_total = 0;
+
+  // sample point start with each 4x4 location
+  for (i = 0; i < height - 8; i += 4, img1 += stride_img1 * 4, img2 += stride_img2 * 4) {
+    for (j = 0; j < width - 8; j += 4) {
+      double v = ssim_8x8(img1 + j, stride_img1, img2 + j, stride_img2);
+      ssim_total += v;
+      samples++;
+    }
+  }
+  ssim_total /= samples;
+  return ssim_total;
+}
+double vp9_calc_ssim(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest,
+                     int lumamask, double *weight) {
+  double a, b, c;
+  double ssimv;
+
+  a = vp9_ssim2(source->y_buffer, dest->y_buffer,
+                source->y_stride, dest->y_stride, source->y_width,
+                source->y_height);
+
+  b = vp9_ssim2(source->u_buffer, dest->u_buffer,
+                source->uv_stride, dest->uv_stride, source->uv_width,
+                source->uv_height);
+
+  c = vp9_ssim2(source->v_buffer, dest->v_buffer,
+                source->uv_stride, dest->uv_stride, source->uv_width,
+                source->uv_height);
+
+  ssimv = a * .8 + .1 * (b + c);
+
+  *weight = 1;
+
+  return ssimv;
+}
+
+double vp9_calc_ssimg(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest,
+                      double *ssim_y, double *ssim_u, double *ssim_v) {
+  double ssim_all = 0;
+  double a, b, c;
+
+  a = vp9_ssim2(source->y_buffer, dest->y_buffer,
+                source->y_stride, dest->y_stride, source->y_width,
+                source->y_height);
+
+  b = vp9_ssim2(source->u_buffer, dest->u_buffer,
+                source->uv_stride, dest->uv_stride, source->uv_width,
+                source->uv_height);
+
+  c = vp9_ssim2(source->v_buffer, dest->v_buffer,
+                source->uv_stride, dest->uv_stride, source->uv_width,
+                source->uv_height);
+  *ssim_y = a;
+  *ssim_u = b;
+  *ssim_v = c;
+  ssim_all = (a * 4 + b + c) / 6;
+
+  return ssim_all;
+}
diff --git a/libvpx/vp9/encoder/vp9_temporal_filter.c b/libvpx/vp9/encoder/vp9_temporal_filter.c
new file mode 100644
index 0000000..47792fc
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_temporal_filter.c
@@ -0,0 +1,529 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+#include <limits.h>
+
+#include "vp9/common/vp9_onyxc_int.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_quantize.h"
+#include "vp9/common/vp9_alloccommon.h"
+#include "vp9/encoder/vp9_mcomp.h"
+#include "vp9/encoder/vp9_firstpass.h"
+#include "vp9/encoder/vp9_psnr.h"
+#include "vpx_scale/vpx_scale.h"
+#include "vp9/common/vp9_extend.h"
+#include "vp9/encoder/vp9_ratectrl.h"
+#include "vp9/common/vp9_quant_common.h"
+#include "vp9/encoder/vp9_segmentation.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/vpx_timer.h"
+
+#define ALT_REF_MC_ENABLED 1    // dis/enable MC in AltRef filtering
+#define ALT_REF_SUBPEL_ENABLED 1 // dis/enable subpel in MC AltRef filtering
+
+static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
+                                            uint8_t *y_mb_ptr,
+                                            uint8_t *u_mb_ptr,
+                                            uint8_t *v_mb_ptr,
+                                            int stride,
+                                            int mv_row,
+                                            int mv_col,
+                                            uint8_t *pred) {
+  const int which_mv = 0;
+  int_mv mv;
+
+  mv.as_mv.row = mv_row;
+  mv.as_mv.col = mv_col;
+
+  vp9_build_inter_predictor(y_mb_ptr, stride,
+                            &pred[0], 16,
+                            &mv,
+                            &xd->scale_factor[which_mv],
+                            16, 16,
+                            which_mv,
+                            &xd->subpix);
+
+  stride = (stride + 1) >> 1;
+
+  vp9_build_inter_predictor_q4(u_mb_ptr, stride,
+                               &pred[256], 8,
+                               &mv,
+                               &xd->scale_factor_uv[which_mv],
+                               8, 8,
+                               which_mv,
+                               &xd->subpix);
+
+  vp9_build_inter_predictor_q4(v_mb_ptr, stride,
+                               &pred[320], 8,
+                               &mv,
+                               &xd->scale_factor_uv[which_mv],
+                               8, 8,
+                               which_mv,
+                               &xd->subpix);
+}
+
+void vp9_temporal_filter_apply_c(uint8_t *frame1,
+                                 unsigned int stride,
+                                 uint8_t *frame2,
+                                 unsigned int block_size,
+                                 int strength,
+                                 int filter_weight,
+                                 unsigned int *accumulator,
+                                 uint16_t *count) {
+  unsigned int i, j, k;
+  int modifier;
+  int byte = 0;
+
+  for (i = 0, k = 0; i < block_size; i++) {
+    for (j = 0; j < block_size; j++, k++) {
+
+      int src_byte = frame1[byte];
+      int pixel_value = *frame2++;
+
+      modifier   = src_byte - pixel_value;
+      // This is an integer approximation of:
+      // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
+      // modifier =  (int)roundf(coeff > 16 ? 0 : 16-coeff);
+      modifier  *= modifier;
+      modifier  *= 3;
+      modifier  += 1 << (strength - 1);
+      modifier >>= strength;
+
+      if (modifier > 16)
+        modifier = 16;
+
+      modifier = 16 - modifier;
+      modifier *= filter_weight;
+
+      count[k] += modifier;
+      accumulator[k] += modifier * pixel_value;
+
+      byte++;
+    }
+
+    byte += stride - block_size;
+  }
+}
+
+#if ALT_REF_MC_ENABLED
+
+static int temporal_filter_find_matching_mb_c(VP9_COMP *cpi,
+                                              uint8_t *arf_frame_buf,
+                                              uint8_t *frame_ptr_buf,
+                                              int stride,
+                                              int error_thresh) {
+  MACROBLOCK *x = &cpi->mb;
+  MACROBLOCKD* const xd = &x->e_mbd;
+  int step_param;
+  int sadpb = x->sadperbit16;
+  int bestsme = INT_MAX;
+
+  int_mv best_ref_mv1;
+  int_mv best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
+  int_mv *ref_mv;
+
+  // Save input state
+  struct buf_2d src = x->plane[0].src;
+  struct buf_2d pre = xd->plane[0].pre[0];
+
+  best_ref_mv1.as_int = 0;
+  best_ref_mv1_full.as_mv.col = best_ref_mv1.as_mv.col >> 3;
+  best_ref_mv1_full.as_mv.row = best_ref_mv1.as_mv.row >> 3;
+
+  // Setup frame pointers
+  x->plane[0].src.buf = arf_frame_buf;
+  x->plane[0].src.stride = stride;
+  xd->plane[0].pre[0].buf = frame_ptr_buf;
+  xd->plane[0].pre[0].stride = stride;
+
+  // Further step/diamond searches as necessary
+  if (cpi->speed < 8)
+    step_param = cpi->sf.first_step + ((cpi->speed > 5) ? 1 : 0);
+  else
+    step_param = cpi->sf.first_step + 2;
+
+  /*cpi->sf.search_method == HEX*/
+  // TODO Check that the 16x16 vf & sdf are selected here
+  // Ignore mv costing by sending NULL pointer instead of cost arrays
+  ref_mv = &x->e_mbd.mode_info_context->bmi[0].as_mv[0];
+  bestsme = vp9_hex_search(x, &best_ref_mv1_full, ref_mv,
+                           step_param, sadpb, &cpi->fn_ptr[BLOCK_16X16],
+                           NULL, NULL, NULL, NULL,
+                           &best_ref_mv1);
+
+#if ALT_REF_SUBPEL_ENABLED
+  // Try sub-pixel MC?
+  // if (bestsme > error_thresh && bestsme < INT_MAX)
+  {
+    int distortion;
+    unsigned int sse;
+    // Ignore mv costing by sending NULL pointer instead of cost array
+    bestsme = cpi->find_fractional_mv_step(x, ref_mv,
+                                           &best_ref_mv1,
+                                           x->errorperbit,
+                                           &cpi->fn_ptr[BLOCK_16X16],
+                                           NULL, NULL,
+                                           &distortion, &sse);
+  }
+#endif
+
+  // Restore input state
+  x->plane[0].src = src;
+  xd->plane[0].pre[0] = pre;
+
+  return bestsme;
+}
+#endif
+
+static void temporal_filter_iterate_c(VP9_COMP *cpi,
+                                      int frame_count,
+                                      int alt_ref_index,
+                                      int strength) {
+  int byte;
+  int frame;
+  int mb_col, mb_row;
+  unsigned int filter_weight;
+  int mb_cols = cpi->common.mb_cols;
+  int mb_rows = cpi->common.mb_rows;
+  int mb_y_offset = 0;
+  int mb_uv_offset = 0;
+  DECLARE_ALIGNED_ARRAY(16, unsigned int, accumulator, 16 * 16 + 8 * 8 + 8 * 8);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, count, 16 * 16 + 8 * 8 + 8 * 8);
+  MACROBLOCKD *mbd = &cpi->mb.e_mbd;
+  YV12_BUFFER_CONFIG *f = cpi->frames[alt_ref_index];
+  uint8_t *dst1, *dst2;
+  DECLARE_ALIGNED_ARRAY(16, uint8_t,  predictor, 16 * 16 + 8 * 8 + 8 * 8);
+
+  // Save input state
+  uint8_t* input_buffer[MAX_MB_PLANE];
+  int i;
+
+  for (i = 0; i < MAX_MB_PLANE; i++)
+    input_buffer[i] = mbd->plane[i].pre[0].buf;
+
+  for (mb_row = 0; mb_row < mb_rows; mb_row++) {
+#if ALT_REF_MC_ENABLED
+    // Source frames are extended to 16 pixels.  This is different than
+    //  L/A/G reference frames that have a border of 32 (VP9BORDERINPIXELS)
+    // A 6/8 tap filter is used for motion search.  This requires 2 pixels
+    //  before and 3 pixels after.  So the largest Y mv on a border would
+    //  then be 16 - VP9_INTERP_EXTEND. The UV blocks are half the size of the
+    //  Y and therefore only extended by 8.  The largest mv that a UV block
+    //  can support is 8 - VP9_INTERP_EXTEND.  A UV mv is half of a Y mv.
+    //  (16 - VP9_INTERP_EXTEND) >> 1 which is greater than
+    //  8 - VP9_INTERP_EXTEND.
+    // To keep the mv in play for both Y and UV planes the max that it
+    //  can be on a border is therefore 16 - (2*VP9_INTERP_EXTEND+1).
+    cpi->mb.mv_row_min = -((mb_row * 16) + (17 - 2 * VP9_INTERP_EXTEND));
+    cpi->mb.mv_row_max = ((cpi->common.mb_rows - 1 - mb_row) * 16)
+                         + (17 - 2 * VP9_INTERP_EXTEND);
+#endif
+
+    for (mb_col = 0; mb_col < mb_cols; mb_col++) {
+      int i, j, k;
+      int stride;
+
+      vpx_memset(accumulator, 0, 384 * sizeof(unsigned int));
+      vpx_memset(count, 0, 384 * sizeof(uint16_t));
+
+#if ALT_REF_MC_ENABLED
+      cpi->mb.mv_col_min = -((mb_col * 16) + (17 - 2 * VP9_INTERP_EXTEND));
+      cpi->mb.mv_col_max = ((cpi->common.mb_cols - 1 - mb_col) * 16)
+                           + (17 - 2 * VP9_INTERP_EXTEND);
+#endif
+
+      for (frame = 0; frame < frame_count; frame++) {
+        if (cpi->frames[frame] == NULL)
+          continue;
+
+        mbd->mode_info_context->bmi[0].as_mv[0].as_mv.row = 0;
+        mbd->mode_info_context->bmi[0].as_mv[0].as_mv.col = 0;
+
+        if (frame == alt_ref_index) {
+          filter_weight = 2;
+        } else {
+          int err = 0;
+#if ALT_REF_MC_ENABLED
+#define THRESH_LOW   10000
+#define THRESH_HIGH  20000
+
+          // Find best match in this frame by MC
+          err = temporal_filter_find_matching_mb_c
+                (cpi,
+                 cpi->frames[alt_ref_index]->y_buffer + mb_y_offset,
+                 cpi->frames[frame]->y_buffer + mb_y_offset,
+                 cpi->frames[frame]->y_stride,
+                 THRESH_LOW);
+#endif
+          // Assign higher weight to matching MB if it's error
+          // score is lower. If not applying MC default behavior
+          // is to weight all MBs equal.
+          filter_weight = err < THRESH_LOW
+                          ? 2 : err < THRESH_HIGH ? 1 : 0;
+        }
+
+        if (filter_weight != 0) {
+          // Construct the predictors
+          temporal_filter_predictors_mb_c
+          (mbd,
+           cpi->frames[frame]->y_buffer + mb_y_offset,
+           cpi->frames[frame]->u_buffer + mb_uv_offset,
+           cpi->frames[frame]->v_buffer + mb_uv_offset,
+           cpi->frames[frame]->y_stride,
+           mbd->mode_info_context->bmi[0].as_mv[0].as_mv.row,
+           mbd->mode_info_context->bmi[0].as_mv[0].as_mv.col,
+           predictor);
+
+          // Apply the filter (YUV)
+          vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
+                                    predictor, 16, strength, filter_weight,
+                                    accumulator, count);
+
+          vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
+                                    predictor + 256, 8, strength, filter_weight,
+                                    accumulator + 256, count + 256);
+
+          vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
+                                    predictor + 320, 8, strength, filter_weight,
+                                    accumulator + 320, count + 320);
+        }
+      }
+
+      // Normalize filter output to produce AltRef frame
+      dst1 = cpi->alt_ref_buffer.y_buffer;
+      stride = cpi->alt_ref_buffer.y_stride;
+      byte = mb_y_offset;
+      for (i = 0, k = 0; i < 16; i++) {
+        for (j = 0; j < 16; j++, k++) {
+          unsigned int pval = accumulator[k] + (count[k] >> 1);
+          pval *= cpi->fixed_divide[count[k]];
+          pval >>= 19;
+
+          dst1[byte] = (uint8_t)pval;
+
+          // move to next pixel
+          byte++;
+        }
+
+        byte += stride - 16;
+      }
+
+      dst1 = cpi->alt_ref_buffer.u_buffer;
+      dst2 = cpi->alt_ref_buffer.v_buffer;
+      stride = cpi->alt_ref_buffer.uv_stride;
+      byte = mb_uv_offset;
+      for (i = 0, k = 256; i < 8; i++) {
+        for (j = 0; j < 8; j++, k++) {
+          int m = k + 64;
+
+          // U
+          unsigned int pval = accumulator[k] + (count[k] >> 1);
+          pval *= cpi->fixed_divide[count[k]];
+          pval >>= 19;
+          dst1[byte] = (uint8_t)pval;
+
+          // V
+          pval = accumulator[m] + (count[m] >> 1);
+          pval *= cpi->fixed_divide[count[m]];
+          pval >>= 19;
+          dst2[byte] = (uint8_t)pval;
+
+          // move to next pixel
+          byte++;
+        }
+
+        byte += stride - 8;
+      }
+
+      mb_y_offset += 16;
+      mb_uv_offset += 8;
+    }
+
+    mb_y_offset += 16 * (f->y_stride - mb_cols);
+    mb_uv_offset += 8 * (f->uv_stride - mb_cols);
+  }
+
+  // Restore input state
+  for (i = 0; i < MAX_MB_PLANE; i++)
+    mbd->plane[i].pre[0].buf = input_buffer[i];
+}
+
+void vp9_temporal_filter_prepare(VP9_COMP *cpi, int distance) {
+  VP9_COMMON *const cm = &cpi->common;
+
+  int frame = 0;
+
+  int frames_to_blur_backward = 0;
+  int frames_to_blur_forward = 0;
+  int frames_to_blur = 0;
+  int start_frame = 0;
+
+  int strength = cpi->active_arnr_strength;
+  int blur_type = cpi->oxcf.arnr_type;
+  int max_frames = cpi->active_arnr_frames;
+
+  const int num_frames_backward = distance;
+  const int num_frames_forward = vp9_lookahead_depth(cpi->lookahead)
+                               - (num_frames_backward + 1);
+
+  switch (blur_type) {
+    case 1:
+      // Backward Blur
+      frames_to_blur_backward = num_frames_backward;
+
+      if (frames_to_blur_backward >= max_frames)
+        frames_to_blur_backward = max_frames - 1;
+
+      frames_to_blur = frames_to_blur_backward + 1;
+      break;
+
+    case 2:
+      // Forward Blur
+
+      frames_to_blur_forward = num_frames_forward;
+
+      if (frames_to_blur_forward >= max_frames)
+        frames_to_blur_forward = max_frames - 1;
+
+      frames_to_blur = frames_to_blur_forward + 1;
+      break;
+
+    case 3:
+    default:
+      // Center Blur
+      frames_to_blur_forward = num_frames_forward;
+      frames_to_blur_backward = num_frames_backward;
+
+      if (frames_to_blur_forward > frames_to_blur_backward)
+        frames_to_blur_forward = frames_to_blur_backward;
+
+      if (frames_to_blur_backward > frames_to_blur_forward)
+        frames_to_blur_backward = frames_to_blur_forward;
+
+      // When max_frames is even we have 1 more frame backward than forward
+      if (frames_to_blur_forward > (max_frames - 1) / 2)
+        frames_to_blur_forward = ((max_frames - 1) / 2);
+
+      if (frames_to_blur_backward > (max_frames / 2))
+        frames_to_blur_backward = (max_frames / 2);
+
+      frames_to_blur = frames_to_blur_backward + frames_to_blur_forward + 1;
+      break;
+  }
+
+  start_frame = distance + frames_to_blur_forward;
+
+#ifdef DEBUGFWG
+  // DEBUG FWG
+  printf("max:%d FBCK:%d FFWD:%d ftb:%d ftbbck:%d ftbfwd:%d sei:%d lasei:%d start:%d"
+, max_frames
+, num_frames_backward
+, num_frames_forward
+, frames_to_blur
+, frames_to_blur_backward
+, frames_to_blur_forward
+, cpi->source_encode_index
+, cpi->last_alt_ref_sei
+, start_frame);
+#endif
+
+  // Setup scaling factors. Scaling on each of the arnr frames is not supported
+  vp9_setup_scale_factors_for_frame(&cpi->mb.e_mbd.scale_factor[0],
+      cm->yv12_fb[cm->new_fb_idx].y_crop_width,
+      cm->yv12_fb[cm->new_fb_idx].y_crop_height,
+      cm->width, cm->height);
+  cpi->mb.e_mbd.scale_factor_uv[0] = cpi->mb.e_mbd.scale_factor[0];
+
+  // Setup frame pointers, NULL indicates frame not included in filter
+  vpx_memset(cpi->frames, 0, max_frames * sizeof(YV12_BUFFER_CONFIG *));
+  for (frame = 0; frame < frames_to_blur; frame++) {
+    int which_buffer = start_frame - frame;
+    struct lookahead_entry *buf = vp9_lookahead_peek(cpi->lookahead,
+                                                     which_buffer);
+    cpi->frames[frames_to_blur - 1 - frame] = &buf->img;
+  }
+
+  temporal_filter_iterate_c(cpi, frames_to_blur, frames_to_blur_backward,
+                            strength);
+}
+
+void configure_arnr_filter(VP9_COMP *cpi, const unsigned int this_frame,
+                           const int group_boost) {
+  int half_gf_int;
+  int frames_after_arf;
+  int frames_bwd = cpi->oxcf.arnr_max_frames - 1;
+  int frames_fwd = cpi->oxcf.arnr_max_frames - 1;
+  int q;
+
+  // Define the arnr filter width for this group of frames:
+  // We only filter frames that lie within a distance of half
+  // the GF interval from the ARF frame. We also have to trap
+  // cases where the filter extends beyond the end of clip.
+  // Note: this_frame->frame has been updated in the loop
+  // so it now points at the ARF frame.
+  half_gf_int = cpi->baseline_gf_interval >> 1;
+  frames_after_arf = (int)(cpi->twopass.total_stats.count - this_frame - 1);
+
+  switch (cpi->oxcf.arnr_type) {
+    case 1:  // Backward filter
+      frames_fwd = 0;
+      if (frames_bwd > half_gf_int)
+        frames_bwd = half_gf_int;
+      break;
+
+    case 2:  // Forward filter
+      if (frames_fwd > half_gf_int)
+        frames_fwd = half_gf_int;
+      if (frames_fwd > frames_after_arf)
+        frames_fwd = frames_after_arf;
+      frames_bwd = 0;
+      break;
+
+    case 3:  // Centered filter
+    default:
+      frames_fwd >>= 1;
+      if (frames_fwd > frames_after_arf)
+        frames_fwd = frames_after_arf;
+      if (frames_fwd > half_gf_int)
+        frames_fwd = half_gf_int;
+
+      frames_bwd = frames_fwd;
+
+      // For even length filter there is one more frame backward
+      // than forward: e.g. len=6 ==> bbbAff, len=7 ==> bbbAfff.
+      if (frames_bwd < half_gf_int)
+        frames_bwd += (cpi->oxcf.arnr_max_frames + 1) & 0x1;
+      break;
+  }
+
+  cpi->active_arnr_frames = frames_bwd + 1 + frames_fwd;
+
+  // Adjust the strength based on active max q
+  q = ((int)vp9_convert_qindex_to_q(cpi->active_worst_quality) >> 1);
+  if (q > 8) {
+    cpi->active_arnr_strength = cpi->oxcf.arnr_strength;
+  } else {
+    cpi->active_arnr_strength = cpi->oxcf.arnr_strength - (8 - q);
+    if (cpi->active_arnr_strength < 0)
+      cpi->active_arnr_strength = 0;
+  }
+
+  // Adjust number of frames in filter and strength based on gf boost level.
+  if (cpi->active_arnr_frames > (group_boost / 150)) {
+    cpi->active_arnr_frames = (group_boost / 150);
+    cpi->active_arnr_frames += !(cpi->active_arnr_frames & 1);
+  }
+  if (cpi->active_arnr_strength > (group_boost / 300)) {
+    cpi->active_arnr_strength = (group_boost / 300);
+  }
+}
diff --git a/libvpx/vp9/encoder/vp9_temporal_filter.h b/libvpx/vp9/encoder/vp9_temporal_filter.h
new file mode 100644
index 0000000..c5f3b46
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_temporal_filter.h
@@ -0,0 +1,18 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_TEMPORAL_FILTER_H_
+#define VP9_ENCODER_VP9_TEMPORAL_FILTER_H_
+
+void vp9_temporal_filter_prepare(VP9_COMP *cpi, int distance);
+void configure_arnr_filter(VP9_COMP *cpi, const unsigned int this_frame,
+                           const int group_boost);
+
+#endif  // VP9_ENCODER_VP9_TEMPORAL_FILTER_H_
diff --git a/libvpx/vp9/encoder/vp9_tokenize.c b/libvpx/vp9/encoder/vp9_tokenize.c
new file mode 100644
index 0000000..0a290e1
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_tokenize.c
@@ -0,0 +1,458 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <math.h>
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/vp9_tokenize.h"
+#include "vpx_mem/vpx_mem.h"
+
+#include "vp9/common/vp9_pred_common.h"
+#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_entropy.h"
+
+/* Global event counters used for accumulating statistics across several
+   compressions, then generating vp9_context.c = initial stats. */
+
+#ifdef ENTROPY_STATS
+vp9_coeff_accum context_counters[TX_SIZE_MAX_SB][BLOCK_TYPES];
+extern vp9_coeff_stats tree_update_hist[TX_SIZE_MAX_SB][BLOCK_TYPES];
+#endif  /* ENTROPY_STATS */
+
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
+static TOKENVALUE dct_value_tokens[DCT_MAX_VALUE * 2];
+const TOKENVALUE *vp9_dct_value_tokens_ptr;
+static int dct_value_cost[DCT_MAX_VALUE * 2];
+const int *vp9_dct_value_cost_ptr;
+
+static void fill_value_tokens() {
+
+  TOKENVALUE *const t = dct_value_tokens + DCT_MAX_VALUE;
+  vp9_extra_bit *const e = vp9_extra_bits;
+
+  int i = -DCT_MAX_VALUE;
+  int sign = 1;
+
+  do {
+    if (!i)
+      sign = 0;
+
+    {
+      const int a = sign ? -i : i;
+      int eb = sign;
+
+      if (a > 4) {
+        int j = 4;
+
+        while (++j < 11  &&  e[j].base_val <= a) {}
+
+        t[i].token = --j;
+        eb |= (a - e[j].base_val) << 1;
+      } else
+        t[i].token = a;
+
+      t[i].extra = eb;
+    }
+
+    // initialize the cost for extra bits for all possible coefficient value.
+    {
+      int cost = 0;
+      vp9_extra_bit *p = vp9_extra_bits + t[i].token;
+
+      if (p->base_val) {
+        const int extra = t[i].extra;
+        const int length = p->len;
+
+        if (length)
+          cost += treed_cost(p->tree, p->prob, extra >> 1, length);
+
+        cost += vp9_cost_bit(vp9_prob_half, extra & 1); /* sign */
+        dct_value_cost[i + DCT_MAX_VALUE] = cost;
+      }
+
+    }
+
+  } while (++i < DCT_MAX_VALUE);
+
+  vp9_dct_value_tokens_ptr = dct_value_tokens + DCT_MAX_VALUE;
+  vp9_dct_value_cost_ptr   = dct_value_cost + DCT_MAX_VALUE;
+}
+
+extern const int *vp9_get_coef_neighbors_handle(const int *scan, int *pad);
+
+struct tokenize_b_args {
+  VP9_COMP *cpi;
+  MACROBLOCKD *xd;
+  TOKENEXTRA **tp;
+  TX_SIZE tx_size;
+  int dry_run;
+};
+
+static void tokenize_b(int plane, int block, BLOCK_SIZE_TYPE bsize,
+                       int ss_txfrm_size, void *arg) {
+  struct tokenize_b_args* const args = arg;
+  VP9_COMP *cpi = args->cpi;
+  MACROBLOCKD *xd = args->xd;
+  TOKENEXTRA **tp = args->tp;
+  PLANE_TYPE type = plane ? PLANE_TYPE_UV : PLANE_TYPE_Y_WITH_DC;
+  TX_SIZE tx_size = ss_txfrm_size / 2;
+  int dry_run = args->dry_run;
+
+  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
+  int pt; /* near block/prev token context index */
+  int c = 0, rc = 0;
+  TOKENEXTRA *t = *tp;        /* store tokens starting here */
+  const int eob = xd->plane[plane].eobs[block];
+  const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[plane].qcoeff, block, 16);
+  const BLOCK_SIZE_TYPE sb_type = (mbmi->sb_type < BLOCK_SIZE_SB8X8) ?
+                                   BLOCK_SIZE_SB8X8 : mbmi->sb_type;
+  const int bwl = b_width_log2(sb_type);
+  const int off = block >> (2 * tx_size);
+  const int mod = bwl - tx_size - xd->plane[plane].subsampling_x;
+  const int aoff = (off & ((1 << mod) - 1)) << tx_size;
+  const int loff = (off >> mod) << tx_size;
+  ENTROPY_CONTEXT *A = xd->plane[plane].above_context + aoff;
+  ENTROPY_CONTEXT *L = xd->plane[plane].left_context + loff;
+  int seg_eob, default_eob, pad;
+  const int segment_id = mbmi->segment_id;
+  const int *scan, *nb;
+  vp9_coeff_count *counts;
+  vp9_coeff_probs_model *coef_probs;
+  const int ref = mbmi->ref_frame[0] != INTRA_FRAME;
+  ENTROPY_CONTEXT above_ec, left_ec;
+  uint8_t token_cache[1024];
+  TX_TYPE tx_type = DCT_DCT;
+  const uint8_t * band_translate;
+  assert((!type && !plane) || (type && plane));
+
+  counts = cpi->coef_counts[tx_size];
+  coef_probs = cpi->common.fc.coef_probs[tx_size];
+  switch (tx_size) {
+    default:
+    case TX_4X4: {
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_4x4(xd, block) : DCT_DCT;
+      above_ec = A[0] != 0;
+      left_ec = L[0] != 0;
+      seg_eob = 16;
+      scan = get_scan_4x4(tx_type);
+      band_translate = vp9_coefband_trans_4x4;
+      break;
+    }
+    case TX_8X8: {
+      const int sz = 1 + b_width_log2(sb_type);
+      const int x = block & ((1 << sz) - 1), y = block - x;
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
+      above_ec = (A[0] + A[1]) != 0;
+      left_ec = (L[0] + L[1]) != 0;
+      seg_eob = 64;
+      scan = get_scan_8x8(tx_type);
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_16X16: {
+      const int sz = 2 + b_width_log2(sb_type);
+      const int x = block & ((1 << sz) - 1), y = block - x;
+      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
+          get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
+      above_ec = (A[0] + A[1] + A[2] + A[3]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3]) != 0;
+      seg_eob = 256;
+      scan = get_scan_16x16(tx_type);
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+    }
+    case TX_32X32:
+      above_ec = (A[0] + A[1] + A[2] + A[3] + A[4] + A[5] + A[6] + A[7]) != 0;
+      left_ec = (L[0] + L[1] + L[2] + L[3] + L[4] + L[5] + L[6] + L[7]) != 0;
+      seg_eob = 1024;
+      scan = vp9_default_scan_32x32;
+      band_translate = vp9_coefband_trans_8x8plus;
+      break;
+  }
+
+  pt = combine_entropy_contexts(above_ec, left_ec);
+  nb = vp9_get_coef_neighbors_handle(scan, &pad);
+  default_eob = seg_eob;
+
+  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP))
+    seg_eob = 0;
+
+  c = 0;
+  do {
+    const int band = get_coef_band(band_translate, c);
+    int token;
+    int v = 0;
+    rc = scan[c];
+    if (c)
+      pt = vp9_get_coef_context(scan, nb, pad, token_cache, c, default_eob);
+    if (c < eob) {
+      v = qcoeff_ptr[rc];
+      assert(-DCT_MAX_VALUE <= v  &&  v < DCT_MAX_VALUE);
+
+      t->extra = vp9_dct_value_tokens_ptr[v].extra;
+      token    = vp9_dct_value_tokens_ptr[v].token;
+    } else {
+      token = DCT_EOB_TOKEN;
+    }
+
+    t->token = token;
+    t->context_tree = coef_probs[type][ref][band][pt];
+    t->skip_eob_node = (c > 0) && (token_cache[scan[c - 1]] == 0);
+
+#if CONFIG_BALANCED_COEFTREE
+    assert(token <= ZERO_TOKEN ||
+           vp9_coef_encodings[t->token].len - t->skip_eob_node > 0);
+#else
+    assert(vp9_coef_encodings[t->token].len - t->skip_eob_node > 0);
+#endif
+
+    if (!dry_run) {
+      ++counts[type][ref][band][pt][token];
+#if CONFIG_BALANCED_COEFTREE
+      if (!t->skip_eob_node && token > ZERO_TOKEN)
+#else
+      if (!t->skip_eob_node)
+#endif
+        ++cpi->common.fc.eob_branch_counts[tx_size][type][ref][band][pt];
+    }
+    token_cache[scan[c]] = vp9_pt_energy_class[token];
+    ++t;
+  } while (c < eob && ++c < seg_eob);
+
+  *tp = t;
+  if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
+    set_contexts_on_border(xd, bsize, plane, tx_size, c, aoff, loff, A, L);
+  } else {
+    for (pt = 0; pt < (1 << tx_size); pt++) {
+      A[pt] = L[pt] = c > 0;
+    }
+  }
+}
+
+struct is_skippable_args {
+  MACROBLOCKD *xd;
+  int *skippable;
+};
+static void is_skippable(int plane, int block,
+                         BLOCK_SIZE_TYPE bsize, int ss_txfrm_size, void *argv) {
+  struct is_skippable_args *args = argv;
+  args->skippable[0] &= (!args->xd->plane[plane].eobs[block]);
+}
+
+int vp9_sb_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
+  int result = 1;
+  struct is_skippable_args args = {xd, &result};
+  foreach_transformed_block(xd, bsize, is_skippable, &args);
+  return result;
+}
+
+int vp9_sby_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
+  int result = 1;
+  struct is_skippable_args args = {xd, &result};
+  foreach_transformed_block_in_plane(xd, bsize, 0,
+                                     is_skippable, &args);
+  return result;
+}
+
+int vp9_sbuv_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
+  int result = 1;
+  struct is_skippable_args args = {xd, &result};
+  foreach_transformed_block_uv(xd, bsize, is_skippable, &args);
+  return result;
+}
+
+void vp9_tokenize_sb(VP9_COMP *cpi,
+                     MACROBLOCKD *xd,
+                     TOKENEXTRA **t,
+                     int dry_run, BLOCK_SIZE_TYPE bsize) {
+  VP9_COMMON * const cm = &cpi->common;
+  MB_MODE_INFO * const mbmi = &xd->mode_info_context->mbmi;
+  TOKENEXTRA *t_backup = *t;
+  const int mb_skip_context = vp9_get_pred_context(cm, xd, PRED_MBSKIP);
+  const int segment_id = mbmi->segment_id;
+  const int skip_inc = !vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
+  const TX_SIZE txfm_size = mbmi->txfm_size;
+  struct tokenize_b_args arg = {
+    cpi, xd, t, txfm_size, dry_run
+  };
+
+  mbmi->mb_skip_coeff = vp9_sb_is_skippable(xd, bsize);
+
+  if (mbmi->mb_skip_coeff) {
+    if (!dry_run)
+      cm->fc.mbskip_count[mb_skip_context][1] += skip_inc;
+    vp9_reset_sb_tokens_context(xd, bsize);
+    if (dry_run)
+      *t = t_backup;
+    return;
+  }
+
+  if (!dry_run)
+    cm->fc.mbskip_count[mb_skip_context][0] += skip_inc;
+
+  foreach_transformed_block(xd, bsize, tokenize_b, &arg);
+
+  if (dry_run)
+    *t = t_backup;
+}
+
+#ifdef ENTROPY_STATS
+void init_context_counters(void) {
+  FILE *f = fopen("context.bin", "rb");
+  if (!f) {
+    vp9_zero(context_counters);
+  } else {
+    fread(context_counters, sizeof(context_counters), 1, f);
+    fclose(f);
+  }
+
+  f = fopen("treeupdate.bin", "rb");
+  if (!f) {
+    vpx_memset(tree_update_hist, 0, sizeof(tree_update_hist));
+  } else {
+    fread(tree_update_hist, sizeof(tree_update_hist), 1, f);
+    fclose(f);
+  }
+}
+
+static void print_counter(FILE *f, vp9_coeff_accum *context_counters,
+                          int block_types, const char *header) {
+  int type, ref, band, pt, t;
+
+  fprintf(f, "static const vp9_coeff_count %s = {\n", header);
+
+#define Comma(X) (X ? "," : "")
+  type = 0;
+  do {
+    ref = 0;
+    fprintf(f, "%s\n  { /* block Type %d */", Comma(type), type);
+    do {
+      fprintf(f, "%s\n    { /* %s */", Comma(type), ref ? "Inter" : "Intra");
+      band = 0;
+      do {
+        fprintf(f, "%s\n      { /* Coeff Band %d */", Comma(band), band);
+        pt = 0;
+        do {
+          fprintf(f, "%s\n        {", Comma(pt));
+
+          t = 0;
+          do {
+            const int64_t x = context_counters[type][ref][band][pt][t];
+            const int y = (int) x;
+
+            assert(x == (int64_t) y);  /* no overflow handling yet */
+            fprintf(f, "%s %d", Comma(t), y);
+          } while (++t < 1 + MAX_ENTROPY_TOKENS);
+          fprintf(f, "}");
+        } while (++pt < PREV_COEF_CONTEXTS);
+        fprintf(f, "\n      }");
+      } while (++band < COEF_BANDS);
+      fprintf(f, "\n    }");
+    } while (++ref < REF_TYPES);
+    fprintf(f, "\n  }");
+  } while (++type < block_types);
+  fprintf(f, "\n};\n");
+}
+
+static void print_probs(FILE *f, vp9_coeff_accum *context_counters,
+                        int block_types, const char *header) {
+  int type, ref, band, pt, t;
+
+  fprintf(f, "static const vp9_coeff_probs %s = {", header);
+
+  type = 0;
+#define Newline(x, spaces) (x ? " " : "\n" spaces)
+  do {
+    fprintf(f, "%s%s{ /* block Type %d */",
+            Comma(type), Newline(type, "  "), type);
+    ref = 0;
+    do {
+      fprintf(f, "%s%s{ /* %s */",
+              Comma(band), Newline(band, "    "), ref ? "Inter" : "Intra");
+      band = 0;
+      do {
+        fprintf(f, "%s%s{ /* Coeff Band %d */",
+                Comma(band), Newline(band, "      "), band);
+        pt = 0;
+        do {
+          unsigned int branch_ct[ENTROPY_NODES][2];
+          unsigned int coef_counts[MAX_ENTROPY_TOKENS + 1];
+          vp9_prob coef_probs[ENTROPY_NODES];
+
+          if (pt >= 3 && band == 0)
+            break;
+          for (t = 0; t < MAX_ENTROPY_TOKENS + 1; ++t)
+            coef_counts[t] = context_counters[type][ref][band][pt][t];
+          vp9_tree_probs_from_distribution(vp9_coef_tree, coef_probs,
+                                           branch_ct, coef_counts, 0);
+          branch_ct[0][1] = coef_counts[MAX_ENTROPY_TOKENS] - branch_ct[0][0];
+          coef_probs[0] = get_binary_prob(branch_ct[0][0], branch_ct[0][1]);
+          fprintf(f, "%s\n      {", Comma(pt));
+
+          t = 0;
+          do {
+            fprintf(f, "%s %3d", Comma(t), coef_probs[t]);
+          } while (++t < ENTROPY_NODES);
+
+          fprintf(f, " }");
+        } while (++pt < PREV_COEF_CONTEXTS);
+        fprintf(f, "\n      }");
+      } while (++band < COEF_BANDS);
+      fprintf(f, "\n    }");
+    } while (++ref < REF_TYPES);
+    fprintf(f, "\n  }");
+  } while (++type < block_types);
+  fprintf(f, "\n};\n");
+}
+
+void print_context_counters() {
+  FILE *f = fopen("vp9_context.c", "w");
+
+  fprintf(f, "#include \"vp9_entropy.h\"\n");
+  fprintf(f, "\n/* *** GENERATED FILE: DO NOT EDIT *** */\n\n");
+
+  /* print counts */
+  print_counter(f, context_counters[TX_4X4], BLOCK_TYPES,
+                "vp9_default_coef_counts_4x4[BLOCK_TYPES]");
+  print_counter(f, context_counters[TX_8X8], BLOCK_TYPES,
+                "vp9_default_coef_counts_8x8[BLOCK_TYPES]");
+  print_counter(f, context_counters[TX_16X16], BLOCK_TYPES,
+                "vp9_default_coef_counts_16x16[BLOCK_TYPES]");
+  print_counter(f, context_counters[TX_32X32], BLOCK_TYPES,
+                "vp9_default_coef_counts_32x32[BLOCK_TYPES]");
+
+  /* print coefficient probabilities */
+  print_probs(f, context_counters[TX_4X4], BLOCK_TYPES,
+              "default_coef_probs_4x4[BLOCK_TYPES]");
+  print_probs(f, context_counters[TX_8X8], BLOCK_TYPES,
+              "default_coef_probs_8x8[BLOCK_TYPES]");
+  print_probs(f, context_counters[TX_16X16], BLOCK_TYPES,
+              "default_coef_probs_16x16[BLOCK_TYPES]");
+  print_probs(f, context_counters[TX_32X32], BLOCK_TYPES,
+              "default_coef_probs_32x32[BLOCK_TYPES]");
+
+  fclose(f);
+
+  f = fopen("context.bin", "wb");
+  fwrite(context_counters, sizeof(context_counters), 1, f);
+  fclose(f);
+}
+#endif
+
+void vp9_tokenize_initialize() {
+  fill_value_tokens();
+}
diff --git a/libvpx/vp9/encoder/vp9_tokenize.h b/libvpx/vp9/encoder/vp9_tokenize.h
new file mode 100644
index 0000000..e7f90c9
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_tokenize.h
@@ -0,0 +1,56 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_TOKENIZE_H_
+#define VP9_ENCODER_VP9_TOKENIZE_H_
+
+#include "vp9/common/vp9_entropy.h"
+#include "vp9/encoder/vp9_block.h"
+
+void vp9_tokenize_initialize();
+
+typedef struct {
+  int16_t token;
+  int16_t extra;
+} TOKENVALUE;
+
+typedef struct {
+  const vp9_prob *context_tree;
+  int16_t         extra;
+  uint8_t         token;
+  uint8_t         skip_eob_node;
+} TOKENEXTRA;
+
+typedef int64_t vp9_coeff_accum[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS]
+                               [MAX_ENTROPY_TOKENS + 1];
+
+int vp9_sb_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize);
+int vp9_sby_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize);
+int vp9_sbuv_is_skippable(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize);
+struct VP9_COMP;
+
+void vp9_tokenize_sb(struct VP9_COMP *cpi, MACROBLOCKD *xd,
+                     TOKENEXTRA **t, int dry_run, BLOCK_SIZE_TYPE bsize);
+
+#ifdef ENTROPY_STATS
+void init_context_counters();
+void print_context_counters();
+
+extern vp9_coeff_accum context_counters[TX_SIZE_MAX_SB][BLOCK_TYPES];
+#endif
+
+extern const int *vp9_dct_value_cost_ptr;
+/* TODO: The Token field should be broken out into a separate char array to
+ *  improve cache locality, since it's needed for costing when the rest of the
+ *  fields are not.
+ */
+extern const TOKENVALUE *vp9_dct_value_tokens_ptr;
+
+#endif  // VP9_ENCODER_VP9_TOKENIZE_H_
diff --git a/libvpx/vp9/encoder/vp9_treewriter.c b/libvpx/vp9/encoder/vp9_treewriter.c
new file mode 100644
index 0000000..e4aed53
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_treewriter.c
@@ -0,0 +1,38 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/encoder/vp9_treewriter.h"
+
+static void cost(int *costs, vp9_tree tree, const vp9_prob *probs,
+                 int i, int c) {
+  const vp9_prob prob = probs[i / 2];
+  int b;
+
+  for (b = 0; b <= 1; ++b) {
+    const int cc = c + vp9_cost_bit(prob, b);
+    const vp9_tree_index ii = tree[i + b];
+
+    if (ii <= 0)
+      costs[-ii] = cc;
+    else
+      cost(costs, tree, probs, ii, cc);
+  }
+}
+
+void vp9_cost_tokens(int *costs, const vp9_prob *probs, vp9_tree tree) {
+  cost(costs, tree, probs, 0, 0);
+}
+
+void vp9_cost_tokens_skip(int *costs, const vp9_prob *probs, vp9_tree tree) {
+  assert(tree[0] <= 0 && tree[1] > 0);
+
+  costs[-tree[0]] = vp9_cost_bit(probs[0], 0);
+  cost(costs, tree, probs, 2, 0);
+}
diff --git a/libvpx/vp9/encoder/vp9_treewriter.h b/libvpx/vp9/encoder/vp9_treewriter.h
new file mode 100644
index 0000000..eeda5cd
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_treewriter.h
@@ -0,0 +1,87 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_VP9_TREEWRITER_H_
+#define VP9_ENCODER_VP9_TREEWRITER_H_
+
+/* Trees map alphabets into huffman-like codes suitable for an arithmetic
+   bit coder.  Timothy S Murphy  11 October 2004 */
+
+#include "vp9/common/vp9_treecoder.h"
+
+#include "vp9/encoder/vp9_boolhuff.h"       /* for now */
+
+
+#define vp9_write_prob(w, v) vp9_write_literal((w), (v), 8)
+
+/* Approximate length of an encoded bool in 256ths of a bit at given prob */
+
+#define vp9_cost_zero(x) (vp9_prob_cost[x])
+#define vp9_cost_one(x) vp9_cost_zero(vp9_complement(x))
+
+#define vp9_cost_bit(x, b) vp9_cost_zero((b) ? vp9_complement(x) : (x))
+
+/* VP8BC version is scaled by 2^20 rather than 2^8; see bool_coder.h */
+
+
+/* Both of these return bits, not scaled bits. */
+static INLINE unsigned int cost_branch256(const unsigned int ct[2],
+                                          vp9_prob p) {
+  return ct[0] * vp9_cost_zero(p) + ct[1] * vp9_cost_one(p);
+}
+
+static INLINE unsigned int cost_branch(const unsigned int ct[2],
+                                       vp9_prob p) {
+  return cost_branch256(ct, p) >> 8;
+}
+
+
+static INLINE void treed_write(vp9_writer *w,
+                               vp9_tree tree, const vp9_prob *probs,
+                               int bits, int len) {
+  vp9_tree_index i = 0;
+
+  do {
+    const int bit = (bits >> --len) & 1;
+    vp9_write(w, bit, probs[i >> 1]);
+    i = tree[i + bit];
+  } while (len);
+}
+
+static INLINE void write_token(vp9_writer *w, vp9_tree tree,
+                               const vp9_prob *probs,
+                               const struct vp9_token *token) {
+  treed_write(w, tree, probs, token->value, token->len);
+}
+
+static INLINE int treed_cost(vp9_tree tree, const vp9_prob *probs,
+                             int bits, int len) {
+  int cost = 0;
+  vp9_tree_index i = 0;
+
+  do {
+    const int bit = (bits >> --len) & 1;
+    cost += vp9_cost_bit(probs[i >> 1], bit);
+    i = tree[i + bit];
+  } while (len);
+
+  return cost;
+}
+
+static INLINE int cost_token(vp9_tree tree, const vp9_prob *probs,
+                             const struct vp9_token *token) {
+  return treed_cost(tree, probs, token->value, token->len);
+}
+
+void vp9_cost_tokens(int *costs, const vp9_prob *probs, vp9_tree tree);
+void vp9_cost_tokens_skip(int *costs, const vp9_prob *probs, vp9_tree tree);
+
+#endif  // VP9_ENCODER_VP9_TREEWRITER_H_
diff --git a/libvpx/vp9/encoder/vp9_variance.h b/libvpx/vp9/encoder/vp9_variance.h
new file mode 100644
index 0000000..38808d7
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_variance.h
@@ -0,0 +1,104 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_VP9_VARIANCE_H_
+#define VP9_ENCODER_VP9_VARIANCE_H_
+
+#include "vpx/vpx_integer.h"
+// #include "./vpx_config.h"
+
+typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
+                                    int source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int ref_stride,
+                                    unsigned int max_sad);
+
+typedef void (*vp9_sad_multi_fn_t)(const uint8_t *src_ptr,
+                                   int source_stride,
+                                   const uint8_t *ref_ptr,
+                                   int  ref_stride,
+                                   unsigned int *sad_array);
+
+typedef void (*vp9_sad_multi1_fn_t)(const uint8_t *src_ptr,
+                                    int source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  ref_stride,
+                                    unsigned int *sad_array);
+
+typedef void (*vp9_sad_multi_d_fn_t)(const uint8_t *src_ptr,
+                                     int source_stride,
+                                     const uint8_t* const ref_ptr[],
+                                     int  ref_stride, unsigned int *sad_array);
+
+typedef unsigned int (*vp9_variance_fn_t)(const uint8_t *src_ptr,
+                                          int source_stride,
+                                          const uint8_t *ref_ptr,
+                                          int ref_stride,
+                                          unsigned int *sse);
+
+typedef unsigned int (*vp9_subpixvariance_fn_t)(const uint8_t *src_ptr,
+                                                int source_stride,
+                                                int xoffset,
+                                                int yoffset,
+                                                const uint8_t *ref_ptr,
+                                                int Refstride,
+                                                unsigned int *sse);
+
+typedef unsigned int (*vp9_subp_avg_variance_fn_t)(const uint8_t *src_ptr,
+                                                   int source_stride,
+                                                   int xoffset,
+                                                   int yoffset,
+                                                   const uint8_t *ref_ptr,
+                                                   int Refstride,
+                                                   unsigned int *sse,
+                                                   const uint8_t *second_pred);
+
+typedef void (*vp9_ssimpf_fn_t)(uint8_t *s, int sp, uint8_t *r,
+                                int rp, unsigned long *sum_s,
+                                unsigned long *sum_r, unsigned long *sum_sq_s,
+                                unsigned long *sum_sq_r,
+                                unsigned long *sum_sxr);
+
+typedef unsigned int (*vp9_getmbss_fn_t)(const short *);
+
+typedef unsigned int (*vp9_get16x16prederror_fn_t)(const uint8_t *src_ptr,
+                                                   int source_stride,
+                                                   const uint8_t *ref_ptr,
+                                                   int  ref_stride);
+
+typedef struct vp9_variance_vtable {
+    vp9_sad_fn_t               sdf;
+    vp9_variance_fn_t          vf;
+    vp9_subpixvariance_fn_t    svf;
+    vp9_subp_avg_variance_fn_t svaf;
+    vp9_variance_fn_t          svf_halfpix_h;
+    vp9_variance_fn_t          svf_halfpix_v;
+    vp9_variance_fn_t          svf_halfpix_hv;
+    vp9_sad_multi_fn_t         sdx3f;
+    vp9_sad_multi1_fn_t        sdx8f;
+    vp9_sad_multi_d_fn_t       sdx4df;
+} vp9_variance_fn_ptr_t;
+
+static void comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,
+                          int height, uint8_t *ref, int ref_stride) {
+  int i, j;
+
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j++) {
+      int tmp;
+      tmp = pred[j] + ref[j];
+      comp_pred[j] = (tmp + 1) >> 1;
+    }
+    comp_pred += width;
+    pred += width;
+    ref += ref_stride;
+  }
+}
+#endif  // VP9_ENCODER_VP9_VARIANCE_H_
diff --git a/libvpx/vp9/encoder/vp9_variance_c.c b/libvpx/vp9/encoder/vp9_variance_c.c
new file mode 100644
index 0000000..23e7767
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_variance_c.c
@@ -0,0 +1,957 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/common/vp9_filter.h"
+#include "vp9/common/vp9_subpelvar.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "./vp9_rtcd.h"
+
+unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
+  unsigned int i, sum = 0;
+
+  for (i = 0; i < 256; i++) {
+    sum += (src_ptr[i] * src_ptr[i]);
+  }
+
+  return sum;
+}
+
+unsigned int vp9_variance64x32_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 64, 32, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 11));
+}
+
+unsigned int vp9_sub_pixel_variance64x32_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 64, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 32, 64, vfilter);
+
+  return vp9_variance64x32(temp2, 64, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance64x32_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 64, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 32, 64, vfilter);
+  comp_avg_pred(temp3, second_pred, 64, 32, temp2, 64);
+  return vp9_variance64x32(temp3, 64, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_variance32x64_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 64, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 11));
+}
+
+unsigned int vp9_sub_pixel_variance32x64_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 65, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 64, 32, vfilter);
+
+  return vp9_variance32x64(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance32x64_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 64);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 65, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 64, 32, vfilter);
+  comp_avg_pred(temp3, second_pred, 32, 64, temp2, 32);
+  return vp9_variance32x64(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_variance32x16_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 16, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 9));
+}
+
+unsigned int vp9_sub_pixel_variance32x16_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 16, 32, vfilter);
+
+  return vp9_variance32x16(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance32x16_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 16);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 16, 32, vfilter);
+  comp_avg_pred(temp3, second_pred, 32, 16, temp2, 32);
+  return vp9_variance32x16(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_variance16x32_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 32, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 9));
+}
+
+unsigned int vp9_sub_pixel_variance16x32_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 32, 16, vfilter);
+
+  return vp9_variance16x32(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance16x32_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 32);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 32, 16, vfilter);
+  comp_avg_pred(temp3, second_pred, 16, 32, temp2, 16);
+  return vp9_variance16x32(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_variance64x64_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 64, 64, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 12));
+}
+
+unsigned int vp9_variance32x32_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 32, 32, &var, &avg);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 10));
+}
+
+unsigned int vp9_variance16x16_c(const uint8_t *src_ptr,
+                                 int  source_stride,
+                                 const uint8_t *ref_ptr,
+                                 int  recon_stride,
+                                 unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 8));
+}
+
+unsigned int vp9_variance8x16_c(const uint8_t *src_ptr,
+                                int  source_stride,
+                                const uint8_t *ref_ptr,
+                                int  recon_stride,
+                                unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 7));
+}
+
+unsigned int vp9_variance16x8_c(const uint8_t *src_ptr,
+                                int  source_stride,
+                                const uint8_t *ref_ptr,
+                                int  recon_stride,
+                                unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 7));
+}
+
+void vp9_get_sse_sum_8x8_c(const uint8_t *src_ptr, int source_stride,
+                       const uint8_t *ref_ptr, int ref_stride,
+                       unsigned int *sse, int *sum) {
+  variance(src_ptr, source_stride, ref_ptr, ref_stride, 8, 8, sse, sum);
+}
+
+unsigned int vp9_variance8x8_c(const uint8_t *src_ptr,
+                               int  source_stride,
+                               const uint8_t *ref_ptr,
+                               int  recon_stride,
+                               unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 6));
+}
+
+unsigned int vp9_variance8x4_c(const uint8_t *src_ptr,
+                               int  source_stride,
+                               const uint8_t *ref_ptr,
+                               int  recon_stride,
+                               unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 4, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 5));
+}
+
+unsigned int vp9_variance4x8_c(const uint8_t *src_ptr,
+                               int  source_stride,
+                               const uint8_t *ref_ptr,
+                               int  recon_stride,
+                               unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 8, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 5));
+}
+
+unsigned int vp9_variance4x4_c(const uint8_t *src_ptr,
+                               int  source_stride,
+                               const uint8_t *ref_ptr,
+                               int  recon_stride,
+                               unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 4));
+}
+
+
+unsigned int vp9_mse16x16_c(const uint8_t *src_ptr,
+                            int  source_stride,
+                            const uint8_t *ref_ptr,
+                            int  recon_stride,
+                            unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, &var, &avg);
+  *sse = var;
+  return var;
+}
+
+unsigned int vp9_mse16x8_c(const uint8_t *src_ptr,
+                           int  source_stride,
+                           const uint8_t *ref_ptr,
+                           int  recon_stride,
+                           unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8, &var, &avg);
+  *sse = var;
+  return var;
+}
+
+unsigned int vp9_mse8x16_c(const uint8_t *src_ptr,
+                           int  source_stride,
+                           const uint8_t *ref_ptr,
+                           int  recon_stride,
+                           unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16, &var, &avg);
+  *sse = var;
+  return var;
+}
+
+unsigned int vp9_mse8x8_c(const uint8_t *src_ptr,
+                          int  source_stride,
+                          const uint8_t *ref_ptr,
+                          int  recon_stride,
+                          unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, &var, &avg);
+  *sse = var;
+  return var;
+}
+
+
+unsigned int vp9_sub_pixel_variance4x4_c(const uint8_t *src_ptr,
+                                         int  src_pixels_per_line,
+                                         int  xoffset,
+                                         int  yoffset,
+                                         const uint8_t *dst_ptr,
+                                         int dst_pixels_per_line,
+                                         unsigned int *sse) {
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+  uint16_t fdata3[5 * 4];  // Temp data bufffer used in filtering
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  // First filter 1d Horizontal
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 5, 4, hfilter);
+
+  // Now filter Verticaly
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 4,  4,  4,  4, vfilter);
+
+  return vp9_variance4x4(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance4x4_c(const uint8_t *src_ptr,
+                                             int  src_pixels_per_line,
+                                             int  xoffset,
+                                             int  yoffset,
+                                             const uint8_t *dst_ptr,
+                                             int dst_pixels_per_line,
+                                             unsigned int *sse,
+                                             const uint8_t *second_pred) {
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 4);  // compound pred buffer
+  uint16_t fdata3[5 * 4];  // Temp data bufffer used in filtering
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  // First filter 1d Horizontal
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 5, 4, hfilter);
+
+  // Now filter Verticaly
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 4,  4,  4,  4, vfilter);
+  comp_avg_pred(temp3, second_pred, 4, 4, temp2, 4);
+  return vp9_variance4x4(temp3, 4, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance8x8_c(const uint8_t *src_ptr,
+                                         int  src_pixels_per_line,
+                                         int  xoffset,
+                                         int  yoffset,
+                                         const uint8_t *dst_ptr,
+                                         int dst_pixels_per_line,
+                                         unsigned int *sse) {
+  uint16_t fdata3[9 * 8];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
+
+  return vp9_variance8x8(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance8x8_c(const uint8_t *src_ptr,
+                                             int  src_pixels_per_line,
+                                             int  xoffset,
+                                             int  yoffset,
+                                             const uint8_t *dst_ptr,
+                                             int dst_pixels_per_line,
+                                             unsigned int *sse,
+                                             const uint8_t *second_pred) {
+  uint16_t fdata3[9 * 8];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 8);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
+  comp_avg_pred(temp3, second_pred, 8, 8, temp2, 8);
+  return vp9_variance8x8(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance16x16_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[17 * 16];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 16, 16, vfilter);
+
+  return vp9_variance16x16(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance16x16_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[17 * 16];
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 16);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 16, 16, vfilter);
+
+  comp_avg_pred(temp3, second_pred, 16, 16, temp2, 16);
+  return vp9_variance16x16(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance64x64_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 65, 64, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 64, 64, vfilter);
+
+  return vp9_variance64x64(temp2, 64, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance64x64_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[65 * 64];  // Temp data bufffer used in filtering
+  uint8_t temp2[68 * 64];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 65, 64, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 64, 64, 64, 64, vfilter);
+  comp_avg_pred(temp3, second_pred, 64, 64, temp2, 64);
+  return vp9_variance64x64(temp3, 64, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance32x32_c(const uint8_t *src_ptr,
+                                           int  src_pixels_per_line,
+                                           int  xoffset,
+                                           int  yoffset,
+                                           const uint8_t *dst_ptr,
+                                           int dst_pixels_per_line,
+                                           unsigned int *sse) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 32, 32, vfilter);
+
+  return vp9_variance32x32(temp2, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance32x32_c(const uint8_t *src_ptr,
+                                               int  src_pixels_per_line,
+                                               int  xoffset,
+                                               int  yoffset,
+                                               const uint8_t *dst_ptr,
+                                               int dst_pixels_per_line,
+                                               unsigned int *sse,
+                                               const uint8_t *second_pred) {
+  uint16_t fdata3[33 * 32];  // Temp data bufffer used in filtering
+  uint8_t temp2[36 * 32];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 32);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 33, 32, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 32, 32, 32, 32, vfilter);
+  comp_avg_pred(temp3, second_pred, 32, 32, temp2, 32);
+  return vp9_variance32x32(temp3, 32, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_variance_halfpixvar16x16_h_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 8, 0,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar32x32_h_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 8, 0,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar64x64_h_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 8, 0,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar16x16_v_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 0, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar32x32_v_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 0, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar64x64_v_c(const uint8_t *src_ptr,
+                                              int  source_stride,
+                                              const uint8_t *ref_ptr,
+                                              int  recon_stride,
+                                              unsigned int *sse) {
+  return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 0, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar16x16_hv_c(const uint8_t *src_ptr,
+                                               int  source_stride,
+                                               const uint8_t *ref_ptr,
+                                               int  recon_stride,
+                                               unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_c(src_ptr, source_stride, 8, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar32x32_hv_c(const uint8_t *src_ptr,
+                                               int  source_stride,
+                                               const uint8_t *ref_ptr,
+                                               int  recon_stride,
+                                               unsigned int *sse) {
+  return vp9_sub_pixel_variance32x32_c(src_ptr, source_stride, 8, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_variance_halfpixvar64x64_hv_c(const uint8_t *src_ptr,
+                                               int  source_stride,
+                                               const uint8_t *ref_ptr,
+                                               int  recon_stride,
+                                               unsigned int *sse) {
+  return vp9_sub_pixel_variance64x64_c(src_ptr, source_stride, 8, 8,
+                                       ref_ptr, recon_stride, sse);
+}
+
+unsigned int vp9_sub_pixel_mse16x16_c(const uint8_t *src_ptr,
+                                      int  src_pixels_per_line,
+                                      int  xoffset,
+                                      int  yoffset,
+                                      const uint8_t *dst_ptr,
+                                      int dst_pixels_per_line,
+                                      unsigned int *sse) {
+  vp9_sub_pixel_variance16x16_c(src_ptr, src_pixels_per_line,
+                                xoffset, yoffset, dst_ptr,
+                                dst_pixels_per_line, sse);
+  return *sse;
+}
+
+unsigned int vp9_sub_pixel_mse32x32_c(const uint8_t *src_ptr,
+                                      int  src_pixels_per_line,
+                                      int  xoffset,
+                                      int  yoffset,
+                                      const uint8_t *dst_ptr,
+                                      int dst_pixels_per_line,
+                                      unsigned int *sse) {
+  vp9_sub_pixel_variance32x32_c(src_ptr, src_pixels_per_line,
+                                xoffset, yoffset, dst_ptr,
+                                dst_pixels_per_line, sse);
+  return *sse;
+}
+
+unsigned int vp9_sub_pixel_mse64x64_c(const uint8_t *src_ptr,
+                                      int  src_pixels_per_line,
+                                      int  xoffset,
+                                      int  yoffset,
+                                      const uint8_t *dst_ptr,
+                                      int dst_pixels_per_line,
+                                      unsigned int *sse) {
+  vp9_sub_pixel_variance64x64_c(src_ptr, src_pixels_per_line,
+                                xoffset, yoffset, dst_ptr,
+                                dst_pixels_per_line, sse);
+  return *sse;
+}
+
+unsigned int vp9_sub_pixel_variance16x8_c(const uint8_t *src_ptr,
+                                          int  src_pixels_per_line,
+                                          int  xoffset,
+                                          int  yoffset,
+                                          const uint8_t *dst_ptr,
+                                          int dst_pixels_per_line,
+                                          unsigned int *sse) {
+  uint16_t fdata3[16 * 9];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 8, 16, vfilter);
+
+  return vp9_variance16x8(temp2, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance16x8_c(const uint8_t *src_ptr,
+                                              int  src_pixels_per_line,
+                                              int  xoffset,
+                                              int  yoffset,
+                                              const uint8_t *dst_ptr,
+                                              int dst_pixels_per_line,
+                                              unsigned int *sse,
+                                              const uint8_t *second_pred) {
+  uint16_t fdata3[16 * 9];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 8);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 16, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 16, 16, 8, 16, vfilter);
+  comp_avg_pred(temp3, second_pred, 16, 8, temp2, 16);
+  return vp9_variance16x8(temp3, 16, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance8x16_c(const uint8_t *src_ptr,
+                                          int  src_pixels_per_line,
+                                          int  xoffset,
+                                          int  yoffset,
+                                          const uint8_t *dst_ptr,
+                                          int dst_pixels_per_line,
+                                          unsigned int *sse) {
+  uint16_t fdata3[9 * 16];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 16, 8, vfilter);
+
+  return vp9_variance8x16(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance8x16_c(const uint8_t *src_ptr,
+                                              int  src_pixels_per_line,
+                                              int  xoffset,
+                                              int  yoffset,
+                                              const uint8_t *dst_ptr,
+                                              int dst_pixels_per_line,
+                                              unsigned int *sse,
+                                              const uint8_t *second_pred) {
+  uint16_t fdata3[9 * 16];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 16);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 17, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 16, 8, vfilter);
+  comp_avg_pred(temp3, second_pred, 8, 16, temp2, 8);
+  return vp9_variance8x16(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance8x4_c(const uint8_t *src_ptr,
+                                         int  src_pixels_per_line,
+                                         int  xoffset,
+                                         int  yoffset,
+                                         const uint8_t *dst_ptr,
+                                         int dst_pixels_per_line,
+                                         unsigned int *sse) {
+  uint16_t fdata3[8 * 5];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 5, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 4, 8, vfilter);
+
+  return vp9_variance8x4(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance8x4_c(const uint8_t *src_ptr,
+                                             int  src_pixels_per_line,
+                                             int  xoffset,
+                                             int  yoffset,
+                                             const uint8_t *dst_ptr,
+                                             int dst_pixels_per_line,
+                                             unsigned int *sse,
+                                             const uint8_t *second_pred) {
+  uint16_t fdata3[8 * 5];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 4);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 5, 8, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 4, 8, vfilter);
+  comp_avg_pred(temp3, second_pred, 8, 4, temp2, 8);
+  return vp9_variance8x4(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_variance4x8_c(const uint8_t *src_ptr,
+                                         int  src_pixels_per_line,
+                                         int  xoffset,
+                                         int  yoffset,
+                                         const uint8_t *dst_ptr,
+                                         int dst_pixels_per_line,
+                                         unsigned int *sse) {
+  uint16_t fdata3[5 * 8];  // Temp data bufffer used in filtering
+  // FIXME(jingning,rbultje): this temp2 buffer probably doesn't need to be
+  // of this big? same issue appears in all other block size settings.
+  uint8_t temp2[20 * 16];
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 4, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 8, 4, vfilter);
+
+  return vp9_variance4x8(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
+}
+
+unsigned int vp9_sub_pixel_avg_variance4x8_c(const uint8_t *src_ptr,
+                                             int  src_pixels_per_line,
+                                             int  xoffset,
+                                             int  yoffset,
+                                             const uint8_t *dst_ptr,
+                                             int dst_pixels_per_line,
+                                             unsigned int *sse,
+                                             const uint8_t *second_pred) {
+  uint16_t fdata3[5 * 8];  // Temp data bufffer used in filtering
+  uint8_t temp2[20 * 16];
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 8);  // compound pred buffer
+  const int16_t *hfilter, *vfilter;
+
+  hfilter = VP9_BILINEAR_FILTERS_2TAP(xoffset);
+  vfilter = VP9_BILINEAR_FILTERS_2TAP(yoffset);
+
+  var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
+                                    1, 9, 4, hfilter);
+  var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 8, 4, vfilter);
+  comp_avg_pred(temp3, second_pred, 4, 8, temp2, 4);
+  return vp9_variance4x8(temp3, 4, dst_ptr, dst_pixels_per_line, sse);
+}
diff --git a/libvpx/vp9/encoder/vp9_write_bit_buffer.h b/libvpx/vp9/encoder/vp9_write_bit_buffer.h
new file mode 100644
index 0000000..6f91cfc
--- /dev/null
+++ b/libvpx/vp9/encoder/vp9_write_bit_buffer.h
@@ -0,0 +1,48 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_BIT_WRITE_BUFFER_H_
+#define VP9_BIT_WRITE_BUFFER_H_
+
+#include <limits.h>
+
+#include "vpx/vpx_integer.h"
+
+struct vp9_write_bit_buffer {
+  uint8_t *bit_buffer;
+  size_t bit_offset;
+};
+
+static size_t vp9_rb_bytes_written(struct vp9_write_bit_buffer *wb) {
+  return wb->bit_offset / CHAR_BIT + (wb->bit_offset % CHAR_BIT > 0);
+}
+
+static void vp9_wb_write_bit(struct vp9_write_bit_buffer *wb, int bit) {
+  const int off = wb->bit_offset;
+  const int p = off / CHAR_BIT;
+  const int q = CHAR_BIT - 1 - off % CHAR_BIT;
+  if (q == CHAR_BIT -1) {
+    wb->bit_buffer[p] = bit << q;
+  } else {
+    wb->bit_buffer[p] &= ~(1 << q);
+    wb->bit_buffer[p] |= bit << q;
+  }
+  wb->bit_offset = off + 1;
+}
+
+static void vp9_wb_write_literal(struct vp9_write_bit_buffer *wb,
+                              int data, int bits) {
+  int bit;
+  for (bit = bits - 1; bit >= 0; bit--)
+    vp9_wb_write_bit(wb, (data >> bit) & 1);
+}
+
+
+#endif  // VP9_BIT_WRITE_BUFFER_H_
diff --git a/libvpx/vp9/encoder/x86/vp9_dct_mmx.asm b/libvpx/vp9/encoder/x86/vp9_dct_mmx.asm
new file mode 100644
index 0000000..54766d8
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_dct_mmx.asm
@@ -0,0 +1,241 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_short_fdct4x4_mmx(short *input, short *output, int pitch)
+global sym(vp9_short_fdct4x4_mmx) PRIVATE
+sym(vp9_short_fdct4x4_mmx):
+    push        rbp
+    mov         rbp,        rsp
+    SHADOW_ARGS_TO_STACK 3
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+        mov         rsi,        arg(0)      ; input
+        mov         rdi,        arg(1)      ; output
+
+        movsxd      rax,        dword ptr arg(2) ;pitch
+
+        lea         rcx,        [rsi + rax*2]
+        ; read the input data
+        movq        mm0,        [rsi]
+        movq        mm1,        [rsi + rax]
+
+        movq        mm2,        [rcx]
+        movq        mm4,        [rcx + rax]
+
+        ; transpose for the first stage
+        movq        mm3,        mm0         ; 00 01 02 03
+        movq        mm5,        mm2         ; 20 21 22 23
+
+        punpcklwd   mm0,        mm1         ; 00 10 01 11
+        punpckhwd   mm3,        mm1         ; 02 12 03 13
+
+        punpcklwd   mm2,        mm4         ; 20 30 21 31
+        punpckhwd   mm5,        mm4         ; 22 32 23 33
+
+        movq        mm1,        mm0         ; 00 10 01 11
+        punpckldq   mm0,        mm2         ; 00 10 20 30
+
+        punpckhdq   mm1,        mm2         ; 01 11 21 31
+
+        movq        mm2,        mm3         ; 02 12 03 13
+        punpckldq   mm2,        mm5         ; 02 12 22 32
+
+        punpckhdq   mm3,        mm5         ; 03 13 23 33
+
+        ; mm0 0
+        ; mm1 1
+        ; mm2 2
+        ; mm3 3
+
+        ; first stage
+        movq        mm5,        mm0
+        movq        mm4,        mm1
+
+        paddw       mm0,        mm3         ; a1 = 0 + 3
+        paddw       mm1,        mm2         ; b1 = 1 + 2
+
+        psubw       mm4,        mm2         ; c1 = 1 - 2
+        psubw       mm5,        mm3         ; d1 = 0 - 3
+
+        psllw       mm5,        3
+        psllw       mm4,        3
+
+        psllw       mm0,        3
+        psllw       mm1,        3
+
+        ; output 0 and 2
+        movq        mm2,        mm0         ; a1
+
+        paddw       mm0,        mm1         ; op[0] = a1 + b1
+        psubw       mm2,        mm1         ; op[2] = a1 - b1
+
+        ; output 1 and 3
+        ; interleave c1, d1
+        movq        mm1,        mm5         ; d1
+        punpcklwd   mm1,        mm4         ; c1 d1
+        punpckhwd   mm5,        mm4         ; c1 d1
+
+        movq        mm3,        mm1
+        movq        mm4,        mm5
+
+        pmaddwd     mm1,        MMWORD PTR[GLOBAL (_5352_2217)]    ; c1*2217 + d1*5352
+        pmaddwd     mm4,        MMWORD PTR[GLOBAL (_5352_2217)]    ; c1*2217 + d1*5352
+
+        pmaddwd     mm3,        MMWORD PTR[GLOBAL(_2217_neg5352)]  ; d1*2217 - c1*5352
+        pmaddwd     mm5,        MMWORD PTR[GLOBAL(_2217_neg5352)]  ; d1*2217 - c1*5352
+
+        paddd       mm1,        MMWORD PTR[GLOBAL(_14500)]
+        paddd       mm4,        MMWORD PTR[GLOBAL(_14500)]
+        paddd       mm3,        MMWORD PTR[GLOBAL(_7500)]
+        paddd       mm5,        MMWORD PTR[GLOBAL(_7500)]
+
+        psrad       mm1,        12          ; (c1 * 2217 + d1 * 5352 +  14500)>>12
+        psrad       mm4,        12          ; (c1 * 2217 + d1 * 5352 +  14500)>>12
+        psrad       mm3,        12          ; (d1 * 2217 - c1 * 5352 +   7500)>>12
+        psrad       mm5,        12          ; (d1 * 2217 - c1 * 5352 +   7500)>>12
+
+        packssdw    mm1,        mm4         ; op[1]
+        packssdw    mm3,        mm5         ; op[3]
+
+        ; done with vertical
+        ; transpose for the second stage
+        movq        mm4,        mm0         ; 00 10 20 30
+        movq        mm5,        mm2         ; 02 12 22 32
+
+        punpcklwd   mm0,        mm1         ; 00 01 10 11
+        punpckhwd   mm4,        mm1         ; 20 21 30 31
+
+        punpcklwd   mm2,        mm3         ; 02 03 12 13
+        punpckhwd   mm5,        mm3         ; 22 23 32 33
+
+        movq        mm1,        mm0         ; 00 01 10 11
+        punpckldq   mm0,        mm2         ; 00 01 02 03
+
+        punpckhdq   mm1,        mm2         ; 01 22 12 13
+
+        movq        mm2,        mm4         ; 20 31 30 31
+        punpckldq   mm2,        mm5         ; 20 21 22 23
+
+        punpckhdq   mm4,        mm5         ; 30 31 32 33
+
+        ; mm0 0
+        ; mm1 1
+        ; mm2 2
+        ; mm3 4
+
+        movq        mm5,        mm0
+        movq        mm3,        mm1
+
+        paddw       mm0,        mm4         ; a1 = 0 + 3
+        paddw       mm1,        mm2         ; b1 = 1 + 2
+
+        psubw       mm3,        mm2         ; c1 = 1 - 2
+        psubw       mm5,        mm4         ; d1 = 0 - 3
+
+        pxor        mm6,        mm6         ; zero out for compare
+
+        pcmpeqw     mm6,        mm5         ; d1 != 0
+
+        pandn       mm6,        MMWORD PTR[GLOBAL(_cmp_mask)]   ; clear upper,
+                                                                ; and keep bit 0 of lower
+
+        ; output 0 and 2
+        movq        mm2,        mm0         ; a1
+
+        paddw       mm0,        mm1         ; a1 + b1
+        psubw       mm2,        mm1         ; a1 - b1
+
+        paddw       mm0,        MMWORD PTR[GLOBAL(_7w)]
+        paddw       mm2,        MMWORD PTR[GLOBAL(_7w)]
+
+        psraw       mm0,        4           ; op[0] = (a1 + b1 + 7)>>4
+        psraw       mm2,        4           ; op[8] = (a1 - b1 + 7)>>4
+
+        movq        MMWORD PTR[rdi + 0 ],  mm0
+        movq        MMWORD PTR[rdi + 16],  mm2
+
+        ; output 1 and 3
+        ; interleave c1, d1
+        movq        mm1,        mm5         ; d1
+        punpcklwd   mm1,        mm3         ; c1 d1
+        punpckhwd   mm5,        mm3         ; c1 d1
+
+        movq        mm3,        mm1
+        movq        mm4,        mm5
+
+        pmaddwd     mm1,        MMWORD PTR[GLOBAL (_5352_2217)]    ; c1*2217 + d1*5352
+        pmaddwd     mm4,        MMWORD PTR[GLOBAL (_5352_2217)]    ; c1*2217 + d1*5352
+
+        pmaddwd     mm3,        MMWORD PTR[GLOBAL(_2217_neg5352)]  ; d1*2217 - c1*5352
+        pmaddwd     mm5,        MMWORD PTR[GLOBAL(_2217_neg5352)]  ; d1*2217 - c1*5352
+
+        paddd       mm1,        MMWORD PTR[GLOBAL(_12000)]
+        paddd       mm4,        MMWORD PTR[GLOBAL(_12000)]
+        paddd       mm3,        MMWORD PTR[GLOBAL(_51000)]
+        paddd       mm5,        MMWORD PTR[GLOBAL(_51000)]
+
+        psrad       mm1,        16          ; (c1 * 2217 + d1 * 5352 +  14500)>>16
+        psrad       mm4,        16          ; (c1 * 2217 + d1 * 5352 +  14500)>>16
+        psrad       mm3,        16          ; (d1 * 2217 - c1 * 5352 +   7500)>>16
+        psrad       mm5,        16          ; (d1 * 2217 - c1 * 5352 +   7500)>>16
+
+        packssdw    mm1,        mm4         ; op[4]
+        packssdw    mm3,        mm5         ; op[12]
+
+        paddw       mm1,        mm6         ; op[4] += (d1!=0)
+
+        movq        MMWORD PTR[rdi + 8 ],  mm1
+        movq        MMWORD PTR[rdi + 24],  mm3
+
+     ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 8
+_5352_2217:
+    dw 5352
+    dw 2217
+    dw 5352
+    dw 2217
+align 8
+_2217_neg5352:
+    dw 2217
+    dw -5352
+    dw 2217
+    dw -5352
+align 8
+_cmp_mask:
+    times 4 dw 1
+align 8
+_7w:
+    times 4 dw 7
+align 8
+_14500:
+    times 2 dd 14500
+align 8
+_7500:
+    times 2 dd 7500
+align 8
+_12000:
+    times 2 dd 12000
+align 8
+_51000:
+    times 2 dd 51000
diff --git a/libvpx/vp9/encoder/x86/vp9_dct_mmx.h b/libvpx/vp9/encoder/x86/vp9_dct_mmx.h
new file mode 100644
index 0000000..3bac7c8
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_dct_mmx.h
@@ -0,0 +1,17 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_X86_VP9_DCT_MMX_H_
+#define VP9_ENCODER_X86_VP9_DCT_MMX_H_
+
+extern void vp9_short_fdct4x4_mmx(short *input, short *output, int pitch);
+
+
+#endif /* VP9_ENCODER_X86_VP9_DCT_MMX_H_ */
diff --git a/libvpx/vp9/encoder/x86/vp9_dct_sse2.c b/libvpx/vp9/encoder/x86/vp9_dct_sse2.c
new file mode 100644
index 0000000..aaacebe
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_dct_sse2.c
@@ -0,0 +1,1000 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  // SSE2
+#include "vp9/common/vp9_idct.h"  // for cospi constants
+
+void vp9_short_fdct4x4_sse2(int16_t *input, int16_t *output, int pitch) {
+  // The 2D transform is done with two passes which are actually pretty
+  // similar. In the first one, we transform the columns and transpose
+  // the results. In the second one, we transform the rows. To achieve that,
+  // as the first pass results are transposed, we tranpose the columns (that
+  // is the transposed rows) and transpose the results (so that it goes back
+  // in normal/row positions).
+  const int stride = pitch >> 1;
+  int pass;
+  // Constants
+  //    When we use them, in one case, they are all the same. In all others
+  //    it's a pair of them that we need to repeat four times. This is done
+  //    by constructing the 32 bit constant corresponding to that pair.
+  const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
+  const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i k__nonzero_bias_a = _mm_setr_epi16(0, 1, 1, 1, 1, 1, 1, 1);
+  const __m128i k__nonzero_bias_b = _mm_setr_epi16(1, 0, 0, 0, 0, 0, 0, 0);
+  const __m128i kOne = _mm_set1_epi16(1);
+  __m128i in0, in1, in2, in3;
+  // Load inputs.
+  {
+    in0  = _mm_loadl_epi64((const __m128i *)(input +  0 * stride));
+    in1  = _mm_loadl_epi64((const __m128i *)(input +  1 * stride));
+    in2  = _mm_loadl_epi64((const __m128i *)(input +  2 * stride));
+    in3  = _mm_loadl_epi64((const __m128i *)(input +  3 * stride));
+    // x = x << 4
+    in0 = _mm_slli_epi16(in0, 4);
+    in1 = _mm_slli_epi16(in1, 4);
+    in2 = _mm_slli_epi16(in2, 4);
+    in3 = _mm_slli_epi16(in3, 4);
+    // if (i == 0 && input[0]) input[0] += 1;
+    {
+      // The mask will only contain wether the first value is zero, all
+      // other comparison will fail as something shifted by 4 (above << 4)
+      // can never be equal to one. To increment in the non-zero case, we
+      // add the mask and one for the first element:
+      //   - if zero, mask = -1, v = v - 1 + 1 = v
+      //   - if non-zero, mask = 0, v = v + 0 + 1 = v + 1
+      __m128i mask = _mm_cmpeq_epi16(in0, k__nonzero_bias_a);
+      in0 = _mm_add_epi16(in0, mask);
+      in0 = _mm_add_epi16(in0, k__nonzero_bias_b);
+    }
+  }
+  // Do the two transform/transpose passes
+  for (pass = 0; pass < 2; ++pass) {
+    // Transform 1/2: Add/substract
+    const __m128i r0 = _mm_add_epi16(in0, in3);
+    const __m128i r1 = _mm_add_epi16(in1, in2);
+    const __m128i r2 = _mm_sub_epi16(in1, in2);
+    const __m128i r3 = _mm_sub_epi16(in0, in3);
+    // Transform 1/2: Interleave to do the multiply by constants which gets us
+    //                into 32 bits.
+    const __m128i t0 = _mm_unpacklo_epi16(r0, r1);
+    const __m128i t2 = _mm_unpacklo_epi16(r2, r3);
+    const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p16_p16);
+    const __m128i u2 = _mm_madd_epi16(t0, k__cospi_p16_m16);
+    const __m128i u4 = _mm_madd_epi16(t2, k__cospi_p24_p08);
+    const __m128i u6 = _mm_madd_epi16(t2, k__cospi_m08_p24);
+    const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+    const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+    const __m128i v4 = _mm_add_epi32(u4, k__DCT_CONST_ROUNDING);
+    const __m128i v6 = _mm_add_epi32(u6, k__DCT_CONST_ROUNDING);
+    const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+    const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+    const __m128i w4 = _mm_srai_epi32(v4, DCT_CONST_BITS);
+    const __m128i w6 = _mm_srai_epi32(v6, DCT_CONST_BITS);
+    // Combine and transpose
+    const __m128i res0 = _mm_packs_epi32(w0, w2);
+    const __m128i res1 = _mm_packs_epi32(w4, w6);
+    // 00 01 02 03 20 21 22 23
+    // 10 11 12 13 30 31 32 33
+    const __m128i tr0_0 = _mm_unpacklo_epi16(res0, res1);
+    const __m128i tr0_1 = _mm_unpackhi_epi16(res0, res1);
+    // 00 10 01 11 02 12 03 13
+    // 20 30 21 31 22 32 23 33
+    in0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
+    in2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+    // 00 10 20 30 01 11 21 31      in0 contains 0 followed by 1
+    // 02 12 22 32 03 13 23 33      in2 contains 2 followed by 3
+    if (0 == pass) {
+      // Extract values in the high part for second pass as transform code
+      // only uses the first four values.
+      in1 = _mm_unpackhi_epi64(in0, in0);
+      in3 = _mm_unpackhi_epi64(in2, in2);
+    } else {
+      // Post-condition output and store it (v + 1) >> 2, taking advantage
+      // of the fact 1/3 are stored just after 0/2.
+      __m128i out01 = _mm_add_epi16(in0, kOne);
+      __m128i out23 = _mm_add_epi16(in2, kOne);
+      out01 = _mm_srai_epi16(out01, 2);
+      out23 = _mm_srai_epi16(out23, 2);
+      _mm_storeu_si128((__m128i *)(output + 0 * 4), out01);
+      _mm_storeu_si128((__m128i *)(output + 2 * 4), out23);
+    }
+  }
+}
+
+void vp9_short_fdct8x4_sse2(int16_t *input, int16_t *output, int pitch) {
+  vp9_short_fdct4x4_sse2(input, output, pitch);
+  vp9_short_fdct4x4_sse2(input + 4, output + 16, pitch);
+}
+
+void vp9_short_fdct8x8_sse2(int16_t *input, int16_t *output, int pitch) {
+  const int stride = pitch >> 1;
+  int pass;
+  // Constants
+  //    When we use them, in one case, they are all the same. In all others
+  //    it's a pair of them that we need to repeat four times. This is done
+  //    by constructing the 32 bit constant corresponding to that pair.
+  const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
+  const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i k__cospi_p28_p04 = pair_set_epi16(cospi_28_64, cospi_4_64);
+  const __m128i k__cospi_m04_p28 = pair_set_epi16(-cospi_4_64, cospi_28_64);
+  const __m128i k__cospi_p12_p20 = pair_set_epi16(cospi_12_64, cospi_20_64);
+  const __m128i k__cospi_m20_p12 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+  const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  // Load input
+  __m128i in0  = _mm_loadu_si128((const __m128i *)(input + 0 * stride));
+  __m128i in1  = _mm_loadu_si128((const __m128i *)(input + 1 * stride));
+  __m128i in2  = _mm_loadu_si128((const __m128i *)(input + 2 * stride));
+  __m128i in3  = _mm_loadu_si128((const __m128i *)(input + 3 * stride));
+  __m128i in4  = _mm_loadu_si128((const __m128i *)(input + 4 * stride));
+  __m128i in5  = _mm_loadu_si128((const __m128i *)(input + 5 * stride));
+  __m128i in6  = _mm_loadu_si128((const __m128i *)(input + 6 * stride));
+  __m128i in7  = _mm_loadu_si128((const __m128i *)(input + 7 * stride));
+  // Pre-condition input (shift by two)
+  in0 = _mm_slli_epi16(in0, 2);
+  in1 = _mm_slli_epi16(in1, 2);
+  in2 = _mm_slli_epi16(in2, 2);
+  in3 = _mm_slli_epi16(in3, 2);
+  in4 = _mm_slli_epi16(in4, 2);
+  in5 = _mm_slli_epi16(in5, 2);
+  in6 = _mm_slli_epi16(in6, 2);
+  in7 = _mm_slli_epi16(in7, 2);
+
+  // We do two passes, first the columns, then the rows. The results of the
+  // first pass are transposed so that the same column code can be reused. The
+  // results of the second pass are also transposed so that the rows (processed
+  // as columns) are put back in row positions.
+  for (pass = 0; pass < 2; pass++) {
+    // To store results of each pass before the transpose.
+    __m128i res0, res1, res2, res3, res4, res5, res6, res7;
+    // Add/substract
+    const __m128i q0 = _mm_add_epi16(in0, in7);
+    const __m128i q1 = _mm_add_epi16(in1, in6);
+    const __m128i q2 = _mm_add_epi16(in2, in5);
+    const __m128i q3 = _mm_add_epi16(in3, in4);
+    const __m128i q4 = _mm_sub_epi16(in3, in4);
+    const __m128i q5 = _mm_sub_epi16(in2, in5);
+    const __m128i q6 = _mm_sub_epi16(in1, in6);
+    const __m128i q7 = _mm_sub_epi16(in0, in7);
+    // Work on first four results
+    {
+      // Add/substract
+      const __m128i r0 = _mm_add_epi16(q0, q3);
+      const __m128i r1 = _mm_add_epi16(q1, q2);
+      const __m128i r2 = _mm_sub_epi16(q1, q2);
+      const __m128i r3 = _mm_sub_epi16(q0, q3);
+      // Interleave to do the multiply by constants which gets us into 32bits
+      const __m128i t0 = _mm_unpacklo_epi16(r0, r1);
+      const __m128i t1 = _mm_unpackhi_epi16(r0, r1);
+      const __m128i t2 = _mm_unpacklo_epi16(r2, r3);
+      const __m128i t3 = _mm_unpackhi_epi16(r2, r3);
+      const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p16_p16);
+      const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p16_p16);
+      const __m128i u2 = _mm_madd_epi16(t0, k__cospi_p16_m16);
+      const __m128i u3 = _mm_madd_epi16(t1, k__cospi_p16_m16);
+      const __m128i u4 = _mm_madd_epi16(t2, k__cospi_p24_p08);
+      const __m128i u5 = _mm_madd_epi16(t3, k__cospi_p24_p08);
+      const __m128i u6 = _mm_madd_epi16(t2, k__cospi_m08_p24);
+      const __m128i u7 = _mm_madd_epi16(t3, k__cospi_m08_p24);
+      // dct_const_round_shift
+      const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+      const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+      const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+      const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+      const __m128i v4 = _mm_add_epi32(u4, k__DCT_CONST_ROUNDING);
+      const __m128i v5 = _mm_add_epi32(u5, k__DCT_CONST_ROUNDING);
+      const __m128i v6 = _mm_add_epi32(u6, k__DCT_CONST_ROUNDING);
+      const __m128i v7 = _mm_add_epi32(u7, k__DCT_CONST_ROUNDING);
+      const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+      const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+      const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+      const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+      const __m128i w4 = _mm_srai_epi32(v4, DCT_CONST_BITS);
+      const __m128i w5 = _mm_srai_epi32(v5, DCT_CONST_BITS);
+      const __m128i w6 = _mm_srai_epi32(v6, DCT_CONST_BITS);
+      const __m128i w7 = _mm_srai_epi32(v7, DCT_CONST_BITS);
+      // Combine
+      res0 = _mm_packs_epi32(w0, w1);
+      res4 = _mm_packs_epi32(w2, w3);
+      res2 = _mm_packs_epi32(w4, w5);
+      res6 = _mm_packs_epi32(w6, w7);
+    }
+    // Work on next four results
+    {
+      // Interleave to do the multiply by constants which gets us into 32bits
+      const __m128i d0 = _mm_unpacklo_epi16(q6, q5);
+      const __m128i d1 = _mm_unpackhi_epi16(q6, q5);
+      const __m128i e0 = _mm_madd_epi16(d0, k__cospi_p16_m16);
+      const __m128i e1 = _mm_madd_epi16(d1, k__cospi_p16_m16);
+      const __m128i e2 = _mm_madd_epi16(d0, k__cospi_p16_p16);
+      const __m128i e3 = _mm_madd_epi16(d1, k__cospi_p16_p16);
+      // dct_const_round_shift
+      const __m128i f0 = _mm_add_epi32(e0, k__DCT_CONST_ROUNDING);
+      const __m128i f1 = _mm_add_epi32(e1, k__DCT_CONST_ROUNDING);
+      const __m128i f2 = _mm_add_epi32(e2, k__DCT_CONST_ROUNDING);
+      const __m128i f3 = _mm_add_epi32(e3, k__DCT_CONST_ROUNDING);
+      const __m128i s0 = _mm_srai_epi32(f0, DCT_CONST_BITS);
+      const __m128i s1 = _mm_srai_epi32(f1, DCT_CONST_BITS);
+      const __m128i s2 = _mm_srai_epi32(f2, DCT_CONST_BITS);
+      const __m128i s3 = _mm_srai_epi32(f3, DCT_CONST_BITS);
+      // Combine
+      const __m128i r0 = _mm_packs_epi32(s0, s1);
+      const __m128i r1 = _mm_packs_epi32(s2, s3);
+      // Add/substract
+      const __m128i x0 = _mm_add_epi16(q4, r0);
+      const __m128i x1 = _mm_sub_epi16(q4, r0);
+      const __m128i x2 = _mm_sub_epi16(q7, r1);
+      const __m128i x3 = _mm_add_epi16(q7, r1);
+      // Interleave to do the multiply by constants which gets us into 32bits
+      const __m128i t0 = _mm_unpacklo_epi16(x0, x3);
+      const __m128i t1 = _mm_unpackhi_epi16(x0, x3);
+      const __m128i t2 = _mm_unpacklo_epi16(x1, x2);
+      const __m128i t3 = _mm_unpackhi_epi16(x1, x2);
+      const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p28_p04);
+      const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p28_p04);
+      const __m128i u2 = _mm_madd_epi16(t0, k__cospi_m04_p28);
+      const __m128i u3 = _mm_madd_epi16(t1, k__cospi_m04_p28);
+      const __m128i u4 = _mm_madd_epi16(t2, k__cospi_p12_p20);
+      const __m128i u5 = _mm_madd_epi16(t3, k__cospi_p12_p20);
+      const __m128i u6 = _mm_madd_epi16(t2, k__cospi_m20_p12);
+      const __m128i u7 = _mm_madd_epi16(t3, k__cospi_m20_p12);
+      // dct_const_round_shift
+      const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+      const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+      const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+      const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+      const __m128i v4 = _mm_add_epi32(u4, k__DCT_CONST_ROUNDING);
+      const __m128i v5 = _mm_add_epi32(u5, k__DCT_CONST_ROUNDING);
+      const __m128i v6 = _mm_add_epi32(u6, k__DCT_CONST_ROUNDING);
+      const __m128i v7 = _mm_add_epi32(u7, k__DCT_CONST_ROUNDING);
+      const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+      const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+      const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+      const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+      const __m128i w4 = _mm_srai_epi32(v4, DCT_CONST_BITS);
+      const __m128i w5 = _mm_srai_epi32(v5, DCT_CONST_BITS);
+      const __m128i w6 = _mm_srai_epi32(v6, DCT_CONST_BITS);
+      const __m128i w7 = _mm_srai_epi32(v7, DCT_CONST_BITS);
+      // Combine
+      res1 = _mm_packs_epi32(w0, w1);
+      res7 = _mm_packs_epi32(w2, w3);
+      res5 = _mm_packs_epi32(w4, w5);
+      res3 = _mm_packs_epi32(w6, w7);
+    }
+    // Transpose the 8x8.
+    {
+      // 00 01 02 03 04 05 06 07
+      // 10 11 12 13 14 15 16 17
+      // 20 21 22 23 24 25 26 27
+      // 30 31 32 33 34 35 36 37
+      // 40 41 42 43 44 45 46 47
+      // 50 51 52 53 54 55 56 57
+      // 60 61 62 63 64 65 66 67
+      // 70 71 72 73 74 75 76 77
+      const __m128i tr0_0 = _mm_unpacklo_epi16(res0, res1);
+      const __m128i tr0_1 = _mm_unpacklo_epi16(res2, res3);
+      const __m128i tr0_2 = _mm_unpackhi_epi16(res0, res1);
+      const __m128i tr0_3 = _mm_unpackhi_epi16(res2, res3);
+      const __m128i tr0_4 = _mm_unpacklo_epi16(res4, res5);
+      const __m128i tr0_5 = _mm_unpacklo_epi16(res6, res7);
+      const __m128i tr0_6 = _mm_unpackhi_epi16(res4, res5);
+      const __m128i tr0_7 = _mm_unpackhi_epi16(res6, res7);
+      // 00 10 01 11 02 12 03 13
+      // 20 30 21 31 22 32 23 33
+      // 04 14 05 15 06 16 07 17
+      // 24 34 25 35 26 36 27 37
+      // 40 50 41 51 42 52 43 53
+      // 60 70 61 71 62 72 63 73
+      // 54 54 55 55 56 56 57 57
+      // 64 74 65 75 66 76 67 77
+      const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
+      const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3);
+      const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+      const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
+      const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
+      const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
+      const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
+      const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);
+      // 00 10 20 30 01 11 21 31
+      // 40 50 60 70 41 51 61 71
+      // 02 12 22 32 03 13 23 33
+      // 42 52 62 72 43 53 63 73
+      // 04 14 24 34 05 15 21 36
+      // 44 54 64 74 45 55 61 76
+      // 06 16 26 36 07 17 27 37
+      // 46 56 66 76 47 57 67 77
+      in0 = _mm_unpacklo_epi64(tr1_0, tr1_4);
+      in1 = _mm_unpackhi_epi64(tr1_0, tr1_4);
+      in2 = _mm_unpacklo_epi64(tr1_2, tr1_6);
+      in3 = _mm_unpackhi_epi64(tr1_2, tr1_6);
+      in4 = _mm_unpacklo_epi64(tr1_1, tr1_5);
+      in5 = _mm_unpackhi_epi64(tr1_1, tr1_5);
+      in6 = _mm_unpacklo_epi64(tr1_3, tr1_7);
+      in7 = _mm_unpackhi_epi64(tr1_3, tr1_7);
+      // 00 10 20 30 40 50 60 70
+      // 01 11 21 31 41 51 61 71
+      // 02 12 22 32 42 52 62 72
+      // 03 13 23 33 43 53 63 73
+      // 04 14 24 34 44 54 64 74
+      // 05 15 25 35 45 55 65 75
+      // 06 16 26 36 46 56 66 76
+      // 07 17 27 37 47 57 67 77
+    }
+  }
+  // Post-condition output and store it
+  {
+    // Post-condition (division by two)
+    //    division of two 16 bits signed numbers using shifts
+    //    n / 2 = (n - (n >> 15)) >> 1
+    const __m128i sign_in0 = _mm_srai_epi16(in0, 15);
+    const __m128i sign_in1 = _mm_srai_epi16(in1, 15);
+    const __m128i sign_in2 = _mm_srai_epi16(in2, 15);
+    const __m128i sign_in3 = _mm_srai_epi16(in3, 15);
+    const __m128i sign_in4 = _mm_srai_epi16(in4, 15);
+    const __m128i sign_in5 = _mm_srai_epi16(in5, 15);
+    const __m128i sign_in6 = _mm_srai_epi16(in6, 15);
+    const __m128i sign_in7 = _mm_srai_epi16(in7, 15);
+    in0 = _mm_sub_epi16(in0, sign_in0);
+    in1 = _mm_sub_epi16(in1, sign_in1);
+    in2 = _mm_sub_epi16(in2, sign_in2);
+    in3 = _mm_sub_epi16(in3, sign_in3);
+    in4 = _mm_sub_epi16(in4, sign_in4);
+    in5 = _mm_sub_epi16(in5, sign_in5);
+    in6 = _mm_sub_epi16(in6, sign_in6);
+    in7 = _mm_sub_epi16(in7, sign_in7);
+    in0 = _mm_srai_epi16(in0, 1);
+    in1 = _mm_srai_epi16(in1, 1);
+    in2 = _mm_srai_epi16(in2, 1);
+    in3 = _mm_srai_epi16(in3, 1);
+    in4 = _mm_srai_epi16(in4, 1);
+    in5 = _mm_srai_epi16(in5, 1);
+    in6 = _mm_srai_epi16(in6, 1);
+    in7 = _mm_srai_epi16(in7, 1);
+    // store results
+    _mm_storeu_si128((__m128i *)(output + 0 * 8), in0);
+    _mm_storeu_si128((__m128i *)(output + 1 * 8), in1);
+    _mm_storeu_si128((__m128i *)(output + 2 * 8), in2);
+    _mm_storeu_si128((__m128i *)(output + 3 * 8), in3);
+    _mm_storeu_si128((__m128i *)(output + 4 * 8), in4);
+    _mm_storeu_si128((__m128i *)(output + 5 * 8), in5);
+    _mm_storeu_si128((__m128i *)(output + 6 * 8), in6);
+    _mm_storeu_si128((__m128i *)(output + 7 * 8), in7);
+  }
+}
+
+void vp9_short_fdct16x16_sse2(int16_t *input, int16_t *output, int pitch) {
+  // The 2D transform is done with two passes which are actually pretty
+  // similar. In the first one, we transform the columns and transpose
+  // the results. In the second one, we transform the rows. To achieve that,
+  // as the first pass results are transposed, we tranpose the columns (that
+  // is the transposed rows) and transpose the results (so that it goes back
+  // in normal/row positions).
+  const int stride = pitch >> 1;
+  int pass;
+  // We need an intermediate buffer between passes.
+  int16_t intermediate[256];
+  int16_t *in = input;
+  int16_t *out = intermediate;
+  // Constants
+  //    When we use them, in one case, they are all the same. In all others
+  //    it's a pair of them that we need to repeat four times. This is done
+  //    by constructing the 32 bit constant corresponding to that pair.
+  const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
+  const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+  const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64);
+  const __m128i k__cospi_m24_m08 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
+  const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+  const __m128i k__cospi_p28_p04 = pair_set_epi16(cospi_28_64, cospi_4_64);
+  const __m128i k__cospi_m04_p28 = pair_set_epi16(-cospi_4_64, cospi_28_64);
+  const __m128i k__cospi_p12_p20 = pair_set_epi16(cospi_12_64, cospi_20_64);
+  const __m128i k__cospi_m20_p12 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+  const __m128i k__cospi_p30_p02 = pair_set_epi16(cospi_30_64, cospi_2_64);
+  const __m128i k__cospi_p14_p18 = pair_set_epi16(cospi_14_64, cospi_18_64);
+  const __m128i k__cospi_m02_p30 = pair_set_epi16(-cospi_2_64, cospi_30_64);
+  const __m128i k__cospi_m18_p14 = pair_set_epi16(-cospi_18_64, cospi_14_64);
+  const __m128i k__cospi_p22_p10 = pair_set_epi16(cospi_22_64, cospi_10_64);
+  const __m128i k__cospi_p06_p26 = pair_set_epi16(cospi_6_64, cospi_26_64);
+  const __m128i k__cospi_m10_p22 = pair_set_epi16(-cospi_10_64, cospi_22_64);
+  const __m128i k__cospi_m26_p06 = pair_set_epi16(-cospi_26_64, cospi_6_64);
+  const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
+  const __m128i kOne = _mm_set1_epi16(1);
+  // Do the two transform/transpose passes
+  for (pass = 0; pass < 2; ++pass) {
+    // We process eight columns (transposed rows in second pass) at a time.
+    int column_start;
+    for (column_start = 0; column_start < 16; column_start += 8) {
+      __m128i in00, in01, in02, in03, in04, in05, in06, in07;
+      __m128i in08, in09, in10, in11, in12, in13, in14, in15;
+      __m128i input0, input1, input2, input3, input4, input5, input6, input7;
+      __m128i step1_0, step1_1, step1_2, step1_3;
+      __m128i step1_4, step1_5, step1_6, step1_7;
+      __m128i step2_1, step2_2, step2_3, step2_4, step2_5, step2_6;
+      __m128i step3_0, step3_1, step3_2, step3_3;
+      __m128i step3_4, step3_5, step3_6, step3_7;
+      __m128i res00, res01, res02, res03, res04, res05, res06, res07;
+      __m128i res08, res09, res10, res11, res12, res13, res14, res15;
+      // Load and pre-condition input.
+      if (0 == pass) {
+        in00  = _mm_loadu_si128((const __m128i *)(in +  0 * stride));
+        in01  = _mm_loadu_si128((const __m128i *)(in +  1 * stride));
+        in02  = _mm_loadu_si128((const __m128i *)(in +  2 * stride));
+        in03  = _mm_loadu_si128((const __m128i *)(in +  3 * stride));
+        in04  = _mm_loadu_si128((const __m128i *)(in +  4 * stride));
+        in05  = _mm_loadu_si128((const __m128i *)(in +  5 * stride));
+        in06  = _mm_loadu_si128((const __m128i *)(in +  6 * stride));
+        in07  = _mm_loadu_si128((const __m128i *)(in +  7 * stride));
+        in08  = _mm_loadu_si128((const __m128i *)(in +  8 * stride));
+        in09  = _mm_loadu_si128((const __m128i *)(in +  9 * stride));
+        in10  = _mm_loadu_si128((const __m128i *)(in + 10 * stride));
+        in11  = _mm_loadu_si128((const __m128i *)(in + 11 * stride));
+        in12  = _mm_loadu_si128((const __m128i *)(in + 12 * stride));
+        in13  = _mm_loadu_si128((const __m128i *)(in + 13 * stride));
+        in14  = _mm_loadu_si128((const __m128i *)(in + 14 * stride));
+        in15  = _mm_loadu_si128((const __m128i *)(in + 15 * stride));
+        // x = x << 2
+        in00 = _mm_slli_epi16(in00, 2);
+        in01 = _mm_slli_epi16(in01, 2);
+        in02 = _mm_slli_epi16(in02, 2);
+        in03 = _mm_slli_epi16(in03, 2);
+        in04 = _mm_slli_epi16(in04, 2);
+        in05 = _mm_slli_epi16(in05, 2);
+        in06 = _mm_slli_epi16(in06, 2);
+        in07 = _mm_slli_epi16(in07, 2);
+        in08 = _mm_slli_epi16(in08, 2);
+        in09 = _mm_slli_epi16(in09, 2);
+        in10 = _mm_slli_epi16(in10, 2);
+        in11 = _mm_slli_epi16(in11, 2);
+        in12 = _mm_slli_epi16(in12, 2);
+        in13 = _mm_slli_epi16(in13, 2);
+        in14 = _mm_slli_epi16(in14, 2);
+        in15 = _mm_slli_epi16(in15, 2);
+      } else {
+        in00  = _mm_loadu_si128((const __m128i *)(in +  0 * 16));
+        in01  = _mm_loadu_si128((const __m128i *)(in +  1 * 16));
+        in02  = _mm_loadu_si128((const __m128i *)(in +  2 * 16));
+        in03  = _mm_loadu_si128((const __m128i *)(in +  3 * 16));
+        in04  = _mm_loadu_si128((const __m128i *)(in +  4 * 16));
+        in05  = _mm_loadu_si128((const __m128i *)(in +  5 * 16));
+        in06  = _mm_loadu_si128((const __m128i *)(in +  6 * 16));
+        in07  = _mm_loadu_si128((const __m128i *)(in +  7 * 16));
+        in08  = _mm_loadu_si128((const __m128i *)(in +  8 * 16));
+        in09  = _mm_loadu_si128((const __m128i *)(in +  9 * 16));
+        in10  = _mm_loadu_si128((const __m128i *)(in + 10 * 16));
+        in11  = _mm_loadu_si128((const __m128i *)(in + 11 * 16));
+        in12  = _mm_loadu_si128((const __m128i *)(in + 12 * 16));
+        in13  = _mm_loadu_si128((const __m128i *)(in + 13 * 16));
+        in14  = _mm_loadu_si128((const __m128i *)(in + 14 * 16));
+        in15  = _mm_loadu_si128((const __m128i *)(in + 15 * 16));
+        // x = (x + 1) >> 2
+        in00 = _mm_add_epi16(in00, kOne);
+        in01 = _mm_add_epi16(in01, kOne);
+        in02 = _mm_add_epi16(in02, kOne);
+        in03 = _mm_add_epi16(in03, kOne);
+        in04 = _mm_add_epi16(in04, kOne);
+        in05 = _mm_add_epi16(in05, kOne);
+        in06 = _mm_add_epi16(in06, kOne);
+        in07 = _mm_add_epi16(in07, kOne);
+        in08 = _mm_add_epi16(in08, kOne);
+        in09 = _mm_add_epi16(in09, kOne);
+        in10 = _mm_add_epi16(in10, kOne);
+        in11 = _mm_add_epi16(in11, kOne);
+        in12 = _mm_add_epi16(in12, kOne);
+        in13 = _mm_add_epi16(in13, kOne);
+        in14 = _mm_add_epi16(in14, kOne);
+        in15 = _mm_add_epi16(in15, kOne);
+        in00 = _mm_srai_epi16(in00, 2);
+        in01 = _mm_srai_epi16(in01, 2);
+        in02 = _mm_srai_epi16(in02, 2);
+        in03 = _mm_srai_epi16(in03, 2);
+        in04 = _mm_srai_epi16(in04, 2);
+        in05 = _mm_srai_epi16(in05, 2);
+        in06 = _mm_srai_epi16(in06, 2);
+        in07 = _mm_srai_epi16(in07, 2);
+        in08 = _mm_srai_epi16(in08, 2);
+        in09 = _mm_srai_epi16(in09, 2);
+        in10 = _mm_srai_epi16(in10, 2);
+        in11 = _mm_srai_epi16(in11, 2);
+        in12 = _mm_srai_epi16(in12, 2);
+        in13 = _mm_srai_epi16(in13, 2);
+        in14 = _mm_srai_epi16(in14, 2);
+        in15 = _mm_srai_epi16(in15, 2);
+      }
+      in += 8;
+      // Calculate input for the first 8 results.
+      {
+        input0 = _mm_add_epi16(in00, in15);
+        input1 = _mm_add_epi16(in01, in14);
+        input2 = _mm_add_epi16(in02, in13);
+        input3 = _mm_add_epi16(in03, in12);
+        input4 = _mm_add_epi16(in04, in11);
+        input5 = _mm_add_epi16(in05, in10);
+        input6 = _mm_add_epi16(in06, in09);
+        input7 = _mm_add_epi16(in07, in08);
+      }
+      // Calculate input for the next 8 results.
+      {
+        step1_0 = _mm_sub_epi16(in07, in08);
+        step1_1 = _mm_sub_epi16(in06, in09);
+        step1_2 = _mm_sub_epi16(in05, in10);
+        step1_3 = _mm_sub_epi16(in04, in11);
+        step1_4 = _mm_sub_epi16(in03, in12);
+        step1_5 = _mm_sub_epi16(in02, in13);
+        step1_6 = _mm_sub_epi16(in01, in14);
+        step1_7 = _mm_sub_epi16(in00, in15);
+      }
+      // Work on the first eight values; fdct8_1d(input, even_results);
+      {
+        // Add/substract
+        const __m128i q0 = _mm_add_epi16(input0, input7);
+        const __m128i q1 = _mm_add_epi16(input1, input6);
+        const __m128i q2 = _mm_add_epi16(input2, input5);
+        const __m128i q3 = _mm_add_epi16(input3, input4);
+        const __m128i q4 = _mm_sub_epi16(input3, input4);
+        const __m128i q5 = _mm_sub_epi16(input2, input5);
+        const __m128i q6 = _mm_sub_epi16(input1, input6);
+        const __m128i q7 = _mm_sub_epi16(input0, input7);
+        // Work on first four results
+        {
+          // Add/substract
+          const __m128i r0 = _mm_add_epi16(q0, q3);
+          const __m128i r1 = _mm_add_epi16(q1, q2);
+          const __m128i r2 = _mm_sub_epi16(q1, q2);
+          const __m128i r3 = _mm_sub_epi16(q0, q3);
+          // Interleave to do the multiply by constants which gets us
+          // into 32 bits.
+          const __m128i t0 = _mm_unpacklo_epi16(r0, r1);
+          const __m128i t1 = _mm_unpackhi_epi16(r0, r1);
+          const __m128i t2 = _mm_unpacklo_epi16(r2, r3);
+          const __m128i t3 = _mm_unpackhi_epi16(r2, r3);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p16_p16);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p16_p16);
+          const __m128i u2 = _mm_madd_epi16(t0, k__cospi_p16_m16);
+          const __m128i u3 = _mm_madd_epi16(t1, k__cospi_p16_m16);
+          const __m128i u4 = _mm_madd_epi16(t2, k__cospi_p24_p08);
+          const __m128i u5 = _mm_madd_epi16(t3, k__cospi_p24_p08);
+          const __m128i u6 = _mm_madd_epi16(t2, k__cospi_m08_p24);
+          const __m128i u7 = _mm_madd_epi16(t3, k__cospi_m08_p24);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i v4 = _mm_add_epi32(u4, k__DCT_CONST_ROUNDING);
+          const __m128i v5 = _mm_add_epi32(u5, k__DCT_CONST_ROUNDING);
+          const __m128i v6 = _mm_add_epi32(u6, k__DCT_CONST_ROUNDING);
+          const __m128i v7 = _mm_add_epi32(u7, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          const __m128i w4 = _mm_srai_epi32(v4, DCT_CONST_BITS);
+          const __m128i w5 = _mm_srai_epi32(v5, DCT_CONST_BITS);
+          const __m128i w6 = _mm_srai_epi32(v6, DCT_CONST_BITS);
+          const __m128i w7 = _mm_srai_epi32(v7, DCT_CONST_BITS);
+          // Combine
+          res00 = _mm_packs_epi32(w0, w1);
+          res08 = _mm_packs_epi32(w2, w3);
+          res04 = _mm_packs_epi32(w4, w5);
+          res12 = _mm_packs_epi32(w6, w7);
+        }
+        // Work on next four results
+        {
+          // Interleave to do the multiply by constants which gets us
+          // into 32 bits.
+          const __m128i d0 = _mm_unpacklo_epi16(q6, q5);
+          const __m128i d1 = _mm_unpackhi_epi16(q6, q5);
+          const __m128i e0 = _mm_madd_epi16(d0, k__cospi_p16_m16);
+          const __m128i e1 = _mm_madd_epi16(d1, k__cospi_p16_m16);
+          const __m128i e2 = _mm_madd_epi16(d0, k__cospi_p16_p16);
+          const __m128i e3 = _mm_madd_epi16(d1, k__cospi_p16_p16);
+          // dct_const_round_shift
+          const __m128i f0 = _mm_add_epi32(e0, k__DCT_CONST_ROUNDING);
+          const __m128i f1 = _mm_add_epi32(e1, k__DCT_CONST_ROUNDING);
+          const __m128i f2 = _mm_add_epi32(e2, k__DCT_CONST_ROUNDING);
+          const __m128i f3 = _mm_add_epi32(e3, k__DCT_CONST_ROUNDING);
+          const __m128i s0 = _mm_srai_epi32(f0, DCT_CONST_BITS);
+          const __m128i s1 = _mm_srai_epi32(f1, DCT_CONST_BITS);
+          const __m128i s2 = _mm_srai_epi32(f2, DCT_CONST_BITS);
+          const __m128i s3 = _mm_srai_epi32(f3, DCT_CONST_BITS);
+          // Combine
+          const __m128i r0 = _mm_packs_epi32(s0, s1);
+          const __m128i r1 = _mm_packs_epi32(s2, s3);
+          // Add/substract
+          const __m128i x0 = _mm_add_epi16(q4, r0);
+          const __m128i x1 = _mm_sub_epi16(q4, r0);
+          const __m128i x2 = _mm_sub_epi16(q7, r1);
+          const __m128i x3 = _mm_add_epi16(q7, r1);
+          // Interleave to do the multiply by constants which gets us
+          // into 32 bits.
+          const __m128i t0 = _mm_unpacklo_epi16(x0, x3);
+          const __m128i t1 = _mm_unpackhi_epi16(x0, x3);
+          const __m128i t2 = _mm_unpacklo_epi16(x1, x2);
+          const __m128i t3 = _mm_unpackhi_epi16(x1, x2);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p28_p04);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p28_p04);
+          const __m128i u2 = _mm_madd_epi16(t0, k__cospi_m04_p28);
+          const __m128i u3 = _mm_madd_epi16(t1, k__cospi_m04_p28);
+          const __m128i u4 = _mm_madd_epi16(t2, k__cospi_p12_p20);
+          const __m128i u5 = _mm_madd_epi16(t3, k__cospi_p12_p20);
+          const __m128i u6 = _mm_madd_epi16(t2, k__cospi_m20_p12);
+          const __m128i u7 = _mm_madd_epi16(t3, k__cospi_m20_p12);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i v4 = _mm_add_epi32(u4, k__DCT_CONST_ROUNDING);
+          const __m128i v5 = _mm_add_epi32(u5, k__DCT_CONST_ROUNDING);
+          const __m128i v6 = _mm_add_epi32(u6, k__DCT_CONST_ROUNDING);
+          const __m128i v7 = _mm_add_epi32(u7, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          const __m128i w4 = _mm_srai_epi32(v4, DCT_CONST_BITS);
+          const __m128i w5 = _mm_srai_epi32(v5, DCT_CONST_BITS);
+          const __m128i w6 = _mm_srai_epi32(v6, DCT_CONST_BITS);
+          const __m128i w7 = _mm_srai_epi32(v7, DCT_CONST_BITS);
+          // Combine
+          res02 = _mm_packs_epi32(w0, w1);
+          res14 = _mm_packs_epi32(w2, w3);
+          res10 = _mm_packs_epi32(w4, w5);
+          res06 = _mm_packs_epi32(w6, w7);
+        }
+      }
+      // Work on the next eight values; step1 -> odd_results
+      {
+        // step 2
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_5, step1_2);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_5, step1_2);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_4, step1_3);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_4, step1_3);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p16_m16);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p16_m16);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_p16_m16);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_p16_m16);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          step2_2 = _mm_packs_epi32(w0, w1);
+          step2_3 = _mm_packs_epi32(w2, w3);
+        }
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_5, step1_2);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_5, step1_2);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_4, step1_3);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_4, step1_3);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p16_p16);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p16_p16);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_p16_p16);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_p16_p16);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          step2_5 = _mm_packs_epi32(w0, w1);
+          step2_4 = _mm_packs_epi32(w2, w3);
+        }
+        // step 3
+        {
+          step3_0 = _mm_add_epi16(step1_0, step2_3);
+          step3_1 = _mm_add_epi16(step1_1, step2_2);
+          step3_2 = _mm_sub_epi16(step1_1, step2_2);
+          step3_3 = _mm_sub_epi16(step1_0, step2_3);
+          step3_4 = _mm_sub_epi16(step1_7, step2_4);
+          step3_5 = _mm_sub_epi16(step1_6, step2_5);
+          step3_6 = _mm_add_epi16(step1_6, step2_5);
+          step3_7 = _mm_add_epi16(step1_7, step2_4);
+        }
+        // step 4
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step3_1, step3_6);
+          const __m128i t1 = _mm_unpackhi_epi16(step3_1, step3_6);
+          const __m128i t2 = _mm_unpacklo_epi16(step3_2, step3_5);
+          const __m128i t3 = _mm_unpackhi_epi16(step3_2, step3_5);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_m08_p24);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_m08_p24);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_m24_m08);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_m24_m08);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          step2_1 = _mm_packs_epi32(w0, w1);
+          step2_2 = _mm_packs_epi32(w2, w3);
+        }
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step3_1, step3_6);
+          const __m128i t1 = _mm_unpackhi_epi16(step3_1, step3_6);
+          const __m128i t2 = _mm_unpacklo_epi16(step3_2, step3_5);
+          const __m128i t3 = _mm_unpackhi_epi16(step3_2, step3_5);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p24_p08);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p24_p08);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_m08_p24);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_m08_p24);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          step2_6 = _mm_packs_epi32(w0, w1);
+          step2_5 = _mm_packs_epi32(w2, w3);
+        }
+        // step 5
+        {
+          step1_0 = _mm_add_epi16(step3_0, step2_1);
+          step1_1 = _mm_sub_epi16(step3_0, step2_1);
+          step1_2 = _mm_sub_epi16(step3_3, step2_2);
+          step1_3 = _mm_add_epi16(step3_3, step2_2);
+          step1_4 = _mm_add_epi16(step3_4, step2_5);
+          step1_5 = _mm_sub_epi16(step3_4, step2_5);
+          step1_6 = _mm_sub_epi16(step3_7, step2_6);
+          step1_7 = _mm_add_epi16(step3_7, step2_6);
+        }
+        // step 6
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_0, step1_7);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_0, step1_7);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_1, step1_6);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_1, step1_6);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p30_p02);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p30_p02);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_p14_p18);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_p14_p18);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          res01 = _mm_packs_epi32(w0, w1);
+          res09 = _mm_packs_epi32(w2, w3);
+        }
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_2, step1_5);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_2, step1_5);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_3, step1_4);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_3, step1_4);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_p22_p10);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_p22_p10);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_p06_p26);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_p06_p26);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          res05 = _mm_packs_epi32(w0, w1);
+          res13 = _mm_packs_epi32(w2, w3);
+        }
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_2, step1_5);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_2, step1_5);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_3, step1_4);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_3, step1_4);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_m10_p22);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_m10_p22);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_m26_p06);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_m26_p06);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          res11 = _mm_packs_epi32(w0, w1);
+          res03 = _mm_packs_epi32(w2, w3);
+        }
+        {
+          const __m128i t0 = _mm_unpacklo_epi16(step1_0, step1_7);
+          const __m128i t1 = _mm_unpackhi_epi16(step1_0, step1_7);
+          const __m128i t2 = _mm_unpacklo_epi16(step1_1, step1_6);
+          const __m128i t3 = _mm_unpackhi_epi16(step1_1, step1_6);
+          const __m128i u0 = _mm_madd_epi16(t0, k__cospi_m02_p30);
+          const __m128i u1 = _mm_madd_epi16(t1, k__cospi_m02_p30);
+          const __m128i u2 = _mm_madd_epi16(t2, k__cospi_m18_p14);
+          const __m128i u3 = _mm_madd_epi16(t3, k__cospi_m18_p14);
+          // dct_const_round_shift
+          const __m128i v0 = _mm_add_epi32(u0, k__DCT_CONST_ROUNDING);
+          const __m128i v1 = _mm_add_epi32(u1, k__DCT_CONST_ROUNDING);
+          const __m128i v2 = _mm_add_epi32(u2, k__DCT_CONST_ROUNDING);
+          const __m128i v3 = _mm_add_epi32(u3, k__DCT_CONST_ROUNDING);
+          const __m128i w0 = _mm_srai_epi32(v0, DCT_CONST_BITS);
+          const __m128i w1 = _mm_srai_epi32(v1, DCT_CONST_BITS);
+          const __m128i w2 = _mm_srai_epi32(v2, DCT_CONST_BITS);
+          const __m128i w3 = _mm_srai_epi32(v3, DCT_CONST_BITS);
+          // Combine
+          res15 = _mm_packs_epi32(w0, w1);
+          res07 = _mm_packs_epi32(w2, w3);
+        }
+      }
+      // Transpose the results, do it as two 8x8 transposes.
+      {
+        // 00 01 02 03 04 05 06 07
+        // 10 11 12 13 14 15 16 17
+        // 20 21 22 23 24 25 26 27
+        // 30 31 32 33 34 35 36 37
+        // 40 41 42 43 44 45 46 47
+        // 50 51 52 53 54 55 56 57
+        // 60 61 62 63 64 65 66 67
+        // 70 71 72 73 74 75 76 77
+        const __m128i tr0_0 = _mm_unpacklo_epi16(res00, res01);
+        const __m128i tr0_1 = _mm_unpacklo_epi16(res02, res03);
+        const __m128i tr0_2 = _mm_unpackhi_epi16(res00, res01);
+        const __m128i tr0_3 = _mm_unpackhi_epi16(res02, res03);
+        const __m128i tr0_4 = _mm_unpacklo_epi16(res04, res05);
+        const __m128i tr0_5 = _mm_unpacklo_epi16(res06, res07);
+        const __m128i tr0_6 = _mm_unpackhi_epi16(res04, res05);
+        const __m128i tr0_7 = _mm_unpackhi_epi16(res06, res07);
+        // 00 10 01 11 02 12 03 13
+        // 20 30 21 31 22 32 23 33
+        // 04 14 05 15 06 16 07 17
+        // 24 34 25 35 26 36 27 37
+        // 40 50 41 51 42 52 43 53
+        // 60 70 61 71 62 72 63 73
+        // 54 54 55 55 56 56 57 57
+        // 64 74 65 75 66 76 67 77
+        const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
+        const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3);
+        const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+        const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
+        const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
+        const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
+        const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
+        const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);
+        // 00 10 20 30 01 11 21 31
+        // 40 50 60 70 41 51 61 71
+        // 02 12 22 32 03 13 23 33
+        // 42 52 62 72 43 53 63 73
+        // 04 14 24 34 05 15 21 36
+        // 44 54 64 74 45 55 61 76
+        // 06 16 26 36 07 17 27 37
+        // 46 56 66 76 47 57 67 77
+        const __m128i tr2_0 = _mm_unpacklo_epi64(tr1_0, tr1_4);
+        const __m128i tr2_1 = _mm_unpackhi_epi64(tr1_0, tr1_4);
+        const __m128i tr2_2 = _mm_unpacklo_epi64(tr1_2, tr1_6);
+        const __m128i tr2_3 = _mm_unpackhi_epi64(tr1_2, tr1_6);
+        const __m128i tr2_4 = _mm_unpacklo_epi64(tr1_1, tr1_5);
+        const __m128i tr2_5 = _mm_unpackhi_epi64(tr1_1, tr1_5);
+        const __m128i tr2_6 = _mm_unpacklo_epi64(tr1_3, tr1_7);
+        const __m128i tr2_7 = _mm_unpackhi_epi64(tr1_3, tr1_7);
+        // 00 10 20 30 40 50 60 70
+        // 01 11 21 31 41 51 61 71
+        // 02 12 22 32 42 52 62 72
+        // 03 13 23 33 43 53 63 73
+        // 04 14 24 34 44 54 64 74
+        // 05 15 25 35 45 55 65 75
+        // 06 16 26 36 46 56 66 76
+        // 07 17 27 37 47 57 67 77
+        _mm_storeu_si128((__m128i *)(out + 0 * 16), tr2_0);
+        _mm_storeu_si128((__m128i *)(out + 1 * 16), tr2_1);
+        _mm_storeu_si128((__m128i *)(out + 2 * 16), tr2_2);
+        _mm_storeu_si128((__m128i *)(out + 3 * 16), tr2_3);
+        _mm_storeu_si128((__m128i *)(out + 4 * 16), tr2_4);
+        _mm_storeu_si128((__m128i *)(out + 5 * 16), tr2_5);
+        _mm_storeu_si128((__m128i *)(out + 6 * 16), tr2_6);
+        _mm_storeu_si128((__m128i *)(out + 7 * 16), tr2_7);
+      }
+      {
+        // 00 01 02 03 04 05 06 07
+        // 10 11 12 13 14 15 16 17
+        // 20 21 22 23 24 25 26 27
+        // 30 31 32 33 34 35 36 37
+        // 40 41 42 43 44 45 46 47
+        // 50 51 52 53 54 55 56 57
+        // 60 61 62 63 64 65 66 67
+        // 70 71 72 73 74 75 76 77
+        const __m128i tr0_0 = _mm_unpacklo_epi16(res08, res09);
+        const __m128i tr0_1 = _mm_unpacklo_epi16(res10, res11);
+        const __m128i tr0_2 = _mm_unpackhi_epi16(res08, res09);
+        const __m128i tr0_3 = _mm_unpackhi_epi16(res10, res11);
+        const __m128i tr0_4 = _mm_unpacklo_epi16(res12, res13);
+        const __m128i tr0_5 = _mm_unpacklo_epi16(res14, res15);
+        const __m128i tr0_6 = _mm_unpackhi_epi16(res12, res13);
+        const __m128i tr0_7 = _mm_unpackhi_epi16(res14, res15);
+        // 00 10 01 11 02 12 03 13
+        // 20 30 21 31 22 32 23 33
+        // 04 14 05 15 06 16 07 17
+        // 24 34 25 35 26 36 27 37
+        // 40 50 41 51 42 52 43 53
+        // 60 70 61 71 62 72 63 73
+        // 54 54 55 55 56 56 57 57
+        // 64 74 65 75 66 76 67 77
+        const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
+        const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3);
+        const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+        const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
+        const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
+        const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
+        const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
+        const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);
+        // 00 10 20 30 01 11 21 31
+        // 40 50 60 70 41 51 61 71
+        // 02 12 22 32 03 13 23 33
+        // 42 52 62 72 43 53 63 73
+        // 04 14 24 34 05 15 21 36
+        // 44 54 64 74 45 55 61 76
+        // 06 16 26 36 07 17 27 37
+        // 46 56 66 76 47 57 67 77
+        const __m128i tr2_0 = _mm_unpacklo_epi64(tr1_0, tr1_4);
+        const __m128i tr2_1 = _mm_unpackhi_epi64(tr1_0, tr1_4);
+        const __m128i tr2_2 = _mm_unpacklo_epi64(tr1_2, tr1_6);
+        const __m128i tr2_3 = _mm_unpackhi_epi64(tr1_2, tr1_6);
+        const __m128i tr2_4 = _mm_unpacklo_epi64(tr1_1, tr1_5);
+        const __m128i tr2_5 = _mm_unpackhi_epi64(tr1_1, tr1_5);
+        const __m128i tr2_6 = _mm_unpacklo_epi64(tr1_3, tr1_7);
+        const __m128i tr2_7 = _mm_unpackhi_epi64(tr1_3, tr1_7);
+        // 00 10 20 30 40 50 60 70
+        // 01 11 21 31 41 51 61 71
+        // 02 12 22 32 42 52 62 72
+        // 03 13 23 33 43 53 63 73
+        // 04 14 24 34 44 54 64 74
+        // 05 15 25 35 45 55 65 75
+        // 06 16 26 36 46 56 66 76
+        // 07 17 27 37 47 57 67 77
+        // Store results
+        _mm_storeu_si128((__m128i *)(out + 8 + 0 * 16), tr2_0);
+        _mm_storeu_si128((__m128i *)(out + 8 + 1 * 16), tr2_1);
+        _mm_storeu_si128((__m128i *)(out + 8 + 2 * 16), tr2_2);
+        _mm_storeu_si128((__m128i *)(out + 8 + 3 * 16), tr2_3);
+        _mm_storeu_si128((__m128i *)(out + 8 + 4 * 16), tr2_4);
+        _mm_storeu_si128((__m128i *)(out + 8 + 5 * 16), tr2_5);
+        _mm_storeu_si128((__m128i *)(out + 8 + 6 * 16), tr2_6);
+        _mm_storeu_si128((__m128i *)(out + 8 + 7 * 16), tr2_7);
+      }
+      out += 8*16;
+    }
+    // Setup in/out for next pass.
+    in = intermediate;
+    out = output;
+  }
+}
diff --git a/libvpx/vp9/encoder/x86/vp9_encodeopt.asm b/libvpx/vp9/encoder/x86/vp9_encodeopt.asm
new file mode 100644
index 0000000..734cb61
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_encodeopt.asm
@@ -0,0 +1,125 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;int vp9_block_error_xmm(short *coeff_ptr,  short *dcoef_ptr)
+global sym(vp9_block_error_xmm) PRIVATE
+sym(vp9_block_error_xmm):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    push rsi
+    push rdi
+    ; end prologue
+
+        mov         rsi,        arg(0) ;coeff_ptr
+        mov         rdi,        arg(1) ;dcoef_ptr
+
+        movdqa      xmm0,       [rsi]
+        movdqa      xmm1,       [rdi]
+
+        movdqa      xmm2,       [rsi+16]
+        movdqa      xmm3,       [rdi+16]
+
+        psubw       xmm0,       xmm1
+        psubw       xmm2,       xmm3
+
+        pmaddwd     xmm0,       xmm0
+        pmaddwd     xmm2,       xmm2
+
+        paddd       xmm0,       xmm2
+
+        pxor        xmm5,       xmm5
+        movdqa      xmm1,       xmm0
+
+        punpckldq   xmm0,       xmm5
+        punpckhdq   xmm1,       xmm5
+
+        paddd       xmm0,       xmm1
+        movdqa      xmm1,       xmm0
+
+        psrldq      xmm0,       8
+        paddd       xmm0,       xmm1
+
+        movq        rax,        xmm0
+
+    pop rdi
+    pop rsi
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;int vp9_block_error_mmx(short *coeff_ptr,  short *dcoef_ptr)
+global sym(vp9_block_error_mmx) PRIVATE
+sym(vp9_block_error_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 2
+    push rsi
+    push rdi
+    ; end prolog
+
+
+        mov         rsi,        arg(0) ;coeff_ptr
+        pxor        mm7,        mm7
+
+        mov         rdi,        arg(1) ;dcoef_ptr
+        movq        mm3,        [rsi]
+
+        movq        mm4,        [rdi]
+        movq        mm5,        [rsi+8]
+
+        movq        mm6,        [rdi+8]
+        pxor        mm1,        mm1 ; from movd mm1, dc ; dc =0
+
+        movq        mm2,        mm7
+        psubw       mm5,        mm6
+
+        por         mm1,        mm2
+        pmaddwd     mm5,        mm5
+
+        pcmpeqw     mm1,        mm7
+        psubw       mm3,        mm4
+
+        pand        mm1,        mm3
+        pmaddwd     mm1,        mm1
+
+        paddd       mm1,        mm5
+        movq        mm3,        [rsi+16]
+
+        movq        mm4,        [rdi+16]
+        movq        mm5,        [rsi+24]
+
+        movq        mm6,        [rdi+24]
+        psubw       mm5,        mm6
+
+        pmaddwd     mm5,        mm5
+        psubw       mm3,        mm4
+
+        pmaddwd     mm3,        mm3
+        paddd       mm3,        mm5
+
+        paddd       mm1,        mm3
+        movq        mm0,        mm1
+
+        psrlq       mm1,        32
+        paddd       mm0,        mm1
+
+        movq        rax,        mm0
+
+    pop rdi
+    pop rsi
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/encoder/x86/vp9_fwalsh_sse2.asm b/libvpx/vp9/encoder/x86/vp9_fwalsh_sse2.asm
new file mode 100644
index 0000000..7bee9ef
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_fwalsh_sse2.asm
@@ -0,0 +1,164 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_short_walsh4x4_sse2(short *input, short *output, int pitch)
+global sym(vp9_short_walsh4x4_sse2) PRIVATE
+sym(vp9_short_walsh4x4_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 3
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov     rsi, arg(0)           ; input
+    mov     rdi, arg(1)           ; output
+    movsxd  rdx, dword ptr arg(2) ; pitch
+
+    ; first for loop
+    movq    xmm0, MMWORD PTR [rsi]           ; load input
+    movq    xmm1, MMWORD PTR [rsi + rdx]
+    lea     rsi,  [rsi + rdx*2]
+    movq    xmm2, MMWORD PTR [rsi]
+    movq    xmm3, MMWORD PTR [rsi + rdx]
+
+    punpcklwd xmm0,  xmm1
+    punpcklwd xmm2,  xmm3
+
+    movdqa    xmm1, xmm0
+    punpckldq xmm0, xmm2           ; ip[1] ip[0]
+    punpckhdq xmm1, xmm2           ; ip[3] ip[2]
+
+    movdqa    xmm2, xmm0
+    paddw     xmm0, xmm1
+    psubw     xmm2, xmm1
+
+    psllw     xmm0, 2              ; d1  a1
+    psllw     xmm2, 2              ; c1  b1
+
+    movdqa    xmm1, xmm0
+    punpcklqdq xmm0, xmm2          ; b1  a1
+    punpckhqdq xmm1, xmm2          ; c1  d1
+
+    pxor      xmm6, xmm6
+    movq      xmm6, xmm0
+    pxor      xmm7, xmm7
+    pcmpeqw   xmm7, xmm6
+    paddw     xmm7, [GLOBAL(c1)]
+
+    movdqa    xmm2, xmm0
+    paddw     xmm0, xmm1           ; b1+c1  a1+d1
+    psubw     xmm2, xmm1           ; b1-c1  a1-d1
+    paddw     xmm0, xmm7           ; b1+c1  a1+d1+(a1!=0)
+
+    ; second for loop
+    ; input: 13  9  5  1 12  8  4  0 (xmm0)
+    ;        14 10  6  2 15 11  7  3 (xmm2)
+    ; after shuffle:
+    ;        13  5  9  1 12  4  8  0 (xmm0)
+    ;        14  6 10  2 15  7 11  3 (xmm1)
+    pshuflw   xmm3, xmm0, 0xd8
+    pshufhw   xmm0, xmm3, 0xd8
+    pshuflw   xmm3, xmm2, 0xd8
+    pshufhw   xmm1, xmm3, 0xd8
+
+    movdqa    xmm2, xmm0
+    pmaddwd   xmm0, [GLOBAL(c1)]    ; d11 a11 d10 a10
+    pmaddwd   xmm2, [GLOBAL(cn1)]   ; c11 b11 c10 b10
+    movdqa    xmm3, xmm1
+    pmaddwd   xmm1, [GLOBAL(c1)]    ; d12 a12 d13 a13
+    pmaddwd   xmm3, [GLOBAL(cn1)]   ; c12 b12 c13 b13
+
+    pshufd    xmm4, xmm0, 0xd8      ; d11 d10 a11 a10
+    pshufd    xmm5, xmm2, 0xd8      ; c11 c10 b11 b10
+    pshufd    xmm6, xmm1, 0x72      ; d13 d12 a13 a12
+    pshufd    xmm7, xmm3, 0x72      ; c13 c12 b13 b12
+
+    movdqa    xmm0, xmm4
+    punpcklqdq xmm0, xmm5           ; b11 b10 a11 a10
+    punpckhqdq xmm4, xmm5           ; c11 c10 d11 d10
+    movdqa    xmm1, xmm6
+    punpcklqdq xmm1, xmm7           ; b13 b12 a13 a12
+    punpckhqdq xmm6, xmm7           ; c13 c12 d13 d12
+
+    movdqa    xmm2, xmm0
+    paddd     xmm0, xmm4            ; b21 b20 a21 a20
+    psubd     xmm2, xmm4            ; c21 c20 d21 d20
+    movdqa    xmm3, xmm1
+    paddd     xmm1, xmm6            ; b23 b22 a23 a22
+    psubd     xmm3, xmm6            ; c23 c22 d23 d22
+
+    pxor      xmm4, xmm4
+    movdqa    xmm5, xmm4
+    pcmpgtd   xmm4, xmm0
+    pcmpgtd   xmm5, xmm2
+    pand      xmm4, [GLOBAL(cd1)]
+    pand      xmm5, [GLOBAL(cd1)]
+
+    pxor      xmm6, xmm6
+    movdqa    xmm7, xmm6
+    pcmpgtd   xmm6, xmm1
+    pcmpgtd   xmm7, xmm3
+    pand      xmm6, [GLOBAL(cd1)]
+    pand      xmm7, [GLOBAL(cd1)]
+
+    paddd     xmm0, xmm4
+    paddd     xmm2, xmm5
+    paddd     xmm0, [GLOBAL(cd3)]
+    paddd     xmm2, [GLOBAL(cd3)]
+    paddd     xmm1, xmm6
+    paddd     xmm3, xmm7
+    paddd     xmm1, [GLOBAL(cd3)]
+    paddd     xmm3, [GLOBAL(cd3)]
+
+    psrad     xmm0, 3
+    psrad     xmm1, 3
+    psrad     xmm2, 3
+    psrad     xmm3, 3
+    movdqa    xmm4, xmm0
+    punpcklqdq xmm0, xmm1           ; a23 a22 a21 a20
+    punpckhqdq xmm4, xmm1           ; b23 b22 b21 b20
+    movdqa    xmm5, xmm2
+    punpckhqdq xmm2, xmm3           ; c23 c22 c21 c20
+    punpcklqdq xmm5, xmm3           ; d23 d22 d21 d20
+
+    packssdw  xmm0, xmm4            ; b23 b22 b21 b20 a23 a22 a21 a20
+    packssdw  xmm2, xmm5            ; d23 d22 d21 d20 c23 c22 c21 c20
+
+    movdqa  XMMWORD PTR [rdi], xmm0
+    movdqa  XMMWORD PTR [rdi + 16], xmm2
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+c1:
+    dw 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001
+align 16
+cn1:
+    dw 0x0001, 0xffff, 0x0001, 0xffff, 0x0001, 0xffff, 0x0001, 0xffff
+align 16
+cd1:
+    dd 0x00000001, 0x00000001, 0x00000001, 0x00000001
+align 16
+cd3:
+    dd 0x00000003, 0x00000003, 0x00000003, 0x00000003
diff --git a/libvpx/vp9/encoder/x86/vp9_mcomp_x86.h b/libvpx/vp9/encoder/x86/vp9_mcomp_x86.h
new file mode 100644
index 0000000..ca80b8b
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_mcomp_x86.h
@@ -0,0 +1,40 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef VP9_ENCODER_X86_VP9_MCOMP_X86_H_
+#define VP9_ENCODER_X86_VP9_MCOMP_X86_H_
+
+#if HAVE_SSE3
+#if !CONFIG_RUNTIME_CPU_DETECT
+
+#undef  vp9_search_full_search
+#define vp9_search_full_search vp9_full_search_sadx3
+
+#undef  vp9_search_refining_search
+#define vp9_search_refining_search vp9_refining_search_sadx4
+
+#undef  vp9_search_diamond_search
+#define vp9_search_diamond_search vp9_diamond_search_sadx4
+
+#endif
+#endif
+
+#if HAVE_SSE4_1
+#if !CONFIG_RUNTIME_CPU_DETECT
+
+#undef  vp9_search_full_search
+#define vp9_search_full_search vp9_full_search_sadx8
+
+#endif
+#endif
+
+#endif
+
diff --git a/libvpx/vp9/encoder/x86/vp9_sad4d_sse2.asm b/libvpx/vp9/encoder/x86/vp9_sad4d_sse2.asm
new file mode 100644
index 0000000..b493628
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad4d_sse2.asm
@@ -0,0 +1,231 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION .text
+
+; PROCESS_4x2x4 first, off_{first,second}_{src,ref}, advance_at_end
+%macro PROCESS_4x2x4 5-6 0
+  movd                  m0, [srcq +%2]
+%if %1 == 1
+  movd                  m6, [ref1q+%3]
+  movd                  m4, [ref2q+%3]
+  movd                  m7, [ref3q+%3]
+  movd                  m5, [ref4q+%3]
+  punpckldq             m0, [srcq +%4]
+  punpckldq             m6, [ref1q+%5]
+  punpckldq             m4, [ref2q+%5]
+  punpckldq             m7, [ref3q+%5]
+  punpckldq             m5, [ref4q+%5]
+  psadbw                m6, m0
+  psadbw                m4, m0
+  psadbw                m7, m0
+  psadbw                m5, m0
+  punpckldq             m6, m4
+  punpckldq             m7, m5
+%else
+  movd                  m1, [ref1q+%3]
+  movd                  m2, [ref2q+%3]
+  movd                  m3, [ref3q+%3]
+  movd                  m4, [ref4q+%3]
+  punpckldq             m0, [srcq +%4]
+  punpckldq             m1, [ref1q+%5]
+  punpckldq             m2, [ref2q+%5]
+  punpckldq             m3, [ref3q+%5]
+  punpckldq             m4, [ref4q+%5]
+  psadbw                m1, m0
+  psadbw                m2, m0
+  psadbw                m3, m0
+  psadbw                m4, m0
+  punpckldq             m1, m2
+  punpckldq             m3, m4
+  paddd                 m6, m1
+  paddd                 m7, m3
+%endif
+%if %6 == 1
+  lea                 srcq, [srcq +src_strideq*2]
+  lea                ref1q, [ref1q+ref_strideq*2]
+  lea                ref2q, [ref2q+ref_strideq*2]
+  lea                ref3q, [ref3q+ref_strideq*2]
+  lea                ref4q, [ref4q+ref_strideq*2]
+%endif
+%endmacro
+
+; PROCESS_8x2x4 first, off_{first,second}_{src,ref}, advance_at_end
+%macro PROCESS_8x2x4 5-6 0
+  movh                  m0, [srcq +%2]
+%if %1 == 1
+  movh                  m4, [ref1q+%3]
+  movh                  m5, [ref2q+%3]
+  movh                  m6, [ref3q+%3]
+  movh                  m7, [ref4q+%3]
+  movhps                m0, [srcq +%4]
+  movhps                m4, [ref1q+%5]
+  movhps                m5, [ref2q+%5]
+  movhps                m6, [ref3q+%5]
+  movhps                m7, [ref4q+%5]
+  psadbw                m4, m0
+  psadbw                m5, m0
+  psadbw                m6, m0
+  psadbw                m7, m0
+%else
+  movh                  m1, [ref1q+%3]
+  movh                  m2, [ref2q+%3]
+  movh                  m3, [ref3q+%3]
+  movhps                m0, [srcq +%4]
+  movhps                m1, [ref1q+%5]
+  movhps                m2, [ref2q+%5]
+  movhps                m3, [ref3q+%5]
+  psadbw                m1, m0
+  psadbw                m2, m0
+  psadbw                m3, m0
+  paddd                 m4, m1
+  movh                  m1, [ref4q+%3]
+  movhps                m1, [ref4q+%5]
+  paddd                 m5, m2
+  paddd                 m6, m3
+  psadbw                m1, m0
+  paddd                 m7, m1
+%endif
+%if %6 == 1
+  lea                 srcq, [srcq +src_strideq*2]
+  lea                ref1q, [ref1q+ref_strideq*2]
+  lea                ref2q, [ref2q+ref_strideq*2]
+  lea                ref3q, [ref3q+ref_strideq*2]
+  lea                ref4q, [ref4q+ref_strideq*2]
+%endif
+%endmacro
+
+; PROCESS_16x2x4 first, off_{first,second}_{src,ref}, advance_at_end
+%macro PROCESS_16x2x4 5-6 0
+  ; 1st 16 px
+  mova                  m0, [srcq +%2]
+%if %1 == 1
+  movu                  m4, [ref1q+%3]
+  movu                  m5, [ref2q+%3]
+  movu                  m6, [ref3q+%3]
+  movu                  m7, [ref4q+%3]
+  psadbw                m4, m0
+  psadbw                m5, m0
+  psadbw                m6, m0
+  psadbw                m7, m0
+%else
+  movu                  m1, [ref1q+%3]
+  movu                  m2, [ref2q+%3]
+  movu                  m3, [ref3q+%3]
+  psadbw                m1, m0
+  psadbw                m2, m0
+  psadbw                m3, m0
+  paddd                 m4, m1
+  movu                  m1, [ref4q+%3]
+  paddd                 m5, m2
+  paddd                 m6, m3
+  psadbw                m1, m0
+  paddd                 m7, m1
+%endif
+
+  ; 2nd 16 px
+  mova                  m0, [srcq +%4]
+  movu                  m1, [ref1q+%5]
+  movu                  m2, [ref2q+%5]
+  movu                  m3, [ref3q+%5]
+  psadbw                m1, m0
+  psadbw                m2, m0
+  psadbw                m3, m0
+  paddd                 m4, m1
+  movu                  m1, [ref4q+%5]
+  paddd                 m5, m2
+  paddd                 m6, m3
+%if %6 == 1
+  lea                 srcq, [srcq +src_strideq*2]
+  lea                ref1q, [ref1q+ref_strideq*2]
+  lea                ref2q, [ref2q+ref_strideq*2]
+  lea                ref3q, [ref3q+ref_strideq*2]
+  lea                ref4q, [ref4q+ref_strideq*2]
+%endif
+  psadbw                m1, m0
+  paddd                 m7, m1
+%endmacro
+
+; PROCESS_32x2x4 first, off_{first,second}_{src,ref}, advance_at_end
+%macro PROCESS_32x2x4 5-6 0
+  PROCESS_16x2x4 %1, %2, %3, %2 + 16, %3 + 16
+  PROCESS_16x2x4  0, %4, %5, %4 + 16, %5 + 16, %6
+%endmacro
+
+; PROCESS_64x2x4 first, off_{first,second}_{src,ref}, advance_at_end
+%macro PROCESS_64x2x4 5-6 0
+  PROCESS_32x2x4 %1, %2, %3, %2 + 32, %3 + 32
+  PROCESS_32x2x4  0, %4, %5, %4 + 32, %5 + 32, %6
+%endmacro
+
+; void vp9_sadNxNx4d_sse2(uint8_t *src,    int src_stride,
+;                         uint8_t *ref[4], int ref_stride,
+;                         unsigned int res[4]);
+; where NxN = 64x64, 32x32, 16x16, 16x8, 8x16 or 8x8
+%macro SADNXN4D 2
+%if UNIX64
+cglobal sad%1x%2x4d, 5, 8, 8, src, src_stride, ref1, ref_stride, \
+                              res, ref2, ref3, ref4
+%else
+cglobal sad%1x%2x4d, 4, 7, 8, src, src_stride, ref1, ref_stride, \
+                              ref2, ref3, ref4
+%endif
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  mov                ref2q, [ref1q+gprsize*1]
+  mov                ref3q, [ref1q+gprsize*2]
+  mov                ref4q, [ref1q+gprsize*3]
+  mov                ref1q, [ref1q+gprsize*0]
+
+  PROCESS_%1x2x4 1, 0, 0, src_strideq, ref_strideq, 1
+%rep (%2-4)/2
+  PROCESS_%1x2x4 0, 0, 0, src_strideq, ref_strideq, 1
+%endrep
+  PROCESS_%1x2x4 0, 0, 0, src_strideq, ref_strideq, 0
+
+%if mmsize == 16
+  pslldq                m5, 4
+  pslldq                m7, 4
+  por                   m4, m5
+  por                   m6, m7
+  mova                  m5, m4
+  mova                  m7, m6
+  punpcklqdq            m4, m6
+  punpckhqdq            m5, m7
+  movifnidn             r4, r4mp
+  paddd                 m4, m5
+  movu                [r4], m4
+  RET
+%else
+  movifnidn             r4, r4mp
+  movq               [r4+0], m6
+  movq               [r4+8], m7
+  RET
+%endif
+%endmacro
+
+INIT_XMM sse2
+SADNXN4D 64, 64
+SADNXN4D 64, 32
+SADNXN4D 32, 64
+SADNXN4D 32, 32
+SADNXN4D 32, 16
+SADNXN4D 16, 32
+SADNXN4D 16, 16
+SADNXN4D 16,  8
+SADNXN4D  8, 16
+SADNXN4D  8,  8
+SADNXN4D  8,  4
+
+INIT_MMX sse
+SADNXN4D  4,  8
+SADNXN4D  4,  4
diff --git a/libvpx/vp9/encoder/x86/vp9_sad_mmx.asm b/libvpx/vp9/encoder/x86/vp9_sad_mmx.asm
new file mode 100644
index 0000000..32fdd23
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad_mmx.asm
@@ -0,0 +1,427 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+global sym(vp9_sad16x16_mmx) PRIVATE
+global sym(vp9_sad8x16_mmx) PRIVATE
+global sym(vp9_sad8x8_mmx) PRIVATE
+global sym(vp9_sad4x4_mmx) PRIVATE
+global sym(vp9_sad16x8_mmx) PRIVATE
+
+;unsigned int vp9_sad16x16_mmx(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride)
+sym(vp9_sad16x16_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        lea             rcx,        [rsi+rax*8]
+
+        lea             rcx,        [rcx+rax*8]
+        pxor            mm7,        mm7
+
+        pxor            mm6,        mm6
+
+.x16x16sad_mmx_loop:
+
+        movq            mm0,        QWORD PTR [rsi]
+        movq            mm2,        QWORD PTR [rsi+8]
+
+        movq            mm1,        QWORD PTR [rdi]
+        movq            mm3,        QWORD PTR [rdi+8]
+
+        movq            mm4,        mm0
+        movq            mm5,        mm2
+
+        psubusb         mm0,        mm1
+        psubusb         mm1,        mm4
+
+        psubusb         mm2,        mm3
+        psubusb         mm3,        mm5
+
+        por             mm0,        mm1
+        por             mm2,        mm3
+
+        movq            mm1,        mm0
+        movq            mm3,        mm2
+
+        punpcklbw       mm0,        mm6
+        punpcklbw       mm2,        mm6
+
+        punpckhbw       mm1,        mm6
+        punpckhbw       mm3,        mm6
+
+        paddw           mm0,        mm2
+        paddw           mm1,        mm3
+
+
+        lea             rsi,        [rsi+rax]
+        add             rdi,        rdx
+
+        paddw           mm7,        mm0
+        paddw           mm7,        mm1
+
+        cmp             rsi,        rcx
+        jne             .x16x16sad_mmx_loop
+
+
+        movq            mm0,        mm7
+
+        punpcklwd       mm0,        mm6
+        punpckhwd       mm7,        mm6
+
+        paddw           mm0,        mm7
+        movq            mm7,        mm0
+
+
+        psrlq           mm0,        32
+        paddw           mm7,        mm0
+
+        movq            rax,        mm7
+
+    pop rdi
+    pop rsi
+    mov rsp, rbp
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_sad8x16_mmx(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride)
+sym(vp9_sad8x16_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        lea             rcx,        [rsi+rax*8]
+
+        lea             rcx,        [rcx+rax*8]
+        pxor            mm7,        mm7
+
+        pxor            mm6,        mm6
+
+.x8x16sad_mmx_loop:
+
+        movq            mm0,        QWORD PTR [rsi]
+        movq            mm1,        QWORD PTR [rdi]
+
+        movq            mm2,        mm0
+        psubusb         mm0,        mm1
+
+        psubusb         mm1,        mm2
+        por             mm0,        mm1
+
+        movq            mm2,        mm0
+        punpcklbw       mm0,        mm6
+
+        punpckhbw       mm2,        mm6
+        lea             rsi,        [rsi+rax]
+
+        add             rdi,        rdx
+        paddw           mm7,        mm0
+
+        paddw           mm7,        mm2
+        cmp             rsi,        rcx
+
+        jne             .x8x16sad_mmx_loop
+
+        movq            mm0,        mm7
+        punpcklwd       mm0,        mm6
+
+        punpckhwd       mm7,        mm6
+        paddw           mm0,        mm7
+
+        movq            mm7,        mm0
+        psrlq           mm0,        32
+
+        paddw           mm7,        mm0
+        movq            rax,        mm7
+
+    pop rdi
+    pop rsi
+    mov rsp, rbp
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_sad8x8_mmx(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride)
+sym(vp9_sad8x8_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        lea             rcx,        [rsi+rax*8]
+        pxor            mm7,        mm7
+
+        pxor            mm6,        mm6
+
+.x8x8sad_mmx_loop:
+
+        movq            mm0,        QWORD PTR [rsi]
+        movq            mm1,        QWORD PTR [rdi]
+
+        movq            mm2,        mm0
+        psubusb         mm0,        mm1
+
+        psubusb         mm1,        mm2
+        por             mm0,        mm1
+
+        movq            mm2,        mm0
+        punpcklbw       mm0,        mm6
+
+        punpckhbw       mm2,        mm6
+        paddw           mm0,        mm2
+
+        lea             rsi,       [rsi+rax]
+        add             rdi,        rdx
+
+        paddw           mm7,       mm0
+        cmp             rsi,        rcx
+
+        jne             .x8x8sad_mmx_loop
+
+        movq            mm0,        mm7
+        punpcklwd       mm0,        mm6
+
+        punpckhwd       mm7,        mm6
+        paddw           mm0,        mm7
+
+        movq            mm7,        mm0
+        psrlq           mm0,        32
+
+        paddw           mm7,        mm0
+        movq            rax,        mm7
+
+    pop rdi
+    pop rsi
+    mov rsp, rbp
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_sad4x4_mmx(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride)
+sym(vp9_sad4x4_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        movd            mm0,        DWORD PTR [rsi]
+        movd            mm1,        DWORD PTR [rdi]
+
+        movd            mm2,        DWORD PTR [rsi+rax]
+        movd            mm3,        DWORD PTR [rdi+rdx]
+
+        punpcklbw       mm0,        mm2
+        punpcklbw       mm1,        mm3
+
+        movq            mm2,        mm0
+        psubusb         mm0,        mm1
+
+        psubusb         mm1,        mm2
+        por             mm0,        mm1
+
+        movq            mm2,        mm0
+        pxor            mm3,        mm3
+
+        punpcklbw       mm0,        mm3
+        punpckhbw       mm2,        mm3
+
+        paddw           mm0,        mm2
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        movd            mm4,        DWORD PTR [rsi]
+        movd            mm5,        DWORD PTR [rdi]
+
+        movd            mm6,        DWORD PTR [rsi+rax]
+        movd            mm7,        DWORD PTR [rdi+rdx]
+
+        punpcklbw       mm4,        mm6
+        punpcklbw       mm5,        mm7
+
+        movq            mm6,        mm4
+        psubusb         mm4,        mm5
+
+        psubusb         mm5,        mm6
+        por             mm4,        mm5
+
+        movq            mm5,        mm4
+        punpcklbw       mm4,        mm3
+
+        punpckhbw       mm5,        mm3
+        paddw           mm4,        mm5
+
+        paddw           mm0,        mm4
+        movq            mm1,        mm0
+
+        punpcklwd       mm0,        mm3
+        punpckhwd       mm1,        mm3
+
+        paddw           mm0,        mm1
+        movq            mm1,        mm0
+
+        psrlq           mm0,        32
+        paddw           mm0,        mm1
+
+        movq            rax,        mm0
+
+    pop rdi
+    pop rsi
+    mov rsp, rbp
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_sad16x8_mmx(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride)
+sym(vp9_sad16x8_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        lea             rcx,        [rsi+rax*8]
+        pxor            mm7,        mm7
+
+        pxor            mm6,        mm6
+
+.x16x8sad_mmx_loop:
+
+        movq            mm0,       [rsi]
+        movq            mm1,       [rdi]
+
+        movq            mm2,        [rsi+8]
+        movq            mm3,        [rdi+8]
+
+        movq            mm4,        mm0
+        movq            mm5,        mm2
+
+        psubusb         mm0,        mm1
+        psubusb         mm1,        mm4
+
+        psubusb         mm2,        mm3
+        psubusb         mm3,        mm5
+
+        por             mm0,        mm1
+        por             mm2,        mm3
+
+        movq            mm1,        mm0
+        movq            mm3,        mm2
+
+        punpcklbw       mm0,        mm6
+        punpckhbw       mm1,        mm6
+
+        punpcklbw       mm2,        mm6
+        punpckhbw       mm3,        mm6
+
+
+        paddw           mm0,        mm2
+        paddw           mm1,        mm3
+
+        paddw           mm0,        mm1
+        lea             rsi,        [rsi+rax]
+
+        add             rdi,        rdx
+        paddw           mm7,        mm0
+
+        cmp             rsi,        rcx
+        jne             .x16x8sad_mmx_loop
+
+        movq            mm0,        mm7
+        punpcklwd       mm0,        mm6
+
+        punpckhwd       mm7,        mm6
+        paddw           mm0,        mm7
+
+        movq            mm7,        mm0
+        psrlq           mm0,        32
+
+        paddw           mm7,        mm0
+        movq            rax,        mm7
+
+    pop rdi
+    pop rsi
+    mov rsp, rbp
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/encoder/x86/vp9_sad_sse2.asm b/libvpx/vp9/encoder/x86/vp9_sad_sse2.asm
new file mode 100644
index 0000000..8fb7d41
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad_sse2.asm
@@ -0,0 +1,211 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION .text
+
+; unsigned int vp9_sad64x64_sse2(uint8_t *src, int src_stride,
+;                                uint8_t *ref, int ref_stride);
+%macro SAD64XN 1
+cglobal sad64x%1, 4, 5, 5, src, src_stride, ref, ref_stride, n_rows
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  mov              n_rowsd, %1
+  pxor                  m0, m0
+.loop:
+  movu                  m1, [refq]
+  movu                  m2, [refq+16]
+  movu                  m3, [refq+32]
+  movu                  m4, [refq+48]
+  psadbw                m1, [srcq]
+  psadbw                m2, [srcq+16]
+  psadbw                m3, [srcq+32]
+  psadbw                m4, [srcq+48]
+  paddd                 m1, m2
+  paddd                 m3, m4
+  add                 refq, ref_strideq
+  paddd                 m0, m1
+  add                 srcq, src_strideq
+  paddd                 m0, m3
+  dec              n_rowsd
+  jg .loop
+
+  movhlps               m1, m0
+  paddd                 m0, m1
+  movd                 eax, m0
+  RET
+%endmacro
+
+INIT_XMM sse2
+SAD64XN 64 ; sad64x64_sse2
+SAD64XN 32 ; sad64x32_sse2
+
+; unsigned int vp9_sad32x32_sse2(uint8_t *src, int src_stride,
+;                                uint8_t *ref, int ref_stride);
+%macro SAD32XN 1
+cglobal sad32x%1, 4, 5, 5, src, src_stride, ref, ref_stride, n_rows
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  mov              n_rowsd, %1/2
+  pxor                  m0, m0
+
+.loop:
+  movu                  m1, [refq]
+  movu                  m2, [refq+16]
+  movu                  m3, [refq+ref_strideq]
+  movu                  m4, [refq+ref_strideq+16]
+  psadbw                m1, [srcq]
+  psadbw                m2, [srcq+16]
+  psadbw                m3, [srcq+src_strideq]
+  psadbw                m4, [srcq+src_strideq+16]
+  paddd                 m1, m2
+  paddd                 m3, m4
+  lea                 refq, [refq+ref_strideq*2]
+  paddd                 m0, m1
+  lea                 srcq, [srcq+src_strideq*2]
+  paddd                 m0, m3
+  dec              n_rowsd
+  jg .loop
+
+  movhlps               m1, m0
+  paddd                 m0, m1
+  movd                 eax, m0
+  RET
+%endmacro
+
+INIT_XMM sse2
+SAD32XN 64 ; sad32x64_sse2
+SAD32XN 32 ; sad32x32_sse2
+SAD32XN 16 ; sad32x16_sse2
+
+; unsigned int vp9_sad16x{8,16}_sse2(uint8_t *src, int src_stride,
+;                                    uint8_t *ref, int ref_stride);
+%macro SAD16XN 1
+cglobal sad16x%1, 4, 7, 5, src, src_stride, ref, ref_stride, \
+                           src_stride3, ref_stride3, n_rows
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  lea         src_stride3q, [src_strideq*3]
+  lea         ref_stride3q, [ref_strideq*3]
+  mov              n_rowsd, %1/4
+  pxor                  m0, m0
+
+.loop:
+  movu                  m1, [refq]
+  movu                  m2, [refq+ref_strideq]
+  movu                  m3, [refq+ref_strideq*2]
+  movu                  m4, [refq+ref_stride3q]
+  psadbw                m1, [srcq]
+  psadbw                m2, [srcq+src_strideq]
+  psadbw                m3, [srcq+src_strideq*2]
+  psadbw                m4, [srcq+src_stride3q]
+  paddd                 m1, m2
+  paddd                 m3, m4
+  lea                 refq, [refq+ref_strideq*4]
+  paddd                 m0, m1
+  lea                 srcq, [srcq+src_strideq*4]
+  paddd                 m0, m3
+  dec              n_rowsd
+  jg .loop
+
+  movhlps               m1, m0
+  paddd                 m0, m1
+  movd                 eax, m0
+  RET
+%endmacro
+
+INIT_XMM sse2
+SAD16XN 32 ; sad16x32_sse2
+SAD16XN 16 ; sad16x16_sse2
+SAD16XN  8 ; sad16x8_sse2
+
+; unsigned int vp9_sad8x{8,16}_sse2(uint8_t *src, int src_stride,
+;                                   uint8_t *ref, int ref_stride);
+%macro SAD8XN 1
+cglobal sad8x%1, 4, 7, 5, src, src_stride, ref, ref_stride, \
+                          src_stride3, ref_stride3, n_rows
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  lea         src_stride3q, [src_strideq*3]
+  lea         ref_stride3q, [ref_strideq*3]
+  mov              n_rowsd, %1/4
+  pxor                  m0, m0
+
+.loop:
+  movh                  m1, [refq]
+  movhps                m1, [refq+ref_strideq]
+  movh                  m2, [refq+ref_strideq*2]
+  movhps                m2, [refq+ref_stride3q]
+  movh                  m3, [srcq]
+  movhps                m3, [srcq+src_strideq]
+  movh                  m4, [srcq+src_strideq*2]
+  movhps                m4, [srcq+src_stride3q]
+  psadbw                m1, m3
+  psadbw                m2, m4
+  lea                 refq, [refq+ref_strideq*4]
+  paddd                 m0, m1
+  lea                 srcq, [srcq+src_strideq*4]
+  paddd                 m0, m2
+  dec              n_rowsd
+  jg .loop
+
+  movhlps               m1, m0
+  paddd                 m0, m1
+  movd                 eax, m0
+  RET
+%endmacro
+
+INIT_XMM sse2
+SAD8XN 16 ; sad8x16_sse2
+SAD8XN  8 ; sad8x8_sse2
+SAD8XN  4 ; sad8x4_sse2
+
+; unsigned int vp9_sad4x{4, 8}_sse(uint8_t *src, int src_stride,
+;                                  uint8_t *ref, int ref_stride);
+%macro SAD4XN 1
+cglobal sad4x%1, 4, 7, 7, src, src_stride, ref, ref_stride, \
+                          src_stride3, ref_stride3, n_rows
+  movsxdifnidn src_strideq, src_strided
+  movsxdifnidn ref_strideq, ref_strided
+  lea         src_stride3q, [src_strideq*3]
+  lea         ref_stride3q, [ref_strideq*3]
+  mov              n_rowsd, %1/4
+  pxor                  m0, m0
+
+.loop:
+  movd                  m1, [refq]
+  movd                  m2, [refq+ref_strideq]
+  movd                  m3, [refq+ref_strideq*2]
+  movd                  m4, [refq+ref_stride3q]
+  punpckldq             m1, m2
+  punpckldq             m3, m4
+  movd                  m2, [srcq]
+  movd                  m5, [srcq+src_strideq]
+  movd                  m4, [srcq+src_strideq*2]
+  movd                  m6, [srcq+src_stride3q]
+  punpckldq             m2, m5
+  punpckldq             m4, m6
+  psadbw                m1, m2
+  psadbw                m3, m4
+  lea                 refq, [refq+ref_strideq*4]
+  paddd                 m0, m1
+  lea                 srcq, [srcq+src_strideq*4]
+  paddd                 m0, m3
+  dec              n_rowsd
+  jg .loop
+
+  movd                 eax, m0
+  RET
+%endmacro
+
+INIT_MMX sse
+SAD4XN  8 ; sad4x8_sse
+SAD4XN  4 ; sad4x4_sse
diff --git a/libvpx/vp9/encoder/x86/vp9_sad_sse3.asm b/libvpx/vp9/encoder/x86/vp9_sad_sse3.asm
new file mode 100644
index 0000000..2b90a5d
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad_sse3.asm
@@ -0,0 +1,378 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro STACK_FRAME_CREATE_X3 0
+%if ABI_IS_32BIT
+  %define     src_ptr       rsi
+  %define     src_stride    rax
+  %define     ref_ptr       rdi
+  %define     ref_stride    rdx
+  %define     end_ptr       rcx
+  %define     ret_var       rbx
+  %define     result_ptr    arg(4)
+  %define     max_err       arg(4)
+  %define     height        dword ptr arg(4)
+    push        rbp
+    mov         rbp,        rsp
+    push        rsi
+    push        rdi
+    push        rbx
+
+    mov         rsi,        arg(0)              ; src_ptr
+    mov         rdi,        arg(2)              ; ref_ptr
+
+    movsxd      rax,        dword ptr arg(1)    ; src_stride
+    movsxd      rdx,        dword ptr arg(3)    ; ref_stride
+%else
+  %if LIBVPX_YASM_WIN64
+    SAVE_XMM 7, u
+    %define     src_ptr     rcx
+    %define     src_stride  rdx
+    %define     ref_ptr     r8
+    %define     ref_stride  r9
+    %define     end_ptr     r10
+    %define     ret_var     r11
+    %define     result_ptr  [rsp+xmm_stack_space+8+4*8]
+    %define     max_err     [rsp+xmm_stack_space+8+4*8]
+    %define     height      dword ptr [rsp+xmm_stack_space+8+4*8]
+  %else
+    %define     src_ptr     rdi
+    %define     src_stride  rsi
+    %define     ref_ptr     rdx
+    %define     ref_stride  rcx
+    %define     end_ptr     r9
+    %define     ret_var     r10
+    %define     result_ptr  r8
+    %define     max_err     r8
+    %define     height      r8
+  %endif
+%endif
+
+%endmacro
+
+%macro STACK_FRAME_DESTROY_X3 0
+  %define     src_ptr
+  %define     src_stride
+  %define     ref_ptr
+  %define     ref_stride
+  %define     end_ptr
+  %define     ret_var
+  %define     result_ptr
+  %define     max_err
+  %define     height
+
+%if ABI_IS_32BIT
+    pop         rbx
+    pop         rdi
+    pop         rsi
+    pop         rbp
+%else
+  %if LIBVPX_YASM_WIN64
+    RESTORE_XMM
+  %endif
+%endif
+    ret
+%endmacro
+
+%macro PROCESS_16X2X3 5
+%if %1==0
+        movdqa          xmm0,       XMMWORD PTR [%2]
+        lddqu           xmm5,       XMMWORD PTR [%3]
+        lddqu           xmm6,       XMMWORD PTR [%3+1]
+        lddqu           xmm7,       XMMWORD PTR [%3+2]
+
+        psadbw          xmm5,       xmm0
+        psadbw          xmm6,       xmm0
+        psadbw          xmm7,       xmm0
+%else
+        movdqa          xmm0,       XMMWORD PTR [%2]
+        lddqu           xmm1,       XMMWORD PTR [%3]
+        lddqu           xmm2,       XMMWORD PTR [%3+1]
+        lddqu           xmm3,       XMMWORD PTR [%3+2]
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endif
+        movdqa          xmm0,       XMMWORD PTR [%2+%4]
+        lddqu           xmm1,       XMMWORD PTR [%3+%5]
+        lddqu           xmm2,       XMMWORD PTR [%3+%5+1]
+        lddqu           xmm3,       XMMWORD PTR [%3+%5+2]
+
+%if %1==0 || %1==1
+        lea             %2,         [%2+%4*2]
+        lea             %3,         [%3+%5*2]
+%endif
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endmacro
+
+%macro PROCESS_8X2X3 5
+%if %1==0
+        movq            mm0,       QWORD PTR [%2]
+        movq            mm5,       QWORD PTR [%3]
+        movq            mm6,       QWORD PTR [%3+1]
+        movq            mm7,       QWORD PTR [%3+2]
+
+        psadbw          mm5,       mm0
+        psadbw          mm6,       mm0
+        psadbw          mm7,       mm0
+%else
+        movq            mm0,       QWORD PTR [%2]
+        movq            mm1,       QWORD PTR [%3]
+        movq            mm2,       QWORD PTR [%3+1]
+        movq            mm3,       QWORD PTR [%3+2]
+
+        psadbw          mm1,       mm0
+        psadbw          mm2,       mm0
+        psadbw          mm3,       mm0
+
+        paddw           mm5,       mm1
+        paddw           mm6,       mm2
+        paddw           mm7,       mm3
+%endif
+        movq            mm0,       QWORD PTR [%2+%4]
+        movq            mm1,       QWORD PTR [%3+%5]
+        movq            mm2,       QWORD PTR [%3+%5+1]
+        movq            mm3,       QWORD PTR [%3+%5+2]
+
+%if %1==0 || %1==1
+        lea             %2,        [%2+%4*2]
+        lea             %3,        [%3+%5*2]
+%endif
+
+        psadbw          mm1,       mm0
+        psadbw          mm2,       mm0
+        psadbw          mm3,       mm0
+
+        paddw           mm5,       mm1
+        paddw           mm6,       mm2
+        paddw           mm7,       mm3
+%endmacro
+
+;void int vp9_sad16x16x3_sse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad16x16x3_sse3) PRIVATE
+sym(vp9_sad16x16x3_sse3):
+
+    STACK_FRAME_CREATE_X3
+
+        PROCESS_16X2X3 0, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 2, src_ptr, ref_ptr, src_stride, ref_stride
+
+        mov             rcx,        result_ptr
+
+        movq            xmm0,       xmm5
+        psrldq          xmm5,       8
+
+        paddw           xmm0,       xmm5
+        movd            [rcx],      xmm0
+;-
+        movq            xmm0,       xmm6
+        psrldq          xmm6,       8
+
+        paddw           xmm0,       xmm6
+        movd            [rcx+4],    xmm0
+;-
+        movq            xmm0,       xmm7
+        psrldq          xmm7,       8
+
+        paddw           xmm0,       xmm7
+        movd            [rcx+8],    xmm0
+
+    STACK_FRAME_DESTROY_X3
+
+;void int vp9_sad16x8x3_sse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad16x8x3_sse3) PRIVATE
+sym(vp9_sad16x8x3_sse3):
+
+    STACK_FRAME_CREATE_X3
+
+        PROCESS_16X2X3 0, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_16X2X3 2, src_ptr, ref_ptr, src_stride, ref_stride
+
+        mov             rcx,        result_ptr
+
+        movq            xmm0,       xmm5
+        psrldq          xmm5,       8
+
+        paddw           xmm0,       xmm5
+        movd            [rcx],      xmm0
+;-
+        movq            xmm0,       xmm6
+        psrldq          xmm6,       8
+
+        paddw           xmm0,       xmm6
+        movd            [rcx+4],    xmm0
+;-
+        movq            xmm0,       xmm7
+        psrldq          xmm7,       8
+
+        paddw           xmm0,       xmm7
+        movd            [rcx+8],    xmm0
+
+    STACK_FRAME_DESTROY_X3
+
+;void int vp9_sad8x16x3_sse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad8x16x3_sse3) PRIVATE
+sym(vp9_sad8x16x3_sse3):
+
+    STACK_FRAME_CREATE_X3
+
+        PROCESS_8X2X3 0, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 2, src_ptr, ref_ptr, src_stride, ref_stride
+
+        mov             rcx,        result_ptr
+
+        punpckldq       mm5,        mm6
+
+        movq            [rcx],      mm5
+        movd            [rcx+8],    mm7
+
+    STACK_FRAME_DESTROY_X3
+
+;void int vp9_sad8x8x3_sse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad8x8x3_sse3) PRIVATE
+sym(vp9_sad8x8x3_sse3):
+
+    STACK_FRAME_CREATE_X3
+
+        PROCESS_8X2X3 0, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 1, src_ptr, ref_ptr, src_stride, ref_stride
+        PROCESS_8X2X3 2, src_ptr, ref_ptr, src_stride, ref_stride
+
+        mov             rcx,        result_ptr
+
+        punpckldq       mm5,        mm6
+
+        movq            [rcx],      mm5
+        movd            [rcx+8],    mm7
+
+    STACK_FRAME_DESTROY_X3
+
+;void int vp9_sad4x4x3_sse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad4x4x3_sse3) PRIVATE
+sym(vp9_sad4x4x3_sse3):
+
+    STACK_FRAME_CREATE_X3
+
+        movd            mm0,        DWORD PTR [src_ptr]
+        movd            mm1,        DWORD PTR [ref_ptr]
+
+        movd            mm2,        DWORD PTR [src_ptr+src_stride]
+        movd            mm3,        DWORD PTR [ref_ptr+ref_stride]
+
+        punpcklbw       mm0,        mm2
+        punpcklbw       mm1,        mm3
+
+        movd            mm4,        DWORD PTR [ref_ptr+1]
+        movd            mm5,        DWORD PTR [ref_ptr+2]
+
+        movd            mm2,        DWORD PTR [ref_ptr+ref_stride+1]
+        movd            mm3,        DWORD PTR [ref_ptr+ref_stride+2]
+
+        psadbw          mm1,        mm0
+
+        punpcklbw       mm4,        mm2
+        punpcklbw       mm5,        mm3
+
+        psadbw          mm4,        mm0
+        psadbw          mm5,        mm0
+
+        lea             src_ptr,    [src_ptr+src_stride*2]
+        lea             ref_ptr,    [ref_ptr+ref_stride*2]
+
+        movd            mm0,        DWORD PTR [src_ptr]
+        movd            mm2,        DWORD PTR [ref_ptr]
+
+        movd            mm3,        DWORD PTR [src_ptr+src_stride]
+        movd            mm6,        DWORD PTR [ref_ptr+ref_stride]
+
+        punpcklbw       mm0,        mm3
+        punpcklbw       mm2,        mm6
+
+        movd            mm3,        DWORD PTR [ref_ptr+1]
+        movd            mm7,        DWORD PTR [ref_ptr+2]
+
+        psadbw          mm2,        mm0
+
+        paddw           mm1,        mm2
+
+        movd            mm2,        DWORD PTR [ref_ptr+ref_stride+1]
+        movd            mm6,        DWORD PTR [ref_ptr+ref_stride+2]
+
+        punpcklbw       mm3,        mm2
+        punpcklbw       mm7,        mm6
+
+        psadbw          mm3,        mm0
+        psadbw          mm7,        mm0
+
+        paddw           mm3,        mm4
+        paddw           mm7,        mm5
+
+        mov             rcx,        result_ptr
+
+        punpckldq       mm1,        mm3
+
+        movq            [rcx],      mm1
+        movd            [rcx+8],    mm7
+
+    STACK_FRAME_DESTROY_X3
diff --git a/libvpx/vp9/encoder/x86/vp9_sad_sse4.asm b/libvpx/vp9/encoder/x86/vp9_sad_sse4.asm
new file mode 100644
index 0000000..faf1768
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad_sse4.asm
@@ -0,0 +1,359 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro PROCESS_16X2X8 1
+%if %1
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        movq            xmm1,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        movq            xmm2,       MMWORD PTR [rdi+16]
+        punpcklqdq      xmm1,       xmm3
+        punpcklqdq      xmm3,       xmm2
+
+        movdqa          xmm2,       xmm1
+        mpsadbw         xmm1,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+
+        psrldq          xmm0,       8
+
+        movdqa          xmm4,       xmm3
+        mpsadbw         xmm3,       xmm0,  0x0
+        mpsadbw         xmm4,       xmm0,  0x5
+
+        paddw           xmm1,       xmm2
+        paddw           xmm1,       xmm3
+        paddw           xmm1,       xmm4
+%else
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        movq            xmm5,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        movq            xmm2,       MMWORD PTR [rdi+16]
+        punpcklqdq      xmm5,       xmm3
+        punpcklqdq      xmm3,       xmm2
+
+        movdqa          xmm2,       xmm5
+        mpsadbw         xmm5,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+
+        psrldq          xmm0,       8
+
+        movdqa          xmm4,       xmm3
+        mpsadbw         xmm3,       xmm0,  0x0
+        mpsadbw         xmm4,       xmm0,  0x5
+
+        paddw           xmm5,       xmm2
+        paddw           xmm5,       xmm3
+        paddw           xmm5,       xmm4
+
+        paddw           xmm1,       xmm5
+%endif
+        movdqa          xmm0,       XMMWORD PTR [rsi + rax]
+        movq            xmm5,       MMWORD PTR [rdi+ rdx]
+        movq            xmm3,       MMWORD PTR [rdi+ rdx+8]
+        movq            xmm2,       MMWORD PTR [rdi+ rdx+16]
+        punpcklqdq      xmm5,       xmm3
+        punpcklqdq      xmm3,       xmm2
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        movdqa          xmm2,       xmm5
+        mpsadbw         xmm5,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+
+        psrldq          xmm0,       8
+        movdqa          xmm4,       xmm3
+        mpsadbw         xmm3,       xmm0,  0x0
+        mpsadbw         xmm4,       xmm0,  0x5
+
+        paddw           xmm5,       xmm2
+        paddw           xmm5,       xmm3
+        paddw           xmm5,       xmm4
+
+        paddw           xmm1,       xmm5
+%endmacro
+
+%macro PROCESS_8X2X8 1
+%if %1
+        movq            xmm0,       MMWORD PTR [rsi]
+        movq            xmm1,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        punpcklqdq      xmm1,       xmm3
+
+        movdqa          xmm2,       xmm1
+        mpsadbw         xmm1,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+        paddw           xmm1,       xmm2
+%else
+        movq            xmm0,       MMWORD PTR [rsi]
+        movq            xmm5,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        punpcklqdq      xmm5,       xmm3
+
+        movdqa          xmm2,       xmm5
+        mpsadbw         xmm5,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+        paddw           xmm5,       xmm2
+
+        paddw           xmm1,       xmm5
+%endif
+        movq            xmm0,       MMWORD PTR [rsi + rax]
+        movq            xmm5,       MMWORD PTR [rdi+ rdx]
+        movq            xmm3,       MMWORD PTR [rdi+ rdx+8]
+        punpcklqdq      xmm5,       xmm3
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        movdqa          xmm2,       xmm5
+        mpsadbw         xmm5,       xmm0,  0x0
+        mpsadbw         xmm2,       xmm0,  0x5
+        paddw           xmm5,       xmm2
+
+        paddw           xmm1,       xmm5
+%endmacro
+
+%macro PROCESS_4X2X8 1
+%if %1
+        movd            xmm0,       [rsi]
+        movq            xmm1,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        punpcklqdq      xmm1,       xmm3
+
+        mpsadbw         xmm1,       xmm0,  0x0
+%else
+        movd            xmm0,       [rsi]
+        movq            xmm5,       MMWORD PTR [rdi]
+        movq            xmm3,       MMWORD PTR [rdi+8]
+        punpcklqdq      xmm5,       xmm3
+
+        mpsadbw         xmm5,       xmm0,  0x0
+
+        paddw           xmm1,       xmm5
+%endif
+        movd            xmm0,       [rsi + rax]
+        movq            xmm5,       MMWORD PTR [rdi+ rdx]
+        movq            xmm3,       MMWORD PTR [rdi+ rdx+8]
+        punpcklqdq      xmm5,       xmm3
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        mpsadbw         xmm5,       xmm0,  0x0
+
+        paddw           xmm1,       xmm5
+%endmacro
+
+%macro WRITE_AS_INTS 0
+    mov             rdi,        arg(4)           ;Results
+    pxor            xmm0, xmm0
+    movdqa          xmm2, xmm1
+    punpcklwd       xmm1, xmm0
+    punpckhwd       xmm2, xmm0
+
+    movdqa          [rdi],    xmm1
+    movdqa          [rdi + 16],    xmm2
+%endmacro
+
+;void vp9_sad16x16x8_sse4(
+;    const unsigned char *src_ptr,
+;    int  src_stride,
+;    const unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned short *sad_array);
+global sym(vp9_sad16x16x8_sse4) PRIVATE
+sym(vp9_sad16x16x8_sse4):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0)           ;src_ptr
+    mov             rdi,        arg(2)           ;ref_ptr
+
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    PROCESS_16X2X8 1
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+
+    WRITE_AS_INTS
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_sad16x8x8_sse4(
+;    const unsigned char *src_ptr,
+;    int  src_stride,
+;    const unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned short *sad_array
+;);
+global sym(vp9_sad16x8x8_sse4) PRIVATE
+sym(vp9_sad16x8x8_sse4):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0)           ;src_ptr
+    mov             rdi,        arg(2)           ;ref_ptr
+
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    PROCESS_16X2X8 1
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+    PROCESS_16X2X8 0
+
+    WRITE_AS_INTS
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_sad8x8x8_sse4(
+;    const unsigned char *src_ptr,
+;    int  src_stride,
+;    const unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned short *sad_array
+;);
+global sym(vp9_sad8x8x8_sse4) PRIVATE
+sym(vp9_sad8x8x8_sse4):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0)           ;src_ptr
+    mov             rdi,        arg(2)           ;ref_ptr
+
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    PROCESS_8X2X8 1
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+
+    WRITE_AS_INTS
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_sad8x16x8_sse4(
+;    const unsigned char *src_ptr,
+;    int  src_stride,
+;    const unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned short *sad_array
+;);
+global sym(vp9_sad8x16x8_sse4) PRIVATE
+sym(vp9_sad8x16x8_sse4):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0)           ;src_ptr
+    mov             rdi,        arg(2)           ;ref_ptr
+
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    PROCESS_8X2X8 1
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+    PROCESS_8X2X8 0
+
+    WRITE_AS_INTS
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_sad4x4x8_c(
+;    const unsigned char *src_ptr,
+;    int  src_stride,
+;    const unsigned char *ref_ptr,
+;    int  ref_stride,
+;    unsigned short *sad_array
+;);
+global sym(vp9_sad4x4x8_sse4) PRIVATE
+sym(vp9_sad4x4x8_sse4):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0)           ;src_ptr
+    mov             rdi,        arg(2)           ;ref_ptr
+
+    movsxd          rax,        dword ptr arg(1) ;src_stride
+    movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+    PROCESS_4X2X8 1
+    PROCESS_4X2X8 0
+
+    WRITE_AS_INTS
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+
diff --git a/libvpx/vp9/encoder/x86/vp9_sad_ssse3.asm b/libvpx/vp9/encoder/x86/vp9_sad_ssse3.asm
new file mode 100644
index 0000000..0cb3542
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_sad_ssse3.asm
@@ -0,0 +1,370 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro PROCESS_16X2X3 1
+%if %1
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        lddqu           xmm5,       XMMWORD PTR [rdi]
+        lddqu           xmm6,       XMMWORD PTR [rdi+1]
+        lddqu           xmm7,       XMMWORD PTR [rdi+2]
+
+        psadbw          xmm5,       xmm0
+        psadbw          xmm6,       xmm0
+        psadbw          xmm7,       xmm0
+%else
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        lddqu           xmm1,       XMMWORD PTR [rdi]
+        lddqu           xmm2,       XMMWORD PTR [rdi+1]
+        lddqu           xmm3,       XMMWORD PTR [rdi+2]
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endif
+        movdqa          xmm0,       XMMWORD PTR [rsi+rax]
+        lddqu           xmm1,       XMMWORD PTR [rdi+rdx]
+        lddqu           xmm2,       XMMWORD PTR [rdi+rdx+1]
+        lddqu           xmm3,       XMMWORD PTR [rdi+rdx+2]
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endmacro
+
+%macro PROCESS_16X2X3_OFFSET 2
+%if %1
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        movdqa          xmm4,       XMMWORD PTR [rdi]
+        movdqa          xmm7,       XMMWORD PTR [rdi+16]
+
+        movdqa          xmm5,       xmm7
+        palignr         xmm5,       xmm4,       %2
+
+        movdqa          xmm6,       xmm7
+        palignr         xmm6,       xmm4,       (%2+1)
+
+        palignr         xmm7,       xmm4,       (%2+2)
+
+        psadbw          xmm5,       xmm0
+        psadbw          xmm6,       xmm0
+        psadbw          xmm7,       xmm0
+%else
+        movdqa          xmm0,       XMMWORD PTR [rsi]
+        movdqa          xmm4,       XMMWORD PTR [rdi]
+        movdqa          xmm3,       XMMWORD PTR [rdi+16]
+
+        movdqa          xmm1,       xmm3
+        palignr         xmm1,       xmm4,       %2
+
+        movdqa          xmm2,       xmm3
+        palignr         xmm2,       xmm4,       (%2+1)
+
+        palignr         xmm3,       xmm4,       (%2+2)
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endif
+        movdqa          xmm0,       XMMWORD PTR [rsi+rax]
+        movdqa          xmm4,       XMMWORD PTR [rdi+rdx]
+        movdqa          xmm3,       XMMWORD PTR [rdi+rdx+16]
+
+        movdqa          xmm1,       xmm3
+        palignr         xmm1,       xmm4,       %2
+
+        movdqa          xmm2,       xmm3
+        palignr         xmm2,       xmm4,       (%2+1)
+
+        palignr         xmm3,       xmm4,       (%2+2)
+
+        lea             rsi,        [rsi+rax*2]
+        lea             rdi,        [rdi+rdx*2]
+
+        psadbw          xmm1,       xmm0
+        psadbw          xmm2,       xmm0
+        psadbw          xmm3,       xmm0
+
+        paddw           xmm5,       xmm1
+        paddw           xmm6,       xmm2
+        paddw           xmm7,       xmm3
+%endmacro
+
+%macro PROCESS_16X16X3_OFFSET 2
+%2_aligned_by_%1:
+
+        sub             rdi,        %1
+
+        PROCESS_16X2X3_OFFSET 1, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+
+        jmp             %2_store_off
+
+%endmacro
+
+%macro PROCESS_16X8X3_OFFSET 2
+%2_aligned_by_%1:
+
+        sub             rdi,        %1
+
+        PROCESS_16X2X3_OFFSET 1, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+        PROCESS_16X2X3_OFFSET 0, %1
+
+        jmp             %2_store_off
+
+%endmacro
+
+;void int vp9_sad16x16x3_ssse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad16x16x3_ssse3) PRIVATE
+sym(vp9_sad16x16x3_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rcx
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        mov             rdx,        0xf
+        and             rdx,        rdi
+
+        jmp .vp9_sad16x16x3_ssse3_skiptable
+.vp9_sad16x16x3_ssse3_jumptable:
+        dd .vp9_sad16x16x3_ssse3_aligned_by_0  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_1  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_2  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_3  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_4  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_5  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_6  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_7  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_8  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_9  - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_10 - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_11 - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_12 - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_13 - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_14 - .vp9_sad16x16x3_ssse3_do_jump
+        dd .vp9_sad16x16x3_ssse3_aligned_by_15 - .vp9_sad16x16x3_ssse3_do_jump
+.vp9_sad16x16x3_ssse3_skiptable:
+
+        call .vp9_sad16x16x3_ssse3_do_jump
+.vp9_sad16x16x3_ssse3_do_jump:
+        pop             rcx                         ; get the address of do_jump
+        mov             rax,  .vp9_sad16x16x3_ssse3_jumptable - .vp9_sad16x16x3_ssse3_do_jump
+        add             rax,  rcx  ; get the absolute address of vp9_sad16x16x3_ssse3_jumptable
+
+        movsxd          rax,  dword [rax + 4*rdx]   ; get the 32 bit offset from the jumptable
+        add             rcx,        rax
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        jmp             rcx
+
+        PROCESS_16X16X3_OFFSET 0,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 1,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 2,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 3,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 4,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 5,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 6,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 7,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 8,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 9,  .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 10, .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 11, .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 12, .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 13, .vp9_sad16x16x3_ssse3
+        PROCESS_16X16X3_OFFSET 14, .vp9_sad16x16x3_ssse3
+
+.vp9_sad16x16x3_ssse3_aligned_by_15:
+        PROCESS_16X2X3 1
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+
+.vp9_sad16x16x3_ssse3_store_off:
+        mov             rdi,        arg(4) ;Results
+
+        movq            xmm0,       xmm5
+        psrldq          xmm5,       8
+
+        paddw           xmm0,       xmm5
+        movd            [rdi],      xmm0
+;-
+        movq            xmm0,       xmm6
+        psrldq          xmm6,       8
+
+        paddw           xmm0,       xmm6
+        movd            [rdi+4],    xmm0
+;-
+        movq            xmm0,       xmm7
+        psrldq          xmm7,       8
+
+        paddw           xmm0,       xmm7
+        movd            [rdi+8],    xmm0
+
+    ; begin epilog
+    pop         rcx
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void int vp9_sad16x8x3_ssse3(
+;    unsigned char *src_ptr,
+;    int  src_stride,
+;    unsigned char *ref_ptr,
+;    int  ref_stride,
+;    int  *results)
+global sym(vp9_sad16x8x3_ssse3) PRIVATE
+sym(vp9_sad16x8x3_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rcx
+    ; end prolog
+
+        mov             rsi,        arg(0) ;src_ptr
+        mov             rdi,        arg(2) ;ref_ptr
+
+        mov             rdx,        0xf
+        and             rdx,        rdi
+
+        jmp .vp9_sad16x8x3_ssse3_skiptable
+.vp9_sad16x8x3_ssse3_jumptable:
+        dd .vp9_sad16x8x3_ssse3_aligned_by_0  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_1  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_2  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_3  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_4  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_5  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_6  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_7  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_8  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_9  - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_10 - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_11 - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_12 - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_13 - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_14 - .vp9_sad16x8x3_ssse3_do_jump
+        dd .vp9_sad16x8x3_ssse3_aligned_by_15 - .vp9_sad16x8x3_ssse3_do_jump
+.vp9_sad16x8x3_ssse3_skiptable:
+
+        call .vp9_sad16x8x3_ssse3_do_jump
+.vp9_sad16x8x3_ssse3_do_jump:
+        pop             rcx                         ; get the address of do_jump
+        mov             rax,  .vp9_sad16x8x3_ssse3_jumptable - .vp9_sad16x8x3_ssse3_do_jump
+        add             rax,  rcx  ; get the absolute address of vp9_sad16x8x3_ssse3_jumptable
+
+        movsxd          rax,  dword [rax + 4*rdx]   ; get the 32 bit offset from the jumptable
+        add             rcx,        rax
+
+        movsxd          rax,        dword ptr arg(1) ;src_stride
+        movsxd          rdx,        dword ptr arg(3) ;ref_stride
+
+        jmp             rcx
+
+        PROCESS_16X8X3_OFFSET 0,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 1,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 2,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 3,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 4,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 5,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 6,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 7,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 8,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 9,  .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 10, .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 11, .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 12, .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 13, .vp9_sad16x8x3_ssse3
+        PROCESS_16X8X3_OFFSET 14, .vp9_sad16x8x3_ssse3
+
+.vp9_sad16x8x3_ssse3_aligned_by_15:
+
+        PROCESS_16X2X3 1
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+        PROCESS_16X2X3 0
+
+.vp9_sad16x8x3_ssse3_store_off:
+        mov             rdi,        arg(4) ;Results
+
+        movq            xmm0,       xmm5
+        psrldq          xmm5,       8
+
+        paddw           xmm0,       xmm5
+        movd            [rdi],      xmm0
+;-
+        movq            xmm0,       xmm6
+        psrldq          xmm6,       8
+
+        paddw           xmm0,       xmm6
+        movd            [rdi+4],    xmm0
+;-
+        movq            xmm0,       xmm7
+        psrldq          xmm7,       8
+
+        paddw           xmm0,       xmm7
+        movd            [rdi+8],    xmm0
+
+    ; begin epilog
+    pop         rcx
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/encoder/x86/vp9_ssim_opt.asm b/libvpx/vp9/encoder/x86/vp9_ssim_opt.asm
new file mode 100644
index 0000000..455d10d
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_ssim_opt.asm
@@ -0,0 +1,216 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+; tabulate_ssim - sums sum_s,sum_r,sum_sq_s,sum_sq_r, sum_sxr
+%macro TABULATE_SSIM 0
+        paddusw         xmm15, xmm3  ; sum_s
+        paddusw         xmm14, xmm4  ; sum_r
+        movdqa          xmm1, xmm3
+        pmaddwd         xmm1, xmm1
+        paddd           xmm13, xmm1 ; sum_sq_s
+        movdqa          xmm2, xmm4
+        pmaddwd         xmm2, xmm2
+        paddd           xmm12, xmm2 ; sum_sq_r
+        pmaddwd         xmm3, xmm4
+        paddd           xmm11, xmm3  ; sum_sxr
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_Q 1
+        movdqa          xmm2,%1
+        punpckldq       %1,xmm0
+        punpckhdq       xmm2,xmm0
+        paddq           %1,xmm2
+        movdqa          xmm2,%1
+        punpcklqdq      %1,xmm0
+        punpckhqdq      xmm2,xmm0
+        paddq           %1,xmm2
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_W 1
+        movdqa          xmm1, %1
+        punpcklwd       %1,xmm0
+        punpckhwd       xmm1,xmm0
+        paddd           %1, xmm1
+        SUM_ACROSS_Q    %1
+%endmacro
+;void ssim_parms_sse2(
+;    unsigned char *s,
+;    int sp,
+;    unsigned char *r,
+;    int rp
+;    unsigned long *sum_s,
+;    unsigned long *sum_r,
+;    unsigned long *sum_sq_s,
+;    unsigned long *sum_sq_r,
+;    unsigned long *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(vp9_ssim_parms_16x16_sse2) PRIVATE
+sym(vp9_ssim_parms_16x16_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 15
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0) ;s
+    mov             rcx,        arg(1) ;sp
+    mov             rdi,        arg(2) ;r
+    mov             rax,        arg(3) ;rp
+
+    pxor            xmm0, xmm0
+    pxor            xmm15,xmm15  ;sum_s
+    pxor            xmm14,xmm14  ;sum_r
+    pxor            xmm13,xmm13  ;sum_sq_s
+    pxor            xmm12,xmm12  ;sum_sq_r
+    pxor            xmm11,xmm11  ;sum_sxr
+
+    mov             rdx, 16      ;row counter
+.NextRow:
+
+    ;grab source and reference pixels
+    movdqu          xmm5, [rsi]
+    movdqu          xmm6, [rdi]
+    movdqa          xmm3, xmm5
+    movdqa          xmm4, xmm6
+    punpckhbw       xmm3, xmm0 ; high_s
+    punpckhbw       xmm4, xmm0 ; high_r
+
+    TABULATE_SSIM
+
+    movdqa          xmm3, xmm5
+    movdqa          xmm4, xmm6
+    punpcklbw       xmm3, xmm0 ; low_s
+    punpcklbw       xmm4, xmm0 ; low_r
+
+    TABULATE_SSIM
+
+    add             rsi, rcx   ; next s row
+    add             rdi, rax   ; next r row
+
+    dec             rdx        ; counter
+    jnz .NextRow
+
+    SUM_ACROSS_W    xmm15
+    SUM_ACROSS_W    xmm14
+    SUM_ACROSS_Q    xmm13
+    SUM_ACROSS_Q    xmm12
+    SUM_ACROSS_Q    xmm11
+
+    mov             rdi,arg(4)
+    movd            [rdi], xmm15;
+    mov             rdi,arg(5)
+    movd            [rdi], xmm14;
+    mov             rdi,arg(6)
+    movd            [rdi], xmm13;
+    mov             rdi,arg(7)
+    movd            [rdi], xmm12;
+    mov             rdi,arg(8)
+    movd            [rdi], xmm11;
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void ssim_parms_sse2(
+;    unsigned char *s,
+;    int sp,
+;    unsigned char *r,
+;    int rp
+;    unsigned long *sum_s,
+;    unsigned long *sum_r,
+;    unsigned long *sum_sq_s,
+;    unsigned long *sum_sq_r,
+;    unsigned long *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(vp9_ssim_parms_8x8_sse2) PRIVATE
+sym(vp9_ssim_parms_8x8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 15
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0) ;s
+    mov             rcx,        arg(1) ;sp
+    mov             rdi,        arg(2) ;r
+    mov             rax,        arg(3) ;rp
+
+    pxor            xmm0, xmm0
+    pxor            xmm15,xmm15  ;sum_s
+    pxor            xmm14,xmm14  ;sum_r
+    pxor            xmm13,xmm13  ;sum_sq_s
+    pxor            xmm12,xmm12  ;sum_sq_r
+    pxor            xmm11,xmm11  ;sum_sxr
+
+    mov             rdx, 8      ;row counter
+.NextRow:
+
+    ;grab source and reference pixels
+    movq            xmm3, [rsi]
+    movq            xmm4, [rdi]
+    punpcklbw       xmm3, xmm0 ; low_s
+    punpcklbw       xmm4, xmm0 ; low_r
+
+    TABULATE_SSIM
+
+    add             rsi, rcx   ; next s row
+    add             rdi, rax   ; next r row
+
+    dec             rdx        ; counter
+    jnz .NextRow
+
+    SUM_ACROSS_W    xmm15
+    SUM_ACROSS_W    xmm14
+    SUM_ACROSS_Q    xmm13
+    SUM_ACROSS_Q    xmm12
+    SUM_ACROSS_Q    xmm11
+
+    mov             rdi,arg(4)
+    movd            [rdi], xmm15;
+    mov             rdi,arg(5)
+    movd            [rdi], xmm14;
+    mov             rdi,arg(6)
+    movd            [rdi], xmm13;
+    mov             rdi,arg(7)
+    movd            [rdi], xmm12;
+    mov             rdi,arg(8)
+    movd            [rdi], xmm11;
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/encoder/x86/vp9_subpel_variance_impl_sse2.asm b/libvpx/vp9/encoder/x86/vp9_subpel_variance_impl_sse2.asm
new file mode 100644
index 0000000..8a2a471
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_subpel_variance_impl_sse2.asm
@@ -0,0 +1,645 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%define xmm_filter_shift            7
+
+;void vp9_filter_block2d_bil_var_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int  xoffset,
+;    int  yoffset,
+;    int *sum,
+;    unsigned int *sumsquared;;
+;
+;)
+global sym(vp9_filter_block2d_bil_var_sse2) PRIVATE
+sym(vp9_filter_block2d_bil_var_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    push rbx
+    ; end prolog
+
+        pxor            xmm6,           xmm6                 ;
+        pxor            xmm7,           xmm7                 ;
+
+        lea             rsi,            [GLOBAL(xmm_bi_rd)]  ; rounding
+        movdqa          xmm4,           XMMWORD PTR [rsi]
+
+        lea             rcx,            [GLOBAL(bilinear_filters_sse2)]
+        movsxd          rax,            dword ptr arg(5)     ; xoffset
+
+        cmp             rax,            0                    ; skip first_pass filter if xoffset=0
+        je              filter_block2d_bil_var_sse2_sp_only
+
+        shl             rax,            5                    ; point to filter coeff with xoffset
+        lea             rax,            [rax + rcx]          ; HFilter
+
+        movsxd          rdx,            dword ptr arg(6)     ; yoffset
+
+        cmp             rdx,            0                    ; skip second_pass filter if yoffset=0
+        je              filter_block2d_bil_var_sse2_fp_only
+
+        shl             rdx,            5
+        lea             rdx,            [rdx + rcx]          ; VFilter
+
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+
+        pxor            xmm0,           xmm0                 ;
+        movq            xmm1,           QWORD PTR [rsi]      ;
+        movq            xmm3,           QWORD PTR [rsi+1]    ;
+
+        punpcklbw       xmm1,           xmm0                 ;
+        pmullw          xmm1,           [rax]                ;
+        punpcklbw       xmm3,           xmm0
+        pmullw          xmm3,           [rax+16]             ;
+
+        paddw           xmm1,           xmm3                 ;
+        paddw           xmm1,           xmm4                 ;
+        psraw           xmm1,           xmm_filter_shift     ;
+        movdqa          xmm5,           xmm1
+
+        movsxd          rbx,            dword ptr arg(1) ;ref_pixels_per_line
+        lea             rsi,            [rsi + rbx]
+%if ABI_IS_32BIT=0
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+filter_block2d_bil_var_sse2_loop:
+        movq            xmm1,           QWORD PTR [rsi]               ;
+        movq            xmm3,           QWORD PTR [rsi+1]             ;
+
+        punpcklbw       xmm1,           xmm0                 ;
+        pmullw          xmm1,           [rax]               ;
+        punpcklbw       xmm3,           xmm0                 ;
+        pmullw          xmm3,           [rax+16]             ;
+
+        paddw           xmm1,           xmm3                 ;
+        paddw           xmm1,           xmm4               ;
+        psraw           xmm1,           xmm_filter_shift    ;
+
+        movdqa          xmm3,           xmm5                 ;
+        movdqa          xmm5,           xmm1                 ;
+
+        pmullw          xmm3,           [rdx]               ;
+        pmullw          xmm1,           [rdx+16]             ;
+        paddw           xmm1,           xmm3                 ;
+        paddw           xmm1,           xmm4                 ;
+        psraw           xmm1,           xmm_filter_shift    ;
+
+        movq            xmm3,           QWORD PTR [rdi]               ;
+        punpcklbw       xmm3,           xmm0                 ;
+
+        psubw           xmm1,           xmm3                 ;
+        paddw           xmm6,           xmm1                 ;
+
+        pmaddwd         xmm1,           xmm1                 ;
+        paddd           xmm7,           xmm1                 ;
+
+        lea             rsi,            [rsi + rbx]          ;ref_pixels_per_line
+%if ABI_IS_32BIT
+        add             rdi,            dword ptr arg(3)     ;src_pixels_per_line
+%else
+        lea             rdi,            [rdi + r9]
+%endif
+
+        sub             rcx,            1                   ;
+        jnz             filter_block2d_bil_var_sse2_loop       ;
+
+        jmp             filter_block2d_bil_variance
+
+filter_block2d_bil_var_sse2_sp_only:
+        movsxd          rdx,            dword ptr arg(6)     ; yoffset
+
+        cmp             rdx,            0                    ; skip all if both xoffset=0 and yoffset=0
+        je              filter_block2d_bil_var_sse2_full_pixel
+
+        shl             rdx,            5
+        lea             rdx,            [rdx + rcx]          ; VFilter
+
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rax,            dword ptr arg(1)     ;ref_pixels_per_line
+
+        pxor            xmm0,           xmm0                 ;
+        movq            xmm1,           QWORD PTR [rsi]      ;
+        punpcklbw       xmm1,           xmm0                 ;
+
+        movsxd          rbx,            dword ptr arg(3)     ;src_pixels_per_line
+        lea             rsi,            [rsi + rax]
+
+filter_block2d_bil_sp_only_loop:
+        movq            xmm3,           QWORD PTR [rsi]             ;
+        punpcklbw       xmm3,           xmm0                 ;
+        movdqa          xmm5,           xmm3
+
+        pmullw          xmm1,           [rdx]               ;
+        pmullw          xmm3,           [rdx+16]             ;
+        paddw           xmm1,           xmm3                 ;
+        paddw           xmm1,           xmm4                 ;
+        psraw           xmm1,           xmm_filter_shift    ;
+
+        movq            xmm3,           QWORD PTR [rdi]               ;
+        punpcklbw       xmm3,           xmm0                 ;
+
+        psubw           xmm1,           xmm3                 ;
+        paddw           xmm6,           xmm1                 ;
+
+        pmaddwd         xmm1,           xmm1                 ;
+        paddd           xmm7,           xmm1                 ;
+
+        movdqa          xmm1,           xmm5                 ;
+        lea             rsi,            [rsi + rax]          ;ref_pixels_per_line
+        lea             rdi,            [rdi + rbx]          ;src_pixels_per_line
+
+        sub             rcx,            1                   ;
+        jnz             filter_block2d_bil_sp_only_loop       ;
+
+        jmp             filter_block2d_bil_variance
+
+filter_block2d_bil_var_sse2_full_pixel:
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rax,            dword ptr arg(1)     ;ref_pixels_per_line
+        movsxd          rbx,            dword ptr arg(3)     ;src_pixels_per_line
+        pxor            xmm0,           xmm0                 ;
+
+filter_block2d_bil_full_pixel_loop:
+        movq            xmm1,           QWORD PTR [rsi]               ;
+        punpcklbw       xmm1,           xmm0                 ;
+
+        movq            xmm2,           QWORD PTR [rdi]               ;
+        punpcklbw       xmm2,           xmm0                 ;
+
+        psubw           xmm1,           xmm2                 ;
+        paddw           xmm6,           xmm1                 ;
+
+        pmaddwd         xmm1,           xmm1                 ;
+        paddd           xmm7,           xmm1                 ;
+
+        lea             rsi,            [rsi + rax]          ;ref_pixels_per_line
+        lea             rdi,            [rdi + rbx]          ;src_pixels_per_line
+
+        sub             rcx,            1                   ;
+        jnz             filter_block2d_bil_full_pixel_loop       ;
+
+        jmp             filter_block2d_bil_variance
+
+filter_block2d_bil_var_sse2_fp_only:
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rdx,            dword ptr arg(1)     ;ref_pixels_per_line
+
+        pxor            xmm0,           xmm0                 ;
+        movsxd          rbx,            dword ptr arg(3)     ;src_pixels_per_line
+
+filter_block2d_bil_fp_only_loop:
+        movq            xmm1,           QWORD PTR [rsi]       ;
+        movq            xmm3,           QWORD PTR [rsi+1]     ;
+
+        punpcklbw       xmm1,           xmm0                 ;
+        pmullw          xmm1,           [rax]               ;
+        punpcklbw       xmm3,           xmm0                 ;
+        pmullw          xmm3,           [rax+16]             ;
+
+        paddw           xmm1,           xmm3                 ;
+        paddw           xmm1,           xmm4  ;
+        psraw           xmm1,           xmm_filter_shift    ;
+
+        movq            xmm3,           QWORD PTR [rdi]     ;
+        punpcklbw       xmm3,           xmm0                 ;
+
+        psubw           xmm1,           xmm3                 ;
+        paddw           xmm6,           xmm1                 ;
+
+        pmaddwd         xmm1,           xmm1                 ;
+        paddd           xmm7,           xmm1                 ;
+        lea             rsi,            [rsi + rdx]
+        lea             rdi,            [rdi + rbx]          ;src_pixels_per_line
+
+        sub             rcx,            1                   ;
+        jnz             filter_block2d_bil_fp_only_loop       ;
+
+        jmp             filter_block2d_bil_variance
+
+filter_block2d_bil_variance:
+        movdq2q         mm6,            xmm6                ;
+        movdq2q         mm7,            xmm7                ;
+
+        psrldq          xmm6,           8
+        psrldq          xmm7,           8
+
+        movdq2q         mm2,            xmm6
+        movdq2q         mm3,            xmm7
+
+        paddw           mm6,            mm2
+        paddd           mm7,            mm3
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rsi,            arg(7) ; sum
+        mov             rdi,            arg(8) ; sumsquared
+
+        movd            [rsi],          mm2    ; xsum
+        movd            [rdi],          mm4    ; xxsum
+
+    ; begin epilog
+    pop rbx
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+;void vp9_half_horiz_vert_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_horiz_vert_variance16x_h_sse2) PRIVATE
+sym(vp9_half_horiz_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3) horizontal line 1
+
+        lea             rsi,            [rsi + rax]
+
+.half_horiz_vert_variance16x_h_1:
+        movdqu          xmm1,           XMMWORD PTR [rsi]     ;
+        movdqu          xmm2,           XMMWORD PTR [rsi+1]   ;
+        pavgb           xmm1,           xmm2                ;  xmm1 = avg(xmm1,xmm3) horizontal line i+1
+
+        pavgb           xmm5,           xmm1                ;  xmm = vertical average of the above
+
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+
+        movq            xmm3,           QWORD PTR [rdi+8]
+        punpcklbw       xmm3,           xmm0
+        psubw           xmm4,           xmm3
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm1                ;  save xmm1 for use on the next row
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             .half_horiz_vert_variance16x_h_1    ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_half_vert_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_vert_variance16x_h_sse2) PRIVATE
+sym(vp9_half_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0)              ;ref_ptr
+
+        mov             rdi,            arg(2)              ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)    ;Height
+        movsxd          rax,            dword ptr arg(1)    ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        lea             rsi,            [rsi + rax          ]
+        pxor            xmm0,           xmm0
+
+.half_vert_variance16x_h_1:
+        movdqu          xmm3,           XMMWORD PTR [rsi]
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm2,           QWORD PTR [rdi]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm5,           xmm2
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm4,           xmm2
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm3
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1
+        jnz             .half_vert_variance16x_h_1
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_half_horiz_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_horiz_variance16x_h_sse2) PRIVATE
+sym(vp9_half_horiz_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+
+.half_horiz_variance16x_h_1:
+        movdqu          xmm5,           XMMWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s15
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s16
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm1,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm1,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        psubw           xmm1,           xmm2
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm1
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm1,           xmm1
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm1
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             .half_horiz_variance16x_h_1         ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+;    short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64};
+align 16
+xmm_bi_rd:
+    times 8 dw 64
+align 16
+bilinear_filters_sse2:
+    dw 128, 128, 128, 128, 128, 128, 128, 128,  0,  0,  0,  0,  0,  0,  0,  0
+    dw 120, 120, 120, 120, 120, 120, 120, 120,  8,  8,  8,  8,  8,  8,  8,  8
+    dw 112, 112, 112, 112, 112, 112, 112, 112, 16, 16, 16, 16, 16, 16, 16, 16
+    dw 104, 104, 104, 104, 104, 104, 104, 104, 24, 24, 24, 24, 24, 24, 24, 24
+    dw 96, 96, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32
+    dw 88, 88, 88, 88, 88, 88, 88, 88, 40, 40, 40, 40, 40, 40, 40, 40
+    dw 80, 80, 80, 80, 80, 80, 80, 80, 48, 48, 48, 48, 48, 48, 48, 48
+    dw 72, 72, 72, 72, 72, 72, 72, 72, 56, 56, 56, 56, 56, 56, 56, 56
+    dw 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
+    dw 56, 56, 56, 56, 56, 56, 56, 56, 72, 72, 72, 72, 72, 72, 72, 72
+    dw 48, 48, 48, 48, 48, 48, 48, 48, 80, 80, 80, 80, 80, 80, 80, 80
+    dw 40, 40, 40, 40, 40, 40, 40, 40, 88, 88, 88, 88, 88, 88, 88, 88
+    dw 32, 32, 32, 32, 32, 32, 32, 32, 96, 96, 96, 96, 96, 96, 96, 96
+    dw 24, 24, 24, 24, 24, 24, 24, 24, 104, 104, 104, 104, 104, 104, 104, 104
+    dw 16, 16, 16, 16, 16, 16, 16, 16, 112, 112, 112, 112, 112, 112, 112, 112
+    dw 8, 8, 8, 8, 8, 8, 8, 8, 120, 120, 120, 120, 120, 120, 120, 120
diff --git a/libvpx/vp9/encoder/x86/vp9_subtract_mmx.asm b/libvpx/vp9/encoder/x86/vp9_subtract_mmx.asm
new file mode 100644
index 0000000..e9eda4f
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_subtract_mmx.asm
@@ -0,0 +1,432 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_subtract_b_mmx_impl(unsigned char *z,  int src_stride,
+;                            short *diff, unsigned char *Predictor,
+;                            int pitch);
+global sym(vp9_subtract_b_mmx_impl) PRIVATE
+sym(vp9_subtract_b_mmx_impl):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push rsi
+    push rdi
+    ; end prolog
+
+
+        mov     rdi,        arg(2) ;diff
+        mov     rax,        arg(3) ;Predictor
+        mov     rsi,        arg(0) ;z
+        movsxd  rdx,        dword ptr arg(1);src_stride;
+        movsxd  rcx,        dword ptr arg(4);pitch
+        pxor    mm7,        mm7
+
+        movd    mm0,        [rsi]
+        movd    mm1,        [rax]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    [rdi],      mm0
+
+
+        movd    mm0,        [rsi+rdx]
+        movd    mm1,        [rax+rcx]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    [rdi+rcx*2],mm0
+
+
+        movd    mm0,        [rsi+rdx*2]
+        movd    mm1,        [rax+rcx*2]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    [rdi+rcx*4],        mm0
+
+        lea     rsi,        [rsi+rdx*2]
+        lea     rcx,        [rcx+rcx*2]
+
+
+
+        movd    mm0,        [rsi+rdx]
+        movd    mm1,        [rax+rcx]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    [rdi+rcx*2],        mm0
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_subtract_mby_mmx(short *diff, unsigned char *src, unsigned char *pred, int stride)
+global sym(vp9_subtract_mby_mmx) PRIVATE
+sym(vp9_subtract_mby_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    ; end prolog
+
+
+            mov         rsi,            arg(1) ;src
+            mov         rdi,            arg(0) ;diff
+
+            mov         rax,            arg(2) ;pred
+            movsxd      rdx,            dword ptr arg(3) ;stride
+
+            mov         rcx,            16
+            pxor        mm0,            mm0
+
+.submby_loop:
+
+            movq        mm1,            [rsi]
+            movq        mm3,            [rax]
+
+            movq        mm2,            mm1
+            movq        mm4,            mm3
+
+            punpcklbw   mm1,            mm0
+            punpcklbw   mm3,            mm0
+
+            punpckhbw   mm2,            mm0
+            punpckhbw   mm4,            mm0
+
+            psubw       mm1,            mm3
+            psubw       mm2,            mm4
+
+            movq        [rdi],          mm1
+            movq        [rdi+8],        mm2
+
+
+            movq        mm1,            [rsi+8]
+            movq        mm3,            [rax+8]
+
+            movq        mm2,            mm1
+            movq        mm4,            mm3
+
+            punpcklbw   mm1,            mm0
+            punpcklbw   mm3,            mm0
+
+            punpckhbw   mm2,            mm0
+            punpckhbw   mm4,            mm0
+
+            psubw       mm1,            mm3
+            psubw       mm2,            mm4
+
+            movq        [rdi+16],       mm1
+            movq        [rdi+24],       mm2
+
+
+            add         rdi,            32
+            add         rax,            16
+
+            lea         rsi,            [rsi+rdx]
+
+            sub         rcx,            1
+            jnz         .submby_loop
+
+    pop rdi
+    pop rsi
+    ; begin epilog
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_subtract_mbuv_mmx(short *diff, unsigned char *usrc, unsigned char *vsrc, unsigned char *pred, int stride)
+global sym(vp9_subtract_mbuv_mmx) PRIVATE
+sym(vp9_subtract_mbuv_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    push rsi
+    push rdi
+    ; end prolog
+
+    ;short *udiff = diff + 256;
+    ;short *vdiff = diff + 320;
+    ;unsigned char *upred = pred + 256;
+    ;unsigned char *vpred = pred + 320;
+
+        ;unsigned char  *z    = usrc;
+        ;unsigned short *diff = udiff;
+        ;unsigned char  *Predictor= upred;
+
+            mov     rdi,        arg(0) ;diff
+            mov     rax,        arg(3) ;pred
+            mov     rsi,        arg(1) ;z = usrc
+            add     rdi,        256*2  ;diff = diff + 256 (shorts)
+            add     rax,        256    ;Predictor = pred + 256
+            movsxd  rdx,        dword ptr arg(4) ;stride;
+            pxor    mm7,        mm7
+
+            movq    mm0,        [rsi]
+            movq    mm1,        [rax]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi],      mm0
+            movq    [rdi+8],    mm3
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+8]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+16],   mm0
+            movq    [rdi+24],   mm3
+
+            movq    mm0,        [rsi+rdx*2]
+            movq    mm1,        [rax+16]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+32],   mm0
+            movq    [rdi+40],   mm3
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+24]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+
+            movq    [rdi+48],   mm0
+            movq    [rdi+56],   mm3
+
+
+            add     rdi,        64
+            add     rax,        32
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi]
+            movq    mm1,        [rax]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi],      mm0
+            movq    [rdi+8],    mm3
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+8]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+16],   mm0
+            movq    [rdi+24],   mm3
+
+            movq    mm0,        [rsi+rdx*2]
+            movq    mm1,        [rax+16]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+32],   mm0
+            movq    [rdi+40],   mm3
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+24]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+
+            movq    [rdi+48],   mm0
+            movq    [rdi+56],   mm3
+
+        ;unsigned char  *z    = vsrc;
+        ;unsigned short *diff = vdiff;
+        ;unsigned char  *Predictor= vpred;
+
+            mov     rdi,        arg(0) ;diff
+            mov     rax,        arg(3) ;pred
+            mov     rsi,        arg(2) ;z = usrc
+            add     rdi,        320*2  ;diff = diff + 320 (shorts)
+            add     rax,        320    ;Predictor = pred + 320
+            movsxd  rdx,        dword ptr arg(4) ;stride;
+            pxor    mm7,        mm7
+
+            movq    mm0,        [rsi]
+            movq    mm1,        [rax]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi],      mm0
+            movq    [rdi+8],    mm3
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+8]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+16],   mm0
+            movq    [rdi+24],   mm3
+
+            movq    mm0,        [rsi+rdx*2]
+            movq    mm1,        [rax+16]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+32],   mm0
+            movq    [rdi+40],   mm3
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+24]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+
+            movq    [rdi+48],   mm0
+            movq    [rdi+56],   mm3
+
+
+            add     rdi,        64
+            add     rax,        32
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi]
+            movq    mm1,        [rax]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi],      mm0
+            movq    [rdi+8],    mm3
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+8]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+16],   mm0
+            movq    [rdi+24],   mm3
+
+            movq    mm0,        [rsi+rdx*2]
+            movq    mm1,        [rax+16]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+            movq    [rdi+32],   mm0
+            movq    [rdi+40],   mm3
+            lea     rsi,        [rsi+rdx*2]
+
+
+            movq    mm0,        [rsi+rdx]
+            movq    mm1,        [rax+24]
+            movq    mm3,        mm0
+            movq    mm4,        mm1
+            punpcklbw   mm0,    mm7
+            punpcklbw   mm1,    mm7
+            punpckhbw   mm3,    mm7
+            punpckhbw   mm4,    mm7
+            psubw   mm0,        mm1
+            psubw   mm3,        mm4
+
+            movq    [rdi+48],   mm0
+            movq    [rdi+56],   mm3
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/libvpx/vp9/encoder/x86/vp9_subtract_sse2.asm b/libvpx/vp9/encoder/x86/vp9_subtract_sse2.asm
new file mode 100644
index 0000000..739d948
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_subtract_sse2.asm
@@ -0,0 +1,356 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_subtract_b_sse2_impl(unsigned char *z,  int src_stride,
+;                            short *diff, unsigned char *Predictor,
+;                            int pitch);
+global sym(vp9_subtract_b_sse2_impl) PRIVATE
+sym(vp9_subtract_b_sse2_impl):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov     rdi,        arg(2) ;diff
+        mov     rax,        arg(3) ;Predictor
+        mov     rsi,        arg(0) ;z
+        movsxd  rdx,        dword ptr arg(1);src_stride;
+        movsxd  rcx,        dword ptr arg(4);pitch
+        pxor    mm7,        mm7
+
+        movd    mm0,        [rsi]
+        movd    mm1,        [rax]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    MMWORD PTR [rdi],      mm0
+
+        movd    mm0,        [rsi+rdx]
+        movd    mm1,        [rax+rcx]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    MMWORD PTR [rdi+rcx*2], mm0
+
+        movd    mm0,        [rsi+rdx*2]
+        movd    mm1,        [rax+rcx*2]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    MMWORD PTR [rdi+rcx*4], mm0
+
+        lea     rsi,        [rsi+rdx*2]
+        lea     rcx,        [rcx+rcx*2]
+
+        movd    mm0,        [rsi+rdx]
+        movd    mm1,        [rax+rcx]
+        punpcklbw   mm0,    mm7
+        punpcklbw   mm1,    mm7
+        psubw   mm0,        mm1
+        movq    MMWORD PTR [rdi+rcx*2], mm0
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_subtract_mby_sse2(short *diff, unsigned char *src, unsigned char *pred, int stride)
+global sym(vp9_subtract_mby_sse2) PRIVATE
+sym(vp9_subtract_mby_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+            mov         rsi,            arg(1) ;src
+            mov         rdi,            arg(0) ;diff
+
+            mov         rax,            arg(2) ;pred
+            movsxd      rdx,            dword ptr arg(3) ;stride
+
+            mov         rcx,            8      ; do two lines at one time
+
+.submby_loop:
+            movdqa      xmm0,           XMMWORD PTR [rsi]   ; src
+            movdqa      xmm1,           XMMWORD PTR [rax]   ; pred
+
+            movdqa      xmm2,           xmm0
+            psubb       xmm0,           xmm1
+
+            pxor        xmm1,           [GLOBAL(t80)]   ;convert to signed values
+            pxor        xmm2,           [GLOBAL(t80)]
+            pcmpgtb     xmm1,           xmm2            ; obtain sign information
+
+            movdqa      xmm2,    xmm0
+            movdqa      xmm3,    xmm1
+            punpcklbw   xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw   xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa      XMMWORD PTR [rdi],   xmm0
+            movdqa      XMMWORD PTR [rdi +16], xmm2
+
+            movdqa      xmm4,           XMMWORD PTR [rsi + rdx]
+            movdqa      xmm5,           XMMWORD PTR [rax + 16]
+
+            movdqa      xmm6,           xmm4
+            psubb       xmm4,           xmm5
+
+            pxor        xmm5,           [GLOBAL(t80)]   ;convert to signed values
+            pxor        xmm6,           [GLOBAL(t80)]
+            pcmpgtb     xmm5,           xmm6            ; obtain sign information
+
+            movdqa      xmm6,    xmm4
+            movdqa      xmm7,    xmm5
+            punpcklbw   xmm4,    xmm5            ; put sign back to subtraction
+            punpckhbw   xmm6,    xmm7            ; put sign back to subtraction
+
+            movdqa      XMMWORD PTR [rdi +32], xmm4
+            movdqa      XMMWORD PTR [rdi +48], xmm6
+
+            add         rdi,            64
+            add         rax,            32
+            lea         rsi,            [rsi+rdx*2]
+
+            sub         rcx,            1
+            jnz         .submby_loop
+
+    pop rdi
+    pop rsi
+    ; begin epilog
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_subtract_mbuv_sse2(short *diff, unsigned char *usrc, unsigned char *vsrc, unsigned char *pred, int stride)
+global sym(vp9_subtract_mbuv_sse2) PRIVATE
+sym(vp9_subtract_mbuv_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 5
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+            mov     rdi,        arg(0) ;diff
+            mov     rax,        arg(3) ;pred
+            mov     rsi,        arg(1) ;z = usrc
+            add     rdi,        256*2  ;diff = diff + 256 (shorts)
+            add     rax,        256    ;Predictor = pred + 256
+            movsxd  rdx,        dword ptr arg(4) ;stride;
+            lea     rcx,        [rdx + rdx*2]
+
+            ;u
+            ;line 0 1
+            movq       xmm0,    MMWORD PTR [rsi]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rdx]
+            movdqa     xmm1,    XMMWORD PTR [rax]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi],   xmm0
+            movdqa     XMMWORD PTR [rdi +16],   xmm2
+
+            ;line 2 3
+            movq       xmm0,    MMWORD PTR [rsi+rdx*2]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rcx]
+            movdqa     xmm1,    XMMWORD PTR [rax+16]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 32],   xmm0
+            movdqa     XMMWORD PTR [rdi + 48],   xmm2
+
+            ;line 4 5
+            lea        rsi,     [rsi + rdx*4]
+
+            movq       xmm0,    MMWORD PTR [rsi]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rdx]
+            movdqa     xmm1,    XMMWORD PTR [rax + 32]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 64],   xmm0
+            movdqa     XMMWORD PTR [rdi + 80],   xmm2
+
+            ;line 6 7
+            movq       xmm0,    MMWORD PTR [rsi+rdx*2]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rcx]
+            movdqa     xmm1,    XMMWORD PTR [rax+ 48]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 96],   xmm0
+            movdqa     XMMWORD PTR [rdi + 112],  xmm2
+
+            ;v
+            mov     rsi,        arg(2) ;z = vsrc
+            add     rdi,        64*2  ;diff = diff + 320 (shorts)
+            add     rax,        64    ;Predictor = pred + 320
+
+            ;line 0 1
+            movq       xmm0,    MMWORD PTR [rsi]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rdx]
+            movdqa     xmm1,    XMMWORD PTR [rax]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi],   xmm0
+            movdqa     XMMWORD PTR [rdi +16],   xmm2
+
+            ;line 2 3
+            movq       xmm0,    MMWORD PTR [rsi+rdx*2]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rcx]
+            movdqa     xmm1,    XMMWORD PTR [rax+16]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 32],   xmm0
+            movdqa     XMMWORD PTR [rdi + 48],   xmm2
+
+            ;line 4 5
+            lea        rsi,     [rsi + rdx*4]
+
+            movq       xmm0,    MMWORD PTR [rsi]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rdx]
+            movdqa     xmm1,    XMMWORD PTR [rax + 32]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 64],   xmm0
+            movdqa     XMMWORD PTR [rdi + 80],   xmm2
+
+            ;line 6 7
+            movq       xmm0,    MMWORD PTR [rsi+rdx*2]  ; src
+            movq       xmm2,    MMWORD PTR [rsi+rcx]
+            movdqa     xmm1,    XMMWORD PTR [rax+ 48]  ; pred
+            punpcklqdq xmm0,    xmm2
+
+            movdqa     xmm2,    xmm0
+            psubb      xmm0,    xmm1            ; subtraction with sign missed
+
+            pxor       xmm1,    [GLOBAL(t80)]   ;convert to signed values
+            pxor       xmm2,    [GLOBAL(t80)]
+            pcmpgtb    xmm1,    xmm2            ; obtain sign information
+
+            movdqa     xmm2,    xmm0
+            movdqa     xmm3,    xmm1
+            punpcklbw  xmm0,    xmm1            ; put sign back to subtraction
+            punpckhbw  xmm2,    xmm3            ; put sign back to subtraction
+
+            movdqa     XMMWORD PTR [rdi + 96],   xmm0
+            movdqa     XMMWORD PTR [rdi + 112],  xmm2
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+t80:
+    times 16 db 0x80
diff --git a/libvpx/vp9/encoder/x86/vp9_temporal_filter_apply_sse2.asm b/libvpx/vp9/encoder/x86/vp9_temporal_filter_apply_sse2.asm
new file mode 100644
index 0000000..d2d13b3
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_temporal_filter_apply_sse2.asm
@@ -0,0 +1,207 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+; void vp9_temporal_filter_apply_sse2 | arg
+;  (unsigned char  *frame1,           |  0
+;   unsigned int    stride,           |  1
+;   unsigned char  *frame2,           |  2
+;   unsigned int    block_size,       |  3
+;   int             strength,         |  4
+;   int             filter_weight,    |  5
+;   unsigned int   *accumulator,      |  6
+;   unsigned short *count)            |  7
+global sym(vp9_temporal_filter_apply_sse2) PRIVATE
+sym(vp9_temporal_filter_apply_sse2):
+
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 8
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push        rsi
+    push        rdi
+    ALIGN_STACK 16, rax
+    %define block_size    0
+    %define strength      16
+    %define filter_weight 32
+    %define rounding_bit  48
+    %define rbp_backup    64
+    %define stack_size    80
+    sub         rsp,           stack_size
+    mov         [rsp + rbp_backup], rbp
+    ; end prolog
+
+        mov         rdx,            arg(3)
+        mov         [rsp + block_size], rdx
+        movd        xmm6,            arg(4)
+        movdqa      [rsp + strength], xmm6 ; where strength is used, all 16 bytes are read
+
+        ; calculate the rounding bit outside the loop
+        ; 0x8000 >> (16 - strength)
+        mov         rdx,            16
+        sub         rdx,            arg(4) ; 16 - strength
+        movq        xmm4,           rdx    ; can't use rdx w/ shift
+        movdqa      xmm5,           [GLOBAL(_const_top_bit)]
+        psrlw       xmm5,           xmm4
+        movdqa      [rsp + rounding_bit], xmm5
+
+        mov         rsi,            arg(0) ; src/frame1
+        mov         rdx,            arg(2) ; predictor frame
+        mov         rdi,            arg(6) ; accumulator
+        mov         rax,            arg(7) ; count
+
+        ; dup the filter weight and store for later
+        movd        xmm0,           arg(5) ; filter_weight
+        pshuflw     xmm0,           xmm0, 0
+        punpcklwd   xmm0,           xmm0
+        movdqa      [rsp + filter_weight], xmm0
+
+        mov         rbp,            arg(1) ; stride
+        pxor        xmm7,           xmm7   ; zero for extraction
+
+        lea         rcx,            [rdx + 16*16*1]
+        cmp         dword ptr [rsp + block_size], 8
+        jne         .temporal_filter_apply_load_16
+        lea         rcx,            [rdx + 8*8*1]
+
+.temporal_filter_apply_load_8:
+        movq        xmm0,           [rsi]  ; first row
+        lea         rsi,            [rsi + rbp] ; += stride
+        punpcklbw   xmm0,           xmm7   ; src[ 0- 7]
+        movq        xmm1,           [rsi]  ; second row
+        lea         rsi,            [rsi + rbp] ; += stride
+        punpcklbw   xmm1,           xmm7   ; src[ 8-15]
+        jmp         .temporal_filter_apply_load_finished
+
+.temporal_filter_apply_load_16:
+        movdqa      xmm0,           [rsi]  ; src (frame1)
+        lea         rsi,            [rsi + rbp] ; += stride
+        movdqa      xmm1,           xmm0
+        punpcklbw   xmm0,           xmm7   ; src[ 0- 7]
+        punpckhbw   xmm1,           xmm7   ; src[ 8-15]
+
+.temporal_filter_apply_load_finished:
+        movdqa      xmm2,           [rdx]  ; predictor (frame2)
+        movdqa      xmm3,           xmm2
+        punpcklbw   xmm2,           xmm7   ; pred[ 0- 7]
+        punpckhbw   xmm3,           xmm7   ; pred[ 8-15]
+
+        ; modifier = src_byte - pixel_value
+        psubw       xmm0,           xmm2   ; src - pred[ 0- 7]
+        psubw       xmm1,           xmm3   ; src - pred[ 8-15]
+
+        ; modifier *= modifier
+        pmullw      xmm0,           xmm0   ; modifer[ 0- 7]^2
+        pmullw      xmm1,           xmm1   ; modifer[ 8-15]^2
+
+        ; modifier *= 3
+        pmullw      xmm0,           [GLOBAL(_const_3w)]
+        pmullw      xmm1,           [GLOBAL(_const_3w)]
+
+        ; modifer += 0x8000 >> (16 - strength)
+        paddw       xmm0,           [rsp + rounding_bit]
+        paddw       xmm1,           [rsp + rounding_bit]
+
+        ; modifier >>= strength
+        psrlw       xmm0,           [rsp + strength]
+        psrlw       xmm1,           [rsp + strength]
+
+        ; modifier = 16 - modifier
+        ; saturation takes care of modifier > 16
+        movdqa      xmm3,           [GLOBAL(_const_16w)]
+        movdqa      xmm2,           [GLOBAL(_const_16w)]
+        psubusw     xmm3,           xmm1
+        psubusw     xmm2,           xmm0
+
+        ; modifier *= filter_weight
+        pmullw      xmm2,           [rsp + filter_weight]
+        pmullw      xmm3,           [rsp + filter_weight]
+
+        ; count
+        movdqa      xmm4,           [rax]
+        movdqa      xmm5,           [rax+16]
+        ; += modifier
+        paddw       xmm4,           xmm2
+        paddw       xmm5,           xmm3
+        ; write back
+        movdqa      [rax],          xmm4
+        movdqa      [rax+16],       xmm5
+        lea         rax,            [rax + 16*2] ; count += 16*(sizeof(short))
+
+        ; load and extract the predictor up to shorts
+        pxor        xmm7,           xmm7
+        movdqa      xmm0,           [rdx]
+        lea         rdx,            [rdx + 16*1] ; pred += 16*(sizeof(char))
+        movdqa      xmm1,           xmm0
+        punpcklbw   xmm0,           xmm7   ; pred[ 0- 7]
+        punpckhbw   xmm1,           xmm7   ; pred[ 8-15]
+
+        ; modifier *= pixel_value
+        pmullw      xmm0,           xmm2
+        pmullw      xmm1,           xmm3
+
+        ; expand to double words
+        movdqa      xmm2,           xmm0
+        punpcklwd   xmm0,           xmm7   ; [ 0- 3]
+        punpckhwd   xmm2,           xmm7   ; [ 4- 7]
+        movdqa      xmm3,           xmm1
+        punpcklwd   xmm1,           xmm7   ; [ 8-11]
+        punpckhwd   xmm3,           xmm7   ; [12-15]
+
+        ; accumulator
+        movdqa      xmm4,           [rdi]
+        movdqa      xmm5,           [rdi+16]
+        movdqa      xmm6,           [rdi+32]
+        movdqa      xmm7,           [rdi+48]
+        ; += modifier
+        paddd       xmm4,           xmm0
+        paddd       xmm5,           xmm2
+        paddd       xmm6,           xmm1
+        paddd       xmm7,           xmm3
+        ; write back
+        movdqa      [rdi],          xmm4
+        movdqa      [rdi+16],       xmm5
+        movdqa      [rdi+32],       xmm6
+        movdqa      [rdi+48],       xmm7
+        lea         rdi,            [rdi + 16*4] ; accumulator += 16*(sizeof(int))
+
+        cmp         rdx,            rcx
+        je          .temporal_filter_apply_epilog
+        pxor        xmm7,           xmm7   ; zero for extraction
+        cmp         dword ptr [rsp + block_size], 16
+        je          .temporal_filter_apply_load_16
+        jmp         .temporal_filter_apply_load_8
+
+.temporal_filter_apply_epilog:
+    ; begin epilog
+    mov         rbp,            [rsp + rbp_backup]
+    add         rsp,            stack_size
+    pop         rsp
+    pop         rdi
+    pop         rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+align 16
+_const_3w:
+    times 8 dw 3
+align 16
+_const_top_bit:
+    times 8 dw 1<<15
+align 16
+_const_16w
+    times 8 dw 16
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_impl_mmx.asm b/libvpx/vp9/encoder/x86/vp9_variance_impl_mmx.asm
new file mode 100644
index 0000000..9f140c9
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_impl_mmx.asm
@@ -0,0 +1,851 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;unsigned int vp9_get_mb_ss_mmx( short *src_ptr )
+global sym(vp9_get_mb_ss_mmx) PRIVATE
+sym(vp9_get_mb_ss_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 8
+    ; end prolog
+
+        mov         rax, arg(0) ;src_ptr
+        mov         rcx, 16
+        pxor        mm4, mm4
+
+.NEXTROW:
+        movq        mm0, [rax]
+        movq        mm1, [rax+8]
+        movq        mm2, [rax+16]
+        movq        mm3, [rax+24]
+        pmaddwd     mm0, mm0
+        pmaddwd     mm1, mm1
+        pmaddwd     mm2, mm2
+        pmaddwd     mm3, mm3
+
+        paddd       mm4, mm0
+        paddd       mm4, mm1
+        paddd       mm4, mm2
+        paddd       mm4, mm3
+
+        add         rax, 32
+        dec         rcx
+        ja          .NEXTROW
+        movq        QWORD PTR [rsp], mm4
+
+        ;return sum[0]+sum[1];
+        movsxd      rax, dword ptr [rsp]
+        movsxd      rcx, dword ptr [rsp+4]
+        add         rax, rcx
+
+
+    ; begin epilog
+    add rsp, 8
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_get8x8var_mmx
+;(
+;    unsigned char *src_ptr,
+;    int  source_stride,
+;    unsigned char *ref_ptr,
+;    int  recon_stride,
+;    unsigned int *SSE,
+;    int *Sum
+;)
+global sym(vp9_get8x8var_mmx) PRIVATE
+sym(vp9_get8x8var_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    push rsi
+    push rdi
+    push rbx
+    sub         rsp, 16
+    ; end prolog
+
+
+        pxor        mm5, mm5                    ; Blank mmx6
+        pxor        mm6, mm6                    ; Blank mmx7
+        pxor        mm7, mm7                    ; Blank mmx7
+
+        mov         rax, arg(0) ;[src_ptr]  ; Load base addresses
+        mov         rbx, arg(2) ;[ref_ptr]
+        movsxd      rcx, dword ptr arg(1) ;[source_stride]
+        movsxd      rdx, dword ptr arg(3) ;[recon_stride]
+
+        ; Row 1
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+
+        ; Row 2
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 3
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 4
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 5
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        ;              movq        mm4, [rbx + rdx]
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 6
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 7
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Row 8
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm2, mm0                    ; Take copies
+        movq        mm3, mm1                    ; Take copies
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        punpckhbw   mm2, mm6                    ; unpack to higher prrcision
+        punpckhbw   mm3, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        psubsw      mm2, mm3                    ; A-B (high order) to MM2
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        paddw       mm5, mm2                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        pmaddwd     mm2, mm2                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        paddd       mm7, mm0                    ; accumulate in mm7
+        paddd       mm7, mm2                    ; accumulate in mm7
+
+        ; Now accumulate the final results.
+        movq        QWORD PTR [rsp+8], mm5      ; copy back accumulated results into normal memory
+        movq        QWORD PTR [rsp], mm7        ; copy back accumulated results into normal memory
+        movsx       rdx, WORD PTR [rsp+8]
+        movsx       rcx, WORD PTR [rsp+10]
+        movsx       rbx, WORD PTR [rsp+12]
+        movsx       rax, WORD PTR [rsp+14]
+        add         rdx, rcx
+        add         rbx, rax
+        add         rdx, rbx    ;XSum
+        movsxd      rax, DWORD PTR [rsp]
+        movsxd      rcx, DWORD PTR [rsp+4]
+        add         rax, rcx    ;XXSum
+        mov         rsi, arg(4) ;SSE
+        mov         rdi, arg(5) ;Sum
+        mov         dword ptr [rsi], eax
+        mov         dword ptr [rdi], edx
+        xor         rax, rax    ; return 0
+
+
+    ; begin epilog
+    add rsp, 16
+    pop rbx
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+;unsigned int
+;vp9_get4x4var_mmx
+;(
+;    unsigned char *src_ptr,
+;    int  source_stride,
+;    unsigned char *ref_ptr,
+;    int  recon_stride,
+;    unsigned int *SSE,
+;    int *Sum
+;)
+global sym(vp9_get4x4var_mmx) PRIVATE
+sym(vp9_get4x4var_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    push rsi
+    push rdi
+    push rbx
+    sub         rsp, 16
+    ; end prolog
+
+
+        pxor        mm5, mm5                    ; Blank mmx6
+        pxor        mm6, mm6                    ; Blank mmx7
+        pxor        mm7, mm7                    ; Blank mmx7
+
+        mov         rax, arg(0) ;[src_ptr]  ; Load base addresses
+        mov         rbx, arg(2) ;[ref_ptr]
+        movsxd      rcx, dword ptr arg(1) ;[source_stride]
+        movsxd      rdx, dword ptr arg(3) ;[recon_stride]
+
+        ; Row 1
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+
+        ; Row 2
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        ; Row 3
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movq        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        ; Row 4
+        movq        mm0, [rax]                  ; Copy eight bytes to mm0
+
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+
+        paddw       mm5, mm0                    ; accumulate differences in mm5
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+
+        ; Now accumulate the final results.
+        movq        QWORD PTR [rsp+8], mm5      ; copy back accumulated results into normal memory
+        movq        QWORD PTR [rsp], mm7        ; copy back accumulated results into normal memory
+        movsx       rdx, WORD PTR [rsp+8]
+        movsx       rcx, WORD PTR [rsp+10]
+        movsx       rbx, WORD PTR [rsp+12]
+        movsx       rax, WORD PTR [rsp+14]
+        add         rdx, rcx
+        add         rbx, rax
+        add         rdx, rbx    ;XSum
+        movsxd      rax, DWORD PTR [rsp]
+        movsxd      rcx, DWORD PTR [rsp+4]
+        add         rax, rcx    ;XXSum
+        mov         rsi, arg(4) ;SSE
+        mov         rdi, arg(5) ;Sum
+        mov         dword ptr [rsi], eax
+        mov         dword ptr [rdi], edx
+        xor         rax, rax    ; return 0
+
+
+    ; begin epilog
+    add rsp, 16
+    pop rbx
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+;unsigned int
+;vp9_get4x4sse_cs_mmx
+;(
+;    unsigned char *src_ptr,
+;    int  source_stride,
+;    unsigned char *ref_ptr,
+;    int  recon_stride
+;)
+global sym(vp9_get4x4sse_cs_mmx) PRIVATE
+sym(vp9_get4x4sse_cs_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    push rsi
+    push rdi
+    push rbx
+    ; end prolog
+
+
+        pxor        mm6, mm6                    ; Blank mmx7
+        pxor        mm7, mm7                    ; Blank mmx7
+
+        mov         rax, arg(0) ;[src_ptr]  ; Load base addresses
+        mov         rbx, arg(2) ;[ref_ptr]
+        movsxd      rcx, dword ptr arg(1) ;[source_stride]
+        movsxd      rdx, dword ptr arg(3) ;[recon_stride]
+        ; Row 1
+        movd        mm0, [rax]                  ; Copy eight bytes to mm0
+        movd        mm1, [rbx]                  ; Copy eight bytes to mm1
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movd        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        ; Row 2
+        movd        mm0, [rax]                  ; Copy eight bytes to mm0
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movd        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        ; Row 3
+        movd        mm0, [rax]                  ; Copy eight bytes to mm0
+        punpcklbw   mm1, mm6
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        add         rbx,rdx                     ; Inc pointer into ref data
+        add         rax,rcx                     ; Inc pointer into the new data
+        movd        mm1, [rbx]                  ; Copy eight bytes to mm1
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        ; Row 4
+        movd        mm0, [rax]                  ; Copy eight bytes to mm0
+        punpcklbw   mm0, mm6                    ; unpack to higher prrcision
+        punpcklbw   mm1, mm6
+        psubsw      mm0, mm1                    ; A-B (low order) to MM0
+        pmaddwd     mm0, mm0                    ; square and accumulate
+        paddd       mm7, mm0                    ; accumulate in mm7
+
+        movq        mm0,    mm7                 ;
+        psrlq       mm7,    32
+
+        paddd       mm0,    mm7
+        movq        rax,    mm0
+
+
+    ; begin epilog
+    pop rbx
+    pop rdi
+    pop rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%define mmx_filter_shift            7
+
+;void vp9_filter_block2d_bil4x4_var_mmx
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned short *HFilter,
+;    unsigned short *VFilter,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_filter_block2d_bil4x4_var_mmx) PRIVATE
+sym(vp9_filter_block2d_bil4x4_var_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 8
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 16
+    ; end prolog
+
+
+        pxor            mm6,            mm6                 ;
+        pxor            mm7,            mm7                 ;
+
+        mov             rax,            arg(4) ;HFilter             ;
+        mov             rdx,            arg(5) ;VFilter             ;
+
+        mov             rsi,            arg(0) ;ref_ptr              ;
+        mov             rdi,            arg(2) ;src_ptr              ;
+
+        mov             rcx,            4                   ;
+        pxor            mm0,            mm0                 ;
+
+        movd            mm1,            [rsi]               ;
+        movd            mm3,            [rsi+1]             ;
+
+        punpcklbw       mm1,            mm0                 ;
+        pmullw          mm1,            [rax]               ;
+
+        punpcklbw       mm3,            mm0                 ;
+        pmullw          mm3,            [rax+8]             ;
+
+        paddw           mm1,            mm3                 ;
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+
+        psraw           mm1,            mmx_filter_shift    ;
+        movq            mm5,            mm1
+
+%if ABI_IS_32BIT
+        add             rsi, dword ptr  arg(1) ;ref_pixels_per_line    ;
+%else
+        movsxd          r8, dword ptr  arg(1) ;ref_pixels_per_line    ;
+        add             rsi, r8
+%endif
+
+.filter_block2d_bil4x4_var_mmx_loop:
+
+        movd            mm1,            [rsi]               ;
+        movd            mm3,            [rsi+1]             ;
+
+        punpcklbw       mm1,            mm0                 ;
+        pmullw          mm1,            [rax]               ;
+
+        punpcklbw       mm3,            mm0                 ;
+        pmullw          mm3,            [rax+8]             ;
+
+        paddw           mm1,            mm3                 ;
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+
+        psraw           mm1,            mmx_filter_shift    ;
+        movq            mm3,            mm5                 ;
+
+        movq            mm5,            mm1                 ;
+        pmullw          mm3,            [rdx]               ;
+
+        pmullw          mm1,            [rdx+8]             ;
+        paddw           mm1,            mm3                 ;
+
+
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+        psraw           mm1,            mmx_filter_shift    ;
+
+        movd            mm3,            [rdi]               ;
+        punpcklbw       mm3,            mm0                 ;
+
+        psubw           mm1,            mm3                 ;
+        paddw           mm6,            mm1                 ;
+
+        pmaddwd         mm1,            mm1                 ;
+        paddd           mm7,            mm1                 ;
+
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1) ;ref_pixels_per_line    ;
+        add             rdi,            dword ptr arg(3) ;src_pixels_per_line    ;
+%else
+        movsxd          r8,             dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line
+        add             rsi,            r8
+        add             rdi,            r9
+%endif
+        sub             rcx,            1                   ;
+        jnz             .filter_block2d_bil4x4_var_mmx_loop       ;
+
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rdi,            arg(6) ;sum
+        mov             rsi,            arg(7) ;sumsquared
+
+        movd            dword ptr [rdi],          mm2                 ;
+        movd            dword ptr [rsi],          mm4                 ;
+
+
+
+    ; begin epilog
+    add rsp, 16
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+
+;void vp9_filter_block2d_bil_var_mmx
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    unsigned short *HFilter,
+;    unsigned short *VFilter,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_filter_block2d_bil_var_mmx) PRIVATE
+sym(vp9_filter_block2d_bil_var_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 16
+    ; end prolog
+
+        pxor            mm6,            mm6                 ;
+        pxor            mm7,            mm7                 ;
+        mov             rax,            arg(5) ;HFilter             ;
+
+        mov             rdx,            arg(6) ;VFilter             ;
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+
+        pxor            mm0,            mm0                 ;
+        movq            mm1,            [rsi]               ;
+
+        movq            mm3,            [rsi+1]             ;
+        movq            mm2,            mm1                 ;
+
+        movq            mm4,            mm3                 ;
+        punpcklbw       mm1,            mm0                 ;
+
+        punpckhbw       mm2,            mm0                 ;
+        pmullw          mm1,            [rax]               ;
+
+        pmullw          mm2,            [rax]               ;
+        punpcklbw       mm3,            mm0                 ;
+
+        punpckhbw       mm4,            mm0                 ;
+        pmullw          mm3,            [rax+8]             ;
+
+        pmullw          mm4,            [rax+8]             ;
+        paddw           mm1,            mm3                 ;
+
+        paddw           mm2,            mm4                 ;
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+
+        psraw           mm1,            mmx_filter_shift    ;
+        paddw           mm2,            [GLOBAL(mmx_bi_rd)] ;
+
+        psraw           mm2,            mmx_filter_shift    ;
+        movq            mm5,            mm1
+
+        packuswb        mm5,            mm2                 ;
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1) ;ref_pixels_per_line
+%else
+        movsxd          r8,             dword ptr arg(1) ;ref_pixels_per_line
+        add             rsi,            r8
+%endif
+
+.filter_block2d_bil_var_mmx_loop:
+
+        movq            mm1,            [rsi]               ;
+        movq            mm3,            [rsi+1]             ;
+
+        movq            mm2,            mm1                 ;
+        movq            mm4,            mm3                 ;
+
+        punpcklbw       mm1,            mm0                 ;
+        punpckhbw       mm2,            mm0                 ;
+
+        pmullw          mm1,            [rax]               ;
+        pmullw          mm2,            [rax]               ;
+
+        punpcklbw       mm3,            mm0                 ;
+        punpckhbw       mm4,            mm0                 ;
+
+        pmullw          mm3,            [rax+8]             ;
+        pmullw          mm4,            [rax+8]             ;
+
+        paddw           mm1,            mm3                 ;
+        paddw           mm2,            mm4                 ;
+
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+        psraw           mm1,            mmx_filter_shift    ;
+
+        paddw           mm2,            [GLOBAL(mmx_bi_rd)] ;
+        psraw           mm2,            mmx_filter_shift    ;
+
+        movq            mm3,            mm5                 ;
+        movq            mm4,            mm5                 ;
+
+        punpcklbw       mm3,            mm0                 ;
+        punpckhbw       mm4,            mm0                 ;
+
+        movq            mm5,            mm1                 ;
+        packuswb        mm5,            mm2                 ;
+
+        pmullw          mm3,            [rdx]               ;
+        pmullw          mm4,            [rdx]               ;
+
+        pmullw          mm1,            [rdx+8]             ;
+        pmullw          mm2,            [rdx+8]             ;
+
+        paddw           mm1,            mm3                 ;
+        paddw           mm2,            mm4                 ;
+
+        paddw           mm1,            [GLOBAL(mmx_bi_rd)] ;
+        paddw           mm2,            [GLOBAL(mmx_bi_rd)] ;
+
+        psraw           mm1,            mmx_filter_shift    ;
+        psraw           mm2,            mmx_filter_shift    ;
+
+        movq            mm3,            [rdi]               ;
+        movq            mm4,            mm3                 ;
+
+        punpcklbw       mm3,            mm0                 ;
+        punpckhbw       mm4,            mm0                 ;
+
+        psubw           mm1,            mm3                 ;
+        psubw           mm2,            mm4                 ;
+
+        paddw           mm6,            mm1                 ;
+        pmaddwd         mm1,            mm1                 ;
+
+        paddw           mm6,            mm2                 ;
+        pmaddwd         mm2,            mm2                 ;
+
+        paddd           mm7,            mm1                 ;
+        paddd           mm7,            mm2                 ;
+
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1) ;ref_pixels_per_line    ;
+        add             rdi,            dword ptr arg(3) ;src_pixels_per_line    ;
+%else
+        movsxd          r8,             dword ptr arg(1) ;ref_pixels_per_line    ;
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line    ;
+        add             rsi,            r8
+        add             rdi,            r9
+%endif
+        sub             rcx,            1                   ;
+        jnz             .filter_block2d_bil_var_mmx_loop       ;
+
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rdi,            arg(7) ;sum
+        mov             rsi,            arg(8) ;sumsquared
+
+        movd            dword ptr [rdi],          mm2                 ;
+        movd            dword ptr [rsi],          mm4                 ;
+
+    ; begin epilog
+    add rsp, 16
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+SECTION_RODATA
+;short mmx_bi_rd[4] = { 64, 64, 64, 64};
+align 16
+mmx_bi_rd:
+    times 4 dw 64
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_impl_sse2.asm b/libvpx/vp9/encoder/x86/vp9_variance_impl_sse2.asm
new file mode 100644
index 0000000..896dd18
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_impl_sse2.asm
@@ -0,0 +1,761 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%define xmm_filter_shift            7
+
+;unsigned int vp9_get_mb_ss_sse2
+;(
+;    short *src_ptr
+;)
+global sym(vp9_get_mb_ss_sse2) PRIVATE
+sym(vp9_get_mb_ss_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 1
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 16
+    ; end prolog
+
+
+        mov         rax, arg(0) ;[src_ptr]
+        mov         rcx, 8
+        pxor        xmm4, xmm4
+
+.NEXTROW:
+        movdqa      xmm0, [rax]
+        movdqa      xmm1, [rax+16]
+        movdqa      xmm2, [rax+32]
+        movdqa      xmm3, [rax+48]
+        pmaddwd     xmm0, xmm0
+        pmaddwd     xmm1, xmm1
+        pmaddwd     xmm2, xmm2
+        pmaddwd     xmm3, xmm3
+
+        paddd       xmm0, xmm1
+        paddd       xmm2, xmm3
+        paddd       xmm4, xmm0
+        paddd       xmm4, xmm2
+
+        add         rax, 0x40
+        dec         rcx
+        ja          .NEXTROW
+
+        movdqa      xmm3,xmm4
+        psrldq      xmm4,8
+        paddd       xmm4,xmm3
+        movdqa      xmm3,xmm4
+        psrldq      xmm4,4
+        paddd       xmm4,xmm3
+        movq        rax,xmm4
+
+
+    ; begin epilog
+    add rsp, 16
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;unsigned int vp9_get16x16var_sse2
+;(
+;    unsigned char   *  src_ptr,
+;    int             source_stride,
+;    unsigned char   *  ref_ptr,
+;    int             recon_stride,
+;    unsigned int    *  SSE,
+;    int             *  Sum
+;)
+global sym(vp9_get16x16var_sse2) PRIVATE
+sym(vp9_get16x16var_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    push rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        mov         rsi,            arg(0) ;[src_ptr]
+        mov         rdi,            arg(2) ;[ref_ptr]
+
+        movsxd      rax,            DWORD PTR arg(1) ;[source_stride]
+        movsxd      rdx,            DWORD PTR arg(3) ;[recon_stride]
+
+        ; Prefetch data
+        lea             rcx,    [rax+rax*2]
+        prefetcht0      [rsi]
+        prefetcht0      [rsi+rax]
+        prefetcht0      [rsi+rax*2]
+        prefetcht0      [rsi+rcx]
+        lea             rbx,    [rsi+rax*4]
+        prefetcht0      [rbx]
+        prefetcht0      [rbx+rax]
+        prefetcht0      [rbx+rax*2]
+        prefetcht0      [rbx+rcx]
+
+        lea             rcx,    [rdx+rdx*2]
+        prefetcht0      [rdi]
+        prefetcht0      [rdi+rdx]
+        prefetcht0      [rdi+rdx*2]
+        prefetcht0      [rdi+rcx]
+        lea             rbx,    [rdi+rdx*4]
+        prefetcht0      [rbx]
+        prefetcht0      [rbx+rdx]
+        prefetcht0      [rbx+rdx*2]
+        prefetcht0      [rbx+rcx]
+
+        pxor        xmm0,           xmm0                        ; clear xmm0 for unpack
+        pxor        xmm7,           xmm7                        ; clear xmm7 for accumulating diffs
+
+        pxor        xmm6,           xmm6                        ; clear xmm6 for accumulating sse
+        mov         rcx,            16
+
+.var16loop:
+        movdqu      xmm1,           XMMWORD PTR [rsi]
+        movdqu      xmm2,           XMMWORD PTR [rdi]
+
+        prefetcht0      [rsi+rax*8]
+        prefetcht0      [rdi+rdx*8]
+
+        movdqa      xmm3,           xmm1
+        movdqa      xmm4,           xmm2
+
+
+        punpcklbw   xmm1,           xmm0
+        punpckhbw   xmm3,           xmm0
+
+        punpcklbw   xmm2,           xmm0
+        punpckhbw   xmm4,           xmm0
+
+
+        psubw       xmm1,           xmm2
+        psubw       xmm3,           xmm4
+
+        paddw       xmm7,           xmm1
+        pmaddwd     xmm1,           xmm1
+
+        paddw       xmm7,           xmm3
+        pmaddwd     xmm3,           xmm3
+
+        paddd       xmm6,           xmm1
+        paddd       xmm6,           xmm3
+
+        add         rsi,            rax
+        add         rdi,            rdx
+
+        sub         rcx,            1
+        jnz         .var16loop
+
+
+        movdqa      xmm1,           xmm6
+        pxor        xmm6,           xmm6
+
+        pxor        xmm5,           xmm5
+        punpcklwd   xmm6,           xmm7
+
+        punpckhwd   xmm5,           xmm7
+        psrad       xmm5,           16
+
+        psrad       xmm6,           16
+        paddd       xmm6,           xmm5
+
+        movdqa      xmm2,           xmm1
+        punpckldq   xmm1,           xmm0
+
+        punpckhdq   xmm2,           xmm0
+        movdqa      xmm7,           xmm6
+
+        paddd       xmm1,           xmm2
+        punpckldq   xmm6,           xmm0
+
+        punpckhdq   xmm7,           xmm0
+        paddd       xmm6,           xmm7
+
+        movdqa      xmm2,           xmm1
+        movdqa      xmm7,           xmm6
+
+        psrldq      xmm1,           8
+        psrldq      xmm6,           8
+
+        paddd       xmm7,           xmm6
+        paddd       xmm1,           xmm2
+
+        mov         rax,            arg(5) ;[Sum]
+        mov         rdi,            arg(4) ;[SSE]
+
+        movd DWORD PTR [rax],       xmm7
+        movd DWORD PTR [rdi],       xmm1
+
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    pop rbx
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+
+
+;unsigned int vp9_get8x8var_sse2
+;(
+;    unsigned char   *  src_ptr,
+;    int             source_stride,
+;    unsigned char   *  ref_ptr,
+;    int             recon_stride,
+;    unsigned int    *  SSE,
+;    int             *  Sum
+;)
+global sym(vp9_get8x8var_sse2) PRIVATE
+sym(vp9_get8x8var_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 16
+    ; end prolog
+
+        mov         rsi,            arg(0) ;[src_ptr]
+        mov         rdi,            arg(2) ;[ref_ptr]
+
+        movsxd      rax,            DWORD PTR arg(1) ;[source_stride]
+        movsxd      rdx,            DWORD PTR arg(3) ;[recon_stride]
+
+        pxor        xmm0,           xmm0                        ; clear xmm0 for unpack
+        pxor        xmm7,           xmm7                        ; clear xmm7 for accumulating diffs
+
+        movq        xmm1,           QWORD PTR [rsi]
+        movq        xmm2,           QWORD PTR [rdi]
+
+        punpcklbw   xmm1,           xmm0
+        punpcklbw   xmm2,           xmm0
+
+        psubsw      xmm1,           xmm2
+        paddw       xmm7,           xmm1
+
+        pmaddwd     xmm1,           xmm1
+
+        movq        xmm2,           QWORD PTR[rsi + rax]
+        movq        xmm3,           QWORD PTR[rdi + rdx]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+
+        movq        xmm2,           QWORD PTR[rsi + rax * 2]
+        movq        xmm3,           QWORD PTR[rdi + rdx * 2]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+
+        lea         rsi,            [rsi + rax * 2]
+        lea         rdi,            [rdi + rdx * 2]
+        movq        xmm2,           QWORD PTR[rsi + rax]
+        movq        xmm3,           QWORD PTR[rdi + rdx]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+        movq        xmm2,           QWORD PTR[rsi + rax *2]
+        movq        xmm3,           QWORD PTR[rdi + rdx *2]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+
+        lea         rsi,            [rsi + rax * 2]
+        lea         rdi,            [rdi + rdx * 2]
+
+
+        movq        xmm2,           QWORD PTR[rsi + rax]
+        movq        xmm3,           QWORD PTR[rdi + rdx]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+        movq        xmm2,           QWORD PTR[rsi + rax *2]
+        movq        xmm3,           QWORD PTR[rdi + rdx *2]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+
+        lea         rsi,            [rsi + rax * 2]
+        lea         rdi,            [rdi + rdx * 2]
+
+        movq        xmm2,           QWORD PTR[rsi + rax]
+        movq        xmm3,           QWORD PTR[rdi + rdx]
+
+        punpcklbw   xmm2,           xmm0
+        punpcklbw   xmm3,           xmm0
+
+        psubsw      xmm2,           xmm3
+        paddw       xmm7,           xmm2
+
+        pmaddwd     xmm2,           xmm2
+        paddd       xmm1,           xmm2
+
+
+        movdqa      xmm6,           xmm7
+        punpcklwd   xmm6,           xmm0
+
+        punpckhwd   xmm7,           xmm0
+        movdqa      xmm2,           xmm1
+
+        paddw       xmm6,           xmm7
+        punpckldq   xmm1,           xmm0
+
+        punpckhdq   xmm2,           xmm0
+        movdqa      xmm7,           xmm6
+
+        paddd       xmm1,           xmm2
+        punpckldq   xmm6,           xmm0
+
+        punpckhdq   xmm7,           xmm0
+        paddw       xmm6,           xmm7
+
+        movdqa      xmm2,           xmm1
+        movdqa      xmm7,           xmm6
+
+        psrldq      xmm1,           8
+        psrldq      xmm6,           8
+
+        paddw       xmm7,           xmm6
+        paddd       xmm1,           xmm2
+
+        mov         rax,            arg(5) ;[Sum]
+        mov         rdi,            arg(4) ;[SSE]
+
+        movq        rdx,            xmm7
+        movsx       rcx,            dx
+
+        mov  dword ptr [rax],       ecx
+        movd DWORD PTR [rdi],       xmm1
+
+    ; begin epilog
+    add rsp, 16
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_half_horiz_vert_variance8x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_horiz_vert_variance8x_h_sse2) PRIVATE
+sym(vp9_half_horiz_vert_variance8x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+%if ABI_IS_32BIT=0
+    movsxd          r8, dword ptr arg(1) ;ref_pixels_per_line
+    movsxd          r9, dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+
+        movq            xmm5,           QWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s8
+        movq            xmm3,           QWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s9
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3) horizontal line 1
+
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1) ;ref_pixels_per_line    ;  next source
+%else
+        add             rsi, r8
+%endif
+
+.half_horiz_vert_variance8x_h_1:
+
+        movq            xmm1,           QWORD PTR [rsi]     ;
+        movq            xmm2,           QWORD PTR [rsi+1]   ;
+        pavgb           xmm1,           xmm2                ;  xmm1 = avg(xmm1,xmm3) horizontal line i+1
+
+        pavgb           xmm5,           xmm1                ;  xmm = vertical average of the above
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d8
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+
+        movdqa          xmm5,           xmm1                ;  save xmm1 for use on the next row
+
+%if ABI_IS_32BIT
+        add             esi,            dword ptr arg(1) ;ref_pixels_per_line    ;  next source
+        add             edi,            dword ptr arg(3) ;src_pixels_per_line    ;  next destination
+%else
+        add             rsi, r8
+        add             rdi, r9
+%endif
+
+        sub             rcx,            1                   ;
+        jnz             .half_horiz_vert_variance8x_h_1     ;
+
+        movdq2q         mm6,            xmm6                ;
+        movdq2q         mm7,            xmm7                ;
+
+        psrldq          xmm6,           8
+        psrldq          xmm7,           8
+
+        movdq2q         mm2,            xmm6
+        movdq2q         mm3,            xmm7
+
+        paddw           mm6,            mm2
+        paddd           mm7,            mm3
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rsi,            arg(5) ; sum
+        mov             rdi,            arg(6) ; sumsquared
+
+        movd            [rsi],          mm2                 ;
+        movd            [rdi],          mm4                 ;
+
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_half_vert_variance8x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_vert_variance8x_h_sse2) PRIVATE
+sym(vp9_half_vert_variance8x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+%if ABI_IS_32BIT=0
+    movsxd          r8, dword ptr arg(1) ;ref_pixels_per_line
+    movsxd          r9, dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+.half_vert_variance8x_h_1:
+        movq            xmm5,           QWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s8
+        movq            xmm3,           QWORD PTR [rsi+rax] ;  xmm3 = s1,s2,s3..s9
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d8
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+
+%if ABI_IS_32BIT
+        add             esi,            dword ptr arg(1) ;ref_pixels_per_line    ;  next source
+        add             edi,            dword ptr arg(3) ;src_pixels_per_line    ;  next destination
+%else
+        add             rsi, r8
+        add             rdi, r9
+%endif
+
+        sub             rcx,            1                   ;
+        jnz             .half_vert_variance8x_h_1          ;
+
+        movdq2q         mm6,            xmm6                ;
+        movdq2q         mm7,            xmm7                ;
+
+        psrldq          xmm6,           8
+        psrldq          xmm7,           8
+
+        movdq2q         mm2,            xmm6
+        movdq2q         mm3,            xmm7
+
+        paddw           mm6,            mm2
+        paddd           mm7,            mm3
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rsi,            arg(5) ; sum
+        mov             rdi,            arg(6) ; sumsquared
+
+        movd            [rsi],          mm2                 ;
+        movd            [rdi],          mm4                 ;
+
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp9_half_horiz_variance8x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp9_half_horiz_variance8x_h_sse2) PRIVATE
+sym(vp9_half_horiz_variance8x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+%if ABI_IS_32BIT=0
+    movsxd          r8, dword ptr arg(1) ;ref_pixels_per_line
+    movsxd          r9, dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+
+        pxor            xmm0,           xmm0                ;
+.half_horiz_variance8x_h_1:
+        movq            xmm5,           QWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s8
+        movq            xmm3,           QWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s9
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d8
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+
+%if ABI_IS_32BIT
+        add             esi,            dword ptr arg(1) ;ref_pixels_per_line    ;  next source
+        add             edi,            dword ptr arg(3) ;src_pixels_per_line    ;  next destination
+%else
+        add             rsi, r8
+        add             rdi, r9
+%endif
+        sub             rcx,            1                   ;
+        jnz             .half_horiz_variance8x_h_1          ;
+
+        movdq2q         mm6,            xmm6                ;
+        movdq2q         mm7,            xmm7                ;
+
+        psrldq          xmm6,           8
+        psrldq          xmm7,           8
+
+        movdq2q         mm2,            xmm6
+        movdq2q         mm3,            xmm7
+
+        paddw           mm6,            mm2
+        paddd           mm7,            mm3
+
+        pxor            mm3,            mm3                 ;
+        pxor            mm2,            mm2                 ;
+
+        punpcklwd       mm2,            mm6                 ;
+        punpckhwd       mm3,            mm6                 ;
+
+        paddd           mm2,            mm3                 ;
+        movq            mm6,            mm2                 ;
+
+        psrlq           mm6,            32                  ;
+        paddd           mm2,            mm6                 ;
+
+        psrad           mm2,            16                  ;
+        movq            mm4,            mm7                 ;
+
+        psrlq           mm4,            32                  ;
+        paddd           mm4,            mm7                 ;
+
+        mov             rsi,            arg(5) ; sum
+        mov             rdi,            arg(6) ; sumsquared
+
+        movd            [rsi],          mm2                 ;
+        movd            [rdi],          mm4                 ;
+
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+SECTION_RODATA
+;    short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64};
+align 16
+xmm_bi_rd:
+    times 8 dw 64
+align 16
+bilinear_filters_sse2:
+    dw 128, 128, 128, 128, 128, 128, 128, 128,  0,  0,  0,  0,  0,  0,  0,  0
+    dw 120, 120, 120, 120, 120, 120, 120, 120,  8,  8,  8,  8,  8,  8,  8,  8
+    dw 112, 112, 112, 112, 112, 112, 112, 112, 16, 16, 16, 16, 16, 16, 16, 16
+    dw 104, 104, 104, 104, 104, 104, 104, 104, 24, 24, 24, 24, 24, 24, 24, 24
+    dw 96, 96, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32
+    dw 88, 88, 88, 88, 88, 88, 88, 88, 40, 40, 40, 40, 40, 40, 40, 40
+    dw 80, 80, 80, 80, 80, 80, 80, 80, 48, 48, 48, 48, 48, 48, 48, 48
+    dw 72, 72, 72, 72, 72, 72, 72, 72, 56, 56, 56, 56, 56, 56, 56, 56
+    dw 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
+    dw 56, 56, 56, 56, 56, 56, 56, 56, 72, 72, 72, 72, 72, 72, 72, 72
+    dw 48, 48, 48, 48, 48, 48, 48, 48, 80, 80, 80, 80, 80, 80, 80, 80
+    dw 40, 40, 40, 40, 40, 40, 40, 40, 88, 88, 88, 88, 88, 88, 88, 88
+    dw 32, 32, 32, 32, 32, 32, 32, 32, 96, 96, 96, 96, 96, 96, 96, 96
+    dw 24, 24, 24, 24, 24, 24, 24, 24, 104, 104, 104, 104, 104, 104, 104, 104
+    dw 16, 16, 16, 16, 16, 16, 16, 16, 112, 112, 112, 112, 112, 112, 112, 112
+    dw 8, 8, 8, 8, 8, 8, 8, 8, 120, 120, 120, 120, 120, 120, 120, 120
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_impl_ssse3.asm b/libvpx/vp9/encoder/x86/vp9_variance_impl_ssse3.asm
new file mode 100644
index 0000000..98a4a16
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_impl_ssse3.asm
@@ -0,0 +1,372 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%define xmm_filter_shift            7
+
+
+;void vp9_filter_block2d_bil_var_ssse3
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int  xoffset,
+;    int  yoffset,
+;    int *sum,
+;    unsigned int *sumsquared;;
+;
+;)
+;Note: The filter coefficient at offset=0 is 128. Since the second register
+;for Pmaddubsw is signed bytes, we must calculate zero offset seperately.
+global sym(vp9_filter_block2d_bil_var_ssse3) PRIVATE
+sym(vp9_filter_block2d_bil_var_ssse3):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6
+        pxor            xmm7,           xmm7
+
+        lea             rcx,            [GLOBAL(bilinear_filters_ssse3)]
+        movsxd          rax,            dword ptr arg(5)     ; xoffset
+
+        cmp             rax,            0                    ; skip first_pass filter if xoffset=0
+        je              .filter_block2d_bil_var_ssse3_sp_only
+
+        shl             rax,            4                    ; point to filter coeff with xoffset
+        lea             rax,            [rax + rcx]          ; HFilter
+
+        movsxd          rdx,            dword ptr arg(6)     ; yoffset
+
+        cmp             rdx,            0                    ; skip second_pass filter if yoffset=0
+        je              .filter_block2d_bil_var_ssse3_fp_only
+
+        shl             rdx,            4
+        lea             rdx,            [rdx + rcx]          ; VFilter
+
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+
+        movdqu          xmm0,           XMMWORD PTR [rsi]
+        movdqu          xmm1,           XMMWORD PTR [rsi+1]
+        movdqa          xmm2,           xmm0
+
+        punpcklbw       xmm0,           xmm1
+        punpckhbw       xmm2,           xmm1
+        pmaddubsw       xmm0,           [rax]
+        pmaddubsw       xmm2,           [rax]
+
+        paddw           xmm0,           [GLOBAL(xmm_bi_rd)]
+        paddw           xmm2,           [GLOBAL(xmm_bi_rd)]
+        psraw           xmm0,           xmm_filter_shift
+        psraw           xmm2,           xmm_filter_shift
+
+        packuswb        xmm0,           xmm2
+
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1) ;ref_pixels_per_line
+%else
+        movsxd          r8,             dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line
+        lea             rsi,            [rsi + r8]
+%endif
+
+.filter_block2d_bil_var_ssse3_loop:
+        movdqu          xmm1,           XMMWORD PTR [rsi]
+        movdqu          xmm2,           XMMWORD PTR [rsi+1]
+        movdqa          xmm3,           xmm1
+
+        punpcklbw       xmm1,           xmm2
+        punpckhbw       xmm3,           xmm2
+        pmaddubsw       xmm1,           [rax]
+        pmaddubsw       xmm3,           [rax]
+
+        paddw           xmm1,           [GLOBAL(xmm_bi_rd)]
+        paddw           xmm3,           [GLOBAL(xmm_bi_rd)]
+        psraw           xmm1,           xmm_filter_shift
+        psraw           xmm3,           xmm_filter_shift
+        packuswb        xmm1,           xmm3
+
+        movdqa          xmm2,           xmm0
+        movdqa          xmm0,           xmm1
+        movdqa          xmm3,           xmm2
+
+        punpcklbw       xmm2,           xmm1
+        punpckhbw       xmm3,           xmm1
+        pmaddubsw       xmm2,           [rdx]
+        pmaddubsw       xmm3,           [rdx]
+
+        paddw           xmm2,           [GLOBAL(xmm_bi_rd)]
+        paddw           xmm3,           [GLOBAL(xmm_bi_rd)]
+        psraw           xmm2,           xmm_filter_shift
+        psraw           xmm3,           xmm_filter_shift
+
+        movq            xmm1,           QWORD PTR [rdi]
+        pxor            xmm4,           xmm4
+        punpcklbw       xmm1,           xmm4
+        movq            xmm5,           QWORD PTR [rdi+8]
+        punpcklbw       xmm5,           xmm4
+
+        psubw           xmm2,           xmm1
+        psubw           xmm3,           xmm5
+        paddw           xmm6,           xmm2
+        paddw           xmm6,           xmm3
+        pmaddwd         xmm2,           xmm2
+        pmaddwd         xmm3,           xmm3
+        paddd           xmm7,           xmm2
+        paddd           xmm7,           xmm3
+
+%if ABI_IS_32BIT
+        add             rsi,            dword ptr arg(1)     ;ref_pixels_per_line
+        add             rdi,            dword ptr arg(3)     ;src_pixels_per_line
+%else
+        lea             rsi,            [rsi + r8]
+        lea             rdi,            [rdi + r9]
+%endif
+
+        sub             rcx,            1
+        jnz             .filter_block2d_bil_var_ssse3_loop
+
+        jmp             .filter_block2d_bil_variance
+
+.filter_block2d_bil_var_ssse3_sp_only:
+        movsxd          rdx,            dword ptr arg(6)     ; yoffset
+
+        cmp             rdx,            0                    ; Both xoffset =0 and yoffset=0
+        je              .filter_block2d_bil_var_ssse3_full_pixel
+
+        shl             rdx,            4
+        lea             rdx,            [rdx + rcx]          ; VFilter
+
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rax,            dword ptr arg(1)     ;ref_pixels_per_line
+
+        movdqu          xmm1,           XMMWORD PTR [rsi]
+        movdqa          xmm0,           xmm1
+
+%if ABI_IS_32BIT=0
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+        lea             rsi,            [rsi + rax]
+
+.filter_block2d_bil_sp_only_loop:
+        movdqu          xmm3,           XMMWORD PTR [rsi]
+        movdqa          xmm2,           xmm1
+        movdqa          xmm0,           xmm3
+
+        punpcklbw       xmm1,           xmm3
+        punpckhbw       xmm2,           xmm3
+        pmaddubsw       xmm1,           [rdx]
+        pmaddubsw       xmm2,           [rdx]
+
+        paddw           xmm1,           [GLOBAL(xmm_bi_rd)]
+        paddw           xmm2,           [GLOBAL(xmm_bi_rd)]
+        psraw           xmm1,           xmm_filter_shift
+        psraw           xmm2,           xmm_filter_shift
+
+        movq            xmm3,           QWORD PTR [rdi]
+        pxor            xmm4,           xmm4
+        punpcklbw       xmm3,           xmm4
+        movq            xmm5,           QWORD PTR [rdi+8]
+        punpcklbw       xmm5,           xmm4
+
+        psubw           xmm1,           xmm3
+        psubw           xmm2,           xmm5
+        paddw           xmm6,           xmm1
+        paddw           xmm6,           xmm2
+        pmaddwd         xmm1,           xmm1
+        pmaddwd         xmm2,           xmm2
+        paddd           xmm7,           xmm1
+        paddd           xmm7,           xmm2
+
+        movdqa          xmm1,           xmm0
+        lea             rsi,            [rsi + rax]          ;ref_pixels_per_line
+
+%if ABI_IS_32BIT
+        add             rdi,            dword ptr arg(3)     ;src_pixels_per_line
+%else
+        lea             rdi,            [rdi + r9]
+%endif
+
+        sub             rcx,            1
+        jnz             .filter_block2d_bil_sp_only_loop
+
+        jmp             .filter_block2d_bil_variance
+
+.filter_block2d_bil_var_ssse3_full_pixel:
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rax,            dword ptr arg(1)     ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)     ;src_pixels_per_line
+        pxor            xmm0,           xmm0
+
+.filter_block2d_bil_full_pixel_loop:
+        movq            xmm1,           QWORD PTR [rsi]
+        punpcklbw       xmm1,           xmm0
+        movq            xmm2,           QWORD PTR [rsi+8]
+        punpcklbw       xmm2,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]
+        punpcklbw       xmm3,           xmm0
+        movq            xmm4,           QWORD PTR [rdi+8]
+        punpcklbw       xmm4,           xmm0
+
+        psubw           xmm1,           xmm3
+        psubw           xmm2,           xmm4
+        paddw           xmm6,           xmm1
+        paddw           xmm6,           xmm2
+        pmaddwd         xmm1,           xmm1
+        pmaddwd         xmm2,           xmm2
+        paddd           xmm7,           xmm1
+        paddd           xmm7,           xmm2
+
+        lea             rsi,            [rsi + rax]          ;ref_pixels_per_line
+        lea             rdi,            [rdi + rdx]          ;src_pixels_per_line
+        sub             rcx,            1
+        jnz             .filter_block2d_bil_full_pixel_loop
+
+        jmp             .filter_block2d_bil_variance
+
+.filter_block2d_bil_var_ssse3_fp_only:
+        mov             rsi,            arg(0)               ;ref_ptr
+        mov             rdi,            arg(2)               ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)     ;Height
+        movsxd          rdx,            dword ptr arg(1)     ;ref_pixels_per_line
+
+        pxor            xmm0,           xmm0
+
+%if ABI_IS_32BIT=0
+        movsxd          r9,             dword ptr arg(3) ;src_pixels_per_line
+%endif
+
+.filter_block2d_bil_fp_only_loop:
+        movdqu          xmm1,           XMMWORD PTR [rsi]
+        movdqu          xmm2,           XMMWORD PTR [rsi+1]
+        movdqa          xmm3,           xmm1
+
+        punpcklbw       xmm1,           xmm2
+        punpckhbw       xmm3,           xmm2
+        pmaddubsw       xmm1,           [rax]
+        pmaddubsw       xmm3,           [rax]
+
+        paddw           xmm1,           [GLOBAL(xmm_bi_rd)]
+        paddw           xmm3,           [GLOBAL(xmm_bi_rd)]
+        psraw           xmm1,           xmm_filter_shift
+        psraw           xmm3,           xmm_filter_shift
+
+        movq            xmm2,           XMMWORD PTR [rdi]
+        pxor            xmm4,           xmm4
+        punpcklbw       xmm2,           xmm4
+        movq            xmm5,           QWORD PTR [rdi+8]
+        punpcklbw       xmm5,           xmm4
+
+        psubw           xmm1,           xmm2
+        psubw           xmm3,           xmm5
+        paddw           xmm6,           xmm1
+        paddw           xmm6,           xmm3
+        pmaddwd         xmm1,           xmm1
+        pmaddwd         xmm3,           xmm3
+        paddd           xmm7,           xmm1
+        paddd           xmm7,           xmm3
+
+        lea             rsi,            [rsi + rdx]
+%if ABI_IS_32BIT
+        add             rdi,            dword ptr arg(3)     ;src_pixels_per_line
+%else
+        lea             rdi,            [rdi + r9]
+%endif
+
+        sub             rcx,            1
+        jnz             .filter_block2d_bil_fp_only_loop
+
+        jmp             .filter_block2d_bil_variance
+
+.filter_block2d_bil_variance:
+        pxor        xmm0,           xmm0
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(7) ;[Sum]
+        mov         rdi,            arg(8) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+SECTION_RODATA
+align 16
+xmm_bi_rd:
+    times 8 dw 64
+align 16
+bilinear_filters_ssse3:
+    times 8 db 128, 0
+    times 8 db 120, 8
+    times 8 db 112, 16
+    times 8 db 104, 24
+    times 8 db  96, 32
+    times 8 db  88, 40
+    times 8 db  80, 48
+    times 8 db  72, 56
+    times 8 db  64, 64
+    times 8 db  56, 72
+    times 8 db  48, 80
+    times 8 db  40, 88
+    times 8 db  32, 96
+    times 8 db  24, 104
+    times 8 db  16, 112
+    times 8 db   8, 120
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_mmx.c b/libvpx/vp9/encoder/x86/vp9_variance_mmx.c
new file mode 100644
index 0000000..bad1cfa
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_mmx.c
@@ -0,0 +1,382 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/common/vp9_pragmas.h"
+#include "vpx_ports/mem.h"
+
+extern void filter_block1d_h6_mmx
+(
+  const unsigned char *src_ptr,
+  unsigned short *output_ptr,
+  unsigned int src_pixels_per_line,
+  unsigned int pixel_step,
+  unsigned int output_height,
+  unsigned int output_width,
+  short *vp7_filter
+);
+extern void filter_block1d_v6_mmx
+(
+  const short *src_ptr,
+  unsigned char *output_ptr,
+  unsigned int pixels_per_line,
+  unsigned int pixel_step,
+  unsigned int output_height,
+  unsigned int output_width,
+  short *vp7_filter
+);
+
+extern unsigned int vp9_get_mb_ss_mmx(const short *src_ptr);
+extern unsigned int vp9_get8x8var_mmx
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+extern unsigned int vp9_get4x4var_mmx
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+extern void vp9_filter_block2d_bil4x4_var_mmx
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  const short *HFilter,
+  const short *VFilter,
+  int *sum,
+  unsigned int *sumsquared
+);
+extern void vp9_filter_block2d_bil_var_mmx
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  const short *HFilter,
+  const short *VFilter,
+  int *sum,
+  unsigned int *sumsquared
+);
+
+
+unsigned int vp9_variance4x4_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  vp9_get4x4var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 4));
+
+}
+
+unsigned int vp9_variance8x8_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &var, &avg);
+  *sse = var;
+
+  return (var - (((unsigned int)avg * avg) >> 6));
+
+}
+
+unsigned int vp9_mse16x16_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int sse0, sse1, sse2, sse3, var;
+  int sum0, sum1, sum2, sum3;
+
+
+  vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
+  vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
+  vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
+  vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
+
+  var = sse0 + sse1 + sse2 + sse3;
+  *sse = var;
+  return var;
+}
+
+
+unsigned int vp9_variance16x16_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int sse0, sse1, sse2, sse3, var;
+  int sum0, sum1, sum2, sum3, avg;
+
+
+  vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
+  vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
+  vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2);
+  vp9_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3);
+
+  var = sse0 + sse1 + sse2 + sse3;
+  avg = sum0 + sum1 + sum2 + sum3;
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 8));
+}
+
+unsigned int vp9_variance16x8_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int sse0, sse1, var;
+  int sum0, sum1, avg;
+
+  vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
+  vp9_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1);
+
+  var = sse0 + sse1;
+  avg = sum0 + sum1;
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 7));
+
+}
+
+
+unsigned int vp9_variance8x16_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int sse0, sse1, var;
+  int sum0, sum1, avg;
+
+  vp9_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0);
+  vp9_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse1, &sum1);
+
+  var = sse0 + sse1;
+  avg = sum0 + sum1;
+  *sse = var;
+
+  return (var - (((unsigned int)avg * avg) >> 7));
+
+}
+
+DECLARE_ALIGNED(16, extern const short, vp9_bilinear_filters_mmx[16][8]);
+
+unsigned int vp9_sub_pixel_variance4x4_mmx
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse)
+
+{
+  int xsum;
+  unsigned int xxsum;
+  vp9_filter_block2d_bil4x4_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum, &xxsum
+  );
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 4));
+}
+
+
+unsigned int vp9_sub_pixel_variance8x8_mmx
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+
+  int xsum;
+  unsigned int xxsum;
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 8,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum, &xxsum
+  );
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 6));
+}
+
+unsigned int vp9_sub_pixel_variance16x16_mmx
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+
+  int xsum0, xsum1;
+  unsigned int xxsum0, xxsum1;
+
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 16,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum0, &xxsum0
+  );
+
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr + 8, src_pixels_per_line,
+    dst_ptr + 8, dst_pixels_per_line, 16,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum1, &xxsum1
+  );
+
+  xsum0 += xsum1;
+  xxsum0 += xxsum1;
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
+
+
+}
+
+unsigned int vp9_sub_pixel_mse16x16_mmx(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  vp9_sub_pixel_variance16x16_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
+  return *sse;
+}
+
+unsigned int vp9_sub_pixel_variance16x8_mmx
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum0, xsum1;
+  unsigned int xxsum0, xxsum1;
+
+
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 8,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum0, &xxsum0
+  );
+
+
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr + 8, src_pixels_per_line,
+    dst_ptr + 8, dst_pixels_per_line, 8,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum1, &xxsum1
+  );
+
+  xsum0 += xsum1;
+  xxsum0 += xxsum1;
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
+}
+
+unsigned int vp9_sub_pixel_variance8x16_mmx
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum;
+  unsigned int xxsum;
+  vp9_filter_block2d_bil_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 16,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum, &xxsum
+  );
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 7));
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_h_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 8, 0,
+                                         ref_ptr, recon_stride, sse);
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_v_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 0, 8,
+                                         ref_ptr, recon_stride, sse);
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_hv_mmx(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  return vp9_sub_pixel_variance16x16_mmx(src_ptr, source_stride, 8, 8,
+                                         ref_ptr, recon_stride, sse);
+}
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_sse2.c b/libvpx/vp9/encoder/x86/vp9_variance_sse2.c
new file mode 100644
index 0000000..67ca925
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_sse2.c
@@ -0,0 +1,773 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/common/vp9_pragmas.h"
+#include "vpx_ports/mem.h"
+
+#define HALFNDX 8
+
+extern void filter_block1d_h6_mmx(const unsigned char *src_ptr, unsigned short *output_ptr, unsigned int src_pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
+extern void filter_block1d_v6_mmx(const short *src_ptr, unsigned char *output_ptr, unsigned int pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
+extern void filter_block1d8_h6_sse2(const unsigned char *src_ptr, unsigned short *output_ptr, unsigned int src_pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
+extern void filter_block1d8_v6_sse2(const short *src_ptr, unsigned char *output_ptr, unsigned int pixels_per_line, unsigned int pixel_step, unsigned int output_height, unsigned int output_width, short *vp7_filter);
+
+extern void vp9_filter_block2d_bil4x4_var_mmx
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  const short *HFilter,
+  const short *VFilter,
+  int *sum,
+  unsigned int *sumsquared
+);
+
+extern unsigned int vp9_get4x4var_mmx
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+
+unsigned int vp9_get_mb_ss_sse2
+(
+  const short *src_ptr
+);
+unsigned int vp9_get16x16var_sse2
+(
+  const unsigned char *src_ptr,
+  int source_stride,
+  const unsigned char *ref_ptr,
+  int recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+unsigned int vp9_get8x8var_sse2
+(
+  const unsigned char *src_ptr,
+  int source_stride,
+  const unsigned char *ref_ptr,
+  int recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+void vp9_filter_block2d_bil_var_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int  xoffset,
+  int  yoffset,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_horiz_vert_variance8x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_horiz_vert_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_horiz_variance8x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_horiz_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_vert_variance8x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+void vp9_half_vert_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+
+DECLARE_ALIGNED(16, extern const short, vp9_bilinear_filters_mmx[16][8]);
+
+typedef unsigned int (*get_var_sse2) (
+  const unsigned char *src_ptr,
+  int source_stride,
+  const unsigned char *ref_ptr,
+  int recon_stride,
+  unsigned int *SSE,
+  int *Sum
+);
+
+static void variance_sse2(const unsigned char *src_ptr, int  source_stride,
+                        const unsigned char *ref_ptr, int  recon_stride,
+                        int  w, int  h, unsigned int *sse, int *sum,
+                        get_var_sse2 var_fn, int block_size) {
+  unsigned int sse0;
+  int sum0;
+  int i, j;
+
+  *sse = 0;
+  *sum = 0;
+
+  for (i = 0; i < h; i += block_size) {
+    for (j = 0; j < w; j += block_size) {
+      var_fn(src_ptr + source_stride * i + j, source_stride,
+             ref_ptr + recon_stride * i + j, recon_stride, &sse0, &sum0);
+      *sse += sse0;
+      *sum += sum0;
+    }
+  }
+}
+
+unsigned int vp9_variance4x4_sse2(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4,
+                  &var, &avg, vp9_get4x4var_mmx, 4);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 4));
+}
+
+unsigned int vp9_variance8x4_sse2(const uint8_t *src_ptr,
+                                  int  source_stride,
+                                  const uint8_t *ref_ptr,
+                                  int  recon_stride,
+                                  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 4,
+                  &var, &avg, vp9_get4x4var_mmx, 4);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 5));
+}
+
+unsigned int vp9_variance4x8_sse2(const uint8_t *src_ptr,
+                                  int  source_stride,
+                                  const uint8_t *ref_ptr,
+                                  int  recon_stride,
+                                  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 4, 8,
+                  &var, &avg, vp9_get4x4var_mmx, 4);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 5));
+}
+
+unsigned int vp9_variance8x8_sse2
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8,
+                  &var, &avg, vp9_get8x8var_sse2, 8);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 6));
+}
+
+unsigned int vp9_variance16x8_sse2
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8,
+                  &var, &avg, vp9_get8x8var_sse2, 8);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 7));
+}
+
+unsigned int vp9_variance8x16_sse2
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16,
+                &var, &avg, vp9_get8x8var_sse2, 8);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 7));
+}
+
+unsigned int vp9_variance16x16_sse2
+(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((unsigned int)avg * avg) >> 8));
+}
+
+unsigned int vp9_mse16x16_wmt(
+  const unsigned char *src_ptr,
+  int  source_stride,
+  const unsigned char *ref_ptr,
+  int  recon_stride,
+  unsigned int *sse) {
+
+  unsigned int sse0;
+  int sum0;
+  vp9_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
+                       &sum0);
+  *sse = sse0;
+  return sse0;
+}
+
+unsigned int vp9_variance32x32_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 32,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 10));
+}
+
+unsigned int vp9_variance32x16_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 16,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 9));
+}
+
+unsigned int vp9_variance16x32_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 32,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 9));
+}
+
+unsigned int vp9_variance64x64_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 64, 64,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 12));
+}
+
+unsigned int vp9_variance64x32_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 64, 32,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 11));
+}
+
+unsigned int vp9_variance32x64_sse2(const uint8_t *src_ptr,
+                                    int  source_stride,
+                                    const uint8_t *ref_ptr,
+                                    int  recon_stride,
+                                    unsigned int *sse) {
+  unsigned int var;
+  int avg;
+
+  variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 64,
+                &var, &avg, vp9_get16x16var_sse2, 16);
+  *sse = var;
+  return (var - (((int64_t)avg * avg) >> 11));
+}
+
+unsigned int vp9_sub_pixel_variance4x4_wmt
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum;
+  unsigned int xxsum;
+  vp9_filter_block2d_bil4x4_var_mmx(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line,
+    vp9_bilinear_filters_mmx[xoffset], vp9_bilinear_filters_mmx[yoffset],
+    &xsum, &xxsum
+  );
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 4));
+}
+
+
+unsigned int vp9_sub_pixel_variance8x8_wmt
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum;
+  unsigned int xxsum;
+
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum, &xxsum);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum, &xxsum);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum, &xxsum);
+  } else {
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      xoffset, yoffset,
+      &xsum, &xxsum);
+  }
+
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 6));
+}
+
+static void sub_pixel_variance16x16_sse2(const uint8_t *src_ptr,
+                                         int src_pixels_per_line,
+                                         int xoffset,
+                                         int yoffset,
+                                         const uint8_t *dst_ptr,
+                                         int dst_pixels_per_line,
+                                         unsigned int *sse, int *avg) {
+  int xsum0, xsum1;
+  unsigned int xxsum0, xxsum1;
+
+  // note we could avoid these if statements if the calling function
+  // just called the appropriate functions inside.
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else {
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      xoffset, yoffset,
+      &xsum0, &xxsum0
+    );
+
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr + 8, src_pixels_per_line,
+      dst_ptr + 8, dst_pixels_per_line, 16,
+      xoffset, yoffset,
+      &xsum1, &xxsum1
+    );
+    xsum0 += xsum1;
+    xxsum0 += xxsum1;
+  }
+
+  *sse = xxsum0;
+  *avg = xsum0;
+}
+
+unsigned int vp9_sub_pixel_variance16x16_sse2(const uint8_t *src_ptr,
+                                              int src_pixels_per_line,
+                                              int xoffset,
+                                              int yoffset,
+                                              const uint8_t *dst_ptr,
+                                              int dst_pixels_per_line,
+                                              unsigned int *sse_ptr) {
+  int avg;
+  unsigned int sse;
+
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse, &avg);
+  *sse_ptr = sse;
+
+  return (sse - (((unsigned int) avg * avg) >> 8));
+}
+
+unsigned int vp9_sub_pixel_variance32x32_sse2(const uint8_t *src_ptr,
+                                              int src_pixels_per_line,
+                                              int xoffset,
+                                              int yoffset,
+                                              const uint8_t *dst_ptr,
+                                              int dst_pixels_per_line,
+                                              unsigned int *sse_ptr) {
+  int avg0, avg1, avg2, avg3;
+  unsigned int sse0, sse1, sse2, sse3;
+
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse0, &avg0);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse1, &avg1);
+  src_ptr += 16 * src_pixels_per_line;
+  dst_ptr += 16 * dst_pixels_per_line;
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse2, &avg2);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse3, &avg3);
+  sse0 += sse1 + sse2 + sse3;
+  avg0 += avg1 + avg2 + avg3;
+  *sse_ptr = sse0;
+
+  return (sse0 - (((unsigned int) avg0 * avg0) >> 10));
+}
+
+unsigned int vp9_sub_pixel_variance64x64_sse2(const uint8_t *src_ptr,
+                                              int src_pixels_per_line,
+                                              int xoffset,
+                                              int yoffset,
+                                              const uint8_t *dst_ptr,
+                                              int dst_pixels_per_line,
+                                              unsigned int *sse_ptr) {
+  int avg0, avg1, avg2, avg3, avg4;
+  unsigned int sse0, sse1, sse2, sse3, sse4;
+
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse0, &avg0);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse1, &avg1);
+  sub_pixel_variance16x16_sse2(src_ptr + 32, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 32, dst_pixels_per_line,
+                               &sse2, &avg2);
+  sub_pixel_variance16x16_sse2(src_ptr + 48, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 48, dst_pixels_per_line,
+                               &sse3, &avg3);
+  src_ptr += 16 * src_pixels_per_line;
+  dst_ptr += 16 * dst_pixels_per_line;
+  avg0 += avg1 + avg2 + avg3;
+  sse0 += sse1 + sse2 + sse3;
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse1, &avg1);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse2, &avg2);
+  sub_pixel_variance16x16_sse2(src_ptr + 32, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 32, dst_pixels_per_line,
+                               &sse3, &avg3);
+  sub_pixel_variance16x16_sse2(src_ptr + 48, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 48, dst_pixels_per_line,
+                               &sse4, &avg4);
+  src_ptr += 16 * src_pixels_per_line;
+  dst_ptr += 16 * dst_pixels_per_line;
+  avg0 += avg1 + avg2 + avg3 + avg4;
+  sse0 += sse1 + sse2 + sse3 + sse4;
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse1, &avg1);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse2, &avg2);
+  sub_pixel_variance16x16_sse2(src_ptr + 32, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 32, dst_pixels_per_line,
+                               &sse3, &avg3);
+  sub_pixel_variance16x16_sse2(src_ptr + 48, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 48, dst_pixels_per_line,
+                               &sse4, &avg4);
+  src_ptr += 16 * src_pixels_per_line;
+  dst_ptr += 16 * dst_pixels_per_line;
+  avg0 += avg1 + avg2 + avg3 + avg4;
+  sse0 += sse1 + sse2 + sse3 + sse4;
+  sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr, dst_pixels_per_line,
+                               &sse1, &avg1);
+  sub_pixel_variance16x16_sse2(src_ptr + 16, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 16, dst_pixels_per_line,
+                               &sse2, &avg2);
+  sub_pixel_variance16x16_sse2(src_ptr + 32, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 32, dst_pixels_per_line,
+                               &sse3, &avg3);
+  sub_pixel_variance16x16_sse2(src_ptr + 48, src_pixels_per_line, xoffset,
+                               yoffset, dst_ptr + 48, dst_pixels_per_line,
+                               &sse4, &avg4);
+  avg0 += avg1 + avg2 + avg3 + avg4;
+  sse0 += sse1 + sse2 + sse3 + sse4;
+  *sse_ptr = sse0;
+
+  return (sse0 - (((unsigned int) avg0 * avg0) >> 12));
+}
+
+unsigned int vp9_sub_pixel_mse16x16_sse2(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  vp9_sub_pixel_variance16x16_sse2(src_ptr, src_pixels_per_line, xoffset,
+                                   yoffset, dst_ptr, dst_pixels_per_line, sse);
+  return *sse;
+}
+
+unsigned int vp9_sub_pixel_variance16x8_wmt
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+
+) {
+  int xsum0, xsum1;
+  unsigned int xxsum0, xxsum1;
+
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else {
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      xoffset, yoffset,
+      &xsum0, &xxsum0);
+
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr + 8, src_pixels_per_line,
+      dst_ptr + 8, dst_pixels_per_line, 8,
+      xoffset, yoffset,
+      &xsum1, &xxsum1);
+    xsum0 += xsum1;
+    xxsum0 += xxsum1;
+  }
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
+}
+
+unsigned int vp9_sub_pixel_variance8x16_wmt
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum;
+  unsigned int xxsum;
+
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum, &xxsum);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum, &xxsum);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance8x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum, &xxsum);
+  } else {
+    vp9_filter_block2d_bil_var_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      xoffset, yoffset,
+      &xsum, &xxsum);
+  }
+
+  *sse = xxsum;
+  return (xxsum - (((unsigned int)xsum * xsum) >> 7));
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_h_wmt(
+  const unsigned char *src_ptr,
+  int  src_pixels_per_line,
+  const unsigned char *dst_ptr,
+  int  dst_pixels_per_line,
+  unsigned int *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  vp9_half_horiz_variance16x_h_sse2(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 16,
+    &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_v_wmt(
+  const unsigned char *src_ptr,
+  int  src_pixels_per_line,
+  const unsigned char *dst_ptr,
+  int  dst_pixels_per_line,
+  unsigned int *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+  vp9_half_vert_variance16x_h_sse2(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 16,
+    &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
+}
+
+
+unsigned int vp9_variance_halfpixvar16x16_hv_wmt(
+  const unsigned char *src_ptr,
+  int  src_pixels_per_line,
+  const unsigned char *dst_ptr,
+  int  dst_pixels_per_line,
+  unsigned int *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  vp9_half_horiz_vert_variance16x_h_sse2(
+    src_ptr, src_pixels_per_line,
+    dst_ptr, dst_pixels_per_line, 16,
+    &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
+}
diff --git a/libvpx/vp9/encoder/x86/vp9_variance_ssse3.c b/libvpx/vp9/encoder/x86/vp9_variance_ssse3.c
new file mode 100644
index 0000000..882acad
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_variance_ssse3.c
@@ -0,0 +1,142 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/common/vp9_pragmas.h"
+#include "vpx_ports/mem.h"
+
+#define HALFNDX 8
+
+extern void vp9_half_horiz_vert_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+extern void vp9_half_horiz_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+extern void vp9_half_vert_variance16x_h_sse2
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int *sum,
+  unsigned int *sumsquared
+);
+extern void vp9_filter_block2d_bil_var_ssse3
+(
+  const unsigned char *ref_ptr,
+  int ref_pixels_per_line,
+  const unsigned char *src_ptr,
+  int src_pixels_per_line,
+  unsigned int Height,
+  int  xoffset,
+  int  yoffset,
+  int *sum,
+  unsigned int *sumsquared
+);
+
+unsigned int vp9_sub_pixel_variance16x16_ssse3
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  // note we could avoid these if statements if the calling function
+  // just called the appropriate functions inside.
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      &xsum0, &xxsum0);
+  } else {
+    vp9_filter_block2d_bil_var_ssse3(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 16,
+      xoffset, yoffset,
+      &xsum0, &xxsum0);
+  }
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 8));
+}
+
+unsigned int vp9_sub_pixel_variance16x8_ssse3
+(
+  const unsigned char  *src_ptr,
+  int  src_pixels_per_line,
+  int  xoffset,
+  int  yoffset,
+  const unsigned char *dst_ptr,
+  int dst_pixels_per_line,
+  unsigned int *sse
+
+) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  if (xoffset == HALFNDX && yoffset == 0) {
+    vp9_half_horiz_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else if (xoffset == 0 && yoffset == HALFNDX) {
+    vp9_half_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else if (xoffset == HALFNDX && yoffset == HALFNDX) {
+    vp9_half_horiz_vert_variance16x_h_sse2(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      &xsum0, &xxsum0);
+  } else {
+    vp9_filter_block2d_bil_var_ssse3(
+      src_ptr, src_pixels_per_line,
+      dst_ptr, dst_pixels_per_line, 8,
+      xoffset, yoffset,
+      &xsum0, &xxsum0);
+  }
+
+  *sse = xxsum0;
+  return (xxsum0 - (((unsigned int)xsum0 * xsum0) >> 7));
+}
diff --git a/libvpx/vp9/encoder/x86/vp9_x86_csystemdependent.c b/libvpx/vp9/encoder/x86/vp9_x86_csystemdependent.c
new file mode 100644
index 0000000..6016e14
--- /dev/null
+++ b/libvpx/vp9/encoder/x86/vp9_x86_csystemdependent.c
@@ -0,0 +1,55 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "./vpx_config.h"
+#include "vpx_ports/x86.h"
+#include "vp9/encoder/vp9_variance.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vp9/encoder/x86/vp9_dct_mmx.h"
+
+// TODO(jimbankoski) Consider rewriting the c to take the same values rather
+// than going through these pointer conversions
+#if 0 && HAVE_MMX
+void vp9_short_fdct8x4_mmx(short *input, short *output, int pitch) {
+  vp9_short_fdct4x4_mmx(input,   output,    pitch);
+  vp9_short_fdct4x4_mmx(input + 4, output + 16, pitch);
+}
+
+void vp9_subtract_b_mmx_impl(unsigned char *z,  int src_stride,
+                             short *diff, unsigned char *predictor,
+                             int pitch);
+void vp9_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch) {
+  unsigned char *z = *(be->base_src) + be->src;
+  unsigned int  src_stride = be->src_stride;
+  short *diff = &be->src_diff[0];
+  unsigned char *predictor = *(bd->base_dst) + bd->dst;
+  // TODO(jingning): The prototype function in c has been changed. Need to
+  // modify the mmx and sse versions.
+  vp9_subtract_b_mmx_impl(z, src_stride, diff, predictor, pitch);
+}
+
+#endif
+
+#if 0 && HAVE_SSE2
+void vp9_subtract_b_sse2_impl(unsigned char *z,  int src_stride,
+                              short *diff, unsigned char *predictor,
+                              int pitch);
+void vp9_subtract_b_sse2(BLOCK *be, BLOCKD *bd, int pitch) {
+  unsigned char *z = *(be->base_src) + be->src;
+  unsigned int  src_stride = be->src_stride;
+  short *diff = &be->src_diff[0];
+  unsigned char *predictor = *(bd->base_dst) + bd->dst;
+  // TODO(jingning): The prototype function in c has been changed. Need to
+  // modify the mmx and sse versions.
+  vp9_subtract_b_sse2_impl(z, src_stride, diff, predictor, pitch);
+}
+
+#endif
diff --git a/libvpx/vp9/exports_dec b/libvpx/vp9/exports_dec
new file mode 100644
index 0000000..0a61fde
--- /dev/null
+++ b/libvpx/vp9/exports_dec
@@ -0,0 +1,2 @@
+data vpx_codec_vp9_dx_algo
+text vpx_codec_vp9_dx
diff --git a/libvpx/vp9/exports_enc b/libvpx/vp9/exports_enc
new file mode 100644
index 0000000..25156e8
--- /dev/null
+++ b/libvpx/vp9/exports_enc
@@ -0,0 +1,4 @@
+data vpx_codec_vp9_cx_algo
+text vpx_codec_vp9_cx
+data vpx_codec_vp9x_cx_algo
+text vpx_codec_vp9x_cx
diff --git a/libvpx/vp9/vp9_common.mk b/libvpx/vp9/vp9_common.mk
new file mode 100644
index 0000000..7a74833
--- /dev/null
+++ b/libvpx/vp9/vp9_common.mk
@@ -0,0 +1,106 @@
+##
+##  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+VP9_COMMON_SRCS-yes += vp9_common.mk
+VP9_COMMON_SRCS-yes += vp9_iface_common.h
+VP9_COMMON_SRCS-yes += common/vp9_pragmas.h
+VP9_COMMON_SRCS-yes += common/vp9_ppflags.h
+VP9_COMMON_SRCS-yes += common/vp9_onyx.h
+VP9_COMMON_SRCS-yes += common/vp9_alloccommon.c
+VP9_COMMON_SRCS-yes += common/vp9_asm_com_offsets.c
+VP9_COMMON_SRCS-yes += common/vp9_convolve.c
+VP9_COMMON_SRCS-yes += common/vp9_convolve.h
+VP9_COMMON_SRCS-yes += common/vp9_debugmodes.c
+VP9_COMMON_SRCS-yes += common/vp9_default_coef_probs.h
+VP9_COMMON_SRCS-yes += common/vp9_entropy.c
+VP9_COMMON_SRCS-yes += common/vp9_entropymode.c
+VP9_COMMON_SRCS-yes += common/vp9_entropymv.c
+VP9_COMMON_SRCS-yes += common/vp9_extend.c
+VP9_COMMON_SRCS-yes += common/vp9_filter.c
+VP9_COMMON_SRCS-yes += common/vp9_filter.h
+VP9_COMMON_SRCS-yes += common/vp9_findnearmv.c
+VP9_COMMON_SRCS-yes += common/generic/vp9_systemdependent.c
+VP9_COMMON_SRCS-yes += common/vp9_idct.c
+VP9_COMMON_SRCS-yes += common/vp9_alloccommon.h
+VP9_COMMON_SRCS-yes += common/vp9_blockd.h
+VP9_COMMON_SRCS-yes += common/vp9_common.h
+VP9_COMMON_SRCS-yes += common/vp9_entropy.h
+VP9_COMMON_SRCS-yes += common/vp9_entropymode.h
+VP9_COMMON_SRCS-yes += common/vp9_entropymv.h
+VP9_COMMON_SRCS-yes += common/vp9_enums.h
+VP9_COMMON_SRCS-yes += common/vp9_extend.h
+VP9_COMMON_SRCS-yes += common/vp9_findnearmv.h
+VP9_COMMON_SRCS-yes += common/vp9_idct.h
+VP9_COMMON_SRCS-yes += common/vp9_loopfilter.h
+VP9_COMMON_SRCS-yes += common/vp9_modecont.h
+VP9_COMMON_SRCS-yes += common/vp9_mv.h
+VP9_COMMON_SRCS-yes += common/vp9_onyxc_int.h
+VP9_COMMON_SRCS-yes += common/vp9_pred_common.h
+VP9_COMMON_SRCS-yes += common/vp9_pred_common.c
+VP9_COMMON_SRCS-yes += common/vp9_quant_common.h
+VP9_COMMON_SRCS-yes += common/vp9_reconinter.h
+VP9_COMMON_SRCS-yes += common/vp9_reconintra.h
+VP9_COMMON_SRCS-yes += common/vp9_rtcd.c
+VP9_COMMON_SRCS-yes += common/vp9_rtcd_defs.sh
+VP9_COMMON_SRCS-yes += common/vp9_sadmxn.h
+VP9_COMMON_SRCS-yes += common/vp9_subpelvar.h
+VP9_COMMON_SRCS-yes += common/vp9_seg_common.h
+VP9_COMMON_SRCS-yes += common/vp9_seg_common.c
+VP9_COMMON_SRCS-yes += common/vp9_systemdependent.h
+VP9_COMMON_SRCS-yes += common/vp9_textblit.h
+VP9_COMMON_SRCS-yes += common/vp9_tile_common.h
+VP9_COMMON_SRCS-yes += common/vp9_tile_common.c
+VP9_COMMON_SRCS-yes += common/vp9_treecoder.h
+VP9_COMMON_SRCS-yes += common/vp9_loopfilter.c
+VP9_COMMON_SRCS-yes += common/vp9_loopfilter_filters.c
+VP9_COMMON_SRCS-yes += common/vp9_mbpitch.c
+VP9_COMMON_SRCS-yes += common/vp9_modecont.c
+VP9_COMMON_SRCS-yes += common/vp9_modecontext.c
+VP9_COMMON_SRCS-yes += common/vp9_mvref_common.c
+VP9_COMMON_SRCS-yes += common/vp9_mvref_common.h
+VP9_COMMON_SRCS-yes += common/vp9_quant_common.c
+VP9_COMMON_SRCS-yes += common/vp9_reconinter.c
+VP9_COMMON_SRCS-yes += common/vp9_reconintra.c
+VP9_COMMON_SRCS-$(CONFIG_POSTPROC_VISUALIZER) += common/vp9_textblit.c
+VP9_COMMON_SRCS-yes += common/vp9_treecoder.c
+VP9_COMMON_SRCS-$(CONFIG_IMPLICIT_SEGMENTATION) += common/vp9_implicit_segmentation.c
+
+VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_loopfilter_x86.h
+VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_postproc_x86.h
+VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_asm_stubs.c
+VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_loopfilter_intrin_sse2.c
+VP9_COMMON_SRCS-$(CONFIG_POSTPROC) += common/vp9_postproc.h
+VP9_COMMON_SRCS-$(CONFIG_POSTPROC) += common/vp9_postproc.c
+VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_iwalsh_mmx.asm
+VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_recon_mmx.asm
+VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_loopfilter_mmx.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_iwalsh_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_loopfilter_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_recon_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_recon_wrapper_sse2.c
+VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
+ifeq ($(CONFIG_POSTPROC),yes)
+VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm
+endif
+
+# common (c)
+ifeq ($(CONFIG_CSM),yes)
+VP9_COMMON_SRCS-yes += common/vp9_maskingmv.c
+VP9_COMMON_SRCS-$(HAVE_SSE3) += common/x86/vp9_mask_sse3.asm
+endif
+
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_sadmxn_sse2.c
+
+$(eval $(call asm_offsets_template,\
+         vp9_asm_com_offsets.asm, $(VP9_PREFIX)common/vp9_asm_com_offsets.c))
+
+$(eval $(call rtcd_h_template,vp9_rtcd,vp9/common/vp9_rtcd_defs.sh))
diff --git a/libvpx/vp9/vp9_cx_iface.c b/libvpx/vp9/vp9_cx_iface.c
new file mode 100644
index 0000000..e5b5089
--- /dev/null
+++ b/libvpx/vp9/vp9_cx_iface.c
@@ -0,0 +1,1153 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include "vpx/vpx_codec.h"
+#include "vpx/internal/vpx_codec_internal.h"
+#include "vpx_version.h"
+#include "vp9/encoder/vp9_onyx_int.h"
+#include "vpx/vp8cx.h"
+#include "vp9/encoder/vp9_firstpass.h"
+#include "vp9/common/vp9_onyx.h"
+#include "vp9/vp9_iface_common.h"
+#include <stdlib.h>
+#include <string.h>
+
+struct vp9_extracfg {
+  struct vpx_codec_pkt_list *pkt_list;
+  int                         cpu_used;                    /** available cpu percentage in 1/16*/
+  unsigned int                enable_auto_alt_ref;           /** if encoder decides to uses alternate reference frame */
+  unsigned int                noise_sensitivity;
+  unsigned int                Sharpness;
+  unsigned int                static_thresh;
+  unsigned int                tile_columns;
+  unsigned int                tile_rows;
+  unsigned int                arnr_max_frames;    /* alt_ref Noise Reduction Max Frame Count */
+  unsigned int                arnr_strength;    /* alt_ref Noise Reduction Strength */
+  unsigned int                arnr_type;        /* alt_ref filter type */
+  unsigned int                experimental;
+  vp8e_tuning                 tuning;
+  unsigned int                cq_level;         /* constrained quality level */
+  unsigned int                rc_max_intra_bitrate_pct;
+  unsigned int                lossless;
+  unsigned int                frame_parallel_decoding_mode;
+};
+
+struct extraconfig_map {
+  int                 usage;
+  struct vp9_extracfg cfg;
+};
+
+static const struct extraconfig_map extracfg_map[] = {
+  {
+    0,
+    {
+      NULL,
+      0,                          /* cpu_used      */
+      0,                          /* enable_auto_alt_ref */
+      0,                          /* noise_sensitivity */
+      0,                          /* Sharpness */
+      0,                          /* static_thresh */
+      0,                          /* tile_columns */
+      0,                          /* tile_rows */
+      0,                          /* arnr_max_frames */
+      3,                          /* arnr_strength */
+      3,                          /* arnr_type*/
+      0,                          /* experimental mode */
+      0,                          /* tuning*/
+      10,                         /* cq_level */
+      0,                          /* rc_max_intra_bitrate_pct */
+      0,                          /* lossless */
+      0,                          /* frame_parallel_decoding_mode */
+    }
+  }
+};
+
+struct vpx_codec_alg_priv {
+  vpx_codec_priv_t        base;
+  vpx_codec_enc_cfg_t     cfg;
+  struct vp9_extracfg     vp8_cfg;
+  VP9_CONFIG              oxcf;
+  VP9_PTR             cpi;
+  unsigned char          *cx_data;
+  unsigned int            cx_data_sz;
+  unsigned char          *pending_cx_data;
+  unsigned int            pending_cx_data_sz;
+  int                     pending_frame_count;
+  uint32_t                pending_frame_sizes[8];
+  uint32_t                pending_frame_magnitude;
+  vpx_image_t             preview_img;
+  vp8_postproc_cfg_t      preview_ppcfg;
+  vpx_codec_pkt_list_decl(64) pkt_list;              // changed to accomendate the maximum number of lagged frames allowed
+  unsigned int                fixed_kf_cntr;
+};
+
+
+static vpx_codec_err_t
+update_error_state(vpx_codec_alg_priv_t                 *ctx,
+                   const struct vpx_internal_error_info *error) {
+  vpx_codec_err_t res;
+
+  if ((res = error->error_code))
+    ctx->base.err_detail = error->has_detail
+                           ? error->detail
+                           : NULL;
+
+  return res;
+}
+
+
+#undef ERROR
+#define ERROR(str) do {\
+    ctx->base.err_detail = str;\
+    return VPX_CODEC_INVALID_PARAM;\
+  } while(0)
+
+#define RANGE_CHECK(p,memb,lo,hi) do {\
+    if(!(((p)->memb == lo || (p)->memb > (lo)) && (p)->memb <= hi)) \
+      ERROR(#memb " out of range ["#lo".."#hi"]");\
+  } while(0)
+
+#define RANGE_CHECK_HI(p,memb,hi) do {\
+    if(!((p)->memb <= (hi))) \
+      ERROR(#memb " out of range [.."#hi"]");\
+  } while(0)
+
+#define RANGE_CHECK_LO(p,memb,lo) do {\
+    if(!((p)->memb >= (lo))) \
+      ERROR(#memb " out of range ["#lo"..]");\
+  } while(0)
+
+#define RANGE_CHECK_BOOL(p,memb) do {\
+    if(!!((p)->memb) != (p)->memb) ERROR(#memb " expected boolean");\
+  } while(0)
+
+static vpx_codec_err_t validate_config(vpx_codec_alg_priv_t      *ctx,
+                                       const vpx_codec_enc_cfg_t *cfg,
+                                       const struct vp9_extracfg *vp8_cfg) {
+  RANGE_CHECK(cfg, g_w,                   1, 65535); /* 16 bits available */
+  RANGE_CHECK(cfg, g_h,                   1, 65535); /* 16 bits available */
+  RANGE_CHECK(cfg, g_timebase.den,        1, 1000000000);
+  RANGE_CHECK(cfg, g_timebase.num,        1, cfg->g_timebase.den);
+  RANGE_CHECK_HI(cfg, g_profile,          3);
+
+  RANGE_CHECK_HI(cfg, rc_max_quantizer,   63);
+  RANGE_CHECK_HI(cfg, rc_min_quantizer,   cfg->rc_max_quantizer);
+  RANGE_CHECK_BOOL(vp8_cfg, lossless);
+  if (vp8_cfg->lossless) {
+    RANGE_CHECK_HI(cfg, rc_max_quantizer, 0);
+    RANGE_CHECK_HI(cfg, rc_min_quantizer, 0);
+  }
+
+  RANGE_CHECK_HI(cfg, g_threads,          64);
+  RANGE_CHECK_HI(cfg, g_lag_in_frames,    MAX_LAG_BUFFERS);
+  RANGE_CHECK(cfg, rc_end_usage,          VPX_VBR, VPX_CQ);
+  RANGE_CHECK_HI(cfg, rc_undershoot_pct,  1000);
+  RANGE_CHECK_HI(cfg, rc_overshoot_pct,   1000);
+  RANGE_CHECK_HI(cfg, rc_2pass_vbr_bias_pct, 100);
+  RANGE_CHECK(cfg, kf_mode,               VPX_KF_DISABLED, VPX_KF_AUTO);
+  // RANGE_CHECK_BOOL(cfg,                 g_delete_firstpassfile);
+  RANGE_CHECK_BOOL(cfg,                   rc_resize_allowed);
+  RANGE_CHECK_HI(cfg, rc_dropframe_thresh,   100);
+  RANGE_CHECK_HI(cfg, rc_resize_up_thresh,   100);
+  RANGE_CHECK_HI(cfg, rc_resize_down_thresh, 100);
+  RANGE_CHECK(cfg,        g_pass,         VPX_RC_ONE_PASS, VPX_RC_LAST_PASS);
+
+  /* VP8 does not support a lower bound on the keyframe interval in
+   * automatic keyframe placement mode.
+   */
+  if (cfg->kf_mode != VPX_KF_DISABLED && cfg->kf_min_dist != cfg->kf_max_dist
+      && cfg->kf_min_dist > 0)
+    ERROR("kf_min_dist not supported in auto mode, use 0 "
+          "or kf_max_dist instead.");
+
+  RANGE_CHECK_BOOL(vp8_cfg,               enable_auto_alt_ref);
+  RANGE_CHECK(vp8_cfg, cpu_used,           -16, 16);
+
+  RANGE_CHECK_HI(vp8_cfg, noise_sensitivity,  6);
+
+  RANGE_CHECK(vp8_cfg, tile_columns, 0, 6);
+  RANGE_CHECK(vp8_cfg, tile_rows, 0, 2);
+  RANGE_CHECK_HI(vp8_cfg, Sharpness,       7);
+  RANGE_CHECK(vp8_cfg, arnr_max_frames, 0, 15);
+  RANGE_CHECK_HI(vp8_cfg, arnr_strength,   6);
+  RANGE_CHECK(vp8_cfg, arnr_type,       1, 3);
+  RANGE_CHECK(vp8_cfg, cq_level, 0, 63);
+
+  if (cfg->g_pass == VPX_RC_LAST_PASS) {
+    size_t           packet_sz = sizeof(FIRSTPASS_STATS);
+    int              n_packets = (int)(cfg->rc_twopass_stats_in.sz / packet_sz);
+    FIRSTPASS_STATS *stats;
+
+    if (!cfg->rc_twopass_stats_in.buf)
+      ERROR("rc_twopass_stats_in.buf not set.");
+
+    if (cfg->rc_twopass_stats_in.sz % packet_sz)
+      ERROR("rc_twopass_stats_in.sz indicates truncated packet.");
+
+    if (cfg->rc_twopass_stats_in.sz < 2 * packet_sz)
+      ERROR("rc_twopass_stats_in requires at least two packets.");
+
+    stats = (void *)((char *)cfg->rc_twopass_stats_in.buf
+                     + (n_packets - 1) * packet_sz);
+
+    if ((int)(stats->count + 0.5) != n_packets - 1)
+      ERROR("rc_twopass_stats_in missing EOS stats packet");
+  }
+
+  return VPX_CODEC_OK;
+}
+
+
+static vpx_codec_err_t validate_img(vpx_codec_alg_priv_t *ctx,
+                                    const vpx_image_t    *img) {
+  switch (img->fmt) {
+    case VPX_IMG_FMT_YV12:
+    case VPX_IMG_FMT_I420:
+    case VPX_IMG_FMT_I422:
+    case VPX_IMG_FMT_I444:
+      break;
+    default:
+      ERROR("Invalid image format. Only YV12, I420, I422, I444 images are "
+            "supported.");
+  }
+
+  if ((img->d_w != ctx->cfg.g_w) || (img->d_h != ctx->cfg.g_h))
+    ERROR("Image size must match encoder init configuration size");
+
+  return VPX_CODEC_OK;
+}
+
+
+static vpx_codec_err_t set_vp9e_config(VP9_CONFIG *oxcf,
+                                       vpx_codec_enc_cfg_t cfg,
+                                       struct vp9_extracfg vp8_cfg) {
+  oxcf->version = cfg.g_profile | (vp8_cfg.experimental ? 0x4 : 0);
+  oxcf->width   = cfg.g_w;
+  oxcf->height  = cfg.g_h;
+  /* guess a frame rate if out of whack, use 30 */
+  oxcf->frame_rate             = (double)(cfg.g_timebase.den) / (double)(cfg.g_timebase.num);
+
+  if (oxcf->frame_rate > 180) {
+    oxcf->frame_rate = 30;
+  }
+
+  switch (cfg.g_pass) {
+    case VPX_RC_ONE_PASS:
+      oxcf->Mode = MODE_BESTQUALITY;
+      break;
+    case VPX_RC_FIRST_PASS:
+      oxcf->Mode = MODE_FIRSTPASS;
+      break;
+    case VPX_RC_LAST_PASS:
+      oxcf->Mode = MODE_SECONDPASS_BEST;
+      break;
+  }
+
+  if (cfg.g_pass == VPX_RC_FIRST_PASS) {
+    oxcf->allow_lag              = 0;
+    oxcf->lag_in_frames           = 0;
+  } else {
+    oxcf->allow_lag              = (cfg.g_lag_in_frames) > 0;
+    oxcf->lag_in_frames           = cfg.g_lag_in_frames;
+  }
+
+  // VBR only supported for now.
+  // CBR code has been deprectated for experimental phase.
+  // CQ mode not yet tested
+  oxcf->end_usage          = USAGE_LOCAL_FILE_PLAYBACK;
+  /*if (cfg.rc_end_usage == VPX_CQ)
+      oxcf->end_usage      = USAGE_CONSTRAINED_QUALITY;
+  else
+      oxcf->end_usage      = USAGE_LOCAL_FILE_PLAYBACK;*/
+
+  oxcf->target_bandwidth       = cfg.rc_target_bitrate;
+  oxcf->rc_max_intra_bitrate_pct = vp8_cfg.rc_max_intra_bitrate_pct;
+
+  oxcf->best_allowed_q          = cfg.rc_min_quantizer;
+  oxcf->worst_allowed_q         = cfg.rc_max_quantizer;
+  oxcf->cq_level                = vp8_cfg.cq_level;
+  oxcf->fixed_q = -1;
+
+  oxcf->under_shoot_pct         = cfg.rc_undershoot_pct;
+  oxcf->over_shoot_pct          = cfg.rc_overshoot_pct;
+
+  oxcf->maximum_buffer_size     = cfg.rc_buf_sz;
+  oxcf->starting_buffer_level   = cfg.rc_buf_initial_sz;
+  oxcf->optimal_buffer_level    = cfg.rc_buf_optimal_sz;
+
+  oxcf->two_pass_vbrbias        = cfg.rc_2pass_vbr_bias_pct;
+  oxcf->two_pass_vbrmin_section  = cfg.rc_2pass_vbr_minsection_pct;
+  oxcf->two_pass_vbrmax_section  = cfg.rc_2pass_vbr_maxsection_pct;
+
+  oxcf->auto_key               = cfg.kf_mode == VPX_KF_AUTO
+                                 && cfg.kf_min_dist != cfg.kf_max_dist;
+  // oxcf->kf_min_dist         = cfg.kf_min_dis;
+  oxcf->key_freq               = cfg.kf_max_dist;
+
+  // oxcf->delete_first_pass_file = cfg.g_delete_firstpassfile;
+  // strcpy(oxcf->first_pass_file, cfg.g_firstpass_file);
+
+  oxcf->cpu_used               =  vp8_cfg.cpu_used;
+  oxcf->encode_breakout        =  vp8_cfg.static_thresh;
+  oxcf->play_alternate         =  vp8_cfg.enable_auto_alt_ref;
+  oxcf->noise_sensitivity      =  vp8_cfg.noise_sensitivity;
+  oxcf->Sharpness             =  vp8_cfg.Sharpness;
+
+  oxcf->two_pass_stats_in        =  cfg.rc_twopass_stats_in;
+  oxcf->output_pkt_list         =  vp8_cfg.pkt_list;
+
+  oxcf->arnr_max_frames = vp8_cfg.arnr_max_frames;
+  oxcf->arnr_strength =  vp8_cfg.arnr_strength;
+  oxcf->arnr_type =      vp8_cfg.arnr_type;
+
+  oxcf->tuning = vp8_cfg.tuning;
+
+  oxcf->tile_columns = vp8_cfg.tile_columns;
+  oxcf->tile_rows = vp8_cfg.tile_rows;
+
+  oxcf->lossless = vp8_cfg.lossless;
+
+  oxcf->error_resilient_mode = cfg.g_error_resilient;
+  oxcf->frame_parallel_decoding_mode = vp8_cfg.frame_parallel_decoding_mode;
+  /*
+  printf("Current VP9 Settings: \n");
+  printf("target_bandwidth: %d\n", oxcf->target_bandwidth);
+  printf("noise_sensitivity: %d\n", oxcf->noise_sensitivity);
+  printf("Sharpness: %d\n",    oxcf->Sharpness);
+  printf("cpu_used: %d\n",  oxcf->cpu_used);
+  printf("Mode: %d\n",     oxcf->Mode);
+  // printf("delete_first_pass_file: %d\n",  oxcf->delete_first_pass_file);
+  printf("auto_key: %d\n",  oxcf->auto_key);
+  printf("key_freq: %d\n", oxcf->key_freq);
+  printf("end_usage: %d\n", oxcf->end_usage);
+  printf("under_shoot_pct: %d\n", oxcf->under_shoot_pct);
+  printf("over_shoot_pct: %d\n", oxcf->over_shoot_pct);
+  printf("starting_buffer_level: %d\n", oxcf->starting_buffer_level);
+  printf("optimal_buffer_level: %d\n",  oxcf->optimal_buffer_level);
+  printf("maximum_buffer_size: %d\n", oxcf->maximum_buffer_size);
+  printf("fixed_q: %d\n",  oxcf->fixed_q);
+  printf("worst_allowed_q: %d\n", oxcf->worst_allowed_q);
+  printf("best_allowed_q: %d\n", oxcf->best_allowed_q);
+  printf("two_pass_vbrbias: %d\n",  oxcf->two_pass_vbrbias);
+  printf("two_pass_vbrmin_section: %d\n", oxcf->two_pass_vbrmin_section);
+  printf("two_pass_vbrmax_section: %d\n", oxcf->two_pass_vbrmax_section);
+  printf("allow_lag: %d\n", oxcf->allow_lag);
+  printf("lag_in_frames: %d\n", oxcf->lag_in_frames);
+  printf("play_alternate: %d\n", oxcf->play_alternate);
+  printf("Version: %d\n", oxcf->Version);
+  printf("encode_breakout: %d\n", oxcf->encode_breakout);
+  printf("error resilient: %d\n", oxcf->error_resilient_mode);
+  printf("frame parallel detokenization: %d\n",
+         oxcf->frame_parallel_decoding_mode);
+  */
+  return VPX_CODEC_OK;
+}
+
+static vpx_codec_err_t vp9e_set_config(vpx_codec_alg_priv_t       *ctx,
+                                       const vpx_codec_enc_cfg_t  *cfg) {
+  vpx_codec_err_t res;
+
+  if ((cfg->g_w != ctx->cfg.g_w) || (cfg->g_h != ctx->cfg.g_h))
+    ERROR("Cannot change width or height after initialization");
+
+  /* Prevent increasing lag_in_frames. This check is stricter than it needs
+   * to be -- the limit is not increasing past the first lag_in_frames
+   * value, but we don't track the initial config, only the last successful
+   * config.
+   */
+  if ((cfg->g_lag_in_frames > ctx->cfg.g_lag_in_frames))
+    ERROR("Cannot increase lag_in_frames");
+
+  res = validate_config(ctx, cfg, &ctx->vp8_cfg);
+
+  if (!res) {
+    ctx->cfg = *cfg;
+    set_vp9e_config(&ctx->oxcf, ctx->cfg, ctx->vp8_cfg);
+    vp9_change_config(ctx->cpi, &ctx->oxcf);
+  }
+
+  return res;
+}
+
+
+int vp9_reverse_trans(int q);
+
+
+static vpx_codec_err_t get_param(vpx_codec_alg_priv_t *ctx,
+                                 int                   ctrl_id,
+                                 va_list               args) {
+  void *arg = va_arg(args, void *);
+
+#define MAP(id, var) case id: *(RECAST(id, arg)) = var; break
+
+  if (!arg)
+    return VPX_CODEC_INVALID_PARAM;
+
+  switch (ctrl_id) {
+      MAP(VP8E_GET_LAST_QUANTIZER, vp9_get_quantizer(ctx->cpi));
+      MAP(VP8E_GET_LAST_QUANTIZER_64,
+          vp9_reverse_trans(vp9_get_quantizer(ctx->cpi)));
+  }
+
+  return VPX_CODEC_OK;
+#undef MAP
+}
+
+
+static vpx_codec_err_t set_param(vpx_codec_alg_priv_t *ctx,
+                                 int                   ctrl_id,
+                                 va_list               args) {
+  vpx_codec_err_t     res  = VPX_CODEC_OK;
+  struct vp9_extracfg xcfg = ctx->vp8_cfg;
+
+#define MAP(id, var) case id: var = CAST(id, args); break;
+
+  switch (ctrl_id) {
+      MAP(VP8E_SET_CPUUSED,               xcfg.cpu_used);
+      MAP(VP8E_SET_ENABLEAUTOALTREF,      xcfg.enable_auto_alt_ref);
+      MAP(VP8E_SET_NOISE_SENSITIVITY,     xcfg.noise_sensitivity);
+      MAP(VP8E_SET_SHARPNESS,             xcfg.Sharpness);
+      MAP(VP8E_SET_STATIC_THRESHOLD,      xcfg.static_thresh);
+      MAP(VP9E_SET_TILE_COLUMNS,          xcfg.tile_columns);
+      MAP(VP9E_SET_TILE_ROWS,             xcfg.tile_rows);
+
+      MAP(VP8E_SET_ARNR_MAXFRAMES,        xcfg.arnr_max_frames);
+      MAP(VP8E_SET_ARNR_STRENGTH,        xcfg.arnr_strength);
+      MAP(VP8E_SET_ARNR_TYPE,        xcfg.arnr_type);
+      MAP(VP8E_SET_TUNING,                xcfg.tuning);
+      MAP(VP8E_SET_CQ_LEVEL,              xcfg.cq_level);
+      MAP(VP8E_SET_MAX_INTRA_BITRATE_PCT, xcfg.rc_max_intra_bitrate_pct);
+      MAP(VP9E_SET_LOSSLESS,              xcfg.lossless);
+      MAP(VP9E_SET_FRAME_PARALLEL_DECODING, xcfg.frame_parallel_decoding_mode);
+  }
+
+  res = validate_config(ctx, &ctx->cfg, &xcfg);
+
+  if (!res) {
+    ctx->vp8_cfg = xcfg;
+    set_vp9e_config(&ctx->oxcf, ctx->cfg, ctx->vp8_cfg);
+    vp9_change_config(ctx->cpi, &ctx->oxcf);
+  }
+
+  return res;
+#undef MAP
+}
+
+
+static vpx_codec_err_t vp9e_common_init(vpx_codec_ctx_t *ctx,
+                                        int              experimental) {
+  vpx_codec_err_t            res = VPX_CODEC_OK;
+  struct vpx_codec_alg_priv *priv;
+  vpx_codec_enc_cfg_t       *cfg;
+  unsigned int               i;
+
+  VP9_PTR optr;
+
+  if (!ctx->priv) {
+    priv = calloc(1, sizeof(struct vpx_codec_alg_priv));
+
+    if (!priv) {
+      return VPX_CODEC_MEM_ERROR;
+    }
+
+    ctx->priv = &priv->base;
+    ctx->priv->sz = sizeof(*ctx->priv);
+    ctx->priv->iface = ctx->iface;
+    ctx->priv->alg_priv = priv;
+    ctx->priv->init_flags = ctx->init_flags;
+    ctx->priv->enc.total_encoders = 1;
+
+    if (ctx->config.enc) {
+      /* Update the reference to the config structure to an
+       * internal copy.
+       */
+      ctx->priv->alg_priv->cfg = *ctx->config.enc;
+      ctx->config.enc = &ctx->priv->alg_priv->cfg;
+    }
+
+    cfg =  &ctx->priv->alg_priv->cfg;
+
+    /* Select the extra vp6 configuration table based on the current
+     * usage value. If the current usage value isn't found, use the
+     * values for usage case 0.
+     */
+    for (i = 0;
+         extracfg_map[i].usage && extracfg_map[i].usage != cfg->g_usage;
+         i++);
+
+    priv->vp8_cfg = extracfg_map[i].cfg;
+    priv->vp8_cfg.pkt_list = &priv->pkt_list.head;
+    priv->vp8_cfg.experimental = experimental;
+
+    // TODO(agrange) Check the limits set on this buffer, or the check that is
+    // applied in vp9e_encode.
+    priv->cx_data_sz = priv->cfg.g_w * priv->cfg.g_h * 3 / 2 * 8;
+//    priv->cx_data_sz = priv->cfg.g_w * priv->cfg.g_h * 3 / 2 * 2;
+
+    if (priv->cx_data_sz < 4096) priv->cx_data_sz = 4096;
+
+    priv->cx_data = malloc(priv->cx_data_sz);
+
+    if (!priv->cx_data) {
+      return VPX_CODEC_MEM_ERROR;
+    }
+
+    vp9_initialize_enc();
+
+    res = validate_config(priv, &priv->cfg, &priv->vp8_cfg);
+
+    if (!res) {
+      set_vp9e_config(&ctx->priv->alg_priv->oxcf,
+                      ctx->priv->alg_priv->cfg,
+                      ctx->priv->alg_priv->vp8_cfg);
+      optr = vp9_create_compressor(&ctx->priv->alg_priv->oxcf);
+
+      if (!optr)
+        res = VPX_CODEC_MEM_ERROR;
+      else
+        ctx->priv->alg_priv->cpi = optr;
+    }
+  }
+
+  return res;
+}
+
+
+static vpx_codec_err_t vp9e_init(vpx_codec_ctx_t *ctx,
+                                 vpx_codec_priv_enc_mr_cfg_t *data) {
+  return vp9e_common_init(ctx, 0);
+}
+
+
+#if CONFIG_EXPERIMENTAL
+static vpx_codec_err_t vp9e_exp_init(vpx_codec_ctx_t *ctx,
+                                     vpx_codec_priv_enc_mr_cfg_t *data) {
+  return vp9e_common_init(ctx, 1);
+}
+#endif
+
+
+static vpx_codec_err_t vp9e_destroy(vpx_codec_alg_priv_t *ctx) {
+
+  free(ctx->cx_data);
+  vp9_remove_compressor(&ctx->cpi);
+  free(ctx);
+  return VPX_CODEC_OK;
+}
+
+static void pick_quickcompress_mode(vpx_codec_alg_priv_t  *ctx,
+                                    unsigned long          duration,
+                                    unsigned long          deadline) {
+  unsigned int new_qc;
+
+  /* Use best quality mode if no deadline is given. */
+  if (deadline)
+    new_qc = MODE_GOODQUALITY;
+  else
+    new_qc = MODE_BESTQUALITY;
+
+  if (ctx->cfg.g_pass == VPX_RC_FIRST_PASS)
+    new_qc = MODE_FIRSTPASS;
+  else if (ctx->cfg.g_pass == VPX_RC_LAST_PASS)
+    new_qc = (new_qc == MODE_BESTQUALITY)
+             ? MODE_SECONDPASS_BEST
+             : MODE_SECONDPASS;
+
+  if (ctx->oxcf.Mode != new_qc) {
+    ctx->oxcf.Mode = new_qc;
+    vp9_change_config(ctx->cpi, &ctx->oxcf);
+  }
+}
+
+
+static int write_superframe_index(vpx_codec_alg_priv_t *ctx) {
+  uint8_t marker = 0xc0;
+  int mag, mask, index_sz;
+
+  assert(ctx->pending_frame_count);
+  assert(ctx->pending_frame_count <= 8);
+
+  /* Add the number of frames to the marker byte */
+  marker |= ctx->pending_frame_count - 1;
+
+  /* Choose the magnitude */
+  for (mag = 0, mask = 0xff; mag < 4; mag++) {
+    if (ctx->pending_frame_magnitude < mask)
+      break;
+    mask <<= 8;
+    mask |= 0xff;
+  }
+  marker |= mag << 3;
+
+  /* Write the index */
+  index_sz = 2 + (mag + 1) * ctx->pending_frame_count;
+  if (ctx->pending_cx_data_sz + index_sz < ctx->cx_data_sz) {
+    uint8_t *x = ctx->pending_cx_data + ctx->pending_cx_data_sz;
+    int i, j;
+
+    *x++ = marker;
+    for (i = 0; i < ctx->pending_frame_count; i++) {
+      int this_sz = ctx->pending_frame_sizes[i];
+
+      for (j = 0; j <= mag; j++) {
+        *x++ = this_sz & 0xff;
+        this_sz >>= 8;
+      }
+    }
+    *x++ = marker;
+    ctx->pending_cx_data_sz += index_sz;
+  }
+  return index_sz;
+}
+
+static vpx_codec_err_t vp9e_encode(vpx_codec_alg_priv_t  *ctx,
+                                   const vpx_image_t     *img,
+                                   vpx_codec_pts_t        pts,
+                                   unsigned long          duration,
+                                   vpx_enc_frame_flags_t  flags,
+                                   unsigned long          deadline) {
+  vpx_codec_err_t res = VPX_CODEC_OK;
+
+  if (img)
+    res = validate_img(ctx, img);
+
+  pick_quickcompress_mode(ctx, duration, deadline);
+  vpx_codec_pkt_list_init(&ctx->pkt_list);
+
+  /* Handle Flags */
+  if (((flags & VP8_EFLAG_NO_UPD_GF) && (flags & VP8_EFLAG_FORCE_GF))
+      || ((flags & VP8_EFLAG_NO_UPD_ARF) && (flags & VP8_EFLAG_FORCE_ARF))) {
+    ctx->base.err_detail = "Conflicting flags.";
+    return VPX_CODEC_INVALID_PARAM;
+  }
+
+  if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF
+               | VP8_EFLAG_NO_REF_ARF)) {
+    int ref = 7;
+
+    if (flags & VP8_EFLAG_NO_REF_LAST)
+      ref ^= VP9_LAST_FLAG;
+
+    if (flags & VP8_EFLAG_NO_REF_GF)
+      ref ^= VP9_GOLD_FLAG;
+
+    if (flags & VP8_EFLAG_NO_REF_ARF)
+      ref ^= VP9_ALT_FLAG;
+
+    vp9_use_as_reference(ctx->cpi, ref);
+  }
+
+  if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF
+               | VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF
+               | VP8_EFLAG_FORCE_ARF)) {
+    int upd = 7;
+
+    if (flags & VP8_EFLAG_NO_UPD_LAST)
+      upd ^= VP9_LAST_FLAG;
+
+    if (flags & VP8_EFLAG_NO_UPD_GF)
+      upd ^= VP9_GOLD_FLAG;
+
+    if (flags & VP8_EFLAG_NO_UPD_ARF)
+      upd ^= VP9_ALT_FLAG;
+
+    vp9_update_reference(ctx->cpi, upd);
+  }
+
+  if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
+    vp9_update_entropy(ctx->cpi, 0);
+  }
+
+  /* Handle fixed keyframe intervals */
+  if (ctx->cfg.kf_mode == VPX_KF_AUTO
+      && ctx->cfg.kf_min_dist == ctx->cfg.kf_max_dist) {
+    if (++ctx->fixed_kf_cntr > ctx->cfg.kf_min_dist) {
+      flags |= VPX_EFLAG_FORCE_KF;
+      ctx->fixed_kf_cntr = 1;
+    }
+  }
+
+  /* Initialize the encoder instance on the first frame*/
+  if (!res && ctx->cpi) {
+    unsigned int lib_flags;
+    YV12_BUFFER_CONFIG sd;
+    int64_t dst_time_stamp, dst_end_time_stamp;
+    unsigned long size, cx_data_sz;
+    unsigned char *cx_data;
+
+    /* Set up internal flags */
+    if (ctx->base.init_flags & VPX_CODEC_USE_PSNR)
+      ((VP9_COMP *)ctx->cpi)->b_calculate_psnr = 1;
+
+    // if (ctx->base.init_flags & VPX_CODEC_USE_OUTPUT_PARTITION)
+    //    ((VP9_COMP *)ctx->cpi)->output_partition = 1;
+
+    /* Convert API flags to internal codec lib flags */
+    lib_flags = (flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
+
+    /* vp8 use 10,000,000 ticks/second as time stamp */
+    dst_time_stamp    = pts * 10000000 * ctx->cfg.g_timebase.num / ctx->cfg.g_timebase.den;
+    dst_end_time_stamp = (pts + duration) * 10000000 * ctx->cfg.g_timebase.num / ctx->cfg.g_timebase.den;
+
+    if (img != NULL) {
+      res = image2yuvconfig(img, &sd);
+
+      if (vp9_receive_raw_frame(ctx->cpi, lib_flags,
+                                &sd, dst_time_stamp, dst_end_time_stamp)) {
+        VP9_COMP *cpi = (VP9_COMP *)ctx->cpi;
+        res = update_error_state(ctx, &cpi->common.error);
+      }
+    }
+
+    cx_data = ctx->cx_data;
+    cx_data_sz = ctx->cx_data_sz;
+    lib_flags = 0;
+
+    /* Any pending invisible frames? */
+    if (ctx->pending_cx_data) {
+      memmove(cx_data, ctx->pending_cx_data, ctx->pending_cx_data_sz);
+      ctx->pending_cx_data = cx_data;
+      cx_data += ctx->pending_cx_data_sz;
+      cx_data_sz -= ctx->pending_cx_data_sz;
+
+      /* TODO: this is a minimal check, the underlying codec doesn't respect
+       * the buffer size anyway.
+       */
+      if (cx_data_sz < ctx->cx_data_sz / 2) {
+        ctx->base.err_detail = "Compressed data buffer too small";
+        return VPX_CODEC_ERROR;
+      }
+    }
+
+    while (cx_data_sz >= ctx->cx_data_sz / 2 &&
+           -1 != vp9_get_compressed_data(ctx->cpi, &lib_flags, &size,
+                                         cx_data, &dst_time_stamp,
+                                         &dst_end_time_stamp, !img)) {
+      if (size) {
+        vpx_codec_pts_t    round, delta;
+        vpx_codec_cx_pkt_t pkt;
+        VP9_COMP *cpi = (VP9_COMP *)ctx->cpi;
+
+        /* Pack invisible frames with the next visible frame */
+        if (!cpi->common.show_frame) {
+          if (!ctx->pending_cx_data)
+            ctx->pending_cx_data = cx_data;
+          ctx->pending_cx_data_sz += size;
+          ctx->pending_frame_sizes[ctx->pending_frame_count++] = size;
+          ctx->pending_frame_magnitude |= size;
+          cx_data += size;
+          cx_data_sz -= size;
+          continue;
+        }
+
+        /* Add the frame packet to the list of returned packets. */
+        round = 1000000 * ctx->cfg.g_timebase.num / 2 - 1;
+        delta = (dst_end_time_stamp - dst_time_stamp);
+        pkt.kind = VPX_CODEC_CX_FRAME_PKT;
+        pkt.data.frame.pts =
+          (dst_time_stamp * ctx->cfg.g_timebase.den + round)
+          / ctx->cfg.g_timebase.num / 10000000;
+        pkt.data.frame.duration = (unsigned long)
+          ((delta * ctx->cfg.g_timebase.den + round)
+          / ctx->cfg.g_timebase.num / 10000000);
+        pkt.data.frame.flags = lib_flags << 16;
+
+        if (lib_flags & FRAMEFLAGS_KEY)
+          pkt.data.frame.flags |= VPX_FRAME_IS_KEY;
+
+        if (!cpi->common.show_frame) {
+          pkt.data.frame.flags |= VPX_FRAME_IS_INVISIBLE;
+
+          // This timestamp should be as close as possible to the
+          // prior PTS so that if a decoder uses pts to schedule when
+          // to do this, we start right after last frame was decoded.
+          // Invisible frames have no duration.
+          pkt.data.frame.pts = ((cpi->last_time_stamp_seen
+                                 * ctx->cfg.g_timebase.den + round)
+                                / ctx->cfg.g_timebase.num / 10000000) + 1;
+          pkt.data.frame.duration = 0;
+        }
+
+        if (cpi->droppable)
+          pkt.data.frame.flags |= VPX_FRAME_IS_DROPPABLE;
+
+        /*if (cpi->output_partition)
+        {
+            int i;
+            const int num_partitions = 1;
+
+            pkt.data.frame.flags |= VPX_FRAME_IS_FRAGMENT;
+
+            for (i = 0; i < num_partitions; ++i)
+            {
+                pkt.data.frame.buf = cx_data;
+                pkt.data.frame.sz = cpi->partition_sz[i];
+                pkt.data.frame.partition_id = i;
+                // don't set the fragment bit for the last partition
+                if (i == (num_partitions - 1))
+                    pkt.data.frame.flags &= ~VPX_FRAME_IS_FRAGMENT;
+                vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
+                cx_data += cpi->partition_sz[i];
+                cx_data_sz -= cpi->partition_sz[i];
+            }
+        }
+        else*/
+        {
+          if (ctx->pending_cx_data) {
+            ctx->pending_frame_sizes[ctx->pending_frame_count++] = size;
+            ctx->pending_frame_magnitude |= size;
+            ctx->pending_cx_data_sz += size;
+            size += write_superframe_index(ctx);
+            pkt.data.frame.buf = ctx->pending_cx_data;
+            pkt.data.frame.sz  = ctx->pending_cx_data_sz;
+            ctx->pending_cx_data = NULL;
+            ctx->pending_cx_data_sz = 0;
+            ctx->pending_frame_count = 0;
+            ctx->pending_frame_magnitude = 0;
+          } else {
+            pkt.data.frame.buf = cx_data;
+            pkt.data.frame.sz  = size;
+          }
+          pkt.data.frame.partition_id = -1;
+          vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
+          cx_data += size;
+          cx_data_sz -= size;
+        }
+
+        // printf("timestamp: %lld, duration: %d\n", pkt->data.frame.pts, pkt->data.frame.duration);
+      }
+    }
+  }
+
+  return res;
+}
+
+
+static const vpx_codec_cx_pkt_t *vp9e_get_cxdata(vpx_codec_alg_priv_t  *ctx,
+                                                 vpx_codec_iter_t      *iter) {
+  return vpx_codec_pkt_list_get(&ctx->pkt_list.head, iter);
+}
+
+static vpx_codec_err_t vp9e_set_reference(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+  vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
+
+  if (data) {
+    vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
+    YV12_BUFFER_CONFIG sd;
+
+    image2yuvconfig(&frame->img, &sd);
+    vp9_set_reference_enc(ctx->cpi, frame->frame_type, &sd);
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+
+}
+
+static vpx_codec_err_t vp9e_copy_reference(vpx_codec_alg_priv_t *ctx,
+                                           int ctr_id,
+                                           va_list args) {
+
+  vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
+
+  if (data) {
+    vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
+    YV12_BUFFER_CONFIG sd;
+
+    image2yuvconfig(&frame->img, &sd);
+    vp9_copy_reference_enc(ctx->cpi, frame->frame_type, &sd);
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+}
+
+static vpx_codec_err_t get_reference(vpx_codec_alg_priv_t *ctx,
+                                     int ctr_id,
+                                     va_list args) {
+  vp9_ref_frame_t *data = va_arg(args, vp9_ref_frame_t *);
+
+  if (data) {
+    YV12_BUFFER_CONFIG* fb;
+
+    vp9_get_reference_enc(ctx->cpi, data->idx, &fb);
+    yuvconfig2image(&data->img, fb, NULL);
+    return VPX_CODEC_OK;
+  } else {
+    return VPX_CODEC_INVALID_PARAM;
+  }
+}
+
+static vpx_codec_err_t vp9e_set_previewpp(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+#if CONFIG_POSTPROC
+  vp8_postproc_cfg_t *data = va_arg(args, vp8_postproc_cfg_t *);
+  (void)ctr_id;
+
+  if (data) {
+    ctx->preview_ppcfg = *((vp8_postproc_cfg_t *)data);
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+#else
+  (void)ctx;
+  (void)ctr_id;
+  (void)args;
+  return VPX_CODEC_INCAPABLE;
+#endif
+}
+
+
+static vpx_image_t *vp9e_get_preview(vpx_codec_alg_priv_t *ctx) {
+
+  YV12_BUFFER_CONFIG sd;
+  vp9_ppflags_t flags = {0};
+
+  if (ctx->preview_ppcfg.post_proc_flag) {
+    flags.post_proc_flag        = ctx->preview_ppcfg.post_proc_flag;
+    flags.deblocking_level      = ctx->preview_ppcfg.deblocking_level;
+    flags.noise_level           = ctx->preview_ppcfg.noise_level;
+  }
+
+  if (0 == vp9_get_preview_raw_frame(ctx->cpi, &sd, &flags)) {
+    yuvconfig2image(&ctx->preview_img, &sd, NULL);
+    return &ctx->preview_img;
+  } else
+    return NULL;
+}
+
+static vpx_codec_err_t vp9e_update_entropy(vpx_codec_alg_priv_t *ctx,
+                                           int ctr_id,
+                                           va_list args) {
+  int update = va_arg(args, int);
+  vp9_update_entropy(ctx->cpi, update);
+  return VPX_CODEC_OK;
+
+}
+
+static vpx_codec_err_t vp9e_update_reference(vpx_codec_alg_priv_t *ctx,
+                                             int ctr_id,
+                                             va_list args) {
+  int update = va_arg(args, int);
+  vp9_update_reference(ctx->cpi, update);
+  return VPX_CODEC_OK;
+}
+
+static vpx_codec_err_t vp9e_use_reference(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+  int reference_flag = va_arg(args, int);
+  vp9_use_as_reference(ctx->cpi, reference_flag);
+  return VPX_CODEC_OK;
+}
+
+static vpx_codec_err_t vp9e_set_roi_map(vpx_codec_alg_priv_t *ctx,
+                                        int ctr_id,
+                                        va_list args) {
+  vpx_roi_map_t *data = va_arg(args, vpx_roi_map_t *);
+
+  if (data) {
+    vpx_roi_map_t *roi = (vpx_roi_map_t *)data;
+
+    if (!vp9_set_roimap(ctx->cpi, roi->roi_map, roi->rows, roi->cols,
+                        roi->delta_q, roi->delta_lf, roi->static_threshold))
+      return VPX_CODEC_OK;
+    else
+      return VPX_CODEC_INVALID_PARAM;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+}
+
+
+static vpx_codec_err_t vp9e_set_activemap(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+  vpx_active_map_t *data = va_arg(args, vpx_active_map_t *);
+
+  if (data) {
+
+    vpx_active_map_t *map = (vpx_active_map_t *)data;
+
+    if (!vp9_set_active_map(ctx->cpi, map->active_map, map->rows, map->cols))
+      return VPX_CODEC_OK;
+    else
+      return VPX_CODEC_INVALID_PARAM;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+}
+
+static vpx_codec_err_t vp9e_set_scalemode(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+
+  vpx_scaling_mode_t *data =  va_arg(args, vpx_scaling_mode_t *);
+
+  if (data) {
+    int res;
+    vpx_scaling_mode_t scalemode = *(vpx_scaling_mode_t *)data;
+    res = vp9_set_internal_size(ctx->cpi, scalemode.h_scaling_mode,
+                                scalemode.v_scaling_mode);
+
+    if (!res) {
+      return VPX_CODEC_OK;
+    } else
+      return VPX_CODEC_INVALID_PARAM;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+}
+
+
+static vpx_codec_ctrl_fn_map_t vp9e_ctf_maps[] = {
+  {VP8_SET_REFERENCE,                 vp9e_set_reference},
+  {VP8_COPY_REFERENCE,                vp9e_copy_reference},
+  {VP8_SET_POSTPROC,                  vp9e_set_previewpp},
+  {VP8E_UPD_ENTROPY,                  vp9e_update_entropy},
+  {VP8E_UPD_REFERENCE,                vp9e_update_reference},
+  {VP8E_USE_REFERENCE,                vp9e_use_reference},
+  {VP8E_SET_ROI_MAP,                  vp9e_set_roi_map},
+  {VP8E_SET_ACTIVEMAP,                vp9e_set_activemap},
+  {VP8E_SET_SCALEMODE,                vp9e_set_scalemode},
+  {VP8E_SET_CPUUSED,                  set_param},
+  {VP8E_SET_NOISE_SENSITIVITY,        set_param},
+  {VP8E_SET_ENABLEAUTOALTREF,         set_param},
+  {VP8E_SET_SHARPNESS,                set_param},
+  {VP8E_SET_STATIC_THRESHOLD,         set_param},
+  {VP9E_SET_TILE_COLUMNS,             set_param},
+  {VP9E_SET_TILE_ROWS,                set_param},
+  {VP8E_GET_LAST_QUANTIZER,           get_param},
+  {VP8E_GET_LAST_QUANTIZER_64,        get_param},
+  {VP8E_SET_ARNR_MAXFRAMES,           set_param},
+  {VP8E_SET_ARNR_STRENGTH,           set_param},
+  {VP8E_SET_ARNR_TYPE,           set_param},
+  {VP8E_SET_TUNING,                   set_param},
+  {VP8E_SET_CQ_LEVEL,                 set_param},
+  {VP8E_SET_MAX_INTRA_BITRATE_PCT,    set_param},
+  {VP9E_SET_LOSSLESS,                 set_param},
+  {VP9_GET_REFERENCE,                 get_reference},
+  { -1, NULL},
+};
+
+static vpx_codec_enc_cfg_map_t vp9e_usage_cfg_map[] = {
+  {
+    0,
+    {
+      0,                  /* g_usage */
+      0,                  /* g_threads */
+      0,                  /* g_profile */
+
+      320,                /* g_width */
+      240,                /* g_height */
+      {1, 30},            /* g_timebase */
+
+      0,                  /* g_error_resilient */
+
+      VPX_RC_ONE_PASS,    /* g_pass */
+
+      0,                  /* g_lag_in_frames */
+
+      0,                  /* rc_dropframe_thresh */
+      0,                  /* rc_resize_allowed */
+      60,                 /* rc_resize_down_thresold */
+      30,                 /* rc_resize_up_thresold */
+
+      VPX_VBR,            /* rc_end_usage */
+#if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
+      {0},                /* rc_twopass_stats_in */
+#endif
+      256,                /* rc_target_bandwidth */
+      4,                  /* rc_min_quantizer */
+      63,                 /* rc_max_quantizer */
+      100,                /* rc_undershoot_pct */
+      100,                /* rc_overshoot_pct */
+
+      6000,               /* rc_max_buffer_size */
+      4000,               /* rc_buffer_initial_size; */
+      5000,               /* rc_buffer_optimal_size; */
+
+      50,                 /* rc_two_pass_vbrbias  */
+      0,                  /* rc_two_pass_vbrmin_section */
+      400,                /* rc_two_pass_vbrmax_section */
+
+      /* keyframing settings (kf) */
+      VPX_KF_AUTO,        /* g_kfmode*/
+      0,                  /* kf_min_dist */
+      9999,               /* kf_max_dist */
+
+#if VPX_ENCODER_ABI_VERSION == (1 + VPX_CODEC_ABI_VERSION)
+      1,                  /* g_delete_first_pass_file */
+      "vp8.fpf"           /* first pass filename */
+#endif
+    }
+  },
+  { -1, {NOT_IMPLEMENTED}}
+};
+
+
+#ifndef VERSION_STRING
+#define VERSION_STRING
+#endif
+CODEC_INTERFACE(vpx_codec_vp9_cx) = {
+  "WebM Project VP9 Encoder" VERSION_STRING,
+  VPX_CODEC_INTERNAL_ABI_VERSION,
+  VPX_CODEC_CAP_ENCODER | VPX_CODEC_CAP_PSNR |
+  VPX_CODEC_CAP_OUTPUT_PARTITION,
+  /* vpx_codec_caps_t          caps; */
+  vp9e_init,          /* vpx_codec_init_fn_t       init; */
+  vp9e_destroy,       /* vpx_codec_destroy_fn_t    destroy; */
+  vp9e_ctf_maps,      /* vpx_codec_ctrl_fn_map_t  *ctrl_maps; */
+  NOT_IMPLEMENTED,    /* vpx_codec_get_mmap_fn_t   get_mmap; */
+  NOT_IMPLEMENTED,    /* vpx_codec_set_mmap_fn_t   set_mmap; */
+  {
+    NOT_IMPLEMENTED,    /* vpx_codec_peek_si_fn_t    peek_si; */
+    NOT_IMPLEMENTED,    /* vpx_codec_get_si_fn_t     get_si; */
+    NOT_IMPLEMENTED,    /* vpx_codec_decode_fn_t     decode; */
+    NOT_IMPLEMENTED,    /* vpx_codec_frame_get_fn_t  frame_get; */
+  },
+  {
+    vp9e_usage_cfg_map, /* vpx_codec_enc_cfg_map_t    peek_si; */
+    vp9e_encode,        /* vpx_codec_encode_fn_t      encode; */
+    vp9e_get_cxdata,    /* vpx_codec_get_cx_data_fn_t   frame_get; */
+    vp9e_set_config,
+    NOT_IMPLEMENTED,
+    vp9e_get_preview,
+  } /* encoder functions */
+};
+
+
+#if CONFIG_EXPERIMENTAL
+
+CODEC_INTERFACE(vpx_codec_vp9x_cx) = {
+  "VP8 Experimental Encoder" VERSION_STRING,
+  VPX_CODEC_INTERNAL_ABI_VERSION,
+  VPX_CODEC_CAP_ENCODER | VPX_CODEC_CAP_PSNR,
+  /* vpx_codec_caps_t          caps; */
+  vp9e_exp_init,      /* vpx_codec_init_fn_t       init; */
+  vp9e_destroy,       /* vpx_codec_destroy_fn_t    destroy; */
+  vp9e_ctf_maps,      /* vpx_codec_ctrl_fn_map_t  *ctrl_maps; */
+  NOT_IMPLEMENTED,    /* vpx_codec_get_mmap_fn_t   get_mmap; */
+  NOT_IMPLEMENTED,    /* vpx_codec_set_mmap_fn_t   set_mmap; */
+  {
+    NOT_IMPLEMENTED,    /* vpx_codec_peek_si_fn_t    peek_si; */
+    NOT_IMPLEMENTED,    /* vpx_codec_get_si_fn_t     get_si; */
+    NOT_IMPLEMENTED,    /* vpx_codec_decode_fn_t     decode; */
+    NOT_IMPLEMENTED,    /* vpx_codec_frame_get_fn_t  frame_get; */
+  },
+  {
+    vp9e_usage_cfg_map, /* vpx_codec_enc_cfg_map_t    peek_si; */
+    vp9e_encode,        /* vpx_codec_encode_fn_t      encode; */
+    vp9e_get_cxdata,    /* vpx_codec_get_cx_data_fn_t   frame_get; */
+    vp9e_set_config,
+    NOT_IMPLEMENTED,
+    vp9e_get_preview,
+  } /* encoder functions */
+};
+#endif
diff --git a/libvpx/vp9/vp9_dx_iface.c b/libvpx/vp9/vp9_dx_iface.c
new file mode 100644
index 0000000..ea6946b
--- /dev/null
+++ b/libvpx/vp9/vp9_dx_iface.c
@@ -0,0 +1,750 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+#include <string.h>
+#include "vpx/vpx_decoder.h"
+#include "vpx/vp8dx.h"
+#include "vpx/internal/vpx_codec_internal.h"
+#include "vpx_version.h"
+#include "decoder/vp9_onyxd.h"
+#include "decoder/vp9_onyxd_int.h"
+#include "vp9/vp9_iface_common.h"
+
+#define VP8_CAP_POSTPROC (CONFIG_POSTPROC ? VPX_CODEC_CAP_POSTPROC : 0)
+typedef vpx_codec_stream_info_t  vp8_stream_info_t;
+
+/* Structures for handling memory allocations */
+typedef enum {
+  VP8_SEG_ALG_PRIV     = 256,
+  VP8_SEG_MAX
+} mem_seg_id_t;
+#define NELEMENTS(x) ((int)(sizeof(x)/sizeof(x[0])))
+
+static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t);
+
+typedef struct {
+  unsigned int   id;
+  unsigned long  sz;
+  unsigned int   align;
+  unsigned int   flags;
+  unsigned long(*calc_sz)(const vpx_codec_dec_cfg_t *, vpx_codec_flags_t);
+} mem_req_t;
+
+static const mem_req_t vp8_mem_req_segs[] = {
+  {VP8_SEG_ALG_PRIV,    0, 8, VPX_CODEC_MEM_ZERO, vp8_priv_sz},
+  {VP8_SEG_MAX, 0, 0, 0, NULL}
+};
+
+struct vpx_codec_alg_priv {
+  vpx_codec_priv_t        base;
+  vpx_codec_mmap_t        mmaps[NELEMENTS(vp8_mem_req_segs) - 1];
+  vpx_codec_dec_cfg_t     cfg;
+  vp8_stream_info_t       si;
+  int                     defer_alloc;
+  int                     decoder_init;
+  VP9D_PTR                pbi;
+  int                     postproc_cfg_set;
+  vp8_postproc_cfg_t      postproc_cfg;
+#if CONFIG_POSTPROC_VISUALIZER
+  unsigned int            dbg_postproc_flag;
+  int                     dbg_color_ref_frame_flag;
+  int                     dbg_color_mb_modes_flag;
+  int                     dbg_color_b_modes_flag;
+  int                     dbg_display_mv_flag;
+#endif
+  vpx_image_t             img;
+  int                     img_setup;
+  int                     img_avail;
+  int                     invert_tile_order;
+};
+
+static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si,
+                                 vpx_codec_flags_t flags) {
+  /* Although this declaration is constant, we can't use it in the requested
+   * segments list because we want to define the requested segments list
+   * before defining the private type (so that the number of memory maps is
+   * known)
+   */
+  (void)si;
+  return sizeof(vpx_codec_alg_priv_t);
+}
+
+
+static void vp8_mmap_dtor(vpx_codec_mmap_t *mmap) {
+  free(mmap->priv);
+}
+
+static vpx_codec_err_t vp8_mmap_alloc(vpx_codec_mmap_t *mmap) {
+  vpx_codec_err_t  res;
+  unsigned int   align;
+
+  align = mmap->align ? mmap->align - 1 : 0;
+
+  if (mmap->flags & VPX_CODEC_MEM_ZERO)
+    mmap->priv = calloc(1, mmap->sz + align);
+  else
+    mmap->priv = malloc(mmap->sz + align);
+
+  res = (mmap->priv) ? VPX_CODEC_OK : VPX_CODEC_MEM_ERROR;
+  mmap->base = (void *)((((uintptr_t)mmap->priv) + align) & ~(uintptr_t)align);
+  mmap->dtor = vp8_mmap_dtor;
+  return res;
+}
+
+static vpx_codec_err_t vp8_validate_mmaps(const vp8_stream_info_t *si,
+                                          const vpx_codec_mmap_t *mmaps,
+                                          vpx_codec_flags_t init_flags) {
+  int i;
+  vpx_codec_err_t res = VPX_CODEC_OK;
+
+  for (i = 0; i < NELEMENTS(vp8_mem_req_segs) - 1; i++) {
+    /* Ensure the segment has been allocated */
+    if (!mmaps[i].base) {
+      res = VPX_CODEC_MEM_ERROR;
+      break;
+    }
+
+    /* Verify variable size segment is big enough for the current si. */
+    if (vp8_mem_req_segs[i].calc_sz) {
+      vpx_codec_dec_cfg_t cfg;
+
+      cfg.w = si->w;
+      cfg.h = si->h;
+
+      if (mmaps[i].sz < vp8_mem_req_segs[i].calc_sz(&cfg, init_flags)) {
+        res = VPX_CODEC_MEM_ERROR;
+        break;
+      }
+    }
+  }
+
+  return res;
+}
+
+static void vp8_init_ctx(vpx_codec_ctx_t *ctx, const vpx_codec_mmap_t *mmap) {
+  int i;
+
+  ctx->priv = mmap->base;
+  ctx->priv->sz = sizeof(*ctx->priv);
+  ctx->priv->iface = ctx->iface;
+  ctx->priv->alg_priv = mmap->base;
+
+  for (i = 0; i < NELEMENTS(ctx->priv->alg_priv->mmaps); i++)
+    ctx->priv->alg_priv->mmaps[i].id = vp8_mem_req_segs[i].id;
+
+  ctx->priv->alg_priv->mmaps[0] = *mmap;
+  ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si);
+  ctx->priv->init_flags = ctx->init_flags;
+
+  if (ctx->config.dec) {
+    /* Update the reference to the config structure to an internal copy. */
+    ctx->priv->alg_priv->cfg = *ctx->config.dec;
+    ctx->config.dec = &ctx->priv->alg_priv->cfg;
+  }
+}
+
+static void *mmap_lkup(vpx_codec_alg_priv_t *ctx, unsigned int id) {
+  int i;
+
+  for (i = 0; i < NELEMENTS(ctx->mmaps); i++)
+    if (ctx->mmaps[i].id == id)
+      return ctx->mmaps[i].base;
+
+  return NULL;
+}
+static void vp8_finalize_mmaps(vpx_codec_alg_priv_t *ctx) {
+  /* nothing to clean up */
+}
+
+static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx,
+                                vpx_codec_priv_enc_mr_cfg_t *data) {
+  vpx_codec_err_t        res = VPX_CODEC_OK;
+
+  /* This function only allocates space for the vpx_codec_alg_priv_t
+   * structure. More memory may be required at the time the stream
+   * information becomes known.
+   */
+  if (!ctx->priv) {
+    vpx_codec_mmap_t mmap;
+
+    mmap.id = vp8_mem_req_segs[0].id;
+    mmap.sz = sizeof(vpx_codec_alg_priv_t);
+    mmap.align = vp8_mem_req_segs[0].align;
+    mmap.flags = vp8_mem_req_segs[0].flags;
+
+    res = vp8_mmap_alloc(&mmap);
+
+    if (!res) {
+      vp8_init_ctx(ctx, &mmap);
+
+      ctx->priv->alg_priv->defer_alloc = 1;
+      /*post processing level initialized to do nothing */
+    }
+  }
+
+  return res;
+}
+
+static vpx_codec_err_t vp8_destroy(vpx_codec_alg_priv_t *ctx) {
+  int i;
+
+  vp9_remove_decompressor(ctx->pbi);
+
+  for (i = NELEMENTS(ctx->mmaps) - 1; i >= 0; i--) {
+    if (ctx->mmaps[i].dtor)
+      ctx->mmaps[i].dtor(&ctx->mmaps[i]);
+  }
+
+  return VPX_CODEC_OK;
+}
+
+static vpx_codec_err_t vp8_peek_si(const uint8_t         *data,
+                                   unsigned int           data_sz,
+                                   vpx_codec_stream_info_t *si) {
+  vpx_codec_err_t res = VPX_CODEC_OK;
+
+  if (data + data_sz <= data)
+    res = VPX_CODEC_INVALID_PARAM;
+  else {
+    si->is_kf = 0;
+
+    if (data_sz >= 8 && (data[0] & 0xD8) == 0x80) { /* I-Frame */
+      const uint8_t *c = data + 1;
+      si->is_kf = 1;
+
+      if (c[0] != SYNC_CODE_0 || c[1] != SYNC_CODE_1 || c[2] != SYNC_CODE_2)
+        res = VPX_CODEC_UNSUP_BITSTREAM;
+
+      si->w = (c[3] << 8) | c[4];
+      si->h = (c[5] << 8) | c[6];
+
+      // printf("w=%d, h=%d\n", si->w, si->h);
+      if (!(si->h | si->w))
+        res = VPX_CODEC_UNSUP_BITSTREAM;
+    } else
+      res = VPX_CODEC_UNSUP_BITSTREAM;
+  }
+
+  return res;
+}
+
+static vpx_codec_err_t vp8_get_si(vpx_codec_alg_priv_t    *ctx,
+                                  vpx_codec_stream_info_t *si) {
+
+  unsigned int sz;
+
+  if (si->sz >= sizeof(vp8_stream_info_t))
+    sz = sizeof(vp8_stream_info_t);
+  else
+    sz = sizeof(vpx_codec_stream_info_t);
+
+  memcpy(si, &ctx->si, sz);
+  si->sz = sz;
+
+  return VPX_CODEC_OK;
+}
+
+
+static vpx_codec_err_t
+update_error_state(vpx_codec_alg_priv_t                 *ctx,
+                   const struct vpx_internal_error_info *error) {
+  vpx_codec_err_t res;
+
+  if ((res = error->error_code))
+    ctx->base.err_detail = error->has_detail
+                           ? error->detail
+                           : NULL;
+
+  return res;
+}
+
+static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t  *ctx,
+                                  const uint8_t        **data,
+                                  unsigned int           data_sz,
+                                  void                  *user_priv,
+                                  long                   deadline) {
+  vpx_codec_err_t res = VPX_CODEC_OK;
+
+  ctx->img_avail = 0;
+
+  /* Determine the stream parameters. Note that we rely on peek_si to
+   * validate that we have a buffer that does not wrap around the top
+   * of the heap.
+   */
+  if (!ctx->si.h)
+    res = ctx->base.iface->dec.peek_si(*data, data_sz, &ctx->si);
+
+
+  /* Perform deferred allocations, if required */
+  if (!res && ctx->defer_alloc) {
+    int i;
+
+    for (i = 1; !res && i < NELEMENTS(ctx->mmaps); i++) {
+      vpx_codec_dec_cfg_t cfg;
+
+      cfg.w = ctx->si.w;
+      cfg.h = ctx->si.h;
+      ctx->mmaps[i].id = vp8_mem_req_segs[i].id;
+      ctx->mmaps[i].sz = vp8_mem_req_segs[i].sz;
+      ctx->mmaps[i].align = vp8_mem_req_segs[i].align;
+      ctx->mmaps[i].flags = vp8_mem_req_segs[i].flags;
+
+      if (!ctx->mmaps[i].sz)
+        ctx->mmaps[i].sz = vp8_mem_req_segs[i].calc_sz(&cfg,
+                                                       ctx->base.init_flags);
+
+      res = vp8_mmap_alloc(&ctx->mmaps[i]);
+    }
+
+    if (!res)
+      vp8_finalize_mmaps(ctx);
+
+    ctx->defer_alloc = 0;
+  }
+
+  /* Initialize the decoder instance on the first frame*/
+  if (!res && !ctx->decoder_init) {
+    res = vp8_validate_mmaps(&ctx->si, ctx->mmaps, ctx->base.init_flags);
+
+    if (!res) {
+      VP9D_CONFIG oxcf;
+      VP9D_PTR optr;
+
+      vp9_initialize_dec();
+
+      oxcf.width = ctx->si.w;
+      oxcf.height = ctx->si.h;
+      oxcf.version = 9;
+      oxcf.postprocess = 0;
+      oxcf.max_threads = ctx->cfg.threads;
+      oxcf.inv_tile_order = ctx->invert_tile_order;
+      optr = vp9_create_decompressor(&oxcf);
+
+      /* If postprocessing was enabled by the application and a
+       * configuration has not been provided, default it.
+       */
+      if (!ctx->postproc_cfg_set
+          && (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) {
+        ctx->postproc_cfg.post_proc_flag =
+          VP8_DEBLOCK | VP8_DEMACROBLOCK;
+        ctx->postproc_cfg.deblocking_level = 4;
+        ctx->postproc_cfg.noise_level = 0;
+      }
+
+      if (!optr)
+        res = VPX_CODEC_ERROR;
+      else
+        ctx->pbi = optr;
+    }
+
+    ctx->decoder_init = 1;
+  }
+
+  if (!res && ctx->pbi) {
+    YV12_BUFFER_CONFIG sd;
+    int64_t time_stamp = 0, time_end_stamp = 0;
+    vp9_ppflags_t flags = {0};
+
+    if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC) {
+      flags.post_proc_flag = ctx->postproc_cfg.post_proc_flag
+#if CONFIG_POSTPROC_VISUALIZER
+
+                             | ((ctx->dbg_color_ref_frame_flag != 0) ? VP9D_DEBUG_CLR_FRM_REF_BLKS : 0)
+                             | ((ctx->dbg_color_mb_modes_flag != 0) ? VP9D_DEBUG_CLR_BLK_MODES : 0)
+                             | ((ctx->dbg_color_b_modes_flag != 0) ? VP9D_DEBUG_CLR_BLK_MODES : 0)
+                             | ((ctx->dbg_display_mv_flag != 0) ? VP9D_DEBUG_DRAW_MV : 0)
+#endif
+;
+      flags.deblocking_level      = ctx->postproc_cfg.deblocking_level;
+      flags.noise_level           = ctx->postproc_cfg.noise_level;
+#if CONFIG_POSTPROC_VISUALIZER
+      flags.display_ref_frame_flag = ctx->dbg_color_ref_frame_flag;
+      flags.display_mb_modes_flag = ctx->dbg_color_mb_modes_flag;
+      flags.display_b_modes_flag  = ctx->dbg_color_b_modes_flag;
+      flags.display_mv_flag       = ctx->dbg_display_mv_flag;
+#endif
+    }
+
+    if (vp9_receive_compressed_data(ctx->pbi, data_sz, data, deadline)) {
+      VP9D_COMP *pbi = (VP9D_COMP *)ctx->pbi;
+      res = update_error_state(ctx, &pbi->common.error);
+    }
+
+    if (!res && 0 == vp9_get_raw_frame(ctx->pbi, &sd, &time_stamp,
+                                       &time_end_stamp, &flags)) {
+      yuvconfig2image(&ctx->img, &sd, user_priv);
+      ctx->img_avail = 1;
+    }
+  }
+
+  return res;
+}
+
+static void parse_superframe_index(const uint8_t *data,
+                                   size_t         data_sz,
+                                   uint32_t       sizes[8],
+                                   int           *count) {
+  uint8_t marker;
+
+  assert(data_sz);
+  marker = data[data_sz - 1];
+  *count = 0;
+
+  if ((marker & 0xe0) == 0xc0) {
+    const uint32_t frames = (marker & 0x7) + 1;
+    const uint32_t mag = ((marker >> 3) & 0x3) + 1;
+    const size_t index_sz = 2 + mag * frames;
+
+    if (data_sz >= index_sz && data[data_sz - index_sz] == marker) {
+      // found a valid superframe index
+      uint32_t i, j;
+      const uint8_t *x = data + data_sz - index_sz + 1;
+
+      for (i = 0; i < frames; i++) {
+        uint32_t this_sz = 0;
+
+        for (j = 0; j < mag; j++)
+          this_sz |= (*x++) << (j * 8);
+        sizes[i] = this_sz;
+      }
+
+      *count = frames;
+    }
+  }
+}
+
+static vpx_codec_err_t vp9_decode(vpx_codec_alg_priv_t  *ctx,
+                                  const uint8_t         *data,
+                                  unsigned int           data_sz,
+                                  void                  *user_priv,
+                                  long                   deadline) {
+  const uint8_t *data_start = data;
+  const uint8_t *data_end = data + data_sz;
+  vpx_codec_err_t res = 0;
+  uint32_t sizes[8];
+  int frames_this_pts, frame_count = 0;
+
+  parse_superframe_index(data, data_sz, sizes, &frames_this_pts);
+
+  do {
+    // Skip over the superframe index, if present
+    if (data_sz && (*data_start & 0xe0) == 0xc0) {
+      const uint8_t marker = *data_start;
+      const uint32_t frames = (marker & 0x7) + 1;
+      const uint32_t mag = ((marker >> 3) & 0x3) + 1;
+      const uint32_t index_sz = 2 + mag * frames;
+
+      if (data_sz >= index_sz && data_start[index_sz - 1] == marker) {
+        data_start += index_sz;
+        data_sz -= index_sz;
+        if (data_start < data_end)
+          continue;
+        else
+          break;
+      }
+    }
+
+    // Use the correct size for this frame, if an index is present.
+    if (frames_this_pts) {
+      uint32_t this_sz = sizes[frame_count];
+
+      if (data_sz < this_sz) {
+        ctx->base.err_detail = "Invalid frame size in index";
+        return VPX_CODEC_CORRUPT_FRAME;
+      }
+
+      data_sz = this_sz;
+      frame_count++;
+    }
+
+    res = decode_one(ctx, &data_start, data_sz, user_priv, deadline);
+    assert(data_start >= data);
+    assert(data_start <= data_end);
+
+    /* Early exit if there was a decode error */
+    if (res)
+      break;
+
+    /* Account for suboptimal termination by the encoder. */
+    while (data_start < data_end && *data_start == 0)
+      data_start++;
+
+    data_sz = data_end - data_start;
+  } while (data_start < data_end);
+  return res;
+}
+
+static vpx_image_t *vp8_get_frame(vpx_codec_alg_priv_t  *ctx,
+                                  vpx_codec_iter_t      *iter) {
+  vpx_image_t *img = NULL;
+
+  if (ctx->img_avail) {
+    /* iter acts as a flip flop, so an image is only returned on the first
+     * call to get_frame.
+     */
+    if (!(*iter)) {
+      img = &ctx->img;
+      *iter = img;
+    }
+  }
+  ctx->img_avail = 0;
+
+  return img;
+}
+
+
+static
+vpx_codec_err_t vp8_xma_get_mmap(const vpx_codec_ctx_t      *ctx,
+                                 vpx_codec_mmap_t           *mmap,
+                                 vpx_codec_iter_t           *iter) {
+  vpx_codec_err_t     res;
+  const mem_req_t  *seg_iter = *iter;
+
+  /* Get address of next segment request */
+  do {
+    if (!seg_iter)
+      seg_iter = vp8_mem_req_segs;
+    else if (seg_iter->id != VP8_SEG_MAX)
+      seg_iter++;
+
+    *iter = (vpx_codec_iter_t)seg_iter;
+
+    if (seg_iter->id != VP8_SEG_MAX) {
+      mmap->id = seg_iter->id;
+      mmap->sz = seg_iter->sz;
+      mmap->align = seg_iter->align;
+      mmap->flags = seg_iter->flags;
+
+      if (!seg_iter->sz)
+        mmap->sz = seg_iter->calc_sz(ctx->config.dec, ctx->init_flags);
+
+      res = VPX_CODEC_OK;
+    } else
+      res = VPX_CODEC_LIST_END;
+  } while (!mmap->sz && res != VPX_CODEC_LIST_END);
+
+  return res;
+}
+
+static vpx_codec_err_t vp8_xma_set_mmap(vpx_codec_ctx_t         *ctx,
+                                        const vpx_codec_mmap_t  *mmap) {
+  vpx_codec_err_t res = VPX_CODEC_MEM_ERROR;
+  int i, done;
+
+  if (!ctx->priv) {
+    if (mmap->id == VP8_SEG_ALG_PRIV) {
+      if (!ctx->priv) {
+        vp8_init_ctx(ctx, mmap);
+        res = VPX_CODEC_OK;
+      }
+    }
+  }
+
+  done = 1;
+
+  if (!res && ctx->priv->alg_priv) {
+    for (i = 0; i < NELEMENTS(ctx->priv->alg_priv->mmaps); i++) {
+      if (ctx->priv->alg_priv->mmaps[i].id == mmap->id)
+        if (!ctx->priv->alg_priv->mmaps[i].base) {
+          ctx->priv->alg_priv->mmaps[i] = *mmap;
+          res = VPX_CODEC_OK;
+        }
+
+      done &= (ctx->priv->alg_priv->mmaps[i].base != NULL);
+    }
+  }
+
+  if (done && !res) {
+    vp8_finalize_mmaps(ctx->priv->alg_priv);
+    res = ctx->iface->init(ctx, NULL);
+  }
+
+  return res;
+}
+
+
+static vpx_codec_err_t vp9_set_reference(vpx_codec_alg_priv_t *ctx,
+                                         int ctr_id,
+                                         va_list args) {
+
+  vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
+
+  if (data) {
+    vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
+    YV12_BUFFER_CONFIG sd;
+
+    image2yuvconfig(&frame->img, &sd);
+
+    return vp9_set_reference_dec(ctx->pbi,
+                                 (VP9_REFFRAME)frame->frame_type, &sd);
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+
+}
+
+static vpx_codec_err_t vp9_copy_reference(vpx_codec_alg_priv_t *ctx,
+                                          int ctr_id,
+                                          va_list args) {
+
+  vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
+
+  if (data) {
+    vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
+    YV12_BUFFER_CONFIG sd;
+
+    image2yuvconfig(&frame->img, &sd);
+
+    return vp9_copy_reference_dec(ctx->pbi,
+                                  (VP9_REFFRAME)frame->frame_type, &sd);
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+
+}
+
+static vpx_codec_err_t get_reference(vpx_codec_alg_priv_t *ctx,
+                                     int ctr_id,
+                                     va_list args) {
+  vp9_ref_frame_t *data = va_arg(args, vp9_ref_frame_t *);
+
+  if (data) {
+    YV12_BUFFER_CONFIG* fb;
+
+    vp9_get_reference_dec(ctx->pbi, data->idx, &fb);
+    yuvconfig2image(&data->img, fb, NULL);
+    return VPX_CODEC_OK;
+  } else {
+    return VPX_CODEC_INVALID_PARAM;
+  }
+}
+
+static vpx_codec_err_t vp8_set_postproc(vpx_codec_alg_priv_t *ctx,
+                                        int ctr_id,
+                                        va_list args) {
+#if CONFIG_POSTPROC
+  vp8_postproc_cfg_t *data = va_arg(args, vp8_postproc_cfg_t *);
+
+  if (data) {
+    ctx->postproc_cfg_set = 1;
+    ctx->postproc_cfg = *((vp8_postproc_cfg_t *)data);
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+
+#else
+  return VPX_CODEC_INCAPABLE;
+#endif
+}
+
+static vpx_codec_err_t vp8_set_dbg_options(vpx_codec_alg_priv_t *ctx,
+                                           int ctrl_id,
+                                           va_list args) {
+#if CONFIG_POSTPROC_VISUALIZER && CONFIG_POSTPROC
+  int data = va_arg(args, int);
+
+#define MAP(id, var) case id: var = data; break;
+
+  switch (ctrl_id) {
+      MAP(VP8_SET_DBG_COLOR_REF_FRAME,   ctx->dbg_color_ref_frame_flag);
+      MAP(VP8_SET_DBG_COLOR_MB_MODES,    ctx->dbg_color_mb_modes_flag);
+      MAP(VP8_SET_DBG_COLOR_B_MODES,     ctx->dbg_color_b_modes_flag);
+      MAP(VP8_SET_DBG_DISPLAY_MV,        ctx->dbg_display_mv_flag);
+  }
+
+  return VPX_CODEC_OK;
+#else
+  return VPX_CODEC_INCAPABLE;
+#endif
+}
+
+static vpx_codec_err_t vp8_get_last_ref_updates(vpx_codec_alg_priv_t *ctx,
+                                                int ctrl_id,
+                                                va_list args) {
+  int *update_info = va_arg(args, int *);
+  VP9D_COMP *pbi = (VP9D_COMP *)ctx->pbi;
+
+  if (update_info) {
+    *update_info = pbi->refresh_frame_flags;
+
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+}
+
+
+static vpx_codec_err_t vp8_get_frame_corrupted(vpx_codec_alg_priv_t *ctx,
+                                               int ctrl_id,
+                                               va_list args) {
+
+  int *corrupted = va_arg(args, int *);
+
+  if (corrupted) {
+    VP9D_COMP *pbi = (VP9D_COMP *)ctx->pbi;
+    *corrupted = pbi->common.frame_to_show->corrupted;
+
+    return VPX_CODEC_OK;
+  } else
+    return VPX_CODEC_INVALID_PARAM;
+
+}
+
+static vpx_codec_err_t set_invert_tile_order(vpx_codec_alg_priv_t *ctx,
+                                             int ctr_id,
+                                             va_list args) {
+  ctx->invert_tile_order = va_arg(args, int);
+  return VPX_CODEC_OK;
+}
+
+static vpx_codec_ctrl_fn_map_t ctf_maps[] = {
+  {VP8_SET_REFERENCE,             vp9_set_reference},
+  {VP8_COPY_REFERENCE,            vp9_copy_reference},
+  {VP8_SET_POSTPROC,              vp8_set_postproc},
+  {VP8_SET_DBG_COLOR_REF_FRAME,   vp8_set_dbg_options},
+  {VP8_SET_DBG_COLOR_MB_MODES,    vp8_set_dbg_options},
+  {VP8_SET_DBG_COLOR_B_MODES,     vp8_set_dbg_options},
+  {VP8_SET_DBG_DISPLAY_MV,        vp8_set_dbg_options},
+  {VP8D_GET_LAST_REF_UPDATES,     vp8_get_last_ref_updates},
+  {VP8D_GET_FRAME_CORRUPTED,      vp8_get_frame_corrupted},
+  {VP9_GET_REFERENCE,             get_reference},
+  {VP9_INVERT_TILE_DECODE_ORDER,  set_invert_tile_order},
+  { -1, NULL},
+};
+
+
+#ifndef VERSION_STRING
+#define VERSION_STRING
+#endif
+CODEC_INTERFACE(vpx_codec_vp9_dx) = {
+  "WebM Project VP9 Decoder" VERSION_STRING,
+  VPX_CODEC_INTERNAL_ABI_VERSION,
+  VPX_CODEC_CAP_DECODER | VP8_CAP_POSTPROC,
+  /* vpx_codec_caps_t          caps; */
+  vp8_init,         /* vpx_codec_init_fn_t       init; */
+  vp8_destroy,      /* vpx_codec_destroy_fn_t    destroy; */
+  ctf_maps,         /* vpx_codec_ctrl_fn_map_t  *ctrl_maps; */
+  vp8_xma_get_mmap, /* vpx_codec_get_mmap_fn_t   get_mmap; */
+  vp8_xma_set_mmap, /* vpx_codec_set_mmap_fn_t   set_mmap; */
+  {
+    vp8_peek_si,      /* vpx_codec_peek_si_fn_t    peek_si; */
+    vp8_get_si,       /* vpx_codec_get_si_fn_t     get_si; */
+    vp9_decode,       /* vpx_codec_decode_fn_t     decode; */
+    vp8_get_frame,    /* vpx_codec_frame_get_fn_t  frame_get; */
+  },
+  {
+    /* encoder functions */
+    NOT_IMPLEMENTED,
+    NOT_IMPLEMENTED,
+    NOT_IMPLEMENTED,
+    NOT_IMPLEMENTED,
+    NOT_IMPLEMENTED,
+    NOT_IMPLEMENTED
+  }
+};
diff --git a/libvpx/vp9/vp9_iface_common.h b/libvpx/vp9/vp9_iface_common.h
new file mode 100644
index 0000000..dc41d77
--- /dev/null
+++ b/libvpx/vp9/vp9_iface_common.h
@@ -0,0 +1,89 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef VP9_VP9_IFACE_COMMON_H_
+#define VP9_VP9_IFACE_COMMON_H_
+
+static void yuvconfig2image(vpx_image_t *img, const YV12_BUFFER_CONFIG  *yv12,
+                            void *user_priv) {
+  /** vpx_img_wrap() doesn't allow specifying independent strides for
+    * the Y, U, and V planes, nor other alignment adjustments that
+    * might be representable by a YV12_BUFFER_CONFIG, so we just
+    * initialize all the fields.*/
+  int bps = 12;
+  if (yv12->uv_height == yv12->y_height) {
+    if (yv12->uv_width == yv12->y_width) {
+      img->fmt = VPX_IMG_FMT_I444;
+      bps = 24;
+    } else {
+      img->fmt = VPX_IMG_FMT_I422;
+      bps = 16;
+    }
+  } else {
+    img->fmt = VPX_IMG_FMT_I420;
+  }
+  img->w = yv12->y_stride;
+  img->h = multiple8(yv12->y_height + 2 * VP9BORDERINPIXELS);
+  img->d_w = yv12->y_crop_width;
+  img->d_h = yv12->y_crop_height;
+  img->x_chroma_shift = yv12->uv_width < yv12->y_width;
+  img->y_chroma_shift = yv12->uv_height < yv12->y_height;
+  img->planes[VPX_PLANE_Y] = yv12->y_buffer;
+  img->planes[VPX_PLANE_U] = yv12->u_buffer;
+  img->planes[VPX_PLANE_V] = yv12->v_buffer;
+  img->planes[VPX_PLANE_ALPHA] = yv12->alpha_buffer;
+  img->stride[VPX_PLANE_Y] = yv12->y_stride;
+  img->stride[VPX_PLANE_U] = yv12->uv_stride;
+  img->stride[VPX_PLANE_V] = yv12->uv_stride;
+  img->stride[VPX_PLANE_ALPHA] = yv12->alpha_stride;
+  img->bps = bps;
+  img->user_priv = user_priv;
+  img->img_data = yv12->buffer_alloc;
+  img->img_data_owner = 0;
+  img->self_allocd = 0;
+}
+
+static vpx_codec_err_t image2yuvconfig(const vpx_image_t *img,
+                                       YV12_BUFFER_CONFIG *yv12) {
+  yv12->y_buffer = img->planes[VPX_PLANE_Y];
+  yv12->u_buffer = img->planes[VPX_PLANE_U];
+  yv12->v_buffer = img->planes[VPX_PLANE_V];
+  yv12->alpha_buffer = img->planes[VPX_PLANE_ALPHA];
+
+  yv12->y_crop_width  = img->d_w;
+  yv12->y_crop_height = img->d_h;
+  yv12->y_width  = img->d_w;
+  yv12->y_height = img->d_h;
+
+  yv12->uv_width = img->x_chroma_shift == 1 ? (1 + yv12->y_width) / 2
+                                            : yv12->y_width;
+  yv12->uv_height = img->y_chroma_shift == 1 ? (1 + yv12->y_height) / 2
+                                             : yv12->y_height;
+
+  yv12->alpha_width = yv12->alpha_buffer ? img->d_w : 0;
+  yv12->alpha_height = yv12->alpha_buffer ? img->d_h : 0;
+
+  yv12->y_stride = img->stride[VPX_PLANE_Y];
+  yv12->uv_stride = img->stride[VPX_PLANE_U];
+  yv12->alpha_stride = yv12->alpha_buffer ? img->stride[VPX_PLANE_ALPHA] : 0;
+
+  yv12->border  = (img->stride[VPX_PLANE_Y] - img->w) / 2;
+  yv12->clrtype = REG_YUV;
+
+#if CONFIG_ALPHA
+  // For development purposes, force alpha to hold the same data a Y for now.
+  yv12->alpha_buffer = yv12->y_buffer;
+  yv12->alpha_width = yv12->y_width;
+  yv12->alpha_height = yv12->y_height;
+  yv12->alpha_stride = yv12->y_stride;
+#endif
+  return VPX_CODEC_OK;
+}
+
+#endif  // VP9_VP9_IFACE_COMMON_H_
diff --git a/libvpx/vp9/vp9cx.mk b/libvpx/vp9/vp9cx.mk
new file mode 100644
index 0000000..4bed6c0
--- /dev/null
+++ b/libvpx/vp9/vp9cx.mk
@@ -0,0 +1,101 @@
+##
+##  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+VP9_CX_EXPORTS += exports_enc
+
+VP9_CX_SRCS-yes += $(VP9_COMMON_SRCS-yes)
+VP9_CX_SRCS-no  += $(VP9_COMMON_SRCS-no)
+VP9_CX_SRCS_REMOVE-yes += $(VP9_COMMON_SRCS_REMOVE-yes)
+VP9_CX_SRCS_REMOVE-no  += $(VP9_COMMON_SRCS_REMOVE-no)
+
+VP9_CX_SRCS-yes += vp9_cx_iface.c
+
+VP9_CX_SRCS-yes += encoder/vp9_bitstream.c
+VP9_CX_SRCS-yes += encoder/vp9_boolhuff.c
+VP9_CX_SRCS-yes += encoder/vp9_dct.c
+VP9_CX_SRCS-yes += encoder/vp9_encodeframe.c
+VP9_CX_SRCS-yes += encoder/vp9_encodeframe.h
+VP9_CX_SRCS-yes += encoder/vp9_encodeintra.c
+VP9_CX_SRCS-yes += encoder/vp9_encodemb.c
+VP9_CX_SRCS-yes += encoder/vp9_encodemv.c
+VP9_CX_SRCS-yes += encoder/vp9_firstpass.c
+VP9_CX_SRCS-yes += encoder/vp9_block.h
+VP9_CX_SRCS-yes += encoder/vp9_boolhuff.h
+VP9_CX_SRCS-yes += encoder/vp9_write_bit_buffer.h
+VP9_CX_SRCS-yes += encoder/vp9_bitstream.h
+VP9_CX_SRCS-yes += encoder/vp9_encodeintra.h
+VP9_CX_SRCS-yes += encoder/vp9_encodemb.h
+VP9_CX_SRCS-yes += encoder/vp9_encodemv.h
+VP9_CX_SRCS-yes += encoder/vp9_firstpass.h
+VP9_CX_SRCS-yes += encoder/vp9_lookahead.c
+VP9_CX_SRCS-yes += encoder/vp9_lookahead.h
+VP9_CX_SRCS-yes += encoder/vp9_mcomp.h
+VP9_CX_SRCS-yes += encoder/vp9_modecosts.h
+VP9_CX_SRCS-yes += encoder/vp9_onyx_int.h
+VP9_CX_SRCS-yes += encoder/vp9_psnr.h
+VP9_CX_SRCS-yes += encoder/vp9_quantize.h
+VP9_CX_SRCS-yes += encoder/vp9_ratectrl.h
+VP9_CX_SRCS-yes += encoder/vp9_rdopt.h
+VP9_CX_SRCS-yes += encoder/vp9_tokenize.h
+VP9_CX_SRCS-yes += encoder/vp9_treewriter.h
+VP9_CX_SRCS-yes += encoder/vp9_variance.h
+VP9_CX_SRCS-yes += encoder/vp9_mcomp.c
+VP9_CX_SRCS-yes += encoder/vp9_modecosts.c
+VP9_CX_SRCS-yes += encoder/vp9_onyx_if.c
+VP9_CX_SRCS-yes += encoder/vp9_picklpf.c
+VP9_CX_SRCS-yes += encoder/vp9_picklpf.h
+VP9_CX_SRCS-yes += encoder/vp9_psnr.c
+VP9_CX_SRCS-yes += encoder/vp9_quantize.c
+VP9_CX_SRCS-yes += encoder/vp9_ratectrl.c
+VP9_CX_SRCS-yes += encoder/vp9_rdopt.c
+VP9_CX_SRCS-yes += encoder/vp9_sad_c.c
+VP9_CX_SRCS-yes += encoder/vp9_segmentation.c
+VP9_CX_SRCS-yes += encoder/vp9_segmentation.h
+VP9_CX_SRCS-$(CONFIG_INTERNAL_STATS) += encoder/vp9_ssim.c
+VP9_CX_SRCS-yes += encoder/vp9_tokenize.c
+VP9_CX_SRCS-yes += encoder/vp9_treewriter.c
+VP9_CX_SRCS-yes += encoder/vp9_variance_c.c
+ifeq ($(CONFIG_POSTPROC),yes)
+VP9_CX_SRCS-$(CONFIG_INTERNAL_STATS) += common/vp9_postproc.h
+VP9_CX_SRCS-$(CONFIG_INTERNAL_STATS) += common/vp9_postproc.c
+endif
+VP9_CX_SRCS-yes += encoder/vp9_temporal_filter.c
+VP9_CX_SRCS-yes += encoder/vp9_temporal_filter.h
+VP9_CX_SRCS-yes += encoder/vp9_mbgraph.c
+VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h
+
+
+VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_mcomp_x86.h
+VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_x86_csystemdependent.c
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_mmx.c
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_impl_mmx.asm
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_sad_mmx.asm
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_dct_mmx.asm
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_dct_mmx.h
+VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_subtract_mmx.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_sse2.c
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_impl_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad4d_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_fwalsh_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_subtract_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_subpel_variance_impl_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_temporal_filter_apply_sse2.asm
+VP9_CX_SRCS-$(HAVE_SSE3) += encoder/x86/vp9_sad_sse3.asm
+VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_sad_ssse3.asm
+VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_variance_ssse3.c
+VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_variance_impl_ssse3.asm
+VP9_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/vp9_sad_sse4.asm
+VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_encodeopt.asm
+VP9_CX_SRCS-$(ARCH_X86_64) += encoder/x86/vp9_ssim_opt.asm
+
+VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct_sse2.c
+
+VP9_CX_SRCS-yes := $(filter-out $(VP9_CX_SRCS_REMOVE-yes),$(VP9_CX_SRCS-yes))
diff --git a/libvpx/vp9/vp9dx.mk b/libvpx/vp9/vp9dx.mk
new file mode 100644
index 0000000..7ae3219
--- /dev/null
+++ b/libvpx/vp9/vp9dx.mk
@@ -0,0 +1,42 @@
+##
+##  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE file in the root of the source
+##  tree. An additional intellectual property rights grant can be found
+##  in the file PATENTS.  All contributing project authors may
+##  be found in the AUTHORS file in the root of the source tree.
+##
+
+VP9_DX_EXPORTS += exports_dec
+
+VP9_DX_SRCS-yes += $(VP9_COMMON_SRCS-yes)
+VP9_DX_SRCS-no  += $(VP9_COMMON_SRCS-no)
+VP9_DX_SRCS_REMOVE-yes += $(VP9_COMMON_SRCS_REMOVE-yes)
+VP9_DX_SRCS_REMOVE-no  += $(VP9_COMMON_SRCS_REMOVE-no)
+
+VP9_DX_SRCS-yes += vp9_dx_iface.c
+
+VP9_DX_SRCS-yes += decoder/vp9_asm_dec_offsets.c
+VP9_DX_SRCS-yes += decoder/vp9_dboolhuff.c
+VP9_DX_SRCS-yes += decoder/vp9_decodemv.c
+VP9_DX_SRCS-yes += decoder/vp9_decodframe.c
+VP9_DX_SRCS-yes += decoder/vp9_decodframe.h
+VP9_DX_SRCS-yes += decoder/vp9_detokenize.c
+VP9_DX_SRCS-yes += decoder/vp9_dboolhuff.h
+VP9_DX_SRCS-yes += decoder/vp9_read_bit_buffer.h
+VP9_DX_SRCS-yes += decoder/vp9_decodemv.h
+VP9_DX_SRCS-yes += decoder/vp9_detokenize.h
+VP9_DX_SRCS-yes += decoder/vp9_onyxd.h
+VP9_DX_SRCS-yes += decoder/vp9_onyxd_int.h
+VP9_DX_SRCS-yes += decoder/vp9_treereader.h
+VP9_DX_SRCS-yes += decoder/vp9_onyxd_if.c
+VP9_DX_SRCS-yes += decoder/vp9_idct_blk.c
+VP9_DX_SRCS-yes += decoder/vp9_idct_blk.h
+
+VP9_DX_SRCS-yes := $(filter-out $(VP9_DX_SRCS_REMOVE-yes),$(VP9_DX_SRCS-yes))
+
+VP9_DX_SRCS-$(HAVE_SSE2) += decoder/x86/vp9_dequantize_sse2.c
+
+$(eval $(call asm_offsets_template,\
+         vp9_asm_dec_offsets.asm, $(VP9_PREFIX)decoder/vp9_asm_dec_offsets.c))
diff --git a/libvpx/vpx/internal/vpx_codec_internal.h b/libvpx/vpx/internal/vpx_codec_internal.h
index 4474331..d7bcd46 100644
--- a/libvpx/vpx/internal/vpx_codec_internal.h
+++ b/libvpx/vpx/internal/vpx_codec_internal.h
@@ -75,7 +75,7 @@
  *     Memory operation failed.
  */
 typedef vpx_codec_err_t (*vpx_codec_init_fn_t)(vpx_codec_ctx_t *ctx,
-                                             vpx_codec_priv_enc_mr_cfg_t *data);
+                                               vpx_codec_priv_enc_mr_cfg_t *data);
 
 /*!\brief destroy function pointer prototype
  *
@@ -109,8 +109,8 @@
  *     Bitstream is parsable and stream information updated
  */
 typedef vpx_codec_err_t (*vpx_codec_peek_si_fn_t)(const uint8_t         *data,
-        unsigned int           data_sz,
-        vpx_codec_stream_info_t *si);
+                                                  unsigned int           data_sz,
+                                                  vpx_codec_stream_info_t *si);
 
 /*!\brief Return information about the current stream.
  *
@@ -126,7 +126,7 @@
  *     Bitstream is parsable and stream information updated
  */
 typedef vpx_codec_err_t (*vpx_codec_get_si_fn_t)(vpx_codec_alg_priv_t    *ctx,
-        vpx_codec_stream_info_t *si);
+                                                 vpx_codec_stream_info_t *si);
 
 /*!\brief control function pointer prototype
  *
@@ -151,8 +151,8 @@
  *     The internal state data was deserialized.
  */
 typedef vpx_codec_err_t (*vpx_codec_control_fn_t)(vpx_codec_alg_priv_t  *ctx,
-        int                  ctrl_id,
-        va_list              ap);
+                                                  int                  ctrl_id,
+                                                  va_list              ap);
 
 /*!\brief control function pointer mapping
  *
@@ -165,10 +165,9 @@
  * mapping. This implies that ctrl_id values chosen by the algorithm
  * \ref MUST be non-zero.
  */
-typedef const struct vpx_codec_ctrl_fn_map
-{
-    int                    ctrl_id;
-    vpx_codec_control_fn_t   fn;
+typedef const struct vpx_codec_ctrl_fn_map {
+  int                    ctrl_id;
+  vpx_codec_control_fn_t   fn;
 } vpx_codec_ctrl_fn_map_t;
 
 /*!\brief decode data function pointer prototype
@@ -192,10 +191,10 @@
  *         for recoverability capabilities.
  */
 typedef vpx_codec_err_t (*vpx_codec_decode_fn_t)(vpx_codec_alg_priv_t  *ctx,
-        const uint8_t         *data,
-        unsigned int     data_sz,
-        void        *user_priv,
-        long         deadline);
+                                                 const uint8_t         *data,
+                                                 unsigned int     data_sz,
+                                                 void        *user_priv,
+                                                 long         deadline);
 
 /*!\brief Decoded frames iterator
  *
@@ -212,8 +211,8 @@
  * \return Returns a pointer to an image, if one is ready for display. Frames
  *         produced will always be in PTS (presentation time stamp) order.
  */
-typedef vpx_image_t*(*vpx_codec_get_frame_fn_t)(vpx_codec_alg_priv_t *ctx,
-        vpx_codec_iter_t     *iter);
+typedef vpx_image_t *(*vpx_codec_get_frame_fn_t)(vpx_codec_alg_priv_t *ctx,
+                                                 vpx_codec_iter_t     *iter);
 
 
 /*\brief eXternal Memory Allocation memory map get iterator
@@ -228,8 +227,8 @@
  *         indicate end-of-list.
  */
 typedef vpx_codec_err_t (*vpx_codec_get_mmap_fn_t)(const vpx_codec_ctx_t      *ctx,
-        vpx_codec_mmap_t           *mmap,
-        vpx_codec_iter_t           *iter);
+                                                   vpx_codec_mmap_t           *mmap,
+                                                   vpx_codec_iter_t           *iter);
 
 
 /*\brief eXternal Memory Allocation memory map set iterator
@@ -245,17 +244,17 @@
  *     The memory map was rejected.
  */
 typedef vpx_codec_err_t (*vpx_codec_set_mmap_fn_t)(vpx_codec_ctx_t         *ctx,
-        const vpx_codec_mmap_t  *mmap);
+                                                   const vpx_codec_mmap_t  *mmap);
 
 
 typedef vpx_codec_err_t (*vpx_codec_encode_fn_t)(vpx_codec_alg_priv_t  *ctx,
-        const vpx_image_t     *img,
-        vpx_codec_pts_t        pts,
-        unsigned long          duration,
-        vpx_enc_frame_flags_t  flags,
-        unsigned long          deadline);
-typedef const vpx_codec_cx_pkt_t*(*vpx_codec_get_cx_data_fn_t)(vpx_codec_alg_priv_t *ctx,
-        vpx_codec_iter_t     *iter);
+                                                 const vpx_image_t     *img,
+                                                 vpx_codec_pts_t        pts,
+                                                 unsigned long          duration,
+                                                 vpx_enc_frame_flags_t  flags,
+                                                 unsigned long          deadline);
+typedef const vpx_codec_cx_pkt_t *(*vpx_codec_get_cx_data_fn_t)(vpx_codec_alg_priv_t *ctx,
+                                                                vpx_codec_iter_t     *iter);
 
 typedef vpx_codec_err_t
 (*vpx_codec_enc_config_set_fn_t)(vpx_codec_alg_priv_t       *ctx,
@@ -268,7 +267,7 @@
 
 typedef vpx_codec_err_t
 (*vpx_codec_enc_mr_get_mem_loc_fn_t)(const vpx_codec_enc_cfg_t     *cfg,
-                                   void **mem_loc);
+                                     void **mem_loc);
 
 /*!\brief usage configuration mapping
  *
@@ -280,10 +279,9 @@
  * one mapping must be present, in addition to the end-of-list.
  *
  */
-typedef const struct vpx_codec_enc_cfg_map
-{
-    int                 usage;
-    vpx_codec_enc_cfg_t cfg;
+typedef const struct vpx_codec_enc_cfg_map {
+  int                 usage;
+  vpx_codec_enc_cfg_t cfg;
 } vpx_codec_enc_cfg_map_t;
 
 #define NOT_IMPLEMENTED 0
@@ -292,44 +290,39 @@
  *
  * All decoders \ref MUST expose a variable of this type.
  */
-struct vpx_codec_iface
-{
-    const char               *name;        /**< Identification String  */
-    int                       abi_version; /**< Implemented ABI version */
-    vpx_codec_caps_t          caps;    /**< Decoder capabilities */
-    vpx_codec_init_fn_t       init;    /**< \copydoc ::vpx_codec_init_fn_t */
-    vpx_codec_destroy_fn_t    destroy;     /**< \copydoc ::vpx_codec_destroy_fn_t */
-    vpx_codec_ctrl_fn_map_t  *ctrl_maps;   /**< \copydoc ::vpx_codec_ctrl_fn_map_t */
-    vpx_codec_get_mmap_fn_t   get_mmap;    /**< \copydoc ::vpx_codec_get_mmap_fn_t */
-    vpx_codec_set_mmap_fn_t   set_mmap;    /**< \copydoc ::vpx_codec_set_mmap_fn_t */
-    struct vpx_codec_dec_iface
-    {
-        vpx_codec_peek_si_fn_t    peek_si;     /**< \copydoc ::vpx_codec_peek_si_fn_t */
-        vpx_codec_get_si_fn_t     get_si;      /**< \copydoc ::vpx_codec_peek_si_fn_t */
-        vpx_codec_decode_fn_t     decode;      /**< \copydoc ::vpx_codec_decode_fn_t */
-        vpx_codec_get_frame_fn_t  get_frame;   /**< \copydoc ::vpx_codec_get_frame_fn_t */
-    } dec;
-    struct vpx_codec_enc_iface
-    {
-        vpx_codec_enc_cfg_map_t           *cfg_maps;      /**< \copydoc ::vpx_codec_enc_cfg_map_t */
-        vpx_codec_encode_fn_t              encode;        /**< \copydoc ::vpx_codec_encode_fn_t */
-        vpx_codec_get_cx_data_fn_t         get_cx_data;   /**< \copydoc ::vpx_codec_get_cx_data_fn_t */
-        vpx_codec_enc_config_set_fn_t      cfg_set;       /**< \copydoc ::vpx_codec_enc_config_set_fn_t */
-        vpx_codec_get_global_headers_fn_t  get_glob_hdrs; /**< \copydoc ::vpx_codec_get_global_headers_fn_t */
-        vpx_codec_get_preview_frame_fn_t   get_preview;   /**< \copydoc ::vpx_codec_get_preview_frame_fn_t */
-        vpx_codec_enc_mr_get_mem_loc_fn_t  mr_get_mem_loc;   /**< \copydoc ::vpx_codec_enc_mr_get_mem_loc_fn_t */
-    } enc;
+struct vpx_codec_iface {
+  const char               *name;        /**< Identification String  */
+  int                       abi_version; /**< Implemented ABI version */
+  vpx_codec_caps_t          caps;    /**< Decoder capabilities */
+  vpx_codec_init_fn_t       init;    /**< \copydoc ::vpx_codec_init_fn_t */
+  vpx_codec_destroy_fn_t    destroy;     /**< \copydoc ::vpx_codec_destroy_fn_t */
+  vpx_codec_ctrl_fn_map_t  *ctrl_maps;   /**< \copydoc ::vpx_codec_ctrl_fn_map_t */
+  vpx_codec_get_mmap_fn_t   get_mmap;    /**< \copydoc ::vpx_codec_get_mmap_fn_t */
+  vpx_codec_set_mmap_fn_t   set_mmap;    /**< \copydoc ::vpx_codec_set_mmap_fn_t */
+  struct vpx_codec_dec_iface {
+    vpx_codec_peek_si_fn_t    peek_si;     /**< \copydoc ::vpx_codec_peek_si_fn_t */
+    vpx_codec_get_si_fn_t     get_si;      /**< \copydoc ::vpx_codec_peek_si_fn_t */
+    vpx_codec_decode_fn_t     decode;      /**< \copydoc ::vpx_codec_decode_fn_t */
+    vpx_codec_get_frame_fn_t  get_frame;   /**< \copydoc ::vpx_codec_get_frame_fn_t */
+  } dec;
+  struct vpx_codec_enc_iface {
+    vpx_codec_enc_cfg_map_t           *cfg_maps;      /**< \copydoc ::vpx_codec_enc_cfg_map_t */
+    vpx_codec_encode_fn_t              encode;        /**< \copydoc ::vpx_codec_encode_fn_t */
+    vpx_codec_get_cx_data_fn_t         get_cx_data;   /**< \copydoc ::vpx_codec_get_cx_data_fn_t */
+    vpx_codec_enc_config_set_fn_t      cfg_set;       /**< \copydoc ::vpx_codec_enc_config_set_fn_t */
+    vpx_codec_get_global_headers_fn_t  get_glob_hdrs; /**< \copydoc ::vpx_codec_get_global_headers_fn_t */
+    vpx_codec_get_preview_frame_fn_t   get_preview;   /**< \copydoc ::vpx_codec_get_preview_frame_fn_t */
+    vpx_codec_enc_mr_get_mem_loc_fn_t  mr_get_mem_loc;   /**< \copydoc ::vpx_codec_enc_mr_get_mem_loc_fn_t */
+  } enc;
 };
 
 /*!\brief Callback function pointer / user data pair storage */
-typedef struct vpx_codec_priv_cb_pair
-{
-    union
-    {
-        vpx_codec_put_frame_cb_fn_t    put_frame;
-        vpx_codec_put_slice_cb_fn_t    put_slice;
-    } u;
-    void                            *user_priv;
+typedef struct vpx_codec_priv_cb_pair {
+  union {
+    vpx_codec_put_frame_cb_fn_t    put_frame;
+    vpx_codec_put_slice_cb_fn_t    put_slice;
+  } u;
+  void                            *user_priv;
 } vpx_codec_priv_cb_pair_t;
 
 
@@ -341,27 +334,24 @@
  * structure can be made the first member of the algorithm specific structure,
  * and the pointer cast to the proper type.
  */
-struct vpx_codec_priv
-{
-    unsigned int                    sz;
-    vpx_codec_iface_t              *iface;
-    struct vpx_codec_alg_priv      *alg_priv;
-    const char                     *err_detail;
-    vpx_codec_flags_t               init_flags;
-    struct
-    {
-        vpx_codec_priv_cb_pair_t    put_frame_cb;
-        vpx_codec_priv_cb_pair_t    put_slice_cb;
-    } dec;
-    struct
-    {
-        int                         tbd;
-        struct vpx_fixed_buf        cx_data_dst_buf;
-        unsigned int                cx_data_pad_before;
-        unsigned int                cx_data_pad_after;
-        vpx_codec_cx_pkt_t          cx_data_pkt;
-        unsigned int                total_encoders;
-    } enc;
+struct vpx_codec_priv {
+  unsigned int                    sz;
+  vpx_codec_iface_t              *iface;
+  struct vpx_codec_alg_priv      *alg_priv;
+  const char                     *err_detail;
+  vpx_codec_flags_t               init_flags;
+  struct {
+    vpx_codec_priv_cb_pair_t    put_frame_cb;
+    vpx_codec_priv_cb_pair_t    put_slice_cb;
+  } dec;
+  struct {
+    int                         tbd;
+    struct vpx_fixed_buf        cx_data_dst_buf;
+    unsigned int                cx_data_pad_before;
+    unsigned int                cx_data_pad_after;
+    vpx_codec_cx_pkt_t          cx_data_pkt;
+    unsigned int                total_encoders;
+  } enc;
 };
 
 /*
@@ -377,32 +367,32 @@
 
 #undef VPX_CTRL_USE_TYPE
 #define VPX_CTRL_USE_TYPE(id, typ) \
-    static typ id##__value(va_list args) {return va_arg(args, typ);} \
-    static typ id##__convert(void *x)\
+  static typ id##__value(va_list args) {return va_arg(args, typ);} \
+  static typ id##__convert(void *x)\
+  {\
+    union\
     {\
-        union\
-        {\
-            void *x;\
-            typ   d;\
-        } u;\
-        u.x = x;\
-        return u.d;\
-    }
+      void *x;\
+      typ   d;\
+    } u;\
+    u.x = x;\
+    return u.d;\
+  }
 
 
 #undef VPX_CTRL_USE_TYPE_DEPRECATED
 #define VPX_CTRL_USE_TYPE_DEPRECATED(id, typ) \
-    static typ id##__value(va_list args) {return va_arg(args, typ);} \
-    static typ id##__convert(void *x)\
+  static typ id##__value(va_list args) {return va_arg(args, typ);} \
+  static typ id##__convert(void *x)\
+  {\
+    union\
     {\
-        union\
-        {\
-            void *x;\
-            typ   d;\
-        } u;\
-        u.x = x;\
-        return u.d;\
-    }
+      void *x;\
+      typ   d;\
+    } u;\
+    u.x = x;\
+    return u.d;\
+  }
 
 #define CAST(id, arg) id##__value(arg)
 #define RECAST(id, x) id##__convert(x)
@@ -418,8 +408,8 @@
  * macro is provided to define this getter function automatically.
  */
 #define CODEC_INTERFACE(id)\
-vpx_codec_iface_t* id(void) { return &id##_algo; }\
-vpx_codec_iface_t  id##_algo
+  vpx_codec_iface_t* id(void) { return &id##_algo; }\
+  vpx_codec_iface_t  id##_algo
 
 
 /* Internal Utility Functions
@@ -427,64 +417,60 @@
  * The following functions are intended to be used inside algorithms as
  * utilities for manipulating vpx_codec_* data structures.
  */
-struct vpx_codec_pkt_list
-{
-    unsigned int            cnt;
-    unsigned int            max;
-    struct vpx_codec_cx_pkt pkts[1];
+struct vpx_codec_pkt_list {
+  unsigned int            cnt;
+  unsigned int            max;
+  struct vpx_codec_cx_pkt pkts[1];
 };
 
 #define vpx_codec_pkt_list_decl(n)\
-    union {struct vpx_codec_pkt_list head;\
-        struct {struct vpx_codec_pkt_list head;\
-            struct vpx_codec_cx_pkt    pkts[n];} alloc;}
+  union {struct vpx_codec_pkt_list head;\
+    struct {struct vpx_codec_pkt_list head;\
+      struct vpx_codec_cx_pkt    pkts[n];} alloc;}
 
 #define vpx_codec_pkt_list_init(m)\
-    (m)->alloc.head.cnt = 0,\
-                          (m)->alloc.head.max = sizeof((m)->alloc.pkts) / sizeof((m)->alloc.pkts[0])
+  (m)->alloc.head.cnt = 0,\
+                        (m)->alloc.head.max = sizeof((m)->alloc.pkts) / sizeof((m)->alloc.pkts[0])
 
 int
 vpx_codec_pkt_list_add(struct vpx_codec_pkt_list *,
                        const struct vpx_codec_cx_pkt *);
 
-const vpx_codec_cx_pkt_t*
+const vpx_codec_cx_pkt_t *
 vpx_codec_pkt_list_get(struct vpx_codec_pkt_list *list,
                        vpx_codec_iter_t           *iter);
 
 
 #include <stdio.h>
 #include <setjmp.h>
-struct vpx_internal_error_info
-{
-    vpx_codec_err_t  error_code;
-    int              has_detail;
-    char             detail[80];
-    int              setjmp;
-    jmp_buf          jmp;
+struct vpx_internal_error_info {
+  vpx_codec_err_t  error_code;
+  int              has_detail;
+  char             detail[80];
+  int              setjmp;
+  jmp_buf          jmp;
 };
 
 static void vpx_internal_error(struct vpx_internal_error_info *info,
                                vpx_codec_err_t                 error,
                                const char                     *fmt,
-                               ...)
-{
-    va_list ap;
+                               ...) {
+  va_list ap;
 
-    info->error_code = error;
-    info->has_detail = 0;
+  info->error_code = error;
+  info->has_detail = 0;
 
-    if (fmt)
-    {
-        size_t  sz = sizeof(info->detail);
+  if (fmt) {
+    size_t  sz = sizeof(info->detail);
 
-        info->has_detail = 1;
-        va_start(ap, fmt);
-        vsnprintf(info->detail, sz - 1, fmt, ap);
-        va_end(ap);
-        info->detail[sz-1] = '\0';
-    }
+    info->has_detail = 1;
+    va_start(ap, fmt);
+    vsnprintf(info->detail, sz - 1, fmt, ap);
+    va_end(ap);
+    info->detail[sz - 1] = '\0';
+  }
 
-    if (info->setjmp)
-        longjmp(info->jmp, info->error_code);
+  if (info->setjmp)
+    longjmp(info->jmp, info->error_code);
 }
 #endif
diff --git a/libvpx/vpx/src/vpx_codec.c b/libvpx/vpx/src/vpx_codec.c
index f1a8b67..61d7f4c 100644
--- a/libvpx/vpx/src/vpx_codec.c
+++ b/libvpx/vpx/src/vpx_codec.c
@@ -20,131 +20,116 @@
 
 #define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var)
 
-int vpx_codec_version(void)
-{
-    return VERSION_PACKED;
+int vpx_codec_version(void) {
+  return VERSION_PACKED;
 }
 
 
-const char *vpx_codec_version_str(void)
-{
-    return VERSION_STRING_NOSP;
+const char *vpx_codec_version_str(void) {
+  return VERSION_STRING_NOSP;
 }
 
 
-const char *vpx_codec_version_extra_str(void)
-{
-    return VERSION_EXTRA;
+const char *vpx_codec_version_extra_str(void) {
+  return VERSION_EXTRA;
 }
 
 
-const char *vpx_codec_iface_name(vpx_codec_iface_t *iface)
-{
-    return iface ? iface->name : "<invalid interface>";
+const char *vpx_codec_iface_name(vpx_codec_iface_t *iface) {
+  return iface ? iface->name : "<invalid interface>";
 }
 
-const char *vpx_codec_err_to_string(vpx_codec_err_t  err)
-{
-    switch (err)
-    {
+const char *vpx_codec_err_to_string(vpx_codec_err_t  err) {
+  switch (err) {
     case VPX_CODEC_OK:
-        return "Success";
+      return "Success";
     case VPX_CODEC_ERROR:
-        return "Unspecified internal error";
+      return "Unspecified internal error";
     case VPX_CODEC_MEM_ERROR:
-        return "Memory allocation error";
+      return "Memory allocation error";
     case VPX_CODEC_ABI_MISMATCH:
-        return "ABI version mismatch";
+      return "ABI version mismatch";
     case VPX_CODEC_INCAPABLE:
-        return "Codec does not implement requested capability";
+      return "Codec does not implement requested capability";
     case VPX_CODEC_UNSUP_BITSTREAM:
-        return "Bitstream not supported by this decoder";
+      return "Bitstream not supported by this decoder";
     case VPX_CODEC_UNSUP_FEATURE:
-        return "Bitstream required feature not supported by this decoder";
+      return "Bitstream required feature not supported by this decoder";
     case VPX_CODEC_CORRUPT_FRAME:
-        return "Corrupt frame detected";
+      return "Corrupt frame detected";
     case  VPX_CODEC_INVALID_PARAM:
-        return "Invalid parameter";
+      return "Invalid parameter";
     case VPX_CODEC_LIST_END:
-        return "End of iterated list";
-    }
+      return "End of iterated list";
+  }
 
-    return "Unrecognized error code";
+  return "Unrecognized error code";
 }
 
-const char *vpx_codec_error(vpx_codec_ctx_t  *ctx)
-{
-    return (ctx) ? vpx_codec_err_to_string(ctx->err)
-           : vpx_codec_err_to_string(VPX_CODEC_INVALID_PARAM);
+const char *vpx_codec_error(vpx_codec_ctx_t  *ctx) {
+  return (ctx) ? vpx_codec_err_to_string(ctx->err)
+         : vpx_codec_err_to_string(VPX_CODEC_INVALID_PARAM);
 }
 
-const char *vpx_codec_error_detail(vpx_codec_ctx_t  *ctx)
-{
-    if (ctx && ctx->err)
-        return ctx->priv ? ctx->priv->err_detail : ctx->err_detail;
+const char *vpx_codec_error_detail(vpx_codec_ctx_t  *ctx) {
+  if (ctx && ctx->err)
+    return ctx->priv ? ctx->priv->err_detail : ctx->err_detail;
 
-    return NULL;
+  return NULL;
 }
 
 
-vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
-{
-    vpx_codec_err_t res;
+vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx) {
+  vpx_codec_err_t res;
 
-    if (!ctx)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv)
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        if (ctx->priv->alg_priv)
-            ctx->iface->destroy(ctx->priv->alg_priv);
+  if (!ctx)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv)
+    res = VPX_CODEC_ERROR;
+  else {
+    if (ctx->priv->alg_priv)
+      ctx->iface->destroy(ctx->priv->alg_priv);
 
-        ctx->iface = NULL;
-        ctx->name = NULL;
-        ctx->priv = NULL;
-        res = VPX_CODEC_OK;
-    }
+    ctx->iface = NULL;
+    ctx->name = NULL;
+    ctx->priv = NULL;
+    res = VPX_CODEC_OK;
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
-vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface)
-{
-    return (iface) ? iface->caps : 0;
+vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface) {
+  return (iface) ? iface->caps : 0;
 }
 
 
 vpx_codec_err_t vpx_codec_control_(vpx_codec_ctx_t  *ctx,
                                    int               ctrl_id,
-                                   ...)
-{
-    vpx_codec_err_t res;
+                                   ...) {
+  vpx_codec_err_t res;
 
-    if (!ctx || !ctrl_id)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv || !ctx->iface->ctrl_maps)
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        vpx_codec_ctrl_fn_map_t *entry;
+  if (!ctx || !ctrl_id)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv || !ctx->iface->ctrl_maps)
+    res = VPX_CODEC_ERROR;
+  else {
+    vpx_codec_ctrl_fn_map_t *entry;
 
-        res = VPX_CODEC_ERROR;
+    res = VPX_CODEC_ERROR;
 
-        for (entry = ctx->iface->ctrl_maps; entry && entry->fn; entry++)
-        {
-            if (!entry->ctrl_id || entry->ctrl_id == ctrl_id)
-            {
-                va_list  ap;
+    for (entry = ctx->iface->ctrl_maps; entry && entry->fn; entry++) {
+      if (!entry->ctrl_id || entry->ctrl_id == ctrl_id) {
+        va_list  ap;
 
-                va_start(ap, ctrl_id);
-                res = entry->fn(ctx->priv->alg_priv, ctrl_id, ap);
-                va_end(ap);
-                break;
-            }
-        }
+        va_start(ap, ctrl_id);
+        res = entry->fn(ctx->priv->alg_priv, ctrl_id, ap);
+        va_end(ap);
+        break;
+      }
     }
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
diff --git a/libvpx/vpx/src/vpx_decoder.c b/libvpx/vpx/src/vpx_decoder.c
index 59a783d..1f575e0 100644
--- a/libvpx/vpx/src/vpx_decoder.c
+++ b/libvpx/vpx/src/vpx_decoder.c
@@ -22,99 +22,91 @@
                                        vpx_codec_iface_t    *iface,
                                        vpx_codec_dec_cfg_t  *cfg,
                                        vpx_codec_flags_t     flags,
-                                       int                   ver)
-{
-    vpx_codec_err_t res;
+                                       int                   ver) {
+  vpx_codec_err_t res;
 
-    if (ver != VPX_DECODER_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if (!ctx || !iface)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_POSTPROC) && !(iface->caps & VPX_CODEC_CAP_POSTPROC))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_ERROR_CONCEALMENT) &&
-            !(iface->caps & VPX_CODEC_CAP_ERROR_CONCEALMENT))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_INPUT_FRAGMENTS) &&
-            !(iface->caps & VPX_CODEC_CAP_INPUT_FRAGMENTS))
-        res = VPX_CODEC_INCAPABLE;
-    else if (!(iface->caps & VPX_CODEC_CAP_DECODER))
-        res = VPX_CODEC_INCAPABLE;
-    else
-    {
-        memset(ctx, 0, sizeof(*ctx));
-        ctx->iface = iface;
-        ctx->name = iface->name;
-        ctx->priv = NULL;
-        ctx->init_flags = flags;
-        ctx->config.dec = cfg;
-        res = VPX_CODEC_OK;
+  if (ver != VPX_DECODER_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if (!ctx || !iface)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_POSTPROC) && !(iface->caps & VPX_CODEC_CAP_POSTPROC))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_ERROR_CONCEALMENT) &&
+           !(iface->caps & VPX_CODEC_CAP_ERROR_CONCEALMENT))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_INPUT_FRAGMENTS) &&
+           !(iface->caps & VPX_CODEC_CAP_INPUT_FRAGMENTS))
+    res = VPX_CODEC_INCAPABLE;
+  else if (!(iface->caps & VPX_CODEC_CAP_DECODER))
+    res = VPX_CODEC_INCAPABLE;
+  else {
+    memset(ctx, 0, sizeof(*ctx));
+    ctx->iface = iface;
+    ctx->name = iface->name;
+    ctx->priv = NULL;
+    ctx->init_flags = flags;
+    ctx->config.dec = cfg;
+    res = VPX_CODEC_OK;
 
-        if (!(flags & VPX_CODEC_USE_XMA))
-        {
-            res = ctx->iface->init(ctx, NULL);
+    if (!(flags & VPX_CODEC_USE_XMA)) {
+      res = ctx->iface->init(ctx, NULL);
 
-            if (res)
-            {
-                ctx->err_detail = ctx->priv ? ctx->priv->err_detail : NULL;
-                vpx_codec_destroy(ctx);
-            }
+      if (res) {
+        ctx->err_detail = ctx->priv ? ctx->priv->err_detail : NULL;
+        vpx_codec_destroy(ctx);
+      }
 
-            if (ctx->priv)
-                ctx->priv->iface = ctx->iface;
-        }
+      if (ctx->priv)
+        ctx->priv->iface = ctx->iface;
     }
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
 vpx_codec_err_t vpx_codec_peek_stream_info(vpx_codec_iface_t       *iface,
-        const uint8_t         *data,
-        unsigned int           data_sz,
-        vpx_codec_stream_info_t *si)
-{
-    vpx_codec_err_t res;
+                                           const uint8_t         *data,
+                                           unsigned int           data_sz,
+                                           vpx_codec_stream_info_t *si) {
+  vpx_codec_err_t res;
 
-    if (!iface || !data || !data_sz || !si
-        || si->sz < sizeof(vpx_codec_stream_info_t))
-        res = VPX_CODEC_INVALID_PARAM;
-    else
-    {
-        /* Set default/unknown values */
-        si->w = 0;
-        si->h = 0;
+  if (!iface || !data || !data_sz || !si
+      || si->sz < sizeof(vpx_codec_stream_info_t))
+    res = VPX_CODEC_INVALID_PARAM;
+  else {
+    /* Set default/unknown values */
+    si->w = 0;
+    si->h = 0;
 
-        res = iface->dec.peek_si(data, data_sz, si);
-    }
+    res = iface->dec.peek_si(data, data_sz, si);
+  }
 
-    return res;
+  return res;
 }
 
 
 vpx_codec_err_t vpx_codec_get_stream_info(vpx_codec_ctx_t         *ctx,
-        vpx_codec_stream_info_t *si)
-{
-    vpx_codec_err_t res;
+                                          vpx_codec_stream_info_t *si) {
+  vpx_codec_err_t res;
 
-    if (!ctx || !si || si->sz < sizeof(vpx_codec_stream_info_t))
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv)
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        /* Set default/unknown values */
-        si->w = 0;
-        si->h = 0;
+  if (!ctx || !si || si->sz < sizeof(vpx_codec_stream_info_t))
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv)
+    res = VPX_CODEC_ERROR;
+  else {
+    /* Set default/unknown values */
+    si->w = 0;
+    si->h = 0;
 
-        res = ctx->iface->dec.get_si(ctx->priv->alg_priv, si);
-    }
+    res = ctx->iface->dec.get_si(ctx->priv->alg_priv, si);
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
@@ -122,126 +114,115 @@
                                  const uint8_t        *data,
                                  unsigned int    data_sz,
                                  void       *user_priv,
-                                 long        deadline)
-{
-    vpx_codec_err_t res;
+                                 long        deadline) {
+  vpx_codec_err_t res;
 
-    /* Sanity checks */
-    /* NULL data ptr allowed if data_sz is 0 too */
-    if (!ctx || (!data && data_sz))
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv)
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        res = ctx->iface->dec.decode(ctx->priv->alg_priv, data, data_sz,
-                                     user_priv, deadline);
-    }
+  /* Sanity checks */
+  /* NULL data ptr allowed if data_sz is 0 too */
+  if (!ctx || (!data && data_sz))
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv)
+    res = VPX_CODEC_ERROR;
+  else {
+    res = ctx->iface->dec.decode(ctx->priv->alg_priv, data, data_sz,
+                                 user_priv, deadline);
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 vpx_image_t *vpx_codec_get_frame(vpx_codec_ctx_t  *ctx,
-                                 vpx_codec_iter_t *iter)
-{
-    vpx_image_t *img;
+                                 vpx_codec_iter_t *iter) {
+  vpx_image_t *img;
 
-    if (!ctx || !iter || !ctx->iface || !ctx->priv)
-        img = NULL;
-    else
-        img = ctx->iface->dec.get_frame(ctx->priv->alg_priv, iter);
+  if (!ctx || !iter || !ctx->iface || !ctx->priv)
+    img = NULL;
+  else
+    img = ctx->iface->dec.get_frame(ctx->priv->alg_priv, iter);
 
-    return img;
+  return img;
 }
 
 
 vpx_codec_err_t vpx_codec_register_put_frame_cb(vpx_codec_ctx_t             *ctx,
-        vpx_codec_put_frame_cb_fn_t  cb,
-        void                      *user_priv)
-{
-    vpx_codec_err_t res;
+                                                vpx_codec_put_frame_cb_fn_t  cb,
+                                                void                      *user_priv) {
+  vpx_codec_err_t res;
 
-    if (!ctx || !cb)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv
-             || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_FRAME))
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        ctx->priv->dec.put_frame_cb.u.put_frame = cb;
-        ctx->priv->dec.put_frame_cb.user_priv = user_priv;
-        res = VPX_CODEC_OK;
-    }
+  if (!ctx || !cb)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv
+           || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_FRAME))
+    res = VPX_CODEC_ERROR;
+  else {
+    ctx->priv->dec.put_frame_cb.u.put_frame = cb;
+    ctx->priv->dec.put_frame_cb.user_priv = user_priv;
+    res = VPX_CODEC_OK;
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
 vpx_codec_err_t vpx_codec_register_put_slice_cb(vpx_codec_ctx_t             *ctx,
-        vpx_codec_put_slice_cb_fn_t  cb,
-        void                      *user_priv)
-{
-    vpx_codec_err_t res;
+                                                vpx_codec_put_slice_cb_fn_t  cb,
+                                                void                      *user_priv) {
+  vpx_codec_err_t res;
 
-    if (!ctx || !cb)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv
-             || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_FRAME))
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        ctx->priv->dec.put_slice_cb.u.put_slice = cb;
-        ctx->priv->dec.put_slice_cb.user_priv = user_priv;
-        res = VPX_CODEC_OK;
-    }
+  if (!ctx || !cb)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv
+           || !(ctx->iface->caps & VPX_CODEC_CAP_PUT_FRAME))
+    res = VPX_CODEC_ERROR;
+  else {
+    ctx->priv->dec.put_slice_cb.u.put_slice = cb;
+    ctx->priv->dec.put_slice_cb.user_priv = user_priv;
+    res = VPX_CODEC_OK;
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
 vpx_codec_err_t vpx_codec_get_mem_map(vpx_codec_ctx_t                *ctx,
                                       vpx_codec_mmap_t               *mmap,
-                                      vpx_codec_iter_t               *iter)
-{
-    vpx_codec_err_t res = VPX_CODEC_OK;
+                                      vpx_codec_iter_t               *iter) {
+  vpx_codec_err_t res = VPX_CODEC_OK;
 
-    if (!ctx || !mmap || !iter || !ctx->iface)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!(ctx->iface->caps & VPX_CODEC_CAP_XMA))
-        res = VPX_CODEC_ERROR;
-    else
-        res = ctx->iface->get_mmap(ctx, mmap, iter);
+  if (!ctx || !mmap || !iter || !ctx->iface)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!(ctx->iface->caps & VPX_CODEC_CAP_XMA))
+    res = VPX_CODEC_ERROR;
+  else
+    res = ctx->iface->get_mmap(ctx, mmap, iter);
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
 vpx_codec_err_t vpx_codec_set_mem_map(vpx_codec_ctx_t   *ctx,
                                       vpx_codec_mmap_t  *mmap,
-                                      unsigned int     num_maps)
-{
-    vpx_codec_err_t res = VPX_CODEC_MEM_ERROR;
+                                      unsigned int     num_maps) {
+  vpx_codec_err_t res = VPX_CODEC_MEM_ERROR;
 
-    if (!ctx || !mmap || !ctx->iface)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!(ctx->iface->caps & VPX_CODEC_CAP_XMA))
-        res = VPX_CODEC_ERROR;
-    else
-    {
-        unsigned int i;
+  if (!ctx || !mmap || !ctx->iface)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!(ctx->iface->caps & VPX_CODEC_CAP_XMA))
+    res = VPX_CODEC_ERROR;
+  else {
+    unsigned int i;
 
-        for (i = 0; i < num_maps; i++, mmap++)
-        {
-            if (!mmap->base)
-                break;
+    for (i = 0; i < num_maps; i++, mmap++) {
+      if (!mmap->base)
+        break;
 
-            /* Everything look ok, set the mmap in the decoder */
-            res = ctx->iface->set_mmap(ctx, mmap);
+      /* Everything look ok, set the mmap in the decoder */
+      res = ctx->iface->set_mmap(ctx, mmap);
 
-            if (res)
-                break;
-        }
+      if (res)
+        break;
     }
+  }
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
diff --git a/libvpx/vpx/src/vpx_encoder.c b/libvpx/vpx/src/vpx_encoder.c
index 73c1c66..3cec895 100644
--- a/libvpx/vpx/src/vpx_encoder.c
+++ b/libvpx/vpx/src/vpx_encoder.c
@@ -24,46 +24,43 @@
                                        vpx_codec_iface_t    *iface,
                                        vpx_codec_enc_cfg_t  *cfg,
                                        vpx_codec_flags_t     flags,
-                                       int                   ver)
-{
-    vpx_codec_err_t res;
+                                       int                   ver) {
+  vpx_codec_err_t res;
 
-    if (ver != VPX_ENCODER_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if (!ctx || !iface || !cfg)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_PSNR)
-             && !(iface->caps & VPX_CODEC_CAP_PSNR))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_OUTPUT_PARTITION)
-             && !(iface->caps & VPX_CODEC_CAP_OUTPUT_PARTITION))
-        res = VPX_CODEC_INCAPABLE;
-    else
-    {
-        ctx->iface = iface;
-        ctx->name = iface->name;
-        ctx->priv = NULL;
-        ctx->init_flags = flags;
-        ctx->config.enc = cfg;
-        res = ctx->iface->init(ctx, NULL);
+  if (ver != VPX_ENCODER_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if (!ctx || !iface || !cfg)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_PSNR)
+           && !(iface->caps & VPX_CODEC_CAP_PSNR))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_OUTPUT_PARTITION)
+           && !(iface->caps & VPX_CODEC_CAP_OUTPUT_PARTITION))
+    res = VPX_CODEC_INCAPABLE;
+  else {
+    ctx->iface = iface;
+    ctx->name = iface->name;
+    ctx->priv = NULL;
+    ctx->init_flags = flags;
+    ctx->config.enc = cfg;
+    res = ctx->iface->init(ctx, NULL);
 
-        if (res)
-        {
-            ctx->err_detail = ctx->priv ? ctx->priv->err_detail : NULL;
-            vpx_codec_destroy(ctx);
-        }
-
-        if (ctx->priv)
-            ctx->priv->iface = ctx->iface;
+    if (res) {
+      ctx->err_detail = ctx->priv ? ctx->priv->err_detail : NULL;
+      vpx_codec_destroy(ctx);
     }
 
-    return SAVE_STATUS(ctx, res);
+    if (ctx->priv)
+      ctx->priv->iface = ctx->iface;
+  }
+
+  return SAVE_STATUS(ctx, res);
 }
 
 vpx_codec_err_t vpx_codec_enc_init_multi_ver(vpx_codec_ctx_t      *ctx,
@@ -72,128 +69,117 @@
                                              int                   num_enc,
                                              vpx_codec_flags_t     flags,
                                              vpx_rational_t       *dsf,
-                                             int                   ver)
-{
-    vpx_codec_err_t res = 0;
+                                             int                   ver) {
+  vpx_codec_err_t res = 0;
 
-    if (ver != VPX_ENCODER_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if (!ctx || !iface || !cfg || (num_enc > 16 || num_enc < 1))
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
-        res = VPX_CODEC_ABI_MISMATCH;
-    else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_PSNR)
-             && !(iface->caps & VPX_CODEC_CAP_PSNR))
-        res = VPX_CODEC_INCAPABLE;
-    else if ((flags & VPX_CODEC_USE_OUTPUT_PARTITION)
-             && !(iface->caps & VPX_CODEC_CAP_OUTPUT_PARTITION))
-        res = VPX_CODEC_INCAPABLE;
-    else
-    {
-        int i;
-        void *mem_loc = NULL;
+  if (ver != VPX_ENCODER_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if (!ctx || !iface || !cfg || (num_enc > 16 || num_enc < 1))
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (iface->abi_version != VPX_CODEC_INTERNAL_ABI_VERSION)
+    res = VPX_CODEC_ABI_MISMATCH;
+  else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_XMA) && !(iface->caps & VPX_CODEC_CAP_XMA))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_PSNR)
+           && !(iface->caps & VPX_CODEC_CAP_PSNR))
+    res = VPX_CODEC_INCAPABLE;
+  else if ((flags & VPX_CODEC_USE_OUTPUT_PARTITION)
+           && !(iface->caps & VPX_CODEC_CAP_OUTPUT_PARTITION))
+    res = VPX_CODEC_INCAPABLE;
+  else {
+    int i;
+    void *mem_loc = NULL;
 
-        if(!(res = iface->enc.mr_get_mem_loc(cfg, &mem_loc)))
-        {
-            for (i = 0; i < num_enc; i++)
-            {
-                vpx_codec_priv_enc_mr_cfg_t mr_cfg;
+    if (!(res = iface->enc.mr_get_mem_loc(cfg, &mem_loc))) {
+      for (i = 0; i < num_enc; i++) {
+        vpx_codec_priv_enc_mr_cfg_t mr_cfg;
 
-                /* Validate down-sampling factor. */
-                if(dsf->num < 1 || dsf->num > 4096 || dsf->den < 1 ||
-                   dsf->den > dsf->num)
-                {
-                    res = VPX_CODEC_INVALID_PARAM;
-                    break;
-                }
-
-                mr_cfg.mr_low_res_mode_info = mem_loc;
-                mr_cfg.mr_total_resolutions = num_enc;
-                mr_cfg.mr_encoder_id = num_enc-1-i;
-                mr_cfg.mr_down_sampling_factor.num = dsf->num;
-                mr_cfg.mr_down_sampling_factor.den = dsf->den;
-
-                /* Force Key-frame synchronization. Namely, encoder at higher
-                 * resolution always use the same frame_type chosen by the
-                 * lowest-resolution encoder.
-                 */
-                if(mr_cfg.mr_encoder_id)
-                    cfg->kf_mode = VPX_KF_DISABLED;
-
-                ctx->iface = iface;
-                ctx->name = iface->name;
-                ctx->priv = NULL;
-                ctx->init_flags = flags;
-                ctx->config.enc = cfg;
-                res = ctx->iface->init(ctx, &mr_cfg);
-
-                if (res)
-                {
-                    const char *error_detail =
-                        ctx->priv ? ctx->priv->err_detail : NULL;
-                    /* Destroy current ctx */
-                    ctx->err_detail = error_detail;
-                    vpx_codec_destroy(ctx);
-
-                    /* Destroy already allocated high-level ctx */
-                    while (i)
-                    {
-                        ctx--;
-                        ctx->err_detail = error_detail;
-                        vpx_codec_destroy(ctx);
-                        i--;
-                    }
-                }
-
-                if (ctx->priv)
-                    ctx->priv->iface = ctx->iface;
-
-                if (res)
-                    break;
-
-                ctx++;
-                cfg++;
-                dsf++;
-            }
+        /* Validate down-sampling factor. */
+        if (dsf->num < 1 || dsf->num > 4096 || dsf->den < 1 ||
+            dsf->den > dsf->num) {
+          res = VPX_CODEC_INVALID_PARAM;
+          break;
         }
-    }
 
-    return SAVE_STATUS(ctx, res);
+        mr_cfg.mr_low_res_mode_info = mem_loc;
+        mr_cfg.mr_total_resolutions = num_enc;
+        mr_cfg.mr_encoder_id = num_enc - 1 - i;
+        mr_cfg.mr_down_sampling_factor.num = dsf->num;
+        mr_cfg.mr_down_sampling_factor.den = dsf->den;
+
+        /* Force Key-frame synchronization. Namely, encoder at higher
+         * resolution always use the same frame_type chosen by the
+         * lowest-resolution encoder.
+         */
+        if (mr_cfg.mr_encoder_id)
+          cfg->kf_mode = VPX_KF_DISABLED;
+
+        ctx->iface = iface;
+        ctx->name = iface->name;
+        ctx->priv = NULL;
+        ctx->init_flags = flags;
+        ctx->config.enc = cfg;
+        res = ctx->iface->init(ctx, &mr_cfg);
+
+        if (res) {
+          const char *error_detail =
+            ctx->priv ? ctx->priv->err_detail : NULL;
+          /* Destroy current ctx */
+          ctx->err_detail = error_detail;
+          vpx_codec_destroy(ctx);
+
+          /* Destroy already allocated high-level ctx */
+          while (i) {
+            ctx--;
+            ctx->err_detail = error_detail;
+            vpx_codec_destroy(ctx);
+            i--;
+          }
+        }
+
+        if (ctx->priv)
+          ctx->priv->iface = ctx->iface;
+
+        if (res)
+          break;
+
+        ctx++;
+        cfg++;
+        dsf++;
+      }
+    }
+  }
+
+  return SAVE_STATUS(ctx, res);
 }
 
 
 vpx_codec_err_t  vpx_codec_enc_config_default(vpx_codec_iface_t    *iface,
-        vpx_codec_enc_cfg_t  *cfg,
-        unsigned int          usage)
-{
-    vpx_codec_err_t res;
-    vpx_codec_enc_cfg_map_t *map;
+                                              vpx_codec_enc_cfg_t  *cfg,
+                                              unsigned int          usage) {
+  vpx_codec_err_t res;
+  vpx_codec_enc_cfg_map_t *map;
 
-    if (!iface || !cfg || usage > INT_MAX)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
-        res = VPX_CODEC_INCAPABLE;
-    else
-    {
-        res = VPX_CODEC_INVALID_PARAM;
+  if (!iface || !cfg || usage > INT_MAX)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!(iface->caps & VPX_CODEC_CAP_ENCODER))
+    res = VPX_CODEC_INCAPABLE;
+  else {
+    res = VPX_CODEC_INVALID_PARAM;
 
-        for (map = iface->enc.cfg_maps; map->usage >= 0; map++)
-        {
-            if (map->usage == (int)usage)
-            {
-                *cfg = map->cfg;
-                cfg->g_usage = usage;
-                res = VPX_CODEC_OK;
-                break;
-            }
-        }
+    for (map = iface->enc.cfg_maps; map->usage >= 0; map++) {
+      if (map->usage == (int)usage) {
+        *cfg = map->cfg;
+        cfg->g_usage = usage;
+        res = VPX_CODEC_OK;
+        break;
+      }
     }
+  }
 
-    return res;
+  return res;
 }
 
 
@@ -203,9 +189,9 @@
  */
 #include "vpx_ports/x86.h"
 #define FLOATING_POINT_INIT() do {\
-        unsigned short x87_orig_mode = x87_set_double_precision();
+    unsigned short x87_orig_mode = x87_set_double_precision();
 #define FLOATING_POINT_RESTORE() \
-    x87_set_control_word(x87_orig_mode); }while(0)
+  x87_set_control_word(x87_orig_mode); }while(0)
 
 
 #else
@@ -219,224 +205,202 @@
                                   vpx_codec_pts_t             pts,
                                   unsigned long               duration,
                                   vpx_enc_frame_flags_t       flags,
-                                  unsigned long               deadline)
-{
-    vpx_codec_err_t res = 0;
+                                  unsigned long               deadline) {
+  vpx_codec_err_t res = 0;
 
-    if (!ctx || (img && !duration))
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!ctx->iface || !ctx->priv)
-        res = VPX_CODEC_ERROR;
-    else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
-        res = VPX_CODEC_INCAPABLE;
-    else
-    {
-        /* Execute in a normalized floating point environment, if the platform
-         * requires it.
-         */
-        unsigned int num_enc =ctx->priv->enc.total_encoders;
+  if (!ctx || (img && !duration))
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!ctx->iface || !ctx->priv)
+    res = VPX_CODEC_ERROR;
+  else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
+    res = VPX_CODEC_INCAPABLE;
+  else {
+    /* Execute in a normalized floating point environment, if the platform
+     * requires it.
+     */
+    unsigned int num_enc = ctx->priv->enc.total_encoders;
 
-        FLOATING_POINT_INIT();
+    FLOATING_POINT_INIT();
 
-        if (num_enc == 1)
-            res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
-                                         duration, flags, deadline);
-        else
-        {
-            /* Multi-resolution encoding:
-             * Encode multi-levels in reverse order. For example,
-             * if mr_total_resolutions = 3, first encode level 2,
-             * then encode level 1, and finally encode level 0.
-             */
-            int i;
+    if (num_enc == 1)
+      res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
+                                   duration, flags, deadline);
+    else {
+      /* Multi-resolution encoding:
+       * Encode multi-levels in reverse order. For example,
+       * if mr_total_resolutions = 3, first encode level 2,
+       * then encode level 1, and finally encode level 0.
+       */
+      int i;
 
-            ctx += num_enc - 1;
-            if (img) img += num_enc - 1;
+      ctx += num_enc - 1;
+      if (img) img += num_enc - 1;
 
-            for (i = num_enc-1; i >= 0; i--)
-            {
-                if ((res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
-                                                  duration, flags, deadline)))
-                    break;
+      for (i = num_enc - 1; i >= 0; i--) {
+        if ((res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
+                                          duration, flags, deadline)))
+          break;
 
-                ctx--;
-                if (img) img--;
-            }
-            ctx++;
-        }
-
-        FLOATING_POINT_RESTORE();
+        ctx--;
+        if (img) img--;
+      }
+      ctx++;
     }
 
-    return SAVE_STATUS(ctx, res);
+    FLOATING_POINT_RESTORE();
+  }
+
+  return SAVE_STATUS(ctx, res);
 }
 
 
 const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t   *ctx,
-        vpx_codec_iter_t  *iter)
-{
-    const vpx_codec_cx_pkt_t *pkt = NULL;
+                                                vpx_codec_iter_t  *iter) {
+  const vpx_codec_cx_pkt_t *pkt = NULL;
 
-    if (ctx)
-    {
-        if (!iter)
-            ctx->err = VPX_CODEC_INVALID_PARAM;
-        else if (!ctx->iface || !ctx->priv)
-            ctx->err = VPX_CODEC_ERROR;
-        else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
-            ctx->err = VPX_CODEC_INCAPABLE;
-        else
-            pkt = ctx->iface->enc.get_cx_data(ctx->priv->alg_priv, iter);
+  if (ctx) {
+    if (!iter)
+      ctx->err = VPX_CODEC_INVALID_PARAM;
+    else if (!ctx->iface || !ctx->priv)
+      ctx->err = VPX_CODEC_ERROR;
+    else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
+      ctx->err = VPX_CODEC_INCAPABLE;
+    else
+      pkt = ctx->iface->enc.get_cx_data(ctx->priv->alg_priv, iter);
+  }
+
+  if (pkt && pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
+    /* If the application has specified a destination area for the
+     * compressed data, and the codec has not placed the data there,
+     * and it fits, copy it.
+     */
+    char *dst_buf = ctx->priv->enc.cx_data_dst_buf.buf;
+
+    if (dst_buf
+        && pkt->data.raw.buf != dst_buf
+        && pkt->data.raw.sz
+        + ctx->priv->enc.cx_data_pad_before
+        + ctx->priv->enc.cx_data_pad_after
+        <= ctx->priv->enc.cx_data_dst_buf.sz) {
+      vpx_codec_cx_pkt_t *modified_pkt = &ctx->priv->enc.cx_data_pkt;
+
+      memcpy(dst_buf + ctx->priv->enc.cx_data_pad_before,
+             pkt->data.raw.buf, pkt->data.raw.sz);
+      *modified_pkt = *pkt;
+      modified_pkt->data.raw.buf = dst_buf;
+      modified_pkt->data.raw.sz += ctx->priv->enc.cx_data_pad_before
+                                   + ctx->priv->enc.cx_data_pad_after;
+      pkt = modified_pkt;
     }
 
-    if (pkt && pkt->kind == VPX_CODEC_CX_FRAME_PKT)
-    {
-        /* If the application has specified a destination area for the
-         * compressed data, and the codec has not placed the data there,
-         * and it fits, copy it.
-         */
-        char *dst_buf = ctx->priv->enc.cx_data_dst_buf.buf;
-
-        if (dst_buf
-            && pkt->data.raw.buf != dst_buf
-            && pkt->data.raw.sz
-            + ctx->priv->enc.cx_data_pad_before
-            + ctx->priv->enc.cx_data_pad_after
-            <= ctx->priv->enc.cx_data_dst_buf.sz)
-        {
-            vpx_codec_cx_pkt_t *modified_pkt = &ctx->priv->enc.cx_data_pkt;
-
-            memcpy(dst_buf + ctx->priv->enc.cx_data_pad_before,
-                   pkt->data.raw.buf, pkt->data.raw.sz);
-            *modified_pkt = *pkt;
-            modified_pkt->data.raw.buf = dst_buf;
-            modified_pkt->data.raw.sz += ctx->priv->enc.cx_data_pad_before
-                                         + ctx->priv->enc.cx_data_pad_after;
-            pkt = modified_pkt;
-        }
-
-        if (dst_buf == pkt->data.raw.buf)
-        {
-            ctx->priv->enc.cx_data_dst_buf.buf = dst_buf + pkt->data.raw.sz;
-            ctx->priv->enc.cx_data_dst_buf.sz -= pkt->data.raw.sz;
-        }
+    if (dst_buf == pkt->data.raw.buf) {
+      ctx->priv->enc.cx_data_dst_buf.buf = dst_buf + pkt->data.raw.sz;
+      ctx->priv->enc.cx_data_dst_buf.sz -= pkt->data.raw.sz;
     }
+  }
 
-    return pkt;
+  return pkt;
 }
 
 
 vpx_codec_err_t vpx_codec_set_cx_data_buf(vpx_codec_ctx_t       *ctx,
-        const vpx_fixed_buf_t *buf,
-        unsigned int           pad_before,
-        unsigned int           pad_after)
-{
-    if (!ctx || !ctx->priv)
-        return VPX_CODEC_INVALID_PARAM;
+                                          const vpx_fixed_buf_t *buf,
+                                          unsigned int           pad_before,
+                                          unsigned int           pad_after) {
+  if (!ctx || !ctx->priv)
+    return VPX_CODEC_INVALID_PARAM;
 
-    if (buf)
-    {
-        ctx->priv->enc.cx_data_dst_buf = *buf;
-        ctx->priv->enc.cx_data_pad_before = pad_before;
-        ctx->priv->enc.cx_data_pad_after = pad_after;
-    }
+  if (buf) {
+    ctx->priv->enc.cx_data_dst_buf = *buf;
+    ctx->priv->enc.cx_data_pad_before = pad_before;
+    ctx->priv->enc.cx_data_pad_after = pad_after;
+  } else {
+    ctx->priv->enc.cx_data_dst_buf.buf = NULL;
+    ctx->priv->enc.cx_data_dst_buf.sz = 0;
+    ctx->priv->enc.cx_data_pad_before = 0;
+    ctx->priv->enc.cx_data_pad_after = 0;
+  }
+
+  return VPX_CODEC_OK;
+}
+
+
+const vpx_image_t *vpx_codec_get_preview_frame(vpx_codec_ctx_t   *ctx) {
+  vpx_image_t *img = NULL;
+
+  if (ctx) {
+    if (!ctx->iface || !ctx->priv)
+      ctx->err = VPX_CODEC_ERROR;
+    else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
+      ctx->err = VPX_CODEC_INCAPABLE;
+    else if (!ctx->iface->enc.get_preview)
+      ctx->err = VPX_CODEC_INCAPABLE;
     else
-    {
-        ctx->priv->enc.cx_data_dst_buf.buf = NULL;
-        ctx->priv->enc.cx_data_dst_buf.sz = 0;
-        ctx->priv->enc.cx_data_pad_before = 0;
-        ctx->priv->enc.cx_data_pad_after = 0;
-    }
+      img = ctx->iface->enc.get_preview(ctx->priv->alg_priv);
+  }
 
-    return VPX_CODEC_OK;
+  return img;
 }
 
 
-const vpx_image_t *vpx_codec_get_preview_frame(vpx_codec_ctx_t   *ctx)
-{
-    vpx_image_t *img = NULL;
+vpx_fixed_buf_t *vpx_codec_get_global_headers(vpx_codec_ctx_t   *ctx) {
+  vpx_fixed_buf_t *buf = NULL;
 
-    if (ctx)
-    {
-        if (!ctx->iface || !ctx->priv)
-            ctx->err = VPX_CODEC_ERROR;
-        else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
-            ctx->err = VPX_CODEC_INCAPABLE;
-        else if (!ctx->iface->enc.get_preview)
-            ctx->err = VPX_CODEC_INCAPABLE;
-        else
-            img = ctx->iface->enc.get_preview(ctx->priv->alg_priv);
-    }
+  if (ctx) {
+    if (!ctx->iface || !ctx->priv)
+      ctx->err = VPX_CODEC_ERROR;
+    else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
+      ctx->err = VPX_CODEC_INCAPABLE;
+    else if (!ctx->iface->enc.get_glob_hdrs)
+      ctx->err = VPX_CODEC_INCAPABLE;
+    else
+      buf = ctx->iface->enc.get_glob_hdrs(ctx->priv->alg_priv);
+  }
 
-    return img;
-}
-
-
-vpx_fixed_buf_t *vpx_codec_get_global_headers(vpx_codec_ctx_t   *ctx)
-{
-    vpx_fixed_buf_t *buf = NULL;
-
-    if (ctx)
-    {
-        if (!ctx->iface || !ctx->priv)
-            ctx->err = VPX_CODEC_ERROR;
-        else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
-            ctx->err = VPX_CODEC_INCAPABLE;
-        else if (!ctx->iface->enc.get_glob_hdrs)
-            ctx->err = VPX_CODEC_INCAPABLE;
-        else
-            buf = ctx->iface->enc.get_glob_hdrs(ctx->priv->alg_priv);
-    }
-
-    return buf;
+  return buf;
 }
 
 
 vpx_codec_err_t  vpx_codec_enc_config_set(vpx_codec_ctx_t            *ctx,
-        const vpx_codec_enc_cfg_t  *cfg)
-{
-    vpx_codec_err_t res;
+                                          const vpx_codec_enc_cfg_t  *cfg) {
+  vpx_codec_err_t res;
 
-    if (!ctx || !ctx->iface || !ctx->priv || !cfg)
-        res = VPX_CODEC_INVALID_PARAM;
-    else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
-        res = VPX_CODEC_INCAPABLE;
-    else
-        res = ctx->iface->enc.cfg_set(ctx->priv->alg_priv, cfg);
+  if (!ctx || !ctx->iface || !ctx->priv || !cfg)
+    res = VPX_CODEC_INVALID_PARAM;
+  else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
+    res = VPX_CODEC_INCAPABLE;
+  else
+    res = ctx->iface->enc.cfg_set(ctx->priv->alg_priv, cfg);
 
-    return SAVE_STATUS(ctx, res);
+  return SAVE_STATUS(ctx, res);
 }
 
 
 int vpx_codec_pkt_list_add(struct vpx_codec_pkt_list *list,
-                           const struct vpx_codec_cx_pkt *pkt)
-{
-    if (list->cnt < list->max)
-    {
-        list->pkts[list->cnt++] = *pkt;
-        return 0;
-    }
+                           const struct vpx_codec_cx_pkt *pkt) {
+  if (list->cnt < list->max) {
+    list->pkts[list->cnt++] = *pkt;
+    return 0;
+  }
 
-    return 1;
+  return 1;
 }
 
 
 const vpx_codec_cx_pkt_t *vpx_codec_pkt_list_get(struct vpx_codec_pkt_list *list,
-        vpx_codec_iter_t           *iter)
-{
-    const vpx_codec_cx_pkt_t *pkt;
+                                                 vpx_codec_iter_t           *iter) {
+  const vpx_codec_cx_pkt_t *pkt;
 
-    if (!(*iter))
-    {
-        *iter = list->pkts;
-    }
+  if (!(*iter)) {
+    *iter = list->pkts;
+  }
 
-    pkt = (const void *) * iter;
+  pkt = (const void *) * iter;
 
-    if ((size_t)(pkt - list->pkts) < list->cnt)
-        *iter = pkt + 1;
-    else
-        pkt = NULL;
+  if ((size_t)(pkt - list->pkts) < list->cnt)
+    *iter = pkt + 1;
+  else
+    pkt = NULL;
 
-    return pkt;
+  return pkt;
 }
diff --git a/libvpx/vpx/src/vpx_image.c b/libvpx/vpx/src/vpx_image.c
index 336b6e2..36eda95 100644
--- a/libvpx/vpx/src/vpx_image.c
+++ b/libvpx/vpx/src/vpx_image.c
@@ -18,30 +18,26 @@
 #define align_addr(addr,align) (void*)(((size_t)(addr) + ((align) - 1)) & (size_t)-(align))
 
 /* Memalign code is copied from vpx_mem.c */
-static void *img_buf_memalign(size_t align, size_t size)
-{
-    void *addr,
-         * x = NULL;
+static void *img_buf_memalign(size_t align, size_t size) {
+  void *addr,
+       * x = NULL;
 
-    addr = malloc(size + align - 1 + ADDRESS_STORAGE_SIZE);
+  addr = malloc(size + align - 1 + ADDRESS_STORAGE_SIZE);
 
-    if (addr)
-    {
-        x = align_addr((unsigned char *)addr + ADDRESS_STORAGE_SIZE, (int)align);
-        /* save the actual malloc address */
-        ((size_t *)x)[-1] = (size_t)addr;
-    }
+  if (addr) {
+    x = align_addr((unsigned char *)addr + ADDRESS_STORAGE_SIZE, (int)align);
+    /* save the actual malloc address */
+    ((size_t *)x)[-1] = (size_t)addr;
+  }
 
-    return x;
+  return x;
 }
 
-static void img_buf_free(void *memblk)
-{
-    if (memblk)
-    {
-        void *addr = (void *)(((size_t *)memblk)[-1]);
-        free(addr);
-    }
+static void img_buf_free(void *memblk) {
+  if (memblk) {
+    void *addr = (void *)(((size_t *)memblk)[-1]);
+    free(addr);
+  }
 }
 
 static vpx_image_t *img_alloc_helper(vpx_image_t  *img,
@@ -50,41 +46,39 @@
                                      unsigned int  d_h,
                                      unsigned int  buf_align,
                                      unsigned int  stride_align,
-                                     unsigned char      *img_data)
-{
+                                     unsigned char      *img_data) {
 
-    unsigned int  h, w, s, xcs, ycs, bps;
-    int           align;
+  unsigned int  h, w, s, xcs, ycs, bps;
+  int           align;
 
-    /* Treat align==0 like align==1 */
-    if (!buf_align)
-        buf_align = 1;
+  /* Treat align==0 like align==1 */
+  if (!buf_align)
+    buf_align = 1;
 
-    /* Validate alignment (must be power of 2) */
-    if (buf_align & (buf_align - 1))
-        goto fail;
+  /* Validate alignment (must be power of 2) */
+  if (buf_align & (buf_align - 1))
+    goto fail;
 
-    /* Treat align==0 like align==1 */
-    if (!stride_align)
-        stride_align = 1;
+  /* Treat align==0 like align==1 */
+  if (!stride_align)
+    stride_align = 1;
 
-    /* Validate alignment (must be power of 2) */
-    if (stride_align & (stride_align - 1))
-        goto fail;
+  /* Validate alignment (must be power of 2) */
+  if (stride_align & (stride_align - 1))
+    goto fail;
 
-    /* Get sample size for this format */
-    switch (fmt)
-    {
+  /* Get sample size for this format */
+  switch (fmt) {
     case VPX_IMG_FMT_RGB32:
     case VPX_IMG_FMT_RGB32_LE:
     case VPX_IMG_FMT_ARGB:
     case VPX_IMG_FMT_ARGB_LE:
-        bps = 32;
-        break;
+      bps = 32;
+      break;
     case VPX_IMG_FMT_RGB24:
     case VPX_IMG_FMT_BGR24:
-        bps = 24;
-        break;
+      bps = 24;
+      break;
     case VPX_IMG_FMT_RGB565:
     case VPX_IMG_FMT_RGB565_LE:
     case VPX_IMG_FMT_RGB555:
@@ -92,108 +86,101 @@
     case VPX_IMG_FMT_UYVY:
     case VPX_IMG_FMT_YUY2:
     case VPX_IMG_FMT_YVYU:
-        bps = 16;
-        break;
+      bps = 16;
+      break;
     case VPX_IMG_FMT_I420:
     case VPX_IMG_FMT_YV12:
     case VPX_IMG_FMT_VPXI420:
     case VPX_IMG_FMT_VPXYV12:
-        bps = 12;
-        break;
+      bps = 12;
+      break;
     default:
-        bps = 16;
-        break;
-    }
+      bps = 16;
+      break;
+  }
 
-    /* Get chroma shift values for this format */
-    switch (fmt)
-    {
+  /* Get chroma shift values for this format */
+  switch (fmt) {
     case VPX_IMG_FMT_I420:
     case VPX_IMG_FMT_YV12:
     case VPX_IMG_FMT_VPXI420:
     case VPX_IMG_FMT_VPXYV12:
-        xcs = 1;
-        break;
+      xcs = 1;
+      break;
     default:
-        xcs = 0;
-        break;
-    }
+      xcs = 0;
+      break;
+  }
 
-    switch (fmt)
-    {
+  switch (fmt) {
     case VPX_IMG_FMT_I420:
     case VPX_IMG_FMT_YV12:
     case VPX_IMG_FMT_VPXI420:
     case VPX_IMG_FMT_VPXYV12:
-        ycs = 1;
-        break;
+      ycs = 1;
+      break;
     default:
-        ycs = 0;
-        break;
-    }
+      ycs = 0;
+      break;
+  }
 
-    /* Calculate storage sizes given the chroma subsampling */
-    align = (1 << xcs) - 1;
-    w = (d_w + align) & ~align;
-    align = (1 << ycs) - 1;
-    h = (d_h + align) & ~align;
-    s = (fmt & VPX_IMG_FMT_PLANAR) ? w : bps * w / 8;
-    s = (s + stride_align - 1) & ~(stride_align - 1);
+  /* Calculate storage sizes given the chroma subsampling */
+  align = (1 << xcs) - 1;
+  w = (d_w + align) & ~align;
+  align = (1 << ycs) - 1;
+  h = (d_h + align) & ~align;
+  s = (fmt & VPX_IMG_FMT_PLANAR) ? w : bps * w / 8;
+  s = (s + stride_align - 1) & ~(stride_align - 1);
 
-    /* Allocate the new image */
+  /* Allocate the new image */
+  if (!img) {
+    img = (vpx_image_t *)calloc(1, sizeof(vpx_image_t));
+
     if (!img)
-    {
-        img = (vpx_image_t *)calloc(1, sizeof(vpx_image_t));
+      goto fail;
 
-        if (!img)
-            goto fail;
+    img->self_allocd = 1;
+  } else {
+    memset(img, 0, sizeof(vpx_image_t));
+  }
 
-        img->self_allocd = 1;
-    }
-    else
-    {
-        memset(img, 0, sizeof(vpx_image_t));
-    }
+  img->img_data = img_data;
 
-    img->img_data = img_data;
+  if (!img_data) {
+    img->img_data = img_buf_memalign(buf_align, ((fmt & VPX_IMG_FMT_PLANAR) ?
+                                                 h * s * bps / 8 : h * s));
+    img->img_data_owner = 1;
+  }
 
-    if (!img_data)
-    {
-        img->img_data = img_buf_memalign(buf_align, ((fmt & VPX_IMG_FMT_PLANAR)?
-                                         h * s * bps / 8 : h * s));
-        img->img_data_owner = 1;
-    }
+  if (!img->img_data)
+    goto fail;
 
-    if (!img->img_data)
-        goto fail;
+  img->fmt = fmt;
+  img->w = w;
+  img->h = h;
+  img->x_chroma_shift = xcs;
+  img->y_chroma_shift = ycs;
+  img->bps = bps;
 
-    img->fmt = fmt;
-    img->w = w;
-    img->h = h;
-    img->x_chroma_shift = xcs;
-    img->y_chroma_shift = ycs;
-    img->bps = bps;
+  /* Calculate strides */
+  img->stride[VPX_PLANE_Y] = img->stride[VPX_PLANE_ALPHA] = s;
+  img->stride[VPX_PLANE_U] = img->stride[VPX_PLANE_V] = s >> xcs;
 
-    /* Calculate strides */
-    img->stride[VPX_PLANE_Y] = img->stride[VPX_PLANE_ALPHA] = s;
-    img->stride[VPX_PLANE_U] = img->stride[VPX_PLANE_V] = s >> xcs;
-
-    /* Default viewport to entire image */
-    if (!vpx_img_set_rect(img, 0, 0, d_w, d_h))
-        return img;
+  /* Default viewport to entire image */
+  if (!vpx_img_set_rect(img, 0, 0, d_w, d_h))
+    return img;
 
 fail:
-    vpx_img_free(img);
-    return NULL;
+  vpx_img_free(img);
+  return NULL;
 }
 
 vpx_image_t *vpx_img_alloc(vpx_image_t  *img,
                            vpx_img_fmt_t fmt,
                            unsigned int  d_w,
                            unsigned int  d_h,
-                           unsigned int  align)
-{
-    return img_alloc_helper(img, fmt, d_w, d_h, align, align, NULL);
+                           unsigned int  align) {
+  return img_alloc_helper(img, fmt, d_w, d_h, align, align, NULL);
 }
 
 vpx_image_t *vpx_img_wrap(vpx_image_t  *img,
@@ -201,105 +188,92 @@
                           unsigned int  d_w,
                           unsigned int  d_h,
                           unsigned int  stride_align,
-                          unsigned char       *img_data)
-{
-    /* By setting buf_align = 1, we don't change buffer alignment in this
-     * function. */
-    return img_alloc_helper(img, fmt, d_w, d_h, 1, stride_align, img_data);
+                          unsigned char       *img_data) {
+  /* By setting buf_align = 1, we don't change buffer alignment in this
+   * function. */
+  return img_alloc_helper(img, fmt, d_w, d_h, 1, stride_align, img_data);
 }
 
 int vpx_img_set_rect(vpx_image_t  *img,
                      unsigned int  x,
                      unsigned int  y,
                      unsigned int  w,
-                     unsigned int  h)
-{
-    unsigned char      *data;
+                     unsigned int  h) {
+  unsigned char      *data;
 
-    if (x + w <= img->w && y + h <= img->h)
-    {
-        img->d_w = w;
-        img->d_h = h;
+  if (x + w <= img->w && y + h <= img->h) {
+    img->d_w = w;
+    img->d_h = h;
 
-        /* Calculate plane pointers */
-        if (!(img->fmt & VPX_IMG_FMT_PLANAR))
-        {
-            img->planes[VPX_PLANE_PACKED] =
-                img->img_data + x * img->bps / 8 + y * img->stride[VPX_PLANE_PACKED];
-        }
-        else
-        {
-            data = img->img_data;
+    /* Calculate plane pointers */
+    if (!(img->fmt & VPX_IMG_FMT_PLANAR)) {
+      img->planes[VPX_PLANE_PACKED] =
+        img->img_data + x * img->bps / 8 + y * img->stride[VPX_PLANE_PACKED];
+    } else {
+      data = img->img_data;
 
-            if (img->fmt & VPX_IMG_FMT_HAS_ALPHA)
-            {
-                img->planes[VPX_PLANE_ALPHA] =
-                    data + x + y * img->stride[VPX_PLANE_ALPHA];
-                data += img->h * img->stride[VPX_PLANE_ALPHA];
-            }
+      if (img->fmt & VPX_IMG_FMT_HAS_ALPHA) {
+        img->planes[VPX_PLANE_ALPHA] =
+          data + x + y * img->stride[VPX_PLANE_ALPHA];
+        data += img->h * img->stride[VPX_PLANE_ALPHA];
+      }
 
-            img->planes[VPX_PLANE_Y] = data + x + y * img->stride[VPX_PLANE_Y];
-            data += img->h * img->stride[VPX_PLANE_Y];
+      img->planes[VPX_PLANE_Y] = data + x + y * img->stride[VPX_PLANE_Y];
+      data += img->h * img->stride[VPX_PLANE_Y];
 
-            if (!(img->fmt & VPX_IMG_FMT_UV_FLIP))
-            {
-                img->planes[VPX_PLANE_U] = data
-                                       + (x >> img->x_chroma_shift)
-                                       + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
-                data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
-                img->planes[VPX_PLANE_V] = data
-                                       + (x >> img->x_chroma_shift)
-                                       + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
-            }
-            else
-            {
-                img->planes[VPX_PLANE_V] = data
-                                       + (x >> img->x_chroma_shift)
-                                       + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
-                data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
-                img->planes[VPX_PLANE_U] = data
-                                       + (x >> img->x_chroma_shift)
-                                       + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
-            }
-        }
-
-        return 0;
+      if (!(img->fmt & VPX_IMG_FMT_UV_FLIP)) {
+        img->planes[VPX_PLANE_U] = data
+                                   + (x >> img->x_chroma_shift)
+                                   + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
+        data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
+        img->planes[VPX_PLANE_V] = data
+                                   + (x >> img->x_chroma_shift)
+                                   + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
+      } else {
+        img->planes[VPX_PLANE_V] = data
+                                   + (x >> img->x_chroma_shift)
+                                   + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
+        data += (img->h >> img->y_chroma_shift) * img->stride[VPX_PLANE_V];
+        img->planes[VPX_PLANE_U] = data
+                                   + (x >> img->x_chroma_shift)
+                                   + (y >> img->y_chroma_shift) * img->stride[VPX_PLANE_U];
+      }
     }
 
-    return -1;
+    return 0;
+  }
+
+  return -1;
 }
 
-void vpx_img_flip(vpx_image_t *img)
-{
-    /* Note: In the calculation pointer adjustment calculation, we want the
-     * rhs to be promoted to a signed type. Section 6.3.1.8 of the ISO C99
-     * standard indicates that if the adjustment parameter is unsigned, the
-     * stride parameter will be promoted to unsigned, causing errors when
-     * the lhs is a larger type than the rhs.
-     */
-    img->planes[VPX_PLANE_Y] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_Y];
-    img->stride[VPX_PLANE_Y] = -img->stride[VPX_PLANE_Y];
+void vpx_img_flip(vpx_image_t *img) {
+  /* Note: In the calculation pointer adjustment calculation, we want the
+   * rhs to be promoted to a signed type. Section 6.3.1.8 of the ISO C99
+   * standard indicates that if the adjustment parameter is unsigned, the
+   * stride parameter will be promoted to unsigned, causing errors when
+   * the lhs is a larger type than the rhs.
+   */
+  img->planes[VPX_PLANE_Y] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_Y];
+  img->stride[VPX_PLANE_Y] = -img->stride[VPX_PLANE_Y];
 
-    img->planes[VPX_PLANE_U] += (signed)((img->d_h >> img->y_chroma_shift) - 1)
-                            * img->stride[VPX_PLANE_U];
-    img->stride[VPX_PLANE_U] = -img->stride[VPX_PLANE_U];
+  img->planes[VPX_PLANE_U] += (signed)((img->d_h >> img->y_chroma_shift) - 1)
+                              * img->stride[VPX_PLANE_U];
+  img->stride[VPX_PLANE_U] = -img->stride[VPX_PLANE_U];
 
-    img->planes[VPX_PLANE_V] += (signed)((img->d_h >> img->y_chroma_shift) - 1)
-                            * img->stride[VPX_PLANE_V];
-    img->stride[VPX_PLANE_V] = -img->stride[VPX_PLANE_V];
+  img->planes[VPX_PLANE_V] += (signed)((img->d_h >> img->y_chroma_shift) - 1)
+                              * img->stride[VPX_PLANE_V];
+  img->stride[VPX_PLANE_V] = -img->stride[VPX_PLANE_V];
 
-    img->planes[VPX_PLANE_ALPHA] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_ALPHA];
-    img->stride[VPX_PLANE_ALPHA] = -img->stride[VPX_PLANE_ALPHA];
+  img->planes[VPX_PLANE_ALPHA] += (signed)(img->d_h - 1) * img->stride[VPX_PLANE_ALPHA];
+  img->stride[VPX_PLANE_ALPHA] = -img->stride[VPX_PLANE_ALPHA];
 }
 
-void vpx_img_free(vpx_image_t *img)
-{
-    if (img)
-    {
-        if (img->img_data && img->img_data_owner)
-            img_buf_free(img->img_data);
+void vpx_img_free(vpx_image_t *img) {
+  if (img) {
+    if (img->img_data && img->img_data_owner)
+      img_buf_free(img->img_data);
 
-        if (img->self_allocd)
-            free(img);
-    }
+    if (img->self_allocd)
+      free(img);
+  }
 }
diff --git a/libvpx/vpx/vp8.h b/libvpx/vpx/vp8.h
index 2952203..0b4cb1b 100644
--- a/libvpx/vpx/vp8.h
+++ b/libvpx/vpx/vp8.h
@@ -36,34 +36,38 @@
  *
  * The set of macros define the control functions of VP8 interface
  */
-enum vp8_com_control_id
-{
-    VP8_SET_REFERENCE           = 1,    /**< pass in an external frame into decoder to be used as reference frame */
-    VP8_COPY_REFERENCE          = 2,    /**< get a copy of reference frame from the decoder */
-    VP8_SET_POSTPROC            = 3,    /**< set the decoder's post processing settings  */
-    VP8_SET_DBG_COLOR_REF_FRAME = 4,    /**< set the reference frames to color for each macroblock */
-    VP8_SET_DBG_COLOR_MB_MODES  = 5,    /**< set which macro block modes to color */
-    VP8_SET_DBG_COLOR_B_MODES   = 6,    /**< set which blocks modes to color */
-    VP8_SET_DBG_DISPLAY_MV      = 7,    /**< set which motion vector modes to draw */
-    VP8_COMMON_CTRL_ID_MAX,
-    VP8_DECODER_CTRL_ID_START   = 256
+enum vp8_com_control_id {
+  VP8_SET_REFERENCE           = 1,    /**< pass in an external frame into decoder to be used as reference frame */
+  VP8_COPY_REFERENCE          = 2,    /**< get a copy of reference frame from the decoder */
+  VP8_SET_POSTPROC            = 3,    /**< set the decoder's post processing settings  */
+  VP8_SET_DBG_COLOR_REF_FRAME = 4,    /**< set the reference frames to color for each macroblock */
+  VP8_SET_DBG_COLOR_MB_MODES  = 5,    /**< set which macro block modes to color */
+  VP8_SET_DBG_COLOR_B_MODES   = 6,    /**< set which blocks modes to color */
+  VP8_SET_DBG_DISPLAY_MV      = 7,    /**< set which motion vector modes to draw */
+
+  /* TODO(jkoleszar): The encoder incorrectly reuses some of these values (5+)
+   * for its control ids. These should be migrated to something like the
+   * VP8_DECODER_CTRL_ID_START range next time we're ready to break the ABI.
+   */
+  VP9_GET_REFERENCE           = 128,  /**< get a pointer to a reference frame */
+  VP8_COMMON_CTRL_ID_MAX,
+  VP8_DECODER_CTRL_ID_START   = 256
 };
 
 /*!\brief post process flags
  *
  * The set of macros define VP8 decoder post processing flags
  */
-enum vp8_postproc_level
-{
-    VP8_NOFILTERING             = 0,
-    VP8_DEBLOCK                 = 1<<0,
-    VP8_DEMACROBLOCK            = 1<<1,
-    VP8_ADDNOISE                = 1<<2,
-    VP8_DEBUG_TXT_FRAME_INFO    = 1<<3, /**< print frame information */
-    VP8_DEBUG_TXT_MBLK_MODES    = 1<<4, /**< print macro block modes over each macro block */
-    VP8_DEBUG_TXT_DC_DIFF       = 1<<5, /**< print dc diff for each macro block */
-    VP8_DEBUG_TXT_RATE_INFO     = 1<<6, /**< print video rate info (encoder only) */
-    VP8_MFQE                    = 1<<10
+enum vp8_postproc_level {
+  VP8_NOFILTERING             = 0,
+  VP8_DEBLOCK                 = 1 << 0,
+  VP8_DEMACROBLOCK            = 1 << 1,
+  VP8_ADDNOISE                = 1 << 2,
+  VP8_DEBUG_TXT_FRAME_INFO    = 1 << 3, /**< print frame information */
+  VP8_DEBUG_TXT_MBLK_MODES    = 1 << 4, /**< print macro block modes over each macro block */
+  VP8_DEBUG_TXT_DC_DIFF       = 1 << 5, /**< print dc diff for each macro block */
+  VP8_DEBUG_TXT_RATE_INFO     = 1 << 6, /**< print video rate info (encoder only) */
+  VP8_MFQE                    = 1 << 10
 };
 
 /*!\brief post process flags
@@ -73,22 +77,20 @@
  * to VP8_DEBLOCK and deblocking_level to 1.
  */
 
-typedef struct vp8_postproc_cfg
-{
-    int post_proc_flag;         /**< the types of post processing to be done, should be combination of "vp8_postproc_level" */
-    int deblocking_level;       /**< the strength of deblocking, valid range [0, 16] */
-    int noise_level;            /**< the strength of additive noise, valid range [0, 16] */
+typedef struct vp8_postproc_cfg {
+  int post_proc_flag;         /**< the types of post processing to be done, should be combination of "vp8_postproc_level" */
+  int deblocking_level;       /**< the strength of deblocking, valid range [0, 16] */
+  int noise_level;            /**< the strength of additive noise, valid range [0, 16] */
 } vp8_postproc_cfg_t;
 
 /*!\brief reference frame type
  *
  * The set of macros define the type of VP8 reference frames
  */
-typedef enum vpx_ref_frame_type
-{
-    VP8_LAST_FRAME = 1,
-    VP8_GOLD_FRAME = 2,
-    VP8_ALTR_FRAME = 4
+typedef enum vpx_ref_frame_type {
+  VP8_LAST_FRAME = 1,
+  VP8_GOLD_FRAME = 2,
+  VP8_ALTR_FRAME = 4
 } vpx_ref_frame_type_t;
 
 /*!\brief reference frame data struct
@@ -96,12 +98,15 @@
  * define the data struct to access vp8 reference frames
  */
 
-typedef struct vpx_ref_frame
-{
-    vpx_ref_frame_type_t  frame_type;   /**< which reference frame */
-    vpx_image_t           img;          /**< reference frame data in image format */
+typedef struct vpx_ref_frame {
+  vpx_ref_frame_type_t  frame_type;   /**< which reference frame */
+  vpx_image_t           img;          /**< reference frame data in image format */
 } vpx_ref_frame_t;
 
+typedef struct vp9_ref_frame {
+  int idx; /**< frame index to get (input) */
+  vpx_image_t  img; /**< img structure to populate (output) */
+} vp9_ref_frame_t;
 
 /*!\brief vp8 decoder control function parameter type
  *
@@ -115,6 +120,7 @@
 VPX_CTRL_USE_TYPE(VP8_SET_DBG_COLOR_MB_MODES,  int)
 VPX_CTRL_USE_TYPE(VP8_SET_DBG_COLOR_B_MODES,   int)
 VPX_CTRL_USE_TYPE(VP8_SET_DBG_DISPLAY_MV,      int)
+VPX_CTRL_USE_TYPE(VP9_GET_REFERENCE,           vp9_ref_frame_t *)
 
 
 /*! @} - end defgroup vp8 */
diff --git a/libvpx/vpx/vp8cx.h b/libvpx/vpx/vp8cx.h
index a3c95d2..f8e2ef9 100644
--- a/libvpx/vpx/vp8cx.h
+++ b/libvpx/vpx/vp8cx.h
@@ -31,7 +31,14 @@
  * @{
  */
 extern vpx_codec_iface_t  vpx_codec_vp8_cx_algo;
-extern vpx_codec_iface_t* vpx_codec_vp8_cx(void);
+extern vpx_codec_iface_t *vpx_codec_vp8_cx(void);
+
+/* TODO(jkoleszar): These move to VP9 in a later patch set. */
+extern vpx_codec_iface_t  vpx_codec_vp9_cx_algo;
+extern vpx_codec_iface_t *vpx_codec_vp9_cx(void);
+extern vpx_codec_iface_t  vpx_codec_vp9x_cx_algo;
+extern vpx_codec_iface_t *vpx_codec_vp9x_cx(void);
+
 /*!@} - end algorithm interface member group*/
 
 
@@ -121,75 +128,80 @@
  *
  * \sa #vpx_codec_control
  */
-enum vp8e_enc_control_id
-{
-    VP8E_UPD_ENTROPY           = 5,  /**< control function to set mode of entropy update in encoder */
-    VP8E_UPD_REFERENCE,              /**< control function to set reference update mode in encoder */
-    VP8E_USE_REFERENCE,              /**< control function to set which reference frame encoder can use */
-    VP8E_SET_ROI_MAP,                /**< control function to pass an ROI map to encoder */
-    VP8E_SET_ACTIVEMAP,              /**< control function to pass an Active map to encoder */
-    VP8E_SET_SCALEMODE         = 11, /**< control function to set encoder scaling mode */
-    /*!\brief control function to set vp8 encoder cpuused
-     *
-     * Changes in this value influences, among others, the encoder's selection
-     * of motion estimation methods. Values greater than 0 will increase encoder
-     * speed at the expense of quality.
-     * The full set of adjustments can be found in
-     * onyx_if.c:vp8_set_speed_features().
-     * \todo List highlights of the changes at various levels.
-     *
-     * \note Valid range: -16..16
-     */
-    VP8E_SET_CPUUSED           = 13,
-    VP8E_SET_ENABLEAUTOALTREF,       /**< control function to enable vp8 to automatic set and use altref frame */
-    VP8E_SET_NOISE_SENSITIVITY,      /**< control function to set noise sensitivity */
-    VP8E_SET_SHARPNESS,              /**< control function to set sharpness */
-    VP8E_SET_STATIC_THRESHOLD,       /**< control function to set the threshold for macroblocks treated static */
-    VP8E_SET_TOKEN_PARTITIONS,       /**< control function to set the number of token partitions  */
-    VP8E_GET_LAST_QUANTIZER,         /**< return the quantizer chosen by the
+enum vp8e_enc_control_id {
+  VP8E_UPD_ENTROPY           = 5,  /**< control function to set mode of entropy update in encoder */
+  VP8E_UPD_REFERENCE,              /**< control function to set reference update mode in encoder */
+  VP8E_USE_REFERENCE,              /**< control function to set which reference frame encoder can use */
+  VP8E_SET_ROI_MAP,                /**< control function to pass an ROI map to encoder */
+  VP8E_SET_ACTIVEMAP,              /**< control function to pass an Active map to encoder */
+  VP8E_SET_SCALEMODE         = 11, /**< control function to set encoder scaling mode */
+  /*!\brief control function to set vp8 encoder cpuused
+   *
+   * Changes in this value influences, among others, the encoder's selection
+   * of motion estimation methods. Values greater than 0 will increase encoder
+   * speed at the expense of quality.
+   * The full set of adjustments can be found in
+   * onyx_if.c:vp8_set_speed_features().
+   * \todo List highlights of the changes at various levels.
+   *
+   * \note Valid range: -16..16
+   */
+  VP8E_SET_CPUUSED           = 13,
+  VP8E_SET_ENABLEAUTOALTREF,       /**< control function to enable vp8 to automatic set and use altref frame */
+  VP8E_SET_NOISE_SENSITIVITY,      /**< control function to set noise sensitivity */
+  VP8E_SET_SHARPNESS,              /**< control function to set sharpness */
+  VP8E_SET_STATIC_THRESHOLD,       /**< control function to set the threshold for macroblocks treated static */
+  VP8E_SET_TOKEN_PARTITIONS,       /**< control function to set the number of token partitions  */
+  VP8E_GET_LAST_QUANTIZER,         /**< return the quantizer chosen by the
                                           encoder for the last frame using the internal
                                           scale */
-    VP8E_GET_LAST_QUANTIZER_64,      /**< return the quantizer chosen by the
+  VP8E_GET_LAST_QUANTIZER_64,      /**< return the quantizer chosen by the
                                           encoder for the last frame, using the 0..63
                                           scale as used by the rc_*_quantizer config
                                           parameters */
-    VP8E_SET_ARNR_MAXFRAMES,         /**< control function to set the max number of frames blurred creating arf*/
-    VP8E_SET_ARNR_STRENGTH ,         /**< control function to set the filter strength for the arf */
-    VP8E_SET_ARNR_TYPE     ,         /**< control function to set the type of filter to use for the arf*/
-    VP8E_SET_TUNING,                 /**< control function to set visual tuning */
-    /*!\brief control function to set constrained quality level
-     *
-     * \attention For this value to be used vpx_codec_enc_cfg_t::g_usage must be
-     *            set to #VPX_CQ.
-     * \note Valid range: 0..63
-     */
-    VP8E_SET_CQ_LEVEL,
+  VP8E_SET_ARNR_MAXFRAMES,         /**< control function to set the max number of frames blurred creating arf*/
+  VP8E_SET_ARNR_STRENGTH,         /**< control function to set the filter strength for the arf */
+  VP8E_SET_ARNR_TYPE,         /**< control function to set the type of filter to use for the arf*/
+  VP8E_SET_TUNING,                 /**< control function to set visual tuning */
+  /*!\brief control function to set constrained quality level
+   *
+   * \attention For this value to be used vpx_codec_enc_cfg_t::g_usage must be
+   *            set to #VPX_CQ.
+   * \note Valid range: 0..63
+   */
+  VP8E_SET_CQ_LEVEL,
 
-    /*!\brief Max data rate for Intra frames
-     *
-     * This value controls additional clamping on the maximum size of a
-     * keyframe. It is expressed as a percentage of the average
-     * per-frame bitrate, with the special (and default) value 0 meaning
-     * unlimited, or no additional clamping beyond the codec's built-in
-     * algorithm.
-     *
-     * For example, to allocate no more than 4.5 frames worth of bitrate
-     * to a keyframe, set this to 450.
-     *
-     */
-    VP8E_SET_MAX_INTRA_BITRATE_PCT
+  /*!\brief Max data rate for Intra frames
+   *
+   * This value controls additional clamping on the maximum size of a
+   * keyframe. It is expressed as a percentage of the average
+   * per-frame bitrate, with the special (and default) value 0 meaning
+   * unlimited, or no additional clamping beyond the codec's built-in
+   * algorithm.
+   *
+   * For example, to allocate no more than 4.5 frames worth of bitrate
+   * to a keyframe, set this to 450.
+   *
+   */
+  VP8E_SET_MAX_INTRA_BITRATE_PCT,
+
+
+  /* TODO(jkoleszar): Move to vp9cx.h */
+  VP9E_SET_LOSSLESS,
+  VP9E_SET_TILE_COLUMNS,
+  VP9E_SET_TILE_ROWS,
+  VP9E_SET_FRAME_PARALLEL_DECODING
 };
 
 /*!\brief vpx 1-D scaling mode
  *
  * This set of constants define 1-D vpx scaling modes
  */
-typedef enum vpx_scaling_mode_1d
-{
-    VP8E_NORMAL      = 0,
-    VP8E_FOURFIVE    = 1,
-    VP8E_THREEFIVE   = 2,
-    VP8E_ONETWO      = 3
+typedef enum vpx_scaling_mode_1d {
+  VP8E_NORMAL      = 0,
+  VP8E_FOURFIVE    = 1,
+  VP8E_THREEFIVE   = 2,
+  VP8E_ONETWO      = 3
 } VPX_SCALING_MODE;
 
 
@@ -199,14 +211,17 @@
  *
  */
 
-typedef struct vpx_roi_map
-{
-    unsigned char *roi_map;      /**< specify an id between 0 and 3 for each 16x16 region within a frame */
-    unsigned int   rows;         /**< number of rows */
-    unsigned int   cols;         /**< number of cols */
-    int     delta_q[4];          /**< quantizer delta [-63, 63] off baseline for regions with id between 0 and 3*/
-    int     delta_lf[4];         /**< loop filter strength delta [-63, 63] for regions with id between 0 and 3 */
-    unsigned int   static_threshold[4];/**< threshold for region to be treated as static */
+typedef struct vpx_roi_map {
+  unsigned char *roi_map;      /**< specify an id between 0 and 3 for each 16x16 region within a frame */
+  unsigned int   rows;         /**< number of rows */
+  unsigned int   cols;         /**< number of cols */
+  // TODO(paulwilkins): broken for VP9 which has 8 segments
+  // q and loop filter deltas for each segment
+  // (see MAX_MB_SEGMENTS)
+  int     delta_q[4];
+  int     delta_lf[4];
+  // Static breakout threshold for each segment
+  unsigned int   static_threshold[4];
 } vpx_roi_map_t;
 
 /*!\brief  vpx active region map
@@ -216,11 +231,10 @@
  */
 
 
-typedef struct vpx_active_map
-{
-    unsigned char  *active_map; /**< specify an on (1) or off (0) each 16x16 region within a frame */
-    unsigned int    rows;       /**< number of rows */
-    unsigned int    cols;       /**< number of cols */
+typedef struct vpx_active_map {
+  unsigned char  *active_map; /**< specify an on (1) or off (0) each 16x16 region within a frame */
+  unsigned int    rows;       /**< number of rows */
+  unsigned int    cols;       /**< number of cols */
 } vpx_active_map_t;
 
 /*!\brief  vpx image scaling mode
@@ -228,10 +242,9 @@
  * This defines the data structure for image scaling mode
  *
  */
-typedef struct vpx_scaling_mode
-{
-    VPX_SCALING_MODE    h_scaling_mode;  /**< horizontal scaling mode */
-    VPX_SCALING_MODE    v_scaling_mode;  /**< vertical scaling mode   */
+typedef struct vpx_scaling_mode {
+  VPX_SCALING_MODE    h_scaling_mode;  /**< horizontal scaling mode */
+  VPX_SCALING_MODE    v_scaling_mode;  /**< vertical scaling mode   */
 } vpx_scaling_mode_t;
 
 /*!\brief VP8 token partition mode
@@ -241,12 +254,11 @@
  *
  */
 
-typedef enum
-{
-    VP8_ONE_TOKENPARTITION   = 0,
-    VP8_TWO_TOKENPARTITION   = 1,
-    VP8_FOUR_TOKENPARTITION  = 2,
-    VP8_EIGHT_TOKENPARTITION = 3
+typedef enum {
+  VP8_ONE_TOKENPARTITION   = 0,
+  VP8_TWO_TOKENPARTITION   = 1,
+  VP8_FOUR_TOKENPARTITION  = 2,
+  VP8_EIGHT_TOKENPARTITION = 3
 } vp8e_token_partitions;
 
 
@@ -255,10 +267,9 @@
  * Changes the encoder to tune for certain types of input material.
  *
  */
-typedef enum
-{
-    VP8_TUNE_PSNR,
-    VP8_TUNE_SSIM
+typedef enum {
+  VP8_TUNE_PSNR,
+  VP8_TUNE_SSIM
 } vp8e_tuning;
 
 
@@ -289,17 +300,22 @@
 VPX_CTRL_USE_TYPE(VP8E_SET_TOKEN_PARTITIONS,   int) /* vp8e_token_partitions */
 
 VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_MAXFRAMES,     unsigned int)
-VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_STRENGTH ,     unsigned int)
-VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_TYPE     ,     unsigned int)
+VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_STRENGTH,     unsigned int)
+VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_TYPE,     unsigned int)
 VPX_CTRL_USE_TYPE(VP8E_SET_TUNING,             int) /* vp8e_tuning */
-VPX_CTRL_USE_TYPE(VP8E_SET_CQ_LEVEL     ,      unsigned int)
+VPX_CTRL_USE_TYPE(VP8E_SET_CQ_LEVEL,      unsigned int)
+
+VPX_CTRL_USE_TYPE(VP9E_SET_TILE_COLUMNS,  int)
+VPX_CTRL_USE_TYPE(VP9E_SET_TILE_ROWS,  int)
 
 VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER,     int *)
 VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER_64,  int *)
 
 VPX_CTRL_USE_TYPE(VP8E_SET_MAX_INTRA_BITRATE_PCT, unsigned int)
 
+VPX_CTRL_USE_TYPE(VP9E_SET_LOSSLESS, unsigned int)
 
+VPX_CTRL_USE_TYPE(VP9E_SET_FRAME_PARALLEL_DECODING, unsigned int)
 /*! @} - end defgroup vp8_encoder */
 #include "vpx_codec_impl_bottom.h"
 #endif
diff --git a/libvpx/vpx/vp8dx.h b/libvpx/vpx/vp8dx.h
index 8661035..7d250cc 100644
--- a/libvpx/vpx/vp8dx.h
+++ b/libvpx/vpx/vp8dx.h
@@ -31,7 +31,11 @@
  * @{
  */
 extern vpx_codec_iface_t  vpx_codec_vp8_dx_algo;
-extern vpx_codec_iface_t* vpx_codec_vp8_dx(void);
+extern vpx_codec_iface_t *vpx_codec_vp8_dx(void);
+
+/* TODO(jkoleszar): These move to VP9 in a later patch set. */
+extern vpx_codec_iface_t  vpx_codec_vp9_dx_algo;
+extern vpx_codec_iface_t *vpx_codec_vp9_dx(void);
 /*!@} - end algorithm interface member group*/
 
 /* Include controls common to both the encoder and decoder */
@@ -45,24 +49,40 @@
  *
  * \sa #vpx_codec_control
  */
-enum vp8_dec_control_id
-{
-    /** control function to get info on which reference frames were updated
-     *  by the last decode
+enum vp8_dec_control_id {
+  /** control function to get info on which reference frames were updated
+   *  by the last decode
+   */
+  VP8D_GET_LAST_REF_UPDATES = VP8_DECODER_CTRL_ID_START,
+
+  /** check if the indicated frame is corrupted */
+  VP8D_GET_FRAME_CORRUPTED,
+
+  /** control function to get info on which reference frames were used
+   *  by the last decode
+   */
+  VP8D_GET_LAST_REF_USED,
+
+  /** decryption function to decrypt encoded buffer data immediately
+   * before decoding. Takes a vp8_decrypt_init, which contains
+   * a callback function and opaque context pointer.
+   */
+  VP8D_SET_DECRYPTOR,
+
+  /** For testing. */
+  VP9_INVERT_TILE_DECODE_ORDER,
+
+  VP8_DECODER_CTRL_ID_MAX
+};
+
+typedef struct vp8_decrypt_init {
+    /** Decrypt n bytes of data from input -> output, using the decrypt_state
+     *  passed in VP8D_SET_DECRYPTOR.
      */
-    VP8D_GET_LAST_REF_UPDATES = VP8_DECODER_CTRL_ID_START,
-
-    /** check if the indicated frame is corrupted */
-    VP8D_GET_FRAME_CORRUPTED,
-
-    /** control function to get info on which reference frames were used
-     *  by the last decode
-     */
-    VP8D_GET_LAST_REF_USED,
-
-    VP8_DECODER_CTRL_ID_MAX
-} ;
-
+    void (*decrypt_cb)(void *decrypt_state, const unsigned char *input,
+                       unsigned char *output, int count);
+    void *decrypt_state;
+} vp8_decrypt_init;
 
 /*!\brief VP8 decoder control function parameter type
  *
@@ -75,6 +95,8 @@
 VPX_CTRL_USE_TYPE(VP8D_GET_LAST_REF_UPDATES,   int *)
 VPX_CTRL_USE_TYPE(VP8D_GET_FRAME_CORRUPTED,    int *)
 VPX_CTRL_USE_TYPE(VP8D_GET_LAST_REF_USED,      int *)
+VPX_CTRL_USE_TYPE(VP8D_SET_DECRYPTOR,          vp8_decrypt_init *)
+VPX_CTRL_USE_TYPE(VP9_INVERT_TILE_DECODE_ORDER, int)
 
 /*! @} - end defgroup vp8_decoder */
 
diff --git a/libvpx/vpx/vpx_codec.h b/libvpx/vpx/vpx_codec.h
index d92e165..2e6f1e7 100644
--- a/libvpx/vpx/vpx_codec.h
+++ b/libvpx/vpx/vpx_codec.h
@@ -45,21 +45,28 @@
 #include "vpx_integer.h"
 #include "vpx_image.h"
 
-    /*!\brief Decorator indicating a function is deprecated */
+  /*!\brief Decorator indicating a function is deprecated */
 #ifndef DEPRECATED
 #if defined(__GNUC__) && __GNUC__
 #define DEPRECATED          __attribute__ ((deprecated))
-#define DECLSPEC_DEPRECATED /**< \copydoc #DEPRECATED */
 #elif defined(_MSC_VER)
 #define DEPRECATED
-#define DECLSPEC_DEPRECATED __declspec(deprecated) /**< \copydoc #DEPRECATED */
 #else
 #define DEPRECATED
+#endif
+#endif  /* DEPRECATED */
+
+#ifndef DECLSPEC_DEPRECATED
+#if defined(__GNUC__) && __GNUC__
+#define DECLSPEC_DEPRECATED /**< \copydoc #DEPRECATED */
+#elif defined(_MSC_VER)
+#define DECLSPEC_DEPRECATED __declspec(deprecated) /**< \copydoc #DEPRECATED */
+#else
 #define DECLSPEC_DEPRECATED /**< \copydoc #DEPRECATED */
 #endif
-#endif
+#endif  /* DECLSPEC_DEPRECATED */
 
-    /*!\brief Decorator indicating a function is potentially unused */
+  /*!\brief Decorator indicating a function is potentially unused */
 #ifdef UNUSED
 #elif __GNUC__
 #define UNUSED __attribute__ ((unused))
@@ -67,312 +74,310 @@
 #define UNUSED
 #endif
 
-    /*!\brief Current ABI version number
-     *
-     * \internal
-     * If this file is altered in any way that changes the ABI, this value
-     * must be bumped.  Examples include, but are not limited to, changing
-     * types, removing or reassigning enums, adding/removing/rearranging
-     * fields to structures
-     */
+  /*!\brief Current ABI version number
+   *
+   * \internal
+   * If this file is altered in any way that changes the ABI, this value
+   * must be bumped.  Examples include, but are not limited to, changing
+   * types, removing or reassigning enums, adding/removing/rearranging
+   * fields to structures
+   */
 #define VPX_CODEC_ABI_VERSION (2 + VPX_IMAGE_ABI_VERSION) /**<\hideinitializer*/
 
-    /*!\brief Algorithm return codes */
-    typedef enum {
-        /*!\brief Operation completed without error */
-        VPX_CODEC_OK,
+  /*!\brief Algorithm return codes */
+  typedef enum {
+    /*!\brief Operation completed without error */
+    VPX_CODEC_OK,
 
-        /*!\brief Unspecified error */
-        VPX_CODEC_ERROR,
+    /*!\brief Unspecified error */
+    VPX_CODEC_ERROR,
 
-        /*!\brief Memory operation failed */
-        VPX_CODEC_MEM_ERROR,
+    /*!\brief Memory operation failed */
+    VPX_CODEC_MEM_ERROR,
 
-        /*!\brief ABI version mismatch */
-        VPX_CODEC_ABI_MISMATCH,
+    /*!\brief ABI version mismatch */
+    VPX_CODEC_ABI_MISMATCH,
 
-        /*!\brief Algorithm does not have required capability */
-        VPX_CODEC_INCAPABLE,
+    /*!\brief Algorithm does not have required capability */
+    VPX_CODEC_INCAPABLE,
 
-        /*!\brief The given bitstream is not supported.
-         *
-         * The bitstream was unable to be parsed at the highest level. The decoder
-         * is unable to proceed. This error \ref SHOULD be treated as fatal to the
-         * stream. */
-        VPX_CODEC_UNSUP_BITSTREAM,
-
-        /*!\brief Encoded bitstream uses an unsupported feature
-         *
-         * The decoder does not implement a feature required by the encoder. This
-         * return code should only be used for features that prevent future
-         * pictures from being properly decoded. This error \ref MAY be treated as
-         * fatal to the stream or \ref MAY be treated as fatal to the current GOP.
-         */
-        VPX_CODEC_UNSUP_FEATURE,
-
-        /*!\brief The coded data for this stream is corrupt or incomplete
-         *
-         * There was a problem decoding the current frame.  This return code
-         * should only be used for failures that prevent future pictures from
-         * being properly decoded. This error \ref MAY be treated as fatal to the
-         * stream or \ref MAY be treated as fatal to the current GOP. If decoding
-         * is continued for the current GOP, artifacts may be present.
-         */
-        VPX_CODEC_CORRUPT_FRAME,
-
-        /*!\brief An application-supplied parameter is not valid.
-         *
-         */
-        VPX_CODEC_INVALID_PARAM,
-
-        /*!\brief An iterator reached the end of list.
-         *
-         */
-        VPX_CODEC_LIST_END
-
-    }
-    vpx_codec_err_t;
-
-
-    /*! \brief Codec capabilities bitfield
+    /*!\brief The given bitstream is not supported.
      *
-     *  Each codec advertises the capabilities it supports as part of its
-     *  ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces
-     *  or functionality, and are not required to be supported.
+     * The bitstream was unable to be parsed at the highest level. The decoder
+     * is unable to proceed. This error \ref SHOULD be treated as fatal to the
+     * stream. */
+    VPX_CODEC_UNSUP_BITSTREAM,
+
+    /*!\brief Encoded bitstream uses an unsupported feature
      *
-     *  The available flags are specified by VPX_CODEC_CAP_* defines.
+     * The decoder does not implement a feature required by the encoder. This
+     * return code should only be used for features that prevent future
+     * pictures from being properly decoded. This error \ref MAY be treated as
+     * fatal to the stream or \ref MAY be treated as fatal to the current GOP.
      */
-    typedef long vpx_codec_caps_t;
+    VPX_CODEC_UNSUP_FEATURE,
+
+    /*!\brief The coded data for this stream is corrupt or incomplete
+     *
+     * There was a problem decoding the current frame.  This return code
+     * should only be used for failures that prevent future pictures from
+     * being properly decoded. This error \ref MAY be treated as fatal to the
+     * stream or \ref MAY be treated as fatal to the current GOP. If decoding
+     * is continued for the current GOP, artifacts may be present.
+     */
+    VPX_CODEC_CORRUPT_FRAME,
+
+    /*!\brief An application-supplied parameter is not valid.
+     *
+     */
+    VPX_CODEC_INVALID_PARAM,
+
+    /*!\brief An iterator reached the end of list.
+     *
+     */
+    VPX_CODEC_LIST_END
+
+  }
+  vpx_codec_err_t;
+
+
+  /*! \brief Codec capabilities bitfield
+   *
+   *  Each codec advertises the capabilities it supports as part of its
+   *  ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces
+   *  or functionality, and are not required to be supported.
+   *
+   *  The available flags are specified by VPX_CODEC_CAP_* defines.
+   */
+  typedef long vpx_codec_caps_t;
 #define VPX_CODEC_CAP_DECODER 0x1 /**< Is a decoder */
 #define VPX_CODEC_CAP_ENCODER 0x2 /**< Is an encoder */
 #define VPX_CODEC_CAP_XMA     0x4 /**< Supports eXternal Memory Allocation */
 
 
-    /*! \brief Initialization-time Feature Enabling
-     *
-     *  Certain codec features must be known at initialization time, to allow for
-     *  proper memory allocation.
-     *
-     *  The available flags are specified by VPX_CODEC_USE_* defines.
-     */
-    typedef long vpx_codec_flags_t;
+  /*! \brief Initialization-time Feature Enabling
+   *
+   *  Certain codec features must be known at initialization time, to allow for
+   *  proper memory allocation.
+   *
+   *  The available flags are specified by VPX_CODEC_USE_* defines.
+   */
+  typedef long vpx_codec_flags_t;
 #define VPX_CODEC_USE_XMA 0x00000001    /**< Use eXternal Memory Allocation mode */
 
 
-    /*!\brief Codec interface structure.
-     *
-     * Contains function pointers and other data private to the codec
-     * implementation. This structure is opaque to the application.
-     */
-    typedef const struct vpx_codec_iface vpx_codec_iface_t;
+  /*!\brief Codec interface structure.
+   *
+   * Contains function pointers and other data private to the codec
+   * implementation. This structure is opaque to the application.
+   */
+  typedef const struct vpx_codec_iface vpx_codec_iface_t;
 
 
-    /*!\brief Codec private data structure.
-     *
-     * Contains data private to the codec implementation. This structure is opaque
-     * to the application.
-     */
-    typedef       struct vpx_codec_priv  vpx_codec_priv_t;
+  /*!\brief Codec private data structure.
+   *
+   * Contains data private to the codec implementation. This structure is opaque
+   * to the application.
+   */
+  typedef       struct vpx_codec_priv  vpx_codec_priv_t;
 
 
-    /*!\brief Iterator
-     *
-     * Opaque storage used for iterating over lists.
-     */
-    typedef const void *vpx_codec_iter_t;
+  /*!\brief Iterator
+   *
+   * Opaque storage used for iterating over lists.
+   */
+  typedef const void *vpx_codec_iter_t;
 
 
-    /*!\brief Codec context structure
-     *
-     * All codecs \ref MUST support this context structure fully. In general,
-     * this data should be considered private to the codec algorithm, and
-     * not be manipulated or examined by the calling application. Applications
-     * may reference the 'name' member to get a printable description of the
-     * algorithm.
-     */
-    typedef struct vpx_codec_ctx
-    {
-        const char              *name;        /**< Printable interface name */
-        vpx_codec_iface_t       *iface;       /**< Interface pointers */
-        vpx_codec_err_t          err;         /**< Last returned error */
-        const char              *err_detail;  /**< Detailed info, if available */
-        vpx_codec_flags_t        init_flags;  /**< Flags passed at init time */
-        union
-        {
-            struct vpx_codec_dec_cfg  *dec;   /**< Decoder Configuration Pointer */
-            struct vpx_codec_enc_cfg  *enc;   /**< Encoder Configuration Pointer */
-            void                      *raw;
-        }                        config;      /**< Configuration pointer aliasing union */
-        vpx_codec_priv_t        *priv;        /**< Algorithm private storage */
-    } vpx_codec_ctx_t;
+  /*!\brief Codec context structure
+   *
+   * All codecs \ref MUST support this context structure fully. In general,
+   * this data should be considered private to the codec algorithm, and
+   * not be manipulated or examined by the calling application. Applications
+   * may reference the 'name' member to get a printable description of the
+   * algorithm.
+   */
+  typedef struct vpx_codec_ctx {
+    const char              *name;        /**< Printable interface name */
+    vpx_codec_iface_t       *iface;       /**< Interface pointers */
+    vpx_codec_err_t          err;         /**< Last returned error */
+    const char              *err_detail;  /**< Detailed info, if available */
+    vpx_codec_flags_t        init_flags;  /**< Flags passed at init time */
+    union {
+      struct vpx_codec_dec_cfg  *dec;   /**< Decoder Configuration Pointer */
+      struct vpx_codec_enc_cfg  *enc;   /**< Encoder Configuration Pointer */
+      void                      *raw;
+    }                        config;      /**< Configuration pointer aliasing union */
+    vpx_codec_priv_t        *priv;        /**< Algorithm private storage */
+  } vpx_codec_ctx_t;
 
 
-    /*
-     * Library Version Number Interface
-     *
-     * For example, see the following sample return values:
-     *     vpx_codec_version()           (1<<16 | 2<<8 | 3)
-     *     vpx_codec_version_str()       "v1.2.3-rc1-16-gec6a1ba"
-     *     vpx_codec_version_extra_str() "rc1-16-gec6a1ba"
-     */
+  /*
+   * Library Version Number Interface
+   *
+   * For example, see the following sample return values:
+   *     vpx_codec_version()           (1<<16 | 2<<8 | 3)
+   *     vpx_codec_version_str()       "v1.2.3-rc1-16-gec6a1ba"
+   *     vpx_codec_version_extra_str() "rc1-16-gec6a1ba"
+   */
 
-    /*!\brief Return the version information (as an integer)
-     *
-     * Returns a packed encoding of the library version number. This will only include
-     * the major.minor.patch component of the version number. Note that this encoded
-     * value should be accessed through the macros provided, as the encoding may change
-     * in the future.
-     *
-     */
-    int vpx_codec_version(void);
+  /*!\brief Return the version information (as an integer)
+   *
+   * Returns a packed encoding of the library version number. This will only include
+   * the major.minor.patch component of the version number. Note that this encoded
+   * value should be accessed through the macros provided, as the encoding may change
+   * in the future.
+   *
+   */
+  int vpx_codec_version(void);
 #define VPX_VERSION_MAJOR(v) ((v>>16)&0xff) /**< extract major from packed version */
 #define VPX_VERSION_MINOR(v) ((v>>8)&0xff)  /**< extract minor from packed version */
 #define VPX_VERSION_PATCH(v) ((v>>0)&0xff)  /**< extract patch from packed version */
 
-    /*!\brief Return the version major number */
+  /*!\brief Return the version major number */
 #define vpx_codec_version_major() ((vpx_codec_version()>>16)&0xff)
 
-    /*!\brief Return the version minor number */
+  /*!\brief Return the version minor number */
 #define vpx_codec_version_minor() ((vpx_codec_version()>>8)&0xff)
 
-    /*!\brief Return the version patch number */
+  /*!\brief Return the version patch number */
 #define vpx_codec_version_patch() ((vpx_codec_version()>>0)&0xff)
 
 
-    /*!\brief Return the version information (as a string)
-     *
-     * Returns a printable string containing the full library version number. This may
-     * contain additional text following the three digit version number, as to indicate
-     * release candidates, prerelease versions, etc.
-     *
-     */
-    const char *vpx_codec_version_str(void);
+  /*!\brief Return the version information (as a string)
+   *
+   * Returns a printable string containing the full library version number. This may
+   * contain additional text following the three digit version number, as to indicate
+   * release candidates, prerelease versions, etc.
+   *
+   */
+  const char *vpx_codec_version_str(void);
 
 
-    /*!\brief Return the version information (as a string)
-     *
-     * Returns a printable "extra string". This is the component of the string returned
-     * by vpx_codec_version_str() following the three digit version number.
-     *
-     */
-    const char *vpx_codec_version_extra_str(void);
+  /*!\brief Return the version information (as a string)
+   *
+   * Returns a printable "extra string". This is the component of the string returned
+   * by vpx_codec_version_str() following the three digit version number.
+   *
+   */
+  const char *vpx_codec_version_extra_str(void);
 
 
-    /*!\brief Return the build configuration
-     *
-     * Returns a printable string containing an encoded version of the build
-     * configuration. This may be useful to vpx support.
-     *
-     */
-    const char *vpx_codec_build_config(void);
+  /*!\brief Return the build configuration
+   *
+   * Returns a printable string containing an encoded version of the build
+   * configuration. This may be useful to vpx support.
+   *
+   */
+  const char *vpx_codec_build_config(void);
 
 
-    /*!\brief Return the name for a given interface
-     *
-     * Returns a human readable string for name of the given codec interface.
-     *
-     * \param[in]    iface     Interface pointer
-     *
-     */
-    const char *vpx_codec_iface_name(vpx_codec_iface_t *iface);
+  /*!\brief Return the name for a given interface
+   *
+   * Returns a human readable string for name of the given codec interface.
+   *
+   * \param[in]    iface     Interface pointer
+   *
+   */
+  const char *vpx_codec_iface_name(vpx_codec_iface_t *iface);
 
 
-    /*!\brief Convert error number to printable string
-     *
-     * Returns a human readable string for the last error returned by the
-     * algorithm. The returned error will be one line and will not contain
-     * any newline characters.
-     *
-     *
-     * \param[in]    err     Error number.
-     *
-     */
-    const char *vpx_codec_err_to_string(vpx_codec_err_t  err);
+  /*!\brief Convert error number to printable string
+   *
+   * Returns a human readable string for the last error returned by the
+   * algorithm. The returned error will be one line and will not contain
+   * any newline characters.
+   *
+   *
+   * \param[in]    err     Error number.
+   *
+   */
+  const char *vpx_codec_err_to_string(vpx_codec_err_t  err);
 
 
-    /*!\brief Retrieve error synopsis for codec context
-     *
-     * Returns a human readable string for the last error returned by the
-     * algorithm. The returned error will be one line and will not contain
-     * any newline characters.
-     *
-     *
-     * \param[in]    ctx     Pointer to this instance's context.
-     *
-     */
-    const char *vpx_codec_error(vpx_codec_ctx_t  *ctx);
+  /*!\brief Retrieve error synopsis for codec context
+   *
+   * Returns a human readable string for the last error returned by the
+   * algorithm. The returned error will be one line and will not contain
+   * any newline characters.
+   *
+   *
+   * \param[in]    ctx     Pointer to this instance's context.
+   *
+   */
+  const char *vpx_codec_error(vpx_codec_ctx_t  *ctx);
 
 
-    /*!\brief Retrieve detailed error information for codec context
-     *
-     * Returns a human readable string providing detailed information about
-     * the last error.
-     *
-     * \param[in]    ctx     Pointer to this instance's context.
-     *
-     * \retval NULL
-     *     No detailed information is available.
-     */
-    const char *vpx_codec_error_detail(vpx_codec_ctx_t  *ctx);
+  /*!\brief Retrieve detailed error information for codec context
+   *
+   * Returns a human readable string providing detailed information about
+   * the last error.
+   *
+   * \param[in]    ctx     Pointer to this instance's context.
+   *
+   * \retval NULL
+   *     No detailed information is available.
+   */
+  const char *vpx_codec_error_detail(vpx_codec_ctx_t  *ctx);
 
 
-    /* REQUIRED FUNCTIONS
-     *
-     * The following functions are required to be implemented for all codecs.
-     * They represent the base case functionality expected of all codecs.
-     */
+  /* REQUIRED FUNCTIONS
+   *
+   * The following functions are required to be implemented for all codecs.
+   * They represent the base case functionality expected of all codecs.
+   */
 
-    /*!\brief Destroy a codec instance
-     *
-     * Destroys a codec context, freeing any associated memory buffers.
-     *
-     * \param[in] ctx   Pointer to this instance's context
-     *
-     * \retval #VPX_CODEC_OK
-     *     The codec algorithm initialized.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Memory allocation failed.
-     */
-    vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx);
+  /*!\brief Destroy a codec instance
+   *
+   * Destroys a codec context, freeing any associated memory buffers.
+   *
+   * \param[in] ctx   Pointer to this instance's context
+   *
+   * \retval #VPX_CODEC_OK
+   *     The codec algorithm initialized.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Memory allocation failed.
+   */
+  vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx);
 
 
-    /*!\brief Get the capabilities of an algorithm.
-     *
-     * Retrieves the capabilities bitfield from the algorithm's interface.
-     *
-     * \param[in] iface   Pointer to the algorithm interface
-     *
-     */
-    vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface);
+  /*!\brief Get the capabilities of an algorithm.
+   *
+   * Retrieves the capabilities bitfield from the algorithm's interface.
+   *
+   * \param[in] iface   Pointer to the algorithm interface
+   *
+   */
+  vpx_codec_caps_t vpx_codec_get_caps(vpx_codec_iface_t *iface);
 
 
-    /*!\brief Control algorithm
-     *
-     * This function is used to exchange algorithm specific data with the codec
-     * instance. This can be used to implement features specific to a particular
-     * algorithm.
-     *
-     * This wrapper function dispatches the request to the helper function
-     * associated with the given ctrl_id. It tries to call this function
-     * transparently, but will return #VPX_CODEC_ERROR if the request could not
-     * be dispatched.
-     *
-     * Note that this function should not be used directly. Call the
-     * #vpx_codec_control wrapper macro instead.
-     *
-     * \param[in]     ctx              Pointer to this instance's context
-     * \param[in]     ctrl_id          Algorithm specific control identifier
-     *
-     * \retval #VPX_CODEC_OK
-     *     The control request was processed.
-     * \retval #VPX_CODEC_ERROR
-     *     The control request was not processed.
-     * \retval #VPX_CODEC_INVALID_PARAM
-     *     The data was not valid.
-     */
-    vpx_codec_err_t vpx_codec_control_(vpx_codec_ctx_t  *ctx,
-                                       int               ctrl_id,
-                                       ...);
+  /*!\brief Control algorithm
+   *
+   * This function is used to exchange algorithm specific data with the codec
+   * instance. This can be used to implement features specific to a particular
+   * algorithm.
+   *
+   * This wrapper function dispatches the request to the helper function
+   * associated with the given ctrl_id. It tries to call this function
+   * transparently, but will return #VPX_CODEC_ERROR if the request could not
+   * be dispatched.
+   *
+   * Note that this function should not be used directly. Call the
+   * #vpx_codec_control wrapper macro instead.
+   *
+   * \param[in]     ctx              Pointer to this instance's context
+   * \param[in]     ctrl_id          Algorithm specific control identifier
+   *
+   * \retval #VPX_CODEC_OK
+   *     The control request was processed.
+   * \retval #VPX_CODEC_ERROR
+   *     The control request was not processed.
+   * \retval #VPX_CODEC_INVALID_PARAM
+   *     The data was not valid.
+   */
+  vpx_codec_err_t vpx_codec_control_(vpx_codec_ctx_t  *ctx,
+                                     int               ctrl_id,
+                                     ...);
 #if defined(VPX_DISABLE_CTRL_TYPECHECKS) && VPX_DISABLE_CTRL_TYPECHECKS
 #    define vpx_codec_control(ctx,id,data) vpx_codec_control_(ctx,id,data)
 #    define VPX_CTRL_USE_TYPE(id, typ)
@@ -380,172 +385,171 @@
 #    define VPX_CTRL_VOID(id, typ)
 
 #else
-    /*!\brief vpx_codec_control wrapper macro
-     *
-     * This macro allows for type safe conversions across the variadic parameter
-     * to vpx_codec_control_().
-     *
-     * \internal
-     * It works by dispatching the call to the control function through a wrapper
-     * function named with the id parameter.
-     */
+  /*!\brief vpx_codec_control wrapper macro
+   *
+   * This macro allows for type safe conversions across the variadic parameter
+   * to vpx_codec_control_().
+   *
+   * \internal
+   * It works by dispatching the call to the control function through a wrapper
+   * function named with the id parameter.
+   */
 #    define vpx_codec_control(ctx,id,data) vpx_codec_control_##id(ctx,id,data)\
-    /**<\hideinitializer*/
+  /**<\hideinitializer*/
 
 
-    /*!\brief vpx_codec_control type definition macro
-     *
-     * This macro allows for type safe conversions across the variadic parameter
-     * to vpx_codec_control_(). It defines the type of the argument for a given
-     * control identifier.
-     *
-     * \internal
-     * It defines a static function with
-     * the correctly typed arguments as a wrapper to the type-unsafe internal
-     * function.
-     */
+  /*!\brief vpx_codec_control type definition macro
+   *
+   * This macro allows for type safe conversions across the variadic parameter
+   * to vpx_codec_control_(). It defines the type of the argument for a given
+   * control identifier.
+   *
+   * \internal
+   * It defines a static function with
+   * the correctly typed arguments as a wrapper to the type-unsafe internal
+   * function.
+   */
 #    define VPX_CTRL_USE_TYPE(id, typ) \
-    static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) UNUSED;\
-    \
-    static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id, typ data) {\
-        return vpx_codec_control_(ctx, ctrl_id, data);\
-    } /**<\hideinitializer*/
+  static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) UNUSED;\
+  \
+  static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id, typ data) {\
+    return vpx_codec_control_(ctx, ctrl_id, data);\
+  } /**<\hideinitializer*/
 
 
-    /*!\brief vpx_codec_control deprecated type definition macro
-     *
-     * Like #VPX_CTRL_USE_TYPE, but indicates that the specified control is
-     * deprecated and should not be used. Consult the documentation for your
-     * codec for more information.
-     *
-     * \internal
-     * It defines a static function with the correctly typed arguments as a
-     * wrapper to the type-unsafe internal function.
-     */
+  /*!\brief vpx_codec_control deprecated type definition macro
+   *
+   * Like #VPX_CTRL_USE_TYPE, but indicates that the specified control is
+   * deprecated and should not be used. Consult the documentation for your
+   * codec for more information.
+   *
+   * \internal
+   * It defines a static function with the correctly typed arguments as a
+   * wrapper to the type-unsafe internal function.
+   */
 #    define VPX_CTRL_USE_TYPE_DEPRECATED(id, typ) \
-    DECLSPEC_DEPRECATED static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) DEPRECATED UNUSED;\
-    \
-    DECLSPEC_DEPRECATED static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id, typ data) {\
-        return vpx_codec_control_(ctx, ctrl_id, data);\
-    } /**<\hideinitializer*/
+  DECLSPEC_DEPRECATED static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t*, int, typ) DEPRECATED UNUSED;\
+  \
+  DECLSPEC_DEPRECATED static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id, typ data) {\
+    return vpx_codec_control_(ctx, ctrl_id, data);\
+  } /**<\hideinitializer*/
 
 
-    /*!\brief vpx_codec_control void type definition macro
-     *
-     * This macro allows for type safe conversions across the variadic parameter
-     * to vpx_codec_control_(). It indicates that a given control identifier takes
-     * no argument.
-     *
-     * \internal
-     * It defines a static function without a data argument as a wrapper to the
-     * type-unsafe internal function.
-     */
+  /*!\brief vpx_codec_control void type definition macro
+   *
+   * This macro allows for type safe conversions across the variadic parameter
+   * to vpx_codec_control_(). It indicates that a given control identifier takes
+   * no argument.
+   *
+   * \internal
+   * It defines a static function without a data argument as a wrapper to the
+   * type-unsafe internal function.
+   */
 #    define VPX_CTRL_VOID(id) \
-    static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t*, int) UNUSED;\
-    \
-    static vpx_codec_err_t \
-    vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id) {\
-        return vpx_codec_control_(ctx, ctrl_id);\
-    } /**<\hideinitializer*/
+  static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t*, int) UNUSED;\
+  \
+  static vpx_codec_err_t \
+  vpx_codec_control_##id(vpx_codec_ctx_t  *ctx, int ctrl_id) {\
+    return vpx_codec_control_(ctx, ctrl_id);\
+  } /**<\hideinitializer*/
 
 
 #endif
 
 
-    /*!\defgroup cap_xma External Memory Allocation Functions
-     *
-     * The following functions are required to be implemented for all codecs
-     * that advertise the VPX_CODEC_CAP_XMA capability. Calling these functions
-     * for codecs that don't advertise this capability will result in an error
-     * code being returned, usually VPX_CODEC_INCAPABLE
-     * @{
-     */
+  /*!\defgroup cap_xma External Memory Allocation Functions
+   *
+   * The following functions are required to be implemented for all codecs
+   * that advertise the VPX_CODEC_CAP_XMA capability. Calling these functions
+   * for codecs that don't advertise this capability will result in an error
+   * code being returned, usually VPX_CODEC_INCAPABLE
+   * @{
+   */
 
 
-    /*!\brief Memory Map Entry
-     *
-     * This structure is used to contain the properties of a memory segment. It
-     * is populated by the codec in the request phase, and by the calling
-     * application once the requested allocation has been performed.
+  /*!\brief Memory Map Entry
+   *
+   * This structure is used to contain the properties of a memory segment. It
+   * is populated by the codec in the request phase, and by the calling
+   * application once the requested allocation has been performed.
+   */
+  typedef struct vpx_codec_mmap {
+    /*
+     * The following members are set by the codec when requesting a segment
      */
-    typedef struct vpx_codec_mmap
-    {
-        /*
-         * The following members are set by the codec when requesting a segment
-         */
-        unsigned int   id;     /**< identifier for the segment's contents */
-        unsigned long  sz;     /**< size of the segment, in bytes */
-        unsigned int   align;  /**< required alignment of the segment, in bytes */
-        unsigned int   flags;  /**< bitfield containing segment properties */
+    unsigned int   id;     /**< identifier for the segment's contents */
+    unsigned long  sz;     /**< size of the segment, in bytes */
+    unsigned int   align;  /**< required alignment of the segment, in bytes */
+    unsigned int   flags;  /**< bitfield containing segment properties */
 #define VPX_CODEC_MEM_ZERO     0x1  /**< Segment must be zeroed by allocation */
 #define VPX_CODEC_MEM_WRONLY   0x2  /**< Segment need not be readable */
 #define VPX_CODEC_MEM_FAST     0x4  /**< Place in fast memory, if available */
 
-        /* The following members are to be filled in by the allocation function */
-        void          *base;   /**< pointer to the allocated segment */
-        void (*dtor)(struct vpx_codec_mmap *map);         /**< destructor to call */
-        void          *priv;   /**< allocator private storage */
-    } vpx_codec_mmap_t; /**< alias for struct vpx_codec_mmap */
+    /* The following members are to be filled in by the allocation function */
+    void          *base;   /**< pointer to the allocated segment */
+    void (*dtor)(struct vpx_codec_mmap *map);         /**< destructor to call */
+    void          *priv;   /**< allocator private storage */
+  } vpx_codec_mmap_t; /**< alias for struct vpx_codec_mmap */
 
 
-    /*!\brief Iterate over the list of segments to allocate.
-     *
-     * Iterates over a list of the segments to allocate. The iterator storage
-     * should be initialized to NULL to start the iteration. Iteration is complete
-     * when this function returns VPX_CODEC_LIST_END. The amount of memory needed to
-     * allocate is dependent upon the size of the encoded stream. In cases where the
-     * stream is not available at allocation time, a fixed size must be requested.
-     * The codec will not be able to operate on streams larger than the size used at
-     * allocation time.
-     *
-     * \param[in]      ctx     Pointer to this instance's context.
-     * \param[out]     mmap    Pointer to the memory map entry to populate.
-     * \param[in,out]  iter    Iterator storage, initialized to NULL
-     *
-     * \retval #VPX_CODEC_OK
-     *     The memory map entry was populated.
-     * \retval #VPX_CODEC_ERROR
-     *     Codec does not support XMA mode.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Unable to determine segment size from stream info.
-     */
-    vpx_codec_err_t vpx_codec_get_mem_map(vpx_codec_ctx_t                *ctx,
-                                          vpx_codec_mmap_t               *mmap,
-                                          vpx_codec_iter_t               *iter);
+  /*!\brief Iterate over the list of segments to allocate.
+   *
+   * Iterates over a list of the segments to allocate. The iterator storage
+   * should be initialized to NULL to start the iteration. Iteration is complete
+   * when this function returns VPX_CODEC_LIST_END. The amount of memory needed to
+   * allocate is dependent upon the size of the encoded stream. In cases where the
+   * stream is not available at allocation time, a fixed size must be requested.
+   * The codec will not be able to operate on streams larger than the size used at
+   * allocation time.
+   *
+   * \param[in]      ctx     Pointer to this instance's context.
+   * \param[out]     mmap    Pointer to the memory map entry to populate.
+   * \param[in,out]  iter    Iterator storage, initialized to NULL
+   *
+   * \retval #VPX_CODEC_OK
+   *     The memory map entry was populated.
+   * \retval #VPX_CODEC_ERROR
+   *     Codec does not support XMA mode.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Unable to determine segment size from stream info.
+   */
+  vpx_codec_err_t vpx_codec_get_mem_map(vpx_codec_ctx_t                *ctx,
+                                        vpx_codec_mmap_t               *mmap,
+                                        vpx_codec_iter_t               *iter);
 
 
-    /*!\brief Identify allocated segments to codec instance
-     *
-     * Stores a list of allocated segments in the codec. Segments \ref MUST be
-     * passed in the order they are read from vpx_codec_get_mem_map(), but may be
-     * passed in groups of any size. Segments \ref MUST be set only once. The
-     * allocation function \ref MUST ensure that the vpx_codec_mmap_t::base member
-     * is non-NULL. If the segment requires cleanup handling (e.g., calling free()
-     * or close()) then the vpx_codec_mmap_t::dtor member \ref MUST be populated.
-     *
-     * \param[in]      ctx     Pointer to this instance's context.
-     * \param[in]      mmaps   Pointer to the first memory map entry in the list.
-     * \param[in]      num_maps  Number of entries being set at this time
-     *
-     * \retval #VPX_CODEC_OK
-     *     The segment was stored in the codec context.
-     * \retval #VPX_CODEC_INCAPABLE
-     *     Codec does not support XMA mode.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Segment base address was not set, or segment was already stored.
+  /*!\brief Identify allocated segments to codec instance
+   *
+   * Stores a list of allocated segments in the codec. Segments \ref MUST be
+   * passed in the order they are read from vpx_codec_get_mem_map(), but may be
+   * passed in groups of any size. Segments \ref MUST be set only once. The
+   * allocation function \ref MUST ensure that the vpx_codec_mmap_t::base member
+   * is non-NULL. If the segment requires cleanup handling (e.g., calling free()
+   * or close()) then the vpx_codec_mmap_t::dtor member \ref MUST be populated.
+   *
+   * \param[in]      ctx     Pointer to this instance's context.
+   * \param[in]      mmaps   Pointer to the first memory map entry in the list.
+   * \param[in]      num_maps  Number of entries being set at this time
+   *
+   * \retval #VPX_CODEC_OK
+   *     The segment was stored in the codec context.
+   * \retval #VPX_CODEC_INCAPABLE
+   *     Codec does not support XMA mode.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Segment base address was not set, or segment was already stored.
 
-     */
-    vpx_codec_err_t  vpx_codec_set_mem_map(vpx_codec_ctx_t   *ctx,
-                                           vpx_codec_mmap_t  *mmaps,
-                                           unsigned int       num_maps);
+   */
+  vpx_codec_err_t  vpx_codec_set_mem_map(vpx_codec_ctx_t   *ctx,
+                                         vpx_codec_mmap_t  *mmaps,
+                                         unsigned int       num_maps);
 
-    /*!@} - end defgroup cap_xma*/
-    /*!@} - end defgroup codec*/
+  /*!@} - end defgroup cap_xma*/
+  /*!@} - end defgroup codec*/
 
 
 #endif
diff --git a/libvpx/vpx/vpx_decoder.h b/libvpx/vpx/vpx_decoder.h
index 1ccf1c5..e7701e5 100644
--- a/libvpx/vpx/vpx_decoder.h
+++ b/libvpx/vpx/vpx_decoder.h
@@ -32,299 +32,302 @@
 #define VPX_DECODER_H
 #include "vpx_codec.h"
 
-    /*!\brief Current ABI version number
-     *
-     * \internal
-     * If this file is altered in any way that changes the ABI, this value
-     * must be bumped.  Examples include, but are not limited to, changing
-     * types, removing or reassigning enums, adding/removing/rearranging
-     * fields to structures
-     */
+  /*!\brief Current ABI version number
+   *
+   * \internal
+   * If this file is altered in any way that changes the ABI, this value
+   * must be bumped.  Examples include, but are not limited to, changing
+   * types, removing or reassigning enums, adding/removing/rearranging
+   * fields to structures
+   */
 #define VPX_DECODER_ABI_VERSION (2 + VPX_CODEC_ABI_VERSION) /**<\hideinitializer*/
 
-    /*! \brief Decoder capabilities bitfield
-     *
-     *  Each decoder advertises the capabilities it supports as part of its
-     *  ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces
-     *  or functionality, and are not required to be supported by a decoder.
-     *
-     *  The available flags are specified by VPX_CODEC_CAP_* defines.
-     */
+  /*! \brief Decoder capabilities bitfield
+   *
+   *  Each decoder advertises the capabilities it supports as part of its
+   *  ::vpx_codec_iface_t interface structure. Capabilities are extra interfaces
+   *  or functionality, and are not required to be supported by a decoder.
+   *
+   *  The available flags are specified by VPX_CODEC_CAP_* defines.
+   */
 #define VPX_CODEC_CAP_PUT_SLICE  0x10000 /**< Will issue put_slice callbacks */
 #define VPX_CODEC_CAP_PUT_FRAME  0x20000 /**< Will issue put_frame callbacks */
 #define VPX_CODEC_CAP_POSTPROC   0x40000 /**< Can postprocess decoded frame */
 #define VPX_CODEC_CAP_ERROR_CONCEALMENT   0x80000 /**< Can conceal errors due to
-                                                       packet loss */
+  packet loss */
 #define VPX_CODEC_CAP_INPUT_FRAGMENTS   0x100000 /**< Can receive encoded frames
-                                                    one fragment at a time */
+  one fragment at a time */
 
-    /*! \brief Initialization-time Feature Enabling
-     *
-     *  Certain codec features must be known at initialization time, to allow for
-     *  proper memory allocation.
-     *
-     *  The available flags are specified by VPX_CODEC_USE_* defines.
-     */
+  /*! \brief Initialization-time Feature Enabling
+   *
+   *  Certain codec features must be known at initialization time, to allow for
+   *  proper memory allocation.
+   *
+   *  The available flags are specified by VPX_CODEC_USE_* defines.
+   */
+#define VPX_CODEC_CAP_FRAME_THREADING   0x200000 /**< Can support frame-based
+                                                      multi-threading */
+
 #define VPX_CODEC_USE_POSTPROC   0x10000 /**< Postprocess decoded frame */
 #define VPX_CODEC_USE_ERROR_CONCEALMENT 0x20000 /**< Conceal errors in decoded
-                                                     frames */
+  frames */
 #define VPX_CODEC_USE_INPUT_FRAGMENTS   0x40000 /**< The input frame should be
-                                                    passed to the decoder one
-                                                    fragment at a time */
+  passed to the decoder one
+  fragment at a time */
+#define VPX_CODEC_USE_FRAME_THREADING   0x80000 /**< Enable frame-based
+                                                     multi-threading */
 
-    /*!\brief Stream properties
-     *
-     * This structure is used to query or set properties of the decoded
-     * stream. Algorithms may extend this structure with data specific
-     * to their bitstream by setting the sz member appropriately.
-     */
-    typedef struct vpx_codec_stream_info
-    {
-        unsigned int sz;     /**< Size of this structure */
-        unsigned int w;      /**< Width (or 0 for unknown/default) */
-        unsigned int h;      /**< Height (or 0 for unknown/default) */
-        unsigned int is_kf;  /**< Current frame is a keyframe */
-    } vpx_codec_stream_info_t;
+  /*!\brief Stream properties
+   *
+   * This structure is used to query or set properties of the decoded
+   * stream. Algorithms may extend this structure with data specific
+   * to their bitstream by setting the sz member appropriately.
+   */
+  typedef struct vpx_codec_stream_info {
+    unsigned int sz;     /**< Size of this structure */
+    unsigned int w;      /**< Width (or 0 for unknown/default) */
+    unsigned int h;      /**< Height (or 0 for unknown/default) */
+    unsigned int is_kf;  /**< Current frame is a keyframe */
+  } vpx_codec_stream_info_t;
 
-    /* REQUIRED FUNCTIONS
-     *
-     * The following functions are required to be implemented for all decoders.
-     * They represent the base case functionality expected of all decoders.
-     */
+  /* REQUIRED FUNCTIONS
+   *
+   * The following functions are required to be implemented for all decoders.
+   * They represent the base case functionality expected of all decoders.
+   */
 
 
-    /*!\brief Initialization Configurations
-     *
-     * This structure is used to pass init time configuration options to the
-     * decoder.
-     */
-    typedef struct vpx_codec_dec_cfg
-    {
-        unsigned int threads; /**< Maximum number of threads to use, default 1 */
-        unsigned int w;      /**< Width */
-        unsigned int h;      /**< Height */
-    } vpx_codec_dec_cfg_t; /**< alias for struct vpx_codec_dec_cfg */
+  /*!\brief Initialization Configurations
+   *
+   * This structure is used to pass init time configuration options to the
+   * decoder.
+   */
+  typedef struct vpx_codec_dec_cfg {
+    unsigned int threads; /**< Maximum number of threads to use, default 1 */
+    unsigned int w;      /**< Width */
+    unsigned int h;      /**< Height */
+  } vpx_codec_dec_cfg_t; /**< alias for struct vpx_codec_dec_cfg */
 
 
-    /*!\brief Initialize a decoder instance
-     *
-     * Initializes a decoder context using the given interface. Applications
-     * should call the vpx_codec_dec_init convenience macro instead of this
-     * function directly, to ensure that the ABI version number parameter
-     * is properly initialized.
-     *
-     * If the library was configured with --disable-multithread, this call
-     * is not thread safe and should be guarded with a lock if being used
-     * in a multithreaded context.
-     *
-     * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
-     * parameter), the storage pointed to by the cfg parameter must be
-     * kept readable and stable until all memory maps have been set.
-     *
-     * \param[in]    ctx     Pointer to this instance's context.
-     * \param[in]    iface   Pointer to the algorithm interface to use.
-     * \param[in]    cfg     Configuration to use, if known. May be NULL.
-     * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
-     * \param[in]    ver     ABI version number. Must be set to
-     *                       VPX_DECODER_ABI_VERSION
-     * \retval #VPX_CODEC_OK
-     *     The decoder algorithm initialized.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Memory allocation failed.
-     */
-    vpx_codec_err_t vpx_codec_dec_init_ver(vpx_codec_ctx_t      *ctx,
-                                           vpx_codec_iface_t    *iface,
-                                           vpx_codec_dec_cfg_t  *cfg,
-                                           vpx_codec_flags_t     flags,
-                                           int                   ver);
+  /*!\brief Initialize a decoder instance
+   *
+   * Initializes a decoder context using the given interface. Applications
+   * should call the vpx_codec_dec_init convenience macro instead of this
+   * function directly, to ensure that the ABI version number parameter
+   * is properly initialized.
+   *
+   * If the library was configured with --disable-multithread, this call
+   * is not thread safe and should be guarded with a lock if being used
+   * in a multithreaded context.
+   *
+   * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
+   * parameter), the storage pointed to by the cfg parameter must be
+   * kept readable and stable until all memory maps have been set.
+   *
+   * \param[in]    ctx     Pointer to this instance's context.
+   * \param[in]    iface   Pointer to the algorithm interface to use.
+   * \param[in]    cfg     Configuration to use, if known. May be NULL.
+   * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
+   * \param[in]    ver     ABI version number. Must be set to
+   *                       VPX_DECODER_ABI_VERSION
+   * \retval #VPX_CODEC_OK
+   *     The decoder algorithm initialized.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Memory allocation failed.
+   */
+  vpx_codec_err_t vpx_codec_dec_init_ver(vpx_codec_ctx_t      *ctx,
+                                         vpx_codec_iface_t    *iface,
+                                         vpx_codec_dec_cfg_t  *cfg,
+                                         vpx_codec_flags_t     flags,
+                                         int                   ver);
 
-    /*!\brief Convenience macro for vpx_codec_dec_init_ver()
-     *
-     * Ensures the ABI version parameter is properly set.
-     */
+  /*!\brief Convenience macro for vpx_codec_dec_init_ver()
+   *
+   * Ensures the ABI version parameter is properly set.
+   */
 #define vpx_codec_dec_init(ctx, iface, cfg, flags) \
-    vpx_codec_dec_init_ver(ctx, iface, cfg, flags, VPX_DECODER_ABI_VERSION)
+  vpx_codec_dec_init_ver(ctx, iface, cfg, flags, VPX_DECODER_ABI_VERSION)
 
 
-    /*!\brief Parse stream info from a buffer
-     *
-     * Performs high level parsing of the bitstream. Construction of a decoder
-     * context is not necessary. Can be used to determine if the bitstream is
-     * of the proper format, and to extract information from the stream.
-     *
-     * \param[in]      iface   Pointer to the algorithm interface
-     * \param[in]      data    Pointer to a block of data to parse
-     * \param[in]      data_sz Size of the data buffer
-     * \param[in,out]  si      Pointer to stream info to update. The size member
-     *                         \ref MUST be properly initialized, but \ref MAY be
-     *                         clobbered by the algorithm. This parameter \ref MAY
-     *                         be NULL.
-     *
-     * \retval #VPX_CODEC_OK
-     *     Bitstream is parsable and stream information updated
-     */
-    vpx_codec_err_t vpx_codec_peek_stream_info(vpx_codec_iface_t       *iface,
-            const uint8_t           *data,
-            unsigned int             data_sz,
-            vpx_codec_stream_info_t *si);
+  /*!\brief Parse stream info from a buffer
+   *
+   * Performs high level parsing of the bitstream. Construction of a decoder
+   * context is not necessary. Can be used to determine if the bitstream is
+   * of the proper format, and to extract information from the stream.
+   *
+   * \param[in]      iface   Pointer to the algorithm interface
+   * \param[in]      data    Pointer to a block of data to parse
+   * \param[in]      data_sz Size of the data buffer
+   * \param[in,out]  si      Pointer to stream info to update. The size member
+   *                         \ref MUST be properly initialized, but \ref MAY be
+   *                         clobbered by the algorithm. This parameter \ref MAY
+   *                         be NULL.
+   *
+   * \retval #VPX_CODEC_OK
+   *     Bitstream is parsable and stream information updated
+   */
+  vpx_codec_err_t vpx_codec_peek_stream_info(vpx_codec_iface_t       *iface,
+                                             const uint8_t           *data,
+                                             unsigned int             data_sz,
+                                             vpx_codec_stream_info_t *si);
 
 
-    /*!\brief Return information about the current stream.
-     *
-     * Returns information about the stream that has been parsed during decoding.
-     *
-     * \param[in]      ctx     Pointer to this instance's context
-     * \param[in,out]  si      Pointer to stream info to update. The size member
-     *                         \ref MUST be properly initialized, but \ref MAY be
-     *                         clobbered by the algorithm. This parameter \ref MAY
-     *                         be NULL.
-     *
-     * \retval #VPX_CODEC_OK
-     *     Bitstream is parsable and stream information updated
-     */
-    vpx_codec_err_t vpx_codec_get_stream_info(vpx_codec_ctx_t         *ctx,
-            vpx_codec_stream_info_t *si);
+  /*!\brief Return information about the current stream.
+   *
+   * Returns information about the stream that has been parsed during decoding.
+   *
+   * \param[in]      ctx     Pointer to this instance's context
+   * \param[in,out]  si      Pointer to stream info to update. The size member
+   *                         \ref MUST be properly initialized, but \ref MAY be
+   *                         clobbered by the algorithm. This parameter \ref MAY
+   *                         be NULL.
+   *
+   * \retval #VPX_CODEC_OK
+   *     Bitstream is parsable and stream information updated
+   */
+  vpx_codec_err_t vpx_codec_get_stream_info(vpx_codec_ctx_t         *ctx,
+                                            vpx_codec_stream_info_t *si);
 
 
-    /*!\brief Decode data
-     *
-     * Processes a buffer of coded data. If the processing results in a new
-     * decoded frame becoming available, PUT_SLICE and PUT_FRAME events may be
-     * generated, as appropriate. Encoded data \ref MUST be passed in DTS (decode
-     * time stamp) order. Frames produced will always be in PTS (presentation
-     * time stamp) order.
-     * If the decoder is configured with VPX_CODEC_USE_INPUT_FRAGMENTS enabled,
-     * data and data_sz can contain a fragment of the encoded frame. Fragment
-     * \#n must contain at least partition \#n, but can also contain subsequent
-     * partitions (\#n+1 - \#n+i), and if so, fragments \#n+1, .., \#n+i must
-     * be empty. When no more data is available, this function should be called
-     * with NULL as data and 0 as data_sz. The memory passed to this function
-     * must be available until the frame has been decoded.
-     *
-     * \param[in] ctx          Pointer to this instance's context
-     * \param[in] data         Pointer to this block of new coded data. If
-     *                         NULL, a VPX_CODEC_CB_PUT_FRAME event is posted
-     *                         for the previously decoded frame.
-     * \param[in] data_sz      Size of the coded data, in bytes.
-     * \param[in] user_priv    Application specific data to associate with
-     *                         this frame.
-     * \param[in] deadline     Soft deadline the decoder should attempt to meet,
-     *                         in us. Set to zero for unlimited.
-     *
-     * \return Returns #VPX_CODEC_OK if the coded data was processed completely
-     *         and future pictures can be decoded without error. Otherwise,
-     *         see the descriptions of the other error codes in ::vpx_codec_err_t
-     *         for recoverability capabilities.
-     */
-    vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t    *ctx,
-                                     const uint8_t        *data,
-                                     unsigned int            data_sz,
-                                     void               *user_priv,
-                                     long                deadline);
+  /*!\brief Decode data
+   *
+   * Processes a buffer of coded data. If the processing results in a new
+   * decoded frame becoming available, PUT_SLICE and PUT_FRAME events may be
+   * generated, as appropriate. Encoded data \ref MUST be passed in DTS (decode
+   * time stamp) order. Frames produced will always be in PTS (presentation
+   * time stamp) order.
+   * If the decoder is configured with VPX_CODEC_USE_INPUT_FRAGMENTS enabled,
+   * data and data_sz can contain a fragment of the encoded frame. Fragment
+   * \#n must contain at least partition \#n, but can also contain subsequent
+   * partitions (\#n+1 - \#n+i), and if so, fragments \#n+1, .., \#n+i must
+   * be empty. When no more data is available, this function should be called
+   * with NULL as data and 0 as data_sz. The memory passed to this function
+   * must be available until the frame has been decoded.
+   *
+   * \param[in] ctx          Pointer to this instance's context
+   * \param[in] data         Pointer to this block of new coded data. If
+   *                         NULL, a VPX_CODEC_CB_PUT_FRAME event is posted
+   *                         for the previously decoded frame.
+   * \param[in] data_sz      Size of the coded data, in bytes.
+   * \param[in] user_priv    Application specific data to associate with
+   *                         this frame.
+   * \param[in] deadline     Soft deadline the decoder should attempt to meet,
+   *                         in us. Set to zero for unlimited.
+   *
+   * \return Returns #VPX_CODEC_OK if the coded data was processed completely
+   *         and future pictures can be decoded without error. Otherwise,
+   *         see the descriptions of the other error codes in ::vpx_codec_err_t
+   *         for recoverability capabilities.
+   */
+  vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t    *ctx,
+                                   const uint8_t        *data,
+                                   unsigned int            data_sz,
+                                   void               *user_priv,
+                                   long                deadline);
 
 
-    /*!\brief Decoded frames iterator
-     *
-     * Iterates over a list of the frames available for display. The iterator
-     * storage should be initialized to NULL to start the iteration. Iteration is
-     * complete when this function returns NULL.
-     *
-     * The list of available frames becomes valid upon completion of the
-     * vpx_codec_decode call, and remains valid until the next call to vpx_codec_decode.
-     *
-     * \param[in]     ctx      Pointer to this instance's context
-     * \param[in,out] iter     Iterator storage, initialized to NULL
-     *
-     * \return Returns a pointer to an image, if one is ready for display. Frames
-     *         produced will always be in PTS (presentation time stamp) order.
-     */
-    vpx_image_t *vpx_codec_get_frame(vpx_codec_ctx_t  *ctx,
-                                     vpx_codec_iter_t *iter);
+  /*!\brief Decoded frames iterator
+   *
+   * Iterates over a list of the frames available for display. The iterator
+   * storage should be initialized to NULL to start the iteration. Iteration is
+   * complete when this function returns NULL.
+   *
+   * The list of available frames becomes valid upon completion of the
+   * vpx_codec_decode call, and remains valid until the next call to vpx_codec_decode.
+   *
+   * \param[in]     ctx      Pointer to this instance's context
+   * \param[in,out] iter     Iterator storage, initialized to NULL
+   *
+   * \return Returns a pointer to an image, if one is ready for display. Frames
+   *         produced will always be in PTS (presentation time stamp) order.
+   */
+  vpx_image_t *vpx_codec_get_frame(vpx_codec_ctx_t  *ctx,
+                                   vpx_codec_iter_t *iter);
 
 
-    /*!\defgroup cap_put_frame Frame-Based Decoding Functions
-     *
-     * The following functions are required to be implemented for all decoders
-     * that advertise the VPX_CODEC_CAP_PUT_FRAME capability. Calling these functions
-     * for codecs that don't advertise this capability will result in an error
-     * code being returned, usually VPX_CODEC_ERROR
-     * @{
-     */
+  /*!\defgroup cap_put_frame Frame-Based Decoding Functions
+   *
+   * The following functions are required to be implemented for all decoders
+   * that advertise the VPX_CODEC_CAP_PUT_FRAME capability. Calling these functions
+   * for codecs that don't advertise this capability will result in an error
+   * code being returned, usually VPX_CODEC_ERROR
+   * @{
+   */
 
-    /*!\brief put frame callback prototype
-     *
-     * This callback is invoked by the decoder to notify the application of
-     * the availability of decoded image data.
-     */
-    typedef void (*vpx_codec_put_frame_cb_fn_t)(void        *user_priv,
-            const vpx_image_t *img);
+  /*!\brief put frame callback prototype
+   *
+   * This callback is invoked by the decoder to notify the application of
+   * the availability of decoded image data.
+   */
+  typedef void (*vpx_codec_put_frame_cb_fn_t)(void        *user_priv,
+                                              const vpx_image_t *img);
 
 
-    /*!\brief Register for notification of frame completion.
-     *
-     * Registers a given function to be called when a decoded frame is
-     * available.
-     *
-     * \param[in] ctx          Pointer to this instance's context
-     * \param[in] cb           Pointer to the callback function
-     * \param[in] user_priv    User's private data
-     *
-     * \retval #VPX_CODEC_OK
-     *     Callback successfully registered.
-     * \retval #VPX_CODEC_ERROR
-     *     Decoder context not initialized, or algorithm not capable of
-     *     posting slice completion.
-     */
-    vpx_codec_err_t vpx_codec_register_put_frame_cb(vpx_codec_ctx_t             *ctx,
-            vpx_codec_put_frame_cb_fn_t  cb,
-            void                        *user_priv);
+  /*!\brief Register for notification of frame completion.
+   *
+   * Registers a given function to be called when a decoded frame is
+   * available.
+   *
+   * \param[in] ctx          Pointer to this instance's context
+   * \param[in] cb           Pointer to the callback function
+   * \param[in] user_priv    User's private data
+   *
+   * \retval #VPX_CODEC_OK
+   *     Callback successfully registered.
+   * \retval #VPX_CODEC_ERROR
+   *     Decoder context not initialized, or algorithm not capable of
+   *     posting slice completion.
+   */
+  vpx_codec_err_t vpx_codec_register_put_frame_cb(vpx_codec_ctx_t             *ctx,
+                                                  vpx_codec_put_frame_cb_fn_t  cb,
+                                                  void                        *user_priv);
 
 
-    /*!@} - end defgroup cap_put_frame */
+  /*!@} - end defgroup cap_put_frame */
 
-    /*!\defgroup cap_put_slice Slice-Based Decoding Functions
-     *
-     * The following functions are required to be implemented for all decoders
-     * that advertise the VPX_CODEC_CAP_PUT_SLICE capability. Calling these functions
-     * for codecs that don't advertise this capability will result in an error
-     * code being returned, usually VPX_CODEC_ERROR
-     * @{
-     */
+  /*!\defgroup cap_put_slice Slice-Based Decoding Functions
+   *
+   * The following functions are required to be implemented for all decoders
+   * that advertise the VPX_CODEC_CAP_PUT_SLICE capability. Calling these functions
+   * for codecs that don't advertise this capability will result in an error
+   * code being returned, usually VPX_CODEC_ERROR
+   * @{
+   */
 
-    /*!\brief put slice callback prototype
-     *
-     * This callback is invoked by the decoder to notify the application of
-     * the availability of partially decoded image data. The
-     */
-    typedef void (*vpx_codec_put_slice_cb_fn_t)(void         *user_priv,
-            const vpx_image_t      *img,
-            const vpx_image_rect_t *valid,
-            const vpx_image_rect_t *update);
+  /*!\brief put slice callback prototype
+   *
+   * This callback is invoked by the decoder to notify the application of
+   * the availability of partially decoded image data. The
+   */
+  typedef void (*vpx_codec_put_slice_cb_fn_t)(void         *user_priv,
+                                              const vpx_image_t      *img,
+                                              const vpx_image_rect_t *valid,
+                                              const vpx_image_rect_t *update);
 
 
-    /*!\brief Register for notification of slice completion.
-     *
-     * Registers a given function to be called when a decoded slice is
-     * available.
-     *
-     * \param[in] ctx          Pointer to this instance's context
-     * \param[in] cb           Pointer to the callback function
-     * \param[in] user_priv    User's private data
-     *
-     * \retval #VPX_CODEC_OK
-     *     Callback successfully registered.
-     * \retval #VPX_CODEC_ERROR
-     *     Decoder context not initialized, or algorithm not capable of
-     *     posting slice completion.
-     */
-    vpx_codec_err_t vpx_codec_register_put_slice_cb(vpx_codec_ctx_t             *ctx,
-            vpx_codec_put_slice_cb_fn_t  cb,
-            void                        *user_priv);
+  /*!\brief Register for notification of slice completion.
+   *
+   * Registers a given function to be called when a decoded slice is
+   * available.
+   *
+   * \param[in] ctx          Pointer to this instance's context
+   * \param[in] cb           Pointer to the callback function
+   * \param[in] user_priv    User's private data
+   *
+   * \retval #VPX_CODEC_OK
+   *     Callback successfully registered.
+   * \retval #VPX_CODEC_ERROR
+   *     Decoder context not initialized, or algorithm not capable of
+   *     posting slice completion.
+   */
+  vpx_codec_err_t vpx_codec_register_put_slice_cb(vpx_codec_ctx_t             *ctx,
+                                                  vpx_codec_put_slice_cb_fn_t  cb,
+                                                  void                        *user_priv);
 
 
-    /*!@} - end defgroup cap_put_slice*/
+  /*!@} - end defgroup cap_put_slice*/
 
-    /*!@} - end defgroup decoder*/
+  /*!@} - end defgroup decoder*/
 
 #endif
 
diff --git a/libvpx/vpx/vpx_encoder.h b/libvpx/vpx/vpx_encoder.h
index 67d9033..ffdbc06 100644
--- a/libvpx/vpx/vpx_encoder.h
+++ b/libvpx/vpx/vpx_encoder.h
@@ -32,902 +32,891 @@
 #define VPX_ENCODER_H
 #include "vpx_codec.h"
 
-/*! Temporal Scalability: Maximum length of the sequence defining frame
- * layer membership
- */
+  /*! Temporal Scalability: Maximum length of the sequence defining frame
+   * layer membership
+   */
 #define VPX_TS_MAX_PERIODICITY 16
 
-/*! Temporal Scalability: Maximum number of coding layers */
+  /*! Temporal Scalability: Maximum number of coding layers */
 #define VPX_TS_MAX_LAYERS       5
 
-/*!\deprecated Use #VPX_TS_MAX_PERIODICITY instead. */
+  /*!\deprecated Use #VPX_TS_MAX_PERIODICITY instead. */
 #define MAX_PERIODICITY VPX_TS_MAX_PERIODICITY
 
-/*!\deprecated Use #VPX_TS_MAX_LAYERS instead. */
+  /*!\deprecated Use #VPX_TS_MAX_LAYERS instead. */
 #define MAX_LAYERS      VPX_TS_MAX_LAYERS
 
-    /*!\brief Current ABI version number
-     *
-     * \internal
-     * If this file is altered in any way that changes the ABI, this value
-     * must be bumped.  Examples include, but are not limited to, changing
-     * types, removing or reassigning enums, adding/removing/rearranging
-     * fields to structures
-     */
+  /*!\brief Current ABI version number
+   *
+   * \internal
+   * If this file is altered in any way that changes the ABI, this value
+   * must be bumped.  Examples include, but are not limited to, changing
+   * types, removing or reassigning enums, adding/removing/rearranging
+   * fields to structures
+   */
 #define VPX_ENCODER_ABI_VERSION (3 + VPX_CODEC_ABI_VERSION) /**<\hideinitializer*/
 
 
-    /*! \brief Encoder capabilities bitfield
-     *
-     *  Each encoder advertises the capabilities it supports as part of its
-     *  ::vpx_codec_iface_t interface structure. Capabilities are extra
-     *  interfaces or functionality, and are not required to be supported
-     *  by an encoder.
-     *
-     *  The available flags are specified by VPX_CODEC_CAP_* defines.
-     */
+  /*! \brief Encoder capabilities bitfield
+   *
+   *  Each encoder advertises the capabilities it supports as part of its
+   *  ::vpx_codec_iface_t interface structure. Capabilities are extra
+   *  interfaces or functionality, and are not required to be supported
+   *  by an encoder.
+   *
+   *  The available flags are specified by VPX_CODEC_CAP_* defines.
+   */
 #define VPX_CODEC_CAP_PSNR  0x10000 /**< Can issue PSNR packets */
 
-    /*! Can output one partition at a time. Each partition is returned in its
-     *  own VPX_CODEC_CX_FRAME_PKT, with the FRAME_IS_FRAGMENT flag set for
-     *  every partition but the last. In this mode all frames are always
-     *  returned partition by partition.
-     */
+  /*! Can output one partition at a time. Each partition is returned in its
+   *  own VPX_CODEC_CX_FRAME_PKT, with the FRAME_IS_FRAGMENT flag set for
+   *  every partition but the last. In this mode all frames are always
+   *  returned partition by partition.
+   */
 #define VPX_CODEC_CAP_OUTPUT_PARTITION  0x20000
 
 
-    /*! \brief Initialization-time Feature Enabling
-     *
-     *  Certain codec features must be known at initialization time, to allow
-     *  for proper memory allocation.
-     *
-     *  The available flags are specified by VPX_CODEC_USE_* defines.
-     */
+  /*! \brief Initialization-time Feature Enabling
+   *
+   *  Certain codec features must be known at initialization time, to allow
+   *  for proper memory allocation.
+   *
+   *  The available flags are specified by VPX_CODEC_USE_* defines.
+   */
 #define VPX_CODEC_USE_PSNR  0x10000 /**< Calculate PSNR on each frame */
 #define VPX_CODEC_USE_OUTPUT_PARTITION  0x20000 /**< Make the encoder output one
-                                                     partition at a time. */
+  partition at a time. */
 
 
-    /*!\brief Generic fixed size buffer structure
-     *
-     * This structure is able to hold a reference to any fixed size buffer.
-     */
-    typedef struct vpx_fixed_buf
-    {
-        void          *buf; /**< Pointer to the data */
-        size_t         sz;  /**< Length of the buffer, in chars */
-    } vpx_fixed_buf_t; /**< alias for struct vpx_fixed_buf */
+  /*!\brief Generic fixed size buffer structure
+   *
+   * This structure is able to hold a reference to any fixed size buffer.
+   */
+  typedef struct vpx_fixed_buf {
+    void          *buf; /**< Pointer to the data */
+    size_t         sz;  /**< Length of the buffer, in chars */
+  } vpx_fixed_buf_t; /**< alias for struct vpx_fixed_buf */
 
 
-    /*!\brief Time Stamp Type
-     *
-     * An integer, which when multiplied by the stream's time base, provides
-     * the absolute time of a sample.
-     */
-    typedef int64_t vpx_codec_pts_t;
+  /*!\brief Time Stamp Type
+   *
+   * An integer, which when multiplied by the stream's time base, provides
+   * the absolute time of a sample.
+   */
+  typedef int64_t vpx_codec_pts_t;
 
 
-    /*!\brief Compressed Frame Flags
-     *
-     * This type represents a bitfield containing information about a compressed
-     * frame that may be useful to an application. The most significant 16 bits
-     * can be used by an algorithm to provide additional detail, for example to
-     * support frame types that are codec specific (MPEG-1 D-frames for example)
-     */
-    typedef uint32_t vpx_codec_frame_flags_t;
+  /*!\brief Compressed Frame Flags
+   *
+   * This type represents a bitfield containing information about a compressed
+   * frame that may be useful to an application. The most significant 16 bits
+   * can be used by an algorithm to provide additional detail, for example to
+   * support frame types that are codec specific (MPEG-1 D-frames for example)
+   */
+  typedef uint32_t vpx_codec_frame_flags_t;
 #define VPX_FRAME_IS_KEY       0x1 /**< frame is the start of a GOP */
 #define VPX_FRAME_IS_DROPPABLE 0x2 /**< frame can be dropped without affecting
-    the stream (no future frame depends on
-                this one) */
+  the stream (no future frame depends on
+              this one) */
 #define VPX_FRAME_IS_INVISIBLE 0x4 /**< frame should be decoded but will not
-    be shown */
+  be shown */
 #define VPX_FRAME_IS_FRAGMENT  0x8 /**< this is a fragment of the encoded
-    frame */
+  frame */
 
-    /*!\brief Error Resilient flags
-     *
-     * These flags define which error resilient features to enable in the
-     * encoder. The flags are specified through the
-     * vpx_codec_enc_cfg::g_error_resilient variable.
-     */
-    typedef uint32_t vpx_codec_er_flags_t;
+  /*!\brief Error Resilient flags
+   *
+   * These flags define which error resilient features to enable in the
+   * encoder. The flags are specified through the
+   * vpx_codec_enc_cfg::g_error_resilient variable.
+   */
+  typedef uint32_t vpx_codec_er_flags_t;
 #define VPX_ERROR_RESILIENT_DEFAULT     0x1 /**< Improve resiliency against
-                                                 losses of whole frames */
+  losses of whole frames */
 #define VPX_ERROR_RESILIENT_PARTITIONS  0x2 /**< The frame partitions are
-                                                 independently decodable by the
-                                                 bool decoder, meaning that
-                                                 partitions can be decoded even
-                                                 though earlier partitions have
-                                                 been lost. Note that intra
-                                                 predicition is still done over
-                                                 the partition boundary. */
+  independently decodable by the
+  bool decoder, meaning that
+  partitions can be decoded even
+  though earlier partitions have
+  been lost. Note that intra
+  predicition is still done over
+  the partition boundary. */
 
-    /*!\brief Encoder output packet variants
-     *
-     * This enumeration lists the different kinds of data packets that can be
-     * returned by calls to vpx_codec_get_cx_data(). Algorithms \ref MAY
-     * extend this list to provide additional functionality.
-     */
-    enum vpx_codec_cx_pkt_kind
-    {
-        VPX_CODEC_CX_FRAME_PKT,    /**< Compressed video frame */
-        VPX_CODEC_STATS_PKT,       /**< Two-pass statistics for this frame */
-        VPX_CODEC_PSNR_PKT,        /**< PSNR statistics for this frame */
-        VPX_CODEC_CUSTOM_PKT = 256 /**< Algorithm extensions  */
-    };
+  /*!\brief Encoder output packet variants
+   *
+   * This enumeration lists the different kinds of data packets that can be
+   * returned by calls to vpx_codec_get_cx_data(). Algorithms \ref MAY
+   * extend this list to provide additional functionality.
+   */
+  enum vpx_codec_cx_pkt_kind {
+    VPX_CODEC_CX_FRAME_PKT,    /**< Compressed video frame */
+    VPX_CODEC_STATS_PKT,       /**< Two-pass statistics for this frame */
+    VPX_CODEC_PSNR_PKT,        /**< PSNR statistics for this frame */
+    VPX_CODEC_CUSTOM_PKT = 256 /**< Algorithm extensions  */
+  };
 
 
-    /*!\brief Encoder output packet
-     *
-     * This structure contains the different kinds of output data the encoder
-     * may produce while compressing a frame.
-     */
-    typedef struct vpx_codec_cx_pkt
-    {
-        enum vpx_codec_cx_pkt_kind  kind; /**< packet variant */
-        union
-        {
-            struct
-            {
-                void                    *buf;      /**< compressed data buffer */
-                size_t                   sz;       /**< length of compressed data */
-                vpx_codec_pts_t          pts;      /**< time stamp to show frame
+  /*!\brief Encoder output packet
+   *
+   * This structure contains the different kinds of output data the encoder
+   * may produce while compressing a frame.
+   */
+  typedef struct vpx_codec_cx_pkt {
+    enum vpx_codec_cx_pkt_kind  kind; /**< packet variant */
+    union {
+      struct {
+        void                    *buf;      /**< compressed data buffer */
+        size_t                   sz;       /**< length of compressed data */
+        vpx_codec_pts_t          pts;      /**< time stamp to show frame
                                                     (in timebase units) */
-                unsigned long            duration; /**< duration to show frame
+        unsigned long            duration; /**< duration to show frame
                                                     (in timebase units) */
-                vpx_codec_frame_flags_t  flags;    /**< flags for this frame */
-                int                      partition_id; /**< the partition id
+        vpx_codec_frame_flags_t  flags;    /**< flags for this frame */
+        int                      partition_id; /**< the partition id
                                               defines the decoding order
                                               of the partitions. Only
                                               applicable when "output partition"
                                               mode is enabled. First partition
                                               has id 0.*/
 
-            } frame;  /**< data for compressed frame packet */
-            struct vpx_fixed_buf twopass_stats;  /**< data for two-pass packet */
-            struct vpx_psnr_pkt
-            {
-                unsigned int samples[4];  /**< Number of samples, total/y/u/v */
-                uint64_t     sse[4];      /**< sum squared error, total/y/u/v */
-                double       psnr[4];     /**< PSNR, total/y/u/v */
-            } psnr;                       /**< data for PSNR packet */
-            struct vpx_fixed_buf raw;     /**< data for arbitrary packets */
+      } frame;  /**< data for compressed frame packet */
+      struct vpx_fixed_buf twopass_stats;  /**< data for two-pass packet */
+      struct vpx_psnr_pkt {
+        unsigned int samples[4];  /**< Number of samples, total/y/u/v */
+        uint64_t     sse[4];      /**< sum squared error, total/y/u/v */
+        double       psnr[4];     /**< PSNR, total/y/u/v */
+      } psnr;                       /**< data for PSNR packet */
+      struct vpx_fixed_buf raw;     /**< data for arbitrary packets */
 
-            /* This packet size is fixed to allow codecs to extend this
-             * interface without having to manage storage for raw packets,
-             * i.e., if it's smaller than 128 bytes, you can store in the
-             * packet list directly.
-             */
-            char pad[128 - sizeof(enum vpx_codec_cx_pkt_kind)]; /**< fixed sz */
-        } data; /**< packet data */
-    } vpx_codec_cx_pkt_t; /**< alias for struct vpx_codec_cx_pkt */
+      /* This packet size is fixed to allow codecs to extend this
+       * interface without having to manage storage for raw packets,
+       * i.e., if it's smaller than 128 bytes, you can store in the
+       * packet list directly.
+       */
+      char pad[128 - sizeof(enum vpx_codec_cx_pkt_kind)]; /**< fixed sz */
+    } data; /**< packet data */
+  } vpx_codec_cx_pkt_t; /**< alias for struct vpx_codec_cx_pkt */
 
 
-    /*!\brief Rational Number
-     *
-     * This structure holds a fractional value.
-     */
-    typedef struct vpx_rational
-    {
-        int num; /**< fraction numerator */
-        int den; /**< fraction denominator */
-    } vpx_rational_t; /**< alias for struct vpx_rational */
+  /*!\brief Rational Number
+   *
+   * This structure holds a fractional value.
+   */
+  typedef struct vpx_rational {
+    int num; /**< fraction numerator */
+    int den; /**< fraction denominator */
+  } vpx_rational_t; /**< alias for struct vpx_rational */
 
 
-    /*!\brief Multi-pass Encoding Pass */
-    enum vpx_enc_pass
-    {
-        VPX_RC_ONE_PASS,   /**< Single pass mode */
-        VPX_RC_FIRST_PASS, /**< First pass of multi-pass mode */
-        VPX_RC_LAST_PASS   /**< Final pass of multi-pass mode */
-    };
+  /*!\brief Multi-pass Encoding Pass */
+  enum vpx_enc_pass {
+    VPX_RC_ONE_PASS,   /**< Single pass mode */
+    VPX_RC_FIRST_PASS, /**< First pass of multi-pass mode */
+    VPX_RC_LAST_PASS   /**< Final pass of multi-pass mode */
+  };
 
 
-    /*!\brief Rate control mode */
-    enum vpx_rc_mode
-    {
-        VPX_VBR, /**< Variable Bit Rate (VBR) mode */
-        VPX_CBR,  /**< Constant Bit Rate (CBR) mode */
-        VPX_CQ   /**< Constant Quality  (CQ)  mode */
-    };
+  /*!\brief Rate control mode */
+  enum vpx_rc_mode {
+    VPX_VBR, /**< Variable Bit Rate (VBR) mode */
+    VPX_CBR,  /**< Constant Bit Rate (CBR) mode */
+    VPX_CQ   /**< Constant Quality  (CQ)  mode */
+  };
 
 
-    /*!\brief Keyframe placement mode.
-     *
-     * This enumeration determines whether keyframes are placed automatically by
-     * the encoder or whether this behavior is disabled. Older releases of this
-     * SDK were implemented such that VPX_KF_FIXED meant keyframes were disabled.
-     * This name is confusing for this behavior, so the new symbols to be used
-     * are VPX_KF_AUTO and VPX_KF_DISABLED.
-     */
-    enum vpx_kf_mode
-    {
-        VPX_KF_FIXED, /**< deprecated, implies VPX_KF_DISABLED */
-        VPX_KF_AUTO,  /**< Encoder determines optimal placement automatically */
-        VPX_KF_DISABLED = 0 /**< Encoder does not place keyframes. */
-    };
+  /*!\brief Keyframe placement mode.
+   *
+   * This enumeration determines whether keyframes are placed automatically by
+   * the encoder or whether this behavior is disabled. Older releases of this
+   * SDK were implemented such that VPX_KF_FIXED meant keyframes were disabled.
+   * This name is confusing for this behavior, so the new symbols to be used
+   * are VPX_KF_AUTO and VPX_KF_DISABLED.
+   */
+  enum vpx_kf_mode {
+    VPX_KF_FIXED, /**< deprecated, implies VPX_KF_DISABLED */
+    VPX_KF_AUTO,  /**< Encoder determines optimal placement automatically */
+    VPX_KF_DISABLED = 0 /**< Encoder does not place keyframes. */
+  };
 
 
-    /*!\brief Encoded Frame Flags
-     *
-     * This type indicates a bitfield to be passed to vpx_codec_encode(), defining
-     * per-frame boolean values. By convention, bits common to all codecs will be
-     * named VPX_EFLAG_*, and bits specific to an algorithm will be named
-     * /algo/_eflag_*. The lower order 16 bits are reserved for common use.
-     */
-    typedef long vpx_enc_frame_flags_t;
+  /*!\brief Encoded Frame Flags
+   *
+   * This type indicates a bitfield to be passed to vpx_codec_encode(), defining
+   * per-frame boolean values. By convention, bits common to all codecs will be
+   * named VPX_EFLAG_*, and bits specific to an algorithm will be named
+   * /algo/_eflag_*. The lower order 16 bits are reserved for common use.
+   */
+  typedef long vpx_enc_frame_flags_t;
 #define VPX_EFLAG_FORCE_KF (1<<0)  /**< Force this frame to be a keyframe */
 
 
-    /*!\brief Encoder configuration structure
-     *
-     * This structure contains the encoder settings that have common representations
-     * across all codecs. This doesn't imply that all codecs support all features,
-     * however.
+  /*!\brief Encoder configuration structure
+   *
+   * This structure contains the encoder settings that have common representations
+   * across all codecs. This doesn't imply that all codecs support all features,
+   * however.
+   */
+  typedef struct vpx_codec_enc_cfg {
+    /*
+     * generic settings (g)
      */
-    typedef struct vpx_codec_enc_cfg
-    {
-        /*
-         * generic settings (g)
-         */
 
-        /*!\brief Algorithm specific "usage" value
-         *
-         * Algorithms may define multiple values for usage, which may convey the
-         * intent of how the application intends to use the stream. If this value
-         * is non-zero, consult the documentation for the codec to determine its
-         * meaning.
-         */
-        unsigned int           g_usage;
-
-
-        /*!\brief Maximum number of threads to use
-         *
-         * For multi-threaded implementations, use no more than this number of
-         * threads. The codec may use fewer threads than allowed. The value
-         * 0 is equivalent to the value 1.
-         */
-        unsigned int           g_threads;
-
-
-        /*!\brief Bitstream profile to use
-         *
-         * Some codecs support a notion of multiple bitstream profiles. Typically
-         * this maps to a set of features that are turned on or off. Often the
-         * profile to use is determined by the features of the intended decoder.
-         * Consult the documentation for the codec to determine the valid values
-         * for this parameter, or set to zero for a sane default.
-         */
-        unsigned int           g_profile;  /**< profile of bitstream to use */
-
-
-
-        /*!\brief Width of the frame
-         *
-         * This value identifies the presentation resolution of the frame,
-         * in pixels. Note that the frames passed as input to the encoder must
-         * have this resolution. Frames will be presented by the decoder in this
-         * resolution, independent of any spatial resampling the encoder may do.
-         */
-        unsigned int           g_w;
-
-
-        /*!\brief Height of the frame
-         *
-         * This value identifies the presentation resolution of the frame,
-         * in pixels. Note that the frames passed as input to the encoder must
-         * have this resolution. Frames will be presented by the decoder in this
-         * resolution, independent of any spatial resampling the encoder may do.
-         */
-        unsigned int           g_h;
-
-
-        /*!\brief Stream timebase units
-         *
-         * Indicates the smallest interval of time, in seconds, used by the stream.
-         * For fixed frame rate material, or variable frame rate material where
-         * frames are timed at a multiple of a given clock (ex: video capture),
-         * the \ref RECOMMENDED method is to set the timebase to the reciprocal
-         * of the frame rate (ex: 1001/30000 for 29.970 Hz NTSC). This allows the
-         * pts to correspond to the frame number, which can be handy. For
-         * re-encoding video from containers with absolute time timestamps, the
-         * \ref RECOMMENDED method is to set the timebase to that of the parent
-         * container or multimedia framework (ex: 1/1000 for ms, as in FLV).
-         */
-        struct vpx_rational    g_timebase;
-
-
-        /*!\brief Enable error resilient modes.
-         *
-         * The error resilient bitfield indicates to the encoder which features
-         * it should enable to take measures for streaming over lossy or noisy
-         * links.
-         */
-        vpx_codec_er_flags_t   g_error_resilient;
-
-
-        /*!\brief Multi-pass Encoding Mode
-         *
-         * This value should be set to the current phase for multi-pass encoding.
-         * For single pass, set to #VPX_RC_ONE_PASS.
-         */
-        enum vpx_enc_pass      g_pass;
-
-
-        /*!\brief Allow lagged encoding
-         *
-         * If set, this value allows the encoder to consume a number of input
-         * frames before producing output frames. This allows the encoder to
-         * base decisions for the current frame on future frames. This does
-         * increase the latency of the encoding pipeline, so it is not appropriate
-         * in all situations (ex: realtime encoding).
-         *
-         * Note that this is a maximum value -- the encoder may produce frames
-         * sooner than the given limit. Set this value to 0 to disable this
-         * feature.
-         */
-        unsigned int           g_lag_in_frames;
-
-
-        /*
-         * rate control settings (rc)
-         */
-
-        /*!\brief Temporal resampling configuration, if supported by the codec.
-         *
-         * Temporal resampling allows the codec to "drop" frames as a strategy to
-         * meet its target data rate. This can cause temporal discontinuities in
-         * the encoded video, which may appear as stuttering during playback. This
-         * trade-off is often acceptable, but for many applications is not. It can
-         * be disabled in these cases.
-         *
-         * Note that not all codecs support this feature. All vpx VPx codecs do.
-         * For other codecs, consult the documentation for that algorithm.
-         *
-         * This threshold is described as a percentage of the target data buffer.
-         * When the data buffer falls below this percentage of fullness, a
-         * dropped frame is indicated. Set the threshold to zero (0) to disable
-         * this feature.
-         */
-        unsigned int           rc_dropframe_thresh;
-
-
-        /*!\brief Enable/disable spatial resampling, if supported by the codec.
-         *
-         * Spatial resampling allows the codec to compress a lower resolution
-         * version of the frame, which is then upscaled by the encoder to the
-         * correct presentation resolution. This increases visual quality at
-         * low data rates, at the expense of CPU time on the encoder/decoder.
-         */
-        unsigned int           rc_resize_allowed;
-
-
-        /*!\brief Spatial resampling up watermark.
-         *
-         * This threshold is described as a percentage of the target data buffer.
-         * When the data buffer rises above this percentage of fullness, the
-         * encoder will step up to a higher resolution version of the frame.
-         */
-        unsigned int           rc_resize_up_thresh;
-
-
-        /*!\brief Spatial resampling down watermark.
-         *
-         * This threshold is described as a percentage of the target data buffer.
-         * When the data buffer falls below this percentage of fullness, the
-         * encoder will step down to a lower resolution version of the frame.
-         */
-        unsigned int           rc_resize_down_thresh;
-
-
-        /*!\brief Rate control algorithm to use.
-         *
-         * Indicates whether the end usage of this stream is to be streamed over
-         * a bandwidth constrained link, indicating that Constant Bit Rate (CBR)
-         * mode should be used, or whether it will be played back on a high
-         * bandwidth link, as from a local disk, where higher variations in
-         * bitrate are acceptable.
-         */
-        enum vpx_rc_mode       rc_end_usage;
-
-
-        /*!\brief Two-pass stats buffer.
-         *
-         * A buffer containing all of the stats packets produced in the first
-         * pass, concatenated.
-         */
-        struct vpx_fixed_buf   rc_twopass_stats_in;
-
-
-        /*!\brief Target data rate
-         *
-         * Target bandwidth to use for this stream, in kilobits per second.
-         */
-        unsigned int           rc_target_bitrate;
-
-
-        /*
-         * quantizer settings
-         */
-
-
-        /*!\brief Minimum (Best Quality) Quantizer
-         *
-         * The quantizer is the most direct control over the quality of the
-         * encoded image. The range of valid values for the quantizer is codec
-         * specific. Consult the documentation for the codec to determine the
-         * values to use. To determine the range programmatically, call
-         * vpx_codec_enc_config_default() with a usage value of 0.
-         */
-        unsigned int           rc_min_quantizer;
-
-
-        /*!\brief Maximum (Worst Quality) Quantizer
-         *
-         * The quantizer is the most direct control over the quality of the
-         * encoded image. The range of valid values for the quantizer is codec
-         * specific. Consult the documentation for the codec to determine the
-         * values to use. To determine the range programmatically, call
-         * vpx_codec_enc_config_default() with a usage value of 0.
-         */
-        unsigned int           rc_max_quantizer;
-
-
-        /*
-         * bitrate tolerance
-         */
-
-
-        /*!\brief Rate control adaptation undershoot control
-         *
-         * This value, expressed as a percentage of the target bitrate,
-         * controls the maximum allowed adaptation speed of the codec.
-         * This factor controls the maximum amount of bits that can
-         * be subtracted from the target bitrate in order to compensate
-         * for prior overshoot.
-         *
-         * Valid values in the range 0-1000.
-         */
-        unsigned int           rc_undershoot_pct;
-
-
-        /*!\brief Rate control adaptation overshoot control
-         *
-         * This value, expressed as a percentage of the target bitrate,
-         * controls the maximum allowed adaptation speed of the codec.
-         * This factor controls the maximum amount of bits that can
-         * be added to the target bitrate in order to compensate for
-         * prior undershoot.
-         *
-         * Valid values in the range 0-1000.
-         */
-        unsigned int           rc_overshoot_pct;
-
-
-        /*
-         * decoder buffer model parameters
-         */
-
-
-        /*!\brief Decoder Buffer Size
-         *
-         * This value indicates the amount of data that may be buffered by the
-         * decoding application. Note that this value is expressed in units of
-         * time (milliseconds). For example, a value of 5000 indicates that the
-         * client will buffer (at least) 5000ms worth of encoded data. Use the
-         * target bitrate (#rc_target_bitrate) to convert to bits/bytes, if
-         * necessary.
-         */
-        unsigned int           rc_buf_sz;
-
-
-        /*!\brief Decoder Buffer Initial Size
-         *
-         * This value indicates the amount of data that will be buffered by the
-         * decoding application prior to beginning playback. This value is
-         * expressed in units of time (milliseconds). Use the target bitrate
-         * (#rc_target_bitrate) to convert to bits/bytes, if necessary.
-         */
-        unsigned int           rc_buf_initial_sz;
-
-
-        /*!\brief Decoder Buffer Optimal Size
-         *
-         * This value indicates the amount of data that the encoder should try
-         * to maintain in the decoder's buffer. This value is expressed in units
-         * of time (milliseconds). Use the target bitrate (#rc_target_bitrate)
-         * to convert to bits/bytes, if necessary.
-         */
-        unsigned int           rc_buf_optimal_sz;
-
-
-        /*
-         * 2 pass rate control parameters
-         */
-
-
-        /*!\brief Two-pass mode CBR/VBR bias
-         *
-         * Bias, expressed on a scale of 0 to 100, for determining target size
-         * for the current frame. The value 0 indicates the optimal CBR mode
-         * value should be used. The value 100 indicates the optimal VBR mode
-         * value should be used. Values in between indicate which way the
-         * encoder should "lean."
-         */
-        unsigned int           rc_2pass_vbr_bias_pct;       /**< RC mode bias between CBR and VBR(0-100: 0->CBR, 100->VBR)   */
-
-
-        /*!\brief Two-pass mode per-GOP minimum bitrate
-         *
-         * This value, expressed as a percentage of the target bitrate, indicates
-         * the minimum bitrate to be used for a single GOP (aka "section")
-         */
-        unsigned int           rc_2pass_vbr_minsection_pct;
-
-
-        /*!\brief Two-pass mode per-GOP maximum bitrate
-         *
-         * This value, expressed as a percentage of the target bitrate, indicates
-         * the maximum bitrate to be used for a single GOP (aka "section")
-         */
-        unsigned int           rc_2pass_vbr_maxsection_pct;
-
-
-        /*
-         * keyframing settings (kf)
-         */
-
-        /*!\brief Keyframe placement mode
-         *
-         * This value indicates whether the encoder should place keyframes at a
-         * fixed interval, or determine the optimal placement automatically
-         * (as governed by the #kf_min_dist and #kf_max_dist parameters)
-         */
-        enum vpx_kf_mode       kf_mode;
-
-
-        /*!\brief Keyframe minimum interval
-         *
-         * This value, expressed as a number of frames, prevents the encoder from
-         * placing a keyframe nearer than kf_min_dist to the previous keyframe. At
-         * least kf_min_dist frames non-keyframes will be coded before the next
-         * keyframe. Set kf_min_dist equal to kf_max_dist for a fixed interval.
-         */
-        unsigned int           kf_min_dist;
-
-
-        /*!\brief Keyframe maximum interval
-         *
-         * This value, expressed as a number of frames, forces the encoder to code
-         * a keyframe if one has not been coded in the last kf_max_dist frames.
-         * A value of 0 implies all frames will be keyframes. Set kf_min_dist
-         * equal to kf_max_dist for a fixed interval.
-         */
-        unsigned int           kf_max_dist;
-
-        /*
-         * Temporal scalability settings (ts)
-         */
-
-        /*!\brief Number of coding layers
-         *
-         * This value specifies the number of coding layers to be used.
-         */
-        unsigned int           ts_number_layers;
-
-        /*!\brief Target bitrate for each layer
-         *
-         * These values specify the target coding bitrate for each coding layer.
-         */
-        unsigned int           ts_target_bitrate[VPX_TS_MAX_LAYERS];
-
-        /*!\brief Frame rate decimation factor for each layer
-         *
-         * These values specify the frame rate decimation factors to apply
-         * to each layer.
-         */
-        unsigned int           ts_rate_decimator[VPX_TS_MAX_LAYERS];
-
-        /*!\brief Length of the sequence defining frame layer membership
-         *
-         * This value specifies the length of the sequence that defines the
-         * membership of frames to layers. For example, if ts_periodicity=8 then
-         * frames are assigned to coding layers with a repeated sequence of
-         * length 8.
-         */
-        unsigned int           ts_periodicity;
-
-        /*!\brief Template defining the membership of frames to coding layers
-         *
-         * This array defines the membership of frames to coding layers. For a
-         * 2-layer encoding that assigns even numbered frames to one layer (0)
-         * and odd numbered frames to a second layer (1) with ts_periodicity=8,
-         * then ts_layer_id = (0,1,0,1,0,1,0,1).
-         */
-        unsigned int           ts_layer_id[VPX_TS_MAX_PERIODICITY];
-    } vpx_codec_enc_cfg_t; /**< alias for struct vpx_codec_enc_cfg */
-
-
-    /*!\brief Initialize an encoder instance
+    /*!\brief Algorithm specific "usage" value
      *
-     * Initializes a encoder context using the given interface. Applications
-     * should call the vpx_codec_enc_init convenience macro instead of this
-     * function directly, to ensure that the ABI version number parameter
-     * is properly initialized.
-     *
-     * If the library was configured with --disable-multithread, this call
-     * is not thread safe and should be guarded with a lock if being used
-     * in a multithreaded context.
-     *
-     * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
-     * parameter), the storage pointed to by the cfg parameter must be
-     * kept readable and stable until all memory maps have been set.
-     *
-     * \param[in]    ctx     Pointer to this instance's context.
-     * \param[in]    iface   Pointer to the algorithm interface to use.
-     * \param[in]    cfg     Configuration to use, if known. May be NULL.
-     * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
-     * \param[in]    ver     ABI version number. Must be set to
-     *                       VPX_ENCODER_ABI_VERSION
-     * \retval #VPX_CODEC_OK
-     *     The decoder algorithm initialized.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Memory allocation failed.
+     * Algorithms may define multiple values for usage, which may convey the
+     * intent of how the application intends to use the stream. If this value
+     * is non-zero, consult the documentation for the codec to determine its
+     * meaning.
      */
-    vpx_codec_err_t vpx_codec_enc_init_ver(vpx_codec_ctx_t      *ctx,
-                                           vpx_codec_iface_t    *iface,
-                                           vpx_codec_enc_cfg_t  *cfg,
-                                           vpx_codec_flags_t     flags,
-                                           int                   ver);
+    unsigned int           g_usage;
 
 
-    /*!\brief Convenience macro for vpx_codec_enc_init_ver()
+    /*!\brief Maximum number of threads to use
      *
-     * Ensures the ABI version parameter is properly set.
+     * For multi-threaded implementations, use no more than this number of
+     * threads. The codec may use fewer threads than allowed. The value
+     * 0 is equivalent to the value 1.
      */
+    unsigned int           g_threads;
+
+
+    /*!\brief Bitstream profile to use
+     *
+     * Some codecs support a notion of multiple bitstream profiles. Typically
+     * this maps to a set of features that are turned on or off. Often the
+     * profile to use is determined by the features of the intended decoder.
+     * Consult the documentation for the codec to determine the valid values
+     * for this parameter, or set to zero for a sane default.
+     */
+    unsigned int           g_profile;  /**< profile of bitstream to use */
+
+
+
+    /*!\brief Width of the frame
+     *
+     * This value identifies the presentation resolution of the frame,
+     * in pixels. Note that the frames passed as input to the encoder must
+     * have this resolution. Frames will be presented by the decoder in this
+     * resolution, independent of any spatial resampling the encoder may do.
+     */
+    unsigned int           g_w;
+
+
+    /*!\brief Height of the frame
+     *
+     * This value identifies the presentation resolution of the frame,
+     * in pixels. Note that the frames passed as input to the encoder must
+     * have this resolution. Frames will be presented by the decoder in this
+     * resolution, independent of any spatial resampling the encoder may do.
+     */
+    unsigned int           g_h;
+
+
+    /*!\brief Stream timebase units
+     *
+     * Indicates the smallest interval of time, in seconds, used by the stream.
+     * For fixed frame rate material, or variable frame rate material where
+     * frames are timed at a multiple of a given clock (ex: video capture),
+     * the \ref RECOMMENDED method is to set the timebase to the reciprocal
+     * of the frame rate (ex: 1001/30000 for 29.970 Hz NTSC). This allows the
+     * pts to correspond to the frame number, which can be handy. For
+     * re-encoding video from containers with absolute time timestamps, the
+     * \ref RECOMMENDED method is to set the timebase to that of the parent
+     * container or multimedia framework (ex: 1/1000 for ms, as in FLV).
+     */
+    struct vpx_rational    g_timebase;
+
+
+    /*!\brief Enable error resilient modes.
+     *
+     * The error resilient bitfield indicates to the encoder which features
+     * it should enable to take measures for streaming over lossy or noisy
+     * links.
+     */
+    vpx_codec_er_flags_t   g_error_resilient;
+
+
+    /*!\brief Multi-pass Encoding Mode
+     *
+     * This value should be set to the current phase for multi-pass encoding.
+     * For single pass, set to #VPX_RC_ONE_PASS.
+     */
+    enum vpx_enc_pass      g_pass;
+
+
+    /*!\brief Allow lagged encoding
+     *
+     * If set, this value allows the encoder to consume a number of input
+     * frames before producing output frames. This allows the encoder to
+     * base decisions for the current frame on future frames. This does
+     * increase the latency of the encoding pipeline, so it is not appropriate
+     * in all situations (ex: realtime encoding).
+     *
+     * Note that this is a maximum value -- the encoder may produce frames
+     * sooner than the given limit. Set this value to 0 to disable this
+     * feature.
+     */
+    unsigned int           g_lag_in_frames;
+
+
+    /*
+     * rate control settings (rc)
+     */
+
+    /*!\brief Temporal resampling configuration, if supported by the codec.
+     *
+     * Temporal resampling allows the codec to "drop" frames as a strategy to
+     * meet its target data rate. This can cause temporal discontinuities in
+     * the encoded video, which may appear as stuttering during playback. This
+     * trade-off is often acceptable, but for many applications is not. It can
+     * be disabled in these cases.
+     *
+     * Note that not all codecs support this feature. All vpx VPx codecs do.
+     * For other codecs, consult the documentation for that algorithm.
+     *
+     * This threshold is described as a percentage of the target data buffer.
+     * When the data buffer falls below this percentage of fullness, a
+     * dropped frame is indicated. Set the threshold to zero (0) to disable
+     * this feature.
+     */
+    unsigned int           rc_dropframe_thresh;
+
+
+    /*!\brief Enable/disable spatial resampling, if supported by the codec.
+     *
+     * Spatial resampling allows the codec to compress a lower resolution
+     * version of the frame, which is then upscaled by the encoder to the
+     * correct presentation resolution. This increases visual quality at
+     * low data rates, at the expense of CPU time on the encoder/decoder.
+     */
+    unsigned int           rc_resize_allowed;
+
+
+    /*!\brief Spatial resampling up watermark.
+     *
+     * This threshold is described as a percentage of the target data buffer.
+     * When the data buffer rises above this percentage of fullness, the
+     * encoder will step up to a higher resolution version of the frame.
+     */
+    unsigned int           rc_resize_up_thresh;
+
+
+    /*!\brief Spatial resampling down watermark.
+     *
+     * This threshold is described as a percentage of the target data buffer.
+     * When the data buffer falls below this percentage of fullness, the
+     * encoder will step down to a lower resolution version of the frame.
+     */
+    unsigned int           rc_resize_down_thresh;
+
+
+    /*!\brief Rate control algorithm to use.
+     *
+     * Indicates whether the end usage of this stream is to be streamed over
+     * a bandwidth constrained link, indicating that Constant Bit Rate (CBR)
+     * mode should be used, or whether it will be played back on a high
+     * bandwidth link, as from a local disk, where higher variations in
+     * bitrate are acceptable.
+     */
+    enum vpx_rc_mode       rc_end_usage;
+
+
+    /*!\brief Two-pass stats buffer.
+     *
+     * A buffer containing all of the stats packets produced in the first
+     * pass, concatenated.
+     */
+    struct vpx_fixed_buf   rc_twopass_stats_in;
+
+
+    /*!\brief Target data rate
+     *
+     * Target bandwidth to use for this stream, in kilobits per second.
+     */
+    unsigned int           rc_target_bitrate;
+
+
+    /*
+     * quantizer settings
+     */
+
+
+    /*!\brief Minimum (Best Quality) Quantizer
+     *
+     * The quantizer is the most direct control over the quality of the
+     * encoded image. The range of valid values for the quantizer is codec
+     * specific. Consult the documentation for the codec to determine the
+     * values to use. To determine the range programmatically, call
+     * vpx_codec_enc_config_default() with a usage value of 0.
+     */
+    unsigned int           rc_min_quantizer;
+
+
+    /*!\brief Maximum (Worst Quality) Quantizer
+     *
+     * The quantizer is the most direct control over the quality of the
+     * encoded image. The range of valid values for the quantizer is codec
+     * specific. Consult the documentation for the codec to determine the
+     * values to use. To determine the range programmatically, call
+     * vpx_codec_enc_config_default() with a usage value of 0.
+     */
+    unsigned int           rc_max_quantizer;
+
+
+    /*
+     * bitrate tolerance
+     */
+
+
+    /*!\brief Rate control adaptation undershoot control
+     *
+     * This value, expressed as a percentage of the target bitrate,
+     * controls the maximum allowed adaptation speed of the codec.
+     * This factor controls the maximum amount of bits that can
+     * be subtracted from the target bitrate in order to compensate
+     * for prior overshoot.
+     *
+     * Valid values in the range 0-1000.
+     */
+    unsigned int           rc_undershoot_pct;
+
+
+    /*!\brief Rate control adaptation overshoot control
+     *
+     * This value, expressed as a percentage of the target bitrate,
+     * controls the maximum allowed adaptation speed of the codec.
+     * This factor controls the maximum amount of bits that can
+     * be added to the target bitrate in order to compensate for
+     * prior undershoot.
+     *
+     * Valid values in the range 0-1000.
+     */
+    unsigned int           rc_overshoot_pct;
+
+
+    /*
+     * decoder buffer model parameters
+     */
+
+
+    /*!\brief Decoder Buffer Size
+     *
+     * This value indicates the amount of data that may be buffered by the
+     * decoding application. Note that this value is expressed in units of
+     * time (milliseconds). For example, a value of 5000 indicates that the
+     * client will buffer (at least) 5000ms worth of encoded data. Use the
+     * target bitrate (#rc_target_bitrate) to convert to bits/bytes, if
+     * necessary.
+     */
+    unsigned int           rc_buf_sz;
+
+
+    /*!\brief Decoder Buffer Initial Size
+     *
+     * This value indicates the amount of data that will be buffered by the
+     * decoding application prior to beginning playback. This value is
+     * expressed in units of time (milliseconds). Use the target bitrate
+     * (#rc_target_bitrate) to convert to bits/bytes, if necessary.
+     */
+    unsigned int           rc_buf_initial_sz;
+
+
+    /*!\brief Decoder Buffer Optimal Size
+     *
+     * This value indicates the amount of data that the encoder should try
+     * to maintain in the decoder's buffer. This value is expressed in units
+     * of time (milliseconds). Use the target bitrate (#rc_target_bitrate)
+     * to convert to bits/bytes, if necessary.
+     */
+    unsigned int           rc_buf_optimal_sz;
+
+
+    /*
+     * 2 pass rate control parameters
+     */
+
+
+    /*!\brief Two-pass mode CBR/VBR bias
+     *
+     * Bias, expressed on a scale of 0 to 100, for determining target size
+     * for the current frame. The value 0 indicates the optimal CBR mode
+     * value should be used. The value 100 indicates the optimal VBR mode
+     * value should be used. Values in between indicate which way the
+     * encoder should "lean."
+     */
+    unsigned int           rc_2pass_vbr_bias_pct;       /**< RC mode bias between CBR and VBR(0-100: 0->CBR, 100->VBR)   */
+
+
+    /*!\brief Two-pass mode per-GOP minimum bitrate
+     *
+     * This value, expressed as a percentage of the target bitrate, indicates
+     * the minimum bitrate to be used for a single GOP (aka "section")
+     */
+    unsigned int           rc_2pass_vbr_minsection_pct;
+
+
+    /*!\brief Two-pass mode per-GOP maximum bitrate
+     *
+     * This value, expressed as a percentage of the target bitrate, indicates
+     * the maximum bitrate to be used for a single GOP (aka "section")
+     */
+    unsigned int           rc_2pass_vbr_maxsection_pct;
+
+
+    /*
+     * keyframing settings (kf)
+     */
+
+    /*!\brief Keyframe placement mode
+     *
+     * This value indicates whether the encoder should place keyframes at a
+     * fixed interval, or determine the optimal placement automatically
+     * (as governed by the #kf_min_dist and #kf_max_dist parameters)
+     */
+    enum vpx_kf_mode       kf_mode;
+
+
+    /*!\brief Keyframe minimum interval
+     *
+     * This value, expressed as a number of frames, prevents the encoder from
+     * placing a keyframe nearer than kf_min_dist to the previous keyframe. At
+     * least kf_min_dist frames non-keyframes will be coded before the next
+     * keyframe. Set kf_min_dist equal to kf_max_dist for a fixed interval.
+     */
+    unsigned int           kf_min_dist;
+
+
+    /*!\brief Keyframe maximum interval
+     *
+     * This value, expressed as a number of frames, forces the encoder to code
+     * a keyframe if one has not been coded in the last kf_max_dist frames.
+     * A value of 0 implies all frames will be keyframes. Set kf_min_dist
+     * equal to kf_max_dist for a fixed interval.
+     */
+    unsigned int           kf_max_dist;
+
+    /*
+     * Temporal scalability settings (ts)
+     */
+
+    /*!\brief Number of coding layers
+     *
+     * This value specifies the number of coding layers to be used.
+     */
+    unsigned int           ts_number_layers;
+
+    /*!\brief Target bitrate for each layer
+     *
+     * These values specify the target coding bitrate for each coding layer.
+     */
+    unsigned int           ts_target_bitrate[VPX_TS_MAX_LAYERS];
+
+    /*!\brief Frame rate decimation factor for each layer
+     *
+     * These values specify the frame rate decimation factors to apply
+     * to each layer.
+     */
+    unsigned int           ts_rate_decimator[VPX_TS_MAX_LAYERS];
+
+    /*!\brief Length of the sequence defining frame layer membership
+     *
+     * This value specifies the length of the sequence that defines the
+     * membership of frames to layers. For example, if ts_periodicity=8 then
+     * frames are assigned to coding layers with a repeated sequence of
+     * length 8.
+     */
+    unsigned int           ts_periodicity;
+
+    /*!\brief Template defining the membership of frames to coding layers
+     *
+     * This array defines the membership of frames to coding layers. For a
+     * 2-layer encoding that assigns even numbered frames to one layer (0)
+     * and odd numbered frames to a second layer (1) with ts_periodicity=8,
+     * then ts_layer_id = (0,1,0,1,0,1,0,1).
+     */
+    unsigned int           ts_layer_id[VPX_TS_MAX_PERIODICITY];
+  } vpx_codec_enc_cfg_t; /**< alias for struct vpx_codec_enc_cfg */
+
+
+  /*!\brief Initialize an encoder instance
+   *
+   * Initializes a encoder context using the given interface. Applications
+   * should call the vpx_codec_enc_init convenience macro instead of this
+   * function directly, to ensure that the ABI version number parameter
+   * is properly initialized.
+   *
+   * If the library was configured with --disable-multithread, this call
+   * is not thread safe and should be guarded with a lock if being used
+   * in a multithreaded context.
+   *
+   * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
+   * parameter), the storage pointed to by the cfg parameter must be
+   * kept readable and stable until all memory maps have been set.
+   *
+   * \param[in]    ctx     Pointer to this instance's context.
+   * \param[in]    iface   Pointer to the algorithm interface to use.
+   * \param[in]    cfg     Configuration to use, if known. May be NULL.
+   * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
+   * \param[in]    ver     ABI version number. Must be set to
+   *                       VPX_ENCODER_ABI_VERSION
+   * \retval #VPX_CODEC_OK
+   *     The decoder algorithm initialized.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Memory allocation failed.
+   */
+  vpx_codec_err_t vpx_codec_enc_init_ver(vpx_codec_ctx_t      *ctx,
+                                         vpx_codec_iface_t    *iface,
+                                         vpx_codec_enc_cfg_t  *cfg,
+                                         vpx_codec_flags_t     flags,
+                                         int                   ver);
+
+
+  /*!\brief Convenience macro for vpx_codec_enc_init_ver()
+   *
+   * Ensures the ABI version parameter is properly set.
+   */
 #define vpx_codec_enc_init(ctx, iface, cfg, flags) \
-    vpx_codec_enc_init_ver(ctx, iface, cfg, flags, VPX_ENCODER_ABI_VERSION)
+  vpx_codec_enc_init_ver(ctx, iface, cfg, flags, VPX_ENCODER_ABI_VERSION)
 
 
-    /*!\brief Initialize multi-encoder instance
-     *
-     * Initializes multi-encoder context using the given interface.
-     * Applications should call the vpx_codec_enc_init_multi convenience macro
-     * instead of this function directly, to ensure that the ABI version number
-     * parameter is properly initialized.
-     *
-     * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
-     * parameter), the storage pointed to by the cfg parameter must be
-     * kept readable and stable until all memory maps have been set.
-     *
-     * \param[in]    ctx     Pointer to this instance's context.
-     * \param[in]    iface   Pointer to the algorithm interface to use.
-     * \param[in]    cfg     Configuration to use, if known. May be NULL.
-     * \param[in]    num_enc Total number of encoders.
-     * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
-     * \param[in]    dsf     Pointer to down-sampling factors.
-     * \param[in]    ver     ABI version number. Must be set to
-     *                       VPX_ENCODER_ABI_VERSION
-     * \retval #VPX_CODEC_OK
-     *     The decoder algorithm initialized.
-     * \retval #VPX_CODEC_MEM_ERROR
-     *     Memory allocation failed.
-     */
-    vpx_codec_err_t vpx_codec_enc_init_multi_ver(vpx_codec_ctx_t      *ctx,
-                                                 vpx_codec_iface_t    *iface,
-                                                 vpx_codec_enc_cfg_t  *cfg,
-                                                 int                   num_enc,
-                                                 vpx_codec_flags_t     flags,
-                                                 vpx_rational_t       *dsf,
-                                                 int                   ver);
+  /*!\brief Initialize multi-encoder instance
+   *
+   * Initializes multi-encoder context using the given interface.
+   * Applications should call the vpx_codec_enc_init_multi convenience macro
+   * instead of this function directly, to ensure that the ABI version number
+   * parameter is properly initialized.
+   *
+   * In XMA mode (activated by setting VPX_CODEC_USE_XMA in the flags
+   * parameter), the storage pointed to by the cfg parameter must be
+   * kept readable and stable until all memory maps have been set.
+   *
+   * \param[in]    ctx     Pointer to this instance's context.
+   * \param[in]    iface   Pointer to the algorithm interface to use.
+   * \param[in]    cfg     Configuration to use, if known. May be NULL.
+   * \param[in]    num_enc Total number of encoders.
+   * \param[in]    flags   Bitfield of VPX_CODEC_USE_* flags
+   * \param[in]    dsf     Pointer to down-sampling factors.
+   * \param[in]    ver     ABI version number. Must be set to
+   *                       VPX_ENCODER_ABI_VERSION
+   * \retval #VPX_CODEC_OK
+   *     The decoder algorithm initialized.
+   * \retval #VPX_CODEC_MEM_ERROR
+   *     Memory allocation failed.
+   */
+  vpx_codec_err_t vpx_codec_enc_init_multi_ver(vpx_codec_ctx_t      *ctx,
+                                               vpx_codec_iface_t    *iface,
+                                               vpx_codec_enc_cfg_t  *cfg,
+                                               int                   num_enc,
+                                               vpx_codec_flags_t     flags,
+                                               vpx_rational_t       *dsf,
+                                               int                   ver);
 
 
-    /*!\brief Convenience macro for vpx_codec_enc_init_multi_ver()
-     *
-     * Ensures the ABI version parameter is properly set.
-     */
+  /*!\brief Convenience macro for vpx_codec_enc_init_multi_ver()
+   *
+   * Ensures the ABI version parameter is properly set.
+   */
 #define vpx_codec_enc_init_multi(ctx, iface, cfg, num_enc, flags, dsf) \
-    vpx_codec_enc_init_multi_ver(ctx, iface, cfg, num_enc, flags, dsf, \
-                                 VPX_ENCODER_ABI_VERSION)
+  vpx_codec_enc_init_multi_ver(ctx, iface, cfg, num_enc, flags, dsf, \
+                               VPX_ENCODER_ABI_VERSION)
 
 
-    /*!\brief Get a default configuration
-     *
-     * Initializes a encoder configuration structure with default values. Supports
-     * the notion of "usages" so that an algorithm may offer different default
-     * settings depending on the user's intended goal. This function \ref SHOULD
-     * be called by all applications to initialize the configuration structure
-     * before specializing the configuration with application specific values.
-     *
-     * \param[in]    iface   Pointer to the algorithm interface to use.
-     * \param[out]   cfg     Configuration buffer to populate
-     * \param[in]    usage   End usage. Set to 0 or use codec specific values.
-     *
-     * \retval #VPX_CODEC_OK
-     *     The configuration was populated.
-     * \retval #VPX_CODEC_INCAPABLE
-     *     Interface is not an encoder interface.
-     * \retval #VPX_CODEC_INVALID_PARAM
-     *     A parameter was NULL, or the usage value was not recognized.
-     */
-    vpx_codec_err_t  vpx_codec_enc_config_default(vpx_codec_iface_t    *iface,
-            vpx_codec_enc_cfg_t  *cfg,
-            unsigned int          usage);
+  /*!\brief Get a default configuration
+   *
+   * Initializes a encoder configuration structure with default values. Supports
+   * the notion of "usages" so that an algorithm may offer different default
+   * settings depending on the user's intended goal. This function \ref SHOULD
+   * be called by all applications to initialize the configuration structure
+   * before specializing the configuration with application specific values.
+   *
+   * \param[in]    iface   Pointer to the algorithm interface to use.
+   * \param[out]   cfg     Configuration buffer to populate
+   * \param[in]    usage   End usage. Set to 0 or use codec specific values.
+   *
+   * \retval #VPX_CODEC_OK
+   *     The configuration was populated.
+   * \retval #VPX_CODEC_INCAPABLE
+   *     Interface is not an encoder interface.
+   * \retval #VPX_CODEC_INVALID_PARAM
+   *     A parameter was NULL, or the usage value was not recognized.
+   */
+  vpx_codec_err_t  vpx_codec_enc_config_default(vpx_codec_iface_t    *iface,
+                                                vpx_codec_enc_cfg_t  *cfg,
+                                                unsigned int          usage);
 
 
-    /*!\brief Set or change configuration
-     *
-     * Reconfigures an encoder instance according to the given configuration.
-     *
-     * \param[in]    ctx     Pointer to this instance's context
-     * \param[in]    cfg     Configuration buffer to use
-     *
-     * \retval #VPX_CODEC_OK
-     *     The configuration was populated.
-     * \retval #VPX_CODEC_INCAPABLE
-     *     Interface is not an encoder interface.
-     * \retval #VPX_CODEC_INVALID_PARAM
-     *     A parameter was NULL, or the usage value was not recognized.
-     */
-    vpx_codec_err_t  vpx_codec_enc_config_set(vpx_codec_ctx_t            *ctx,
-            const vpx_codec_enc_cfg_t  *cfg);
+  /*!\brief Set or change configuration
+   *
+   * Reconfigures an encoder instance according to the given configuration.
+   *
+   * \param[in]    ctx     Pointer to this instance's context
+   * \param[in]    cfg     Configuration buffer to use
+   *
+   * \retval #VPX_CODEC_OK
+   *     The configuration was populated.
+   * \retval #VPX_CODEC_INCAPABLE
+   *     Interface is not an encoder interface.
+   * \retval #VPX_CODEC_INVALID_PARAM
+   *     A parameter was NULL, or the usage value was not recognized.
+   */
+  vpx_codec_err_t  vpx_codec_enc_config_set(vpx_codec_ctx_t            *ctx,
+                                            const vpx_codec_enc_cfg_t  *cfg);
 
 
-    /*!\brief Get global stream headers
-     *
-     * Retrieves a stream level global header packet, if supported by the codec.
-     *
-     * \param[in]    ctx     Pointer to this instance's context
-     *
-     * \retval NULL
-     *     Encoder does not support global header
-     * \retval Non-NULL
-     *     Pointer to buffer containing global header packet
-     */
-    vpx_fixed_buf_t *vpx_codec_get_global_headers(vpx_codec_ctx_t   *ctx);
+  /*!\brief Get global stream headers
+   *
+   * Retrieves a stream level global header packet, if supported by the codec.
+   *
+   * \param[in]    ctx     Pointer to this instance's context
+   *
+   * \retval NULL
+   *     Encoder does not support global header
+   * \retval Non-NULL
+   *     Pointer to buffer containing global header packet
+   */
+  vpx_fixed_buf_t *vpx_codec_get_global_headers(vpx_codec_ctx_t   *ctx);
 
 
 #define VPX_DL_REALTIME     (1)        /**< deadline parameter analogous to
-    *   VPx REALTIME mode. */
+  *   VPx REALTIME mode. */
 #define VPX_DL_GOOD_QUALITY (1000000)  /**< deadline parameter analogous to
-    *   VPx GOOD QUALITY mode. */
+  *   VPx GOOD QUALITY mode. */
 #define VPX_DL_BEST_QUALITY (0)        /**< deadline parameter analogous to
-    *   VPx BEST QUALITY mode. */
-    /*!\brief Encode a frame
-     *
-     * Encodes a video frame at the given "presentation time." The presentation
-     * time stamp (PTS) \ref MUST be strictly increasing.
-     *
-     * The encoder supports the notion of a soft real-time deadline. Given a
-     * non-zero value to the deadline parameter, the encoder will make a "best
-     * effort" guarantee to  return before the given time slice expires. It is
-     * implicit that limiting the available time to encode will degrade the
-     * output quality. The encoder can be given an unlimited time to produce the
-     * best possible frame by specifying a deadline of '0'. This deadline
-     * supercedes the VPx notion of "best quality, good quality, realtime".
-     * Applications that wish to map these former settings to the new deadline
-     * based system can use the symbols #VPX_DL_REALTIME, #VPX_DL_GOOD_QUALITY,
-     * and #VPX_DL_BEST_QUALITY.
-     *
-     * When the last frame has been passed to the encoder, this function should
-     * continue to be called, with the img parameter set to NULL. This will
-     * signal the end-of-stream condition to the encoder and allow it to encode
-     * any held buffers. Encoding is complete when vpx_codec_encode() is called
-     * and vpx_codec_get_cx_data() returns no data.
-     *
-     * \param[in]    ctx       Pointer to this instance's context
-     * \param[in]    img       Image data to encode, NULL to flush.
-     * \param[in]    pts       Presentation time stamp, in timebase units.
-     * \param[in]    duration  Duration to show frame, in timebase units.
-     * \param[in]    flags     Flags to use for encoding this frame.
-     * \param[in]    deadline  Time to spend encoding, in microseconds. (0=infinite)
-     *
-     * \retval #VPX_CODEC_OK
-     *     The configuration was populated.
-     * \retval #VPX_CODEC_INCAPABLE
-     *     Interface is not an encoder interface.
-     * \retval #VPX_CODEC_INVALID_PARAM
-     *     A parameter was NULL, the image format is unsupported, etc.
-     */
-    vpx_codec_err_t  vpx_codec_encode(vpx_codec_ctx_t            *ctx,
-                                      const vpx_image_t          *img,
-                                      vpx_codec_pts_t             pts,
-                                      unsigned long               duration,
-                                      vpx_enc_frame_flags_t       flags,
-                                      unsigned long               deadline);
+  *   VPx BEST QUALITY mode. */
+  /*!\brief Encode a frame
+   *
+   * Encodes a video frame at the given "presentation time." The presentation
+   * time stamp (PTS) \ref MUST be strictly increasing.
+   *
+   * The encoder supports the notion of a soft real-time deadline. Given a
+   * non-zero value to the deadline parameter, the encoder will make a "best
+   * effort" guarantee to  return before the given time slice expires. It is
+   * implicit that limiting the available time to encode will degrade the
+   * output quality. The encoder can be given an unlimited time to produce the
+   * best possible frame by specifying a deadline of '0'. This deadline
+   * supercedes the VPx notion of "best quality, good quality, realtime".
+   * Applications that wish to map these former settings to the new deadline
+   * based system can use the symbols #VPX_DL_REALTIME, #VPX_DL_GOOD_QUALITY,
+   * and #VPX_DL_BEST_QUALITY.
+   *
+   * When the last frame has been passed to the encoder, this function should
+   * continue to be called, with the img parameter set to NULL. This will
+   * signal the end-of-stream condition to the encoder and allow it to encode
+   * any held buffers. Encoding is complete when vpx_codec_encode() is called
+   * and vpx_codec_get_cx_data() returns no data.
+   *
+   * \param[in]    ctx       Pointer to this instance's context
+   * \param[in]    img       Image data to encode, NULL to flush.
+   * \param[in]    pts       Presentation time stamp, in timebase units.
+   * \param[in]    duration  Duration to show frame, in timebase units.
+   * \param[in]    flags     Flags to use for encoding this frame.
+   * \param[in]    deadline  Time to spend encoding, in microseconds. (0=infinite)
+   *
+   * \retval #VPX_CODEC_OK
+   *     The configuration was populated.
+   * \retval #VPX_CODEC_INCAPABLE
+   *     Interface is not an encoder interface.
+   * \retval #VPX_CODEC_INVALID_PARAM
+   *     A parameter was NULL, the image format is unsupported, etc.
+   */
+  vpx_codec_err_t  vpx_codec_encode(vpx_codec_ctx_t            *ctx,
+                                    const vpx_image_t          *img,
+                                    vpx_codec_pts_t             pts,
+                                    unsigned long               duration,
+                                    vpx_enc_frame_flags_t       flags,
+                                    unsigned long               deadline);
 
-    /*!\brief Set compressed data output buffer
-     *
-     * Sets the buffer that the codec should output the compressed data
-     * into. This call effectively sets the buffer pointer returned in the
-     * next VPX_CODEC_CX_FRAME_PKT packet. Subsequent packets will be
-     * appended into this buffer. The buffer is preserved across frames,
-     * so applications must periodically call this function after flushing
-     * the accumulated compressed data to disk or to the network to reset
-     * the pointer to the buffer's head.
-     *
-     * `pad_before` bytes will be skipped before writing the compressed
-     * data, and `pad_after` bytes will be appended to the packet. The size
-     * of the packet will be the sum of the size of the actual compressed
-     * data, pad_before, and pad_after. The padding bytes will be preserved
-     * (not overwritten).
-     *
-     * Note that calling this function does not guarantee that the returned
-     * compressed data will be placed into the specified buffer. In the
-     * event that the encoded data will not fit into the buffer provided,
-     * the returned packet \ref MAY point to an internal buffer, as it would
-     * if this call were never used. In this event, the output packet will
-     * NOT have any padding, and the application must free space and copy it
-     * to the proper place. This is of particular note in configurations
-     * that may output multiple packets for a single encoded frame (e.g., lagged
-     * encoding) or if the application does not reset the buffer periodically.
-     *
-     * Applications may restore the default behavior of the codec providing
-     * the compressed data buffer by calling this function with a NULL
-     * buffer.
-     *
-     * Applications \ref MUSTNOT call this function during iteration of
-     * vpx_codec_get_cx_data().
-     *
-     * \param[in]    ctx         Pointer to this instance's context
-     * \param[in]    buf         Buffer to store compressed data into
-     * \param[in]    pad_before  Bytes to skip before writing compressed data
-     * \param[in]    pad_after   Bytes to skip after writing compressed data
-     *
-     * \retval #VPX_CODEC_OK
-     *     The buffer was set successfully.
-     * \retval #VPX_CODEC_INVALID_PARAM
-     *     A parameter was NULL, the image format is unsupported, etc.
-     */
-    vpx_codec_err_t vpx_codec_set_cx_data_buf(vpx_codec_ctx_t       *ctx,
-            const vpx_fixed_buf_t *buf,
-            unsigned int           pad_before,
-            unsigned int           pad_after);
+  /*!\brief Set compressed data output buffer
+   *
+   * Sets the buffer that the codec should output the compressed data
+   * into. This call effectively sets the buffer pointer returned in the
+   * next VPX_CODEC_CX_FRAME_PKT packet. Subsequent packets will be
+   * appended into this buffer. The buffer is preserved across frames,
+   * so applications must periodically call this function after flushing
+   * the accumulated compressed data to disk or to the network to reset
+   * the pointer to the buffer's head.
+   *
+   * `pad_before` bytes will be skipped before writing the compressed
+   * data, and `pad_after` bytes will be appended to the packet. The size
+   * of the packet will be the sum of the size of the actual compressed
+   * data, pad_before, and pad_after. The padding bytes will be preserved
+   * (not overwritten).
+   *
+   * Note that calling this function does not guarantee that the returned
+   * compressed data will be placed into the specified buffer. In the
+   * event that the encoded data will not fit into the buffer provided,
+   * the returned packet \ref MAY point to an internal buffer, as it would
+   * if this call were never used. In this event, the output packet will
+   * NOT have any padding, and the application must free space and copy it
+   * to the proper place. This is of particular note in configurations
+   * that may output multiple packets for a single encoded frame (e.g., lagged
+   * encoding) or if the application does not reset the buffer periodically.
+   *
+   * Applications may restore the default behavior of the codec providing
+   * the compressed data buffer by calling this function with a NULL
+   * buffer.
+   *
+   * Applications \ref MUSTNOT call this function during iteration of
+   * vpx_codec_get_cx_data().
+   *
+   * \param[in]    ctx         Pointer to this instance's context
+   * \param[in]    buf         Buffer to store compressed data into
+   * \param[in]    pad_before  Bytes to skip before writing compressed data
+   * \param[in]    pad_after   Bytes to skip after writing compressed data
+   *
+   * \retval #VPX_CODEC_OK
+   *     The buffer was set successfully.
+   * \retval #VPX_CODEC_INVALID_PARAM
+   *     A parameter was NULL, the image format is unsupported, etc.
+   */
+  vpx_codec_err_t vpx_codec_set_cx_data_buf(vpx_codec_ctx_t       *ctx,
+                                            const vpx_fixed_buf_t *buf,
+                                            unsigned int           pad_before,
+                                            unsigned int           pad_after);
 
 
-    /*!\brief Encoded data iterator
-     *
-     * Iterates over a list of data packets to be passed from the encoder to the
-     * application. The different kinds of packets available are enumerated in
-     * #vpx_codec_cx_pkt_kind.
-     *
-     * #VPX_CODEC_CX_FRAME_PKT packets should be passed to the application's
-     * muxer. Multiple compressed frames may be in the list.
-     * #VPX_CODEC_STATS_PKT packets should be appended to a global buffer.
-     *
-     * The application \ref MUST silently ignore any packet kinds that it does
-     * not recognize or support.
-     *
-     * The data buffers returned from this function are only guaranteed to be
-     * valid until the application makes another call to any vpx_codec_* function.
-     *
-     * \param[in]     ctx      Pointer to this instance's context
-     * \param[in,out] iter     Iterator storage, initialized to NULL
-     *
-     * \return Returns a pointer to an output data packet (compressed frame data,
-     *         two-pass statistics, etc.) or NULL to signal end-of-list.
-     *
-     */
-    const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t   *ctx,
-            vpx_codec_iter_t  *iter);
+  /*!\brief Encoded data iterator
+   *
+   * Iterates over a list of data packets to be passed from the encoder to the
+   * application. The different kinds of packets available are enumerated in
+   * #vpx_codec_cx_pkt_kind.
+   *
+   * #VPX_CODEC_CX_FRAME_PKT packets should be passed to the application's
+   * muxer. Multiple compressed frames may be in the list.
+   * #VPX_CODEC_STATS_PKT packets should be appended to a global buffer.
+   *
+   * The application \ref MUST silently ignore any packet kinds that it does
+   * not recognize or support.
+   *
+   * The data buffers returned from this function are only guaranteed to be
+   * valid until the application makes another call to any vpx_codec_* function.
+   *
+   * \param[in]     ctx      Pointer to this instance's context
+   * \param[in,out] iter     Iterator storage, initialized to NULL
+   *
+   * \return Returns a pointer to an output data packet (compressed frame data,
+   *         two-pass statistics, etc.) or NULL to signal end-of-list.
+   *
+   */
+  const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t   *ctx,
+                                                  vpx_codec_iter_t  *iter);
 
 
-    /*!\brief Get Preview Frame
-     *
-     * Returns an image that can be used as a preview. Shows the image as it would
-     * exist at the decompressor. The application \ref MUST NOT write into this
-     * image buffer.
-     *
-     * \param[in]     ctx      Pointer to this instance's context
-     *
-     * \return Returns a pointer to a preview image, or NULL if no image is
-     *         available.
-     *
-     */
-    const vpx_image_t *vpx_codec_get_preview_frame(vpx_codec_ctx_t   *ctx);
+  /*!\brief Get Preview Frame
+   *
+   * Returns an image that can be used as a preview. Shows the image as it would
+   * exist at the decompressor. The application \ref MUST NOT write into this
+   * image buffer.
+   *
+   * \param[in]     ctx      Pointer to this instance's context
+   *
+   * \return Returns a pointer to a preview image, or NULL if no image is
+   *         available.
+   *
+   */
+  const vpx_image_t *vpx_codec_get_preview_frame(vpx_codec_ctx_t   *ctx);
 
 
-    /*!@} - end defgroup encoder*/
+  /*!@} - end defgroup encoder*/
 
 #endif
 #ifdef __cplusplus
diff --git a/libvpx/vpx/vpx_image.h b/libvpx/vpx/vpx_image.h
index 3e42447..c304bac 100644
--- a/libvpx/vpx/vpx_image.h
+++ b/libvpx/vpx/vpx_image.h
@@ -20,14 +20,14 @@
 #ifndef VPX_IMAGE_H
 #define VPX_IMAGE_H
 
-    /*!\brief Current ABI version number
-     *
-     * \internal
-     * If this file is altered in any way that changes the ABI, this value
-     * must be bumped.  Examples include, but are not limited to, changing
-     * types, removing or reassigning enums, adding/removing/rearranging
-     * fields to structures
-     */
+  /*!\brief Current ABI version number
+   *
+   * \internal
+   * If this file is altered in any way that changes the ABI, this value
+   * must be bumped.  Examples include, but are not limited to, changing
+   * types, removing or reassigning enums, adding/removing/rearranging
+   * fields to structures
+   */
 #define VPX_IMAGE_ABI_VERSION (1) /**<\hideinitializer*/
 
 
@@ -36,41 +36,43 @@
 #define VPX_IMG_FMT_HAS_ALPHA  0x400  /**< Image has an alpha channel component */
 
 
-    /*!\brief List of supported image formats */
-    typedef enum vpx_img_fmt {
-        VPX_IMG_FMT_NONE,
-        VPX_IMG_FMT_RGB24,   /**< 24 bit per pixel packed RGB */
-        VPX_IMG_FMT_RGB32,   /**< 32 bit per pixel packed 0RGB */
-        VPX_IMG_FMT_RGB565,  /**< 16 bit per pixel, 565 */
-        VPX_IMG_FMT_RGB555,  /**< 16 bit per pixel, 555 */
-        VPX_IMG_FMT_UYVY,    /**< UYVY packed YUV */
-        VPX_IMG_FMT_YUY2,    /**< YUYV packed YUV */
-        VPX_IMG_FMT_YVYU,    /**< YVYU packed YUV */
-        VPX_IMG_FMT_BGR24,   /**< 24 bit per pixel packed BGR */
-        VPX_IMG_FMT_RGB32_LE, /**< 32 bit packed BGR0 */
-        VPX_IMG_FMT_ARGB,     /**< 32 bit packed ARGB, alpha=255 */
-        VPX_IMG_FMT_ARGB_LE,  /**< 32 bit packed BGRA, alpha=255 */
-        VPX_IMG_FMT_RGB565_LE,  /**< 16 bit per pixel, gggbbbbb rrrrrggg */
-        VPX_IMG_FMT_RGB555_LE,  /**< 16 bit per pixel, gggbbbbb 0rrrrrgg */
-        VPX_IMG_FMT_YV12    = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 1, /**< planar YVU */
-        VPX_IMG_FMT_I420    = VPX_IMG_FMT_PLANAR | 2,
-        VPX_IMG_FMT_VPXYV12 = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 3, /** < planar 4:2:0 format with vpx color space */
-        VPX_IMG_FMT_VPXI420 = VPX_IMG_FMT_PLANAR | 4   /** < planar 4:2:0 format with vpx color space */
-    }
-    vpx_img_fmt_t; /**< alias for enum vpx_img_fmt */
+  /*!\brief List of supported image formats */
+  typedef enum vpx_img_fmt {
+    VPX_IMG_FMT_NONE,
+    VPX_IMG_FMT_RGB24,   /**< 24 bit per pixel packed RGB */
+    VPX_IMG_FMT_RGB32,   /**< 32 bit per pixel packed 0RGB */
+    VPX_IMG_FMT_RGB565,  /**< 16 bit per pixel, 565 */
+    VPX_IMG_FMT_RGB555,  /**< 16 bit per pixel, 555 */
+    VPX_IMG_FMT_UYVY,    /**< UYVY packed YUV */
+    VPX_IMG_FMT_YUY2,    /**< YUYV packed YUV */
+    VPX_IMG_FMT_YVYU,    /**< YVYU packed YUV */
+    VPX_IMG_FMT_BGR24,   /**< 24 bit per pixel packed BGR */
+    VPX_IMG_FMT_RGB32_LE, /**< 32 bit packed BGR0 */
+    VPX_IMG_FMT_ARGB,     /**< 32 bit packed ARGB, alpha=255 */
+    VPX_IMG_FMT_ARGB_LE,  /**< 32 bit packed BGRA, alpha=255 */
+    VPX_IMG_FMT_RGB565_LE,  /**< 16 bit per pixel, gggbbbbb rrrrrggg */
+    VPX_IMG_FMT_RGB555_LE,  /**< 16 bit per pixel, gggbbbbb 0rrrrrgg */
+    VPX_IMG_FMT_YV12    = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 1, /**< planar YVU */
+    VPX_IMG_FMT_I420    = VPX_IMG_FMT_PLANAR | 2,
+    VPX_IMG_FMT_VPXYV12 = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_UV_FLIP | 3, /** < planar 4:2:0 format with vpx color space */
+    VPX_IMG_FMT_VPXI420 = VPX_IMG_FMT_PLANAR | 4,
+    VPX_IMG_FMT_I422    = VPX_IMG_FMT_PLANAR | 5,
+    VPX_IMG_FMT_I444    = VPX_IMG_FMT_PLANAR | 6,
+    VPX_IMG_FMT_444A    = VPX_IMG_FMT_PLANAR | VPX_IMG_FMT_HAS_ALPHA | 7
+  } vpx_img_fmt_t; /**< alias for enum vpx_img_fmt */
 
 #if !defined(VPX_CODEC_DISABLE_COMPAT) || !VPX_CODEC_DISABLE_COMPAT
 #define IMG_FMT_PLANAR         VPX_IMG_FMT_PLANAR     /**< \deprecated Use #VPX_IMG_FMT_PLANAR */
 #define IMG_FMT_UV_FLIP        VPX_IMG_FMT_UV_FLIP    /**< \deprecated Use #VPX_IMG_FMT_UV_FLIP */
 #define IMG_FMT_HAS_ALPHA      VPX_IMG_FMT_HAS_ALPHA  /**< \deprecated Use #VPX_IMG_FMT_HAS_ALPHA */
 
-    /*!\brief Deprecated list of supported image formats
-     * \deprecated New code should use #vpx_img_fmt
-     */
+  /*!\brief Deprecated list of supported image formats
+   * \deprecated New code should use #vpx_img_fmt
+   */
 #define img_fmt   vpx_img_fmt
-    /*!\brief alias for enum img_fmt.
-     * \deprecated New code should use #vpx_img_fmt_t
-     */
+  /*!\brief alias for enum img_fmt.
+   * \deprecated New code should use #vpx_img_fmt_t
+   */
 #define img_fmt_t vpx_img_fmt_t
 
 #define IMG_FMT_NONE       VPX_IMG_FMT_NONE       /**< \deprecated Use #VPX_IMG_FMT_NONE */
@@ -93,24 +95,23 @@
 #define IMG_FMT_VPXI420    VPX_IMG_FMT_VPXI420    /**< \deprecated Use #VPX_IMG_FMT_VPXI420 */
 #endif /* VPX_CODEC_DISABLE_COMPAT */
 
-    /**\brief Image Descriptor */
-    typedef struct vpx_image
-    {
-        vpx_img_fmt_t fmt; /**< Image Format */
+  /**\brief Image Descriptor */
+  typedef struct vpx_image {
+    vpx_img_fmt_t fmt; /**< Image Format */
 
-        /* Image storage dimensions */
-        unsigned int  w;   /**< Stored image width */
-        unsigned int  h;   /**< Stored image height */
+    /* Image storage dimensions */
+    unsigned int  w;   /**< Stored image width */
+    unsigned int  h;   /**< Stored image height */
 
-        /* Image display dimensions */
-        unsigned int  d_w;   /**< Displayed image width */
-        unsigned int  d_h;   /**< Displayed image height */
+    /* Image display dimensions */
+    unsigned int  d_w;   /**< Displayed image width */
+    unsigned int  d_h;   /**< Displayed image height */
 
-        /* Chroma subsampling info */
-        unsigned int  x_chroma_shift;   /**< subsampling order, X */
-        unsigned int  y_chroma_shift;   /**< subsampling order, Y */
+    /* Chroma subsampling info */
+    unsigned int  x_chroma_shift;   /**< subsampling order, X */
+    unsigned int  y_chroma_shift;   /**< subsampling order, Y */
 
-        /* Image data pointers. */
+    /* Image data pointers. */
 #define VPX_PLANE_PACKED 0   /**< To be used for all packed formats */
 #define VPX_PLANE_Y      0   /**< Y (Luminance) plane */
 #define VPX_PLANE_U      1   /**< U (Chroma) plane */
@@ -123,119 +124,118 @@
 #define PLANE_V          VPX_PLANE_V
 #define PLANE_ALPHA      VPX_PLANE_ALPHA
 #endif
-        unsigned char *planes[4];  /**< pointer to the top left pixel for each plane */
-        int      stride[4];  /**< stride between rows for each plane */
+    unsigned char *planes[4];  /**< pointer to the top left pixel for each plane */
+    int      stride[4];  /**< stride between rows for each plane */
 
-        int     bps; /**< bits per sample (for packed formats) */
+    int     bps; /**< bits per sample (for packed formats) */
 
-        /* The following member may be set by the application to associate data
-         * with this image.
-         */
-        void    *user_priv; /**< may be set by the application to associate data
+    /* The following member may be set by the application to associate data
+     * with this image.
+     */
+    void    *user_priv; /**< may be set by the application to associate data
                          *   with this image. */
 
-        /* The following members should be treated as private. */
-        unsigned char *img_data;       /**< private */
-        int      img_data_owner; /**< private */
-        int      self_allocd;    /**< private */
-    } vpx_image_t; /**< alias for struct vpx_image */
+    /* The following members should be treated as private. */
+    unsigned char *img_data;       /**< private */
+    int      img_data_owner; /**< private */
+    int      self_allocd;    /**< private */
+  } vpx_image_t; /**< alias for struct vpx_image */
 
-    /**\brief Representation of a rectangle on a surface */
-    typedef struct vpx_image_rect
-    {
-        unsigned int x; /**< leftmost column */
-        unsigned int y; /**< topmost row */
-        unsigned int w; /**< width */
-        unsigned int h; /**< height */
-    } vpx_image_rect_t; /**< alias for struct vpx_image_rect */
+  /**\brief Representation of a rectangle on a surface */
+  typedef struct vpx_image_rect {
+    unsigned int x; /**< leftmost column */
+    unsigned int y; /**< topmost row */
+    unsigned int w; /**< width */
+    unsigned int h; /**< height */
+  } vpx_image_rect_t; /**< alias for struct vpx_image_rect */
 
-    /*!\brief Open a descriptor, allocating storage for the underlying image
-     *
-     * Returns a descriptor for storing an image of the given format. The
-     * storage for the descriptor is allocated on the heap.
-     *
-     * \param[in]    img       Pointer to storage for descriptor. If this parameter
-     *                         is NULL, the storage for the descriptor will be
-     *                         allocated on the heap.
-     * \param[in]    fmt       Format for the image
-     * \param[in]    d_w       Width of the image
-     * \param[in]    d_h       Height of the image
-     * \param[in]    align     Alignment, in bytes, of the image buffer and
-     *                         each row in the image(stride).
-     *
-     * \return Returns a pointer to the initialized image descriptor. If the img
-     *         parameter is non-null, the value of the img parameter will be
-     *         returned.
-     */
-    vpx_image_t *vpx_img_alloc(vpx_image_t  *img,
-                               vpx_img_fmt_t fmt,
-                               unsigned int d_w,
-                               unsigned int d_h,
-                               unsigned int align);
+  /*!\brief Open a descriptor, allocating storage for the underlying image
+   *
+   * Returns a descriptor for storing an image of the given format. The
+   * storage for the descriptor is allocated on the heap.
+   *
+   * \param[in]    img       Pointer to storage for descriptor. If this parameter
+   *                         is NULL, the storage for the descriptor will be
+   *                         allocated on the heap.
+   * \param[in]    fmt       Format for the image
+   * \param[in]    d_w       Width of the image
+   * \param[in]    d_h       Height of the image
+   * \param[in]    align     Alignment, in bytes, of the image buffer and
+   *                         each row in the image(stride).
+   *
+   * \return Returns a pointer to the initialized image descriptor. If the img
+   *         parameter is non-null, the value of the img parameter will be
+   *         returned.
+   */
+  vpx_image_t *vpx_img_alloc(vpx_image_t  *img,
+                             vpx_img_fmt_t fmt,
+                             unsigned int d_w,
+                             unsigned int d_h,
+                             unsigned int align);
 
-    /*!\brief Open a descriptor, using existing storage for the underlying image
-     *
-     * Returns a descriptor for storing an image of the given format. The
-     * storage for descriptor has been allocated elsewhere, and a descriptor is
-     * desired to "wrap" that storage.
-     *
-     * \param[in]    img       Pointer to storage for descriptor. If this parameter
-     *                         is NULL, the storage for the descriptor will be
-     *                         allocated on the heap.
-     * \param[in]    fmt       Format for the image
-     * \param[in]    d_w       Width of the image
-     * \param[in]    d_h       Height of the image
-     * \param[in]    align     Alignment, in bytes, of each row in the image.
-     * \param[in]    img_data  Storage to use for the image
-     *
-     * \return Returns a pointer to the initialized image descriptor. If the img
-     *         parameter is non-null, the value of the img parameter will be
-     *         returned.
-     */
-    vpx_image_t *vpx_img_wrap(vpx_image_t  *img,
-                              vpx_img_fmt_t fmt,
-                              unsigned int d_w,
-                              unsigned int d_h,
-                              unsigned int align,
-                              unsigned char      *img_data);
+  /*!\brief Open a descriptor, using existing storage for the underlying image
+   *
+   * Returns a descriptor for storing an image of the given format. The
+   * storage for descriptor has been allocated elsewhere, and a descriptor is
+   * desired to "wrap" that storage.
+   *
+   * \param[in]    img       Pointer to storage for descriptor. If this parameter
+   *                         is NULL, the storage for the descriptor will be
+   *                         allocated on the heap.
+   * \param[in]    fmt       Format for the image
+   * \param[in]    d_w       Width of the image
+   * \param[in]    d_h       Height of the image
+   * \param[in]    align     Alignment, in bytes, of each row in the image.
+   * \param[in]    img_data  Storage to use for the image
+   *
+   * \return Returns a pointer to the initialized image descriptor. If the img
+   *         parameter is non-null, the value of the img parameter will be
+   *         returned.
+   */
+  vpx_image_t *vpx_img_wrap(vpx_image_t  *img,
+                            vpx_img_fmt_t fmt,
+                            unsigned int d_w,
+                            unsigned int d_h,
+                            unsigned int align,
+                            unsigned char      *img_data);
 
 
-    /*!\brief Set the rectangle identifying the displayed portion of the image
-     *
-     * Updates the displayed rectangle (aka viewport) on the image surface to
-     * match the specified coordinates and size.
-     *
-     * \param[in]    img       Image descriptor
-     * \param[in]    x         leftmost column
-     * \param[in]    y         topmost row
-     * \param[in]    w         width
-     * \param[in]    h         height
-     *
-     * \return 0 if the requested rectangle is valid, nonzero otherwise.
-     */
-    int vpx_img_set_rect(vpx_image_t  *img,
-                         unsigned int  x,
-                         unsigned int  y,
-                         unsigned int  w,
-                         unsigned int  h);
+  /*!\brief Set the rectangle identifying the displayed portion of the image
+   *
+   * Updates the displayed rectangle (aka viewport) on the image surface to
+   * match the specified coordinates and size.
+   *
+   * \param[in]    img       Image descriptor
+   * \param[in]    x         leftmost column
+   * \param[in]    y         topmost row
+   * \param[in]    w         width
+   * \param[in]    h         height
+   *
+   * \return 0 if the requested rectangle is valid, nonzero otherwise.
+   */
+  int vpx_img_set_rect(vpx_image_t  *img,
+                       unsigned int  x,
+                       unsigned int  y,
+                       unsigned int  w,
+                       unsigned int  h);
 
 
-    /*!\brief Flip the image vertically (top for bottom)
-     *
-     * Adjusts the image descriptor's pointers and strides to make the image
-     * be referenced upside-down.
-     *
-     * \param[in]    img       Image descriptor
-     */
-    void vpx_img_flip(vpx_image_t *img);
+  /*!\brief Flip the image vertically (top for bottom)
+   *
+   * Adjusts the image descriptor's pointers and strides to make the image
+   * be referenced upside-down.
+   *
+   * \param[in]    img       Image descriptor
+   */
+  void vpx_img_flip(vpx_image_t *img);
 
-    /*!\brief Close an image descriptor
-     *
-     * Frees all allocated storage associated with an image descriptor.
-     *
-     * \param[in]    img       Image descriptor
-     */
-    void vpx_img_free(vpx_image_t *img);
+  /*!\brief Close an image descriptor
+   *
+   * Frees all allocated storage associated with an image descriptor.
+   *
+   * \param[in]    img       Image descriptor
+   */
+  void vpx_img_free(vpx_image_t *img);
 
 #endif
 #ifdef __cplusplus
diff --git a/libvpx/vpx/vpx_integer.h b/libvpx/vpx/vpx_integer.h
index 218bca7..0ccc96c 100644
--- a/libvpx/vpx/vpx_integer.h
+++ b/libvpx/vpx/vpx_integer.h
@@ -27,6 +27,9 @@
 #if (defined(_MSC_VER) && (_MSC_VER < 1600))
 typedef signed __int64   int64_t;
 typedef unsigned __int64 uint64_t;
+#define INT64_MAX _I64_MAX
+#define INT16_MAX _I16_MAX
+#define INT16_MIN _I16_MIN
 #endif
 
 #ifndef _UINTPTR_T_DEFINED
diff --git a/libvpx/vpx_mem/include/vpx_mem_intrnl.h b/libvpx/vpx_mem/include/vpx_mem_intrnl.h
index 63c6b77..60b5165 100644
--- a/libvpx/vpx_mem/include/vpx_mem_intrnl.h
+++ b/libvpx/vpx_mem/include/vpx_mem_intrnl.h
@@ -11,7 +11,7 @@
 
 #ifndef __VPX_MEM_INTRNL_H__
 #define __VPX_MEM_INTRNL_H__
-#include "vpx_config.h"
+#include "./vpx_config.h"
 
 #ifndef CONFIG_MEM_MANAGER
 # if defined(VXWORKS)
@@ -47,8 +47,8 @@
 #ifndef DEFAULT_ALIGNMENT
 # if defined(VXWORKS)
 #  define DEFAULT_ALIGNMENT        32        /*default addr alignment to use in
-                                               calls to vpx_* functions other
-                                               than vpx_memalign*/
+calls to vpx_* functions other
+than vpx_memalign*/
 # else
 #  define DEFAULT_ALIGNMENT        1
 # endif
@@ -60,9 +60,9 @@
 
 #if CONFIG_MEM_TRACKER
 # define TRY_BOUNDS_CHECK         1        /*when set to 1 pads each allocation,
-                                             integrity can be checked using
-                                             vpx_memory_tracker_check_integrity
-                                             or on free by defining*/
+integrity can be checked using
+vpx_memory_tracker_check_integrity
+or on free by defining*/
 /*TRY_BOUNDS_CHECK_ON_FREE*/
 #else
 # define TRY_BOUNDS_CHECK         0
@@ -70,13 +70,13 @@
 
 #if TRY_BOUNDS_CHECK
 # define TRY_BOUNDS_CHECK_ON_FREE 0          /*checks mem integrity on every
-                                               free, very expensive*/
+free, very expensive*/
 # define BOUNDS_CHECK_VALUE       0xdeadbeef /*value stored before/after ea.
-                                               mem addr for bounds checking*/
+mem addr for bounds checking*/
 # define BOUNDS_CHECK_PAD_SIZE    32         /*size of the padding before and
-                                               after ea allocation to be filled
-                                               with BOUNDS_CHECK_VALUE.
-                                               this should be a multiple of 4*/
+after ea allocation to be filled
+with BOUNDS_CHECK_VALUE.
+this should be a multiple of 4*/
 #else
 # define BOUNDS_CHECK_VALUE       0
 # define BOUNDS_CHECK_PAD_SIZE    0
diff --git a/libvpx/vpx_mem/include/vpx_mem_tracker.h b/libvpx/vpx_mem/include/vpx_mem_tracker.h
index ef2b29b..3be0d2d 100644
--- a/libvpx/vpx_mem/include/vpx_mem_tracker.h
+++ b/libvpx/vpx_mem/include/vpx_mem_tracker.h
@@ -23,158 +23,157 @@
 
 #include <stdarg.h>
 
-struct mem_block
-{
-    size_t addr;
-    unsigned int size,
-             line;
-    char *file;
-    struct mem_block *prev,
-            * next;
+struct mem_block {
+  size_t addr;
+  unsigned int size,
+           line;
+  char *file;
+  struct mem_block *prev,
+      * next;
 
-    int padded; // This mem_block has padding for integrity checks.
-    // As of right now, this should only be 0 if
-    // using vpx_mem_alloc to allocate cache memory.
-    // 2005-01-11 tjf
+  int padded; // This mem_block has padding for integrity checks.
+  // As of right now, this should only be 0 if
+  // using vpx_mem_alloc to allocate cache memory.
+  // 2005-01-11 tjf
 };
 
 #if defined(__cplusplus)
 extern "C" {
 #endif
 
-    /*
-        vpx_memory_tracker_init(int padding_size, int pad_value)
-          padding_size - the size of the padding before and after each mem addr.
-                         Values > 0 indicate that integrity checks can be performed
-                         by inspecting these areas.
-          pad_value - the initial value within the padding area before and after
-                      each mem addr.
+  /*
+      vpx_memory_tracker_init(int padding_size, int pad_value)
+        padding_size - the size of the padding before and after each mem addr.
+                       Values > 0 indicate that integrity checks can be performed
+                       by inspecting these areas.
+        pad_value - the initial value within the padding area before and after
+                    each mem addr.
 
-        Initializes the memory tracker interface. Should be called before any
-        other calls to the memory tracker.
-    */
-    int vpx_memory_tracker_init(int padding_size, int pad_value);
+      Initializes the memory tracker interface. Should be called before any
+      other calls to the memory tracker.
+  */
+  int vpx_memory_tracker_init(int padding_size, int pad_value);
 
-    /*
-        vpx_memory_tracker_destroy()
-        Deinitializes the memory tracker interface
-    */
-    void vpx_memory_tracker_destroy();
+  /*
+      vpx_memory_tracker_destroy()
+      Deinitializes the memory tracker interface
+  */
+  void vpx_memory_tracker_destroy();
 
-    /*
-        vpx_memory_tracker_add(size_t addr, unsigned int size,
-                             char * file, unsigned int line)
-          addr - memory address to be added to list
-          size - size of addr
-          file - the file addr was referenced from
-          line - the line in file addr was referenced from
-        Adds memory address addr, it's size, file and line it came from
-        to the memory tracker allocation table
-    */
-    void vpx_memory_tracker_add(size_t addr, unsigned int size,
-                                char *file, unsigned int line,
-                                int padded);
+  /*
+      vpx_memory_tracker_add(size_t addr, unsigned int size,
+                           char * file, unsigned int line)
+        addr - memory address to be added to list
+        size - size of addr
+        file - the file addr was referenced from
+        line - the line in file addr was referenced from
+      Adds memory address addr, it's size, file and line it came from
+      to the memory tracker allocation table
+  */
+  void vpx_memory_tracker_add(size_t addr, unsigned int size,
+                              char *file, unsigned int line,
+                              int padded);
 
-    /*
-        vpx_memory_tracker_add(size_t addr, unsigned int size, char * file, unsigned int line)
-          addr - memory address to be added to be removed
-          padded - if 0, disables bounds checking on this memory block even if bounds
-          checking is enabled. (for example, when allocating cache memory, we still want
-          to check for memory leaks, but we do not waste cache space for bounds check padding)
-        Removes the specified address from the memory tracker's allocation
-        table
-        Return:
-          0: on success
-          -1: if memory allocation table's mutex could not be locked
-          -2: if the addr was not found in the list
-    */
-    int vpx_memory_tracker_remove(size_t addr);
+  /*
+      vpx_memory_tracker_add(size_t addr, unsigned int size, char * file, unsigned int line)
+        addr - memory address to be added to be removed
+        padded - if 0, disables bounds checking on this memory block even if bounds
+        checking is enabled. (for example, when allocating cache memory, we still want
+        to check for memory leaks, but we do not waste cache space for bounds check padding)
+      Removes the specified address from the memory tracker's allocation
+      table
+      Return:
+        0: on success
+        -1: if memory allocation table's mutex could not be locked
+        -2: if the addr was not found in the list
+  */
+  int vpx_memory_tracker_remove(size_t addr);
 
-    /*
-        vpx_memory_tracker_find(unsigned int addr)
-          addr - address to be found in the memory tracker's
-                 allocation table
-        Return:
-            If found, pointer to the memory block that matches addr
-            NULL otherwise
-    */
-    struct mem_block *vpx_memory_tracker_find(size_t addr);
+  /*
+      vpx_memory_tracker_find(unsigned int addr)
+        addr - address to be found in the memory tracker's
+               allocation table
+      Return:
+          If found, pointer to the memory block that matches addr
+          NULL otherwise
+  */
+  struct mem_block *vpx_memory_tracker_find(size_t addr);
 
-    /*
-        vpx_memory_tracker_dump()
-        Dumps the current contents of the memory
-        tracker allocation table
-    */
-    void vpx_memory_tracker_dump();
+  /*
+      vpx_memory_tracker_dump()
+      Dumps the current contents of the memory
+      tracker allocation table
+  */
+  void vpx_memory_tracker_dump();
 
-    /*
-        vpx_memory_tracker_check_integrity()
-        If a padding_size was provided to vpx_memory_tracker_init()
-        This function will verify that the region before and after each
-        memory address contains the specified pad_value. Should the check
-        fail, the filename and line of the check will be printed out.
-    */
-    void vpx_memory_tracker_check_integrity(char *file, unsigned int line);
+  /*
+      vpx_memory_tracker_check_integrity()
+      If a padding_size was provided to vpx_memory_tracker_init()
+      This function will verify that the region before and after each
+      memory address contains the specified pad_value. Should the check
+      fail, the filename and line of the check will be printed out.
+  */
+  void vpx_memory_tracker_check_integrity(char *file, unsigned int line);
 
-    /*
-        vpx_memory_tracker_set_log_type
-          type - value representing the logging type to use
-          option - type specific option. This will be interpreted differently
-                   based on the type.
-        Sets the logging type for the memory tracker.
-        Values currently supported:
-          0: if option is NULL, log to stderr, otherwise interpret option as a
-             filename and attempt to open it.
-          1: Use output_debug_string (WIN32 only), option ignored
-        Return:
-          0: on success
-          -1: if the logging type could not be set, because the value was invalid
-              or because a file could not be opened
-    */
-    int vpx_memory_tracker_set_log_type(int type, char *option);
+  /*
+      vpx_memory_tracker_set_log_type
+        type - value representing the logging type to use
+        option - type specific option. This will be interpreted differently
+                 based on the type.
+      Sets the logging type for the memory tracker.
+      Values currently supported:
+        0: if option is NULL, log to stderr, otherwise interpret option as a
+           filename and attempt to open it.
+        1: Use output_debug_string (WIN32 only), option ignored
+      Return:
+        0: on success
+        -1: if the logging type could not be set, because the value was invalid
+            or because a file could not be opened
+  */
+  int vpx_memory_tracker_set_log_type(int type, char *option);
 
-    /*
-        vpx_memory_tracker_set_log_func
-          userdata - ptr to be passed to the supplied logfunc, can be NULL
-          logfunc - the logging function to be used to output data from
-                    vpx_memory_track_dump/check_integrity
-        Sets a logging function to be used by the memory tracker.
-        Return:
-          0: on success
-          -1: if the logging type could not be set because logfunc was NULL
-    */
-    int vpx_memory_tracker_set_log_func(void *userdata,
-                                        void(*logfunc)(void *userdata,
-                                                const char *fmt, va_list args));
+  /*
+      vpx_memory_tracker_set_log_func
+        userdata - ptr to be passed to the supplied logfunc, can be NULL
+        logfunc - the logging function to be used to output data from
+                  vpx_memory_track_dump/check_integrity
+      Sets a logging function to be used by the memory tracker.
+      Return:
+        0: on success
+        -1: if the logging type could not be set because logfunc was NULL
+  */
+  int vpx_memory_tracker_set_log_func(void *userdata,
+                                      void(*logfunc)(void *userdata,
+                                                     const char *fmt, va_list args));
 
-    /* Wrappers to standard library functions. */
-    typedef void*(* mem_track_malloc_func)(size_t);
-    typedef void*(* mem_track_calloc_func)(size_t, size_t);
-    typedef void*(* mem_track_realloc_func)(void *, size_t);
-    typedef void (* mem_track_free_func)(void *);
-    typedef void*(* mem_track_memcpy_func)(void *, const void *, size_t);
-    typedef void*(* mem_track_memset_func)(void *, int, size_t);
-    typedef void*(* mem_track_memmove_func)(void *, const void *, size_t);
+  /* Wrappers to standard library functions. */
+  typedef void *(* mem_track_malloc_func)(size_t);
+  typedef void *(* mem_track_calloc_func)(size_t, size_t);
+  typedef void *(* mem_track_realloc_func)(void *, size_t);
+  typedef void (* mem_track_free_func)(void *);
+  typedef void *(* mem_track_memcpy_func)(void *, const void *, size_t);
+  typedef void *(* mem_track_memset_func)(void *, int, size_t);
+  typedef void *(* mem_track_memmove_func)(void *, const void *, size_t);
 
-    /*
-        vpx_memory_tracker_set_functions
+  /*
+      vpx_memory_tracker_set_functions
 
-        Sets the function pointers for the standard library functions.
+      Sets the function pointers for the standard library functions.
 
-        Return:
-          0: on success
-          -1: if the use global function pointers is not set.
-    */
-    int vpx_memory_tracker_set_functions(mem_track_malloc_func g_malloc_l
-                                         , mem_track_calloc_func g_calloc_l
-                                         , mem_track_realloc_func g_realloc_l
-                                         , mem_track_free_func g_free_l
-                                         , mem_track_memcpy_func g_memcpy_l
-                                         , mem_track_memset_func g_memset_l
-                                         , mem_track_memmove_func g_memmove_l);
+      Return:
+        0: on success
+        -1: if the use global function pointers is not set.
+  */
+  int vpx_memory_tracker_set_functions(mem_track_malloc_func g_malloc_l
+, mem_track_calloc_func g_calloc_l
+, mem_track_realloc_func g_realloc_l
+, mem_track_free_func g_free_l
+, mem_track_memcpy_func g_memcpy_l
+, mem_track_memset_func g_memset_l
+, mem_track_memmove_func g_memmove_l);
 
 #if defined(__cplusplus)
 }
 #endif
 
-#endif //__VPX_MEM_TRACKER_H__
+#endif // __VPX_MEM_TRACKER_H__
diff --git a/libvpx/vpx_mem/memory_manager/hmm_alloc.c b/libvpx/vpx_mem/memory_manager/hmm_alloc.c
index 22c4a54..ab3562d 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_alloc.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_alloc.c
@@ -15,46 +15,44 @@
 
 #include "hmm_intrnl.h"
 
-void *U(alloc)(U(descriptor) *desc, U(size_aau) n)
-{
+void *U(alloc)(U(descriptor) *desc, U(size_aau) n) {
 #ifdef HMM_AUDIT_FAIL
 
-    if (desc->avl_tree_root)
-        AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+  if (desc->avl_tree_root)
+    AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 #endif
 
-        if (desc->last_freed)
-        {
+    if (desc->last_freed) {
 #ifdef HMM_AUDIT_FAIL
-            AUDIT_BLOCK(desc->last_freed)
+      AUDIT_BLOCK(desc->last_freed)
 #endif
 
-            U(into_free_collection)(desc, (head_record *)(desc->last_freed));
+      U(into_free_collection)(desc, (head_record *)(desc->last_freed));
 
-            desc->last_freed = 0;
-        }
-
-    /* Add space for block header. */
-    n += HEAD_AAUS;
-
-    /* Convert n from number of address alignment units to block alignment
-    ** units. */
-    n = DIV_ROUND_UP(n, HMM_BLOCK_ALIGN_UNIT);
-
-    if (n < MIN_BLOCK_BAUS)
-        n = MIN_BLOCK_BAUS;
-
-    {
-        /* Search for the first node of the bin containing the smallest
-        ** block big enough to satisfy request. */
-        ptr_record *ptr_rec_ptr =
-            U(avl_search)(
-                (U(avl_avl) *) & (desc->avl_tree_root), (U(size_bau)) n,
-                AVL_GREATER_EQUAL);
-
-        /* If an approprate bin is found, satisfy the allocation request,
-        ** otherwise return null pointer. */
-        return(ptr_rec_ptr ?
-               U(alloc_from_bin)(desc, ptr_rec_ptr, (U(size_bau)) n) : 0);
+      desc->last_freed = 0;
     }
+
+  /* Add space for block header. */
+  n += HEAD_AAUS;
+
+  /* Convert n from number of address alignment units to block alignment
+  ** units. */
+  n = DIV_ROUND_UP(n, HMM_BLOCK_ALIGN_UNIT);
+
+  if (n < MIN_BLOCK_BAUS)
+    n = MIN_BLOCK_BAUS;
+
+  {
+    /* Search for the first node of the bin containing the smallest
+    ** block big enough to satisfy request. */
+    ptr_record *ptr_rec_ptr =
+      U(avl_search)(
+        (U(avl_avl) *) & (desc->avl_tree_root), (U(size_bau)) n,
+        AVL_GREATER_EQUAL);
+
+    /* If an approprate bin is found, satisfy the allocation request,
+    ** otherwise return null pointer. */
+    return(ptr_rec_ptr ?
+           U(alloc_from_bin)(desc, ptr_rec_ptr, (U(size_bau)) n) : 0);
+  }
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_base.c b/libvpx/vpx_mem/memory_manager/hmm_base.c
index ad1da03..0eff59d 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_base.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_base.c
@@ -15,58 +15,53 @@
 
 #include "hmm_intrnl.h"
 
-void U(init)(U(descriptor) *desc)
-{
-    desc->avl_tree_root = 0;
-    desc->last_freed = 0;
+void U(init)(U(descriptor) *desc) {
+  desc->avl_tree_root = 0;
+  desc->last_freed = 0;
 }
 
 /* Remove a free block from a bin's doubly-linked list when it is not,
 ** the first block in the bin.
 */
 void U(dll_remove)(
-    /* Pointer to pointer record in the block to be removed. */
-    ptr_record *to_remove)
-{
-    to_remove->prev->next = to_remove->next;
+  /* Pointer to pointer record in the block to be removed. */
+  ptr_record *to_remove) {
+  to_remove->prev->next = to_remove->next;
 
-    if (to_remove->next)
-        to_remove->next->prev = to_remove->prev;
+  if (to_remove->next)
+    to_remove->next->prev = to_remove->prev;
 }
 
 /* Put a block into the free collection of a heap.
 */
 void U(into_free_collection)(
-    /* Pointer to heap descriptor. */
-    U(descriptor) *desc,
-    /* Pointer to head record of block. */
-    head_record *head_ptr)
-{
-    ptr_record *ptr_rec_ptr = HEAD_TO_PTR_REC(head_ptr);
+  /* Pointer to heap descriptor. */
+  U(descriptor) *desc,
+  /* Pointer to head record of block. */
+  head_record *head_ptr) {
+  ptr_record *ptr_rec_ptr = HEAD_TO_PTR_REC(head_ptr);
 
-    ptr_record *bin_front_ptr =
-        U(avl_insert)((U(avl_avl) *) & (desc->avl_tree_root), ptr_rec_ptr);
+  ptr_record *bin_front_ptr =
+    U(avl_insert)((U(avl_avl) *) & (desc->avl_tree_root), ptr_rec_ptr);
 
-    if (bin_front_ptr != ptr_rec_ptr)
-    {
-        /* The block was not inserted into the AVL tree because there is
-        ** already a bin for the size of the block. */
+  if (bin_front_ptr != ptr_rec_ptr) {
+    /* The block was not inserted into the AVL tree because there is
+    ** already a bin for the size of the block. */
 
-        MARK_SUCCESSIVE_BLOCK_IN_FREE_BIN(head_ptr)
-        ptr_rec_ptr->self = ptr_rec_ptr;
+    MARK_SUCCESSIVE_BLOCK_IN_FREE_BIN(head_ptr)
+    ptr_rec_ptr->self = ptr_rec_ptr;
 
-        /* Make the block the new second block in the bin's doubly-linked
-        ** list. */
-        ptr_rec_ptr->prev = bin_front_ptr;
-        ptr_rec_ptr->next = bin_front_ptr->next;
-        bin_front_ptr->next = ptr_rec_ptr;
+    /* Make the block the new second block in the bin's doubly-linked
+    ** list. */
+    ptr_rec_ptr->prev = bin_front_ptr;
+    ptr_rec_ptr->next = bin_front_ptr->next;
+    bin_front_ptr->next = ptr_rec_ptr;
 
-        if (ptr_rec_ptr->next)
-            ptr_rec_ptr->next->prev = ptr_rec_ptr;
-    }
-    else
-        /* Block is first block in new bin. */
-        ptr_rec_ptr->next = 0;
+    if (ptr_rec_ptr->next)
+      ptr_rec_ptr->next->prev = ptr_rec_ptr;
+  } else
+    /* Block is first block in new bin. */
+    ptr_rec_ptr->next = 0;
 }
 
 /* Allocate a block from a given bin.  Returns a pointer to the payload
@@ -74,268 +69,245 @@
 ** to calling this function.
 */
 void *U(alloc_from_bin)(
-    /* Pointer to heap descriptor. */
-    U(descriptor) *desc,
-    /* Pointer to pointer record of first block in bin. */
-    ptr_record *bin_front_ptr,
-    /* Number of BAUs needed in the allocated block.  If the block taken
-    ** from the bin is significantly larger than the number of BAUs needed,
-    ** the "extra" BAUs are split off to form a new free block. */
-    U(size_bau) n_baus)
-{
-    head_record *head_ptr;
-    U(size_bau) rem_baus;
+  /* Pointer to heap descriptor. */
+  U(descriptor) *desc,
+  /* Pointer to pointer record of first block in bin. */
+  ptr_record *bin_front_ptr,
+  /* Number of BAUs needed in the allocated block.  If the block taken
+  ** from the bin is significantly larger than the number of BAUs needed,
+  ** the "extra" BAUs are split off to form a new free block. */
+  U(size_bau) n_baus) {
+  head_record *head_ptr;
+  U(size_bau) rem_baus;
 
-    if (bin_front_ptr->next)
-    {
-        /* There are multiple blocks in this bin.  Use the 2nd block in
-        ** the bin to avoid needless change to the AVL tree.
-        */
+  if (bin_front_ptr->next) {
+    /* There are multiple blocks in this bin.  Use the 2nd block in
+    ** the bin to avoid needless change to the AVL tree.
+    */
 
-        ptr_record *ptr_rec_ptr = bin_front_ptr->next;
-        head_ptr = PTR_REC_TO_HEAD(ptr_rec_ptr);
+    ptr_record *ptr_rec_ptr = bin_front_ptr->next;
+    head_ptr = PTR_REC_TO_HEAD(ptr_rec_ptr);
 
 #ifdef AUDIT_FAIL
-        AUDIT_BLOCK(head_ptr)
+    AUDIT_BLOCK(head_ptr)
 #endif
 
-        U(dll_remove)(ptr_rec_ptr);
-    }
-    else
-    {
-        /* There is only one block in the bin, so it has to be removed
-        ** from the AVL tree.
-        */
+    U(dll_remove)(ptr_rec_ptr);
+  } else {
+    /* There is only one block in the bin, so it has to be removed
+    ** from the AVL tree.
+    */
 
-        head_ptr = PTR_REC_TO_HEAD(bin_front_ptr);
+    head_ptr = PTR_REC_TO_HEAD(bin_front_ptr);
 
-        U(avl_remove)(
-            (U(avl_avl) *) &(desc->avl_tree_root), BLOCK_BAUS(head_ptr));
-    }
+    U(avl_remove)(
+      (U(avl_avl) *) & (desc->avl_tree_root), BLOCK_BAUS(head_ptr));
+  }
 
-    MARK_BLOCK_ALLOCATED(head_ptr)
+  MARK_BLOCK_ALLOCATED(head_ptr)
 
-    rem_baus = BLOCK_BAUS(head_ptr) - n_baus;
+  rem_baus = BLOCK_BAUS(head_ptr) - n_baus;
 
-    if (rem_baus >= MIN_BLOCK_BAUS)
-    {
-        /* Since there are enough "extra" BAUs, split them off to form
-        ** a new free block.
-        */
+  if (rem_baus >= MIN_BLOCK_BAUS) {
+    /* Since there are enough "extra" BAUs, split them off to form
+    ** a new free block.
+    */
 
-        head_record *rem_head_ptr =
-            (head_record *) BAUS_FORWARD(head_ptr, n_baus);
+    head_record *rem_head_ptr =
+      (head_record *) BAUS_FORWARD(head_ptr, n_baus);
 
-        /* Change the next block's header to reflect the fact that the
-        ** block preceeding it is now smaller.
-        */
-        SET_PREV_BLOCK_BAUS(
-            BAUS_FORWARD(head_ptr, head_ptr->block_size), rem_baus)
+    /* Change the next block's header to reflect the fact that the
+    ** block preceeding it is now smaller.
+    */
+    SET_PREV_BLOCK_BAUS(
+      BAUS_FORWARD(head_ptr, head_ptr->block_size), rem_baus)
 
-        head_ptr->block_size = n_baus;
+    head_ptr->block_size = n_baus;
 
-        rem_head_ptr->previous_block_size = n_baus;
-        rem_head_ptr->block_size = rem_baus;
+    rem_head_ptr->previous_block_size = n_baus;
+    rem_head_ptr->block_size = rem_baus;
 
-        desc->last_freed = rem_head_ptr;
-    }
+    desc->last_freed = rem_head_ptr;
+  }
 
-    return(HEAD_TO_PTR_REC(head_ptr));
+  return(HEAD_TO_PTR_REC(head_ptr));
 }
 
 /* Take a block out of the free collection.
 */
 void U(out_of_free_collection)(
-    /* Descriptor of heap that block is in. */
-    U(descriptor) *desc,
-    /* Pointer to head of block to take out of free collection. */
-    head_record *head_ptr)
-{
-    ptr_record *ptr_rec_ptr = HEAD_TO_PTR_REC(head_ptr);
+  /* Descriptor of heap that block is in. */
+  U(descriptor) *desc,
+  /* Pointer to head of block to take out of free collection. */
+  head_record *head_ptr) {
+  ptr_record *ptr_rec_ptr = HEAD_TO_PTR_REC(head_ptr);
 
-    if (ptr_rec_ptr->self == ptr_rec_ptr)
-        /* Block is not the front block in its bin, so all we have to
-        ** do is take it out of the bin's doubly-linked list. */
-        U(dll_remove)(ptr_rec_ptr);
+  if (ptr_rec_ptr->self == ptr_rec_ptr)
+    /* Block is not the front block in its bin, so all we have to
+    ** do is take it out of the bin's doubly-linked list. */
+    U(dll_remove)(ptr_rec_ptr);
+  else {
+    ptr_record *next = ptr_rec_ptr->next;
+
+    if (next)
+      /* Block is the front block in its bin, and there is at least
+      ** one other block in the bin.  Substitute the next block for
+      ** the front block. */
+      U(avl_subst)((U(avl_avl) *) & (desc->avl_tree_root), next);
     else
-    {
-        ptr_record *next = ptr_rec_ptr->next;
-
-        if (next)
-            /* Block is the front block in its bin, and there is at least
-            ** one other block in the bin.  Substitute the next block for
-            ** the front block. */
-            U(avl_subst)((U(avl_avl) *) &(desc->avl_tree_root), next);
-        else
-            /* Block is the front block in its bin, but there is no other
-            ** block in the bin.  Eliminate the bin. */
-            U(avl_remove)(
-                (U(avl_avl) *) &(desc->avl_tree_root), BLOCK_BAUS(head_ptr));
-    }
+      /* Block is the front block in its bin, but there is no other
+      ** block in the bin.  Eliminate the bin. */
+      U(avl_remove)(
+        (U(avl_avl) *) & (desc->avl_tree_root), BLOCK_BAUS(head_ptr));
+  }
 }
 
-void U(free)(U(descriptor) *desc, void *payload_ptr)
-{
-    /* Flags if coalesce with adjacent block. */
-    int coalesce;
+void U(free)(U(descriptor) *desc, void *payload_ptr) {
+  /* Flags if coalesce with adjacent block. */
+  int coalesce;
 
-    head_record *fwd_head_ptr;
-    head_record *free_head_ptr = PTR_REC_TO_HEAD(payload_ptr);
+  head_record *fwd_head_ptr;
+  head_record *free_head_ptr = PTR_REC_TO_HEAD(payload_ptr);
 
-    desc->num_baus_can_shrink = 0;
+  desc->num_baus_can_shrink = 0;
 
 #ifdef HMM_AUDIT_FAIL
 
-    AUDIT_BLOCK(free_head_ptr)
+  AUDIT_BLOCK(free_head_ptr)
 
-    /* Make sure not freeing an already free block. */
-    if (!IS_BLOCK_ALLOCATED(free_head_ptr))
-        HMM_AUDIT_FAIL
+  /* Make sure not freeing an already free block. */
+  if (!IS_BLOCK_ALLOCATED(free_head_ptr))
+    HMM_AUDIT_FAIL
 
-        if (desc->avl_tree_root)
-            /* Audit root block in AVL tree. */
-            AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+    if (desc->avl_tree_root)
+      /* Audit root block in AVL tree. */
+      AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 
 #endif
 
-            fwd_head_ptr =
-                (head_record *) BAUS_FORWARD(free_head_ptr, free_head_ptr->block_size);
+      fwd_head_ptr =
+        (head_record *) BAUS_FORWARD(free_head_ptr, free_head_ptr->block_size);
 
-    if (free_head_ptr->previous_block_size)
-    {
-        /* Coalesce with backward block if possible. */
+  if (free_head_ptr->previous_block_size) {
+    /* Coalesce with backward block if possible. */
 
-        head_record *bkwd_head_ptr =
-            (head_record *) BAUS_BACKWARD(
-                free_head_ptr, free_head_ptr->previous_block_size);
+    head_record *bkwd_head_ptr =
+      (head_record *) BAUS_BACKWARD(
+        free_head_ptr, free_head_ptr->previous_block_size);
 
 #ifdef HMM_AUDIT_FAIL
-        AUDIT_BLOCK(bkwd_head_ptr)
+    AUDIT_BLOCK(bkwd_head_ptr)
 #endif
 
-        if (bkwd_head_ptr == (head_record *)(desc->last_freed))
-        {
-            desc->last_freed = 0;
-            coalesce = 1;
-        }
-        else if (IS_BLOCK_ALLOCATED(bkwd_head_ptr))
-            coalesce = 0;
-        else
-        {
-            U(out_of_free_collection)(desc, bkwd_head_ptr);
-            coalesce = 1;
-        }
-
-        if (coalesce)
-        {
-            bkwd_head_ptr->block_size += free_head_ptr->block_size;
-            SET_PREV_BLOCK_BAUS(fwd_head_ptr, BLOCK_BAUS(bkwd_head_ptr))
-            free_head_ptr = bkwd_head_ptr;
-        }
+    if (bkwd_head_ptr == (head_record *)(desc->last_freed)) {
+      desc->last_freed = 0;
+      coalesce = 1;
+    } else if (IS_BLOCK_ALLOCATED(bkwd_head_ptr))
+      coalesce = 0;
+    else {
+      U(out_of_free_collection)(desc, bkwd_head_ptr);
+      coalesce = 1;
     }
 
-    if (fwd_head_ptr->block_size == 0)
-    {
-        /* Block to be freed is last block before dummy end-of-chunk block. */
+    if (coalesce) {
+      bkwd_head_ptr->block_size += free_head_ptr->block_size;
+      SET_PREV_BLOCK_BAUS(fwd_head_ptr, BLOCK_BAUS(bkwd_head_ptr))
+      free_head_ptr = bkwd_head_ptr;
+    }
+  }
+
+  if (fwd_head_ptr->block_size == 0) {
+    /* Block to be freed is last block before dummy end-of-chunk block. */
+    desc->end_of_shrinkable_chunk =
+      BAUS_FORWARD(fwd_head_ptr, DUMMY_END_BLOCK_BAUS);
+    desc->num_baus_can_shrink = BLOCK_BAUS(free_head_ptr);
+
+    if (PREV_BLOCK_BAUS(free_head_ptr) == 0)
+      /* Free block is the entire chunk, so shrinking can eliminate
+      ** entire chunk including dummy end block. */
+      desc->num_baus_can_shrink += DUMMY_END_BLOCK_BAUS;
+  } else {
+    /* Coalesce with forward block if possible. */
+
+#ifdef HMM_AUDIT_FAIL
+    AUDIT_BLOCK(fwd_head_ptr)
+#endif
+
+    if (fwd_head_ptr == (head_record *)(desc->last_freed)) {
+      desc->last_freed = 0;
+      coalesce = 1;
+    } else if (IS_BLOCK_ALLOCATED(fwd_head_ptr))
+      coalesce = 0;
+    else {
+      U(out_of_free_collection)(desc, fwd_head_ptr);
+      coalesce = 1;
+    }
+
+    if (coalesce) {
+      free_head_ptr->block_size += fwd_head_ptr->block_size;
+
+      fwd_head_ptr =
+        (head_record *) BAUS_FORWARD(
+          fwd_head_ptr, BLOCK_BAUS(fwd_head_ptr));
+
+      SET_PREV_BLOCK_BAUS(fwd_head_ptr, BLOCK_BAUS(free_head_ptr))
+
+      if (fwd_head_ptr->block_size == 0) {
+        /* Coalesced block to be freed is last block before dummy
+        ** end-of-chunk block. */
         desc->end_of_shrinkable_chunk =
-            BAUS_FORWARD(fwd_head_ptr, DUMMY_END_BLOCK_BAUS);
+          BAUS_FORWARD(fwd_head_ptr, DUMMY_END_BLOCK_BAUS);
         desc->num_baus_can_shrink = BLOCK_BAUS(free_head_ptr);
 
         if (PREV_BLOCK_BAUS(free_head_ptr) == 0)
-            /* Free block is the entire chunk, so shrinking can eliminate
-            ** entire chunk including dummy end block. */
-            desc->num_baus_can_shrink += DUMMY_END_BLOCK_BAUS;
+          /* Free block is the entire chunk, so shrinking can
+          ** eliminate entire chunk including dummy end block. */
+          desc->num_baus_can_shrink += DUMMY_END_BLOCK_BAUS;
+      }
     }
-    else
-    {
-        /* Coalesce with forward block if possible. */
+  }
+
+  if (desc->last_freed) {
+    /* There is a last freed block, but it is not adjacent to the
+    ** block being freed by this call to free, so put the last
+    ** freed block into the free collection.
+    */
 
 #ifdef HMM_AUDIT_FAIL
-        AUDIT_BLOCK(fwd_head_ptr)
+    AUDIT_BLOCK(desc->last_freed)
 #endif
 
-        if (fwd_head_ptr == (head_record *)(desc->last_freed))
-        {
-            desc->last_freed = 0;
-            coalesce = 1;
-        }
-        else if (IS_BLOCK_ALLOCATED(fwd_head_ptr))
-            coalesce = 0;
-        else
-        {
-            U(out_of_free_collection)(desc, fwd_head_ptr);
-            coalesce = 1;
-        }
+    U(into_free_collection)(desc, (head_record *)(desc->last_freed));
+  }
 
-        if (coalesce)
-        {
-            free_head_ptr->block_size += fwd_head_ptr->block_size;
-
-            fwd_head_ptr =
-                (head_record *) BAUS_FORWARD(
-                    fwd_head_ptr, BLOCK_BAUS(fwd_head_ptr));
-
-            SET_PREV_BLOCK_BAUS(fwd_head_ptr, BLOCK_BAUS(free_head_ptr))
-
-            if (fwd_head_ptr->block_size == 0)
-            {
-                /* Coalesced block to be freed is last block before dummy
-                ** end-of-chunk block. */
-                desc->end_of_shrinkable_chunk =
-                    BAUS_FORWARD(fwd_head_ptr, DUMMY_END_BLOCK_BAUS);
-                desc->num_baus_can_shrink = BLOCK_BAUS(free_head_ptr);
-
-                if (PREV_BLOCK_BAUS(free_head_ptr) == 0)
-                    /* Free block is the entire chunk, so shrinking can
-                    ** eliminate entire chunk including dummy end block. */
-                    desc->num_baus_can_shrink += DUMMY_END_BLOCK_BAUS;
-            }
-        }
-    }
-
-    if (desc->last_freed)
-    {
-        /* There is a last freed block, but it is not adjacent to the
-        ** block being freed by this call to free, so put the last
-        ** freed block into the free collection.
-        */
-
-#ifdef HMM_AUDIT_FAIL
-        AUDIT_BLOCK(desc->last_freed)
-#endif
-
-        U(into_free_collection)(desc, (head_record *)(desc->last_freed));
-    }
-
-    desc->last_freed = free_head_ptr;
+  desc->last_freed = free_head_ptr;
 }
 
-void U(new_chunk)(U(descriptor) *desc, void *start, U(size_bau) n_baus)
-{
+void U(new_chunk)(U(descriptor) *desc, void *start, U(size_bau) n_baus) {
 #ifdef HMM_AUDIT_FAIL
 
-    if (desc->avl_tree_root)
-        /* Audit root block in AVL tree. */
-        AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+  if (desc->avl_tree_root)
+    /* Audit root block in AVL tree. */
+    AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 #endif
 
 #undef HEAD_PTR
 #define HEAD_PTR ((head_record *) start)
 
-        /* Make the chunk one big free block followed by a dummy end block.
-        */
+    /* Make the chunk one big free block followed by a dummy end block.
+    */
 
-        n_baus -= DUMMY_END_BLOCK_BAUS;
+    n_baus -= DUMMY_END_BLOCK_BAUS;
 
-    HEAD_PTR->previous_block_size = 0;
-    HEAD_PTR->block_size = n_baus;
+  HEAD_PTR->previous_block_size = 0;
+  HEAD_PTR->block_size = n_baus;
 
-    U(into_free_collection)(desc, HEAD_PTR);
+  U(into_free_collection)(desc, HEAD_PTR);
 
-    /* Set up the dummy end block. */
-    start = BAUS_FORWARD(start, n_baus);
-    HEAD_PTR->previous_block_size = n_baus;
-    HEAD_PTR->block_size = 0;
+  /* Set up the dummy end block. */
+  start = BAUS_FORWARD(start, n_baus);
+  HEAD_PTR->previous_block_size = n_baus;
+  HEAD_PTR->block_size = 0;
 
 #undef HEAD_PTR
 }
@@ -345,12 +317,11 @@
 /* Function that does audit fail actions defined my preprocessor symbol,
 ** and returns a dummy integer value.
 */
-int U(audit_block_fail_dummy_return)(void)
-{
-    HMM_AUDIT_FAIL
+int U(audit_block_fail_dummy_return)(void) {
+  HMM_AUDIT_FAIL
 
-    /* Dummy return. */
-    return(0);
+  /* Dummy return. */
+  return(0);
 }
 
 #endif
@@ -372,9 +343,9 @@
 */
 
 #define AVL_GET_LESS(H, ACCESS) \
-    (((ACCESS) ? AUDIT_BLOCK_AS_EXPR(PTR_REC_TO_HEAD(H)) : 0), (H)->self)
+  (((ACCESS) ? AUDIT_BLOCK_AS_EXPR(PTR_REC_TO_HEAD(H)) : 0), (H)->self)
 #define AVL_GET_GREATER(H, ACCESS) \
-    (((ACCESS) ? AUDIT_BLOCK_AS_EXPR(PTR_REC_TO_HEAD(H)) : 0), (H)->prev)
+  (((ACCESS) ? AUDIT_BLOCK_AS_EXPR(PTR_REC_TO_HEAD(H)) : 0), (H)->prev)
 
 #else
 
@@ -396,39 +367,39 @@
 */
 
 #define AVL_GET_BALANCE_FACTOR(H) \
-    ((((head_record *) (PTR_REC_TO_HEAD(H)))->block_size & \
-      HIGH_BIT_BAU_SIZE) ? \
-     (((head_record *) (PTR_REC_TO_HEAD(H)))->previous_block_size & \
-      HIGH_BIT_BAU_SIZE ? 0 : -1) : 1)
+  ((((head_record *) (PTR_REC_TO_HEAD(H)))->block_size & \
+    HIGH_BIT_BAU_SIZE) ? \
+   (((head_record *) (PTR_REC_TO_HEAD(H)))->previous_block_size & \
+    HIGH_BIT_BAU_SIZE ? 0 : -1) : 1)
 
 #define AVL_SET_BALANCE_FACTOR(H, BF) \
-    {                         \
-        register head_record *p =               \
-                                                (head_record *) PTR_REC_TO_HEAD(H);       \
-        register int bal_f = (BF);              \
-        \
-        if (bal_f <= 0)                 \
-            p->block_size |= HIGH_BIT_BAU_SIZE;       \
-        else                        \
-            p->block_size &= ~HIGH_BIT_BAU_SIZE;      \
-        if (bal_f >= 0)                 \
-            p->previous_block_size |= HIGH_BIT_BAU_SIZE;  \
-        else                        \
-            p->previous_block_size &= ~HIGH_BIT_BAU_SIZE; \
-    }
+  {                         \
+    register head_record *p =               \
+                                            (head_record *) PTR_REC_TO_HEAD(H);       \
+    register int bal_f = (BF);              \
+    \
+    if (bal_f <= 0)                 \
+      p->block_size |= HIGH_BIT_BAU_SIZE;       \
+    else                        \
+      p->block_size &= ~HIGH_BIT_BAU_SIZE;      \
+    if (bal_f >= 0)                 \
+      p->previous_block_size |= HIGH_BIT_BAU_SIZE;  \
+    else                        \
+      p->previous_block_size &= ~HIGH_BIT_BAU_SIZE; \
+  }
 
 #define COMPARE_KEY_KEY(K1, K2) ((K1) == (K2) ? 0 : ((K1) > (K2) ? 1 : -1))
 
 #define AVL_COMPARE_KEY_NODE(K, H) \
-    COMPARE_KEY_KEY(K, BLOCK_BAUS(PTR_REC_TO_HEAD(H)))
+  COMPARE_KEY_KEY(K, BLOCK_BAUS(PTR_REC_TO_HEAD(H)))
 
 #define AVL_COMPARE_NODE_NODE(H1, H2) \
-    COMPARE_KEY_KEY(BLOCK_BAUS(PTR_REC_TO_HEAD(H1)), \
-                    BLOCK_BAUS(PTR_REC_TO_HEAD(H2)))
+  COMPARE_KEY_KEY(BLOCK_BAUS(PTR_REC_TO_HEAD(H1)), \
+                  BLOCK_BAUS(PTR_REC_TO_HEAD(H2)))
 
 #define AVL_NULL ((ptr_record *) 0)
 
 #define AVL_IMPL_MASK \
-    ( AVL_IMPL_INSERT | AVL_IMPL_SEARCH | AVL_IMPL_REMOVE | AVL_IMPL_SUBST )
+  ( AVL_IMPL_INSERT | AVL_IMPL_SEARCH | AVL_IMPL_REMOVE | AVL_IMPL_SUBST )
 
 #include "cavl_impl.h"
diff --git a/libvpx/vpx_mem/memory_manager/hmm_dflt_abort.c b/libvpx/vpx_mem/memory_manager/hmm_dflt_abort.c
index d92435c..51c3cc2 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_dflt_abort.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_dflt_abort.c
@@ -29,26 +29,25 @@
 
 /* Print abort message, file and line.  Terminate execution.
 */
-void hmm_dflt_abort(const char *file, const char *line)
-{
-    /* Avoid use of printf(), which is more likely to use heap. */
+void hmm_dflt_abort(const char *file, const char *line) {
+  /* Avoid use of printf(), which is more likely to use heap. */
 
-    if (entered)
+  if (entered)
 
-        /* The standard I/O functions called a heap function and caused
-        ** an indirect recursive call to this function.  So we'll have
-        ** to just exit without printing a message.  */
-        while (1);
-
-    entered = 1;
-
-    fputs("\n_abort - Heap corruption\n" "File: ", stderr);
-    fputs(file, stderr);
-    fputs("  Line: ", stderr);
-    fputs(line, stderr);
-    fputs("\n\n", stderr);
-    fputs("hmm_dflt_abort: while(1)!!!\n", stderr);
-    fflush(stderr);
-
+    /* The standard I/O functions called a heap function and caused
+    ** an indirect recursive call to this function.  So we'll have
+    ** to just exit without printing a message.  */
     while (1);
+
+  entered = 1;
+
+  fputs("\n_abort - Heap corruption\n" "File: ", stderr);
+  fputs(file, stderr);
+  fputs("  Line: ", stderr);
+  fputs(line, stderr);
+  fputs("\n\n", stderr);
+  fputs("hmm_dflt_abort: while(1)!!!\n", stderr);
+  fflush(stderr);
+
+  while (1);
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_grow.c b/libvpx/vpx_mem/memory_manager/hmm_grow.c
index 9a4b6e4..0e86373 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_grow.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_grow.c
@@ -15,36 +15,35 @@
 
 #include "hmm_intrnl.h"
 
-void U(grow_chunk)(U(descriptor) *desc, void *end, U(size_bau) n_baus)
-{
+void U(grow_chunk)(U(descriptor) *desc, void *end, U(size_bau) n_baus) {
 #undef HEAD_PTR
 #define HEAD_PTR ((head_record *) end)
 
-    end = BAUS_BACKWARD(end, DUMMY_END_BLOCK_BAUS);
+  end = BAUS_BACKWARD(end, DUMMY_END_BLOCK_BAUS);
 
 #ifdef HMM_AUDIT_FAIL
 
-    if (HEAD_PTR->block_size != 0)
-        /* Chunk does not have valid dummy end block. */
-        HMM_AUDIT_FAIL
+  if (HEAD_PTR->block_size != 0)
+    /* Chunk does not have valid dummy end block. */
+    HMM_AUDIT_FAIL
 
 #endif
 
-        /* Create a new block that absorbs the old dummy end block. */
-        HEAD_PTR->block_size = n_baus;
+    /* Create a new block that absorbs the old dummy end block. */
+    HEAD_PTR->block_size = n_baus;
 
-    /* Set up the new dummy end block. */
-    {
-        head_record *dummy = (head_record *) BAUS_FORWARD(end, n_baus);
-        dummy->previous_block_size = n_baus;
-        dummy->block_size = 0;
-    }
+  /* Set up the new dummy end block. */
+  {
+    head_record *dummy = (head_record *) BAUS_FORWARD(end, n_baus);
+    dummy->previous_block_size = n_baus;
+    dummy->block_size = 0;
+  }
 
-    /* Simply free the new block, allowing it to coalesce with any
-    ** free block at that was the last block in the chunk prior to
-    ** growth.
-    */
-    U(free)(desc, HEAD_TO_PTR_REC(end));
+  /* Simply free the new block, allowing it to coalesce with any
+  ** free block at that was the last block in the chunk prior to
+  ** growth.
+  */
+  U(free)(desc, HEAD_TO_PTR_REC(end));
 
 #undef HEAD_PTR
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_largest.c b/libvpx/vpx_mem/memory_manager/hmm_largest.c
index c3c6f2c..192758d 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_largest.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_largest.c
@@ -15,46 +15,43 @@
 
 #include "hmm_intrnl.h"
 
-U(size_aau) U(largest_available)(U(descriptor) *desc)
-{
-    U(size_bau) largest;
+U(size_aau) U(largest_available)(U(descriptor) *desc) {
+  U(size_bau) largest;
 
-    if (!(desc->avl_tree_root))
-        largest = 0;
-    else
-    {
+  if (!(desc->avl_tree_root))
+    largest = 0;
+  else {
 #ifdef HMM_AUDIT_FAIL
-        /* Audit root block in AVL tree. */
-        AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+    /* Audit root block in AVL tree. */
+    AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 #endif
 
-        largest =
-            BLOCK_BAUS(
-                PTR_REC_TO_HEAD(
-                    U(avl_search)(
-                        (U(avl_avl) *) & (desc->avl_tree_root),
-                        (U(size_bau)) ~(U(size_bau)) 0, AVL_LESS)));
-    }
+    largest =
+      BLOCK_BAUS(
+        PTR_REC_TO_HEAD(
+          U(avl_search)(
+            (U(avl_avl) *) & (desc->avl_tree_root),
+            (U(size_bau)) ~(U(size_bau)) 0, AVL_LESS)));
+  }
 
-    if (desc->last_freed)
-    {
-        /* Size of last freed block. */
-        register U(size_bau) lf_size;
+  if (desc->last_freed) {
+    /* Size of last freed block. */
+    register U(size_bau) lf_size;
 
 #ifdef HMM_AUDIT_FAIL
-        AUDIT_BLOCK(desc->last_freed)
+    AUDIT_BLOCK(desc->last_freed)
 #endif
 
-        lf_size = BLOCK_BAUS(desc->last_freed);
+    lf_size = BLOCK_BAUS(desc->last_freed);
 
-        if (lf_size > largest)
-            largest = lf_size;
-    }
+    if (lf_size > largest)
+      largest = lf_size;
+  }
 
-    /* Convert largest size to AAUs and subract head size leaving payload
-    ** size.
-    */
-    return(largest ?
-           ((largest * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT)) - HEAD_AAUS) :
-           0);
+  /* Convert largest size to AAUs and subract head size leaving payload
+  ** size.
+  */
+  return(largest ?
+         ((largest * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT)) - HEAD_AAUS) :
+         0);
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_resize.c b/libvpx/vpx_mem/memory_manager/hmm_resize.c
index f90da96..baa5a8f 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_resize.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_resize.c
@@ -15,105 +15,100 @@
 
 #include "hmm_intrnl.h"
 
-int U(resize)(U(descriptor) *desc, void *mem, U(size_aau) n)
-{
-    U(size_aau) i;
-    head_record *next_head_ptr;
-    head_record *head_ptr = PTR_REC_TO_HEAD(mem);
+int U(resize)(U(descriptor) *desc, void *mem, U(size_aau) n) {
+  U(size_aau) i;
+  head_record *next_head_ptr;
+  head_record *head_ptr = PTR_REC_TO_HEAD(mem);
 
-    /* Flag. */
-    int next_block_free;
+  /* Flag. */
+  int next_block_free;
 
-    /* Convert n from desired block size in AAUs to BAUs. */
-    n += HEAD_AAUS;
-    n = DIV_ROUND_UP(n, HMM_BLOCK_ALIGN_UNIT);
+  /* Convert n from desired block size in AAUs to BAUs. */
+  n += HEAD_AAUS;
+  n = DIV_ROUND_UP(n, HMM_BLOCK_ALIGN_UNIT);
 
-    if (n < MIN_BLOCK_BAUS)
-        n = MIN_BLOCK_BAUS;
+  if (n < MIN_BLOCK_BAUS)
+    n = MIN_BLOCK_BAUS;
 
 #ifdef HMM_AUDIT_FAIL
 
-    AUDIT_BLOCK(head_ptr)
+  AUDIT_BLOCK(head_ptr)
 
-    if (!IS_BLOCK_ALLOCATED(head_ptr))
-        HMM_AUDIT_FAIL
+  if (!IS_BLOCK_ALLOCATED(head_ptr))
+    HMM_AUDIT_FAIL
 
-        if (desc->avl_tree_root)
-            AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+    if (desc->avl_tree_root)
+      AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 
 #endif
 
-            i = head_ptr->block_size;
+      i = head_ptr->block_size;
+
+  next_head_ptr =
+    (head_record *) BAUS_FORWARD(head_ptr, head_ptr->block_size);
+
+  next_block_free =
+    (next_head_ptr == desc->last_freed) ||
+    !IS_BLOCK_ALLOCATED(next_head_ptr);
+
+  if (next_block_free)
+    /* Block can expand into next free block. */
+    i += BLOCK_BAUS(next_head_ptr);
+
+  if (n > i)
+    /* Not enough room for block to expand. */
+    return(-1);
+
+  if (next_block_free) {
+#ifdef HMM_AUDIT_FAIL
+    AUDIT_BLOCK(next_head_ptr)
+#endif
+
+    if (next_head_ptr == desc->last_freed)
+      desc->last_freed = 0;
+    else
+      U(out_of_free_collection)(desc, next_head_ptr);
 
     next_head_ptr =
-        (head_record *) BAUS_FORWARD(head_ptr, head_ptr->block_size);
+      (head_record *) BAUS_FORWARD(head_ptr, (U(size_bau)) i);
+  }
 
-    next_block_free =
-        (next_head_ptr == desc->last_freed) ||
-        !IS_BLOCK_ALLOCATED(next_head_ptr);
+  /* Set i to number of "extra" BAUs. */
+  i -= n;
 
-    if (next_block_free)
-        /* Block can expand into next free block. */
-        i += BLOCK_BAUS(next_head_ptr);
+  if (i < MIN_BLOCK_BAUS)
+    /* Not enough extra BAUs to be a block on their own, so just keep them
+    ** in the block being resized.
+    */
+  {
+    n += i;
+    i = n;
+  } else {
+    /* There are enough "leftover" BAUs in the next block to
+    ** form a remainder block. */
 
-    if (n > i)
-        /* Not enough room for block to expand. */
-        return(-1);
+    head_record *rem_head_ptr;
 
-    if (next_block_free)
-    {
+    rem_head_ptr = (head_record *) BAUS_FORWARD(head_ptr, n);
+
+    rem_head_ptr->previous_block_size = (U(size_bau)) n;
+    rem_head_ptr->block_size = (U(size_bau)) i;
+
+    if (desc->last_freed) {
 #ifdef HMM_AUDIT_FAIL
-        AUDIT_BLOCK(next_head_ptr)
+      AUDIT_BLOCK(desc->last_freed)
 #endif
 
-        if (next_head_ptr == desc->last_freed)
-            desc->last_freed = 0;
-        else
-            U(out_of_free_collection)(desc, next_head_ptr);
+      U(into_free_collection)(desc, (head_record *)(desc->last_freed));
 
-        next_head_ptr =
-            (head_record *) BAUS_FORWARD(head_ptr, (U(size_bau)) i);
+      desc->last_freed = 0;
     }
 
-    /* Set i to number of "extra" BAUs. */
-    i -= n;
+    desc->last_freed = rem_head_ptr;
+  }
 
-    if (i < MIN_BLOCK_BAUS)
-        /* Not enough extra BAUs to be a block on their own, so just keep them
-        ** in the block being resized.
-        */
-    {
-        n += i;
-        i = n;
-    }
-    else
-    {
-        /* There are enough "leftover" BAUs in the next block to
-        ** form a remainder block. */
+  head_ptr->block_size = (U(size_bau)) n;
+  next_head_ptr->previous_block_size = (U(size_bau)) i;
 
-        head_record *rem_head_ptr;
-
-        rem_head_ptr = (head_record *) BAUS_FORWARD(head_ptr, n);
-
-        rem_head_ptr->previous_block_size = (U(size_bau)) n;
-        rem_head_ptr->block_size = (U(size_bau)) i;
-
-        if (desc->last_freed)
-        {
-#ifdef HMM_AUDIT_FAIL
-            AUDIT_BLOCK(desc->last_freed)
-#endif
-
-            U(into_free_collection)(desc, (head_record *)(desc->last_freed));
-
-            desc->last_freed = 0;
-        }
-
-        desc->last_freed = rem_head_ptr;
-    }
-
-    head_ptr->block_size = (U(size_bau)) n;
-    next_head_ptr->previous_block_size = (U(size_bau)) i;
-
-    return(0);
+  return(0);
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_shrink.c b/libvpx/vpx_mem/memory_manager/hmm_shrink.c
index 78fe268..f80aeea 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_shrink.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_shrink.c
@@ -15,97 +15,89 @@
 
 #include "hmm_intrnl.h"
 
-void U(shrink_chunk)(U(descriptor) *desc, U(size_bau) n_baus_to_shrink)
-{
-    head_record *dummy_end_block = (head_record *)
-                                   BAUS_BACKWARD(desc->end_of_shrinkable_chunk, DUMMY_END_BLOCK_BAUS);
+void U(shrink_chunk)(U(descriptor) *desc, U(size_bau) n_baus_to_shrink) {
+  head_record *dummy_end_block = (head_record *)
+                                 BAUS_BACKWARD(desc->end_of_shrinkable_chunk, DUMMY_END_BLOCK_BAUS);
 
 #ifdef HMM_AUDIT_FAIL
 
-    if (dummy_end_block->block_size != 0)
-        /* Chunk does not have valid dummy end block. */
-        HMM_AUDIT_FAIL
+  if (dummy_end_block->block_size != 0)
+    /* Chunk does not have valid dummy end block. */
+    HMM_AUDIT_FAIL
 
 #endif
 
-        if (n_baus_to_shrink)
-        {
-            head_record *last_block = (head_record *)
-                                      BAUS_BACKWARD(
-                                          dummy_end_block, dummy_end_block->previous_block_size);
+    if (n_baus_to_shrink) {
+      head_record *last_block = (head_record *)
+                                BAUS_BACKWARD(
+                                  dummy_end_block, dummy_end_block->previous_block_size);
 
 #ifdef HMM_AUDIT_FAIL
-            AUDIT_BLOCK(last_block)
+      AUDIT_BLOCK(last_block)
 #endif
 
-            if (last_block == desc->last_freed)
-            {
-                U(size_bau) bs = BLOCK_BAUS(last_block);
+      if (last_block == desc->last_freed) {
+        U(size_bau) bs = BLOCK_BAUS(last_block);
 
-                /* Chunk will not be shrunk out of existence if
-                ** 1.  There is at least one allocated block in the chunk
-                **     and the amount to shrink is exactly the size of the
-                **     last block, OR
-                ** 2.  After the last block is shrunk, there will be enough
-                **     BAUs left in it to form a minimal size block. */
-                int chunk_will_survive =
-                    (PREV_BLOCK_BAUS(last_block) && (n_baus_to_shrink == bs)) ||
-                    (n_baus_to_shrink <= (U(size_bau))(bs - MIN_BLOCK_BAUS));
+        /* Chunk will not be shrunk out of existence if
+        ** 1.  There is at least one allocated block in the chunk
+        **     and the amount to shrink is exactly the size of the
+        **     last block, OR
+        ** 2.  After the last block is shrunk, there will be enough
+        **     BAUs left in it to form a minimal size block. */
+        int chunk_will_survive =
+          (PREV_BLOCK_BAUS(last_block) && (n_baus_to_shrink == bs)) ||
+          (n_baus_to_shrink <= (U(size_bau))(bs - MIN_BLOCK_BAUS));
 
-                if (chunk_will_survive ||
-                    (!PREV_BLOCK_BAUS(last_block) &&
-                     (n_baus_to_shrink ==
-                      (U(size_bau))(bs + DUMMY_END_BLOCK_BAUS))))
-                {
-                    desc->last_freed = 0;
+        if (chunk_will_survive ||
+            (!PREV_BLOCK_BAUS(last_block) &&
+             (n_baus_to_shrink ==
+              (U(size_bau))(bs + DUMMY_END_BLOCK_BAUS)))) {
+          desc->last_freed = 0;
 
-                    if (chunk_will_survive)
-                    {
-                        bs -= n_baus_to_shrink;
+          if (chunk_will_survive) {
+            bs -= n_baus_to_shrink;
 
-                        if (bs)
-                        {
-                            /* The last (non-dummy) block was not completely
-                            ** eliminated by the shrink. */
+            if (bs) {
+              /* The last (non-dummy) block was not completely
+              ** eliminated by the shrink. */
 
-                            last_block->block_size = bs;
+              last_block->block_size = bs;
 
-                            /* Create new dummy end record.
-                            */
-                            dummy_end_block =
-                                (head_record *) BAUS_FORWARD(last_block, bs);
-                            dummy_end_block->previous_block_size = bs;
-                            dummy_end_block->block_size = 0;
+              /* Create new dummy end record.
+              */
+              dummy_end_block =
+                (head_record *) BAUS_FORWARD(last_block, bs);
+              dummy_end_block->previous_block_size = bs;
+              dummy_end_block->block_size = 0;
 
 #ifdef HMM_AUDIT_FAIL
 
-                            if (desc->avl_tree_root)
-                                AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
+              if (desc->avl_tree_root)
+                AUDIT_BLOCK(PTR_REC_TO_HEAD(desc->avl_tree_root))
 #endif
 
-                                U(into_free_collection)(desc, last_block);
-                        }
-                        else
-                        {
-                            /* The last (non-dummy) block was completely
-                            ** eliminated by the shrink.  Make its head
-                            ** the new dummy end block.
-                            */
-                            last_block->block_size = 0;
-                            last_block->previous_block_size &= ~HIGH_BIT_BAU_SIZE;
-                        }
-                    }
-                }
-
-#ifdef HMM_AUDIT_FAIL
-                else
-                    HMM_AUDIT_FAIL
-#endif
-                }
-
-#ifdef HMM_AUDIT_FAIL
-            else
-                HMM_AUDIT_FAIL
-#endif
+                U(into_free_collection)(desc, last_block);
+            } else {
+              /* The last (non-dummy) block was completely
+              ** eliminated by the shrink.  Make its head
+              ** the new dummy end block.
+              */
+              last_block->block_size = 0;
+              last_block->previous_block_size &= ~HIGH_BIT_BAU_SIZE;
             }
+          }
+        }
+
+#ifdef HMM_AUDIT_FAIL
+        else
+          HMM_AUDIT_FAIL
+#endif
+        }
+
+#ifdef HMM_AUDIT_FAIL
+      else
+        HMM_AUDIT_FAIL
+#endif
+      }
 }
diff --git a/libvpx/vpx_mem/memory_manager/hmm_true.c b/libvpx/vpx_mem/memory_manager/hmm_true.c
index 3f7be8f..4428c3e 100644
--- a/libvpx/vpx_mem/memory_manager/hmm_true.c
+++ b/libvpx/vpx_mem/memory_manager/hmm_true.c
@@ -15,18 +15,17 @@
 
 #include "hmm_intrnl.h"
 
-U(size_aau) U(true_size)(void *payload_ptr)
-{
-    register  head_record *head_ptr = PTR_REC_TO_HEAD(payload_ptr);
+U(size_aau) U(true_size)(void *payload_ptr) {
+  register  head_record *head_ptr = PTR_REC_TO_HEAD(payload_ptr);
 
 #ifdef HMM_AUDIT_FAIL
-    AUDIT_BLOCK(head_ptr)
+  AUDIT_BLOCK(head_ptr)
 #endif
 
-    /* Convert block size from BAUs to AAUs.  Subtract head size, leaving
-    ** payload size.
-    */
-    return(
-              (BLOCK_BAUS(head_ptr) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT)) -
-              HEAD_AAUS);
+  /* Convert block size from BAUs to AAUs.  Subtract head size, leaving
+  ** payload size.
+  */
+  return(
+          (BLOCK_BAUS(head_ptr) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT)) -
+          HEAD_AAUS);
 }
diff --git a/libvpx/vpx_mem/memory_manager/include/cavl_if.h b/libvpx/vpx_mem/memory_manager/include/cavl_if.h
index 1b2c9b7..ec6e525 100644
--- a/libvpx/vpx_mem/memory_manager/include/cavl_if.h
+++ b/libvpx/vpx_mem/memory_manager/include/cavl_if.h
@@ -32,13 +32,12 @@
 #ifndef AVL_SEARCH_TYPE_DEFINED_
 #define AVL_SEARCH_TYPE_DEFINED_
 
-typedef enum
-{
-    AVL_EQUAL = 1,
-    AVL_LESS = 2,
-    AVL_GREATER = 4,
-    AVL_LESS_EQUAL = AVL_EQUAL | AVL_LESS,
-    AVL_GREATER_EQUAL = AVL_EQUAL | AVL_GREATER
+typedef enum {
+  AVL_EQUAL = 1,
+  AVL_LESS = 2,
+  AVL_GREATER = 4,
+  AVL_LESS_EQUAL = AVL_EQUAL | AVL_LESS,
+  AVL_GREATER_EQUAL = AVL_EQUAL | AVL_GREATER
 }
 avl_search_type;
 
@@ -75,15 +74,14 @@
 
 #endif
 
-typedef struct
-{
+typedef struct {
 #ifdef AVL_INSIDE_STRUCT
 
-    AVL_INSIDE_STRUCT
+  AVL_INSIDE_STRUCT
 
 #endif
 
-    AVL_HANDLE root;
+  AVL_HANDLE root;
 }
 L_(avl);
 
@@ -108,7 +106,7 @@
 #ifdef AVL_BUILD_ITER_TYPE
 
 L_SC int L_(build)(
-    L_(avl) *tree, AVL_BUILD_ITER_TYPE p, L_SIZE num_nodes);
+  L_(avl) *tree, AVL_BUILD_ITER_TYPE p, L_SIZE num_nodes);
 
 #endif
 
@@ -153,7 +151,7 @@
 /* Maximum depth may be more than number of bits in a long. */
 
 #define L_BIT_ARR_DEFN(NAME) \
-    unsigned long NAME[((AVL_MAX_DEPTH) + L_LONG_BIT - 1) / L_LONG_BIT];
+  unsigned long NAME[((AVL_MAX_DEPTH) + L_LONG_BIT - 1) / L_LONG_BIT];
 
 #else
 
@@ -164,29 +162,28 @@
 #endif
 
 /* Iterator structure. */
-typedef struct
-{
-    /* Tree being iterated over. */
-    L_(avl) *tree_;
+typedef struct {
+  /* Tree being iterated over. */
+  L_(avl) *tree_;
 
-    /* Records a path into the tree.  If bit n is true, indicates
-    ** take greater branch from the nth node in the path, otherwise
-    ** take the less branch.  bit 0 gives branch from root, and
-    ** so on. */
-    L_BIT_ARR_DEFN(branch)
+  /* Records a path into the tree.  If bit n is true, indicates
+  ** take greater branch from the nth node in the path, otherwise
+  ** take the less branch.  bit 0 gives branch from root, and
+  ** so on. */
+  L_BIT_ARR_DEFN(branch)
 
-    /* Zero-based depth of path into tree. */
-    unsigned depth;
+  /* Zero-based depth of path into tree. */
+  unsigned depth;
 
-    /* Handles of nodes in path from root to current node (returned by *). */
-    AVL_HANDLE path_h[(AVL_MAX_DEPTH) - 1];
+  /* Handles of nodes in path from root to current node (returned by *). */
+  AVL_HANDLE path_h[(AVL_MAX_DEPTH) - 1];
 }
 L_(iter);
 
 /* Iterator function prototypes. */
 
 L_SC void L_(start_iter)(
-    L_(avl) *tree, L_(iter) *iter, AVL_KEY k, avl_search_type st);
+  L_(avl) *tree, L_(iter) *iter, AVL_KEY k, avl_search_type st);
 
 L_SC void L_(start_iter_least)(L_(avl) *tree, L_(iter) *iter);
 
diff --git a/libvpx/vpx_mem/memory_manager/include/cavl_impl.h b/libvpx/vpx_mem/memory_manager/include/cavl_impl.h
index 5e165dd..cf7deb7 100644
--- a/libvpx/vpx_mem/memory_manager/include/cavl_impl.h
+++ b/libvpx/vpx_mem/memory_manager/include/cavl_impl.h
@@ -110,16 +110,16 @@
 #define L_BIT_ARR_DEFN(NAME) unsigned long NAME[L_BIT_ARR_LONGS];
 
 #define L_BIT_ARR_VAL(BIT_ARR, BIT_NUM) \
-    ((BIT_ARR)[(BIT_NUM) / L_LONG_BIT] & (1L << ((BIT_NUM) % L_LONG_BIT)))
+  ((BIT_ARR)[(BIT_NUM) / L_LONG_BIT] & (1L << ((BIT_NUM) % L_LONG_BIT)))
 
 #define L_BIT_ARR_0(BIT_ARR, BIT_NUM) \
-    (BIT_ARR)[(BIT_NUM) / L_LONG_BIT] &= ~(1L << ((BIT_NUM) % L_LONG_BIT));
+  (BIT_ARR)[(BIT_NUM) / L_LONG_BIT] &= ~(1L << ((BIT_NUM) % L_LONG_BIT));
 
 #define L_BIT_ARR_1(BIT_ARR, BIT_NUM) \
-    (BIT_ARR)[(BIT_NUM) / L_LONG_BIT] |= 1L << ((BIT_NUM) % L_LONG_BIT);
+  (BIT_ARR)[(BIT_NUM) / L_LONG_BIT] |= 1L << ((BIT_NUM) % L_LONG_BIT);
 
 #define L_BIT_ARR_ALL(BIT_ARR, BIT_VAL) \
-    { int i = L_BIT_ARR_LONGS; do (BIT_ARR)[--i] = 0L - (BIT_VAL); while(i); }
+  { int i = L_BIT_ARR_LONGS; do (BIT_ARR)[--i] = 0L - (BIT_VAL); while(i); }
 
 #else /* The bit array can definitely fit in one long */
 
@@ -138,7 +138,7 @@
 #ifdef AVL_READ_ERRORS_HAPPEN
 
 #define L_CHECK_READ_ERROR(ERROR_RETURN) \
-    { if (AVL_READ_ERROR) return(ERROR_RETURN); }
+  { if (AVL_READ_ERROR) return(ERROR_RETURN); }
 
 #else
 
@@ -179,18 +179,16 @@
 
 #if (L_IMPL_MASK & AVL_IMPL_INIT)
 
-L_SC void L_(init)(L_(avl) *l_tree)
-{
-    l_tree->root = AVL_NULL;
+L_SC void L_(init)(L_(avl) *l_tree) {
+  l_tree->root = AVL_NULL;
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_IS_EMPTY)
 
-L_SC int L_(is_empty)(L_(avl) *l_tree)
-{
-    return(l_tree->root == AVL_NULL);
+L_SC int L_(is_empty)(L_(avl) *l_tree) {
+  return(l_tree->root == AVL_NULL);
 }
 
 #endif
@@ -201,358 +199,305 @@
 
 /* Balances subtree, returns handle of root node of subtree after balancing.
 */
-L_SC AVL_HANDLE L_(balance)(L_BALANCE_PARAM_DECL_PREFIX AVL_HANDLE bal_h)
-{
-    AVL_HANDLE deep_h;
+L_SC AVL_HANDLE L_(balance)(L_BALANCE_PARAM_DECL_PREFIX AVL_HANDLE bal_h) {
+  AVL_HANDLE deep_h;
 
-    /* Either the "greater than" or the "less than" subtree of
-    ** this node has to be 2 levels deeper (or else it wouldn't
-    ** need balancing).
-    */
-    if (AVL_GET_BALANCE_FACTOR(bal_h) > 0)
-    {
-        /* "Greater than" subtree is deeper. */
+  /* Either the "greater than" or the "less than" subtree of
+  ** this node has to be 2 levels deeper (or else it wouldn't
+  ** need balancing).
+  */
+  if (AVL_GET_BALANCE_FACTOR(bal_h) > 0) {
+    /* "Greater than" subtree is deeper. */
 
-        deep_h = AVL_GET_GREATER(bal_h, 1);
+    deep_h = AVL_GET_GREATER(bal_h, 1);
 
-        L_CHECK_READ_ERROR(AVL_NULL)
+    L_CHECK_READ_ERROR(AVL_NULL)
 
-        if (AVL_GET_BALANCE_FACTOR(deep_h) < 0)
-        {
-            int bf;
+    if (AVL_GET_BALANCE_FACTOR(deep_h) < 0) {
+      int bf;
 
-            AVL_HANDLE old_h = bal_h;
-            bal_h = AVL_GET_LESS(deep_h, 1);
-            L_CHECK_READ_ERROR(AVL_NULL)
-            AVL_SET_GREATER(old_h, AVL_GET_LESS(bal_h, 1))
-            AVL_SET_LESS(deep_h, AVL_GET_GREATER(bal_h, 1))
-            AVL_SET_LESS(bal_h, old_h)
-            AVL_SET_GREATER(bal_h, deep_h)
+      AVL_HANDLE old_h = bal_h;
+      bal_h = AVL_GET_LESS(deep_h, 1);
+      L_CHECK_READ_ERROR(AVL_NULL)
+      AVL_SET_GREATER(old_h, AVL_GET_LESS(bal_h, 1))
+      AVL_SET_LESS(deep_h, AVL_GET_GREATER(bal_h, 1))
+      AVL_SET_LESS(bal_h, old_h)
+      AVL_SET_GREATER(bal_h, deep_h)
 
-            bf = AVL_GET_BALANCE_FACTOR(bal_h);
+      bf = AVL_GET_BALANCE_FACTOR(bal_h);
 
-            if (bf != 0)
-            {
-                if (bf > 0)
-                {
-                    AVL_SET_BALANCE_FACTOR(old_h, -1)
-                    AVL_SET_BALANCE_FACTOR(deep_h, 0)
-                }
-                else
-                {
-                    AVL_SET_BALANCE_FACTOR(deep_h, 1)
-                    AVL_SET_BALANCE_FACTOR(old_h, 0)
-                }
-
-                AVL_SET_BALANCE_FACTOR(bal_h, 0)
-            }
-            else
-            {
-                AVL_SET_BALANCE_FACTOR(old_h, 0)
-                AVL_SET_BALANCE_FACTOR(deep_h, 0)
-            }
+      if (bf != 0) {
+        if (bf > 0) {
+          AVL_SET_BALANCE_FACTOR(old_h, -1)
+          AVL_SET_BALANCE_FACTOR(deep_h, 0)
+        } else {
+          AVL_SET_BALANCE_FACTOR(deep_h, 1)
+          AVL_SET_BALANCE_FACTOR(old_h, 0)
         }
-        else
-        {
-            AVL_SET_GREATER(bal_h, AVL_GET_LESS(deep_h, 0))
-            AVL_SET_LESS(deep_h, bal_h)
 
-            if (AVL_GET_BALANCE_FACTOR(deep_h) == 0)
-            {
-                AVL_SET_BALANCE_FACTOR(deep_h, -1)
-                AVL_SET_BALANCE_FACTOR(bal_h, 1)
-            }
-            else
-            {
-                AVL_SET_BALANCE_FACTOR(deep_h, 0)
-                AVL_SET_BALANCE_FACTOR(bal_h, 0)
-            }
+        AVL_SET_BALANCE_FACTOR(bal_h, 0)
+      } else {
+        AVL_SET_BALANCE_FACTOR(old_h, 0)
+        AVL_SET_BALANCE_FACTOR(deep_h, 0)
+      }
+    } else {
+      AVL_SET_GREATER(bal_h, AVL_GET_LESS(deep_h, 0))
+      AVL_SET_LESS(deep_h, bal_h)
 
-            bal_h = deep_h;
-        }
+      if (AVL_GET_BALANCE_FACTOR(deep_h) == 0) {
+        AVL_SET_BALANCE_FACTOR(deep_h, -1)
+        AVL_SET_BALANCE_FACTOR(bal_h, 1)
+      } else {
+        AVL_SET_BALANCE_FACTOR(deep_h, 0)
+        AVL_SET_BALANCE_FACTOR(bal_h, 0)
+      }
+
+      bal_h = deep_h;
     }
-    else
-    {
-        /* "Less than" subtree is deeper. */
+  } else {
+    /* "Less than" subtree is deeper. */
 
-        deep_h = AVL_GET_LESS(bal_h, 1);
-        L_CHECK_READ_ERROR(AVL_NULL)
+    deep_h = AVL_GET_LESS(bal_h, 1);
+    L_CHECK_READ_ERROR(AVL_NULL)
 
-        if (AVL_GET_BALANCE_FACTOR(deep_h) > 0)
-        {
-            int bf;
-            AVL_HANDLE old_h = bal_h;
-            bal_h = AVL_GET_GREATER(deep_h, 1);
-            L_CHECK_READ_ERROR(AVL_NULL)
-            AVL_SET_LESS(old_h, AVL_GET_GREATER(bal_h, 0))
-            AVL_SET_GREATER(deep_h, AVL_GET_LESS(bal_h, 0))
-            AVL_SET_GREATER(bal_h, old_h)
-            AVL_SET_LESS(bal_h, deep_h)
+    if (AVL_GET_BALANCE_FACTOR(deep_h) > 0) {
+      int bf;
+      AVL_HANDLE old_h = bal_h;
+      bal_h = AVL_GET_GREATER(deep_h, 1);
+      L_CHECK_READ_ERROR(AVL_NULL)
+      AVL_SET_LESS(old_h, AVL_GET_GREATER(bal_h, 0))
+      AVL_SET_GREATER(deep_h, AVL_GET_LESS(bal_h, 0))
+      AVL_SET_GREATER(bal_h, old_h)
+      AVL_SET_LESS(bal_h, deep_h)
 
-            bf = AVL_GET_BALANCE_FACTOR(bal_h);
+      bf = AVL_GET_BALANCE_FACTOR(bal_h);
 
-            if (bf != 0)
-            {
-                if (bf < 0)
-                {
-                    AVL_SET_BALANCE_FACTOR(old_h, 1)
-                    AVL_SET_BALANCE_FACTOR(deep_h, 0)
-                }
-                else
-                {
-                    AVL_SET_BALANCE_FACTOR(deep_h, -1)
-                    AVL_SET_BALANCE_FACTOR(old_h, 0)
-                }
-
-                AVL_SET_BALANCE_FACTOR(bal_h, 0)
-            }
-            else
-            {
-                AVL_SET_BALANCE_FACTOR(old_h, 0)
-                AVL_SET_BALANCE_FACTOR(deep_h, 0)
-            }
+      if (bf != 0) {
+        if (bf < 0) {
+          AVL_SET_BALANCE_FACTOR(old_h, 1)
+          AVL_SET_BALANCE_FACTOR(deep_h, 0)
+        } else {
+          AVL_SET_BALANCE_FACTOR(deep_h, -1)
+          AVL_SET_BALANCE_FACTOR(old_h, 0)
         }
-        else
-        {
-            AVL_SET_LESS(bal_h, AVL_GET_GREATER(deep_h, 0))
-            AVL_SET_GREATER(deep_h, bal_h)
 
-            if (AVL_GET_BALANCE_FACTOR(deep_h) == 0)
-            {
-                AVL_SET_BALANCE_FACTOR(deep_h, 1)
-                AVL_SET_BALANCE_FACTOR(bal_h, -1)
-            }
-            else
-            {
-                AVL_SET_BALANCE_FACTOR(deep_h, 0)
-                AVL_SET_BALANCE_FACTOR(bal_h, 0)
-            }
+        AVL_SET_BALANCE_FACTOR(bal_h, 0)
+      } else {
+        AVL_SET_BALANCE_FACTOR(old_h, 0)
+        AVL_SET_BALANCE_FACTOR(deep_h, 0)
+      }
+    } else {
+      AVL_SET_LESS(bal_h, AVL_GET_GREATER(deep_h, 0))
+      AVL_SET_GREATER(deep_h, bal_h)
 
-            bal_h = deep_h;
-        }
+      if (AVL_GET_BALANCE_FACTOR(deep_h) == 0) {
+        AVL_SET_BALANCE_FACTOR(deep_h, 1)
+        AVL_SET_BALANCE_FACTOR(bal_h, -1)
+      } else {
+        AVL_SET_BALANCE_FACTOR(deep_h, 0)
+        AVL_SET_BALANCE_FACTOR(bal_h, 0)
+      }
+
+      bal_h = deep_h;
     }
+  }
 
-    return(bal_h);
+  return(bal_h);
 }
 
-L_SC AVL_HANDLE L_(insert)(L_(avl) *l_tree, AVL_HANDLE h)
-{
-    AVL_SET_LESS(h, AVL_NULL)
-    AVL_SET_GREATER(h, AVL_NULL)
-    AVL_SET_BALANCE_FACTOR(h, 0)
+L_SC AVL_HANDLE L_(insert)(L_(avl) *l_tree, AVL_HANDLE h) {
+  AVL_SET_LESS(h, AVL_NULL)
+  AVL_SET_GREATER(h, AVL_NULL)
+  AVL_SET_BALANCE_FACTOR(h, 0)
 
-    if (l_tree->root == AVL_NULL)
-        l_tree->root = h;
-    else
-    {
-        /* Last unbalanced node encountered in search for insertion point. */
-        AVL_HANDLE unbal = AVL_NULL;
-        /* Parent of last unbalanced node. */
-        AVL_HANDLE parent_unbal = AVL_NULL;
-        /* Balance factor of last unbalanced node. */
-        int unbal_bf;
+  if (l_tree->root == AVL_NULL)
+    l_tree->root = h;
+  else {
+    /* Last unbalanced node encountered in search for insertion point. */
+    AVL_HANDLE unbal = AVL_NULL;
+    /* Parent of last unbalanced node. */
+    AVL_HANDLE parent_unbal = AVL_NULL;
+    /* Balance factor of last unbalanced node. */
+    int unbal_bf;
 
-        /* Zero-based depth in tree. */
-        unsigned depth = 0, unbal_depth = 0;
+    /* Zero-based depth in tree. */
+    unsigned depth = 0, unbal_depth = 0;
 
-        /* Records a path into the tree.  If bit n is true, indicates
-        ** take greater branch from the nth node in the path, otherwise
-        ** take the less branch.  bit 0 gives branch from root, and
-        ** so on. */
-        L_BIT_ARR_DEFN(branch)
+    /* Records a path into the tree.  If bit n is true, indicates
+    ** take greater branch from the nth node in the path, otherwise
+    ** take the less branch.  bit 0 gives branch from root, and
+    ** so on. */
+    L_BIT_ARR_DEFN(branch)
 
-        AVL_HANDLE hh = l_tree->root;
-        AVL_HANDLE parent = AVL_NULL;
-        int cmp;
+    AVL_HANDLE hh = l_tree->root;
+    AVL_HANDLE parent = AVL_NULL;
+    int cmp;
 
-        do
-        {
-            if (AVL_GET_BALANCE_FACTOR(hh) != 0)
-            {
-                unbal = hh;
-                parent_unbal = parent;
-                unbal_depth = depth;
-            }
+    do {
+      if (AVL_GET_BALANCE_FACTOR(hh) != 0) {
+        unbal = hh;
+        parent_unbal = parent;
+        unbal_depth = depth;
+      }
 
-            cmp = AVL_COMPARE_NODE_NODE(h, hh);
+      cmp = AVL_COMPARE_NODE_NODE(h, hh);
 
-            if (cmp == 0)
-                /* Duplicate key. */
-                return(hh);
+      if (cmp == 0)
+        /* Duplicate key. */
+        return(hh);
 
-            parent = hh;
+      parent = hh;
 
-            if (cmp > 0)
-            {
-                hh = AVL_GET_GREATER(hh, 1);
-                L_BIT_ARR_1(branch, depth)
-            }
-            else
-            {
-                hh = AVL_GET_LESS(hh, 1);
-                L_BIT_ARR_0(branch, depth)
-            }
+      if (cmp > 0) {
+        hh = AVL_GET_GREATER(hh, 1);
+        L_BIT_ARR_1(branch, depth)
+      } else {
+        hh = AVL_GET_LESS(hh, 1);
+        L_BIT_ARR_0(branch, depth)
+      }
 
-            L_CHECK_READ_ERROR(AVL_NULL)
-            depth++;
-        }
-        while (hh != AVL_NULL);
+      L_CHECK_READ_ERROR(AVL_NULL)
+      depth++;
+    } while (hh != AVL_NULL);
 
-        /*  Add node to insert as leaf of tree. */
-        if (cmp < 0)
-            AVL_SET_LESS(parent, h)
-            else
-                AVL_SET_GREATER(parent, h)
+    /*  Add node to insert as leaf of tree. */
+    if (cmp < 0)
+      AVL_SET_LESS(parent, h)
+      else
+        AVL_SET_GREATER(parent, h)
 
-                depth = unbal_depth;
+        depth = unbal_depth;
 
-        if (unbal == AVL_NULL)
-            hh = l_tree->root;
-        else
-        {
-            cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
-            depth++;
-            unbal_bf = AVL_GET_BALANCE_FACTOR(unbal);
+    if (unbal == AVL_NULL)
+      hh = l_tree->root;
+    else {
+      cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
+      depth++;
+      unbal_bf = AVL_GET_BALANCE_FACTOR(unbal);
 
-            if (cmp < 0)
-                unbal_bf--;
-            else  /* cmp > 0 */
-                unbal_bf++;
+      if (cmp < 0)
+        unbal_bf--;
+      else  /* cmp > 0 */
+        unbal_bf++;
 
-            hh = cmp < 0 ? AVL_GET_LESS(unbal, 1) : AVL_GET_GREATER(unbal, 1);
-            L_CHECK_READ_ERROR(AVL_NULL)
+      hh = cmp < 0 ? AVL_GET_LESS(unbal, 1) : AVL_GET_GREATER(unbal, 1);
+      L_CHECK_READ_ERROR(AVL_NULL)
 
-            if ((unbal_bf != -2) && (unbal_bf != 2))
-            {
-                /* No rebalancing of tree is necessary. */
-                AVL_SET_BALANCE_FACTOR(unbal, unbal_bf)
-                unbal = AVL_NULL;
-            }
-        }
-
-        if (hh != AVL_NULL)
-            while (h != hh)
-            {
-                cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
-                depth++;
-
-                if (cmp < 0)
-                {
-                    AVL_SET_BALANCE_FACTOR(hh, -1)
-                    hh = AVL_GET_LESS(hh, 1);
-                }
-                else /* cmp > 0 */
-                {
-                    AVL_SET_BALANCE_FACTOR(hh, 1)
-                    hh = AVL_GET_GREATER(hh, 1);
-                }
-
-                L_CHECK_READ_ERROR(AVL_NULL)
-            }
-
-        if (unbal != AVL_NULL)
-        {
-            unbal = L_(balance)(L_BALANCE_PARAM_CALL_PREFIX unbal);
-            L_CHECK_READ_ERROR(AVL_NULL)
-
-            if (parent_unbal == AVL_NULL)
-                l_tree->root = unbal;
-            else
-            {
-                depth = unbal_depth - 1;
-                cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
-
-                if (cmp < 0)
-                    AVL_SET_LESS(parent_unbal, unbal)
-                    else  /* cmp > 0 */
-                        AVL_SET_GREATER(parent_unbal, unbal)
-                    }
-        }
-
+      if ((unbal_bf != -2) && (unbal_bf != 2)) {
+        /* No rebalancing of tree is necessary. */
+        AVL_SET_BALANCE_FACTOR(unbal, unbal_bf)
+        unbal = AVL_NULL;
+      }
     }
 
-    return(h);
+    if (hh != AVL_NULL)
+      while (h != hh) {
+        cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
+        depth++;
+
+        if (cmp < 0) {
+          AVL_SET_BALANCE_FACTOR(hh, -1)
+          hh = AVL_GET_LESS(hh, 1);
+        } else { /* cmp > 0 */
+          AVL_SET_BALANCE_FACTOR(hh, 1)
+          hh = AVL_GET_GREATER(hh, 1);
+        }
+
+        L_CHECK_READ_ERROR(AVL_NULL)
+      }
+
+    if (unbal != AVL_NULL) {
+      unbal = L_(balance)(L_BALANCE_PARAM_CALL_PREFIX unbal);
+      L_CHECK_READ_ERROR(AVL_NULL)
+
+      if (parent_unbal == AVL_NULL)
+        l_tree->root = unbal;
+      else {
+        depth = unbal_depth - 1;
+        cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
+
+        if (cmp < 0)
+          AVL_SET_LESS(parent_unbal, unbal)
+          else  /* cmp > 0 */
+            AVL_SET_GREATER(parent_unbal, unbal)
+          }
+    }
+
+  }
+
+  return(h);
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_SEARCH)
 
-L_SC AVL_HANDLE L_(search)(L_(avl) *l_tree, AVL_KEY k, avl_search_type st)
-{
-    int cmp, target_cmp;
-    AVL_HANDLE match_h = AVL_NULL;
-    AVL_HANDLE h = l_tree->root;
+L_SC AVL_HANDLE L_(search)(L_(avl) *l_tree, AVL_KEY k, avl_search_type st) {
+  int cmp, target_cmp;
+  AVL_HANDLE match_h = AVL_NULL;
+  AVL_HANDLE h = l_tree->root;
 
-    if (st & AVL_LESS)
-        target_cmp = 1;
-    else if (st & AVL_GREATER)
-        target_cmp = -1;
-    else
-        target_cmp = 0;
+  if (st & AVL_LESS)
+    target_cmp = 1;
+  else if (st & AVL_GREATER)
+    target_cmp = -1;
+  else
+    target_cmp = 0;
 
-    while (h != AVL_NULL)
-    {
-        cmp = AVL_COMPARE_KEY_NODE(k, h);
+  while (h != AVL_NULL) {
+    cmp = AVL_COMPARE_KEY_NODE(k, h);
 
-        if (cmp == 0)
-        {
-            if (st & AVL_EQUAL)
-            {
-                match_h = h;
-                break;
-            }
+    if (cmp == 0) {
+      if (st & AVL_EQUAL) {
+        match_h = h;
+        break;
+      }
 
-            cmp = -target_cmp;
-        }
-        else if (target_cmp != 0)
-            if (!((cmp ^ target_cmp) & L_MASK_HIGH_BIT))
-                /* cmp and target_cmp are both positive or both negative. */
-                match_h = h;
+      cmp = -target_cmp;
+    } else if (target_cmp != 0)
+      if (!((cmp ^ target_cmp) & L_MASK_HIGH_BIT))
+        /* cmp and target_cmp are both positive or both negative. */
+        match_h = h;
 
-        h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
-        L_CHECK_READ_ERROR(AVL_NULL)
-    }
+    h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
+    L_CHECK_READ_ERROR(AVL_NULL)
+  }
 
-    return(match_h);
+  return(match_h);
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_SEARCH_LEAST)
 
-L_SC AVL_HANDLE L_(search_least)(L_(avl) *l_tree)
-{
-    AVL_HANDLE h = l_tree->root;
-    AVL_HANDLE parent = AVL_NULL;
+L_SC AVL_HANDLE L_(search_least)(L_(avl) *l_tree) {
+  AVL_HANDLE h = l_tree->root;
+  AVL_HANDLE parent = AVL_NULL;
 
-    while (h != AVL_NULL)
-    {
-        parent = h;
-        h = AVL_GET_LESS(h, 1);
-        L_CHECK_READ_ERROR(AVL_NULL)
-    }
+  while (h != AVL_NULL) {
+    parent = h;
+    h = AVL_GET_LESS(h, 1);
+    L_CHECK_READ_ERROR(AVL_NULL)
+  }
 
-    return(parent);
+  return(parent);
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_SEARCH_GREATEST)
 
-L_SC AVL_HANDLE L_(search_greatest)(L_(avl) *l_tree)
-{
-    AVL_HANDLE h = l_tree->root;
-    AVL_HANDLE parent = AVL_NULL;
+L_SC AVL_HANDLE L_(search_greatest)(L_(avl) *l_tree) {
+  AVL_HANDLE h = l_tree->root;
+  AVL_HANDLE parent = AVL_NULL;
 
-    while (h != AVL_NULL)
-    {
-        parent = h;
-        h = AVL_GET_GREATER(h, 1);
-        L_CHECK_READ_ERROR(AVL_NULL)
-    }
+  while (h != AVL_NULL) {
+    parent = h;
+    h = AVL_GET_GREATER(h, 1);
+    L_CHECK_READ_ERROR(AVL_NULL)
+  }
 
-    return(parent);
+  return(parent);
 }
 
 #endif
@@ -564,284 +509,253 @@
 */
 L_SC AVL_HANDLE L_(balance)(L_BALANCE_PARAM_DECL_PREFIX AVL_HANDLE bal_h);
 
-L_SC AVL_HANDLE L_(remove)(L_(avl) *l_tree, AVL_KEY k)
-{
-    /* Zero-based depth in tree. */
-    unsigned depth = 0, rm_depth;
+L_SC AVL_HANDLE L_(remove)(L_(avl) *l_tree, AVL_KEY k) {
+  /* Zero-based depth in tree. */
+  unsigned depth = 0, rm_depth;
 
-    /* Records a path into the tree.  If bit n is true, indicates
-    ** take greater branch from the nth node in the path, otherwise
-    ** take the less branch.  bit 0 gives branch from root, and
-    ** so on. */
-    L_BIT_ARR_DEFN(branch)
+  /* Records a path into the tree.  If bit n is true, indicates
+  ** take greater branch from the nth node in the path, otherwise
+  ** take the less branch.  bit 0 gives branch from root, and
+  ** so on. */
+  L_BIT_ARR_DEFN(branch)
 
-    AVL_HANDLE h = l_tree->root;
-    AVL_HANDLE parent = AVL_NULL;
-    AVL_HANDLE child;
-    AVL_HANDLE path;
-    int cmp, cmp_shortened_sub_with_path;
-    int reduced_depth;
-    int bf;
-    AVL_HANDLE rm;
-    AVL_HANDLE parent_rm;
+  AVL_HANDLE h = l_tree->root;
+  AVL_HANDLE parent = AVL_NULL;
+  AVL_HANDLE child;
+  AVL_HANDLE path;
+  int cmp, cmp_shortened_sub_with_path;
+  int reduced_depth;
+  int bf;
+  AVL_HANDLE rm;
+  AVL_HANDLE parent_rm;
 
-    for (; ;)
-    {
-        if (h == AVL_NULL)
-            /* No node in tree with given key. */
-            return(AVL_NULL);
+  for (;;) {
+    if (h == AVL_NULL)
+      /* No node in tree with given key. */
+      return(AVL_NULL);
 
-        cmp = AVL_COMPARE_KEY_NODE(k, h);
+    cmp = AVL_COMPARE_KEY_NODE(k, h);
 
-        if (cmp == 0)
-            /* Found node to remove. */
-            break;
+    if (cmp == 0)
+      /* Found node to remove. */
+      break;
 
-        parent = h;
+    parent = h;
 
-        if (cmp > 0)
-        {
-            h = AVL_GET_GREATER(h, 1);
-            L_BIT_ARR_1(branch, depth)
-        }
-        else
-        {
-            h = AVL_GET_LESS(h, 1);
-            L_BIT_ARR_0(branch, depth)
-        }
-
-        L_CHECK_READ_ERROR(AVL_NULL)
-        depth++;
-        cmp_shortened_sub_with_path = cmp;
-    }
-
-    rm = h;
-    parent_rm = parent;
-    rm_depth = depth;
-
-    /* If the node to remove is not a leaf node, we need to get a
-    ** leaf node, or a node with a single leaf as its child, to put
-    ** in the place of the node to remove.  We will get the greatest
-    ** node in the less subtree (of the node to remove), or the least
-    ** node in the greater subtree.  We take the leaf node from the
-    ** deeper subtree, if there is one. */
-
-    if (AVL_GET_BALANCE_FACTOR(h) < 0)
-    {
-        child = AVL_GET_LESS(h, 1);
-        L_BIT_ARR_0(branch, depth)
-        cmp = -1;
-    }
-    else
-    {
-        child = AVL_GET_GREATER(h, 1);
-        L_BIT_ARR_1(branch, depth)
-        cmp = 1;
+    if (cmp > 0) {
+      h = AVL_GET_GREATER(h, 1);
+      L_BIT_ARR_1(branch, depth)
+    } else {
+      h = AVL_GET_LESS(h, 1);
+      L_BIT_ARR_0(branch, depth)
     }
 
     L_CHECK_READ_ERROR(AVL_NULL)
     depth++;
+    cmp_shortened_sub_with_path = cmp;
+  }
 
-    if (child != AVL_NULL)
-    {
-        cmp = -cmp;
+  rm = h;
+  parent_rm = parent;
+  rm_depth = depth;
 
-        do
-        {
-            parent = h;
-            h = child;
+  /* If the node to remove is not a leaf node, we need to get a
+  ** leaf node, or a node with a single leaf as its child, to put
+  ** in the place of the node to remove.  We will get the greatest
+  ** node in the less subtree (of the node to remove), or the least
+  ** node in the greater subtree.  We take the leaf node from the
+  ** deeper subtree, if there is one. */
 
-            if (cmp < 0)
-            {
-                child = AVL_GET_LESS(h, 1);
-                L_BIT_ARR_0(branch, depth)
-            }
-            else
-            {
-                child = AVL_GET_GREATER(h, 1);
-                L_BIT_ARR_1(branch, depth)
-            }
+  if (AVL_GET_BALANCE_FACTOR(h) < 0) {
+    child = AVL_GET_LESS(h, 1);
+    L_BIT_ARR_0(branch, depth)
+    cmp = -1;
+  } else {
+    child = AVL_GET_GREATER(h, 1);
+    L_BIT_ARR_1(branch, depth)
+    cmp = 1;
+  }
 
-            L_CHECK_READ_ERROR(AVL_NULL)
-            depth++;
-        }
-        while (child != AVL_NULL);
+  L_CHECK_READ_ERROR(AVL_NULL)
+  depth++;
 
-        if (parent == rm)
-            /* Only went through do loop once.  Deleted node will be replaced
-            ** in the tree structure by one of its immediate children. */
-            cmp_shortened_sub_with_path = -cmp;
+  if (child != AVL_NULL) {
+    cmp = -cmp;
+
+    do {
+      parent = h;
+      h = child;
+
+      if (cmp < 0) {
+        child = AVL_GET_LESS(h, 1);
+        L_BIT_ARR_0(branch, depth)
+      } else {
+        child = AVL_GET_GREATER(h, 1);
+        L_BIT_ARR_1(branch, depth)
+      }
+
+      L_CHECK_READ_ERROR(AVL_NULL)
+      depth++;
+    } while (child != AVL_NULL);
+
+    if (parent == rm)
+      /* Only went through do loop once.  Deleted node will be replaced
+      ** in the tree structure by one of its immediate children. */
+      cmp_shortened_sub_with_path = -cmp;
+    else
+      cmp_shortened_sub_with_path = cmp;
+
+    /* Get the handle of the opposite child, which may not be null. */
+    child = cmp > 0 ? AVL_GET_LESS(h, 0) : AVL_GET_GREATER(h, 0);
+  }
+
+  if (parent == AVL_NULL)
+    /* There were only 1 or 2 nodes in this tree. */
+    l_tree->root = child;
+  else if (cmp_shortened_sub_with_path < 0)
+    AVL_SET_LESS(parent, child)
+    else
+      AVL_SET_GREATER(parent, child)
+
+      /* "path" is the parent of the subtree being eliminated or reduced
+      ** from a depth of 2 to 1.  If "path" is the node to be removed, we
+      ** set path to the node we're about to poke into the position of the
+      ** node to be removed. */
+      path = parent == rm ? h : parent;
+
+  if (h != rm) {
+    /* Poke in the replacement for the node to be removed. */
+    AVL_SET_LESS(h, AVL_GET_LESS(rm, 0))
+    AVL_SET_GREATER(h, AVL_GET_GREATER(rm, 0))
+    AVL_SET_BALANCE_FACTOR(h, AVL_GET_BALANCE_FACTOR(rm))
+
+    if (parent_rm == AVL_NULL)
+      l_tree->root = h;
+    else {
+      depth = rm_depth - 1;
+
+      if (L_BIT_ARR_VAL(branch, depth))
+        AVL_SET_GREATER(parent_rm, h)
         else
-            cmp_shortened_sub_with_path = cmp;
+          AVL_SET_LESS(parent_rm, h)
+        }
+  }
 
-        /* Get the handle of the opposite child, which may not be null. */
-        child = cmp > 0 ? AVL_GET_LESS(h, 0) : AVL_GET_GREATER(h, 0);
+  if (path != AVL_NULL) {
+    /* Create a temporary linked list from the parent of the path node
+    ** to the root node. */
+    h = l_tree->root;
+    parent = AVL_NULL;
+    depth = 0;
+
+    while (h != path) {
+      if (L_BIT_ARR_VAL(branch, depth)) {
+        child = AVL_GET_GREATER(h, 1);
+        AVL_SET_GREATER(h, parent)
+      } else {
+        child = AVL_GET_LESS(h, 1);
+        AVL_SET_LESS(h, parent)
+      }
+
+      L_CHECK_READ_ERROR(AVL_NULL)
+      depth++;
+      parent = h;
+      h = child;
     }
 
-    if (parent == AVL_NULL)
-        /* There were only 1 or 2 nodes in this tree. */
-        l_tree->root = child;
-    else if (cmp_shortened_sub_with_path < 0)
-        AVL_SET_LESS(parent, child)
-        else
-            AVL_SET_GREATER(parent, child)
+    /* Climb from the path node to the root node using the linked
+    ** list, restoring the tree structure and rebalancing as necessary.
+    */
+    reduced_depth = 1;
+    cmp = cmp_shortened_sub_with_path;
 
-            /* "path" is the parent of the subtree being eliminated or reduced
-            ** from a depth of 2 to 1.  If "path" is the node to be removed, we
-            ** set path to the node we're about to poke into the position of the
-            ** node to be removed. */
-            path = parent == rm ? h : parent;
+    for (;;) {
+      if (reduced_depth) {
+        bf = AVL_GET_BALANCE_FACTOR(h);
 
-    if (h != rm)
-    {
-        /* Poke in the replacement for the node to be removed. */
-        AVL_SET_LESS(h, AVL_GET_LESS(rm, 0))
-        AVL_SET_GREATER(h, AVL_GET_GREATER(rm, 0))
-        AVL_SET_BALANCE_FACTOR(h, AVL_GET_BALANCE_FACTOR(rm))
+        if (cmp < 0)
+          bf++;
+        else  /* cmp > 0 */
+          bf--;
 
-        if (parent_rm == AVL_NULL)
-            l_tree->root = h;
-        else
-        {
-            depth = rm_depth - 1;
+        if ((bf == -2) || (bf == 2)) {
+          h = L_(balance)(L_BALANCE_PARAM_CALL_PREFIX h);
+          L_CHECK_READ_ERROR(AVL_NULL)
+          bf = AVL_GET_BALANCE_FACTOR(h);
+        } else
+          AVL_SET_BALANCE_FACTOR(h, bf)
+          reduced_depth = (bf == 0);
+      }
 
-            if (L_BIT_ARR_VAL(branch, depth))
-                AVL_SET_GREATER(parent_rm, h)
-                else
-                    AVL_SET_LESS(parent_rm, h)
-                }
+      if (parent == AVL_NULL)
+        break;
+
+      child = h;
+      h = parent;
+      depth--;
+      cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
+
+      if (cmp < 0) {
+        parent = AVL_GET_LESS(h, 1);
+        AVL_SET_LESS(h, child)
+      } else {
+        parent = AVL_GET_GREATER(h, 1);
+        AVL_SET_GREATER(h, child)
+      }
+
+      L_CHECK_READ_ERROR(AVL_NULL)
     }
 
-    if (path != AVL_NULL)
-    {
-        /* Create a temporary linked list from the parent of the path node
-        ** to the root node. */
-        h = l_tree->root;
-        parent = AVL_NULL;
-        depth = 0;
+    l_tree->root = h;
+  }
 
-        while (h != path)
-        {
-            if (L_BIT_ARR_VAL(branch, depth))
-            {
-                child = AVL_GET_GREATER(h, 1);
-                AVL_SET_GREATER(h, parent)
-            }
-            else
-            {
-                child = AVL_GET_LESS(h, 1);
-                AVL_SET_LESS(h, parent)
-            }
-
-            L_CHECK_READ_ERROR(AVL_NULL)
-            depth++;
-            parent = h;
-            h = child;
-        }
-
-        /* Climb from the path node to the root node using the linked
-        ** list, restoring the tree structure and rebalancing as necessary.
-        */
-        reduced_depth = 1;
-        cmp = cmp_shortened_sub_with_path;
-
-        for (; ;)
-        {
-            if (reduced_depth)
-            {
-                bf = AVL_GET_BALANCE_FACTOR(h);
-
-                if (cmp < 0)
-                    bf++;
-                else  /* cmp > 0 */
-                    bf--;
-
-                if ((bf == -2) || (bf == 2))
-                {
-                    h = L_(balance)(L_BALANCE_PARAM_CALL_PREFIX h);
-                    L_CHECK_READ_ERROR(AVL_NULL)
-                    bf = AVL_GET_BALANCE_FACTOR(h);
-                }
-                else
-                    AVL_SET_BALANCE_FACTOR(h, bf)
-                    reduced_depth = (bf == 0);
-            }
-
-            if (parent == AVL_NULL)
-                break;
-
-            child = h;
-            h = parent;
-            depth--;
-            cmp = L_BIT_ARR_VAL(branch, depth) ? 1 : -1;
-
-            if (cmp < 0)
-            {
-                parent = AVL_GET_LESS(h, 1);
-                AVL_SET_LESS(h, child)
-            }
-            else
-            {
-                parent = AVL_GET_GREATER(h, 1);
-                AVL_SET_GREATER(h, child)
-            }
-
-            L_CHECK_READ_ERROR(AVL_NULL)
-        }
-
-        l_tree->root = h;
-    }
-
-    return(rm);
+  return(rm);
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_SUBST)
 
-L_SC AVL_HANDLE L_(subst)(L_(avl) *l_tree, AVL_HANDLE new_node)
-{
-    AVL_HANDLE h = l_tree->root;
-    AVL_HANDLE parent = AVL_NULL;
-    int cmp, last_cmp;
+L_SC AVL_HANDLE L_(subst)(L_(avl) *l_tree, AVL_HANDLE new_node) {
+  AVL_HANDLE h = l_tree->root;
+  AVL_HANDLE parent = AVL_NULL;
+  int cmp, last_cmp;
 
-    /* Search for node already in tree with same key. */
-    for (; ;)
-    {
-        if (h == AVL_NULL)
-            /* No node in tree with same key as new node. */
-            return(AVL_NULL);
+  /* Search for node already in tree with same key. */
+  for (;;) {
+    if (h == AVL_NULL)
+      /* No node in tree with same key as new node. */
+      return(AVL_NULL);
 
-        cmp = AVL_COMPARE_NODE_NODE(new_node, h);
+    cmp = AVL_COMPARE_NODE_NODE(new_node, h);
 
-        if (cmp == 0)
-            /* Found the node to substitute new one for. */
-            break;
+    if (cmp == 0)
+      /* Found the node to substitute new one for. */
+      break;
 
-        last_cmp = cmp;
-        parent = h;
-        h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
-        L_CHECK_READ_ERROR(AVL_NULL)
-    }
+    last_cmp = cmp;
+    parent = h;
+    h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
+    L_CHECK_READ_ERROR(AVL_NULL)
+  }
 
-    /* Copy tree housekeeping fields from node in tree to new node. */
-    AVL_SET_LESS(new_node, AVL_GET_LESS(h, 0))
-    AVL_SET_GREATER(new_node, AVL_GET_GREATER(h, 0))
-    AVL_SET_BALANCE_FACTOR(new_node, AVL_GET_BALANCE_FACTOR(h))
+  /* Copy tree housekeeping fields from node in tree to new node. */
+  AVL_SET_LESS(new_node, AVL_GET_LESS(h, 0))
+  AVL_SET_GREATER(new_node, AVL_GET_GREATER(h, 0))
+  AVL_SET_BALANCE_FACTOR(new_node, AVL_GET_BALANCE_FACTOR(h))
 
-    if (parent == AVL_NULL)
-        /* New node is also new root. */
-        l_tree->root = new_node;
-    else
-    {
-        /* Make parent point to new node. */
-        if (last_cmp < 0)
-            AVL_SET_LESS(parent, new_node)
-            else
-                AVL_SET_GREATER(parent, new_node)
-            }
+  if (parent == AVL_NULL)
+    /* New node is also new root. */
+    l_tree->root = new_node;
+  else {
+    /* Make parent point to new node. */
+    if (last_cmp < 0)
+      AVL_SET_LESS(parent, new_node)
+      else
+        AVL_SET_GREATER(parent, new_node)
+      }
 
-    return(h);
+  return(h);
 }
 
 #endif
@@ -851,144 +765,136 @@
 #if (L_IMPL_MASK & AVL_IMPL_BUILD)
 
 L_SC int L_(build)(
-    L_(avl) *l_tree, AVL_BUILD_ITER_TYPE p, L_SIZE num_nodes)
-{
-    /* Gives path to subtree being built.  If bit n is false, branch
-    ** less from the node at depth n, if true branch greater. */
-    L_BIT_ARR_DEFN(branch)
+  L_(avl) *l_tree, AVL_BUILD_ITER_TYPE p, L_SIZE num_nodes) {
+  /* Gives path to subtree being built.  If bit n is false, branch
+  ** less from the node at depth n, if true branch greater. */
+  L_BIT_ARR_DEFN(branch)
 
-    /* If bit n is true, then for the current subtree at depth n, its
-    ** greater subtree has one more node than its less subtree. */
-    L_BIT_ARR_DEFN(rem)
+  /* If bit n is true, then for the current subtree at depth n, its
+  ** greater subtree has one more node than its less subtree. */
+  L_BIT_ARR_DEFN(rem)
 
-    /* Depth of root node of current subtree. */
-    unsigned depth = 0;
+  /* Depth of root node of current subtree. */
+  unsigned depth = 0;
 
-    /* Number of nodes in current subtree. */
-    L_SIZE num_sub = num_nodes;
+  /* Number of nodes in current subtree. */
+  L_SIZE num_sub = num_nodes;
 
-    /* The algorithm relies on a stack of nodes whose less subtree has
-    ** been built, but whose greater subtree has not yet been built.
-    ** The stack is implemented as linked list.  The nodes are linked
-    ** together by having the "greater" handle of a node set to the
-    ** next node in the list.  "less_parent" is the handle of the first
-    ** node in the list. */
-    AVL_HANDLE less_parent = AVL_NULL;
+  /* The algorithm relies on a stack of nodes whose less subtree has
+  ** been built, but whose greater subtree has not yet been built.
+  ** The stack is implemented as linked list.  The nodes are linked
+  ** together by having the "greater" handle of a node set to the
+  ** next node in the list.  "less_parent" is the handle of the first
+  ** node in the list. */
+  AVL_HANDLE less_parent = AVL_NULL;
 
-    /* h is root of current subtree, child is one of its children. */
-    AVL_HANDLE h;
-    AVL_HANDLE child;
+  /* h is root of current subtree, child is one of its children. */
+  AVL_HANDLE h;
+  AVL_HANDLE child;
 
-    if (num_nodes == 0)
-    {
-        l_tree->root = AVL_NULL;
-        return(1);
+  if (num_nodes == 0) {
+    l_tree->root = AVL_NULL;
+    return(1);
+  }
+
+  for (;;) {
+    while (num_sub > 2) {
+      /* Subtract one for root of subtree. */
+      num_sub--;
+
+      if (num_sub & 1)
+        L_BIT_ARR_1(rem, depth)
+        else
+          L_BIT_ARR_0(rem, depth)
+          L_BIT_ARR_0(branch, depth)
+          depth++;
+
+      num_sub >>= 1;
     }
 
-    for (; ;)
-    {
-        while (num_sub > 2)
-        {
-            /* Subtract one for root of subtree. */
-            num_sub--;
+    if (num_sub == 2) {
+      /* Build a subtree with two nodes, slanting to greater.
+      ** I arbitrarily chose to always have the extra node in the
+      ** greater subtree when there is an odd number of nodes to
+      ** split between the two subtrees. */
 
-            if (num_sub & 1)
-                L_BIT_ARR_1(rem, depth)
-                else
-                    L_BIT_ARR_0(rem, depth)
-                    L_BIT_ARR_0(branch, depth)
-                    depth++;
+      h = AVL_BUILD_ITER_VAL(p);
+      L_CHECK_READ_ERROR(0)
+      AVL_BUILD_ITER_INCR(p)
+      child = AVL_BUILD_ITER_VAL(p);
+      L_CHECK_READ_ERROR(0)
+      AVL_BUILD_ITER_INCR(p)
+      AVL_SET_LESS(child, AVL_NULL)
+      AVL_SET_GREATER(child, AVL_NULL)
+      AVL_SET_BALANCE_FACTOR(child, 0)
+      AVL_SET_GREATER(h, child)
+      AVL_SET_LESS(h, AVL_NULL)
+      AVL_SET_BALANCE_FACTOR(h, 1)
+    } else { /* num_sub == 1 */
+      /* Build a subtree with one node. */
 
-            num_sub >>= 1;
+      h = AVL_BUILD_ITER_VAL(p);
+      L_CHECK_READ_ERROR(0)
+      AVL_BUILD_ITER_INCR(p)
+      AVL_SET_LESS(h, AVL_NULL)
+      AVL_SET_GREATER(h, AVL_NULL)
+      AVL_SET_BALANCE_FACTOR(h, 0)
+    }
+
+    while (depth) {
+      depth--;
+
+      if (!L_BIT_ARR_VAL(branch, depth))
+        /* We've completed a less subtree. */
+        break;
+
+      /* We've completed a greater subtree, so attach it to
+      ** its parent (that is less than it).  We pop the parent
+      ** off the stack of less parents. */
+      child = h;
+      h = less_parent;
+      less_parent = AVL_GET_GREATER(h, 1);
+      L_CHECK_READ_ERROR(0)
+      AVL_SET_GREATER(h, child)
+      /* num_sub = 2 * (num_sub - rem[depth]) + rem[depth] + 1 */
+      num_sub <<= 1;
+      num_sub += L_BIT_ARR_VAL(rem, depth) ? 0 : 1;
+
+      if (num_sub & (num_sub - 1))
+        /* num_sub is not a power of 2. */
+        AVL_SET_BALANCE_FACTOR(h, 0)
+        else
+          /* num_sub is a power of 2. */
+          AVL_SET_BALANCE_FACTOR(h, 1)
         }
 
-        if (num_sub == 2)
-        {
-            /* Build a subtree with two nodes, slanting to greater.
-            ** I arbitrarily chose to always have the extra node in the
-            ** greater subtree when there is an odd number of nodes to
-            ** split between the two subtrees. */
+    if (num_sub == num_nodes)
+      /* We've completed the full tree. */
+      break;
 
-            h = AVL_BUILD_ITER_VAL(p);
-            L_CHECK_READ_ERROR(0)
-            AVL_BUILD_ITER_INCR(p)
-            child = AVL_BUILD_ITER_VAL(p);
-            L_CHECK_READ_ERROR(0)
-            AVL_BUILD_ITER_INCR(p)
-            AVL_SET_LESS(child, AVL_NULL)
-            AVL_SET_GREATER(child, AVL_NULL)
-            AVL_SET_BALANCE_FACTOR(child, 0)
-            AVL_SET_GREATER(h, child)
-            AVL_SET_LESS(h, AVL_NULL)
-            AVL_SET_BALANCE_FACTOR(h, 1)
-        }
-        else  /* num_sub == 1 */
-        {
-            /* Build a subtree with one node. */
+    /* The subtree we've completed is the less subtree of the
+    ** next node in the sequence. */
 
-            h = AVL_BUILD_ITER_VAL(p);
-            L_CHECK_READ_ERROR(0)
-            AVL_BUILD_ITER_INCR(p)
-            AVL_SET_LESS(h, AVL_NULL)
-            AVL_SET_GREATER(h, AVL_NULL)
-            AVL_SET_BALANCE_FACTOR(h, 0)
-        }
+    child = h;
+    h = AVL_BUILD_ITER_VAL(p);
+    L_CHECK_READ_ERROR(0)
+    AVL_BUILD_ITER_INCR(p)
+    AVL_SET_LESS(h, child)
 
-        while (depth)
-        {
-            depth--;
+    /* Put h into stack of less parents. */
+    AVL_SET_GREATER(h, less_parent)
+    less_parent = h;
 
-            if (!L_BIT_ARR_VAL(branch, depth))
-                /* We've completed a less subtree. */
-                break;
+    /* Proceed to creating greater than subtree of h. */
+    L_BIT_ARR_1(branch, depth)
+    num_sub += L_BIT_ARR_VAL(rem, depth) ? 1 : 0;
+    depth++;
 
-            /* We've completed a greater subtree, so attach it to
-            ** its parent (that is less than it).  We pop the parent
-            ** off the stack of less parents. */
-            child = h;
-            h = less_parent;
-            less_parent = AVL_GET_GREATER(h, 1);
-            L_CHECK_READ_ERROR(0)
-            AVL_SET_GREATER(h, child)
-            /* num_sub = 2 * (num_sub - rem[depth]) + rem[depth] + 1 */
-            num_sub <<= 1;
-            num_sub += L_BIT_ARR_VAL(rem, depth) ? 0 : 1;
+  } /* end for (;; ) */
 
-            if (num_sub & (num_sub - 1))
-                /* num_sub is not a power of 2. */
-                AVL_SET_BALANCE_FACTOR(h, 0)
-                else
-                    /* num_sub is a power of 2. */
-                    AVL_SET_BALANCE_FACTOR(h, 1)
-                }
+  l_tree->root = h;
 
-        if (num_sub == num_nodes)
-            /* We've completed the full tree. */
-            break;
-
-        /* The subtree we've completed is the less subtree of the
-        ** next node in the sequence. */
-
-        child = h;
-        h = AVL_BUILD_ITER_VAL(p);
-        L_CHECK_READ_ERROR(0)
-        AVL_BUILD_ITER_INCR(p)
-        AVL_SET_LESS(h, child)
-
-        /* Put h into stack of less parents. */
-        AVL_SET_GREATER(h, less_parent)
-        less_parent = h;
-
-        /* Proceed to creating greater than subtree of h. */
-        L_BIT_ARR_1(branch, depth)
-        num_sub += L_BIT_ARR_VAL(rem, depth) ? 1 : 0;
-        depth++;
-
-    } /* end for ( ; ; ) */
-
-    l_tree->root = h;
-
-    return(1);
+  return(1);
 }
 
 #endif
@@ -1001,9 +907,8 @@
 ** invalid.   (Depth is zero-base.)  It's not necessary to initialize
 ** iterators prior to passing them to the "start" function.
 */
-L_SC void L_(init_iter)(L_(iter) *iter)
-{
-    iter->depth = ~0;
+L_SC void L_(init_iter)(L_(iter) *iter) {
+  iter->depth = ~0;
 }
 
 #endif
@@ -1011,7 +916,7 @@
 #ifdef AVL_READ_ERRORS_HAPPEN
 
 #define L_CHECK_READ_ERROR_INV_DEPTH \
-    { if (AVL_READ_ERROR) { iter->depth = ~0; return; } }
+  { if (AVL_READ_ERROR) { iter->depth = ~0; return; } }
 
 #else
 
@@ -1022,174 +927,157 @@
 #if (L_IMPL_MASK & AVL_IMPL_START_ITER)
 
 L_SC void L_(start_iter)(
-    L_(avl) *l_tree, L_(iter) *iter, AVL_KEY k, avl_search_type st)
-{
-    AVL_HANDLE h = l_tree->root;
-    unsigned d = 0;
-    int cmp, target_cmp;
+  L_(avl) *l_tree, L_(iter) *iter, AVL_KEY k, avl_search_type st) {
+  AVL_HANDLE h = l_tree->root;
+  unsigned d = 0;
+  int cmp, target_cmp;
 
-    /* Save the tree that we're going to iterate through in a
-    ** member variable. */
-    iter->tree_ = l_tree;
+  /* Save the tree that we're going to iterate through in a
+  ** member variable. */
+  iter->tree_ = l_tree;
 
-    iter->depth = ~0;
+  iter->depth = ~0;
+
+  if (h == AVL_NULL)
+    /* Tree is empty. */
+    return;
+
+  if (st & AVL_LESS)
+    /* Key can be greater than key of starting node. */
+    target_cmp = 1;
+  else if (st & AVL_GREATER)
+    /* Key can be less than key of starting node. */
+    target_cmp = -1;
+  else
+    /* Key must be same as key of starting node. */
+    target_cmp = 0;
+
+  for (;;) {
+    cmp = AVL_COMPARE_KEY_NODE(k, h);
+
+    if (cmp == 0) {
+      if (st & AVL_EQUAL) {
+        /* Equal node was sought and found as starting node. */
+        iter->depth = d;
+        break;
+      }
+
+      cmp = -target_cmp;
+    } else if (target_cmp != 0)
+      if (!((cmp ^ target_cmp) & L_MASK_HIGH_BIT))
+        /* cmp and target_cmp are both negative or both positive. */
+        iter->depth = d;
+
+    h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
+    L_CHECK_READ_ERROR_INV_DEPTH
 
     if (h == AVL_NULL)
-        /* Tree is empty. */
-        return;
+      break;
 
-    if (st & AVL_LESS)
-        /* Key can be greater than key of starting node. */
-        target_cmp = 1;
-    else if (st & AVL_GREATER)
-        /* Key can be less than key of starting node. */
-        target_cmp = -1;
-    else
-        /* Key must be same as key of starting node. */
-        target_cmp = 0;
-
-    for (; ;)
-    {
-        cmp = AVL_COMPARE_KEY_NODE(k, h);
-
-        if (cmp == 0)
-        {
-            if (st & AVL_EQUAL)
-            {
-                /* Equal node was sought and found as starting node. */
-                iter->depth = d;
-                break;
-            }
-
-            cmp = -target_cmp;
-        }
-        else if (target_cmp != 0)
-            if (!((cmp ^ target_cmp) & L_MASK_HIGH_BIT))
-                /* cmp and target_cmp are both negative or both positive. */
-                iter->depth = d;
-
-        h = cmp < 0 ? AVL_GET_LESS(h, 1) : AVL_GET_GREATER(h, 1);
-        L_CHECK_READ_ERROR_INV_DEPTH
-
-        if (h == AVL_NULL)
-            break;
-
-        if (cmp > 0)
-            L_BIT_ARR_1(iter->branch, d)
-            else
-                L_BIT_ARR_0(iter->branch, d)
-                iter->path_h[d++] = h;
-    }
+    if (cmp > 0)
+      L_BIT_ARR_1(iter->branch, d)
+      else
+        L_BIT_ARR_0(iter->branch, d)
+        iter->path_h[d++] = h;
+  }
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_START_ITER_LEAST)
 
-L_SC void L_(start_iter_least)(L_(avl) *l_tree, L_(iter) *iter)
-{
-    AVL_HANDLE h = l_tree->root;
+L_SC void L_(start_iter_least)(L_(avl) *l_tree, L_(iter) *iter) {
+  AVL_HANDLE h = l_tree->root;
 
-    iter->tree_ = l_tree;
+  iter->tree_ = l_tree;
 
-    iter->depth = ~0;
+  iter->depth = ~0;
 
-    L_BIT_ARR_ALL(iter->branch, 0)
+  L_BIT_ARR_ALL(iter->branch, 0)
 
-    while (h != AVL_NULL)
-    {
-        if (iter->depth != ~0)
-            iter->path_h[iter->depth] = h;
+  while (h != AVL_NULL) {
+    if (iter->depth != ~0)
+      iter->path_h[iter->depth] = h;
 
-        iter->depth++;
-        h = AVL_GET_LESS(h, 1);
-        L_CHECK_READ_ERROR_INV_DEPTH
-    }
+    iter->depth++;
+    h = AVL_GET_LESS(h, 1);
+    L_CHECK_READ_ERROR_INV_DEPTH
+  }
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_START_ITER_GREATEST)
 
-L_SC void L_(start_iter_greatest)(L_(avl) *l_tree, L_(iter) *iter)
-{
-    AVL_HANDLE h = l_tree->root;
+L_SC void L_(start_iter_greatest)(L_(avl) *l_tree, L_(iter) *iter) {
+  AVL_HANDLE h = l_tree->root;
 
-    iter->tree_ = l_tree;
+  iter->tree_ = l_tree;
 
-    iter->depth = ~0;
+  iter->depth = ~0;
 
-    L_BIT_ARR_ALL(iter->branch, 1)
+  L_BIT_ARR_ALL(iter->branch, 1)
 
-    while (h != AVL_NULL)
-    {
-        if (iter->depth != ~0)
-            iter->path_h[iter->depth] = h;
+  while (h != AVL_NULL) {
+    if (iter->depth != ~0)
+      iter->path_h[iter->depth] = h;
 
-        iter->depth++;
-        h = AVL_GET_GREATER(h, 1);
-        L_CHECK_READ_ERROR_INV_DEPTH
-    }
+    iter->depth++;
+    h = AVL_GET_GREATER(h, 1);
+    L_CHECK_READ_ERROR_INV_DEPTH
+  }
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_GET_ITER)
 
-L_SC AVL_HANDLE L_(get_iter)(L_(iter) *iter)
-{
-    if (iter->depth == ~0)
-        return(AVL_NULL);
+L_SC AVL_HANDLE L_(get_iter)(L_(iter) *iter) {
+  if (iter->depth == ~0)
+    return(AVL_NULL);
 
-    return(iter->depth == 0 ?
-           iter->tree_->root : iter->path_h[iter->depth - 1]);
+  return(iter->depth == 0 ?
+         iter->tree_->root : iter->path_h[iter->depth - 1]);
 }
 
 #endif
 
 #if (L_IMPL_MASK & AVL_IMPL_INCR_ITER)
 
-L_SC void L_(incr_iter)(L_(iter) *iter)
-{
+L_SC void L_(incr_iter)(L_(iter) *iter) {
 #define l_tree (iter->tree_)
 
-    if (iter->depth != ~0)
-    {
-        AVL_HANDLE h =
-            AVL_GET_GREATER((iter->depth == 0 ?
-                             iter->tree_->root : iter->path_h[iter->depth - 1]), 1);
+  if (iter->depth != ~0) {
+    AVL_HANDLE h =
+      AVL_GET_GREATER((iter->depth == 0 ?
+                       iter->tree_->root : iter->path_h[iter->depth - 1]), 1);
+    L_CHECK_READ_ERROR_INV_DEPTH
+
+    if (h == AVL_NULL)
+      do {
+        if (iter->depth == 0) {
+          iter->depth = ~0;
+          break;
+        }
+
+        iter->depth--;
+      } while (L_BIT_ARR_VAL(iter->branch, iter->depth));
+    else {
+      L_BIT_ARR_1(iter->branch, iter->depth)
+      iter->path_h[iter->depth++] = h;
+
+      for (;;) {
+        h = AVL_GET_LESS(h, 1);
         L_CHECK_READ_ERROR_INV_DEPTH
 
         if (h == AVL_NULL)
-            do
-            {
-                if (iter->depth == 0)
-                {
-                    iter->depth = ~0;
-                    break;
-                }
+          break;
 
-                iter->depth--;
-            }
-            while (L_BIT_ARR_VAL(iter->branch, iter->depth));
-        else
-        {
-            L_BIT_ARR_1(iter->branch, iter->depth)
-            iter->path_h[iter->depth++] = h;
-
-            for (; ;)
-            {
-                h = AVL_GET_LESS(h, 1);
-                L_CHECK_READ_ERROR_INV_DEPTH
-
-                if (h == AVL_NULL)
-                    break;
-
-                L_BIT_ARR_0(iter->branch, iter->depth)
-                iter->path_h[iter->depth++] = h;
-            }
-        }
+        L_BIT_ARR_0(iter->branch, iter->depth)
+        iter->path_h[iter->depth++] = h;
+      }
     }
+  }
 
 #undef l_tree
 }
@@ -1198,47 +1086,40 @@
 
 #if (L_IMPL_MASK & AVL_IMPL_DECR_ITER)
 
-L_SC void L_(decr_iter)(L_(iter) *iter)
-{
+L_SC void L_(decr_iter)(L_(iter) *iter) {
 #define l_tree (iter->tree_)
 
-    if (iter->depth != ~0)
-    {
-        AVL_HANDLE h =
-            AVL_GET_LESS((iter->depth == 0 ?
-                          iter->tree_->root : iter->path_h[iter->depth - 1]), 1);
+  if (iter->depth != ~0) {
+    AVL_HANDLE h =
+      AVL_GET_LESS((iter->depth == 0 ?
+                    iter->tree_->root : iter->path_h[iter->depth - 1]), 1);
+    L_CHECK_READ_ERROR_INV_DEPTH
+
+    if (h == AVL_NULL)
+      do {
+        if (iter->depth == 0) {
+          iter->depth = ~0;
+          break;
+        }
+
+        iter->depth--;
+      } while (!L_BIT_ARR_VAL(iter->branch, iter->depth));
+    else {
+      L_BIT_ARR_0(iter->branch, iter->depth)
+      iter->path_h[iter->depth++] = h;
+
+      for (;;) {
+        h = AVL_GET_GREATER(h, 1);
         L_CHECK_READ_ERROR_INV_DEPTH
 
         if (h == AVL_NULL)
-            do
-            {
-                if (iter->depth == 0)
-                {
-                    iter->depth = ~0;
-                    break;
-                }
+          break;
 
-                iter->depth--;
-            }
-            while (!L_BIT_ARR_VAL(iter->branch, iter->depth));
-        else
-        {
-            L_BIT_ARR_0(iter->branch, iter->depth)
-            iter->path_h[iter->depth++] = h;
-
-            for (; ;)
-            {
-                h = AVL_GET_GREATER(h, 1);
-                L_CHECK_READ_ERROR_INV_DEPTH
-
-                if (h == AVL_NULL)
-                    break;
-
-                L_BIT_ARR_1(iter->branch, iter->depth)
-                iter->path_h[iter->depth++] = h;
-            }
-        }
+        L_BIT_ARR_1(iter->branch, iter->depth)
+        iter->path_h[iter->depth++] = h;
+      }
     }
+  }
 
 #undef l_tree
 }
diff --git a/libvpx/vpx_mem/memory_manager/include/heapmm.h b/libvpx/vpx_mem/memory_manager/include/heapmm.h
index 33004ca..4934c2d 100644
--- a/libvpx/vpx_mem/memory_manager/include/heapmm.h
+++ b/libvpx/vpx_mem/memory_manager/include/heapmm.h
@@ -81,30 +81,29 @@
 #include "hmm_cnfg.h"
 
 /* Heap descriptor. */
-typedef struct HMM_UNIQUE(structure)
-{
-    /* private: */
+typedef struct HMM_UNIQUE(structure) {
+  /* private: */
 
-    /* Pointer to (payload of) root node in AVL tree.  This field should
-    ** really be the AVL tree descriptor (type avl_avl).  But (in the
-    ** instantiation of the AVL tree generic package used in package) the
-    ** AVL tree descriptor simply contains a pointer to the root.  So,
-    ** whenever a pointer to the AVL tree descriptor is needed, I use the
-    ** cast:
-    **
-    ** (avl_avl *) &(heap_desc->avl_tree_root)
-    **
-    ** (where heap_desc is a pointer to a heap descriptor).  This trick
-    ** allows me to avoid including cavl_if.h in this external header. */
-    void *avl_tree_root;
+  /* Pointer to (payload of) root node in AVL tree.  This field should
+  ** really be the AVL tree descriptor (type avl_avl).  But (in the
+  ** instantiation of the AVL tree generic package used in package) the
+  ** AVL tree descriptor simply contains a pointer to the root.  So,
+  ** whenever a pointer to the AVL tree descriptor is needed, I use the
+  ** cast:
+  **
+  ** (avl_avl *) &(heap_desc->avl_tree_root)
+  **
+  ** (where heap_desc is a pointer to a heap descriptor).  This trick
+  ** allows me to avoid including cavl_if.h in this external header. */
+  void *avl_tree_root;
 
-    /* Pointer to first byte of last block freed, after any coalescing. */
-    void *last_freed;
+  /* Pointer to first byte of last block freed, after any coalescing. */
+  void *last_freed;
 
-    /* public: */
+  /* public: */
 
-    HMM_UNIQUE(size_bau) num_baus_can_shrink;
-    void *end_of_shrinkable_chunk;
+  HMM_UNIQUE(size_bau) num_baus_can_shrink;
+  void *end_of_shrinkable_chunk;
 }
 HMM_UNIQUE(descriptor);
 
@@ -113,41 +112,41 @@
 void HMM_UNIQUE(init)(HMM_UNIQUE(descriptor) *desc);
 
 void *HMM_UNIQUE(alloc)(
-    HMM_UNIQUE(descriptor) *desc, HMM_UNIQUE(size_aau) num_addr_align_units);
+  HMM_UNIQUE(descriptor) *desc, HMM_UNIQUE(size_aau) num_addr_align_units);
 
 /* NOT YET IMPLEMENTED */
 void *HMM_UNIQUE(greedy_alloc)(
-    HMM_UNIQUE(descriptor) *desc, HMM_UNIQUE(size_aau) needed_addr_align_units,
-    HMM_UNIQUE(size_aau) coveted_addr_align_units);
+  HMM_UNIQUE(descriptor) *desc, HMM_UNIQUE(size_aau) needed_addr_align_units,
+  HMM_UNIQUE(size_aau) coveted_addr_align_units);
 
 int HMM_UNIQUE(resize)(
-    HMM_UNIQUE(descriptor) *desc, void *mem,
-    HMM_UNIQUE(size_aau) num_addr_align_units);
+  HMM_UNIQUE(descriptor) *desc, void *mem,
+  HMM_UNIQUE(size_aau) num_addr_align_units);
 
 /* NOT YET IMPLEMENTED */
 int HMM_UNIQUE(greedy_resize)(
-    HMM_UNIQUE(descriptor) *desc, void *mem,
-    HMM_UNIQUE(size_aau) needed_addr_align_units,
-    HMM_UNIQUE(size_aau) coveted_addr_align_units);
+  HMM_UNIQUE(descriptor) *desc, void *mem,
+  HMM_UNIQUE(size_aau) needed_addr_align_units,
+  HMM_UNIQUE(size_aau) coveted_addr_align_units);
 
 void HMM_UNIQUE(free)(HMM_UNIQUE(descriptor) *desc, void *mem);
 
 HMM_UNIQUE(size_aau) HMM_UNIQUE(true_size)(void *mem);
 
 HMM_UNIQUE(size_aau) HMM_UNIQUE(largest_available)(
-    HMM_UNIQUE(descriptor) *desc);
+  HMM_UNIQUE(descriptor) *desc);
 
 void HMM_UNIQUE(new_chunk)(
-    HMM_UNIQUE(descriptor) *desc, void *start_of_chunk,
-    HMM_UNIQUE(size_bau) num_block_align_units);
+  HMM_UNIQUE(descriptor) *desc, void *start_of_chunk,
+  HMM_UNIQUE(size_bau) num_block_align_units);
 
 void HMM_UNIQUE(grow_chunk)(
-    HMM_UNIQUE(descriptor) *desc, void *end_of_chunk,
-    HMM_UNIQUE(size_bau) num_block_align_units);
+  HMM_UNIQUE(descriptor) *desc, void *end_of_chunk,
+  HMM_UNIQUE(size_bau) num_block_align_units);
 
 /* NOT YET IMPLEMENTED */
 void HMM_UNIQUE(shrink_chunk)(
-    HMM_UNIQUE(descriptor) *desc,
-    HMM_UNIQUE(size_bau) num_block_align_units);
+  HMM_UNIQUE(descriptor) *desc,
+  HMM_UNIQUE(size_bau) num_block_align_units);
 
 #endif /* defined HMM_PROCESS */
diff --git a/libvpx/vpx_mem/memory_manager/include/hmm_cnfg.h b/libvpx/vpx_mem/memory_manager/include/hmm_cnfg.h
index 30b9f50..2c3391d 100644
--- a/libvpx/vpx_mem/memory_manager/include/hmm_cnfg.h
+++ b/libvpx/vpx_mem/memory_manager/include/hmm_cnfg.h
@@ -45,8 +45,8 @@
 #define HMM_UNIQUE(BASE) hmm_ ## BASE
 
 /* Number of bytes in an Address Alignment Unit (AAU). */
-//fwg
-//#define HMM_ADDR_ALIGN_UNIT sizeof(int)
+// fwg
+// #define HMM_ADDR_ALIGN_UNIT sizeof(int)
 #define HMM_ADDR_ALIGN_UNIT 32
 
 /* Number of AAUs in a Block Alignment Unit (BAU). */
@@ -65,7 +65,7 @@
 ** statement.  If you remove the definition of this macro, no self-auditing
 ** will be performed. */
 #define HMM_AUDIT_FAIL \
-    hmm_dflt_abort(__FILE__, HMM_SYM_TO_STRING(__LINE__));
+  hmm_dflt_abort(__FILE__, HMM_SYM_TO_STRING(__LINE__));
 
 #elif HMM_CNFG_NUM == 0
 
@@ -90,8 +90,8 @@
 extern unsigned HMM_UNIQUE(fail_line);
 
 #define HMM_AUDIT_FAIL \
-    { HMM_UNIQUE(fail_file) = __FILE__; HMM_UNIQUE(fail_line) = __LINE__; \
-        longjmp(HMM_UNIQUE(jmp_buf), 1); }
+  { HMM_UNIQUE(fail_file) = __FILE__; HMM_UNIQUE(fail_line) = __LINE__; \
+    longjmp(HMM_UNIQUE(jmp_buf), 1); }
 
 #elif HMM_CNFG_NUM == 1
 
diff --git a/libvpx/vpx_mem/memory_manager/include/hmm_intrnl.h b/libvpx/vpx_mem/memory_manager/include/hmm_intrnl.h
index 5d62abc..27cefe4 100644
--- a/libvpx/vpx_mem/memory_manager/include/hmm_intrnl.h
+++ b/libvpx/vpx_mem/memory_manager/include/hmm_intrnl.h
@@ -26,34 +26,32 @@
 
 /* Mask of high bit of variable of size_bau type. */
 #define HIGH_BIT_BAU_SIZE \
-    ((U(size_bau)) ~ (((U(size_bau)) ~ (U(size_bau)) 0) >> 1))
+  ((U(size_bau)) ~ (((U(size_bau)) ~ (U(size_bau)) 0) >> 1))
 
 /* Add a given number of AAUs to pointer. */
 #define AAUS_FORWARD(PTR, AAU_OFFSET) \
-    (((char *) (PTR)) + ((AAU_OFFSET) * ((U(size_aau)) HMM_ADDR_ALIGN_UNIT)))
+  (((char *) (PTR)) + ((AAU_OFFSET) * ((U(size_aau)) HMM_ADDR_ALIGN_UNIT)))
 
 /* Subtract a given number of AAUs from pointer. */
 #define AAUS_BACKWARD(PTR, AAU_OFFSET) \
-    (((char *) (PTR)) - ((AAU_OFFSET) * ((U(size_aau)) HMM_ADDR_ALIGN_UNIT)))
+  (((char *) (PTR)) - ((AAU_OFFSET) * ((U(size_aau)) HMM_ADDR_ALIGN_UNIT)))
 
 /* Add a given number of BAUs to a pointer. */
 #define BAUS_FORWARD(PTR, BAU_OFFSET) \
-    AAUS_FORWARD((PTR), (BAU_OFFSET) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT))
+  AAUS_FORWARD((PTR), (BAU_OFFSET) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT))
 
 /* Subtract a given number of BAUs to a pointer. */
 #define BAUS_BACKWARD(PTR, BAU_OFFSET) \
-    AAUS_BACKWARD((PTR), (BAU_OFFSET) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT))
+  AAUS_BACKWARD((PTR), (BAU_OFFSET) * ((U(size_aau)) HMM_BLOCK_ALIGN_UNIT))
 
-typedef struct head_struct
-{
-    /* Sizes in Block Alignment Units. */
-    HMM_UNIQUE(size_bau) previous_block_size, block_size;
+typedef struct head_struct {
+  /* Sizes in Block Alignment Units. */
+  HMM_UNIQUE(size_bau) previous_block_size, block_size;
 }
 head_record;
 
-typedef struct ptr_struct
-{
-    struct ptr_struct *self, *prev, *next;
+typedef struct ptr_struct {
+  struct ptr_struct *self, *prev, *next;
 }
 ptr_record;
 
@@ -71,50 +69,50 @@
 
 /* Minimum number of BAUs in a block (allowing room for the pointer record. */
 #define MIN_BLOCK_BAUS \
-    DIV_ROUND_UP(HEAD_AAUS + PTR_RECORD_AAUS, HMM_BLOCK_ALIGN_UNIT)
+  DIV_ROUND_UP(HEAD_AAUS + PTR_RECORD_AAUS, HMM_BLOCK_ALIGN_UNIT)
 
 /* Return number of BAUs in block (masking off high bit containing block
 ** status). */
 #define BLOCK_BAUS(HEAD_PTR) \
-    (((head_record *) (HEAD_PTR))->block_size & ~HIGH_BIT_BAU_SIZE)
+  (((head_record *) (HEAD_PTR))->block_size & ~HIGH_BIT_BAU_SIZE)
 
 /* Return number of BAUs in previous block (masking off high bit containing
 ** block status). */
 #define PREV_BLOCK_BAUS(HEAD_PTR) \
-    (((head_record *) (HEAD_PTR))->previous_block_size & ~HIGH_BIT_BAU_SIZE)
+  (((head_record *) (HEAD_PTR))->previous_block_size & ~HIGH_BIT_BAU_SIZE)
 
 /* Set number of BAUs in previous block, preserving high bit containing
 ** block status. */
 #define SET_PREV_BLOCK_BAUS(HEAD_PTR, N_BAUS) \
-    { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
-        h_ptr->previous_block_size &= HIGH_BIT_BAU_SIZE; \
-        h_ptr->previous_block_size |= (N_BAUS); }
+  { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
+    h_ptr->previous_block_size &= HIGH_BIT_BAU_SIZE; \
+    h_ptr->previous_block_size |= (N_BAUS); }
 
 /* Convert pointer to pointer record of block to pointer to block's head
 ** record. */
 #define PTR_REC_TO_HEAD(PTR_REC_PTR) \
-    ((head_record *) AAUS_BACKWARD(PTR_REC_PTR, HEAD_AAUS))
+  ((head_record *) AAUS_BACKWARD(PTR_REC_PTR, HEAD_AAUS))
 
 /* Convert pointer to block head to pointer to block's pointer record. */
 #define HEAD_TO_PTR_REC(HEAD_PTR) \
-    ((ptr_record *) AAUS_FORWARD(HEAD_PTR, HEAD_AAUS))
+  ((ptr_record *) AAUS_FORWARD(HEAD_PTR, HEAD_AAUS))
 
 /* Returns non-zero if block is allocated. */
 #define IS_BLOCK_ALLOCATED(HEAD_PTR) \
-    (((((head_record *) (HEAD_PTR))->block_size | \
-       ((head_record *) (HEAD_PTR))->previous_block_size) & \
-      HIGH_BIT_BAU_SIZE) == 0)
+  (((((head_record *) (HEAD_PTR))->block_size | \
+     ((head_record *) (HEAD_PTR))->previous_block_size) & \
+    HIGH_BIT_BAU_SIZE) == 0)
 
 #define MARK_BLOCK_ALLOCATED(HEAD_PTR) \
-    { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
-        h_ptr->block_size &= ~HIGH_BIT_BAU_SIZE; \
-        h_ptr->previous_block_size &= ~HIGH_BIT_BAU_SIZE; }
+  { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
+    h_ptr->block_size &= ~HIGH_BIT_BAU_SIZE; \
+    h_ptr->previous_block_size &= ~HIGH_BIT_BAU_SIZE; }
 
 /* Mark a block as free when it is not the first block in a bin (and
 ** therefore not a node in the AVL tree). */
 #define MARK_SUCCESSIVE_BLOCK_IN_FREE_BIN(HEAD_PTR) \
-    { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
-        h_ptr->block_size |= HIGH_BIT_BAU_SIZE; }
+  { register head_record *h_ptr = (head_record *) (HEAD_PTR); \
+    h_ptr->block_size |= HIGH_BIT_BAU_SIZE; }
 
 /* Prototypes for internal functions implemented in one file and called in
 ** another.
@@ -125,7 +123,7 @@
 void U(out_of_free_collection)(U(descriptor) *desc, head_record *head_ptr);
 
 void *U(alloc_from_bin)(
-    U(descriptor) *desc, ptr_record *bin_front_ptr, U(size_bau) n_baus);
+  U(descriptor) *desc, ptr_record *bin_front_ptr, U(size_bau) n_baus);
 
 #ifdef HMM_AUDIT_FAIL
 
@@ -137,12 +135,12 @@
 /* Auditing a block consists of checking that the size in its head
 ** matches the previous block size in the head of the next block. */
 #define AUDIT_BLOCK_AS_EXPR(HEAD_PTR) \
-    ((BLOCK_BAUS(HEAD_PTR) == \
-      PREV_BLOCK_BAUS(BAUS_FORWARD(HEAD_PTR, BLOCK_BAUS(HEAD_PTR)))) ? \
-     0 : U(audit_block_fail_dummy_return)())
+  ((BLOCK_BAUS(HEAD_PTR) == \
+    PREV_BLOCK_BAUS(BAUS_FORWARD(HEAD_PTR, BLOCK_BAUS(HEAD_PTR)))) ? \
+   0 : U(audit_block_fail_dummy_return)())
 
 #define AUDIT_BLOCK(HEAD_PTR) \
-    { void *h_ptr = (HEAD_PTR); AUDIT_BLOCK_AS_EXPR(h_ptr); }
+  { void *h_ptr = (HEAD_PTR); AUDIT_BLOCK_AS_EXPR(h_ptr); }
 
 #endif
 
diff --git a/libvpx/vpx_mem/vpx_mem.c b/libvpx/vpx_mem/vpx_mem.c
index eade432..059248b 100644
--- a/libvpx/vpx_mem/vpx_mem.c
+++ b/libvpx/vpx_mem/vpx_mem.c
@@ -51,15 +51,14 @@
 #endif /*CONFIG_MEM_MANAGER*/
 
 #if USE_GLOBAL_FUNCTION_POINTERS
-struct GLOBAL_FUNC_POINTERS
-{
-    g_malloc_func g_malloc;
-    g_calloc_func g_calloc;
-    g_realloc_func g_realloc;
-    g_free_func g_free;
-    g_memcpy_func g_memcpy;
-    g_memset_func g_memset;
-    g_memmove_func g_memmove;
+struct GLOBAL_FUNC_POINTERS {
+  g_malloc_func g_malloc;
+  g_calloc_func g_calloc;
+  g_realloc_func g_realloc;
+  g_free_func g_free;
+  g_memcpy_func g_memcpy;
+  g_memset_func g_memset;
+  g_memmove_func g_memmove;
 } *g_func = NULL;
 
 # define VPX_MALLOC_L  g_func->g_malloc
@@ -77,346 +76,314 @@
 # define VPX_MEMMOVE_L memmove
 #endif /* USE_GLOBAL_FUNCTION_POINTERS */
 
-unsigned int vpx_mem_get_version()
-{
-    unsigned int ver = ((unsigned int)(unsigned char)VPX_MEM_VERSION_CHIEF << 24 |
-                        (unsigned int)(unsigned char)VPX_MEM_VERSION_MAJOR << 16 |
-                        (unsigned int)(unsigned char)VPX_MEM_VERSION_MINOR << 8  |
-                        (unsigned int)(unsigned char)VPX_MEM_VERSION_PATCH);
-    return ver;
+unsigned int vpx_mem_get_version() {
+  unsigned int ver = ((unsigned int)(unsigned char)VPX_MEM_VERSION_CHIEF << 24 |
+                      (unsigned int)(unsigned char)VPX_MEM_VERSION_MAJOR << 16 |
+                      (unsigned int)(unsigned char)VPX_MEM_VERSION_MINOR << 8  |
+                      (unsigned int)(unsigned char)VPX_MEM_VERSION_PATCH);
+  return ver;
 }
 
-int vpx_mem_set_heap_size(size_t size)
-{
-    int ret = -1;
+int vpx_mem_set_heap_size(size_t size) {
+  int ret = -1;
 
 #if CONFIG_MEM_MANAGER
 #if MM_DYNAMIC_MEMORY
 
-    if (!g_mng_memory_allocated && size)
-    {
-        g_mm_memory_size = size;
-        ret = 0;
-    }
-    else
-        ret = -3;
+  if (!g_mng_memory_allocated && size) {
+    g_mm_memory_size = size;
+    ret = 0;
+  } else
+    ret = -3;
 
 #else
-    ret = -2;
+  ret = -2;
 #endif
 #else
-    (void)size;
+  (void)size;
 #endif
 
-    return ret;
+  return ret;
 }
 
-void *vpx_memalign(size_t align, size_t size)
-{
-    void *addr,
-         * x = NULL;
+void *vpx_memalign(size_t align, size_t size) {
+  void *addr,
+       * x = NULL;
 
 #if CONFIG_MEM_MANAGER
-    int number_aau;
+  int number_aau;
 
-    if (vpx_mm_create_heap_memory() < 0)
-    {
-        _P(printf("[vpx][mm] ERROR vpx_memalign() Couldn't create memory for Heap.\n");)
-    }
+  if (vpx_mm_create_heap_memory() < 0) {
+    _P(printf("[vpx][mm] ERROR vpx_memalign() Couldn't create memory for Heap.\n");)
+  }
 
-    number_aau = ((size + align - 1 + ADDRESS_STORAGE_SIZE) >>
-                  SHIFT_HMM_ADDR_ALIGN_UNIT) + 1;
+  number_aau = ((size + align - 1 + ADDRESS_STORAGE_SIZE) >>
+                SHIFT_HMM_ADDR_ALIGN_UNIT) + 1;
 
-    addr = hmm_alloc(&hmm_d, number_aau);
+  addr = hmm_alloc(&hmm_d, number_aau);
 #else
-    addr = VPX_MALLOC_L(size + align - 1 + ADDRESS_STORAGE_SIZE);
+  addr = VPX_MALLOC_L(size + align - 1 + ADDRESS_STORAGE_SIZE);
 #endif /*CONFIG_MEM_MANAGER*/
 
-    if (addr)
-    {
-        x = align_addr((unsigned char *)addr + ADDRESS_STORAGE_SIZE, (int)align);
-        /* save the actual malloc address */
-        ((size_t *)x)[-1] = (size_t)addr;
-    }
+  if (addr) {
+    x = align_addr((unsigned char *)addr + ADDRESS_STORAGE_SIZE, (int)align);
+    /* save the actual malloc address */
+    ((size_t *)x)[-1] = (size_t)addr;
+  }
 
-    return x;
+  return x;
 }
 
-void *vpx_malloc(size_t size)
-{
-    return vpx_memalign(DEFAULT_ALIGNMENT, size);
+void *vpx_malloc(size_t size) {
+  return vpx_memalign(DEFAULT_ALIGNMENT, size);
 }
 
-void *vpx_calloc(size_t num, size_t size)
-{
-    void *x;
+void *vpx_calloc(size_t num, size_t size) {
+  void *x;
 
-    x = vpx_memalign(DEFAULT_ALIGNMENT, num * size);
+  x = vpx_memalign(DEFAULT_ALIGNMENT, num * size);
 
-    if (x)
-        VPX_MEMSET_L(x, 0, num * size);
+  if (x)
+    VPX_MEMSET_L(x, 0, num * size);
 
-    return x;
+  return x;
 }
 
-void *vpx_realloc(void *memblk, size_t size)
-{
-    void *addr,
-         * new_addr = NULL;
-    int align = DEFAULT_ALIGNMENT;
+void *vpx_realloc(void *memblk, size_t size) {
+  void *addr,
+       * new_addr = NULL;
+  int align = DEFAULT_ALIGNMENT;
 
-    /*
-    The realloc() function changes the size of the object pointed to by
-    ptr to the size specified by size, and returns a pointer to the
-    possibly moved block. The contents are unchanged up to the lesser
-    of the new and old sizes. If ptr is null, realloc() behaves like
-    malloc() for the specified size. If size is zero (0) and ptr is
-    not a null pointer, the object pointed to is freed.
-    */
-    if (!memblk)
-        new_addr = vpx_malloc(size);
-    else if (!size)
-        vpx_free(memblk);
-    else
-    {
-        addr   = (void *)(((size_t *)memblk)[-1]);
-        memblk = NULL;
+  /*
+  The realloc() function changes the size of the object pointed to by
+  ptr to the size specified by size, and returns a pointer to the
+  possibly moved block. The contents are unchanged up to the lesser
+  of the new and old sizes. If ptr is null, realloc() behaves like
+  malloc() for the specified size. If size is zero (0) and ptr is
+  not a null pointer, the object pointed to is freed.
+  */
+  if (!memblk)
+    new_addr = vpx_malloc(size);
+  else if (!size)
+    vpx_free(memblk);
+  else {
+    addr   = (void *)(((size_t *)memblk)[-1]);
+    memblk = NULL;
 
 #if CONFIG_MEM_MANAGER
-        new_addr = vpx_mm_realloc(addr, size + align + ADDRESS_STORAGE_SIZE);
+    new_addr = vpx_mm_realloc(addr, size + align + ADDRESS_STORAGE_SIZE);
 #else
-        new_addr = VPX_REALLOC_L(addr, size + align + ADDRESS_STORAGE_SIZE);
+    new_addr = VPX_REALLOC_L(addr, size + align + ADDRESS_STORAGE_SIZE);
 #endif
 
-        if (new_addr)
-        {
-            addr = new_addr;
-            new_addr = (void *)(((size_t)
-                                 ((unsigned char *)new_addr + ADDRESS_STORAGE_SIZE) + (align - 1)) &
-                                (size_t) - align);
-            /* save the actual malloc address */
-            ((size_t *)new_addr)[-1] = (size_t)addr;
-        }
+    if (new_addr) {
+      addr = new_addr;
+      new_addr = (void *)(((size_t)
+                           ((unsigned char *)new_addr + ADDRESS_STORAGE_SIZE) + (align - 1)) &
+                          (size_t) - align);
+      /* save the actual malloc address */
+      ((size_t *)new_addr)[-1] = (size_t)addr;
     }
+  }
 
-    return new_addr;
+  return new_addr;
 }
 
-void vpx_free(void *memblk)
-{
-    if (memblk)
-    {
-        void *addr = (void *)(((size_t *)memblk)[-1]);
+void vpx_free(void *memblk) {
+  if (memblk) {
+    void *addr = (void *)(((size_t *)memblk)[-1]);
 #if CONFIG_MEM_MANAGER
-        hmm_free(&hmm_d, addr);
+    hmm_free(&hmm_d, addr);
 #else
-        VPX_FREE_L(addr);
+    VPX_FREE_L(addr);
 #endif
-    }
+  }
 }
 
 #if CONFIG_MEM_TRACKER
-void *xvpx_memalign(size_t align, size_t size, char *file, int line)
-{
+void *xvpx_memalign(size_t align, size_t size, char *file, int line) {
 #if TRY_BOUNDS_CHECK
-    unsigned char *x_bounds;
+  unsigned char *x_bounds;
 #endif
 
-    void *x;
+  void *x;
 
-    if (g_alloc_count == 0)
-    {
+  if (g_alloc_count == 0) {
 #if TRY_BOUNDS_CHECK
-        int i_rv = vpx_memory_tracker_init(BOUNDS_CHECK_PAD_SIZE, BOUNDS_CHECK_VALUE);
+    int i_rv = vpx_memory_tracker_init(BOUNDS_CHECK_PAD_SIZE, BOUNDS_CHECK_VALUE);
 #else
-        int i_rv = vpx_memory_tracker_init(0, 0);
+    int i_rv = vpx_memory_tracker_init(0, 0);
 #endif
 
-        if (i_rv < 0)
-        {
-            _P(printf("ERROR xvpx_malloc MEM_TRACK_USAGE error vpx_memory_tracker_init().\n");)
-        }
+    if (i_rv < 0) {
+      _P(printf("ERROR xvpx_malloc MEM_TRACK_USAGE error vpx_memory_tracker_init().\n");)
     }
+  }
 
 #if TRY_BOUNDS_CHECK
-    {
-        int i;
-        unsigned int tempme = BOUNDS_CHECK_VALUE;
+  {
+    int i;
+    unsigned int tempme = BOUNDS_CHECK_VALUE;
 
-        x_bounds = vpx_memalign(align, size + (BOUNDS_CHECK_PAD_SIZE * 2));
+    x_bounds = vpx_memalign(align, size + (BOUNDS_CHECK_PAD_SIZE * 2));
 
-        if (x_bounds)
-        {
-            /*we're aligning the address twice here but to keep things
-              consistent we want to have the padding come before the stored
-              address so no matter what free function gets called we will
-              attempt to free the correct address*/
-            x_bounds = (unsigned char *)(((size_t *)x_bounds)[-1]);
-            x = align_addr(x_bounds + BOUNDS_CHECK_PAD_SIZE + ADDRESS_STORAGE_SIZE,
-                           (int)align);
-            /* save the actual malloc address */
-            ((size_t *)x)[-1] = (size_t)x_bounds;
+    if (x_bounds) {
+      /*we're aligning the address twice here but to keep things
+        consistent we want to have the padding come before the stored
+        address so no matter what free function gets called we will
+        attempt to free the correct address*/
+      x_bounds = (unsigned char *)(((size_t *)x_bounds)[-1]);
+      x = align_addr(x_bounds + BOUNDS_CHECK_PAD_SIZE + ADDRESS_STORAGE_SIZE,
+                     (int)align);
+      /* save the actual malloc address */
+      ((size_t *)x)[-1] = (size_t)x_bounds;
 
-            for (i = 0; i < BOUNDS_CHECK_PAD_SIZE; i += sizeof(unsigned int))
-            {
-                VPX_MEMCPY_L(x_bounds + i, &tempme, sizeof(unsigned int));
-                VPX_MEMCPY_L((unsigned char *)x + size + i,
-                             &tempme, sizeof(unsigned int));
-            }
-        }
-        else
-            x = NULL;
-    }
+      for (i = 0; i < BOUNDS_CHECK_PAD_SIZE; i += sizeof(unsigned int)) {
+        VPX_MEMCPY_L(x_bounds + i, &tempme, sizeof(unsigned int));
+        VPX_MEMCPY_L((unsigned char *)x + size + i,
+                     &tempme, sizeof(unsigned int));
+      }
+    } else
+      x = NULL;
+  }
 #else
-    x = vpx_memalign(align, size);
+  x = vpx_memalign(align, size);
 #endif /*TRY_BOUNDS_CHECK*/
 
-    g_alloc_count++;
+  g_alloc_count++;
 
-    vpx_memory_tracker_add((size_t)x, (unsigned int)size, file, line, 1);
+  vpx_memory_tracker_add((size_t)x, (unsigned int)size, file, line, 1);
 
-    return x;
+  return x;
 }
 
-void *xvpx_malloc(size_t size, char *file, int line)
-{
-    return xvpx_memalign(DEFAULT_ALIGNMENT, size, file, line);
+void *xvpx_malloc(size_t size, char *file, int line) {
+  return xvpx_memalign(DEFAULT_ALIGNMENT, size, file, line);
 }
 
-void *xvpx_calloc(size_t num, size_t size, char *file, int line)
-{
-    void *x = xvpx_memalign(DEFAULT_ALIGNMENT, num * size, file, line);
+void *xvpx_calloc(size_t num, size_t size, char *file, int line) {
+  void *x = xvpx_memalign(DEFAULT_ALIGNMENT, num * size, file, line);
 
-    if (x)
-        VPX_MEMSET_L(x, 0, num * size);
+  if (x)
+    VPX_MEMSET_L(x, 0, num * size);
 
-    return x;
+  return x;
 }
 
-void *xvpx_realloc(void *memblk, size_t size, char *file, int line)
-{
-    struct mem_block *p = NULL;
-    int orig_size = 0,
-        orig_line = 0;
-    char *orig_file = NULL;
+void *xvpx_realloc(void *memblk, size_t size, char *file, int line) {
+  struct mem_block *p = NULL;
+  int orig_size = 0,
+      orig_line = 0;
+  char *orig_file = NULL;
 
 #if TRY_BOUNDS_CHECK
-    unsigned char *x_bounds = memblk ?
-                              (unsigned char *)(((size_t *)memblk)[-1]) :
-                              NULL;
+  unsigned char *x_bounds = memblk ?
+                            (unsigned char *)(((size_t *)memblk)[-1]) :
+                            NULL;
 #endif
 
-    void *x;
+  void *x;
 
-    if (g_alloc_count == 0)
-    {
+  if (g_alloc_count == 0) {
 #if TRY_BOUNDS_CHECK
 
-        if (!vpx_memory_tracker_init(BOUNDS_CHECK_PAD_SIZE, BOUNDS_CHECK_VALUE))
+    if (!vpx_memory_tracker_init(BOUNDS_CHECK_PAD_SIZE, BOUNDS_CHECK_VALUE))
 #else
-        if (!vpx_memory_tracker_init(0, 0))
+    if (!vpx_memory_tracker_init(0, 0))
 #endif
-        {
-            _P(printf("ERROR xvpx_malloc MEM_TRACK_USAGE error vpx_memory_tracker_init().\n");)
-        }
-    }
-
-    if ((p = vpx_memory_tracker_find((size_t)memblk)))
     {
-        orig_size = p->size;
-        orig_file = p->file;
-        orig_line = p->line;
+      _P(printf("ERROR xvpx_malloc MEM_TRACK_USAGE error vpx_memory_tracker_init().\n");)
     }
+  }
 
+  if ((p = vpx_memory_tracker_find((size_t)memblk))) {
+    orig_size = p->size;
+    orig_file = p->file;
+    orig_line = p->line;
+  }
+
+#if TRY_BOUNDS_CHECK_ON_FREE
+  vpx_memory_tracker_check_integrity(file, line);
+#endif
+
+  /* have to do this regardless of success, because
+   * the memory that does get realloc'd may change
+   * the bounds values of this block
+   */
+  vpx_memory_tracker_remove((size_t)memblk);
+
+#if TRY_BOUNDS_CHECK
+  {
+    int i;
+    unsigned int tempme = BOUNDS_CHECK_VALUE;
+
+    x_bounds = vpx_realloc(memblk, size + (BOUNDS_CHECK_PAD_SIZE * 2));
+
+    if (x_bounds) {
+      x_bounds = (unsigned char *)(((size_t *)x_bounds)[-1]);
+      x = align_addr(x_bounds + BOUNDS_CHECK_PAD_SIZE + ADDRESS_STORAGE_SIZE,
+                     (int)DEFAULT_ALIGNMENT);
+      /* save the actual malloc address */
+      ((size_t *)x)[-1] = (size_t)x_bounds;
+
+      for (i = 0; i < BOUNDS_CHECK_PAD_SIZE; i += sizeof(unsigned int)) {
+        VPX_MEMCPY_L(x_bounds + i, &tempme, sizeof(unsigned int));
+        VPX_MEMCPY_L((unsigned char *)x + size + i,
+                     &tempme, sizeof(unsigned int));
+      }
+    } else
+      x = NULL;
+  }
+#else
+  x = vpx_realloc(memblk, size);
+#endif /*TRY_BOUNDS_CHECK*/
+
+  if (!memblk) ++g_alloc_count;
+
+  if (x)
+    vpx_memory_tracker_add((size_t)x, (unsigned int)size, file, line, 1);
+  else
+    vpx_memory_tracker_add((size_t)memblk, orig_size, orig_file, orig_line, 1);
+
+  return x;
+}
+
+void xvpx_free(void *p_address, char *file, int line) {
+#if TRY_BOUNDS_CHECK
+  unsigned char *p_bounds_address = (unsigned char *)p_address;
+  /*p_bounds_address -= BOUNDS_CHECK_PAD_SIZE;*/
+#endif
+
+#if !TRY_BOUNDS_CHECK_ON_FREE
+  (void)file;
+  (void)line;
+#endif
+
+  if (p_address) {
 #if TRY_BOUNDS_CHECK_ON_FREE
     vpx_memory_tracker_check_integrity(file, line);
 #endif
 
-    /* have to do this regardless of success, because
-     * the memory that does get realloc'd may change
-     * the bounds values of this block
+    /* if the addr isn't found in the list, assume it was allocated via
+     * vpx_ calls not xvpx_, therefore it does not contain any padding
      */
-    vpx_memory_tracker_remove((size_t)memblk);
+    if (vpx_memory_tracker_remove((size_t)p_address) == -2) {
+      p_bounds_address = p_address;
+      _P(fprintf(stderr, "[vpx_mem][xvpx_free] addr: %p not found in"
+                 " list; freed from file:%s"
+                 " line:%d\n", p_address, file, line));
+    } else
+      --g_alloc_count;
 
 #if TRY_BOUNDS_CHECK
-    {
-        int i;
-        unsigned int tempme = BOUNDS_CHECK_VALUE;
-
-        x_bounds = vpx_realloc(memblk, size + (BOUNDS_CHECK_PAD_SIZE * 2));
-
-        if (x_bounds)
-        {
-            x_bounds = (unsigned char *)(((size_t *)x_bounds)[-1]);
-            x = align_addr(x_bounds + BOUNDS_CHECK_PAD_SIZE + ADDRESS_STORAGE_SIZE,
-                           (int)DEFAULT_ALIGNMENT);
-            /* save the actual malloc address */
-            ((size_t *)x)[-1] = (size_t)x_bounds;
-
-            for (i = 0; i < BOUNDS_CHECK_PAD_SIZE; i += sizeof(unsigned int))
-            {
-                VPX_MEMCPY_L(x_bounds + i, &tempme, sizeof(unsigned int));
-                VPX_MEMCPY_L((unsigned char *)x + size + i,
-                             &tempme, sizeof(unsigned int));
-            }
-        }
-        else
-            x = NULL;
-    }
+    vpx_free(p_bounds_address);
 #else
-    x = vpx_realloc(memblk, size);
-#endif /*TRY_BOUNDS_CHECK*/
-
-    if (!memblk) ++g_alloc_count;
-
-    if (x)
-        vpx_memory_tracker_add((size_t)x, (unsigned int)size, file, line, 1);
-    else
-        vpx_memory_tracker_add((size_t)memblk, orig_size, orig_file, orig_line, 1);
-
-    return x;
-}
-
-void xvpx_free(void *p_address, char *file, int line)
-{
-#if TRY_BOUNDS_CHECK
-    unsigned char *p_bounds_address = (unsigned char *)p_address;
-    /*p_bounds_address -= BOUNDS_CHECK_PAD_SIZE;*/
+    vpx_free(p_address);
 #endif
 
-#if !TRY_BOUNDS_CHECK_ON_FREE
-    (void)file;
-    (void)line;
-#endif
-
-    if (p_address)
-    {
-#if TRY_BOUNDS_CHECK_ON_FREE
-        vpx_memory_tracker_check_integrity(file, line);
-#endif
-
-        /* if the addr isn't found in the list, assume it was allocated via
-         * vpx_ calls not xvpx_, therefore it does not contain any padding
-         */
-        if (vpx_memory_tracker_remove((size_t)p_address) == -2)
-        {
-            p_bounds_address = p_address;
-            _P(fprintf(stderr, "[vpx_mem][xvpx_free] addr: %p not found in"
-                       " list; freed from file:%s"
-                       " line:%d\n", p_address, file, line));
-        }
-        else
-            --g_alloc_count;
-
-#if TRY_BOUNDS_CHECK
-        vpx_free(p_bounds_address);
-#else
-        vpx_free(p_address);
-#endif
-
-        if (!g_alloc_count)
-            vpx_memory_tracker_destroy();
-    }
+    if (!g_alloc_count)
+      vpx_memory_tracker_destroy();
+  }
 }
 
 #endif /*CONFIG_MEM_TRACKER*/
@@ -426,297 +393,265 @@
 #include <task_lib.h> /*for task_delay()*/
 /* This function is only used to get a stack trace of the player
 object so we can se where we are having a problem. */
-static int get_my_tt(int task)
-{
-    tt(task);
+static int get_my_tt(int task) {
+  tt(task);
 
-    return 0;
+  return 0;
 }
 
-static void vx_sleep(int msec)
-{
-    int ticks_to_sleep = 0;
+static void vx_sleep(int msec) {
+  int ticks_to_sleep = 0;
 
-    if (msec)
-    {
-        int msec_per_tick = 1000 / sys_clk_rate_get();
+  if (msec) {
+    int msec_per_tick = 1000 / sys_clk_rate_get();
 
-        if (msec < msec_per_tick)
-            ticks_to_sleep++;
-        else
-            ticks_to_sleep = msec / msec_per_tick;
-    }
+    if (msec < msec_per_tick)
+      ticks_to_sleep++;
+    else
+      ticks_to_sleep = msec / msec_per_tick;
+  }
 
-    task_delay(ticks_to_sleep);
+  task_delay(ticks_to_sleep);
 }
 #endif
 #endif
 
-void *vpx_memcpy(void *dest, const void *source, size_t length)
-{
+void *vpx_memcpy(void *dest, const void *source, size_t length) {
 #if CONFIG_MEM_CHECKS
 
-    if (((int)dest < 0x4000) || ((int)source < 0x4000))
-    {
-        _P(printf("WARNING: vpx_memcpy dest:0x%x source:0x%x len:%d\n", (int)dest, (int)source, length);)
+  if (((int)dest < 0x4000) || ((int)source < 0x4000)) {
+    _P(printf("WARNING: vpx_memcpy dest:0x%x source:0x%x len:%d\n", (int)dest, (int)source, length);)
 
 #if defined(VXWORKS)
-        sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
+    sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
 
-        vx_sleep(10000);
+    vx_sleep(10000);
 #endif
-    }
+  }
 
 #endif
 
-    return VPX_MEMCPY_L(dest, source, length);
+  return VPX_MEMCPY_L(dest, source, length);
 }
 
-void *vpx_memset(void *dest, int val, size_t length)
-{
+void *vpx_memset(void *dest, int val, size_t length) {
 #if CONFIG_MEM_CHECKS
 
-    if ((int)dest < 0x4000)
-    {
-        _P(printf("WARNING: vpx_memset dest:0x%x val:%d len:%d\n", (int)dest, val, length);)
+  if ((int)dest < 0x4000) {
+    _P(printf("WARNING: vpx_memset dest:0x%x val:%d len:%d\n", (int)dest, val, length);)
 
 #if defined(VXWORKS)
-        sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
+    sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
 
-        vx_sleep(10000);
+    vx_sleep(10000);
 #endif
-    }
+  }
 
 #endif
 
-    return VPX_MEMSET_L(dest, val, length);
+  return VPX_MEMSET_L(dest, val, length);
 }
 
-void *vpx_memmove(void *dest, const void *src, size_t count)
-{
+void *vpx_memmove(void *dest, const void *src, size_t count) {
 #if CONFIG_MEM_CHECKS
 
-    if (((int)dest < 0x4000) || ((int)src < 0x4000))
-    {
-        _P(printf("WARNING: vpx_memmove dest:0x%x src:0x%x count:%d\n", (int)dest, (int)src, count);)
+  if (((int)dest < 0x4000) || ((int)src < 0x4000)) {
+    _P(printf("WARNING: vpx_memmove dest:0x%x src:0x%x count:%d\n", (int)dest, (int)src, count);)
 
 #if defined(VXWORKS)
-        sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
+    sp(get_my_tt, task_id_self(), 0, 0, 0, 0, 0, 0, 0, 0);
 
-        vx_sleep(10000);
+    vx_sleep(10000);
 #endif
-    }
+  }
 
 #endif
 
-    return VPX_MEMMOVE_L(dest, src, count);
+  return VPX_MEMMOVE_L(dest, src, count);
 }
 
 #if CONFIG_MEM_MANAGER
 
-static int vpx_mm_create_heap_memory()
-{
-    int i_rv = 0;
+static int vpx_mm_create_heap_memory() {
+  int i_rv = 0;
 
-    if (!g_mng_memory_allocated)
-    {
+  if (!g_mng_memory_allocated) {
 #if MM_DYNAMIC_MEMORY
-        g_p_mng_memory_raw =
-            (unsigned char *)malloc(g_mm_memory_size + HMM_ADDR_ALIGN_UNIT);
+    g_p_mng_memory_raw =
+      (unsigned char *)malloc(g_mm_memory_size + HMM_ADDR_ALIGN_UNIT);
 
-        if (g_p_mng_memory_raw)
-        {
-            g_p_mng_memory = (unsigned char *)((((unsigned int)g_p_mng_memory_raw) +
-                                                HMM_ADDR_ALIGN_UNIT - 1) &
-                                               -(int)HMM_ADDR_ALIGN_UNIT);
+    if (g_p_mng_memory_raw) {
+      g_p_mng_memory = (unsigned char *)((((unsigned int)g_p_mng_memory_raw) +
+                                          HMM_ADDR_ALIGN_UNIT - 1) &
+                                         -(int)HMM_ADDR_ALIGN_UNIT);
 
-            _P(printf("[vpx][mm] total memory size:%d g_p_mng_memory_raw:0x%x g_p_mng_memory:0x%x\n"
-                      , g_mm_memory_size + HMM_ADDR_ALIGN_UNIT
-                      , (unsigned int)g_p_mng_memory_raw
-                      , (unsigned int)g_p_mng_memory);)
-        }
-        else
-        {
-            _P(printf("[vpx][mm] Couldn't allocate memory:%d for vpx memory manager.\n"
-                      , g_mm_memory_size);)
+      _P(printf("[vpx][mm] total memory size:%d g_p_mng_memory_raw:0x%x g_p_mng_memory:0x%x\n"
+, g_mm_memory_size + HMM_ADDR_ALIGN_UNIT
+, (unsigned int)g_p_mng_memory_raw
+, (unsigned int)g_p_mng_memory);)
+    } else {
+      _P(printf("[vpx][mm] Couldn't allocate memory:%d for vpx memory manager.\n"
+, g_mm_memory_size);)
 
-            i_rv = -1;
-        }
-
-        if (g_p_mng_memory)
-#endif
-        {
-            int chunk_size = 0;
-
-            g_mng_memory_allocated = 1;
-
-            hmm_init(&hmm_d);
-
-            chunk_size = g_mm_memory_size >> SHIFT_HMM_ADDR_ALIGN_UNIT;
-
-            chunk_size -= DUMMY_END_BLOCK_BAUS;
-
-            _P(printf("[vpx][mm] memory size:%d for vpx memory manager. g_p_mng_memory:0x%x  chunk_size:%d\n"
-                      , g_mm_memory_size
-                      , (unsigned int)g_p_mng_memory
-                      , chunk_size);)
-
-            hmm_new_chunk(&hmm_d, (void *)g_p_mng_memory, chunk_size);
-        }
-
-#if MM_DYNAMIC_MEMORY
-        else
-        {
-            _P(printf("[vpx][mm] Couldn't allocate memory:%d for vpx memory manager.\n"
-                      , g_mm_memory_size);)
-
-            i_rv = -1;
-        }
-
-#endif
+      i_rv = -1;
     }
 
-    return i_rv;
+    if (g_p_mng_memory)
+#endif
+    {
+      int chunk_size = 0;
+
+      g_mng_memory_allocated = 1;
+
+      hmm_init(&hmm_d);
+
+      chunk_size = g_mm_memory_size >> SHIFT_HMM_ADDR_ALIGN_UNIT;
+
+      chunk_size -= DUMMY_END_BLOCK_BAUS;
+
+      _P(printf("[vpx][mm] memory size:%d for vpx memory manager. g_p_mng_memory:0x%x  chunk_size:%d\n"
+, g_mm_memory_size
+, (unsigned int)g_p_mng_memory
+, chunk_size);)
+
+      hmm_new_chunk(&hmm_d, (void *)g_p_mng_memory, chunk_size);
+    }
+
+#if MM_DYNAMIC_MEMORY
+    else {
+      _P(printf("[vpx][mm] Couldn't allocate memory:%d for vpx memory manager.\n"
+, g_mm_memory_size);)
+
+      i_rv = -1;
+    }
+
+#endif
+  }
+
+  return i_rv;
 }
 
-static void *vpx_mm_realloc(void *memblk, size_t size)
-{
-    void *p_ret = NULL;
+static void *vpx_mm_realloc(void *memblk, size_t size) {
+  void *p_ret = NULL;
 
-    if (vpx_mm_create_heap_memory() < 0)
-    {
-        _P(printf("[vpx][mm] ERROR vpx_mm_realloc() Couldn't create memory for Heap.\n");)
-    }
-    else
-    {
-        int i_rv = 0;
-        int old_num_aaus;
-        int new_num_aaus;
+  if (vpx_mm_create_heap_memory() < 0) {
+    _P(printf("[vpx][mm] ERROR vpx_mm_realloc() Couldn't create memory for Heap.\n");)
+  } else {
+    int i_rv = 0;
+    int old_num_aaus;
+    int new_num_aaus;
 
-        old_num_aaus = hmm_true_size(memblk);
+    old_num_aaus = hmm_true_size(memblk);
+    new_num_aaus = (size >> SHIFT_HMM_ADDR_ALIGN_UNIT) + 1;
+
+    if (old_num_aaus == new_num_aaus) {
+      p_ret = memblk;
+    } else {
+      i_rv = hmm_resize(&hmm_d, memblk, new_num_aaus);
+
+      if (i_rv == 0) {
+        p_ret = memblk;
+      } else {
+        /* Error. Try to malloc and then copy data. */
+        void *p_from_malloc;
+
         new_num_aaus = (size >> SHIFT_HMM_ADDR_ALIGN_UNIT) + 1;
+        p_from_malloc  = hmm_alloc(&hmm_d, new_num_aaus);
 
-        if (old_num_aaus == new_num_aaus)
-        {
-            p_ret = memblk;
+        if (p_from_malloc) {
+          vpx_memcpy(p_from_malloc, memblk, size);
+          hmm_free(&hmm_d, memblk);
+
+          p_ret = p_from_malloc;
         }
-        else
-        {
-            i_rv = hmm_resize(&hmm_d, memblk, new_num_aaus);
-
-            if (i_rv == 0)
-            {
-                p_ret = memblk;
-            }
-            else
-            {
-                /* Error. Try to malloc and then copy data. */
-                void *p_from_malloc;
-
-                new_num_aaus = (size >> SHIFT_HMM_ADDR_ALIGN_UNIT) + 1;
-                p_from_malloc  = hmm_alloc(&hmm_d, new_num_aaus);
-
-                if (p_from_malloc)
-                {
-                    vpx_memcpy(p_from_malloc, memblk, size);
-                    hmm_free(&hmm_d, memblk);
-
-                    p_ret = p_from_malloc;
-                }
-            }
-        }
+      }
     }
+  }
 
-    return p_ret;
+  return p_ret;
 }
 #endif /*CONFIG_MEM_MANAGER*/
 
 #if USE_GLOBAL_FUNCTION_POINTERS
 # if CONFIG_MEM_TRACKER
 extern int vpx_memory_tracker_set_functions(g_malloc_func g_malloc_l
-        , g_calloc_func g_calloc_l
-        , g_realloc_func g_realloc_l
-        , g_free_func g_free_l
-        , g_memcpy_func g_memcpy_l
-        , g_memset_func g_memset_l
-        , g_memmove_func g_memmove_l);
+, g_calloc_func g_calloc_l
+, g_realloc_func g_realloc_l
+, g_free_func g_free_l
+, g_memcpy_func g_memcpy_l
+, g_memset_func g_memset_l
+, g_memmove_func g_memmove_l);
 # endif
 #endif /*USE_GLOBAL_FUNCTION_POINTERS*/
 int vpx_mem_set_functions(g_malloc_func g_malloc_l
-                          , g_calloc_func g_calloc_l
-                          , g_realloc_func g_realloc_l
-                          , g_free_func g_free_l
-                          , g_memcpy_func g_memcpy_l
-                          , g_memset_func g_memset_l
-                          , g_memmove_func g_memmove_l)
-{
+, g_calloc_func g_calloc_l
+, g_realloc_func g_realloc_l
+, g_free_func g_free_l
+, g_memcpy_func g_memcpy_l
+, g_memset_func g_memset_l
+, g_memmove_func g_memmove_l) {
 #if USE_GLOBAL_FUNCTION_POINTERS
 
-    /* If use global functions is turned on then the
-    application must set the global functions before
-    it does anything else or vpx_mem will have
-    unpredictable results. */
-    if (!g_func)
-    {
-        g_func = (struct GLOBAL_FUNC_POINTERS *)
-                 g_malloc_l(sizeof(struct GLOBAL_FUNC_POINTERS));
+  /* If use global functions is turned on then the
+  application must set the global functions before
+  it does anything else or vpx_mem will have
+  unpredictable results. */
+  if (!g_func) {
+    g_func = (struct GLOBAL_FUNC_POINTERS *)
+             g_malloc_l(sizeof(struct GLOBAL_FUNC_POINTERS));
 
-        if (!g_func)
-        {
-            return -1;
-        }
+    if (!g_func) {
+      return -1;
     }
+  }
 
 #if CONFIG_MEM_TRACKER
-    {
-        int rv = 0;
-        rv = vpx_memory_tracker_set_functions(g_malloc_l
-                                              , g_calloc_l
-                                              , g_realloc_l
-                                              , g_free_l
-                                              , g_memcpy_l
-                                              , g_memset_l
-                                              , g_memmove_l);
+  {
+    int rv = 0;
+    rv = vpx_memory_tracker_set_functions(g_malloc_l
+, g_calloc_l
+, g_realloc_l
+, g_free_l
+, g_memcpy_l
+, g_memset_l
+, g_memmove_l);
 
-        if (rv < 0)
-        {
-            return rv;
-        }
+    if (rv < 0) {
+      return rv;
     }
+  }
 #endif
 
-    g_func->g_malloc  = g_malloc_l;
-    g_func->g_calloc  = g_calloc_l;
-    g_func->g_realloc = g_realloc_l;
-    g_func->g_free    = g_free_l;
-    g_func->g_memcpy  = g_memcpy_l;
-    g_func->g_memset  = g_memset_l;
-    g_func->g_memmove = g_memmove_l;
+  g_func->g_malloc  = g_malloc_l;
+  g_func->g_calloc  = g_calloc_l;
+  g_func->g_realloc = g_realloc_l;
+  g_func->g_free    = g_free_l;
+  g_func->g_memcpy  = g_memcpy_l;
+  g_func->g_memset  = g_memset_l;
+  g_func->g_memmove = g_memmove_l;
 
-    return 0;
+  return 0;
 #else
-    (void)g_malloc_l;
-    (void)g_calloc_l;
-    (void)g_realloc_l;
-    (void)g_free_l;
-    (void)g_memcpy_l;
-    (void)g_memset_l;
-    (void)g_memmove_l;
-    return -1;
+  (void)g_malloc_l;
+  (void)g_calloc_l;
+  (void)g_realloc_l;
+  (void)g_free_l;
+  (void)g_memcpy_l;
+  (void)g_memset_l;
+  (void)g_memmove_l;
+  return -1;
 #endif
 }
 
-int vpx_mem_unset_functions()
-{
+int vpx_mem_unset_functions() {
 #if USE_GLOBAL_FUNCTION_POINTERS
 
-    if (g_func)
-    {
-        g_free_func temp_free = g_func->g_free;
-        temp_free(g_func);
-        g_func = NULL;
-    }
+  if (g_func) {
+    g_free_func temp_free = g_func->g_free;
+    temp_free(g_func);
+    g_func = NULL;
+  }
 
 #endif
-    return 0;
+  return 0;
 }
diff --git a/libvpx/vpx_mem/vpx_mem.h b/libvpx/vpx_mem/vpx_mem.h
index 749eaa4..c7321a9 100644
--- a/libvpx/vpx_mem/vpx_mem.h
+++ b/libvpx/vpx_mem/vpx_mem.h
@@ -12,6 +12,7 @@
 #ifndef __VPX_MEM_H__
 #define __VPX_MEM_H__
 
+#include "vpx_config.h"
 #if defined(__uClinux__)
 # include <lddk.h>
 #endif
@@ -30,11 +31,11 @@
 #endif
 #ifndef VPX_CHECK_MEM_FUNCTIONS
 # define VPX_CHECK_MEM_FUNCTIONS   0  /* enable basic safety checks in _memcpy,
-                                         _memset, and _memmove */
+_memset, and _memmove */
 #endif
 #ifndef REPLACE_BUILTIN_FUNCTIONS
 # define REPLACE_BUILTIN_FUNCTIONS 0  /* replace builtin functions with their
-                                         vpx_ equivalents */
+vpx_ equivalents */
 #endif
 
 #include <stdlib.h>
@@ -44,70 +45,63 @@
 extern "C" {
 #endif
 
-    /*
-        vpx_mem_get_version()
-        provided for runtime version checking. Returns an unsigned int of the form
-        CHIEF | MAJOR | MINOR | PATCH, where the chief version number is the high
-        order byte.
-    */
-    unsigned int vpx_mem_get_version(void);
+  /*
+      vpx_mem_get_version()
+      provided for runtime version checking. Returns an unsigned int of the form
+      CHIEF | MAJOR | MINOR | PATCH, where the chief version number is the high
+      order byte.
+  */
+  unsigned int vpx_mem_get_version(void);
 
-    /*
-        vpx_mem_set_heap_size(size_t size)
-          size - size in bytes for the memory manager to allocate for its heap
-        Sets the memory manager's initial heap size
-        Return:
-          0: on success
-          -1: if memory manager calls have not been included in the vpx_mem lib
-          -2: if the memory manager has been compiled to use static memory
-          -3: if the memory manager has already allocated its heap
-    */
-    int vpx_mem_set_heap_size(size_t size);
+  /*
+      vpx_mem_set_heap_size(size_t size)
+        size - size in bytes for the memory manager to allocate for its heap
+      Sets the memory manager's initial heap size
+      Return:
+        0: on success
+        -1: if memory manager calls have not been included in the vpx_mem lib
+        -2: if the memory manager has been compiled to use static memory
+        -3: if the memory manager has already allocated its heap
+  */
+  int vpx_mem_set_heap_size(size_t size);
 
-    void *vpx_memalign(size_t align, size_t size);
-    void *vpx_malloc(size_t size);
-    void *vpx_calloc(size_t num, size_t size);
-    void *vpx_realloc(void *memblk, size_t size);
-    void vpx_free(void *memblk);
+  void *vpx_memalign(size_t align, size_t size);
+  void *vpx_malloc(size_t size);
+  void *vpx_calloc(size_t num, size_t size);
+  void *vpx_realloc(void *memblk, size_t size);
+  void vpx_free(void *memblk);
 
-    void *vpx_memcpy(void *dest, const void *src, size_t length);
-    void *vpx_memset(void *dest, int val, size_t length);
-    void *vpx_memmove(void *dest, const void *src, size_t count);
+  void *vpx_memcpy(void *dest, const void *src, size_t length);
+  void *vpx_memset(void *dest, int val, size_t length);
+  void *vpx_memmove(void *dest, const void *src, size_t count);
 
-    /* special memory functions */
-    void *vpx_mem_alloc(int id, size_t size, size_t align);
-    void vpx_mem_free(int id, void *mem, size_t size);
+  /* special memory functions */
+  void *vpx_mem_alloc(int id, size_t size, size_t align);
+  void vpx_mem_free(int id, void *mem, size_t size);
 
-    /* Wrappers to standard library functions. */
-    typedef void*(* g_malloc_func)(size_t);
-    typedef void*(* g_calloc_func)(size_t, size_t);
-    typedef void*(* g_realloc_func)(void *, size_t);
-    typedef void (* g_free_func)(void *);
-    typedef void*(* g_memcpy_func)(void *, const void *, size_t);
-    typedef void*(* g_memset_func)(void *, int, size_t);
-    typedef void*(* g_memmove_func)(void *, const void *, size_t);
+  /* Wrappers to standard library functions. */
+  typedef void *(* g_malloc_func)(size_t);
+  typedef void *(* g_calloc_func)(size_t, size_t);
+  typedef void *(* g_realloc_func)(void *, size_t);
+  typedef void (* g_free_func)(void *);
+  typedef void *(* g_memcpy_func)(void *, const void *, size_t);
+  typedef void *(* g_memset_func)(void *, int, size_t);
+  typedef void *(* g_memmove_func)(void *, const void *, size_t);
 
-    int vpx_mem_set_functions(g_malloc_func g_malloc_l
-                              , g_calloc_func g_calloc_l
-                              , g_realloc_func g_realloc_l
-                              , g_free_func g_free_l
-                              , g_memcpy_func g_memcpy_l
-                              , g_memset_func g_memset_l
-                              , g_memmove_func g_memmove_l);
-    int vpx_mem_unset_functions(void);
+  int vpx_mem_set_functions(g_malloc_func g_malloc_l
+, g_calloc_func g_calloc_l
+, g_realloc_func g_realloc_l
+, g_free_func g_free_l
+, g_memcpy_func g_memcpy_l
+, g_memset_func g_memset_l
+, g_memmove_func g_memmove_l);
+  int vpx_mem_unset_functions(void);
 
 
-    /* some defines for backward compatibility */
+  /* some defines for backward compatibility */
 #define DMEM_GENERAL 0
 
-#define duck_memalign(X,Y,Z) vpx_memalign(X,Y)
-#define duck_malloc(X,Y) vpx_malloc(X)
-#define duck_calloc(X,Y,Z) vpx_calloc(X,Y)
-#define duck_realloc  vpx_realloc
-#define duck_free     vpx_free
-#define duck_memcpy   vpx_memcpy
-#define duck_memmove  vpx_memmove
-#define duck_memset   vpx_memset
+// (*)<
 
 #if REPLACE_BUILTIN_FUNCTIONS
 # ifndef __VPX_MEM_C__
@@ -124,13 +118,13 @@
 
 #if CONFIG_MEM_TRACKER
 #include <stdarg.h>
-    /*from vpx_mem/vpx_mem_tracker.c*/
-    extern void vpx_memory_tracker_dump();
-    extern void vpx_memory_tracker_check_integrity(char *file, unsigned int line);
-    extern int vpx_memory_tracker_set_log_type(int type, char *option);
-    extern int vpx_memory_tracker_set_log_func(void *userdata,
-            void(*logfunc)(void *userdata,
-                           const char *fmt, va_list args));
+  /*from vpx_mem/vpx_mem_tracker.c*/
+  extern void vpx_memory_tracker_dump();
+  extern void vpx_memory_tracker_check_integrity(char *file, unsigned int line);
+  extern int vpx_memory_tracker_set_log_type(int type, char *option);
+  extern int vpx_memory_tracker_set_log_func(void *userdata,
+                                             void(*logfunc)(void *userdata,
+                                                            const char *fmt, va_list args));
 # ifndef __VPX_MEM_C__
 #  define vpx_memalign(align, size) xvpx_memalign((align), (size), __FILE__, __LINE__)
 #  define vpx_malloc(size)          xvpx_malloc((size), __FILE__, __LINE__)
@@ -142,13 +136,13 @@
 #  define vpx_mem_free(id,mem,size) xvpx_mem_free(id, mem, size, __FILE__, __LINE__)
 # endif
 
-    void *xvpx_memalign(size_t align, size_t size, char *file, int line);
-    void *xvpx_malloc(size_t size, char *file, int line);
-    void *xvpx_calloc(size_t num, size_t size, char *file, int line);
-    void *xvpx_realloc(void *memblk, size_t size, char *file, int line);
-    void xvpx_free(void *memblk, char *file, int line);
-    void *xvpx_mem_alloc(int id, size_t size, size_t align, char *file, int line);
-    void xvpx_mem_free(int id, void *mem, size_t size, char *file, int line);
+  void *xvpx_memalign(size_t align, size_t size, char *file, int line);
+  void *xvpx_malloc(size_t size, char *file, int line);
+  void *xvpx_calloc(size_t num, size_t size, char *file, int line);
+  void *xvpx_realloc(void *memblk, size_t size, char *file, int line);
+  void xvpx_free(void *memblk, char *file, int line);
+  void *xvpx_mem_alloc(int id, size_t size, size_t align, char *file, int line);
+  void xvpx_mem_free(int id, void *mem, size_t size, char *file, int line);
 
 #else
 # ifndef __VPX_MEM_C__
diff --git a/libvpx/vpx_mem/vpx_mem_tracker.c b/libvpx/vpx_mem/vpx_mem_tracker.c
index b37076e..613e8a1 100644
--- a/libvpx/vpx_mem/vpx_mem_tracker.c
+++ b/libvpx/vpx_mem/vpx_mem_tracker.c
@@ -22,7 +22,7 @@
    in the memory_tracker struct as well as calls to create/destroy/lock/unlock
    the mutex in vpx_memory_tracker_init/Destroy and memory_tracker_lock_mutex/unlock_mutex
 */
-#include "vpx_config.h"
+#include "./vpx_config.h"
 
 #if defined(__uClinux__)
 # include <lddk.h>
@@ -40,20 +40,20 @@
 
 #include <stdio.h>
 #include <stdlib.h>
-#include <string.h> //VXWORKS doesn't have a malloc/memory.h file,
-//this should pull in malloc,free,etc.
+#include <string.h> // VXWORKS doesn't have a malloc/memory.h file,
+// this should pull in malloc,free,etc.
 #include <stdarg.h>
 
 #include "include/vpx_mem_tracker.h"
 
-#undef vpx_malloc   //undefine any vpx_mem macros that may affect calls to
-#undef vpx_free     //memory functions in this file
+#undef vpx_malloc   // undefine any vpx_mem macros that may affect calls to
+#undef vpx_free     // memory functions in this file
 #undef vpx_memcpy
 #undef vpx_memset
 
 
 #ifndef USE_GLOBAL_FUNCTION_POINTERS
-# define USE_GLOBAL_FUNCTION_POINTERS   0  //use function pointers instead of compiled functions.
+# define USE_GLOBAL_FUNCTION_POINTERS   0  // use function pointers instead of compiled functions.
 #endif
 
 #if USE_GLOBAL_FUNCTION_POINTERS
@@ -94,39 +94,37 @@
 #endif
 
 #ifndef VPX_NO_GLOBALS
-struct memory_tracker
-{
-    struct mem_block *head,
-            * tail;
-    int len,
-        totalsize;
-    unsigned int current_allocated,
-             max_allocated;
+struct memory_tracker {
+  struct mem_block *head,
+      * tail;
+  int len,
+      totalsize;
+  unsigned int current_allocated,
+           max_allocated;
 
 #if HAVE_PTHREAD_H
-    pthread_mutex_t mutex;
+  pthread_mutex_t mutex;
 #elif defined(WIN32) || defined(_WIN32_WCE)
-    HANDLE mutex;
+  HANDLE mutex;
 #elif defined(VXWORKS)
-    SEM_ID mutex;
+  SEM_ID mutex;
 #elif defined(NO_MUTEX)
 #else
 #error "No mutex type defined for this platform!"
 #endif
 
-    int padding_size,
-        pad_value;
+  int padding_size,
+      pad_value;
 };
 
-static struct memory_tracker memtrack;   //our global memory allocation list
-static int g_b_mem_tracker_inited = 0;     //indicates whether the global list has
-//been initialized (1:yes/0:no)
-static struct
-{
-    FILE *file;
-    int type;
-    void (*func)(void *userdata, const char *fmt, va_list args);
-    void *userdata;
+static struct memory_tracker memtrack;   // our global memory allocation list
+static int g_b_mem_tracker_inited = 0;     // indicates whether the global list has
+// been initialized (1:yes/0:no)
+static struct {
+  FILE *file;
+  int type;
+  void (*func)(void *userdata, const char *fmt, va_list args);
+  void *userdata;
 } g_logging = {NULL, 0, NULL, NULL};
 #else
 # include "vpx_global_handling.h"
@@ -157,60 +155,54 @@
     Initializes global memory tracker structure
     Allocates the head of the list
 */
-int vpx_memory_tracker_init(int padding_size, int pad_value)
-{
-    if (!g_b_mem_tracker_inited)
-    {
-        if ((memtrack.head = (struct mem_block *)
-                             MEM_TRACK_MALLOC(sizeof(struct mem_block))))
-        {
-            int ret;
+int vpx_memory_tracker_init(int padding_size, int pad_value) {
+  if (!g_b_mem_tracker_inited) {
+    if ((memtrack.head = (struct mem_block *)
+                         MEM_TRACK_MALLOC(sizeof(struct mem_block)))) {
+      int ret;
 
-            MEM_TRACK_MEMSET(memtrack.head, 0, sizeof(struct mem_block));
+      MEM_TRACK_MEMSET(memtrack.head, 0, sizeof(struct mem_block));
 
-            memtrack.tail = memtrack.head;
+      memtrack.tail = memtrack.head;
 
-            memtrack.current_allocated = 0;
-            memtrack.max_allocated     = 0;
+      memtrack.current_allocated = 0;
+      memtrack.max_allocated     = 0;
 
-            memtrack.padding_size = padding_size;
-            memtrack.pad_value    = pad_value;
+      memtrack.padding_size = padding_size;
+      memtrack.pad_value    = pad_value;
 
 #if HAVE_PTHREAD_H
-            ret = pthread_mutex_init(&memtrack.mutex,
-                                     NULL);            /*mutex attributes (NULL=default)*/
+      ret = pthread_mutex_init(&memtrack.mutex,
+                               NULL);            /*mutex attributes (NULL=default)*/
 #elif defined(WIN32) || defined(_WIN32_WCE)
-            memtrack.mutex = CreateMutex(NULL,   /*security attributes*/
-                                          FALSE,  /*we don't want initial ownership*/
-                                          NULL);  /*mutex name*/
-            ret = !memtrack.mutex;
+      memtrack.mutex = CreateMutex(NULL,   /*security attributes*/
+                                   FALSE,  /*we don't want initial ownership*/
+                                   NULL);  /*mutex name*/
+      ret = !memtrack.mutex;
 #elif defined(VXWORKS)
-            memtrack.mutex = sem_bcreate(SEM_Q_FIFO, /*SEM_Q_FIFO non-priority based mutex*/
-                                         SEM_FULL);  /*SEM_FULL initial state is unlocked*/
-            ret = !memtrack.mutex;
+      memtrack.mutex = sem_bcreate(SEM_Q_FIFO, /*SEM_Q_FIFO non-priority based mutex*/
+                                   SEM_FULL);  /*SEM_FULL initial state is unlocked*/
+      ret = !memtrack.mutex;
 #elif defined(NO_MUTEX)
-            ret = 0;
+      ret = 0;
 #endif
 
-            if (ret)
-            {
-                memtrack_log("vpx_memory_tracker_init: Error creating mutex!\n");
+      if (ret) {
+        memtrack_log("vpx_memory_tracker_init: Error creating mutex!\n");
 
-                MEM_TRACK_FREE(memtrack.head);
-                memtrack.head = NULL;
-            }
-            else
-            {
-                memtrack_log("Memory Tracker init'd, v."vpx_mem_tracker_version" pad_size:%d pad_val:0x%x %d\n"
-                             , padding_size
-                             , pad_value
-                             , pad_value);
-                g_b_mem_tracker_inited = 1;
-            }
-        }
+        MEM_TRACK_FREE(memtrack.head);
+        memtrack.head = NULL;
+      } else {
+        memtrack_log("Memory Tracker init'd, v."vpx_mem_tracker_version" pad_size:%d pad_val:0x%x %d\n"
+, padding_size
+, pad_value
+, pad_value);
+        g_b_mem_tracker_inited = 1;
+      }
     }
+  }
 
-    return g_b_mem_tracker_inited;
+  return g_b_mem_tracker_inited;
 }
 
 /*
@@ -218,39 +210,35 @@
     If our global struct was initialized zeros out all its members,
     frees memory and destroys it's mutex
 */
-void vpx_memory_tracker_destroy()
-{
-    if (!memory_tracker_lock_mutex())
-    {
-        struct mem_block *p  = memtrack.head,
-                                  * p2 = memtrack.head;
+void vpx_memory_tracker_destroy() {
+  if (!memory_tracker_lock_mutex()) {
+    struct mem_block *p  = memtrack.head,
+                          * p2 = memtrack.head;
 
-        memory_tracker_dump();
+    memory_tracker_dump();
 
-        while (p)
-    {
-            p2 = p;
-            p  = p->next;
+    while (p) {
+      p2 = p;
+      p  = p->next;
 
-            MEM_TRACK_FREE(p2);
-        }
-
-        memtrack.head              = NULL;
-        memtrack.tail              = NULL;
-        memtrack.len               = 0;
-        memtrack.current_allocated = 0;
-        memtrack.max_allocated     = 0;
-
-        if (!g_logging.type && g_logging.file && g_logging.file != stderr)
-        {
-            fclose(g_logging.file);
-            g_logging.file = NULL;
-        }
-
-        memory_tracker_unlock_mutex();
-
-        g_b_mem_tracker_inited = 0;
+      MEM_TRACK_FREE(p2);
     }
+
+    memtrack.head              = NULL;
+    memtrack.tail              = NULL;
+    memtrack.len               = 0;
+    memtrack.current_allocated = 0;
+    memtrack.max_allocated     = 0;
+
+    if (!g_logging.type && g_logging.file && g_logging.file != stderr) {
+      fclose(g_logging.file);
+      g_logging.file = NULL;
+    }
+
+    memory_tracker_unlock_mutex();
+
+    g_b_mem_tracker_inited = 0;
+  }
 }
 
 /*
@@ -265,9 +253,8 @@
 */
 void vpx_memory_tracker_add(size_t addr, unsigned int size,
                             char *file, unsigned int line,
-                            int padded)
-{
-    memory_tracker_add(addr, size, file, line, padded);
+                            int padded) {
+  memory_tracker_add(addr, size, file, line, padded);
 }
 
 /*
@@ -278,9 +265,8 @@
     Return:
       Same as described for memory_tracker_remove
 */
-int vpx_memory_tracker_remove(size_t addr)
-{
-    return memory_tracker_remove(addr);
+int vpx_memory_tracker_remove(size_t addr) {
+  return memory_tracker_remove(addr);
 }
 
 /*
@@ -290,17 +276,15 @@
         If found, pointer to the memory block that matches addr
         NULL otherwise
 */
-struct mem_block *vpx_memory_tracker_find(size_t addr)
-{
-    struct mem_block *p = NULL;
+struct mem_block *vpx_memory_tracker_find(size_t addr) {
+  struct mem_block *p = NULL;
 
-    if (!memory_tracker_lock_mutex())
-    {
-        p = memory_tracker_find(addr);
-        memory_tracker_unlock_mutex();
-    }
+  if (!memory_tracker_lock_mutex()) {
+    p = memory_tracker_find(addr);
+    memory_tracker_unlock_mutex();
+  }
 
-    return p;
+  return p;
 }
 
 /*
@@ -309,13 +293,11 @@
     library function to dump the current contents of the
     global memory allocation list
 */
-void vpx_memory_tracker_dump()
-{
-    if (!memory_tracker_lock_mutex())
-    {
-        memory_tracker_dump();
-        memory_tracker_unlock_mutex();
-    }
+void vpx_memory_tracker_dump() {
+  if (!memory_tracker_lock_mutex()) {
+    memory_tracker_dump();
+    memory_tracker_unlock_mutex();
+  }
 }
 
 /*
@@ -326,13 +308,11 @@
     integrity check function to inspect every address in the global
     memory allocation list
 */
-void vpx_memory_tracker_check_integrity(char *file, unsigned int line)
-{
-    if (!memory_tracker_lock_mutex())
-    {
-        memory_tracker_check_integrity(file, line);
-        memory_tracker_unlock_mutex();
-    }
+void vpx_memory_tracker_check_integrity(char *file, unsigned int line) {
+  if (!memory_tracker_lock_mutex()) {
+    memory_tracker_check_integrity(file, line);
+    memory_tracker_unlock_mutex();
+  }
 }
 
 /*
@@ -344,43 +324,38 @@
       -1: if the logging type could not be set, because the value was invalid
           or because a file could not be opened
 */
-int vpx_memory_tracker_set_log_type(int type, char *option)
-{
-    int ret = -1;
+int vpx_memory_tracker_set_log_type(int type, char *option) {
+  int ret = -1;
 
-    switch (type)
-    {
+  switch (type) {
     case 0:
-        g_logging.type = 0;
+      g_logging.type = 0;
 
-        if (!option)
-        {
-            g_logging.file = stderr;
-            ret = 0;
-        }
-        else
-        {
-            if ((g_logging.file = fopen((char *)option, "w")))
-                ret = 0;
-        }
+      if (!option) {
+        g_logging.file = stderr;
+        ret = 0;
+      } else {
+        if ((g_logging.file = fopen((char *)option, "w")))
+          ret = 0;
+      }
 
-        break;
+      break;
 #if defined(WIN32) && !defined(_WIN32_WCE)
     case 1:
-        g_logging.type = type;
-        ret = 0;
-        break;
+      g_logging.type = type;
+      ret = 0;
+      break;
 #endif
     default:
-        break;
-    }
+      break;
+  }
 
-    //output the version to the new logging destination
-    if (!ret)
-        memtrack_log("Memory Tracker logging initialized, "
-                     "Memory Tracker v."vpx_mem_tracker_version"\n");
+  // output the version to the new logging destination
+  if (!ret)
+    memtrack_log("Memory Tracker logging initialized, "
+                 "Memory Tracker v."vpx_mem_tracker_version"\n");
 
-    return ret;
+  return ret;
 }
 
 /*
@@ -392,24 +367,22 @@
 */
 int vpx_memory_tracker_set_log_func(void *userdata,
                                     void(*logfunc)(void *userdata,
-                                            const char *fmt, va_list args))
-{
-    int ret = -1;
+                                                   const char *fmt, va_list args)) {
+  int ret = -1;
 
-    if (logfunc)
-    {
-        g_logging.type     = -1;
-        g_logging.userdata = userdata;
-        g_logging.func     = logfunc;
-        ret = 0;
-    }
+  if (logfunc) {
+    g_logging.type     = -1;
+    g_logging.userdata = userdata;
+    g_logging.func     = logfunc;
+    ret = 0;
+  }
 
-    //output the version to the new logging destination
-    if (!ret)
-        memtrack_log("Memory Tracker logging initialized, "
-                     "Memory Tracker v."vpx_mem_tracker_version"\n");
+  // output the version to the new logging destination
+  if (!ret)
+    memtrack_log("Memory Tracker logging initialized, "
+                 "Memory Tracker v."vpx_mem_tracker_version"\n");
 
-    return ret;
+  return ret;
 }
 
 /*
@@ -425,79 +398,73 @@
  *
 */
 
-static void memtrack_log(const char *fmt, ...)
-{
-    va_list list;
+static void memtrack_log(const char *fmt, ...) {
+  va_list list;
 
-    va_start(list, fmt);
+  va_start(list, fmt);
 
-    switch (g_logging.type)
-    {
+  switch (g_logging.type) {
     case -1:
 
-        if (g_logging.func)
-            g_logging.func(g_logging.userdata, fmt, list);
+      if (g_logging.func)
+        g_logging.func(g_logging.userdata, fmt, list);
 
-        break;
+      break;
     case 0:
 
-        if (g_logging.file)
-        {
-            vfprintf(g_logging.file, fmt, list);
-            fflush(g_logging.file);
-        }
+      if (g_logging.file) {
+        vfprintf(g_logging.file, fmt, list);
+        fflush(g_logging.file);
+      }
 
-        break;
+      break;
 #if defined(WIN32) && !defined(_WIN32_WCE)
-    case 1:
-    {
-        char temp[1024];
-        _vsnprintf(temp, sizeof(temp) / sizeof(char) - 1, fmt, list);
-        OutputDebugString(temp);
+    case 1: {
+      char temp[1024];
+      _vsnprintf(temp, sizeof(temp) / sizeof(char) - 1, fmt, list);
+      OutputDebugString(temp);
     }
     break;
 #endif
     default:
-        break;
-    }
+      break;
+  }
 
-    va_end(list);
+  va_end(list);
 }
 
 /*
     memory_tracker_dump()
     Dumps the current contents of the global memory allocation list
 */
-static void memory_tracker_dump()
-{
-    int i = 0;
-    struct mem_block *p = (memtrack.head ? memtrack.head->next : NULL);
+static void memory_tracker_dump() {
+  int i = 0;
+  struct mem_block *p = (memtrack.head ? memtrack.head->next : NULL);
 
-    memtrack_log("\n_currently Allocated= %d; Max allocated= %d\n",
-                 memtrack.current_allocated, memtrack.max_allocated);
+  memtrack_log("\n_currently Allocated= %d; Max allocated= %d\n",
+               memtrack.current_allocated, memtrack.max_allocated);
 
-    while (p)
-    {
+  while (p) {
 #if defined(WIN32) && !defined(_WIN32_WCE)
 
-        /*when using outputdebugstring, output filenames so they
-          can be clicked to be opened in visual studio*/
-        if (g_logging.type == 1)
-            memtrack_log("memblocks[%d].addr= 0x%.8x, memblocks[%d].size= %d, file:\n"
-                         "  %s(%d):\n", i,
-                         p->addr, i, p->size,
-                         p->file, p->line);
-        else
+    /*when using outputdebugstring, output filenames so they
+      can be clicked to be opened in visual studio*/
+    if (g_logging.type == 1)
+      memtrack_log("memblocks[%d].addr= 0x%.8x, memblocks[%d].size= %d, file:\n"
+                   "  %s(%d):\n", i,
+                   p->addr, i, p->size,
+                   p->file, p->line);
+    else
 #endif
-            memtrack_log("memblocks[%d].addr= 0x%.8x, memblocks[%d].size= %d, file: %s, line: %d\n", i,
-                         p->addr, i, p->size,
-                         p->file, p->line);
+      memtrack_log("memblocks[%d].addr= 0x%.8x, memblocks[%d].size= %d, file: %s, line: %d\n", i,
+                   p->addr, i, p->size,
+                   p->file, p->line);
 
-        p = p->next;
-        ++i;
-    }
+    p = p->next;
+    ++i;
+  }
 
-    memtrack_log("\n");
+  memtrack_log("\n");
 }
 
 /*
@@ -508,55 +475,49 @@
     this function will check ea. addr in the list verifying that
     addr-padding_size and addr+padding_size is filled with pad_value
 */
-static void memory_tracker_check_integrity(char *file, unsigned int line)
-{
-    if (memtrack.padding_size)
-    {
-        int i,
-            index = 0;
-        unsigned char *p_show_me,
-                 * p_show_me2;
-        unsigned int tempme = memtrack.pad_value,
-                     dead1,
-                     dead2;
-        unsigned char *x_bounds;
-        struct mem_block *p = memtrack.head->next;
+static void memory_tracker_check_integrity(char *file, unsigned int line) {
+  if (memtrack.padding_size) {
+    int i,
+        index = 0;
+    unsigned char *p_show_me,
+             * p_show_me2;
+    unsigned int tempme = memtrack.pad_value,
+                 dead1,
+                 dead2;
+    unsigned char *x_bounds;
+    struct mem_block *p = memtrack.head->next;
 
-        while (p)
-        {
-            //x_bounds = (unsigned char*)p->addr;
-            //back up VPX_BYTE_ALIGNMENT
-            //x_bounds -= memtrack.padding_size;
+    while (p) {
+      // x_bounds = (unsigned char*)p->addr;
+      // back up VPX_BYTE_ALIGNMENT
+      // x_bounds -= memtrack.padding_size;
 
-            if (p->padded)   // can the bounds be checked?
-            {
-                /*yes, move to the address that was actually allocated
-                by the vpx_* calls*/
-                x_bounds = (unsigned char *)(((size_t *)p->addr)[-1]);
+      if (p->padded) { // can the bounds be checked?
+        /*yes, move to the address that was actually allocated
+        by the vpx_* calls*/
+        x_bounds = (unsigned char *)(((size_t *)p->addr)[-1]);
 
-                for (i = 0; i < memtrack.padding_size; i += sizeof(unsigned int))
-                {
-                    p_show_me = (x_bounds + i);
-                    p_show_me2 = (unsigned char *)(p->addr + p->size + i);
+        for (i = 0; i < memtrack.padding_size; i += sizeof(unsigned int)) {
+          p_show_me = (x_bounds + i);
+          p_show_me2 = (unsigned char *)(p->addr + p->size + i);
 
-                    MEM_TRACK_MEMCPY(&dead1, p_show_me, sizeof(unsigned int));
-                    MEM_TRACK_MEMCPY(&dead2, p_show_me2, sizeof(unsigned int));
+          MEM_TRACK_MEMCPY(&dead1, p_show_me, sizeof(unsigned int));
+          MEM_TRACK_MEMCPY(&dead2, p_show_me2, sizeof(unsigned int));
 
-                    if ((dead1 != tempme) || (dead2 != tempme))
-                    {
-                        memtrack_log("\n[vpx_mem integrity check failed]:\n"
-                                     "    index[%d,%d] {%s:%d} addr=0x%x, size=%d,"
-                                     " file: %s, line: %d c0:0x%x c1:0x%x\n",
-                                     index, i, file, line, p->addr, p->size, p->file,
-                                     p->line, dead1, dead2);
-                    }
-                }
-            }
-
-            ++index;
-            p = p->next;
+          if ((dead1 != tempme) || (dead2 != tempme)) {
+            memtrack_log("\n[vpx_mem integrity check failed]:\n"
+                         "    index[%d,%d] {%s:%d} addr=0x%x, size=%d,"
+                         " file: %s, line: %d c0:0x%x c1:0x%x\n",
+                         index, i, file, line, p->addr, p->size, p->file,
+                         p->line, dead1, dead2);
+          }
         }
+      }
+
+      ++index;
+      p = p->next;
     }
+  }
 }
 
 /*
@@ -568,43 +529,38 @@
 */
 void memory_tracker_add(size_t addr, unsigned int size,
                         char *file, unsigned int line,
-                        int padded)
-{
-    if (!memory_tracker_lock_mutex())
-    {
-        struct mem_block *p;
+                        int padded) {
+  if (!memory_tracker_lock_mutex()) {
+    struct mem_block *p;
 
-        p = MEM_TRACK_MALLOC(sizeof(struct mem_block));
+    p = MEM_TRACK_MALLOC(sizeof(struct mem_block));
 
-        if (p)
-        {
-            p->prev       = memtrack.tail;
-            p->prev->next = p;
-            p->addr       = addr;
-            p->size       = size;
-            p->line       = line;
-            p->file       = file;
-            p->padded     = padded;
-            p->next       = NULL;
+    if (p) {
+      p->prev       = memtrack.tail;
+      p->prev->next = p;
+      p->addr       = addr;
+      p->size       = size;
+      p->line       = line;
+      p->file       = file;
+      p->padded     = padded;
+      p->next       = NULL;
 
-            memtrack.tail = p;
+      memtrack.tail = p;
 
-            memtrack.current_allocated += size;
+      memtrack.current_allocated += size;
 
-            if (memtrack.current_allocated > memtrack.max_allocated)
-                memtrack.max_allocated = memtrack.current_allocated;
+      if (memtrack.current_allocated > memtrack.max_allocated)
+        memtrack.max_allocated = memtrack.current_allocated;
 
-            //memtrack_log("memory_tracker_add: added addr=0x%.8x\n", addr);
+      // memtrack_log("memory_tracker_add: added addr=0x%.8x\n", addr);
 
-            memory_tracker_unlock_mutex();
-        }
-        else
-        {
-            memtrack_log("memory_tracker_add: error allocating memory!\n");
-            memory_tracker_unlock_mutex();
-            vpx_memory_tracker_destroy();
-        }
+      memory_tracker_unlock_mutex();
+    } else {
+      memtrack_log("memory_tracker_add: error allocating memory!\n");
+      memory_tracker_unlock_mutex();
+      vpx_memory_tracker_destroy();
     }
+  }
 }
 
 /*
@@ -617,41 +573,36 @@
       -1: if the mutex could not be locked
       -2: if the addr was not found in the list
 */
-int memory_tracker_remove(size_t addr)
-{
-    int ret = -1;
+int memory_tracker_remove(size_t addr) {
+  int ret = -1;
 
-    if (!memory_tracker_lock_mutex())
-    {
-        struct mem_block *p;
+  if (!memory_tracker_lock_mutex()) {
+    struct mem_block *p;
 
-        if ((p = memory_tracker_find(addr)))
-        {
-            memtrack.current_allocated -= p->size;
+    if ((p = memory_tracker_find(addr))) {
+      memtrack.current_allocated -= p->size;
 
-            p->prev->next = p->next;
+      p->prev->next = p->next;
 
-            if (p->next)
-                p->next->prev = p->prev;
-            else
-                memtrack.tail = p->prev;
+      if (p->next)
+        p->next->prev = p->prev;
+      else
+        memtrack.tail = p->prev;
 
-            ret = 0;
-            MEM_TRACK_FREE(p);
-        }
-        else
-        {
-            if (addr)
-                memtrack_log("memory_tracker_remove(): addr not found in list,"
-                             " 0x%.8x\n", addr);
+      ret = 0;
+      MEM_TRACK_FREE(p);
+    } else {
+      if (addr)
+        memtrack_log("memory_tracker_remove(): addr not found in list,"
+                     " 0x%.8x\n", addr);
 
-            ret = -2;
-        }
-
-        memory_tracker_unlock_mutex();
+      ret = -2;
     }
 
-    return ret;
+    memory_tracker_unlock_mutex();
+  }
+
+  return ret;
 }
 
 /*
@@ -662,19 +613,17 @@
           the need for repeated locking and unlocking as in Remove
     Returns: pointer to the mem block if found, NULL otherwise
 */
-static struct mem_block *memory_tracker_find(size_t addr)
-{
-    struct mem_block *p = NULL;
+static struct mem_block *memory_tracker_find(size_t addr) {
+  struct mem_block *p = NULL;
 
-    if (memtrack.head)
-    {
-        p = memtrack.head->next;
+  if (memtrack.head) {
+    p = memtrack.head->next;
 
-        while (p && (p->addr != addr))
-            p = p->next;
-    }
+    while (p && (p->addr != addr))
+      p = p->next;
+  }
 
-    return p;
+  return p;
 }
 
 
@@ -687,28 +636,25 @@
        <0: Failure, either the mutex was not initialized
            or the call to lock the mutex failed
 */
-static int memory_tracker_lock_mutex()
-{
-    int ret = -1;
+static int memory_tracker_lock_mutex() {
+  int ret = -1;
 
-    if (g_b_mem_tracker_inited)
-    {
+  if (g_b_mem_tracker_inited) {
 
 #if HAVE_PTHREAD_H
-        ret = pthread_mutex_lock(&memtrack.mutex);
+    ret = pthread_mutex_lock(&memtrack.mutex);
 #elif defined(WIN32) || defined(_WIN32_WCE)
-        ret = WaitForSingleObject(memtrack.mutex, INFINITE);
+    ret = WaitForSingleObject(memtrack.mutex, INFINITE);
 #elif defined(VXWORKS)
-        ret = sem_take(memtrack.mutex, WAIT_FOREVER);
+    ret = sem_take(memtrack.mutex, WAIT_FOREVER);
 #endif
 
-        if (ret)
-        {
-            memtrack_log("memory_tracker_lock_mutex: mutex lock failed\n");
-        }
+    if (ret) {
+      memtrack_log("memory_tracker_lock_mutex: mutex lock failed\n");
     }
+  }
 
-    return ret;
+  return ret;
 }
 
 /*
@@ -719,28 +665,25 @@
        <0: Failure, either the mutex was not initialized
            or the call to unlock the mutex failed
 */
-static int memory_tracker_unlock_mutex()
-{
-    int ret = -1;
+static int memory_tracker_unlock_mutex() {
+  int ret = -1;
 
-    if (g_b_mem_tracker_inited)
-    {
+  if (g_b_mem_tracker_inited) {
 
 #if HAVE_PTHREAD_H
-        ret = pthread_mutex_unlock(&memtrack.mutex);
+    ret = pthread_mutex_unlock(&memtrack.mutex);
 #elif defined(WIN32) || defined(_WIN32_WCE)
-        ret = !ReleaseMutex(memtrack.mutex);
+    ret = !ReleaseMutex(memtrack.mutex);
 #elif defined(VXWORKS)
-        ret = sem_give(memtrack.mutex);
+    ret = sem_give(memtrack.mutex);
 #endif
 
-        if (ret)
-        {
-            memtrack_log("memory_tracker_unlock_mutex: mutex unlock failed\n");
-        }
+    if (ret) {
+      memtrack_log("memory_tracker_unlock_mutex: mutex unlock failed\n");
     }
+  }
 
-    return ret;
+  return ret;
 }
 #endif
 
@@ -754,45 +697,44 @@
       -1: if the use global function pointers is not set.
 */
 int vpx_memory_tracker_set_functions(mem_track_malloc_func g_malloc_l
-                                     , mem_track_calloc_func g_calloc_l
-                                     , mem_track_realloc_func g_realloc_l
-                                     , mem_track_free_func g_free_l
-                                     , mem_track_memcpy_func g_memcpy_l
-                                     , mem_track_memset_func g_memset_l
-                                     , mem_track_memmove_func g_memmove_l)
-{
+, mem_track_calloc_func g_calloc_l
+, mem_track_realloc_func g_realloc_l
+, mem_track_free_func g_free_l
+, mem_track_memcpy_func g_memcpy_l
+, mem_track_memset_func g_memset_l
+, mem_track_memmove_func g_memmove_l) {
 #if USE_GLOBAL_FUNCTION_POINTERS
 
-    if (g_malloc_l)
-        g_malloc = g_malloc_l;
+  if (g_malloc_l)
+    g_malloc = g_malloc_l;
 
-    if (g_calloc_l)
-        g_calloc = g_calloc_l;
+  if (g_calloc_l)
+    g_calloc = g_calloc_l;
 
-    if (g_realloc_l)
-        g_realloc = g_realloc_l;
+  if (g_realloc_l)
+    g_realloc = g_realloc_l;
 
-    if (g_free_l)
-        g_free = g_free_l;
+  if (g_free_l)
+    g_free = g_free_l;
 
-    if (g_memcpy_l)
-        g_memcpy = g_memcpy_l;
+  if (g_memcpy_l)
+    g_memcpy = g_memcpy_l;
 
-    if (g_memset_l)
-        g_memset = g_memset_l;
+  if (g_memset_l)
+    g_memset = g_memset_l;
 
-    if (g_memmove_l)
-        g_memmove = g_memmove_l;
+  if (g_memmove_l)
+    g_memmove = g_memmove_l;
 
-    return 0;
+  return 0;
 #else
-    (void)g_malloc_l;
-    (void)g_calloc_l;
-    (void)g_realloc_l;
-    (void)g_free_l;
-    (void)g_memcpy_l;
-    (void)g_memset_l;
-    (void)g_memmove_l;
-    return -1;
+  (void)g_malloc_l;
+  (void)g_calloc_l;
+  (void)g_realloc_l;
+  (void)g_free_l;
+  (void)g_memcpy_l;
+  (void)g_memset_l;
+  (void)g_memmove_l;
+  return -1;
 #endif
 }
diff --git a/libvpx/vpx_ports/arm_cpudetect.c b/libvpx/vpx_ports/arm_cpudetect.c
index 8ff95a1..542ff67 100644
--- a/libvpx/vpx_ports/arm_cpudetect.c
+++ b/libvpx/vpx_ports/arm_cpudetect.c
@@ -12,50 +12,45 @@
 #include <string.h>
 #include "arm.h"
 
-static int arm_cpu_env_flags(int *flags)
-{
-    char *env;
-    env = getenv("VPX_SIMD_CAPS");
-    if (env && *env)
-    {
-        *flags = (int)strtol(env, NULL, 0);
-        return 0;
-    }
-    *flags = 0;
-    return -1;
+static int arm_cpu_env_flags(int *flags) {
+  char *env;
+  env = getenv("VPX_SIMD_CAPS");
+  if (env && *env) {
+    *flags = (int)strtol(env, NULL, 0);
+    return 0;
+  }
+  *flags = 0;
+  return -1;
 }
 
-static int arm_cpu_env_mask(void)
-{
-    char *env;
-    env = getenv("VPX_SIMD_CAPS_MASK");
-    return env && *env ? (int)strtol(env, NULL, 0) : ~0;
+static int arm_cpu_env_mask(void) {
+  char *env;
+  env = getenv("VPX_SIMD_CAPS_MASK");
+  return env && *env ? (int)strtol(env, NULL, 0) : ~0;
 }
 
 #if !CONFIG_RUNTIME_CPU_DETECT
 
-int arm_cpu_caps(void)
-{
+int arm_cpu_caps(void) {
   /* This function should actually be a no-op. There is no way to adjust any of
    * these because the RTCD tables do not exist: the functions are called
    * statically */
-    int flags;
-    int mask;
-    if (!arm_cpu_env_flags(&flags))
-    {
-        return flags;
-    }
-    mask = arm_cpu_env_mask();
+  int flags;
+  int mask;
+  if (!arm_cpu_env_flags(&flags)) {
+    return flags;
+  }
+  mask = arm_cpu_env_mask();
 #if HAVE_EDSP
-    flags |= HAS_EDSP;
+  flags |= HAS_EDSP;
 #endif /* HAVE_EDSP */
 #if HAVE_MEDIA
-    flags |= HAS_MEDIA;
+  flags |= HAS_MEDIA;
 #endif /* HAVE_MEDIA */
 #if HAVE_NEON
-    flags |= HAS_NEON;
+  flags |= HAS_NEON;
 #endif /* HAVE_NEON */
-    return flags & mask;
+  return flags & mask;
 }
 
 #elif defined(_MSC_VER) /* end !CONFIG_RUNTIME_CPU_DETECT */
@@ -64,156 +59,134 @@
 #define WIN32_EXTRA_LEAN
 #include <windows.h>
 
-int arm_cpu_caps(void)
-{
-    int flags;
-    int mask;
-    if (!arm_cpu_env_flags(&flags))
-    {
-        return flags;
-    }
-    mask = arm_cpu_env_mask();
-    /* MSVC has no inline __asm support for ARM, but it does let you __emit
-     *  instructions via their assembled hex code.
-     * All of these instructions should be essentially nops.
-     */
+int arm_cpu_caps(void) {
+  int flags;
+  int mask;
+  if (!arm_cpu_env_flags(&flags)) {
+    return flags;
+  }
+  mask = arm_cpu_env_mask();
+  /* MSVC has no inline __asm support for ARM, but it does let you __emit
+   *  instructions via their assembled hex code.
+   * All of these instructions should be essentially nops.
+   */
 #if HAVE_EDSP
-    if (mask & HAS_EDSP)
-    {
-        __try
-        {
-            /*PLD [r13]*/
-            __emit(0xF5DDF000);
-            flags |= HAS_EDSP;
-        }
-        __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION)
-        {
-            /*Ignore exception.*/
-        }
+  if (mask & HAS_EDSP) {
+    __try {
+      /*PLD [r13]*/
+      __emit(0xF5DDF000);
+      flags |= HAS_EDSP;
+    } __except (GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) {
+      /*Ignore exception.*/
     }
+  }
 #if HAVE_MEDIA
-    if (mask & HAS_MEDIA)
-        __try
-        {
-            /*SHADD8 r3,r3,r3*/
-            __emit(0xE6333F93);
-            flags |= HAS_MEDIA;
-        }
-        __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION)
-        {
-            /*Ignore exception.*/
-        }
-    }
+  if (mask & HAS_MEDIA)
+    __try {
+      /*SHADD8 r3,r3,r3*/
+      __emit(0xE6333F93);
+      flags |= HAS_MEDIA;
+    } __except (GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) {
+    /*Ignore exception.*/
+  }
+}
 #if HAVE_NEON
-    if (mask & HAS_NEON)
-    {
-        __try
-        {
-            /*VORR q0,q0,q0*/
-            __emit(0xF2200150);
-            flags |= HAS_NEON;
-        }
-        __except(GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION)
-        {
-            /*Ignore exception.*/
-        }
-    }
+if (mask &HAS_NEON) {
+  __try {
+    /*VORR q0,q0,q0*/
+    __emit(0xF2200150);
+    flags |= HAS_NEON;
+  } __except (GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION) {
+    /*Ignore exception.*/
+  }
+}
 #endif /* HAVE_NEON */
 #endif /* HAVE_MEDIA */
 #endif /* HAVE_EDSP */
-    return flags & mask;
+return flags & mask;
 }
 
 #elif defined(__ANDROID__) /* end _MSC_VER */
 #include <cpu-features.h>
 
-int arm_cpu_caps(void)
-{
-    int flags;
-    int mask;
-    uint64_t features;
-    if (!arm_cpu_env_flags(&flags))
-    {
-        return flags;
-    }
-    mask = arm_cpu_env_mask();
-    features = android_getCpuFeatures();
+int arm_cpu_caps(void) {
+  int flags;
+  int mask;
+  uint64_t features;
+  if (!arm_cpu_env_flags(&flags)) {
+    return flags;
+  }
+  mask = arm_cpu_env_mask();
+  features = android_getCpuFeatures();
 
 #if HAVE_EDSP
-    flags |= HAS_EDSP;
+  flags |= HAS_EDSP;
 #endif /* HAVE_EDSP */
 #if HAVE_MEDIA
-    flags |= HAS_MEDIA;
+  flags |= HAS_MEDIA;
 #endif /* HAVE_MEDIA */
 #if HAVE_NEON
-    if (features & ANDROID_CPU_ARM_FEATURE_NEON)
-        flags |= HAS_NEON;
+  if (features & ANDROID_CPU_ARM_FEATURE_NEON)
+    flags |= HAS_NEON;
 #endif /* HAVE_NEON */
-    return flags & mask;
+  return flags & mask;
 }
 
 #elif defined(__linux__) /* end __ANDROID__ */
+
 #include <stdio.h>
 
-int arm_cpu_caps(void)
-{
-    FILE *fin;
-    int flags;
-    int mask;
-    if (!arm_cpu_env_flags(&flags))
-    {
-        return flags;
-    }
-    mask = arm_cpu_env_mask();
-    /* Reading /proc/self/auxv would be easier, but that doesn't work reliably
-     *  on Android.
-     * This also means that detection will fail in Scratchbox.
+int arm_cpu_caps(void) {
+  FILE *fin;
+  int flags;
+  int mask;
+  if (!arm_cpu_env_flags(&flags)) {
+    return flags;
+  }
+  mask = arm_cpu_env_mask();
+  /* Reading /proc/self/auxv would be easier, but that doesn't work reliably
+   *  on Android.
+   * This also means that detection will fail in Scratchbox.
+   */
+  fin = fopen("/proc/cpuinfo", "r");
+  if (fin != NULL) {
+    /* 512 should be enough for anybody (it's even enough for all the flags
+     * that x86 has accumulated... so far).
      */
-    fin = fopen("/proc/cpuinfo","r");
-    if(fin != NULL)
-    {
-        /* 512 should be enough for anybody (it's even enough for all the flags
-         * that x86 has accumulated... so far).
-         */
-        char buf[512];
-        while (fgets(buf, 511, fin) != NULL)
-        {
+    char buf[512];
+    while (fgets(buf, 511, fin) != NULL) {
 #if HAVE_EDSP || HAVE_NEON
-            if (memcmp(buf, "Features", 8) == 0)
-            {
-                char *p;
+      if (memcmp(buf, "Features", 8) == 0) {
+        char *p;
 #if HAVE_EDSP
-                p=strstr(buf, " edsp");
-                if (p != NULL && (p[5] == ' ' || p[5] == '\n'))
-                {
-                    flags |= HAS_EDSP;
-                }
+        p = strstr(buf, " edsp");
+        if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
+          flags |= HAS_EDSP;
+        }
 #if HAVE_NEON
-                p = strstr(buf, " neon");
-                if (p != NULL && (p[5] == ' ' || p[5] == '\n'))
-                {
-                    flags |= HAS_NEON;
-                }
+        p = strstr(buf, " neon");
+        if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
+          flags |= HAS_NEON;
+        }
 #endif /* HAVE_NEON */
 #endif /* HAVE_EDSP */
-            }
+      }
 #endif /* HAVE_EDSP || HAVE_NEON */
 #if HAVE_MEDIA
-            if (memcmp(buf, "CPU architecture:",17) == 0){
-                int version;
-                version = atoi(buf+17);
-                if (version >= 6)
-                {
-                    flags |= HAS_MEDIA;
-                }
-            }
-#endif /* HAVE_MEDIA */
+      if (memcmp(buf, "CPU architecture:", 17) == 0) {
+        int version;
+        version = atoi(buf + 17);
+        if (version >= 6) {
+          flags |= HAS_MEDIA;
         }
-        fclose(fin);
+      }
+#endif /* HAVE_MEDIA */
     }
-    return flags & mask;
+    fclose(fin);
+  }
+  return flags & mask;
 }
 #else /* end __linux__ */
 #error "--enable-runtime-cpu-detect selected, but no CPU detection method " \
- "available for your platform. Reconfigure with --disable-runtime-cpu-detect."
+"available for your platform. Reconfigure with --disable-runtime-cpu-detect."
 #endif
diff --git a/libvpx/vpx_ports/asm_offsets.h b/libvpx/vpx_ports/asm_offsets.h
index 7b6ae4a..d3a3e5a 100644
--- a/libvpx/vpx_ports/asm_offsets.h
+++ b/libvpx/vpx_ports/asm_offsets.h
@@ -15,8 +15,8 @@
 #include <stddef.h>
 
 #define ct_assert(name,cond) \
-    static void assert_##name(void) UNUSED;\
-    static void assert_##name(void) {switch(0){case 0:case !!(cond):;}}
+  static void assert_##name(void) UNUSED;\
+  static void assert_##name(void) {switch(0){case 0:case !!(cond):;}}
 
 #if INLINE_ASM
 #define DEFINE(sym, val) asm("\n" #sym " EQU %0" : : "i" (val))
diff --git a/libvpx/vpx_ports/config.h b/libvpx/vpx_ports/config.h
new file mode 100644
index 0000000..1abe70d
--- /dev/null
+++ b/libvpx/vpx_ports/config.h
@@ -0,0 +1,10 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "vpx_config.h"
diff --git a/libvpx/vpx_ports/emmintrin_compat.h b/libvpx/vpx_ports/emmintrin_compat.h
new file mode 100644
index 0000000..782d603
--- /dev/null
+++ b/libvpx/vpx_ports/emmintrin_compat.h
@@ -0,0 +1,55 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VPX_PORTS_EMMINTRIN_COMPAT_H
+#define VPX_PORTS_EMMINTRIN_COMPAT_H
+
+#if defined(__GNUC__) && __GNUC__ < 4
+/* From emmintrin.h (gcc 4.5.3) */
+/* Casts between various SP, DP, INT vector types.  Note that these do no
+   conversion of values, they just change the type.  */
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castpd_ps(__m128d __A)
+{
+  return (__m128) __A;
+}
+
+extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castpd_si128(__m128d __A)
+{
+  return (__m128i) __A;
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castps_pd(__m128 __A)
+{
+  return (__m128d) __A;
+}
+
+extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castps_si128(__m128 __A)
+{
+  return (__m128i) __A;
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castsi128_ps(__m128i __A)
+{
+  return (__m128) __A;
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_castsi128_pd(__m128i __A)
+{
+  return (__m128d) __A;
+}
+#endif
+
+#endif
diff --git a/libvpx/vpx_ports/emms.asm b/libvpx/vpx_ports/emms.asm
index efad1a5..db8da28 100644
--- a/libvpx/vpx_ports/emms.asm
+++ b/libvpx/vpx_ports/emms.asm
@@ -18,7 +18,7 @@
     ret
 
 
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
 global sym(vpx_winx64_fldcw) PRIVATE
 sym(vpx_winx64_fldcw):
     sub   rsp, 8
diff --git a/libvpx/vpx_ports/mem.h b/libvpx/vpx_ports/mem.h
index 29e507f..62b86bb 100644
--- a/libvpx/vpx_ports/mem.h
+++ b/libvpx/vpx_ports/mem.h
@@ -11,6 +11,7 @@
 
 #ifndef VPX_PORTS_MEM_H
 #define VPX_PORTS_MEM_H
+
 #include "vpx_config.h"
 #include "vpx/vpx_integer.h"
 
@@ -31,8 +32,8 @@
  * within the array.
  */
 #define DECLARE_ALIGNED_ARRAY(a,typ,val,n)\
-typ val##_[(n)+(a)/sizeof(typ)+1];\
-typ *val = (typ*)((((intptr_t)val##_)+(a)-1)&((intptr_t)-(a)))
+  typ val##_[(n)+(a)/sizeof(typ)+1];\
+  typ *val = (typ*)((((intptr_t)val##_)+(a)-1)&((intptr_t)-(a)))
 
 
 /* Indicates that the usage of the specified variable has been audited to assure
diff --git a/libvpx/vpx_ports/mem_ops.h b/libvpx/vpx_ports/mem_ops.h
index dec28d5..2d44a3a 100644
--- a/libvpx/vpx_ports/mem_ops.h
+++ b/libvpx/vpx_ports/mem_ops.h
@@ -60,88 +60,82 @@
 
 #undef  mem_get_be16
 #define mem_get_be16 mem_ops_wrap_symbol(mem_get_be16)
-static unsigned MEM_VALUE_T mem_get_be16(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_be16(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[0] << 8;
-    val |= mem[1];
-    return val;
+  val = mem[0] << 8;
+  val |= mem[1];
+  return val;
 }
 
 #undef  mem_get_be24
 #define mem_get_be24 mem_ops_wrap_symbol(mem_get_be24)
-static unsigned MEM_VALUE_T mem_get_be24(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_be24(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[0] << 16;
-    val |= mem[1] << 8;
-    val |= mem[2];
-    return val;
+  val = mem[0] << 16;
+  val |= mem[1] << 8;
+  val |= mem[2];
+  return val;
 }
 
 #undef  mem_get_be32
 #define mem_get_be32 mem_ops_wrap_symbol(mem_get_be32)
-static unsigned MEM_VALUE_T mem_get_be32(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_be32(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[0] << 24;
-    val |= mem[1] << 16;
-    val |= mem[2] << 8;
-    val |= mem[3];
-    return val;
+  val = mem[0] << 24;
+  val |= mem[1] << 16;
+  val |= mem[2] << 8;
+  val |= mem[3];
+  return val;
 }
 
 #undef  mem_get_le16
 #define mem_get_le16 mem_ops_wrap_symbol(mem_get_le16)
-static unsigned MEM_VALUE_T mem_get_le16(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_le16(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[1] << 8;
-    val |= mem[0];
-    return val;
+  val = mem[1] << 8;
+  val |= mem[0];
+  return val;
 }
 
 #undef  mem_get_le24
 #define mem_get_le24 mem_ops_wrap_symbol(mem_get_le24)
-static unsigned MEM_VALUE_T mem_get_le24(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_le24(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[2] << 16;
-    val |= mem[1] << 8;
-    val |= mem[0];
-    return val;
+  val = mem[2] << 16;
+  val |= mem[1] << 8;
+  val |= mem[0];
+  return val;
 }
 
 #undef  mem_get_le32
 #define mem_get_le32 mem_ops_wrap_symbol(mem_get_le32)
-static unsigned MEM_VALUE_T mem_get_le32(const void *vmem)
-{
-    unsigned MEM_VALUE_T  val;
-    const MAU_T          *mem = (const MAU_T *)vmem;
+static unsigned MEM_VALUE_T mem_get_le32(const void *vmem) {
+  unsigned MEM_VALUE_T  val;
+  const MAU_T          *mem = (const MAU_T *)vmem;
 
-    val = mem[3] << 24;
-    val |= mem[2] << 16;
-    val |= mem[1] << 8;
-    val |= mem[0];
-    return val;
+  val = mem[3] << 24;
+  val |= mem[2] << 16;
+  val |= mem[1] << 8;
+  val |= mem[0];
+  return val;
 }
 
 #define mem_get_s_generic(end,sz) \
-    static signed MEM_VALUE_T mem_get_s##end##sz(const void *vmem) {\
-        const MAU_T *mem = (const MAU_T*)vmem;\
-        signed MEM_VALUE_T val = mem_get_##end##sz(mem);\
-        return (val << (MEM_VALUE_T_SZ_BITS - sz)) >> (MEM_VALUE_T_SZ_BITS - sz);\
-    }
+  static signed MEM_VALUE_T mem_get_s##end##sz(const void *vmem) {\
+    const MAU_T *mem = (const MAU_T*)vmem;\
+    signed MEM_VALUE_T val = mem_get_##end##sz(mem);\
+    return (val << (MEM_VALUE_T_SZ_BITS - sz)) >> (MEM_VALUE_T_SZ_BITS - sz);\
+  }
 
 #undef  mem_get_sbe16
 #define mem_get_sbe16 mem_ops_wrap_symbol(mem_get_sbe16)
@@ -169,66 +163,60 @@
 
 #undef  mem_put_be16
 #define mem_put_be16 mem_ops_wrap_symbol(mem_put_be16)
-static void mem_put_be16(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_be16(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >> 8) & 0xff;
-    mem[1] = (val >> 0) & 0xff;
+  mem[0] = (val >> 8) & 0xff;
+  mem[1] = (val >> 0) & 0xff;
 }
 
 #undef  mem_put_be24
 #define mem_put_be24 mem_ops_wrap_symbol(mem_put_be24)
-static void mem_put_be24(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_be24(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >> 16) & 0xff;
-    mem[1] = (val >>  8) & 0xff;
-    mem[2] = (val >>  0) & 0xff;
+  mem[0] = (val >> 16) & 0xff;
+  mem[1] = (val >>  8) & 0xff;
+  mem[2] = (val >>  0) & 0xff;
 }
 
 #undef  mem_put_be32
 #define mem_put_be32 mem_ops_wrap_symbol(mem_put_be32)
-static void mem_put_be32(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_be32(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >> 24) & 0xff;
-    mem[1] = (val >> 16) & 0xff;
-    mem[2] = (val >>  8) & 0xff;
-    mem[3] = (val >>  0) & 0xff;
+  mem[0] = (val >> 24) & 0xff;
+  mem[1] = (val >> 16) & 0xff;
+  mem[2] = (val >>  8) & 0xff;
+  mem[3] = (val >>  0) & 0xff;
 }
 
 #undef  mem_put_le16
 #define mem_put_le16 mem_ops_wrap_symbol(mem_put_le16)
-static void mem_put_le16(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_le16(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >>  0) & 0xff;
-    mem[1] = (val >>  8) & 0xff;
+  mem[0] = (val >>  0) & 0xff;
+  mem[1] = (val >>  8) & 0xff;
 }
 
 #undef  mem_put_le24
 #define mem_put_le24 mem_ops_wrap_symbol(mem_put_le24)
-static void mem_put_le24(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_le24(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >>  0) & 0xff;
-    mem[1] = (val >>  8) & 0xff;
-    mem[2] = (val >> 16) & 0xff;
+  mem[0] = (val >>  0) & 0xff;
+  mem[1] = (val >>  8) & 0xff;
+  mem[2] = (val >> 16) & 0xff;
 }
 
 #undef  mem_put_le32
 #define mem_put_le32 mem_ops_wrap_symbol(mem_put_le32)
-static void mem_put_le32(void *vmem, MEM_VALUE_T val)
-{
-    MAU_T *mem = (MAU_T *)vmem;
+static void mem_put_le32(void *vmem, MEM_VALUE_T val) {
+  MAU_T *mem = (MAU_T *)vmem;
 
-    mem[0] = (val >>  0) & 0xff;
-    mem[1] = (val >>  8) & 0xff;
-    mem[2] = (val >> 16) & 0xff;
-    mem[3] = (val >> 24) & 0xff;
+  mem[0] = (val >>  0) & 0xff;
+  mem[1] = (val >>  8) & 0xff;
+  mem[2] = (val >> 16) & 0xff;
+  mem[3] = (val >> 24) & 0xff;
 }
diff --git a/libvpx/vpx_ports/mem_ops_aligned.h b/libvpx/vpx_ports/mem_ops_aligned.h
index fca653a..0100300 100644
--- a/libvpx/vpx_ports/mem_ops_aligned.h
+++ b/libvpx/vpx_ports/mem_ops_aligned.h
@@ -24,61 +24,61 @@
  * could redefine these macros.
  */
 #define swap_endian_16(val,raw) do {\
-        val = ((raw>>8) & 0x00ff) \
-              | ((raw<<8) & 0xff00);\
-    } while(0)
+    val = ((raw>>8) & 0x00ff) \
+          | ((raw<<8) & 0xff00);\
+  } while(0)
 #define swap_endian_32(val,raw) do {\
-        val = ((raw>>24) & 0x000000ff) \
-              | ((raw>>8)  & 0x0000ff00) \
-              | ((raw<<8)  & 0x00ff0000) \
-              | ((raw<<24) & 0xff000000); \
-    } while(0)
+    val = ((raw>>24) & 0x000000ff) \
+          | ((raw>>8)  & 0x0000ff00) \
+          | ((raw<<8)  & 0x00ff0000) \
+          | ((raw<<24) & 0xff000000); \
+  } while(0)
 #define swap_endian_16_se(val,raw) do {\
-        swap_endian_16(val,raw);\
-        val = ((val << 16) >> 16);\
-    } while(0)
+    swap_endian_16(val,raw);\
+    val = ((val << 16) >> 16);\
+  } while(0)
 #define swap_endian_32_se(val,raw) swap_endian_32(val,raw)
 
 #define mem_get_ne_aligned_generic(end,sz) \
-    static unsigned MEM_VALUE_T mem_get_##end##sz##_aligned(const void *vmem) {\
-        const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\
-        return *mem;\
-    }
+  static unsigned MEM_VALUE_T mem_get_##end##sz##_aligned(const void *vmem) {\
+    const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\
+    return *mem;\
+  }
 
 #define mem_get_sne_aligned_generic(end,sz) \
-    static signed MEM_VALUE_T mem_get_s##end##sz##_aligned(const void *vmem) {\
-        const int##sz##_t *mem = (const int##sz##_t *)vmem;\
-        return *mem;\
-    }
+  static signed MEM_VALUE_T mem_get_s##end##sz##_aligned(const void *vmem) {\
+    const int##sz##_t *mem = (const int##sz##_t *)vmem;\
+    return *mem;\
+  }
 
 #define mem_get_se_aligned_generic(end,sz) \
-    static unsigned MEM_VALUE_T mem_get_##end##sz##_aligned(const void *vmem) {\
-        const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\
-        unsigned MEM_VALUE_T val, raw = *mem;\
-        swap_endian_##sz(val,raw);\
-        return val;\
-    }
+  static unsigned MEM_VALUE_T mem_get_##end##sz##_aligned(const void *vmem) {\
+    const uint##sz##_t *mem = (const uint##sz##_t *)vmem;\
+    unsigned MEM_VALUE_T val, raw = *mem;\
+    swap_endian_##sz(val,raw);\
+    return val;\
+  }
 
 #define mem_get_sse_aligned_generic(end,sz) \
-    static signed MEM_VALUE_T mem_get_s##end##sz##_aligned(const void *vmem) {\
-        const int##sz##_t *mem = (const int##sz##_t *)vmem;\
-        unsigned MEM_VALUE_T val, raw = *mem;\
-        swap_endian_##sz##_se(val,raw);\
-        return val;\
-    }
+  static signed MEM_VALUE_T mem_get_s##end##sz##_aligned(const void *vmem) {\
+    const int##sz##_t *mem = (const int##sz##_t *)vmem;\
+    unsigned MEM_VALUE_T val, raw = *mem;\
+    swap_endian_##sz##_se(val,raw);\
+    return val;\
+  }
 
 #define mem_put_ne_aligned_generic(end,sz) \
-    static void mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\
-        uint##sz##_t *mem = (uint##sz##_t *)vmem;\
-        *mem = (uint##sz##_t)val;\
-    }
+  static void mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\
+    uint##sz##_t *mem = (uint##sz##_t *)vmem;\
+    *mem = (uint##sz##_t)val;\
+  }
 
 #define mem_put_se_aligned_generic(end,sz) \
-    static void mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\
-        uint##sz##_t *mem = (uint##sz##_t *)vmem, raw;\
-        swap_endian_##sz(raw,val);\
-        *mem = (uint##sz##_t)raw;\
-    }
+  static void mem_put_##end##sz##_aligned(void *vmem, MEM_VALUE_T val) {\
+    uint##sz##_t *mem = (uint##sz##_t *)vmem, raw;\
+    swap_endian_##sz(raw,val);\
+    *mem = (uint##sz##_t)raw;\
+  }
 
 #include "vpx_config.h"
 #if CONFIG_BIG_ENDIAN
diff --git a/libvpx/vpx_ports/vpx_once.h b/libvpx/vpx_ports/vpx_once.h
new file mode 100644
index 0000000..16a735c
--- /dev/null
+++ b/libvpx/vpx_ports/vpx_once.h
@@ -0,0 +1,97 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "vpx_config.h"
+
+#if CONFIG_MULTITHREAD && defined(_WIN32)
+#include <windows.h>
+#include <stdlib.h>
+static void once(void (*func)(void))
+{
+    static CRITICAL_SECTION *lock;
+    static LONG waiters;
+    static int done;
+    void *lock_ptr = &lock;
+
+    /* If the initialization is complete, return early. This isn't just an
+     * optimization, it prevents races on the destruction of the global
+     * lock.
+     */
+    if(done)
+        return;
+
+    InterlockedIncrement(&waiters);
+
+    /* Get a lock. We create one and try to make it the one-true-lock,
+     * throwing it away if we lost the race.
+     */
+
+    {
+        /* Scope to protect access to new_lock */
+        CRITICAL_SECTION *new_lock = malloc(sizeof(CRITICAL_SECTION));
+        InitializeCriticalSection(new_lock);
+        if (InterlockedCompareExchangePointer(lock_ptr, new_lock, NULL) != NULL)
+        {
+            DeleteCriticalSection(new_lock);
+            free(new_lock);
+        }
+    }
+
+    /* At this point, we have a lock that can be synchronized on. We don't
+     * care which thread actually performed the allocation.
+     */
+
+    EnterCriticalSection(lock);
+
+    if (!done)
+    {
+        func();
+        done = 1;
+    }
+
+    LeaveCriticalSection(lock);
+
+    /* Last one out should free resources. The destructed objects are
+     * protected by checking if(done) above.
+     */
+    if(!InterlockedDecrement(&waiters))
+    {
+        DeleteCriticalSection(lock);
+        free(lock);
+        lock = NULL;
+    }
+}
+
+
+#elif CONFIG_MULTITHREAD && HAVE_PTHREAD_H
+#include <pthread.h>
+static void once(void (*func)(void))
+{
+    static pthread_once_t lock = PTHREAD_ONCE_INIT;
+    pthread_once(&lock, func);
+}
+
+
+#else
+/* No-op version that performs no synchronization. vp8_rtcd() is idempotent,
+ * so as long as your platform provides atomic loads/stores of pointers
+ * no synchronization is strictly necessary.
+ */
+
+static void once(void (*func)(void))
+{
+    static int done;
+
+    if(!done)
+    {
+        func();
+        done = 1;
+    }
+}
+#endif
diff --git a/libvpx/vpx_ports/vpx_timer.h b/libvpx/vpx_ports/vpx_timer.h
index d07e086..cdad9ef 100644
--- a/libvpx/vpx_ports/vpx_timer.h
+++ b/libvpx/vpx_ports/vpx_timer.h
@@ -32,65 +32,61 @@
 /* timersub is not provided by msys at this time. */
 #ifndef timersub
 #define timersub(a, b, result) \
-    do { \
-        (result)->tv_sec = (a)->tv_sec - (b)->tv_sec; \
-        (result)->tv_usec = (a)->tv_usec - (b)->tv_usec; \
-        if ((result)->tv_usec < 0) { \
-            --(result)->tv_sec; \
-            (result)->tv_usec += 1000000; \
-        } \
-    } while (0)
+  do { \
+    (result)->tv_sec = (a)->tv_sec - (b)->tv_sec; \
+    (result)->tv_usec = (a)->tv_usec - (b)->tv_usec; \
+    if ((result)->tv_usec < 0) { \
+      --(result)->tv_sec; \
+      (result)->tv_usec += 1000000; \
+    } \
+  } while (0)
 #endif
 #endif
 
 
-struct vpx_usec_timer
-{
+struct vpx_usec_timer {
 #if defined(_WIN32)
-    LARGE_INTEGER  begin, end;
+  LARGE_INTEGER  begin, end;
 #else
-    struct timeval begin, end;
+  struct timeval begin, end;
 #endif
 };
 
 
 static void
-vpx_usec_timer_start(struct vpx_usec_timer *t)
-{
+vpx_usec_timer_start(struct vpx_usec_timer *t) {
 #if defined(_WIN32)
-    QueryPerformanceCounter(&t->begin);
+  QueryPerformanceCounter(&t->begin);
 #else
-    gettimeofday(&t->begin, NULL);
+  gettimeofday(&t->begin, NULL);
 #endif
 }
 
 
 static void
-vpx_usec_timer_mark(struct vpx_usec_timer *t)
-{
+vpx_usec_timer_mark(struct vpx_usec_timer *t) {
 #if defined(_WIN32)
-    QueryPerformanceCounter(&t->end);
+  QueryPerformanceCounter(&t->end);
 #else
-    gettimeofday(&t->end, NULL);
+  gettimeofday(&t->end, NULL);
 #endif
 }
 
 
 static int64_t
-vpx_usec_timer_elapsed(struct vpx_usec_timer *t)
-{
+vpx_usec_timer_elapsed(struct vpx_usec_timer *t) {
 #if defined(_WIN32)
-    LARGE_INTEGER freq, diff;
+  LARGE_INTEGER freq, diff;
 
-    diff.QuadPart = t->end.QuadPart - t->begin.QuadPart;
+  diff.QuadPart = t->end.QuadPart - t->begin.QuadPart;
 
-    QueryPerformanceFrequency(&freq);
-    return diff.QuadPart * 1000000 / freq.QuadPart;
+  QueryPerformanceFrequency(&freq);
+  return diff.QuadPart * 1000000 / freq.QuadPart;
 #else
-    struct timeval diff;
+  struct timeval diff;
 
-    timersub(&t->end, &t->begin, &diff);
-    return diff.tv_sec * 1000000 + diff.tv_usec;
+  timersub(&t->end, &t->begin, &diff);
+  return diff.tv_sec * 1000000 + diff.tv_usec;
 #endif
 }
 
@@ -101,9 +97,8 @@
 #define timersub(a, b, result)
 #endif
 
-struct vpx_usec_timer
-{
-    void *dummy;
+struct vpx_usec_timer {
+  void *dummy;
 };
 
 static void
@@ -113,7 +108,9 @@
 vpx_usec_timer_mark(struct vpx_usec_timer *t) { }
 
 static long
-vpx_usec_timer_elapsed(struct vpx_usec_timer *t) { return 0; }
+vpx_usec_timer_elapsed(struct vpx_usec_timer *t) {
+  return 0;
+}
 
 #endif /* CONFIG_OS_SUPPORT */
 
diff --git a/libvpx/vpx_ports/vpxtypes.h b/libvpx/vpx_ports/vpxtypes.h
deleted file mode 100644
index f2fb089..0000000
--- a/libvpx/vpx_ports/vpxtypes.h
+++ /dev/null
@@ -1,167 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-#ifndef __VPXTYPES_H__
-#define __VPXTYPES_H__
-
-#include "vpx_config.h"
-
-//#include <sys/types.h>
-#ifdef _MSC_VER
-# include <basetsd.h>
-typedef SSIZE_T ssize_t;
-#endif
-
-#if defined(HAVE_STDINT_H) && HAVE_STDINT_H
-/* C99 types are preferred to vpx integer types */
-# include <stdint.h>
-#endif
-
-/*!\defgroup basetypes Base Types
-  @{*/
-#if !defined(HAVE_STDINT_H) && !defined(INT_T_DEFINED)
-# ifdef STRICTTYPES
-typedef signed char  int8_t;
-typedef signed short int16_t;
-typedef signed int   int32_t;
-# else
-typedef char         int8_t;
-typedef short        int16_t;
-typedef int          int32_t;
-# endif
-typedef unsigned char  uint8_t;
-typedef unsigned short uint16_t;
-typedef unsigned int   uint32_t;
-#endif
-
-typedef int8_t     vpxs8;
-typedef uint8_t    vpxu8;
-typedef int16_t    vpxs16;
-typedef uint16_t   vpxu16;
-typedef int32_t    vpxs32;
-typedef uint32_t   vpxu32;
-typedef int32_t    vpxbool;
-
-enum {vpxfalse, vpxtrue};
-
-/*!\def OTC
-   \brief a macro suitable for declaring a constant #vpxtc*/
-/*!\def VPXTC
-   \brief printf format string suitable for printing an #vpxtc*/
-#ifdef UNICODE
-# ifdef NO_WCHAR
-#  error "no non-wchar support added yet"
-# else
-#  include <wchar.h>
-typedef wchar_t vpxtc;
-#  define OTC(str) L ## str
-#  define VPXTC "ls"
-# endif /*NO_WCHAR*/
-#else
-typedef char vpxtc;
-# define OTC(str) (vpxtc*)str
-# define VPXTC "s"
-#endif /*UNICODE*/
-/*@} end - base types*/
-
-/*!\addtogroup basetypes
-  @{*/
-/*!\def VPX64
-   \brief printf format string suitable for printing an #vpxs64*/
-#if defined(HAVE_STDINT_H)
-# define VPX64 PRId64
-typedef int64_t vpxs64;
-#elif defined(HASLONGLONG)
-# undef  PRId64
-# define PRId64 "lld"
-# define VPX64 PRId64
-typedef long long vpxs64;
-#elif defined(WIN32) || defined(_WIN32_WCE)
-# undef  PRId64
-# define PRId64 "I64d"
-# define VPX64 PRId64
-typedef __int64 vpxs64;
-typedef unsigned __int64 vpxu64;
-#elif defined(__uClinux__) && defined(CHIP_DM642)
-# include <lddk.h>
-# undef  PRId64
-# define PRId64 "lld"
-# define VPX64 PRId64
-typedef long vpxs64;
-#else
-# error "64 bit integer type undefined for this platform!"
-#endif
-#if !defined(HAVE_STDINT_H) && !defined(INT_T_DEFINED)
-typedef vpxs64 int64_t;
-typedef vpxu64 uint64_t;
-#endif
-/*!@} end - base types*/
-
-/*!\ingroup basetypes
-   \brief Common return type*/
-typedef enum
-{
-    VPX_NOT_FOUND        = -404,
-    VPX_BUFFER_EMPTY     = -202,
-    VPX_BUFFER_FULL      = -201,
-
-    VPX_CONNREFUSED      = -102,
-    VPX_TIMEDOUT         = -101,
-    VPX_WOULDBLOCK       = -100,
-
-    VPX_NET_ERROR        = -9,
-    VPX_INVALID_VERSION  = -8,
-    VPX_INPROGRESS       = -7,
-    VPX_NOT_SUPP         = -6,
-    VPX_NO_MEM           = -3,
-    VPX_INVALID_PARAMS   = -2,
-    VPX_ERROR            = -1,
-    VPX_OK               = 0,
-    VPX_DONE             = 1
-} vpxsc;
-
-#if defined(WIN32) || defined(_WIN32_WCE)
-# define DLLIMPORT __declspec(dllimport)
-# define DLLEXPORT __declspec(dllexport)
-# define DLLLOCAL
-#elif defined(LINUX)
-# define DLLIMPORT
-/*visibility attribute support is available in 3.4 and later.
-  see: http://gcc.gnu.org/wiki/Visibility for more info*/
-# if defined(__GNUC__) && ((__GNUC__<<16|(__GNUC_MINOR__&0xff)) >= (3<<16|4))
-#  define GCC_HASCLASSVISIBILITY
-# endif /*defined(__GNUC__) && __GNUC_PREREQ(3,4)*/
-# ifdef GCC_HASCLASSVISIBILITY
-#  define DLLEXPORT   __attribute__ ((visibility("default")))
-#  define DLLLOCAL __attribute__ ((visibility("hidden")))
-# else
-#  define DLLEXPORT
-#  define DLLLOCAL
-# endif /*GCC_HASCLASSVISIBILITY*/
-#endif /*platform ifdefs*/
-
-#endif /*__VPXTYPES_H__*/
-
-#undef VPXAPI
-/*!\def VPXAPI
-   \brief library calling convention/storage class attributes.
-
-   Specifies whether the function is imported through a dll
-   or is from a static library.*/
-#ifdef VPXDLL
-# ifdef VPXDLLEXPORT
-#  define VPXAPI DLLEXPORT
-# else
-#  define VPXAPI DLLIMPORT
-# endif /*VPXDLLEXPORT*/
-#else
-# define VPXAPI
-#endif /*VPXDLL*/
diff --git a/libvpx/vpx_ports/x86.h b/libvpx/vpx_ports/x86.h
index 9dd8c4b..b009c35 100644
--- a/libvpx/vpx_ports/x86.h
+++ b/libvpx/vpx_ports/x86.h
@@ -14,80 +14,79 @@
 #include <stdlib.h>
 #include "vpx_config.h"
 
-typedef enum
-{
-    VPX_CPU_UNKNOWN = -1,
-    VPX_CPU_AMD,
-    VPX_CPU_AMD_OLD,
-    VPX_CPU_CENTAUR,
-    VPX_CPU_CYRIX,
-    VPX_CPU_INTEL,
-    VPX_CPU_NEXGEN,
-    VPX_CPU_NSC,
-    VPX_CPU_RISE,
-    VPX_CPU_SIS,
-    VPX_CPU_TRANSMETA,
-    VPX_CPU_TRANSMETA_OLD,
-    VPX_CPU_UMC,
-    VPX_CPU_VIA,
+typedef enum {
+  VPX_CPU_UNKNOWN = -1,
+  VPX_CPU_AMD,
+  VPX_CPU_AMD_OLD,
+  VPX_CPU_CENTAUR,
+  VPX_CPU_CYRIX,
+  VPX_CPU_INTEL,
+  VPX_CPU_NEXGEN,
+  VPX_CPU_NSC,
+  VPX_CPU_RISE,
+  VPX_CPU_SIS,
+  VPX_CPU_TRANSMETA,
+  VPX_CPU_TRANSMETA_OLD,
+  VPX_CPU_UMC,
+  VPX_CPU_VIA,
 
-    VPX_CPU_LAST
+  VPX_CPU_LAST
 }  vpx_cpu_t;
 
-#if defined(__GNUC__) && __GNUC__
+#if defined(__GNUC__) && __GNUC__ || defined(__ANDROID__)
 #if ARCH_X86_64
 #define cpuid(func,ax,bx,cx,dx)\
-    __asm__ __volatile__ (\
-                          "cpuid           \n\t" \
-                          : "=a" (ax), "=b" (bx), "=c" (cx), "=d" (dx) \
-                          : "a"  (func));
+  __asm__ __volatile__ (\
+                        "cpuid           \n\t" \
+                        : "=a" (ax), "=b" (bx), "=c" (cx), "=d" (dx) \
+                        : "a"  (func));
 #else
 #define cpuid(func,ax,bx,cx,dx)\
-    __asm__ __volatile__ (\
-                          "mov %%ebx, %%edi   \n\t" \
-                          "cpuid              \n\t" \
-                          "xchg %%edi, %%ebx  \n\t" \
-                          : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
-                          : "a" (func));
+  __asm__ __volatile__ (\
+                        "mov %%ebx, %%edi   \n\t" \
+                        "cpuid              \n\t" \
+                        "xchg %%edi, %%ebx  \n\t" \
+                        : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
+                        : "a" (func));
 #endif
-#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
+#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) /* end __GNUC__ or __ANDROID__*/
 #if ARCH_X86_64
 #define cpuid(func,ax,bx,cx,dx)\
-    asm volatile (\
-                  "xchg %rsi, %rbx \n\t" \
-                  "cpuid           \n\t" \
-                  "movl %ebx, %edi \n\t" \
-                  "xchg %rsi, %rbx \n\t" \
-                  : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
-                  : "a"  (func));
+  asm volatile (\
+                "xchg %rsi, %rbx \n\t" \
+                "cpuid           \n\t" \
+                "movl %ebx, %edi \n\t" \
+                "xchg %rsi, %rbx \n\t" \
+                : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
+                : "a"  (func));
 #else
 #define cpuid(func,ax,bx,cx,dx)\
-    asm volatile (\
-                  "pushl %ebx       \n\t" \
-                  "cpuid            \n\t" \
-                  "movl %ebx, %edi  \n\t" \
-                  "popl %ebx        \n\t" \
-                  : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
-                  : "a" (func));
+  asm volatile (\
+                "pushl %ebx       \n\t" \
+                "cpuid            \n\t" \
+                "movl %ebx, %edi  \n\t" \
+                "popl %ebx        \n\t" \
+                : "=a" (ax), "=D" (bx), "=c" (cx), "=d" (dx) \
+                : "a" (func));
 #endif
-#else
+#else /* end __SUNPRO__ */
 #if ARCH_X86_64
 void __cpuid(int CPUInfo[4], int info_type);
 #pragma intrinsic(__cpuid)
 #define cpuid(func,a,b,c,d) do{\
-        int regs[4];\
-        __cpuid(regs,func); a=regs[0];  b=regs[1];  c=regs[2];  d=regs[3];\
-    } while(0)
+    int regs[4];\
+    __cpuid(regs,func); a=regs[0];  b=regs[1];  c=regs[2];  d=regs[3];\
+  } while(0)
 #else
 #define cpuid(func,a,b,c,d)\
-    __asm mov eax, func\
-    __asm cpuid\
-    __asm mov a, eax\
-    __asm mov b, ebx\
-    __asm mov c, ecx\
-    __asm mov d, edx
+  __asm mov eax, func\
+  __asm cpuid\
+  __asm mov a, eax\
+  __asm mov b, ebx\
+  __asm mov c, ecx\
+  __asm mov d, edx
 #endif
-#endif
+#endif /* end others */
 
 #define HAS_MMX   0x01
 #define HAS_SSE   0x02
@@ -100,47 +99,46 @@
 #endif
 
 static int
-x86_simd_caps(void)
-{
-    unsigned int flags = 0;
-    unsigned int mask = ~0;
-    unsigned int reg_eax, reg_ebx, reg_ecx, reg_edx;
-    char *env;
-    (void)reg_ebx;
+x86_simd_caps(void) {
+  unsigned int flags = 0;
+  unsigned int mask = ~0;
+  unsigned int reg_eax, reg_ebx, reg_ecx, reg_edx;
+  char *env;
+  (void)reg_ebx;
 
-    /* See if the CPU capabilities are being overridden by the environment */
-    env = getenv("VPX_SIMD_CAPS");
+  /* See if the CPU capabilities are being overridden by the environment */
+  env = getenv("VPX_SIMD_CAPS");
 
-    if (env && *env)
-        return (int)strtol(env, NULL, 0);
+  if (env && *env)
+    return (int)strtol(env, NULL, 0);
 
-    env = getenv("VPX_SIMD_CAPS_MASK");
+  env = getenv("VPX_SIMD_CAPS_MASK");
 
-    if (env && *env)
-        mask = strtol(env, NULL, 0);
+  if (env && *env)
+    mask = strtol(env, NULL, 0);
 
-    /* Ensure that the CPUID instruction supports extended features */
-    cpuid(0, reg_eax, reg_ebx, reg_ecx, reg_edx);
+  /* Ensure that the CPUID instruction supports extended features */
+  cpuid(0, reg_eax, reg_ebx, reg_ecx, reg_edx);
 
-    if (reg_eax < 1)
-        return 0;
+  if (reg_eax < 1)
+    return 0;
 
-    /* Get the standard feature flags */
-    cpuid(1, reg_eax, reg_ebx, reg_ecx, reg_edx);
+  /* Get the standard feature flags */
+  cpuid(1, reg_eax, reg_ebx, reg_ecx, reg_edx);
 
-    if (reg_edx & BIT(23)) flags |= HAS_MMX;
+  if (reg_edx & BIT(23)) flags |= HAS_MMX;
 
-    if (reg_edx & BIT(25)) flags |= HAS_SSE; /* aka xmm */
+  if (reg_edx & BIT(25)) flags |= HAS_SSE; /* aka xmm */
 
-    if (reg_edx & BIT(26)) flags |= HAS_SSE2; /* aka wmt */
+  if (reg_edx & BIT(26)) flags |= HAS_SSE2; /* aka wmt */
 
-    if (reg_ecx & BIT(0))  flags |= HAS_SSE3;
+  if (reg_ecx & BIT(0))  flags |= HAS_SSE3;
 
-    if (reg_ecx & BIT(9))  flags |= HAS_SSSE3;
+  if (reg_ecx & BIT(9))  flags |= HAS_SSSE3;
 
-    if (reg_ecx & BIT(19)) flags |= HAS_SSE4_1;
+  if (reg_ecx & BIT(19)) flags |= HAS_SSE4_1;
 
-    return flags & mask;
+  return flags & mask;
 }
 
 vpx_cpu_t vpx_x86_vendor(void);
@@ -150,21 +148,20 @@
 #pragma intrinsic(__rdtsc)
 #endif
 static unsigned int
-x86_readtsc(void)
-{
+x86_readtsc(void) {
 #if defined(__GNUC__) && __GNUC__
-    unsigned int tsc;
-    __asm__ __volatile__("rdtsc\n\t":"=a"(tsc):);
-    return tsc;
+  unsigned int tsc;
+  __asm__ __volatile__("rdtsc\n\t":"=a"(tsc):);
+  return tsc;
 #elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
-    unsigned int tsc;
-    asm volatile("rdtsc\n\t":"=a"(tsc):);
-    return tsc;
+  unsigned int tsc;
+  asm volatile("rdtsc\n\t":"=a"(tsc):);
+  return tsc;
 #else
 #if ARCH_X86_64
-    return (unsigned int)__rdtsc();
+  return (unsigned int)__rdtsc();
 #else
-    __asm  rdtsc;
+  __asm  rdtsc;
 #endif
 #endif
 }
@@ -172,45 +169,41 @@
 
 #if defined(__GNUC__) && __GNUC__
 #define x86_pause_hint()\
-    __asm__ __volatile__ ("pause \n\t")
+  __asm__ __volatile__ ("pause \n\t")
 #elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
 #define x86_pause_hint()\
-    asm volatile ("pause \n\t")
+  asm volatile ("pause \n\t")
 #else
 #if ARCH_X86_64
 #define x86_pause_hint()\
-    _mm_pause();
+  _mm_pause();
 #else
 #define x86_pause_hint()\
-    __asm pause
+  __asm pause
 #endif
 #endif
 
 #if defined(__GNUC__) && __GNUC__
 static void
-x87_set_control_word(unsigned short mode)
-{
-    __asm__ __volatile__("fldcw %0" : : "m"(*&mode));
+x87_set_control_word(unsigned short mode) {
+  __asm__ __volatile__("fldcw %0" : : "m"(*&mode));
 }
 static unsigned short
-x87_get_control_word(void)
-{
-    unsigned short mode;
-    __asm__ __volatile__("fstcw %0\n\t":"=m"(*&mode):);
+x87_get_control_word(void) {
+  unsigned short mode;
+  __asm__ __volatile__("fstcw %0\n\t":"=m"(*&mode):);
     return mode;
 }
 #elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
 static void
-x87_set_control_word(unsigned short mode)
-{
-    asm volatile("fldcw %0" : : "m"(*&mode));
+x87_set_control_word(unsigned short mode) {
+  asm volatile("fldcw %0" : : "m"(*&mode));
 }
 static unsigned short
-x87_get_control_word(void)
-{
-    unsigned short mode;
-    asm volatile("fstcw %0\n\t":"=m"(*&mode):);
-    return mode;
+x87_get_control_word(void) {
+  unsigned short mode;
+  asm volatile("fstcw %0\n\t":"=m"(*&mode):);
+  return mode;
 }
 #elif ARCH_X86_64
 /* No fldcw intrinsics on Windows x64, punt to external asm */
@@ -220,25 +213,22 @@
 #define x87_get_control_word vpx_winx64_fstcw
 #else
 static void
-x87_set_control_word(unsigned short mode)
-{
-    __asm { fldcw mode }
+x87_set_control_word(unsigned short mode) {
+  __asm { fldcw mode }
 }
 static unsigned short
-x87_get_control_word(void)
-{
-    unsigned short mode;
-    __asm { fstcw mode }
-    return mode;
+x87_get_control_word(void) {
+  unsigned short mode;
+  __asm { fstcw mode }
+  return mode;
 }
 #endif
 
 static unsigned short
-x87_set_double_precision(void)
-{
-    unsigned short mode = x87_get_control_word();
-    x87_set_control_word((mode&~0x300) | 0x200);
-    return mode;
+x87_set_double_precision(void) {
+  unsigned short mode = x87_get_control_word();
+  x87_set_control_word((mode&~0x300) | 0x200);
+  return mode;
 }
 
 
diff --git a/libvpx/vpx_ports/x86_abi_support.asm b/libvpx/vpx_ports/x86_abi_support.asm
index 0c9fe37..eccbfa3 100644
--- a/libvpx/vpx_ports/x86_abi_support.asm
+++ b/libvpx/vpx_ports/x86_abi_support.asm
@@ -78,6 +78,17 @@
 %endif
 
 
+; LIBVPX_YASM_WIN64
+; Set LIBVPX_YASM_WIN64 if output is Windows 64bit so the code will work if x64
+; or win64 is defined on the Yasm command line.
+%ifidn __OUTPUT_FORMAT__,win64
+%define LIBVPX_YASM_WIN64 1
+%elifidn __OUTPUT_FORMAT__,x64
+%define LIBVPX_YASM_WIN64 1
+%else
+%define LIBVPX_YASM_WIN64 0
+%endif
+
 ; sym()
 ; Return the proper symbol name for the target ABI.
 ;
@@ -90,7 +101,7 @@
 %define sym(x) x
 %elifidn __OUTPUT_FORMAT__,elfx32
 %define sym(x) x
-%elifidn __OUTPUT_FORMAT__,x64
+%elif LIBVPX_YASM_WIN64
 %define sym(x) x
 %else
 %define sym(x) _ %+ x
@@ -114,7 +125,7 @@
     %define PRIVATE :hidden
   %elifidn __OUTPUT_FORMAT__,elfx32
     %define PRIVATE :hidden
-  %elifidn __OUTPUT_FORMAT__,x64
+  %elif LIBVPX_YASM_WIN64
     %define PRIVATE
   %else
     %define PRIVATE :private_extern
@@ -131,7 +142,7 @@
 %else
   ; 64 bit ABI passes arguments in registers. This is a workaround to get up
   ; and running quickly. Relies on SHADOW_ARGS_TO_STACK
-  %ifidn __OUTPUT_FORMAT__,x64
+  %if LIBVPX_YASM_WIN64
     %define arg(x) [rbp+16+8*x]
   %else
     %define arg(x) [rbp-8-8*x]
@@ -230,6 +241,12 @@
   %elifidn __OUTPUT_FORMAT__,elfx32
     %define WRT_PLT wrt ..plt
     %define HIDDEN_DATA(x) x:data hidden
+  %elifidn __OUTPUT_FORMAT__,macho64
+    %ifdef CHROMIUM
+      %define HIDDEN_DATA(x) x:private_extern
+    %else
+      %define HIDDEN_DATA(x) x
+    %endif
   %else
     %define HIDDEN_DATA(x) x
   %endif
@@ -251,7 +268,7 @@
   %endm
   %define UNSHADOW_ARGS
 %else
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
   %macro SHADOW_ARGS_TO_STACK 1 ; argc
     %if %1 > 0
         mov arg(0),rcx
@@ -307,7 +324,7 @@
 ; Win64 ABI requires 16 byte stack alignment, but then pushes an 8 byte return
 ; value. Typically we follow this up with 'push rbp' - re-aligning the stack -
 ; but in some cases this is not done and unaligned movs must be used.
-%ifidn __OUTPUT_FORMAT__,x64
+%if LIBVPX_YASM_WIN64
 %macro SAVE_XMM 1-2 a
   %if %1 < 6
     %error Only xmm registers 6-15 must be preserved
diff --git a/libvpx/vpx_ports/x86_cpuid.c b/libvpx/vpx_ports/x86_cpuid.c
index ce64033..fe86cfc 100644
--- a/libvpx/vpx_ports/x86_cpuid.c
+++ b/libvpx/vpx_ports/x86_cpuid.c
@@ -11,43 +11,39 @@
 #include <string.h>
 #include "x86.h"
 
-struct cpuid_vendors
-{
-    char vendor_string[12];
-    vpx_cpu_t vendor_id;
+struct cpuid_vendors {
+  char vendor_string[12];
+  vpx_cpu_t vendor_id;
 };
 
-static struct cpuid_vendors cpuid_vendor_list[VPX_CPU_LAST] =
-{
-    { "AuthenticAMD", VPX_CPU_AMD           },
-    { "AMDisbetter!", VPX_CPU_AMD_OLD       },
-    { "CentaurHauls", VPX_CPU_CENTAUR       },
-    { "CyrixInstead", VPX_CPU_CYRIX         },
-    { "GenuineIntel", VPX_CPU_INTEL         },
-    { "NexGenDriven", VPX_CPU_NEXGEN        },
-    { "Geode by NSC", VPX_CPU_NSC           },
-    { "RiseRiseRise", VPX_CPU_RISE          },
-    { "SiS SiS SiS ", VPX_CPU_SIS           },
-    { "GenuineTMx86", VPX_CPU_TRANSMETA     },
-    { "TransmetaCPU", VPX_CPU_TRANSMETA_OLD },
-    { "UMC UMC UMC ", VPX_CPU_UMC           },
-    { "VIA VIA VIA ", VPX_CPU_VIA           },
+static struct cpuid_vendors cpuid_vendor_list[VPX_CPU_LAST] = {
+  { "AuthenticAMD", VPX_CPU_AMD           },
+  { "AMDisbetter!", VPX_CPU_AMD_OLD       },
+  { "CentaurHauls", VPX_CPU_CENTAUR       },
+  { "CyrixInstead", VPX_CPU_CYRIX         },
+  { "GenuineIntel", VPX_CPU_INTEL         },
+  { "NexGenDriven", VPX_CPU_NEXGEN        },
+  { "Geode by NSC", VPX_CPU_NSC           },
+  { "RiseRiseRise", VPX_CPU_RISE          },
+  { "SiS SiS SiS ", VPX_CPU_SIS           },
+  { "GenuineTMx86", VPX_CPU_TRANSMETA     },
+  { "TransmetaCPU", VPX_CPU_TRANSMETA_OLD },
+  { "UMC UMC UMC ", VPX_CPU_UMC           },
+  { "VIA VIA VIA ", VPX_CPU_VIA           },
 };
 
-vpx_cpu_t vpx_x86_vendor(void)
-{
-    unsigned int reg_eax;
-    unsigned int vs[3];
-    int i;
+vpx_cpu_t vpx_x86_vendor(void) {
+  unsigned int reg_eax;
+  unsigned int vs[3];
+  int i;
 
-    /* Get the Vendor String from the CPU */
-    cpuid(0, reg_eax, vs[0], vs[2], vs[1]);
+  /* Get the Vendor String from the CPU */
+  cpuid(0, reg_eax, vs[0], vs[2], vs[1]);
 
-    for (i = 0; i < VPX_CPU_LAST; i++)
-    {
-        if (strncmp ((const char *)vs, cpuid_vendor_list[i].vendor_string, 12) == 0)
-           return (cpuid_vendor_list[i].vendor_id);
-    }
+  for (i = 0; i < VPX_CPU_LAST; i++) {
+    if (strncmp((const char *)vs, cpuid_vendor_list[i].vendor_string, 12) == 0)
+      return (cpuid_vendor_list[i].vendor_id);
+  }
 
-    return VPX_CPU_UNKNOWN;
+  return VPX_CPU_UNKNOWN;
 }
diff --git a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copy_y_neon.asm b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copy_y_neon.asm
index 9189641..cc1789a 100644
--- a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copy_y_neon.asm
+++ b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copy_y_neon.asm
@@ -15,7 +15,7 @@
     REQUIRE8
     PRESERVE8
 
-    INCLUDE asm_com_offsets.asm
+    INCLUDE vpx_scale_asm_offsets.asm
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 
diff --git a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copyframe_func_neon.asm b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copyframe_func_neon.asm
index e55d076..3f17883 100644
--- a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copyframe_func_neon.asm
+++ b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copyframe_func_neon.asm
@@ -14,7 +14,7 @@
     REQUIRE8
     PRESERVE8
 
-    INCLUDE asm_com_offsets.asm
+    INCLUDE vpx_scale_asm_offsets.asm
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 
diff --git a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copysrcframe_func_neon.asm b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copysrcframe_func_neon.asm
index ec64dbc..d452ad2 100644
--- a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copysrcframe_func_neon.asm
+++ b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_copysrcframe_func_neon.asm
@@ -14,7 +14,7 @@
     REQUIRE8
     PRESERVE8
 
-    INCLUDE asm_com_offsets.asm
+    INCLUDE vpx_scale_asm_offsets.asm
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 ;Note: This function is used to copy source data in src_buffer[i] at beginning of
diff --git a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm
index ebc4242..b2eb9eb 100644
--- a/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm
+++ b/libvpx/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm
@@ -14,7 +14,7 @@
     REQUIRE8
     PRESERVE8
 
-    INCLUDE asm_com_offsets.asm
+    INCLUDE vpx_scale_asm_offsets.asm
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 ;void vp8_yv12_extend_frame_borders_neon (YV12_BUFFER_CONFIG *ybf);
diff --git a/libvpx/vpx_scale/arm/neon/yv12extend_arm.c b/libvpx/vpx_scale/arm/neon/yv12extend_arm.c
index eabd495..4535b8f 100644
--- a/libvpx/vpx_scale/arm/neon/yv12extend_arm.c
+++ b/libvpx/vpx_scale/arm/neon/yv12extend_arm.c
@@ -8,7 +8,7 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "./vpx_rtcd.h"
+#include "./vpx_scale_rtcd.h"
 
 extern void vp8_yv12_copy_frame_func_neon(struct yv12_buffer_config *src_ybc,
                                           struct yv12_buffer_config *dst_ybc);
diff --git a/libvpx/vpx_scale/generic/bicubic_scaler.c b/libvpx/vpx_scale/generic/bicubic_scaler.c
deleted file mode 100644
index c116740..0000000
--- a/libvpx/vpx_scale/generic/bicubic_scaler.c
+++ /dev/null
@@ -1,569 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-#include <float.h>
-#include <math.h>
-#include <stdio.h>
-#include "vpx_mem/vpx_mem.h"
-#include "vpxscale_arbitrary.h"
-
-#define FIXED_POINT
-
-#define MAX_IN_WIDTH        800
-#define MAX_IN_HEIGHT       600
-#define MAX_OUT_WIDTH       800
-#define MAX_OUT_HEIGHT      600
-#define MAX_OUT_DIMENSION   ((MAX_OUT_WIDTH > MAX_OUT_HEIGHT) ? \
-                             MAX_OUT_WIDTH : MAX_OUT_HEIGHT)
-
-BICUBIC_SCALER_STRUCT g_b_scaler;
-static int g_first_time = 1;
-
-#pragma DATA_SECTION(g_hbuf, "VP6_HEAP")
-#pragma DATA_ALIGN (g_hbuf, 32);
-unsigned char g_hbuf[MAX_OUT_DIMENSION];
-
-#pragma DATA_SECTION(g_hbuf_uv, "VP6_HEAP")
-#pragma DATA_ALIGN (g_hbuf_uv, 32);
-unsigned char g_hbuf_uv[MAX_OUT_DIMENSION];
-
-
-#ifdef FIXED_POINT
-static int a_i = 0.6 * 65536;
-#else
-static float a = -0.6;
-#endif
-
-#ifdef FIXED_POINT
-//         3     2
-// C0 = a*t - a*t
-//
-static short c0_fixed(unsigned int t) {
-  // put t in Q16 notation
-  unsigned short v1, v2;
-
-  // Q16
-  v1 = (a_i * t) >> 16;
-  v1 = (v1 * t) >> 16;
-
-  // Q16
-  v2 = (a_i * t) >> 16;
-  v2 = (v2 * t) >> 16;
-  v2 = (v2 * t) >> 16;
-
-  // Q12
-  return -((v1 - v2) >> 4);
-}
-
-//                     2          3
-// C1 = a*t + (3-2*a)*t  - (2-a)*t
-//
-static short c1_fixed(unsigned int t) {
-  unsigned short v1, v2, v3;
-  unsigned short two, three;
-
-  // Q16
-  v1 = (a_i * t) >> 16;
-
-  // Q13
-  two = 2 << 13;
-  v2 = two - (a_i >> 3);
-  v2 = (v2 * t) >> 16;
-  v2 = (v2 * t) >> 16;
-  v2 = (v2 * t) >> 16;
-
-  // Q13
-  three = 3 << 13;
-  v3 = three - (2 * (a_i >> 3));
-  v3 = (v3 * t) >> 16;
-  v3 = (v3 * t) >> 16;
-
-  // Q12
-  return (((v1 >> 3) - v2 + v3) >> 1);
-
-}
-
-//                 2          3
-// C2 = 1 - (3-a)*t  + (2-a)*t
-//
-static short c2_fixed(unsigned int t) {
-  unsigned short v1, v2, v3;
-  unsigned short two, three;
-
-  // Q13
-  v1 = 1 << 13;
-
-  // Q13
-  three = 3 << 13;
-  v2 = three - (a_i >> 3);
-  v2 = (v2 * t) >> 16;
-  v2 = (v2 * t) >> 16;
-
-  // Q13
-  two = 2 << 13;
-  v3 = two - (a_i >> 3);
-  v3 = (v3 * t) >> 16;
-  v3 = (v3 * t) >> 16;
-  v3 = (v3 * t) >> 16;
-
-  // Q12
-  return (v1 - v2 + v3) >> 1;
-}
-
-//                 2      3
-// C3 = a*t - 2*a*t  + a*t
-//
-static short c3_fixed(unsigned int t) {
-  int v1, v2, v3;
-
-  // Q16
-  v1 = (a_i * t) >> 16;
-
-  // Q15
-  v2 = 2 * (a_i >> 1);
-  v2 = (v2 * t) >> 16;
-  v2 = (v2 * t) >> 16;
-
-  // Q16
-  v3 = (a_i * t) >> 16;
-  v3 = (v3 * t) >> 16;
-  v3 = (v3 * t) >> 16;
-
-  // Q12
-  return ((v2 - (v1 >> 1) - (v3 >> 1)) >> 3);
-}
-#else
-//          3     2
-// C0 = -a*t + a*t
-//
-float C0(float t) {
-  return -a * t * t * t + a * t * t;
-}
-
-//                      2          3
-// C1 = -a*t + (2*a+3)*t  - (a+2)*t
-//
-float C1(float t) {
-  return -(a + 2.0f) * t * t * t + (2.0f * a + 3.0f) * t * t - a * t;
-}
-
-//                 2          3
-// C2 = 1 - (a+3)*t  + (a+2)*t
-//
-float C2(float t) {
-  return (a + 2.0f) * t * t * t - (a + 3.0f) * t * t + 1.0f;
-}
-
-//                 2      3
-// C3 = a*t - 2*a*t  + a*t
-//
-float C3(float t) {
-  return a * t * t * t - 2.0f * a * t * t + a * t;
-}
-#endif
-
-#if 0
-int compare_real_fixed() {
-  int i, errors = 0;
-  float mult = 1.0 / 10000.0;
-  unsigned int fixed_mult = mult * 4294967296;// 65536;
-  unsigned int phase_offset_int;
-  float phase_offset_real;
-
-  for (i = 0; i < 10000; i++) {
-    int fixed0, fixed1, fixed2, fixed3, fixed_total;
-    int real0, real1, real2, real3, real_total;
-
-    phase_offset_real = (float)i * mult;
-    phase_offset_int = (fixed_mult * i) >> 16;
-//      phase_offset_int = phase_offset_real * 65536;
-
-    fixed0 = c0_fixed(phase_offset_int);
-    real0 = C0(phase_offset_real) * 4096.0;
-
-    if ((abs(fixed0) > (abs(real0) + 1)) || (abs(fixed0) < (abs(real0) - 1)))
-      errors++;
-
-    fixed1 = c1_fixed(phase_offset_int);
-    real1 = C1(phase_offset_real) * 4096.0;
-
-    if ((abs(fixed1) > (abs(real1) + 1)) || (abs(fixed1) < (abs(real1) - 1)))
-      errors++;
-
-    fixed2 = c2_fixed(phase_offset_int);
-    real2 = C2(phase_offset_real) * 4096.0;
-
-    if ((abs(fixed2) > (abs(real2) + 1)) || (abs(fixed2) < (abs(real2) - 1)))
-      errors++;
-
-    fixed3 = c3_fixed(phase_offset_int);
-    real3 = C3(phase_offset_real) * 4096.0;
-
-    if ((abs(fixed3) > (abs(real3) + 1)) || (abs(fixed3) < (abs(real3) - 1)))
-      errors++;
-
-    fixed_total = fixed0 + fixed1 + fixed2 + fixed3;
-    real_total = real0 + real1 + real2 + real3;
-
-    if ((fixed_total > 4097) || (fixed_total < 4094))
-      errors++;
-
-    if ((real_total > 4097) || (real_total < 4095))
-      errors++;
-  }
-
-  return errors;
-}
-#endif
-
-// Find greatest common denominator between two integers.  Method used here is
-//  slow compared to Euclid's algorithm, but does not require any division.
-int gcd(int a, int b) {
-  // Problem with this algorithm is that if a or b = 0 this function
-  //  will never exit.  Don't want to return 0 because any computation
-  //  that was based on a common denoninator and tried to reduce by
-  //  dividing by 0 would fail.  Best solution that could be thought of
-  //  would to be fail by returing a 1;
-  if (a <= 0 || b <= 0)
-    return 1;
-
-  while (a != b) {
-    if (b > a)
-      b = b - a;
-    else {
-      int tmp = a;// swap large and
-      a = b; // small
-      b = tmp;
-    }
-  }
-
-  return b;
-}
-
-void bicubic_coefficient_init() {
-  vpx_memset(&g_b_scaler, 0, sizeof(BICUBIC_SCALER_STRUCT));
-  g_first_time = 0;
-}
-
-void bicubic_coefficient_destroy() {
-  if (!g_first_time) {
-    vpx_free(g_b_scaler.l_w);
-
-    vpx_free(g_b_scaler.l_h);
-
-    vpx_free(g_b_scaler.l_h_uv);
-
-    vpx_free(g_b_scaler.c_w);
-
-    vpx_free(g_b_scaler.c_h);
-
-    vpx_free(g_b_scaler.c_h_uv);
-
-    vpx_memset(&g_b_scaler, 0, sizeof(BICUBIC_SCALER_STRUCT));
-  }
-}
-
-// Create the coeffients that will be used for the cubic interpolation.
-//  Because scaling does not have to be equal in the vertical and horizontal
-//  regimes the phase offsets will be different.  There are 4 coefficents
-//  for each point, two on each side.  The layout is that there are the
-//  4 coefficents for each phase in the array and then the next phase.
-int bicubic_coefficient_setup(int in_width, int in_height, int out_width, int out_height) {
-  int i;
-#ifdef FIXED_POINT
-  int phase_offset_int;
-  unsigned int fixed_mult;
-  int product_val = 0;
-#else
-  float phase_offset;
-#endif
-  int gcd_w, gcd_h, gcd_h_uv, d_w, d_h, d_h_uv;
-
-  if (g_first_time)
-    bicubic_coefficient_init();
-
-
-  // check to see if the coefficents have already been set up correctly
-  if ((in_width == g_b_scaler.in_width) && (in_height == g_b_scaler.in_height)
-      && (out_width == g_b_scaler.out_width) && (out_height == g_b_scaler.out_height))
-    return 0;
-
-  g_b_scaler.in_width = in_width;
-  g_b_scaler.in_height = in_height;
-  g_b_scaler.out_width = out_width;
-  g_b_scaler.out_height = out_height;
-
-  // Don't want to allow crazy scaling, just try and prevent a catastrophic
-  //  failure here.  Want to fail after setting the member functions so if
-  //  if the scaler is called the member functions will not scale.
-  if (out_width <= 0 || out_height <= 0)
-    return -1;
-
-  // reduce in/out width and height ratios using the gcd
-  gcd_w = gcd(out_width, in_width);
-  gcd_h = gcd(out_height, in_height);
-  gcd_h_uv = gcd(out_height, in_height / 2);
-
-  // the numerator width and height are to be saved in
-  //  globals so they can be used during the scaling process
-  //  without having to be recalculated.
-  g_b_scaler.nw = out_width / gcd_w;
-  d_w = in_width / gcd_w;
-
-  g_b_scaler.nh = out_height / gcd_h;
-  d_h = in_height / gcd_h;
-
-  g_b_scaler.nh_uv = out_height / gcd_h_uv;
-  d_h_uv = (in_height / 2) / gcd_h_uv;
-
-  // allocate memory for the coefficents
-  vpx_free(g_b_scaler.l_w);
-
-  vpx_free(g_b_scaler.l_h);
-
-  vpx_free(g_b_scaler.l_h_uv);
-
-  g_b_scaler.l_w = (short *)vpx_memalign(32, out_width * 2);
-  g_b_scaler.l_h = (short *)vpx_memalign(32, out_height * 2);
-  g_b_scaler.l_h_uv = (short *)vpx_memalign(32, out_height * 2);
-
-  vpx_free(g_b_scaler.c_w);
-
-  vpx_free(g_b_scaler.c_h);
-
-  vpx_free(g_b_scaler.c_h_uv);
-
-  g_b_scaler.c_w = (short *)vpx_memalign(32, g_b_scaler.nw * 4 * 2);
-  g_b_scaler.c_h = (short *)vpx_memalign(32, g_b_scaler.nh * 4 * 2);
-  g_b_scaler.c_h_uv = (short *)vpx_memalign(32, g_b_scaler.nh_uv * 4 * 2);
-
-  g_b_scaler.hbuf = g_hbuf;
-  g_b_scaler.hbuf_uv = g_hbuf_uv;
-
-  // Set up polyphase filter taps.  This needs to be done before
-  //  the scaling because of the floating point math required.  The
-  //  coefficients are multiplied by 2^12 so that fixed point math
-  //  can be used in the main scaling loop.
-#ifdef FIXED_POINT
-  fixed_mult = (1.0 / (float)g_b_scaler.nw) * 4294967296;
-
-  product_val = 0;
-
-  for (i = 0; i < g_b_scaler.nw; i++) {
-    if (product_val > g_b_scaler.nw)
-      product_val -= g_b_scaler.nw;
-
-    phase_offset_int = (fixed_mult * product_val) >> 16;
-
-    g_b_scaler.c_w[i * 4]   = c3_fixed(phase_offset_int);
-    g_b_scaler.c_w[i * 4 + 1] = c2_fixed(phase_offset_int);
-    g_b_scaler.c_w[i * 4 + 2] = c1_fixed(phase_offset_int);
-    g_b_scaler.c_w[i * 4 + 3] = c0_fixed(phase_offset_int);
-
-    product_val += d_w;
-  }
-
-
-  fixed_mult = (1.0 / (float)g_b_scaler.nh) * 4294967296;
-
-  product_val = 0;
-
-  for (i = 0; i < g_b_scaler.nh; i++) {
-    if (product_val > g_b_scaler.nh)
-      product_val -= g_b_scaler.nh;
-
-    phase_offset_int = (fixed_mult * product_val) >> 16;
-
-    g_b_scaler.c_h[i * 4]   = c0_fixed(phase_offset_int);
-    g_b_scaler.c_h[i * 4 + 1] = c1_fixed(phase_offset_int);
-    g_b_scaler.c_h[i * 4 + 2] = c2_fixed(phase_offset_int);
-    g_b_scaler.c_h[i * 4 + 3] = c3_fixed(phase_offset_int);
-
-    product_val += d_h;
-  }
-
-  fixed_mult = (1.0 / (float)g_b_scaler.nh_uv) * 4294967296;
-
-  product_val = 0;
-
-  for (i = 0; i < g_b_scaler.nh_uv; i++) {
-    if (product_val > g_b_scaler.nh_uv)
-      product_val -= g_b_scaler.nh_uv;
-
-    phase_offset_int = (fixed_mult * product_val) >> 16;
-
-    g_b_scaler.c_h_uv[i * 4]   = c0_fixed(phase_offset_int);
-    g_b_scaler.c_h_uv[i * 4 + 1] = c1_fixed(phase_offset_int);
-    g_b_scaler.c_h_uv[i * 4 + 2] = c2_fixed(phase_offset_int);
-    g_b_scaler.c_h_uv[i * 4 + 3] = c3_fixed(phase_offset_int);
-
-    product_val += d_h_uv;
-  }
-
-#else
-
-  for (i = 0; i < g_nw; i++) {
-    phase_offset = (float)((i * d_w) % g_nw) / (float)g_nw;
-    g_c_w[i * 4]   = (C3(phase_offset) * 4096.0);
-    g_c_w[i * 4 + 1] = (C2(phase_offset) * 4096.0);
-    g_c_w[i * 4 + 2] = (C1(phase_offset) * 4096.0);
-    g_c_w[i * 4 + 3] = (C0(phase_offset) * 4096.0);
-  }
-
-  for (i = 0; i < g_nh; i++) {
-    phase_offset = (float)((i * d_h) % g_nh) / (float)g_nh;
-    g_c_h[i * 4]   = (C0(phase_offset) * 4096.0);
-    g_c_h[i * 4 + 1] = (C1(phase_offset) * 4096.0);
-    g_c_h[i * 4 + 2] = (C2(phase_offset) * 4096.0);
-    g_c_h[i * 4 + 3] = (C3(phase_offset) * 4096.0);
-  }
-
-  for (i = 0; i < g_nh_uv; i++) {
-    phase_offset = (float)((i * d_h_uv) % g_nh_uv) / (float)g_nh_uv;
-    g_c_h_uv[i * 4]   = (C0(phase_offset) * 4096.0);
-    g_c_h_uv[i * 4 + 1] = (C1(phase_offset) * 4096.0);
-    g_c_h_uv[i * 4 + 2] = (C2(phase_offset) * 4096.0);
-    g_c_h_uv[i * 4 + 3] = (C3(phase_offset) * 4096.0);
-  }
-
-#endif
-
-  // Create an array that corresponds input lines to output lines.
-  //  This doesn't require floating point math, but it does require
-  //  a division and because hardware division is not present that
-  //  is a call.
-  for (i = 0; i < out_width; i++) {
-    g_b_scaler.l_w[i] = (i * d_w) / g_b_scaler.nw;
-
-    if ((g_b_scaler.l_w[i] + 2) <= in_width)
-      g_b_scaler.max_usable_out_width = i;
-
-  }
-
-  for (i = 0; i < out_height + 1; i++) {
-    g_b_scaler.l_h[i] = (i * d_h) / g_b_scaler.nh;
-    g_b_scaler.l_h_uv[i] = (i * d_h_uv) / g_b_scaler.nh_uv;
-  }
-
-  return 0;
-}
-
-int bicubic_scale(int in_width, int in_height, int in_stride,
-                  int out_width, int out_height, int out_stride,
-                  unsigned char *input_image, unsigned char *output_image) {
-  short *RESTRICT l_w, * RESTRICT l_h;
-  short *RESTRICT c_w, * RESTRICT c_h;
-  unsigned char *RESTRICT ip, * RESTRICT op;
-  unsigned char *RESTRICT hbuf;
-  int h, w, lw, lh;
-  int temp_sum;
-  int phase_offset_w, phase_offset_h;
-
-  c_w = g_b_scaler.c_w;
-  c_h = g_b_scaler.c_h;
-
-  op = output_image;
-
-  l_w = g_b_scaler.l_w;
-  l_h = g_b_scaler.l_h;
-
-  phase_offset_h = 0;
-
-  for (h = 0; h < out_height; h++) {
-    // select the row to work on
-    lh = l_h[h];
-    ip = input_image + (in_stride * lh);
-
-    // vp8_filter the row vertically into an temporary buffer.
-    //  If the phase offset == 0 then all the multiplication
-    //  is going to result in the output equalling the input.
-    //  So instead point the temporary buffer to the input.
-    //  Also handle the boundry condition of not being able to
-    //  filter that last lines.
-    if (phase_offset_h && (lh < in_height - 2)) {
-      hbuf = g_b_scaler.hbuf;
-
-      for (w = 0; w < in_width; w++) {
-        temp_sum =  c_h[phase_offset_h * 4 + 3] * ip[w - in_stride];
-        temp_sum += c_h[phase_offset_h * 4 + 2] * ip[w];
-        temp_sum += c_h[phase_offset_h * 4 + 1] * ip[w + in_stride];
-        temp_sum += c_h[phase_offset_h * 4]   * ip[w + 2 * in_stride];
-
-        hbuf[w] = temp_sum >> 12;
-      }
-    } else
-      hbuf = ip;
-
-    // increase the phase offset for the next time around.
-    if (++phase_offset_h >= g_b_scaler.nh)
-      phase_offset_h = 0;
-
-    // now filter and expand it horizontally into the final
-    //  output buffer
-    phase_offset_w = 0;
-
-    for (w = 0; w < out_width; w++) {
-      // get the index to use to expand the image
-      lw = l_w[w];
-
-      temp_sum =  c_w[phase_offset_w * 4]   * hbuf[lw - 1];
-      temp_sum += c_w[phase_offset_w * 4 + 1] * hbuf[lw];
-      temp_sum += c_w[phase_offset_w * 4 + 2] * hbuf[lw + 1];
-      temp_sum += c_w[phase_offset_w * 4 + 3] * hbuf[lw + 2];
-      temp_sum = temp_sum >> 12;
-
-      if (++phase_offset_w >= g_b_scaler.nw)
-        phase_offset_w = 0;
-
-      // boundry conditions
-      if ((lw + 2) >= in_width)
-        temp_sum = hbuf[lw];
-
-      if (lw == 0)
-        temp_sum = hbuf[0];
-
-      op[w] = temp_sum;
-    }
-
-    op += out_stride;
-  }
-
-  return 0;
-}
-
-void bicubic_scale_frame_reset() {
-  g_b_scaler.out_width = 0;
-  g_b_scaler.out_height = 0;
-}
-
-void bicubic_scale_frame(YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
-                         int new_width, int new_height) {
-
-  dst->y_width = new_width;
-  dst->y_height = new_height;
-  dst->uv_width = new_width / 2;
-  dst->uv_height = new_height / 2;
-
-  dst->y_stride = dst->y_width;
-  dst->uv_stride = dst->uv_width;
-
-  bicubic_scale(src->y_width, src->y_height, src->y_stride,
-                new_width, new_height, dst->y_stride,
-                src->y_buffer, dst->y_buffer);
-
-  bicubic_scale(src->uv_width, src->uv_height, src->uv_stride,
-                new_width / 2, new_height / 2, dst->uv_stride,
-                src->u_buffer, dst->u_buffer);
-
-  bicubic_scale(src->uv_width, src->uv_height, src->uv_stride,
-                new_width / 2, new_height / 2, dst->uv_stride,
-                src->v_buffer, dst->v_buffer);
-}
diff --git a/libvpx/vpx_scale/generic/gen_scalers.c b/libvpx/vpx_scale/generic/gen_scalers.c
index 60c21fb..5f355c5 100644
--- a/libvpx/vpx_scale/generic/gen_scalers.c
+++ b/libvpx/vpx_scale/generic/gen_scalers.c
@@ -9,7 +9,7 @@
  */
 
 
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vpx_mem/vpx_mem.h"
 /****************************************************************************
 *  Imports
@@ -17,688 +17,6 @@
 
 /****************************************************************************
  *
- *  ROUTINE       : vp8_horizontal_line_4_5_scale_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to source data.
- *                  unsigned int source_width    : Stride of source.
- *                  unsigned char *dest         : Pointer to destination data.
- *                  unsigned int dest_width      : Stride of destination (NOT USED).
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Copies horizontal line of pixels from source to
- *                  destination scaling up by 4 to 5.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source,
-                                     unsigned int source_width,
-                                     unsigned char *dest,
-                                     unsigned int dest_width) {
-  unsigned i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-  const unsigned char *src = source;
-
-  (void) dest_width;
-
-  for (i = 0; i < source_width - 4; i += 4) {
-    a = src[0];
-    b = src[1];
-    des [0] = (unsigned char) a;
-    des [1] = (unsigned char)((a * 51 + 205 * b + 128) >> 8);
-    c = src[2] * 154;
-    a = src[3];
-    des [2] = (unsigned char)((b * 102 + c + 128) >> 8);
-    des [3] = (unsigned char)((c + 102 * a + 128) >> 8);
-    b = src[4];
-    des [4] = (unsigned char)((a * 205 + 51 * b + 128) >> 8);
-
-    src += 4;
-    des += 5;
-  }
-
-  a = src[0];
-  b = src[1];
-  des [0] = (unsigned char)(a);
-  des [1] = (unsigned char)((a * 51 + 205 * b + 128) >> 8);
-  c = src[2] * 154;
-  a = src[3];
-  des [2] = (unsigned char)((b * 102 + c + 128) >> 8);
-  des [3] = (unsigned char)((c + 102 * a + 128) >> 8);
-  des [4] = (unsigned char)(a);
-
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_vertical_band_4_5_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales vertical band of pixels by scale 4 to 5. The
- *                  height of the band scaled is 4-pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band.
- *
- ****************************************************************************/
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest,
-                                   unsigned int dest_pitch,
-                                   unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c, d;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; i++) {
-    a = des [0];
-    b = des [dest_pitch];
-
-    des[dest_pitch] = (unsigned char)((a * 51 + 205 * b + 128) >> 8);
-
-    c = des[dest_pitch * 2] * 154;
-    d = des[dest_pitch * 3];
-
-    des [dest_pitch * 2] = (unsigned char)((b * 102 + c + 128) >> 8);
-    des [dest_pitch * 3] = (unsigned char)((c + 102 * d + 128) >> 8);
-
-    /* First line in next band */
-    a = des [dest_pitch * 5];
-    des [dest_pitch * 4] = (unsigned char)((d * 205 + 51 * a + 128) >> 8);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_last_vertical_band_4_5_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales last vertical band of pixels by scale 4 to 5. The
- *                  height of the band scaled is 4-pixels.
- *
- *  SPECIAL NOTES : The routine does not have available the first line of
- *                  the band below the current band, since this is the
- *                  last band.
- *
- ****************************************************************************/
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest,
-                                        unsigned int dest_pitch,
-                                        unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c, d;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; ++i) {
-    a = des[0];
-    b = des[dest_pitch];
-
-    des[dest_pitch] = (unsigned char)((a * 51 + 205 * b + 128) >> 8);
-
-    c = des[dest_pitch * 2] * 154;
-    d = des[dest_pitch * 3];
-
-    des [dest_pitch * 2] = (unsigned char)((b * 102 + c + 128) >> 8);
-    des [dest_pitch * 3] = (unsigned char)((c + 102 * d + 128) >> 8);
-
-    /* No other line for interplation of this line, so .. */
-    des[dest_pitch * 4] = (unsigned char) d;
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_horizontal_line_2_3_scale_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to source data.
- *                  unsigned int source_width    : Stride of source.
- *                  unsigned char *dest         : Pointer to destination data.
- *                  unsigned int dest_width      : Stride of destination (NOT USED).
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Copies horizontal line of pixels from source to
- *                  destination scaling up by 2 to 3.
- *
- *  SPECIAL NOTES : None.
- *
- *
- ****************************************************************************/
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source,
-                                     unsigned int source_width,
-                                     unsigned char *dest,
-                                     unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-  const unsigned char *src = source;
-
-  (void) dest_width;
-
-  for (i = 0; i < source_width - 2; i += 2) {
-    a = src[0];
-    b = src[1];
-    c = src[2];
-
-    des [0] = (unsigned char)(a);
-    des [1] = (unsigned char)((a * 85 + 171 * b + 128) >> 8);
-    des [2] = (unsigned char)((b * 171 + 85 * c + 128) >> 8);
-
-    src += 2;
-    des += 3;
-  }
-
-  a = src[0];
-  b = src[1];
-  des [0] = (unsigned char)(a);
-  des [1] = (unsigned char)((a * 85 + 171 * b + 128) >> 8);
-  des [2] = (unsigned char)(b);
-}
-
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_vertical_band_2_3_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales vertical band of pixels by scale 2 to 3. The
- *                  height of the band scaled is 2-pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band.
- *
- ****************************************************************************/
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest,
-                                   unsigned int dest_pitch,
-                                   unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; i++) {
-    a = des [0];
-    b = des [dest_pitch];
-    c = des[dest_pitch * 3];
-    des [dest_pitch  ] = (unsigned char)((a * 85 + 171 * b + 128) >> 8);
-    des [dest_pitch * 2] = (unsigned char)((b * 171 + 85 * c + 128) >> 8);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_last_vertical_band_2_3_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales last vertical band of pixels by scale 2 to 3. The
- *                  height of the band scaled is 2-pixels.
- *
- *  SPECIAL NOTES : The routine does not have available the first line of
- *                  the band below the current band, since this is the
- *                  last band.
- *
- ****************************************************************************/
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest,
-                                        unsigned int dest_pitch,
-                                        unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; ++i) {
-    a = des [0];
-    b = des [dest_pitch];
-
-    des [dest_pitch  ] = (unsigned char)((a * 85 + 171 * b + 128) >> 8);
-    des [dest_pitch * 2] = (unsigned char)(b);
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_horizontal_line_3_5_scale_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to source data.
- *                  unsigned int source_width    : Stride of source.
- *                  unsigned char *dest         : Pointer to destination data.
- *                  unsigned int dest_width      : Stride of destination (NOT USED).
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Copies horizontal line of pixels from source to
- *                  destination scaling up by 3 to 5.
- *
- *  SPECIAL NOTES : None.
- *
- *
- ****************************************************************************/
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source,
-                                     unsigned int source_width,
-                                     unsigned char *dest,
-                                     unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-  const unsigned char *src = source;
-
-  (void) dest_width;
-
-  for (i = 0; i < source_width - 3; i += 3) {
-    a = src[0];
-    b = src[1];
-    des [0] = (unsigned char)(a);
-    des [1] = (unsigned char)((a * 102 + 154 * b + 128) >> 8);
-
-    c = src[2];
-    des [2] = (unsigned char)((b * 205 + c * 51 + 128) >> 8);
-    des [3] = (unsigned char)((b * 51 + c * 205 + 128) >> 8);
-
-    a = src[3];
-    des [4] = (unsigned char)((c * 154 + a * 102 + 128) >> 8);
-
-    src += 3;
-    des += 5;
-  }
-
-  a = src[0];
-  b = src[1];
-  des [0] = (unsigned char)(a);
-
-  des [1] = (unsigned char)((a * 102 + 154 * b + 128) >> 8);
-  c = src[2];
-  des [2] = (unsigned char)((b * 205 + c * 51 + 128) >> 8);
-  des [3] = (unsigned char)((b * 51 + c * 205 + 128) >> 8);
-
-  des [4] = (unsigned char)(c);
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_vertical_band_3_5_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales vertical band of pixels by scale 3 to 5. The
- *                  height of the band scaled is 3-pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band.
- *
- ****************************************************************************/
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest,
-                                   unsigned int dest_pitch,
-                                   unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; i++) {
-    a = des [0];
-    b = des [dest_pitch];
-    des [dest_pitch] = (unsigned char)((a * 102 + 154 * b + 128) >> 8);
-
-    c = des[dest_pitch * 2];
-    des [dest_pitch * 2] = (unsigned char)((b * 205 + c * 51 + 128) >> 8);
-    des [dest_pitch * 3] = (unsigned char)((b * 51 + c * 205 + 128) >> 8);
-
-    /* First line in next band... */
-    a = des [dest_pitch * 5];
-    des [dest_pitch * 4] = (unsigned char)((c * 154 + a * 102 + 128) >> 8);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_last_vertical_band_3_5_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales last vertical band of pixels by scale 3 to 5. The
- *                  height of the band scaled is 3-pixels.
- *
- *  SPECIAL NOTES : The routine does not have available the first line of
- *                  the band below the current band, since this is the
- *                  last band.
- *
- ****************************************************************************/
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest,
-                                        unsigned int dest_pitch,
-                                        unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; ++i) {
-    a = des [0];
-    b = des [dest_pitch];
-
-    des [ dest_pitch ] = (unsigned char)((a * 102 + 154 * b + 128) >> 8);
-
-    c = des[dest_pitch * 2];
-    des [dest_pitch * 2] = (unsigned char)((b * 205 + c * 51 + 128) >> 8);
-    des [dest_pitch * 3] = (unsigned char)((b * 51 + c * 205 + 128) >> 8);
-
-    /* No other line for interplation of this line, so .. */
-    des [ dest_pitch * 4 ] = (unsigned char)(c);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_horizontal_line_3_4_scale_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to source data.
- *                  unsigned int source_width    : Stride of source.
- *                  unsigned char *dest         : Pointer to destination data.
- *                  unsigned int dest_width      : Stride of destination (NOT USED).
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Copies horizontal line of pixels from source to
- *                  destination scaling up by 3 to 4.
- *
- *  SPECIAL NOTES : None.
- *
- *
- ****************************************************************************/
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source,
-                                     unsigned int source_width,
-                                     unsigned char *dest,
-                                     unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-  const unsigned char *src = source;
-
-  (void) dest_width;
-
-  for (i = 0; i < source_width - 3; i += 3) {
-    a = src[0];
-    b = src[1];
-    des [0] = (unsigned char)(a);
-    des [1] = (unsigned char)((a * 64 + b * 192 + 128) >> 8);
-
-    c = src[2];
-    des [2] = (unsigned char)((b + c + 1) >> 1);
-
-    a = src[3];
-    des [3] = (unsigned char)((c * 192 + a * 64 + 128) >> 8);
-
-    src += 3;
-    des += 4;
-  }
-
-  a = src[0];
-  b = src[1];
-  des [0] = (unsigned char)(a);
-  des [1] = (unsigned char)((a * 64 + b * 192 + 128) >> 8);
-
-  c = src[2];
-  des [2] = (unsigned char)((b + c + 1) >> 1);
-  des [3] = (unsigned char)(c);
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_vertical_band_3_4_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales vertical band of pixels by scale 3 to 4. The
- *                  height of the band scaled is 3-pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band.
- *
- ****************************************************************************/
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest,
-                                   unsigned int dest_pitch,
-                                   unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; i++) {
-    a = des [0];
-    b = des [dest_pitch];
-    des [dest_pitch]   = (unsigned char)((a * 64 + b * 192 + 128) >> 8);
-
-    c = des[dest_pitch * 2];
-    des [dest_pitch * 2] = (unsigned char)((b + c + 1) >> 1);
-
-    /* First line in next band... */
-    a = des [dest_pitch * 4];
-    des [dest_pitch * 3] = (unsigned char)((c * 192 + a * 64 + 128) >> 8);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_last_vertical_band_3_4_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales last vertical band of pixels by scale 3 to 4. The
- *                  height of the band scaled is 3-pixels.
- *
- *  SPECIAL NOTES : The routine does not have available the first line of
- *                  the band below the current band, since this is the
- *                  last band.
- *
- ****************************************************************************/
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest,
-                                        unsigned int dest_pitch,
-                                        unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b, c;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; ++i) {
-    a = des [0];
-    b = des [dest_pitch];
-
-    des [dest_pitch]   = (unsigned char)((a * 64 + b * 192 + 128) >> 8);
-
-    c = des[dest_pitch * 2];
-    des [dest_pitch * 2] = (unsigned char)((b + c + 1) >> 1);
-
-    /* No other line for interplation of this line, so .. */
-    des [dest_pitch * 3] = (unsigned char)(c);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_horizontal_line_1_2_scale_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to source data.
- *                  unsigned int source_width    : Stride of source.
- *                  unsigned char *dest         : Pointer to destination data.
- *                  unsigned int dest_width      : Stride of destination (NOT USED).
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Copies horizontal line of pixels from source to
- *                  destination scaling up by 1 to 2.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source,
-                                     unsigned int source_width,
-                                     unsigned char *dest,
-                                     unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b;
-  unsigned char *des = dest;
-  const unsigned char *src = source;
-
-  (void) dest_width;
-
-  for (i = 0; i < source_width - 1; i += 1) {
-    a = src[0];
-    b = src[1];
-    des [0] = (unsigned char)(a);
-    des [1] = (unsigned char)((a + b + 1) >> 1);
-    src += 1;
-    des += 2;
-  }
-
-  a = src[0];
-  des [0] = (unsigned char)(a);
-  des [1] = (unsigned char)(a);
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_vertical_band_1_2_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales vertical band of pixels by scale 1 to 2. The
- *                  height of the band scaled is 1-pixel.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band.
- *
- ****************************************************************************/
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest,
-                                   unsigned int dest_pitch,
-                                   unsigned int dest_width) {
-  unsigned int i;
-  unsigned int a, b;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; i++) {
-    a = des [0];
-    b = des [dest_pitch * 2];
-
-    des[dest_pitch] = (unsigned char)((a + b + 1) >> 1);
-
-    des++;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_last_vertical_band_1_2_scale_c
- *
- *  INPUTS        : unsigned char *dest    : Pointer to destination data.
- *                  unsigned int dest_pitch : Stride of destination data.
- *                  unsigned int dest_width : Width of destination data.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Scales last vertical band of pixels by scale 1 to 2. The
- *                  height of the band scaled is 1-pixel.
- *
- *  SPECIAL NOTES : The routine does not have available the first line of
- *                  the band below the current band, since this is the
- *                  last band.
- *
- ****************************************************************************/
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest,
-                                        unsigned int dest_pitch,
-                                        unsigned int dest_width) {
-  unsigned int i;
-  unsigned char *des = dest;
-
-  for (i = 0; i < dest_width; ++i) {
-    des[dest_pitch] = des[0];
-    des++;
-  }
-}
-
-
-
-
-
-/****************************************************************************
- *
- *  ROUTINE       : vp8_horizontal_line_4_5_scale_c
  *
  *  INPUTS        : const unsigned char *source : Pointer to source data.
  *                  unsigned int source_width    : Stride of source.
diff --git a/libvpx/vpx_scale/generic/vpx_scale.c b/libvpx/vpx_scale/generic/vpx_scale.c
new file mode 100644
index 0000000..8044d2a
--- /dev/null
+++ b/libvpx/vpx_scale/generic/vpx_scale.c
@@ -0,0 +1,530 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/****************************************************************************
+ *
+ *   Module Title :     scale.c
+ *
+ *   Description  :     Image scaling functions.
+ *
+ ***************************************************************************/
+
+/****************************************************************************
+*  Header Files
+****************************************************************************/
+#include "./vpx_scale_rtcd.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/yv12config.h"
+
+typedef struct {
+  int     expanded_frame_width;
+  int     expanded_frame_height;
+
+  int HScale;
+  int HRatio;
+  int VScale;
+  int VRatio;
+
+  YV12_BUFFER_CONFIG *src_yuv_config;
+  YV12_BUFFER_CONFIG *dst_yuv_config;
+
+} SCALE_VARS;
+
+/****************************************************************************
+ *
+ *  ROUTINE       : scale1d_2t1_i
+ *
+ *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
+ *                  int source_step              : Number of pixels to step on in source.
+ *                  unsigned int source_scale    : Scale for source (UNUSED).
+ *                  unsigned int source_length   : Length of source (UNUSED).
+ *                  unsigned char *dest         : Pointer to output data array.
+ *                  int dest_step                : Number of pixels to step on in destination.
+ *                  unsigned int dest_scale      : Scale for destination (UNUSED).
+ *                  unsigned int dest_length     : Length of destination.
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Performs 2-to-1 interpolated scaling.
+ *
+ *  SPECIAL NOTES : None.
+ *
+ ****************************************************************************/
+static
+void scale1d_2t1_i
+(
+  const unsigned char *source,
+  int source_step,
+  unsigned int source_scale,
+  unsigned int source_length,
+  unsigned char *dest,
+  int dest_step,
+  unsigned int dest_scale,
+  unsigned int dest_length
+) {
+  unsigned int i, j;
+  unsigned int temp;
+  int source_pitch = source_step;
+  (void) source_length;
+  (void) source_scale;
+  (void) dest_scale;
+
+  source_step *= 2;
+  dest[0] = source[0];
+
+  for (i = dest_step, j = source_step; i < dest_length * dest_step; i += dest_step, j += source_step) {
+    temp = 8;
+    temp += 3 * source[j - source_pitch];
+    temp += 10 * source[j];
+    temp += 3 * source[j + source_pitch];
+    temp >>= 4;
+    dest[i] = (char)(temp);
+  }
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : scale1d_2t1_ps
+ *
+ *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
+ *                  int source_step              : Number of pixels to step on in source.
+ *                  unsigned int source_scale    : Scale for source (UNUSED).
+ *                  unsigned int source_length   : Length of source (UNUSED).
+ *                  unsigned char *dest         : Pointer to output data array.
+ *                  int dest_step                : Number of pixels to step on in destination.
+ *                  unsigned int dest_scale      : Scale for destination (UNUSED).
+ *                  unsigned int dest_length     : Length of destination.
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Performs 2-to-1 point subsampled scaling.
+ *
+ *  SPECIAL NOTES : None.
+ *
+ ****************************************************************************/
+static
+void scale1d_2t1_ps
+(
+  const unsigned char *source,
+  int source_step,
+  unsigned int source_scale,
+  unsigned int source_length,
+  unsigned char *dest,
+  int dest_step,
+  unsigned int dest_scale,
+  unsigned int dest_length
+) {
+  unsigned int i, j;
+
+  (void) source_length;
+  (void) source_scale;
+  (void) dest_scale;
+
+  source_step *= 2;
+  j = 0;
+
+  for (i = 0; i < dest_length * dest_step; i += dest_step, j += source_step)
+    dest[i] = source[j];
+}
+/****************************************************************************
+ *
+ *  ROUTINE       : scale1d_c
+ *
+ *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
+ *                  int source_step              : Number of pixels to step on in source.
+ *                  unsigned int source_scale    : Scale for source.
+ *                  unsigned int source_length   : Length of source (UNUSED).
+ *                  unsigned char *dest         : Pointer to output data array.
+ *                  int dest_step                : Number of pixels to step on in destination.
+ *                  unsigned int dest_scale      : Scale for destination.
+ *                  unsigned int dest_length     : Length of destination.
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Performs linear interpolation in one dimension.
+ *
+ *  SPECIAL NOTES : None.
+ *
+ ****************************************************************************/
+static
+void scale1d_c
+(
+  const unsigned char *source,
+  int source_step,
+  unsigned int source_scale,
+  unsigned int source_length,
+  unsigned char *dest,
+  int dest_step,
+  unsigned int dest_scale,
+  unsigned int dest_length
+) {
+  unsigned int i;
+  unsigned int round_value = dest_scale / 2;
+  unsigned int left_modifier = dest_scale;
+  unsigned int right_modifier = 0;
+  unsigned char left_pixel = *source;
+  unsigned char right_pixel = *(source + source_step);
+
+  (void) source_length;
+
+  /* These asserts are needed if there are boundary issues... */
+  /*assert ( dest_scale > source_scale );*/
+  /*assert ( (source_length-1) * dest_scale >= (dest_length-1) * source_scale );*/
+
+  for (i = 0; i < dest_length * dest_step; i += dest_step) {
+    dest[i] = (char)((left_modifier * left_pixel + right_modifier * right_pixel + round_value) / dest_scale);
+
+    right_modifier += source_scale;
+
+    while (right_modifier > dest_scale) {
+      right_modifier -= dest_scale;
+      source += source_step;
+      left_pixel = *source;
+      right_pixel = *(source + source_step);
+    }
+
+    left_modifier = dest_scale - right_modifier;
+  }
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : Scale2D
+ *
+ *  INPUTS        : const unsigned char *source  : Pointer to data to be scaled.
+ *                  int source_pitch              : Stride of source image.
+ *                  unsigned int source_width     : Width of input image.
+ *                  unsigned int source_height    : Height of input image.
+ *                  unsigned char *dest          : Pointer to output data array.
+ *                  int dest_pitch                : Stride of destination image.
+ *                  unsigned int dest_width       : Width of destination image.
+ *                  unsigned int dest_height      : Height of destination image.
+ *                  unsigned char *temp_area      : Pointer to temp work area.
+ *                  unsigned char temp_area_height : Height of temp work area.
+ *                  unsigned int hscale          : Horizontal scale factor numerator.
+ *                  unsigned int hratio          : Horizontal scale factor denominator.
+ *                  unsigned int vscale          : Vertical scale factor numerator.
+ *                  unsigned int vratio          : Vertical scale factor denominator.
+ *                  unsigned int interlaced      : Interlace flag.
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Performs 2-tap linear interpolation in two dimensions.
+ *
+ *  SPECIAL NOTES : Expansion is performed one band at a time to help with
+ *                  caching.
+ *
+ ****************************************************************************/
+static
+void Scale2D
+(
+  /*const*/
+  unsigned char *source,
+  int source_pitch,
+  unsigned int source_width,
+  unsigned int source_height,
+  unsigned char *dest,
+  int dest_pitch,
+  unsigned int dest_width,
+  unsigned int dest_height,
+  unsigned char *temp_area,
+  unsigned char temp_area_height,
+  unsigned int hscale,
+  unsigned int hratio,
+  unsigned int vscale,
+  unsigned int vratio,
+  unsigned int interlaced
+) {
+  /*unsigned*/
+  int i, j, k;
+  int bands;
+  int dest_band_height;
+  int source_band_height;
+
+  typedef void (*Scale1D)(const unsigned char * source, int source_step, unsigned int source_scale, unsigned int source_length,
+                          unsigned char * dest, int dest_step, unsigned int dest_scale, unsigned int dest_length);
+
+  Scale1D Scale1Dv = scale1d_c;
+  Scale1D Scale1Dh = scale1d_c;
+
+  void (*horiz_line_scale)(const unsigned char *, unsigned int, unsigned char *, unsigned int) = NULL;
+  void (*vert_band_scale)(unsigned char *, unsigned int, unsigned char *, unsigned int, unsigned int) = NULL;
+
+  int ratio_scalable = 1;
+  int interpolation = 0;
+
+  unsigned char *source_base; /* = (unsigned char *) ((source_pitch >= 0) ? source : (source + ((source_height-1) * source_pitch))); */
+  unsigned char *line_src;
+
+
+  source_base = (unsigned char *)source;
+
+  if (source_pitch < 0) {
+    int offset;
+
+    offset = (source_height - 1);
+    offset *= source_pitch;
+
+    source_base += offset;
+  }
+
+  /* find out the ratio for each direction */
+  switch (hratio * 10 / hscale) {
+    case 8:
+      /* 4-5 Scale in Width direction */
+      horiz_line_scale = vp8_horizontal_line_5_4_scale;
+      break;
+    case 6:
+      /* 3-5 Scale in Width direction */
+      horiz_line_scale = vp8_horizontal_line_5_3_scale;
+      break;
+    case 5:
+      /* 1-2 Scale in Width direction */
+      horiz_line_scale = vp8_horizontal_line_2_1_scale;
+      break;
+    default:
+      /* The ratio is not acceptable now */
+      /* throw("The ratio is not acceptable for now!"); */
+      ratio_scalable = 0;
+      break;
+  }
+
+  switch (vratio * 10 / vscale) {
+    case 8:
+      /* 4-5 Scale in vertical direction */
+      vert_band_scale     = vp8_vertical_band_5_4_scale;
+      source_band_height  = 5;
+      dest_band_height    = 4;
+      break;
+    case 6:
+      /* 3-5 Scale in vertical direction */
+      vert_band_scale     = vp8_vertical_band_5_3_scale;
+      source_band_height  = 5;
+      dest_band_height    = 3;
+      break;
+    case 5:
+      /* 1-2 Scale in vertical direction */
+
+      if (interlaced) {
+        /* if the content is interlaced, point sampling is used */
+        vert_band_scale     = vp8_vertical_band_2_1_scale;
+      } else {
+
+        interpolation = 1;
+        /* if the content is progressive, interplo */
+        vert_band_scale     = vp8_vertical_band_2_1_scale_i;
+
+      }
+
+      source_band_height  = 2;
+      dest_band_height    = 1;
+      break;
+    default:
+      /* The ratio is not acceptable now */
+      /* throw("The ratio is not acceptable for now!"); */
+      ratio_scalable = 0;
+      break;
+  }
+
+  if (ratio_scalable) {
+    if (source_height == dest_height) {
+      /* for each band of the image */
+      for (k = 0; k < (int)dest_height; k++) {
+        horiz_line_scale(source, source_width, dest, dest_width);
+        source += source_pitch;
+        dest   += dest_pitch;
+      }
+
+      return;
+    }
+
+    if (interpolation) {
+      if (source < source_base)
+        source = source_base;
+
+      horiz_line_scale(source, source_width, temp_area, dest_width);
+    }
+
+    for (k = 0; k < (int)(dest_height + dest_band_height - 1) / dest_band_height; k++) {
+      /* scale one band horizontally */
+      for (i = 0; i < source_band_height; i++) {
+        /* Trap case where we could read off the base of the source buffer */
+
+        line_src = (unsigned char *)source + i * source_pitch;
+
+        if (line_src < source_base)
+          line_src = source_base;
+
+        horiz_line_scale(line_src, source_width,
+                         temp_area + (i + 1)*dest_pitch, dest_width);
+      }
+
+      /* Vertical scaling is in place */
+      vert_band_scale(temp_area + dest_pitch, dest_pitch, dest, dest_pitch, dest_width);
+
+      if (interpolation)
+        vpx_memcpy(temp_area, temp_area + source_band_height * dest_pitch, dest_width);
+
+      /* Next band... */
+      source += (unsigned long) source_band_height  * source_pitch;
+      dest   += (unsigned long) dest_band_height * dest_pitch;
+    }
+
+    return;
+  }
+
+  if (hscale == 2 && hratio == 1)
+    Scale1Dh = scale1d_2t1_ps;
+
+  if (vscale == 2 && vratio == 1) {
+    if (interlaced)
+      Scale1Dv = scale1d_2t1_ps;
+    else
+      Scale1Dv = scale1d_2t1_i;
+  }
+
+  if (source_height == dest_height) {
+    /* for each band of the image */
+    for (k = 0; k < (int)dest_height; k++) {
+      Scale1Dh(source, 1, hscale, source_width + 1, dest, 1, hratio, dest_width);
+      source += source_pitch;
+      dest   += dest_pitch;
+    }
+
+    return;
+  }
+
+  if (dest_height > source_height) {
+    dest_band_height   = temp_area_height - 1;
+    source_band_height = dest_band_height * source_height / dest_height;
+  } else {
+    source_band_height = temp_area_height - 1;
+    dest_band_height   = source_band_height * vratio / vscale;
+  }
+
+  /* first row needs to be done so that we can stay one row ahead for vertical zoom */
+  Scale1Dh(source, 1, hscale, source_width + 1, temp_area, 1, hratio, dest_width);
+
+  /* for each band of the image */
+  bands = (dest_height + dest_band_height - 1) / dest_band_height;
+
+  for (k = 0; k < bands; k++) {
+    /* scale one band horizontally */
+    for (i = 1; i < source_band_height + 1; i++) {
+      if (k * source_band_height + i < (int) source_height) {
+        Scale1Dh(source + i * source_pitch, 1, hscale, source_width + 1,
+                 temp_area + i * dest_pitch, 1, hratio, dest_width);
+      } else { /*  Duplicate the last row */
+        /* copy temp_area row 0 over from last row in the past */
+        vpx_memcpy(temp_area + i * dest_pitch, temp_area + (i - 1)*dest_pitch, dest_pitch);
+      }
+    }
+
+    /* scale one band vertically */
+    for (j = 0; j < (int)dest_width; j++) {
+      Scale1Dv(&temp_area[j], dest_pitch, vscale, source_band_height + 1,
+               &dest[j], dest_pitch, vratio, dest_band_height);
+    }
+
+    /* copy temp_area row 0 over from last row in the past */
+    vpx_memcpy(temp_area, temp_area + source_band_height * dest_pitch, dest_pitch);
+
+    /* move to the next band */
+    source += source_band_height * source_pitch;
+    dest   += dest_band_height * dest_pitch;
+  }
+}
+
+/****************************************************************************
+ *
+ *  ROUTINE       : vpx_scale_frame
+ *
+ *  INPUTS        : YV12_BUFFER_CONFIG *src       : Pointer to frame to be scaled.
+ *                  YV12_BUFFER_CONFIG *dst       : Pointer to buffer to hold scaled frame.
+ *                  unsigned char *temp_area      : Pointer to temp work area.
+ *                  unsigned char temp_area_height : Height of temp work area.
+ *                  unsigned int hscale          : Horizontal scale factor numerator.
+ *                  unsigned int hratio          : Horizontal scale factor denominator.
+ *                  unsigned int vscale          : Vertical scale factor numerator.
+ *                  unsigned int vratio          : Vertical scale factor denominator.
+ *                  unsigned int interlaced      : Interlace flag.
+ *
+ *  OUTPUTS       : None.
+ *
+ *  RETURNS       : void
+ *
+ *  FUNCTION      : Performs 2-tap linear interpolation in two dimensions.
+ *
+ *  SPECIAL NOTES : Expansion is performed one band at a time to help with
+ *                  caching.
+ *
+ ****************************************************************************/
+void vpx_scale_frame
+(
+  YV12_BUFFER_CONFIG *src,
+  YV12_BUFFER_CONFIG *dst,
+  unsigned char *temp_area,
+  unsigned char temp_height,
+  unsigned int hscale,
+  unsigned int hratio,
+  unsigned int vscale,
+  unsigned int vratio,
+  unsigned int interlaced
+) {
+  int i;
+  int dw = (hscale - 1 + src->y_width * hratio) / hscale;
+  int dh = (vscale - 1 + src->y_height * vratio) / vscale;
+
+  /* call our internal scaling routines!! */
+  Scale2D((unsigned char *) src->y_buffer, src->y_stride, src->y_width, src->y_height,
+          (unsigned char *) dst->y_buffer, dst->y_stride, dw, dh,
+          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
+
+  if (dw < (int)dst->y_width)
+    for (i = 0; i < dh; i++)
+      vpx_memset(dst->y_buffer + i * dst->y_stride + dw - 1, dst->y_buffer[i * dst->y_stride + dw - 2], dst->y_width - dw + 1);
+
+  if (dh < (int)dst->y_height)
+    for (i = dh - 1; i < (int)dst->y_height; i++)
+      vpx_memcpy(dst->y_buffer + i * dst->y_stride, dst->y_buffer + (dh - 2) * dst->y_stride, dst->y_width + 1);
+
+  Scale2D((unsigned char *) src->u_buffer, src->uv_stride, src->uv_width, src->uv_height,
+          (unsigned char *) dst->u_buffer, dst->uv_stride, dw / 2, dh / 2,
+          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
+
+  if (dw / 2 < (int)dst->uv_width)
+    for (i = 0; i < dst->uv_height; i++)
+      vpx_memset(dst->u_buffer + i * dst->uv_stride + dw / 2 - 1, dst->u_buffer[i * dst->uv_stride + dw / 2 - 2], dst->uv_width - dw / 2 + 1);
+
+  if (dh / 2 < (int)dst->uv_height)
+    for (i = dh / 2 - 1; i < (int)dst->y_height / 2; i++)
+      vpx_memcpy(dst->u_buffer + i * dst->uv_stride, dst->u_buffer + (dh / 2 - 2)*dst->uv_stride, dst->uv_width);
+
+  Scale2D((unsigned char *) src->v_buffer, src->uv_stride, src->uv_width, src->uv_height,
+          (unsigned char *) dst->v_buffer, dst->uv_stride, dw / 2, dh / 2,
+          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
+
+  if (dw / 2 < (int)dst->uv_width)
+    for (i = 0; i < dst->uv_height; i++)
+      vpx_memset(dst->v_buffer + i * dst->uv_stride + dw / 2 - 1, dst->v_buffer[i * dst->uv_stride + dw / 2 - 2], dst->uv_width - dw / 2 + 1);
+
+  if (dh / 2 < (int) dst->uv_height)
+    for (i = dh / 2 - 1; i < (int)dst->y_height / 2; i++)
+      vpx_memcpy(dst->v_buffer + i * dst->uv_stride, dst->v_buffer + (dh / 2 - 2)*dst->uv_stride, dst->uv_width);
+}
diff --git a/libvpx/vpx_scale/generic/vpxscale.c b/libvpx/vpx_scale/generic/vpxscale.c
deleted file mode 100644
index 7de85ca..0000000
--- a/libvpx/vpx_scale/generic/vpxscale.c
+++ /dev/null
@@ -1,1010 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/****************************************************************************
- *
- *   Module Title :     scale.c
- *
- *   Description  :     Image scaling functions.
- *
- ***************************************************************************/
-
-/****************************************************************************
-*  Header Files
-****************************************************************************/
-#include "./vpx_rtcd.h"
-#include "vpx_mem/vpx_mem.h"
-#include "vpx_scale/yv12config.h"
-#include "vpx_scale/scale_mode.h"
-
-typedef struct {
-  int     expanded_frame_width;
-  int     expanded_frame_height;
-
-  int HScale;
-  int HRatio;
-  int VScale;
-  int VRatio;
-
-  YV12_BUFFER_CONFIG *src_yuv_config;
-  YV12_BUFFER_CONFIG *dst_yuv_config;
-
-} SCALE_VARS;
-
-/****************************************************************************
- *
- *  ROUTINE       :     horizontal_line_copy
- *
- *  INPUTS        :     None
- *
- *
- *  OUTPUTS       :     None.
- *
- *  RETURNS       :     None
- *
- *  FUNCTION      :     1 to 1 scaling up for a horizontal line of pixles
- *
- *  SPECIAL NOTES :     None.
- *
- *  ERRORS        :     None.
- *
- ****************************************************************************/
-static
-void horizontal_line_copy(
-  const unsigned char *source,
-  unsigned int source_width,
-  unsigned char *dest,
-  unsigned int dest_width
-) {
-  (void) dest_width;
-
-  duck_memcpy(dest, source, source_width);
-}
-/****************************************************************************
- *
- *  ROUTINE       :     null_scale
- *
- *  INPUTS        :     None
- *
- *
- *  OUTPUTS       :     None.
- *
- *  RETURNS       :     None
- *
- *  FUNCTION      :     1 to 1 scaling up for a vertical band
- *
- *  SPECIAL NOTES :     None.
- *
- *  ERRORS        :     None.
- *
- ****************************************************************************/
-static
-void null_scale(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  (void) dest;
-  (void) dest_pitch;
-  (void) dest_width;
-
-  return;
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : scale1d_2t1_i
- *
- *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
- *                  int source_step              : Number of pixels to step on in source.
- *                  unsigned int source_scale    : Scale for source (UNUSED).
- *                  unsigned int source_length   : Length of source (UNUSED).
- *                  unsigned char *dest         : Pointer to output data array.
- *                  int dest_step                : Number of pixels to step on in destination.
- *                  unsigned int dest_scale      : Scale for destination (UNUSED).
- *                  unsigned int dest_length     : Length of destination.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Performs 2-to-1 interpolated scaling.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void scale1d_2t1_i
-(
-  const unsigned char *source,
-  int source_step,
-  unsigned int source_scale,
-  unsigned int source_length,
-  unsigned char *dest,
-  int dest_step,
-  unsigned int dest_scale,
-  unsigned int dest_length
-) {
-  unsigned int i, j;
-  unsigned int temp;
-  int source_pitch = source_step;
-  (void) source_length;
-  (void) source_scale;
-  (void) dest_scale;
-
-  source_step *= 2;
-  dest[0] = source[0];
-
-  for (i = dest_step, j = source_step; i < dest_length * dest_step; i += dest_step, j += source_step) {
-    temp = 8;
-    temp += 3 * source[j - source_pitch];
-    temp += 10 * source[j];
-    temp += 3 * source[j + source_pitch];
-    temp >>= 4;
-    dest[i] = (char)(temp);
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : scale1d_2t1_ps
- *
- *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
- *                  int source_step              : Number of pixels to step on in source.
- *                  unsigned int source_scale    : Scale for source (UNUSED).
- *                  unsigned int source_length   : Length of source (UNUSED).
- *                  unsigned char *dest         : Pointer to output data array.
- *                  int dest_step                : Number of pixels to step on in destination.
- *                  unsigned int dest_scale      : Scale for destination (UNUSED).
- *                  unsigned int dest_length     : Length of destination.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Performs 2-to-1 point subsampled scaling.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void scale1d_2t1_ps
-(
-  const unsigned char *source,
-  int source_step,
-  unsigned int source_scale,
-  unsigned int source_length,
-  unsigned char *dest,
-  int dest_step,
-  unsigned int dest_scale,
-  unsigned int dest_length
-) {
-  unsigned int i, j;
-
-  (void) source_length;
-  (void) source_scale;
-  (void) dest_scale;
-
-  source_step *= 2;
-  j = 0;
-
-  for (i = 0; i < dest_length * dest_step; i += dest_step, j += source_step)
-    dest[i] = source[j];
-}
-/****************************************************************************
- *
- *  ROUTINE       : scale1d_c
- *
- *  INPUTS        : const unsigned char *source : Pointer to data to be scaled.
- *                  int source_step              : Number of pixels to step on in source.
- *                  unsigned int source_scale    : Scale for source.
- *                  unsigned int source_length   : Length of source (UNUSED).
- *                  unsigned char *dest         : Pointer to output data array.
- *                  int dest_step                : Number of pixels to step on in destination.
- *                  unsigned int dest_scale      : Scale for destination.
- *                  unsigned int dest_length     : Length of destination.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Performs linear interpolation in one dimension.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void scale1d_c
-(
-  const unsigned char *source,
-  int source_step,
-  unsigned int source_scale,
-  unsigned int source_length,
-  unsigned char *dest,
-  int dest_step,
-  unsigned int dest_scale,
-  unsigned int dest_length
-) {
-  unsigned int i;
-  unsigned int round_value = dest_scale / 2;
-  unsigned int left_modifier = dest_scale;
-  unsigned int right_modifier = 0;
-  unsigned char left_pixel = *source;
-  unsigned char right_pixel = *(source + source_step);
-
-  (void) source_length;
-
-  /* These asserts are needed if there are boundary issues... */
-  /*assert ( dest_scale > source_scale );*/
-  /*assert ( (source_length-1) * dest_scale >= (dest_length-1) * source_scale );*/
-
-  for (i = 0; i < dest_length * dest_step; i += dest_step) {
-    dest[i] = (char)((left_modifier * left_pixel + right_modifier * right_pixel + round_value) / dest_scale);
-
-    right_modifier += source_scale;
-
-    while (right_modifier > dest_scale) {
-      right_modifier -= dest_scale;
-      source += source_step;
-      left_pixel = *source;
-      right_pixel = *(source + source_step);
-    }
-
-    left_modifier = dest_scale - right_modifier;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : Scale2D
- *
- *  INPUTS        : const unsigned char *source  : Pointer to data to be scaled.
- *                  int source_pitch              : Stride of source image.
- *                  unsigned int source_width     : Width of input image.
- *                  unsigned int source_height    : Height of input image.
- *                  unsigned char *dest          : Pointer to output data array.
- *                  int dest_pitch                : Stride of destination image.
- *                  unsigned int dest_width       : Width of destination image.
- *                  unsigned int dest_height      : Height of destination image.
- *                  unsigned char *temp_area      : Pointer to temp work area.
- *                  unsigned char temp_area_height : Height of temp work area.
- *                  unsigned int hscale          : Horizontal scale factor numerator.
- *                  unsigned int hratio          : Horizontal scale factor denominator.
- *                  unsigned int vscale          : Vertical scale factor numerator.
- *                  unsigned int vratio          : Vertical scale factor denominator.
- *                  unsigned int interlaced      : Interlace flag.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Performs 2-tap linear interpolation in two dimensions.
- *
- *  SPECIAL NOTES : Expansion is performed one band at a time to help with
- *                  caching.
- *
- ****************************************************************************/
-static
-void Scale2D
-(
-  /*const*/
-  unsigned char *source,
-  int source_pitch,
-  unsigned int source_width,
-  unsigned int source_height,
-  unsigned char *dest,
-  int dest_pitch,
-  unsigned int dest_width,
-  unsigned int dest_height,
-  unsigned char *temp_area,
-  unsigned char temp_area_height,
-  unsigned int hscale,
-  unsigned int hratio,
-  unsigned int vscale,
-  unsigned int vratio,
-  unsigned int interlaced
-) {
-  /*unsigned*/
-  int i, j, k;
-  int bands;
-  int dest_band_height;
-  int source_band_height;
-
-  typedef void (*Scale1D)(const unsigned char * source, int source_step, unsigned int source_scale, unsigned int source_length,
-                          unsigned char * dest, int dest_step, unsigned int dest_scale, unsigned int dest_length);
-
-  Scale1D Scale1Dv = scale1d_c;
-  Scale1D Scale1Dh = scale1d_c;
-
-  void (*horiz_line_scale)(const unsigned char *, unsigned int, unsigned char *, unsigned int) = NULL;
-  void (*vert_band_scale)(unsigned char *, unsigned int, unsigned char *, unsigned int, unsigned int) = NULL;
-
-  int ratio_scalable = 1;
-  int interpolation = 0;
-
-  unsigned char *source_base; /* = (unsigned char *) ((source_pitch >= 0) ? source : (source + ((source_height-1) * source_pitch))); */
-  unsigned char *line_src;
-
-
-  source_base = (unsigned char *)source;
-
-  if (source_pitch < 0) {
-    int offset;
-
-    offset = (source_height - 1);
-    offset *= source_pitch;
-
-    source_base += offset;
-  }
-
-  /* find out the ratio for each direction */
-  switch (hratio * 10 / hscale) {
-    case 8:
-      /* 4-5 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_5_4_scale;
-      break;
-    case 6:
-      /* 3-5 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_5_3_scale;
-      break;
-    case 5:
-      /* 1-2 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_2_1_scale;
-      break;
-    default:
-      /* The ratio is not acceptable now */
-      /* throw("The ratio is not acceptable for now!"); */
-      ratio_scalable = 0;
-      break;
-  }
-
-  switch (vratio * 10 / vscale) {
-    case 8:
-      /* 4-5 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_5_4_scale;
-      source_band_height  = 5;
-      dest_band_height    = 4;
-      break;
-    case 6:
-      /* 3-5 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_5_3_scale;
-      source_band_height  = 5;
-      dest_band_height    = 3;
-      break;
-    case 5:
-      /* 1-2 Scale in vertical direction */
-
-      if (interlaced) {
-        /* if the content is interlaced, point sampling is used */
-        vert_band_scale     = vp8_vertical_band_2_1_scale;
-      } else {
-
-        interpolation = 1;
-        /* if the content is progressive, interplo */
-        vert_band_scale     = vp8_vertical_band_2_1_scale_i;
-
-      }
-
-      source_band_height  = 2;
-      dest_band_height    = 1;
-      break;
-    default:
-      /* The ratio is not acceptable now */
-      /* throw("The ratio is not acceptable for now!"); */
-      ratio_scalable = 0;
-      break;
-  }
-
-  if (ratio_scalable) {
-    if (source_height == dest_height) {
-      /* for each band of the image */
-      for (k = 0; k < (int)dest_height; k++) {
-        horiz_line_scale(source, source_width, dest, dest_width);
-        source += source_pitch;
-        dest   += dest_pitch;
-      }
-
-      return;
-    }
-
-    if (interpolation) {
-      if (source < source_base)
-        source = source_base;
-
-      horiz_line_scale(source, source_width, temp_area, dest_width);
-    }
-
-    for (k = 0; k < (int)(dest_height + dest_band_height - 1) / dest_band_height; k++) {
-      /* scale one band horizontally */
-      for (i = 0; i < source_band_height; i++) {
-        /* Trap case where we could read off the base of the source buffer */
-
-        line_src = (unsigned char *)source + i * source_pitch;
-
-        if (line_src < source_base)
-          line_src = source_base;
-
-        horiz_line_scale(line_src, source_width,
-                         temp_area + (i + 1)*dest_pitch, dest_width);
-      }
-
-      /* Vertical scaling is in place */
-      vert_band_scale(temp_area + dest_pitch, dest_pitch, dest, dest_pitch, dest_width);
-
-      if (interpolation)
-        vpx_memcpy(temp_area, temp_area + source_band_height * dest_pitch, dest_width);
-
-      /* Next band... */
-      source += (unsigned long) source_band_height  * source_pitch;
-      dest   += (unsigned long) dest_band_height * dest_pitch;
-    }
-
-    return;
-  }
-
-  if (hscale == 2 && hratio == 1)
-    Scale1Dh = scale1d_2t1_ps;
-
-  if (vscale == 2 && vratio == 1) {
-    if (interlaced)
-      Scale1Dv = scale1d_2t1_ps;
-    else
-      Scale1Dv = scale1d_2t1_i;
-  }
-
-  if (source_height == dest_height) {
-    /* for each band of the image */
-    for (k = 0; k < (int)dest_height; k++) {
-      Scale1Dh(source, 1, hscale, source_width + 1, dest, 1, hratio, dest_width);
-      source += source_pitch;
-      dest   += dest_pitch;
-    }
-
-    return;
-  }
-
-  if (dest_height > source_height) {
-    dest_band_height   = temp_area_height - 1;
-    source_band_height = dest_band_height * source_height / dest_height;
-  } else {
-    source_band_height = temp_area_height - 1;
-    dest_band_height   = source_band_height * vratio / vscale;
-  }
-
-  /* first row needs to be done so that we can stay one row ahead for vertical zoom */
-  Scale1Dh(source, 1, hscale, source_width + 1, temp_area, 1, hratio, dest_width);
-
-  /* for each band of the image */
-  bands = (dest_height + dest_band_height - 1) / dest_band_height;
-
-  for (k = 0; k < bands; k++) {
-    /* scale one band horizontally */
-    for (i = 1; i < source_band_height + 1; i++) {
-      if (k * source_band_height + i < (int) source_height) {
-        Scale1Dh(source + i * source_pitch, 1, hscale, source_width + 1,
-                 temp_area + i * dest_pitch, 1, hratio, dest_width);
-      } else { /*  Duplicate the last row */
-        /* copy temp_area row 0 over from last row in the past */
-        duck_memcpy(temp_area + i * dest_pitch, temp_area + (i - 1)*dest_pitch, dest_pitch);
-      }
-    }
-
-    /* scale one band vertically */
-    for (j = 0; j < (int)dest_width; j++) {
-      Scale1Dv(&temp_area[j], dest_pitch, vscale, source_band_height + 1,
-               &dest[j], dest_pitch, vratio, dest_band_height);
-    }
-
-    /* copy temp_area row 0 over from last row in the past */
-    duck_memcpy(temp_area, temp_area + source_band_height * dest_pitch, dest_pitch);
-
-    /* move to the next band */
-    source += source_band_height * source_pitch;
-    dest   += dest_band_height * dest_pitch;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       :
- *
- *  INPUTS        : YV12_BUFFER_CONFIG *src       : Pointer to frame to be scaled.
- *                  YV12_BUFFER_CONFIG *dst       : Pointer to buffer to hold scaled frame.
- *                  unsigned char *temp_area      : Pointer to temp work area.
- *                  unsigned char temp_area_height : Height of temp work area.
- *                  unsigned int hscale          : Horizontal scale factor numerator.
- *                  unsigned int hratio          : Horizontal scale factor denominator.
- *                  unsigned int vscale          : Vertical scale factor numerator.
- *                  unsigned int vratio          : Vertical scale factor denominator.
- *                  unsigned int interlaced      : Interlace flag.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Performs 2-tap linear interpolation in two dimensions.
- *
- *  SPECIAL NOTES : Expansion is performed one band at a time to help with
- *                  caching.
- *
- ****************************************************************************/
-void vp8_scale_frame
-(
-  YV12_BUFFER_CONFIG *src,
-  YV12_BUFFER_CONFIG *dst,
-  unsigned char *temp_area,
-  unsigned char temp_height,
-  unsigned int hscale,
-  unsigned int hratio,
-  unsigned int vscale,
-  unsigned int vratio,
-  unsigned int interlaced
-) {
-  int i;
-  int dw = (hscale - 1 + src->y_width * hratio) / hscale;
-  int dh = (vscale - 1 + src->y_height * vratio) / vscale;
-
-  /* call our internal scaling routines!! */
-  Scale2D((unsigned char *) src->y_buffer, src->y_stride, src->y_width, src->y_height,
-          (unsigned char *) dst->y_buffer, dst->y_stride, dw, dh,
-          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
-
-  if (dw < (int)dst->y_width)
-    for (i = 0; i < dh; i++)
-      duck_memset(dst->y_buffer + i * dst->y_stride + dw - 1, dst->y_buffer[i * dst->y_stride + dw - 2], dst->y_width - dw + 1);
-
-  if (dh < (int)dst->y_height)
-    for (i = dh - 1; i < (int)dst->y_height; i++)
-      duck_memcpy(dst->y_buffer + i * dst->y_stride, dst->y_buffer + (dh - 2) * dst->y_stride, dst->y_width + 1);
-
-  Scale2D((unsigned char *) src->u_buffer, src->uv_stride, src->uv_width, src->uv_height,
-          (unsigned char *) dst->u_buffer, dst->uv_stride, dw / 2, dh / 2,
-          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
-
-  if (dw / 2 < (int)dst->uv_width)
-    for (i = 0; i < dst->uv_height; i++)
-      duck_memset(dst->u_buffer + i * dst->uv_stride + dw / 2 - 1, dst->u_buffer[i * dst->uv_stride + dw / 2 - 2], dst->uv_width - dw / 2 + 1);
-
-  if (dh / 2 < (int)dst->uv_height)
-    for (i = dh / 2 - 1; i < (int)dst->y_height / 2; i++)
-      duck_memcpy(dst->u_buffer + i * dst->uv_stride, dst->u_buffer + (dh / 2 - 2)*dst->uv_stride, dst->uv_width);
-
-  Scale2D((unsigned char *) src->v_buffer, src->uv_stride, src->uv_width, src->uv_height,
-          (unsigned char *) dst->v_buffer, dst->uv_stride, dw / 2, dh / 2,
-          temp_area, temp_height, hscale, hratio, vscale, vratio, interlaced);
-
-  if (dw / 2 < (int)dst->uv_width)
-    for (i = 0; i < dst->uv_height; i++)
-      duck_memset(dst->v_buffer + i * dst->uv_stride + dw / 2 - 1, dst->v_buffer[i * dst->uv_stride + dw / 2 - 2], dst->uv_width - dw / 2 + 1);
-
-  if (dh / 2 < (int) dst->uv_height)
-    for (i = dh / 2 - 1; i < (int)dst->y_height / 2; i++)
-      duck_memcpy(dst->v_buffer + i * dst->uv_stride, dst->v_buffer + (dh / 2 - 2)*dst->uv_stride, dst->uv_width);
-}
-/****************************************************************************
- *
- *  ROUTINE       : any_ratio_2d_scale
- *
- *  INPUTS        : SCALE_INSTANCE *si      : Pointer to post-processor instance (NOT USED).
- *                  const unsigned char *source : Pointer to source image.
- *                  unsigned int source_pitch    : Stride of source image.
- *                  unsigned int source_width    : Width of source image.
- *                  unsigned int source_height   : Height of source image (NOT USED).
- *                  unsigned char *dest         : Pointer to destination image.
- *                  unsigned int dest_pitch      : Stride of destination image.
- *                  unsigned int dest_width      : Width of destination image.
- *                  unsigned int dest_height     : Height of destination image.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : int: 1 if image scaled, 0 if image could not be scaled.
- *
- *  FUNCTION      : Scale the image with changing apect ratio.
- *
- *  SPECIAL NOTES : This scaling is a bi-linear scaling. Need to re-work the
- *                  whole function for new scaling algorithm.
- *
- ****************************************************************************/
-static
-int any_ratio_2d_scale
-(
-  SCALE_VARS *si,
-  const unsigned char *source,
-  int source_pitch,
-  unsigned int source_width,
-  unsigned int source_height,
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width,
-  unsigned int dest_height
-) {
-  unsigned int i, k;
-  unsigned int src_band_height  = 0;
-  unsigned int dest_band_height = 0;
-
-  /* suggested scale factors */
-  int hs = si->HScale;
-  int hr = si->HRatio;
-  int vs = si->VScale;
-  int vr = si->VRatio;
-
-  /* assume the ratios are scalable instead of should be centered */
-  int ratio_scalable = 1;
-
-  const unsigned char *source_base = ((source_pitch >= 0) ? source : (source + ((source_height - 1) * source_pitch)));
-  const unsigned char *line_src;
-
-  void (*horiz_line_scale)(const unsigned char *, unsigned int, unsigned char *, unsigned int) = NULL;
-  void (*vert_band_scale)(unsigned char *, unsigned int, unsigned int) = NULL;
-  void (*last_vert_band_scale)(unsigned char *, unsigned int, unsigned int) = NULL;
-
-  (void) si;
-
-  /* find out the ratio for each direction */
-  switch (hr * 30 / hs) {
-    case 24:
-      /* 4-5 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_4_5_scale;
-      break;
-    case 22:
-      /* 3-4 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_3_4_scale;
-      break;
-
-    case 20:
-      /* 4-5 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_2_3_scale;
-      break;
-    case 18:
-      /* 3-5 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_3_5_scale;
-      break;
-    case 15:
-      /* 1-2 Scale in Width direction */
-      horiz_line_scale = vp8_horizontal_line_1_2_scale;
-      break;
-    case 30:
-      /* no scale in Width direction */
-      horiz_line_scale = horizontal_line_copy;
-      break;
-    default:
-      /* The ratio is not acceptable now */
-      /* throw("The ratio is not acceptable for now!"); */
-      ratio_scalable = 0;
-      break;
-  }
-
-  switch (vr * 30 / vs) {
-    case 24:
-      /* 4-5 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_4_5_scale;
-      last_vert_band_scale = vp8_last_vertical_band_4_5_scale;
-      src_band_height     = 4;
-      dest_band_height    = 5;
-      break;
-    case 22:
-      /* 3-4 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_3_4_scale;
-      last_vert_band_scale = vp8_last_vertical_band_3_4_scale;
-      src_band_height     = 3;
-      dest_band_height    = 4;
-      break;
-    case 20:
-      /* 2-3 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_2_3_scale;
-      last_vert_band_scale = vp8_last_vertical_band_2_3_scale;
-      src_band_height     = 2;
-      dest_band_height    = 3;
-      break;
-    case 18:
-      /* 3-5 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_3_5_scale;
-      last_vert_band_scale = vp8_last_vertical_band_3_5_scale;
-      src_band_height     = 3;
-      dest_band_height    = 5;
-      break;
-    case 15:
-      /* 1-2 Scale in vertical direction */
-      vert_band_scale     = vp8_vertical_band_1_2_scale;
-      last_vert_band_scale = vp8_last_vertical_band_1_2_scale;
-      src_band_height     = 1;
-      dest_band_height    = 2;
-      break;
-    case 30:
-      /* no scale in Width direction */
-      vert_band_scale     = null_scale;
-      last_vert_band_scale = null_scale;
-      src_band_height     = 4;
-      dest_band_height    = 4;
-      break;
-    default:
-      /* The ratio is not acceptable now */
-      /* throw("The ratio is not acceptable for now!"); */
-      ratio_scalable = 0;
-      break;
-  }
-
-  if (ratio_scalable == 0)
-    return ratio_scalable;
-
-  horiz_line_scale(source, source_width, dest, dest_width);
-
-  /* except last band */
-  for (k = 0; k < (dest_height + dest_band_height - 1) / dest_band_height - 1; k++) {
-    /* scale one band horizontally */
-    for (i = 1; i < src_band_height; i++) {
-      /* Trap case where we could read off the base of the source buffer */
-      line_src = source + i * source_pitch;
-
-      if (line_src < source_base)
-        line_src = source_base;
-
-      horiz_line_scale(line_src, source_width,
-                       dest + i * dest_pitch, dest_width);
-    }
-
-    /* first line of next band */
-    /* Trap case where we could read off the base of the source buffer */
-    line_src = source + src_band_height * source_pitch;
-
-    if (line_src < source_base)
-      line_src = source_base;
-
-    horiz_line_scale(line_src, source_width,
-                     dest + dest_band_height * dest_pitch,
-                     dest_width);
-
-    /* Vertical scaling is in place */
-    vert_band_scale(dest, dest_pitch, dest_width);
-
-    /* Next band... */
-    source += src_band_height  * source_pitch;
-    dest   += dest_band_height * dest_pitch;
-  }
-
-  /* scale one band horizontally */
-  for (i = 1; i < src_band_height; i++) {
-    /* Trap case where we could read off the base of the source buffer */
-    line_src = source + i * source_pitch;
-
-    if (line_src < source_base)
-      line_src = source_base;
-
-    horiz_line_scale(line_src, source_width,
-                     dest + i * dest_pitch,
-                     dest_width);
-  }
-
-  /* Vertical scaling is in place */
-  last_vert_band_scale(dest, dest_pitch, dest_width);
-
-  return ratio_scalable;
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : any_ratio_frame_scale
- *
- *  INPUTS        : SCALE_INSTANCE *si       : Pointer to post-processor instance (NOT USED).
- *                  unsigned char *frame_buffer           : Pointer to source image.
- *                  int YOffset                : Offset from start of buffer to Y samples.
- *                  int UVOffset               : Offset from start of buffer to UV samples.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : int: 1 if image scaled, 0 if image could not be scaled.
- *
- *  FUNCTION      : Scale the image with changing apect ratio.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-int any_ratio_frame_scale(SCALE_VARS *scale_vars, int YOffset, int UVOffset) {
-  int i;
-  int ew;
-  int eh;
-
-  /* suggested scale factors */
-  int hs = scale_vars->HScale;
-  int hr = scale_vars->HRatio;
-  int vs = scale_vars->VScale;
-  int vr = scale_vars->VRatio;
-
-  int ratio_scalable = 1;
-
-  int sw = (scale_vars->expanded_frame_width * hr + hs - 1) / hs;
-  int sh = (scale_vars->expanded_frame_height * vr + vs - 1) / vs;
-  int dw = scale_vars->expanded_frame_width;
-  int dh = scale_vars->expanded_frame_height;
-  YV12_BUFFER_CONFIG *src_yuv_config = scale_vars->src_yuv_config;
-  YV12_BUFFER_CONFIG *dst_yuv_config = scale_vars->dst_yuv_config;
-
-  if (hr == 3)
-    ew = (sw + 2) / 3 * 3 * hs / hr;
-  else
-    ew = (sw + 7) / 8 * 8 * hs / hr;
-
-  if (vr == 3)
-    eh = (sh + 2) / 3 * 3 * vs / vr;
-  else
-    eh = (sh + 7) / 8 * 8 * vs / vr;
-
-  ratio_scalable = any_ratio_2d_scale(scale_vars,
-                                      (const unsigned char *)src_yuv_config->y_buffer,
-                                      src_yuv_config->y_stride, sw, sh,
-                                      (unsigned char *) dst_yuv_config->y_buffer + YOffset,
-                                      dst_yuv_config->y_stride, dw, dh);
-
-  for (i = 0; i < eh; i++)
-    duck_memset(dst_yuv_config->y_buffer + YOffset + i * dst_yuv_config->y_stride + dw, 0, ew - dw);
-
-  for (i = dh; i < eh; i++)
-    duck_memset(dst_yuv_config->y_buffer + YOffset + i * dst_yuv_config->y_stride, 0, ew);
-
-  if (ratio_scalable == 0)
-    return ratio_scalable;
-
-  sw = (sw + 1) >> 1;
-  sh = (sh + 1) >> 1;
-  dw = (dw + 1) >> 1;
-  dh = (dh + 1) >> 1;
-
-  any_ratio_2d_scale(scale_vars,
-                     (const unsigned char *)src_yuv_config->u_buffer,
-                     src_yuv_config->y_stride / 2, sw, sh,
-                     (unsigned char *)dst_yuv_config->u_buffer + UVOffset,
-                     dst_yuv_config->uv_stride, dw, dh);
-
-  any_ratio_2d_scale(scale_vars,
-                     (const unsigned char *)src_yuv_config->v_buffer,
-                     src_yuv_config->y_stride / 2, sw, sh,
-                     (unsigned char *)dst_yuv_config->v_buffer + UVOffset,
-                     dst_yuv_config->uv_stride, dw, dh);
-
-  return ratio_scalable;
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : center_image
- *
- *  INPUTS        : SCALE_INSTANCE *si       : Pointer to post-processor instance.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Centers the image without scaling in the output buffer.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static void
-center_image(YV12_BUFFER_CONFIG *src_yuv_config, YV12_BUFFER_CONFIG *dst_yuv_config) {
-  int i;
-  int row_offset, col_offset;
-  unsigned char *src_data_pointer;
-  unsigned char *dst_data_pointer;
-
-  /* center values */
-  row_offset = (dst_yuv_config->y_height - src_yuv_config->y_height) / 2;
-  col_offset = (dst_yuv_config->y_width - src_yuv_config->y_width) / 2;
-
-  /* Y's */
-  src_data_pointer = src_yuv_config->y_buffer;
-  dst_data_pointer = (unsigned char *)dst_yuv_config->y_buffer + (row_offset * dst_yuv_config->y_stride) + col_offset;
-
-  for (i = 0; i < src_yuv_config->y_height; i++) {
-    duck_memcpy(dst_data_pointer, src_data_pointer, src_yuv_config->y_width);
-    dst_data_pointer += dst_yuv_config->y_stride;
-    src_data_pointer += src_yuv_config->y_stride;
-  }
-
-  row_offset /= 2;
-  col_offset /= 2;
-
-  /* U's */
-  src_data_pointer = src_yuv_config->u_buffer;
-  dst_data_pointer = (unsigned char *)dst_yuv_config->u_buffer + (row_offset * dst_yuv_config->uv_stride) + col_offset;
-
-  for (i = 0; i < src_yuv_config->uv_height; i++) {
-    duck_memcpy(dst_data_pointer, src_data_pointer, src_yuv_config->uv_width);
-    dst_data_pointer += dst_yuv_config->uv_stride;
-    src_data_pointer += src_yuv_config->uv_stride;
-  }
-
-  /* V's */
-  src_data_pointer = src_yuv_config->v_buffer;
-  dst_data_pointer = (unsigned char *)dst_yuv_config->v_buffer + (row_offset * dst_yuv_config->uv_stride) + col_offset;
-
-  for (i = 0; i < src_yuv_config->uv_height; i++) {
-    duck_memcpy(dst_data_pointer, src_data_pointer, src_yuv_config->uv_width);
-    dst_data_pointer += dst_yuv_config->uv_stride;
-    src_data_pointer += src_yuv_config->uv_stride;
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : scale_or_center
- *
- *  INPUTS        : SCALE_INSTANCE *si       : Pointer to post-processor instance.
- *
- *
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Decides to scale or center image in scale buffer for blit
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-void
-vp8_yv12_scale_or_center
-(
-  YV12_BUFFER_CONFIG *src_yuv_config,
-  YV12_BUFFER_CONFIG *dst_yuv_config,
-  int expanded_frame_width,
-  int expanded_frame_height,
-  int scaling_mode,
-  int HScale,
-  int HRatio,
-  int VScale,
-  int VRatio
-) {
-  /*if ( ppi->post_processing_level )
-        update_umvborder ( ppi, frame_buffer );*/
-
-
-  switch (scaling_mode) {
-    case SCALE_TO_FIT:
-    case MAINTAIN_ASPECT_RATIO: {
-      SCALE_VARS scale_vars;
-      /* center values */
-#if 1
-      int row = (dst_yuv_config->y_height - expanded_frame_height) / 2;
-      int col = (dst_yuv_config->y_width  - expanded_frame_width) / 2;
-      /*int YOffset  = row * dst_yuv_config->y_width + col;
-      int UVOffset = (row>>1) * dst_yuv_config->uv_width + (col>>1);*/
-      int YOffset  = row * dst_yuv_config->y_stride + col;
-      int UVOffset = (row >> 1) * dst_yuv_config->uv_stride + (col >> 1);
-#else
-      int row = (src_yuv_config->y_height - expanded_frame_height) / 2;
-      int col = (src_yuv_config->y_width  - expanded_frame_width) / 2;
-      int YOffset  = row * src_yuv_config->y_width + col;
-      int UVOffset = (row >> 1) * src_yuv_config->uv_width + (col >> 1);
-#endif
-
-      scale_vars.dst_yuv_config = dst_yuv_config;
-      scale_vars.src_yuv_config = src_yuv_config;
-      scale_vars.HScale = HScale;
-      scale_vars.HRatio = HRatio;
-      scale_vars.VScale = VScale;
-      scale_vars.VRatio = VRatio;
-      scale_vars.expanded_frame_width = expanded_frame_width;
-      scale_vars.expanded_frame_height = expanded_frame_height;
-
-      /* perform center and scale */
-      any_ratio_frame_scale(&scale_vars, YOffset, UVOffset);
-
-      break;
-    }
-    case CENTER:
-      center_image(src_yuv_config, dst_yuv_config);
-      break;
-
-    default:
-      break;
-  }
-}
diff --git a/libvpx/vpx_scale/generic/yv12config.c b/libvpx/vpx_scale/generic/yv12config.c
index 4cb2a41..754a615 100644
--- a/libvpx/vpx_scale/generic/yv12config.c
+++ b/libvpx/vpx_scale/generic/yv12config.c
@@ -8,7 +8,7 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
+#include "./vpx_config.h"
 #include "vpx_scale/yv12config.h"
 #include "vpx_mem/vpx_mem.h"
 
@@ -35,36 +35,139 @@
   return 0;
 }
 
-/****************************************************************************
- *
- ****************************************************************************/
-int
-vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, int border) {
-  /*NOTE:*/
-
+int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                                  int width, int height, int border) {
   if (ybf) {
-    int y_stride = ((width + 2 * border) + 31) & ~31;
-    int yplane_size = (height + 2 * border) * y_stride;
-    int uv_width = width >> 1;
-    int uv_height = height >> 1;
+    int aligned_width = (width + 15) & ~15;
+    int aligned_height = (height + 15) & ~15;
+    int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
+    int yplane_size = (aligned_height + 2 * border) * y_stride;
+    int uv_width = aligned_width >> 1;
+    int uv_height = aligned_height >> 1;
     /** There is currently a bunch of code which assumes
       *  uv_stride == y_stride/2, so enforce this here. */
     int uv_stride = y_stride >> 1;
     int uvplane_size = (uv_height + border) * uv_stride;
+    const int frame_size = yplane_size + 2 * uvplane_size;
 
-    vp8_yv12_de_alloc_frame_buffer(ybf);
+    if (!ybf->buffer_alloc) {
+      ybf->buffer_alloc = vpx_memalign(32, frame_size);
+      ybf->buffer_alloc_sz = frame_size;
+    }
 
-    /** Only support allocating buffers that have a height and width that
-      *  are multiples of 16, and a border that's a multiple of 32.
-      * The border restriction is required to get 16-byte alignment of the
-      *  start of the chroma rows without intoducing an arbitrary gap
-      *  between planes, which would break the semantics of things like
-      *  vpx_img_set_rect(). */
-    if ((width & 0xf) | (height & 0xf) | (border & 0x1f))
+    if (!ybf->buffer_alloc || ybf->buffer_alloc_sz < frame_size)
+      return -1;
+
+    /* Only support allocating buffers that have a border that's a multiple
+     * of 32. The border restriction is required to get 16-byte alignment of
+     * the start of the chroma rows without intoducing an arbitrary gap
+     * between planes, which would break the semantics of things like
+     * vpx_img_set_rect(). */
+    if (border & 0x1f)
       return -3;
 
-    ybf->y_width  = width;
-    ybf->y_height = height;
+    ybf->y_crop_width = width;
+    ybf->y_crop_height = height;
+    ybf->y_width  = aligned_width;
+    ybf->y_height = aligned_height;
+    ybf->y_stride = y_stride;
+
+    ybf->uv_width = uv_width;
+    ybf->uv_height = uv_height;
+    ybf->uv_stride = uv_stride;
+
+    ybf->alpha_width = 0;
+    ybf->alpha_height = 0;
+    ybf->alpha_stride = 0;
+
+    ybf->border = border;
+    ybf->frame_size = frame_size;
+
+    ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border;
+    ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2  * uv_stride) + border / 2;
+    ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2  * uv_stride) + border / 2;
+    ybf->alpha_buffer = NULL;
+
+    ybf->corrupted = 0; /* assume not currupted by errors */
+    return 0;
+  }
+  return -2;
+}
+
+int vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                                int width, int height, int border) {
+  if (ybf) {
+    vp8_yv12_de_alloc_frame_buffer(ybf);
+    return vp8_yv12_realloc_frame_buffer(ybf, width, height, border);
+  }
+  return -2;
+}
+
+#if CONFIG_VP9
+// TODO(jkoleszar): Maybe replace this with struct vpx_image
+
+int vp9_free_frame_buffer(YV12_BUFFER_CONFIG *ybf) {
+  if (ybf) {
+    vpx_free(ybf->buffer_alloc);
+
+    /* buffer_alloc isn't accessed by most functions.  Rather y_buffer,
+      u_buffer and v_buffer point to buffer_alloc and are used.  Clear out
+      all of this so that a freed pointer isn't inadvertently used */
+    vpx_memset(ybf, 0, sizeof(YV12_BUFFER_CONFIG));
+  } else {
+    return -1;
+  }
+
+  return 0;
+}
+
+int vp9_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                             int width, int height,
+                             int ss_x, int ss_y, int border) {
+  if (ybf) {
+    const int aligned_width = (width + 7) & ~7;
+    const int aligned_height = (height + 7) & ~7;
+    const int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
+    const int yplane_size = (aligned_height + 2 * border) * y_stride;
+    const int uv_width = aligned_width >> ss_x;
+    const int uv_height = aligned_height >> ss_y;
+    const int uv_stride = y_stride >> ss_x;
+    const int uv_border_w = border >> ss_x;
+    const int uv_border_h = border >> ss_y;
+    const int uvplane_size = (uv_height + 2 * uv_border_h) * uv_stride;
+#if CONFIG_ALPHA
+    const int alpha_width = aligned_width;
+    const int alpha_height = aligned_height;
+    const int alpha_stride = y_stride;
+    const int alpha_border_w = border;
+    const int alpha_border_h = border;
+    const int alpha_plane_size = (alpha_height + 2 * alpha_border_h) *
+                                 alpha_stride;
+    const int frame_size = yplane_size + 2 * uvplane_size +
+                           alpha_plane_size;
+#else
+    const int frame_size = yplane_size + 2 * uvplane_size;
+#endif
+    if (!ybf->buffer_alloc) {
+      ybf->buffer_alloc = vpx_memalign(32, frame_size);
+      ybf->buffer_alloc_sz = frame_size;
+    }
+
+    if (!ybf->buffer_alloc || ybf->buffer_alloc_sz < frame_size)
+      return -1;
+
+    /* Only support allocating buffers that have a border that's a multiple
+     * of 32. The border restriction is required to get 16-byte alignment of
+     * the start of the chroma rows without intoducing an arbitrary gap
+     * between planes, which would break the semantics of things like
+     * vpx_img_set_rect(). */
+    if (border & 0x1f)
+      return -3;
+
+    ybf->y_crop_width = width;
+    ybf->y_crop_height = height;
+    ybf->y_width  = aligned_width;
+    ybf->y_height = aligned_height;
     ybf->y_stride = y_stride;
 
     ybf->uv_width = uv_width;
@@ -72,21 +175,34 @@
     ybf->uv_stride = uv_stride;
 
     ybf->border = border;
-    ybf->frame_size = yplane_size + 2 * uvplane_size;
-
-    ybf->buffer_alloc = (unsigned char *) vpx_memalign(32, ybf->frame_size);
-
-    if (ybf->buffer_alloc == NULL)
-      return -1;
+    ybf->frame_size = frame_size;
 
     ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border;
-    ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2  * uv_stride) + border / 2;
-    ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2  * uv_stride) + border / 2;
+    ybf->u_buffer = ybf->buffer_alloc + yplane_size +
+                    (uv_border_h * uv_stride) + uv_border_w;
+    ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size +
+                    (uv_border_h * uv_stride) + uv_border_w;
 
+#if CONFIG_ALPHA
+    ybf->alpha_width = alpha_width;
+    ybf->alpha_height = alpha_height;
+    ybf->alpha_stride = alpha_stride;
+    ybf->alpha_buffer = ybf->buffer_alloc + yplane_size + 2 * uvplane_size +
+                        (alpha_border_h * alpha_stride) + alpha_border_w;
+#endif
     ybf->corrupted = 0; /* assume not currupted by errors */
-  } else {
-    return -2;
+    return 0;
   }
-
-  return 0;
+  return -2;
 }
+
+int vp9_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                           int width, int height,
+                           int ss_x, int ss_y, int border) {
+  if (ybf) {
+    vp9_free_frame_buffer(ybf);
+    return vp9_realloc_frame_buffer(ybf, width, height, ss_x, ss_y, border);
+  }
+  return -2;
+}
+#endif
diff --git a/libvpx/vpx_scale/generic/yv12extend.c b/libvpx/vpx_scale/generic/yv12extend.c
index 247078c..c38fb80 100644
--- a/libvpx/vpx_scale/generic/yv12extend.c
+++ b/libvpx/vpx_scale/generic/yv12extend.c
@@ -8,10 +8,11 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
+#include <assert.h>
+#include "./vpx_config.h"
 #include "vpx_scale/yv12config.h"
 #include "vpx_mem/vpx_mem.h"
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 
 /****************************************************************************
 *  Exports
@@ -20,179 +21,110 @@
 /****************************************************************************
  *
  ****************************************************************************/
+static void extend_plane(uint8_t *s,       /* source */
+                         int sp,           /* source pitch */
+                         int w,            /* width */
+                         int h,            /* height */
+                         int et,           /* extend top border */
+                         int el,           /* extend left border */
+                         int eb,           /* extend bottom border */
+                         int er) {         /* extend right border */
+  int i;
+  uint8_t *src_ptr1, *src_ptr2;
+  uint8_t *dest_ptr1, *dest_ptr2;
+  int linesize;
+
+  /* copy the left and right most columns out */
+  src_ptr1 = s;
+  src_ptr2 = s + w - 1;
+  dest_ptr1 = s - el;
+  dest_ptr2 = s + w;
+
+  for (i = 0; i < h; i++) {
+    vpx_memset(dest_ptr1, src_ptr1[0], el);
+    vpx_memset(dest_ptr2, src_ptr2[0], er);
+    src_ptr1  += sp;
+    src_ptr2  += sp;
+    dest_ptr1 += sp;
+    dest_ptr2 += sp;
+  }
+
+  /* Now copy the top and bottom lines into each line of the respective
+   * borders
+   */
+  src_ptr1 = s - el;
+  src_ptr2 = s + sp * (h - 1) - el;
+  dest_ptr1 = s + sp * (-et) - el;
+  dest_ptr2 = s + sp * (h) - el;
+  linesize = el + er + w;
+
+  for (i = 0; i < et; i++) {
+    vpx_memcpy(dest_ptr1, src_ptr1, linesize);
+    dest_ptr1 += sp;
+  }
+
+  for (i = 0; i < eb; i++) {
+    vpx_memcpy(dest_ptr2, src_ptr2, linesize);
+    dest_ptr2 += sp;
+  }
+}
+
 void
 vp8_yv12_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf) {
-  int i;
-  unsigned char *src_ptr1, *src_ptr2;
-  unsigned char *dest_ptr1, *dest_ptr2;
+  assert(ybf->y_height - ybf->y_crop_height < 16);
+  assert(ybf->y_width - ybf->y_crop_width < 16);
+  assert(ybf->y_height - ybf->y_crop_height >= 0);
+  assert(ybf->y_width - ybf->y_crop_width >= 0);
 
-  unsigned int Border;
-  int plane_stride;
-  int plane_height;
-  int plane_width;
+  extend_plane(ybf->y_buffer, ybf->y_stride,
+               ybf->y_crop_width, ybf->y_crop_height,
+               ybf->border, ybf->border,
+               ybf->border + ybf->y_height - ybf->y_crop_height,
+               ybf->border + ybf->y_width - ybf->y_crop_width);
 
-  /***********/
-  /* Y Plane */
-  /***********/
-  Border = ybf->border;
-  plane_stride = ybf->y_stride;
-  plane_height = ybf->y_height;
-  plane_width = ybf->y_width;
+  extend_plane(ybf->u_buffer, ybf->uv_stride,
+               (ybf->y_crop_width + 1) / 2, (ybf->y_crop_height + 1) / 2,
+               ybf->border / 2, ybf->border / 2,
+               (ybf->border + ybf->y_height - ybf->y_crop_height + 1) / 2,
+               (ybf->border + ybf->y_width - ybf->y_crop_width + 1) / 2);
 
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->y_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->y_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)Border; i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-
-  /***********/
-  /* U Plane */
-  /***********/
-  plane_stride = ybf->uv_stride;
-  plane_height = ybf->uv_height;
-  plane_width = ybf->uv_width;
-  Border /= 2;
-
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->u_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->u_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)(Border); i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /***********/
-  /* V Plane */
-  /***********/
-
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->v_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->v_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)(Border); i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
+  extend_plane(ybf->v_buffer, ybf->uv_stride,
+               (ybf->y_crop_width + 1) / 2, (ybf->y_crop_height + 1) / 2,
+               ybf->border / 2, ybf->border / 2,
+               (ybf->border + ybf->y_height - ybf->y_crop_height + 1) / 2,
+               (ybf->border + ybf->y_width - ybf->y_crop_width + 1) / 2);
 }
 
+#if CONFIG_VP9
+void vp9_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf,
+                                int subsampling_x, int subsampling_y) {
+  const int c_w = (ybf->y_crop_width + subsampling_x) >> subsampling_x;
+  const int c_h = (ybf->y_crop_height + subsampling_y) >> subsampling_y;
+  const int c_et = ybf->border >> subsampling_y;
+  const int c_el = ybf->border >> subsampling_x;
+  const int c_eb = (ybf->border + ybf->y_height - ybf->y_crop_height +
+                    subsampling_y) >> subsampling_y;
+  const int c_er = (ybf->border + ybf->y_width - ybf->y_crop_width +
+                    subsampling_x) >> subsampling_x;
 
-static void
-extend_frame_borders_yonly_c(YV12_BUFFER_CONFIG *ybf) {
-  int i;
-  unsigned char *src_ptr1, *src_ptr2;
-  unsigned char *dest_ptr1, *dest_ptr2;
+  assert(ybf->y_height - ybf->y_crop_height < 16);
+  assert(ybf->y_width - ybf->y_crop_width < 16);
+  assert(ybf->y_height - ybf->y_crop_height >= 0);
+  assert(ybf->y_width - ybf->y_crop_width >= 0);
 
-  unsigned int Border;
-  int plane_stride;
-  int plane_height;
-  int plane_width;
+  extend_plane(ybf->y_buffer, ybf->y_stride,
+               ybf->y_crop_width, ybf->y_crop_height,
+               ybf->border, ybf->border,
+               ybf->border + ybf->y_height - ybf->y_crop_height,
+               ybf->border + ybf->y_width - ybf->y_crop_width);
 
-  /***********/
-  /* Y Plane */
-  /***********/
-  Border = ybf->border;
-  plane_stride = ybf->y_stride;
-  plane_height = ybf->y_height;
-  plane_width = ybf->y_width;
+  extend_plane(ybf->u_buffer, ybf->uv_stride,
+               c_w, c_h, c_et, c_el, c_eb, c_er);
 
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->y_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->y_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)Border; i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  plane_stride /= 2;
-  plane_height /= 2;
-  plane_width /= 2;
-  Border /= 2;
-
+  extend_plane(ybf->v_buffer, ybf->uv_stride,
+               c_w, c_h, c_et, c_el, c_eb, c_er);
 }
-
-
+#endif
 
 /****************************************************************************
  *
@@ -216,6 +148,14 @@
   int row;
   unsigned char *source, *dest;
 
+#if 0
+  /* These assertions are valid in the codec, but the libvpx-tester uses
+   * this code slightly differently.
+   */
+  assert(src_ybc->y_width == dst_ybc->y_width);
+  assert(src_ybc->y_height == dst_ybc->y_height);
+#endif
+
   source = src_ybc->y_buffer;
   dest = dst_ybc->y_buffer;
 
diff --git a/libvpx/vpx_scale/generic/yv12extend_generic.h b/libvpx/vpx_scale/generic/yv12extend_generic.h
deleted file mode 100644
index cc2a554..0000000
--- a/libvpx/vpx_scale/generic/yv12extend_generic.h
+++ /dev/null
@@ -1,25 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-#ifndef YV12_EXTEND_GENERIC_H
-#define YV12_EXTEND_GENERIC_H
-
-#include "vpx_scale/yv12config.h"
-
-    void vp8_yv12_extend_frame_borders(YV12_BUFFER_CONFIG *ybf);
-
-    /* Copy Y,U,V buffer data from src to dst, filling border of dst as well. */
-    void vp8_yv12_copy_frame(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc);
-
-    /* Copy Y buffer data from src_ybc to dst_ybc without filling border data */
-    void vp8_yv12_copy_y_c(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc);
-
-#endif /* YV12_EXTEND_GENERIC_H */
diff --git a/libvpx/vpx_scale/include/generic/vpxscale_arbitrary.h b/libvpx/vpx_scale/include/generic/vpxscale_arbitrary.h
deleted file mode 100644
index c535252..0000000
--- a/libvpx/vpx_scale/include/generic/vpxscale_arbitrary.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-#ifndef __VPX_SCALE_ARBITRARY_H__
-#define __VPX_SCALE_ARBITRARY_H__
-
-#include "vpx_scale/yv12config.h"
-
-typedef struct {
-  int in_width;
-  int in_height;
-
-  int out_width;
-  int out_height;
-  int max_usable_out_width;
-
-  // numerator for the width and height
-  int nw;
-  int nh;
-  int nh_uv;
-
-  // output to input correspondance array
-  short *l_w;
-  short *l_h;
-  short *l_h_uv;
-
-  // polyphase coefficients
-  short *c_w;
-  short *c_h;
-  short *c_h_uv;
-
-  // buffer for horizontal filtering.
-  unsigned char *hbuf;
-  unsigned char *hbuf_uv;
-} BICUBIC_SCALER_STRUCT;
-
-int bicubic_coefficient_setup(int in_width, int in_height, int out_width, int out_height);
-int bicubic_scale(int in_width, int in_height, int in_stride,
-                  int out_width, int out_height, int out_stride,
-                  unsigned char *input_image, unsigned char *output_image);
-void bicubic_scale_frame_reset();
-void bicubic_scale_frame(YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
-                         int new_width, int new_height);
-void bicubic_coefficient_init();
-void bicubic_coefficient_destroy();
-
-#endif /* __VPX_SCALE_ARBITRARY_H__ */
diff --git a/libvpx/vpx_scale/include/generic/vpxscale_depricated.h b/libvpx/vpx_scale/include/generic/vpxscale_depricated.h
deleted file mode 100644
index 3f7fe0f..0000000
--- a/libvpx/vpx_scale/include/generic/vpxscale_depricated.h
+++ /dev/null
@@ -1,34 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/****************************************************************************
-*
-*   Module Title :     postp.h
-*
-*   Description  :     Post processor interface
-*
-****************************************************************************/
-#ifndef VPXSCALE_H
-#define VPXSCALE_H
-
-extern void (*vp8_vertical_band_4_5_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-extern void (*vp8_last_vertical_band_4_5_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-extern void (*vp8_vertical_band_3_5_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-extern void (*vp8_last_vertical_band_3_5_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-extern void (*vp8_horizontal_line_1_2_scale)(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-extern void (*vp8_horizontal_line_3_5_scale)(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-extern void (*vp8_horizontal_line_4_5_scale)(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-extern void (*vp8_vertical_band_1_2_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-extern void (*vp8_last_vertical_band_1_2_scale)(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-
-extern void  dmachine_specific_config(int mmx_enabled, int xmm_enabled, int wmt_enabled);
-
-#endif
diff --git a/libvpx/vpx_scale/scale_mode.h b/libvpx/vpx_scale/scale_mode.h
deleted file mode 100644
index 5581385..0000000
--- a/libvpx/vpx_scale/scale_mode.h
+++ /dev/null
@@ -1,28 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/****************************************************************************
-*
-*****************************************************************************
-*/
-
-#ifndef SCALE_MODE_H
-#define SCALE_MODE_H
-
-typedef enum {
-  MAINTAIN_ASPECT_RATIO   = 0x0,
-  SCALE_TO_FIT            = 0x1,
-  CENTER                  = 0x2,
-  OTHER                   = 0x3
-} SCALE_MODE;
-
-
-#endif
diff --git a/libvpx/vpx_scale/vpxscale.h b/libvpx/vpx_scale/vpx_scale.h
similarity index 61%
rename from libvpx/vpx_scale/vpxscale.h
rename to libvpx/vpx_scale/vpx_scale.h
index 3c2194d..9ddf62e 100644
--- a/libvpx/vpx_scale/vpxscale.h
+++ b/libvpx/vpx_scale/vpx_scale.h
@@ -14,17 +14,7 @@
 
 #include "vpx_scale/yv12config.h"
 
-extern void vp8_yv12_scale_or_center(YV12_BUFFER_CONFIG *src_yuv_config,
-                                     YV12_BUFFER_CONFIG *dst_yuv_config,
-                                     int expanded_frame_width,
-                                     int expanded_frame_height,
-                                     int scaling_mode,
-                                     int HScale,
-                                     int HRatio,
-                                     int VScale,
-                                     int VRatio);
-
-extern void vp8_scale_frame(YV12_BUFFER_CONFIG *src,
+extern void vpx_scale_frame(YV12_BUFFER_CONFIG *src,
                             YV12_BUFFER_CONFIG *dst,
                             unsigned char *temp_area,
                             unsigned char temp_height,
diff --git a/libvpx/vpx_scale/vpx_scale.mk b/libvpx/vpx_scale/vpx_scale.mk
index dc89478..76c11e7 100644
--- a/libvpx/vpx_scale/vpx_scale.mk
+++ b/libvpx/vpx_scale/vpx_scale.mk
@@ -1,12 +1,13 @@
 SCALE_SRCS-yes += vpx_scale.mk
-SCALE_SRCS-yes += scale_mode.h
 SCALE_SRCS-yes += yv12config.h
-SCALE_SRCS-yes += vpxscale.h
-SCALE_SRCS-yes += generic/vpxscale.c
+SCALE_SRCS-yes += vpx_scale.h
+SCALE_SRCS-yes += generic/vpx_scale.c
 SCALE_SRCS-yes += generic/yv12config.c
 SCALE_SRCS-yes += generic/yv12extend.c
-SCALE_SRCS-yes += generic/yv12extend_generic.h
 SCALE_SRCS-$(CONFIG_SPATIAL_RESAMPLING) += generic/gen_scalers.c
+SCALE_SRCS-yes += vpx_scale_asm_offsets.c
+SCALE_SRCS-yes += vpx_scale_rtcd.c
+SCALE_SRCS-yes += vpx_scale_rtcd.sh
 
 #neon
 SCALE_SRCS-$(HAVE_NEON)  += arm/neon/vp8_vpxyv12_copyframe_func_neon$(ASM)
@@ -16,3 +17,8 @@
 SCALE_SRCS-$(HAVE_NEON)  += arm/neon/yv12extend_arm.c
 
 SCALE_SRCS-no += $(SCALE_SRCS_REMOVE-yes)
+
+$(eval $(call asm_offsets_template,\
+	         vpx_scale_asm_offsets.asm, vpx_scale/vpx_scale_asm_offsets.c))
+
+$(eval $(call rtcd_h_template,vpx_scale_rtcd,vpx_scale/vpx_scale_rtcd.sh))
diff --git a/libvpx/vp8/common/asm_com_offsets.c b/libvpx/vpx_scale/vpx_scale_asm_offsets.c
similarity index 65%
rename from libvpx/vp8/common/asm_com_offsets.c
rename to libvpx/vpx_scale/vpx_scale_asm_offsets.c
index ae22b5f..caa9e80 100644
--- a/libvpx/vp8/common/asm_com_offsets.c
+++ b/libvpx/vpx_scale/vpx_scale_asm_offsets.c
@@ -9,15 +9,10 @@
  */
 
 
-#include "vpx_config.h"
+#include "./vpx_config.h"
 #include "vpx/vpx_codec.h"
 #include "vpx_ports/asm_offsets.h"
 #include "vpx_scale/yv12config.h"
-#include "vp8/common/blockd.h"
-
-#if CONFIG_POSTPROC
-#include "postproc.h"
-#endif /* CONFIG_POSTPROC */
 
 BEGIN
 
@@ -34,38 +29,12 @@
 DEFINE(yv12_buffer_config_border,               offsetof(YV12_BUFFER_CONFIG, border));
 DEFINE(VP8BORDERINPIXELS_VAL,                   VP8BORDERINPIXELS);
 
-#if CONFIG_POSTPROC
-/* mfqe.c / filter_by_weight */
-DEFINE(MFQE_PRECISION_VAL,                      MFQE_PRECISION);
-#endif /* CONFIG_POSTPROC */
-
 END
 
 /* add asserts for any offset that is not supported by assembly code */
 /* add asserts for any size that is not supported by assembly code */
 
-#if HAVE_MEDIA
-/* switch case in vp8_intra4x4_predict_armv6 is based on these enumerated values */
-ct_assert(B_DC_PRED, B_DC_PRED == 0);
-ct_assert(B_TM_PRED, B_TM_PRED == 1);
-ct_assert(B_VE_PRED, B_VE_PRED == 2);
-ct_assert(B_HE_PRED, B_HE_PRED == 3);
-ct_assert(B_LD_PRED, B_LD_PRED == 4);
-ct_assert(B_RD_PRED, B_RD_PRED == 5);
-ct_assert(B_VR_PRED, B_VR_PRED == 6);
-ct_assert(B_VL_PRED, B_VL_PRED == 7);
-ct_assert(B_HD_PRED, B_HD_PRED == 8);
-ct_assert(B_HU_PRED, B_HU_PRED == 9);
-#endif
-
 #if HAVE_NEON
 /* vp8_yv12_extend_frame_borders_neon makes several assumptions based on this */
 ct_assert(VP8BORDERINPIXELS_VAL, VP8BORDERINPIXELS == 32)
 #endif
-
-#if HAVE_SSE2
-#if CONFIG_POSTPROC
-/* vp8_filter_by_weight16x16 and 8x8 */
-ct_assert(MFQE_PRECISION_VAL, MFQE_PRECISION == 4)
-#endif /* CONFIG_POSTPROC */
-#endif /* HAVE_SSE2 */
diff --git a/libvpx/vpx_scale/vpx_scale_rtcd.c b/libvpx/vpx_scale/vpx_scale_rtcd.c
new file mode 100644
index 0000000..656a22f
--- /dev/null
+++ b/libvpx/vpx_scale/vpx_scale_rtcd.c
@@ -0,0 +1,18 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "vpx_config.h"
+#define RTCD_C
+#include "vpx_scale_rtcd.h"
+#include "vpx_ports/vpx_once.h"
+
+void vpx_scale_rtcd()
+{
+    once(setup_rtcd_internal);
+}
diff --git a/libvpx/vpx_scale/vpx_scale_rtcd.sh b/libvpx/vpx_scale/vpx_scale_rtcd.sh
new file mode 100644
index 0000000..b4f8907
--- /dev/null
+++ b/libvpx/vpx_scale/vpx_scale_rtcd.sh
@@ -0,0 +1,31 @@
+vpx_scale_forward_decls() {
+cat <<EOF
+struct yv12_buffer_config;
+EOF
+}
+forward_decls vpx_scale_forward_decls
+
+# Scaler functions
+if [ "CONFIG_SPATIAL_RESAMPLING" != "yes" ]; then
+    prototype void vp8_horizontal_line_5_4_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
+    prototype void vp8_vertical_band_5_4_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
+    prototype void vp8_horizontal_line_5_3_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
+    prototype void vp8_vertical_band_5_3_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
+    prototype void vp8_horizontal_line_2_1_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
+    prototype void vp8_vertical_band_2_1_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
+    prototype void vp8_vertical_band_2_1_scale_i "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
+fi
+
+prototype void vp8_yv12_extend_frame_borders "struct yv12_buffer_config *ybf"
+specialize vp8_yv12_extend_frame_borders neon
+
+prototype void vp8_yv12_copy_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
+specialize vp8_yv12_copy_frame neon
+
+prototype void vp8_yv12_copy_y "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
+specialize vp8_yv12_copy_y neon
+
+if [ "$CONFIG_VP9" = "yes" ]; then
+    prototype void vp9_extend_frame_borders "struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y"
+    specialize vp9_extend_frame_borders
+fi
diff --git a/libvpx/vpx_scale/win32/scaleopt.c b/libvpx/vpx_scale/win32/scaleopt.c
index 2d96cc7..4336ece 100644
--- a/libvpx/vpx_scale/win32/scaleopt.c
+++ b/libvpx/vpx_scale/win32/scaleopt.c
@@ -18,1184 +18,14 @@
 ****************************************************************************/
 #include "pragmas.h"
 
-
-
 /****************************************************************************
 *  Module Statics
 ****************************************************************************/
-__declspec(align(16)) const static unsigned short one_fifth[]  = { 51, 51, 51, 51 };
-__declspec(align(16)) const static unsigned short two_fifths[] = { 102, 102, 102, 102 };
-__declspec(align(16)) const static unsigned short three_fifths[] = { 154, 154, 154, 154 };
-__declspec(align(16)) const static unsigned short four_fifths[] = { 205, 205, 205, 205 };
 __declspec(align(16)) const static unsigned short round_values[] = { 128, 128, 128, 128 };
-__declspec(align(16)) const static unsigned short four_ones[] = { 1, 1, 1, 1};
-__declspec(align(16)) const static unsigned short const45_2[] = {205, 154, 102,  51 };
-__declspec(align(16)) const static unsigned short const45_1[] = { 51, 102, 154, 205 };
-__declspec(align(16)) const static unsigned char  mask45[] = { 0, 0, 0, 0, 0, 0, 255, 0};
-__declspec(align(16)) const static unsigned short const35_2[] = { 154,  51, 205, 102 };
-__declspec(align(16)) const static unsigned short const35_1[] = { 102, 205,  51, 154 };
 
-
-
-#include "vpx_scale/vpxscale.h"
+#include "vpx_scale/vpx_scale.h"
 #include "vpx_mem/vpx_mem.h"
 
-/****************************************************************************
- *
- *  ROUTINE       : horizontal_line_3_5_scale_mmx
- *
- *  INPUTS        : const unsigned char *source :
- *                  unsigned int source_width    :
- *                  unsigned char *dest         :
- *                  unsigned int dest_width      :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 3 to 5 up-scaling of a horizontal line of pixels.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void horizontal_line_3_5_scale_mmx
-(
-  const unsigned char *source,
-  unsigned int source_width,
-  unsigned char *dest,
-  unsigned int dest_width
-) {
-  (void) dest_width;
-
-  __asm {
-
-    push ebx
-
-    mov         esi,    source
-    mov         edi,    dest
-
-    mov         ecx,    source_width
-    lea         edx,    [esi+ecx-3];
-
-    movq        mm5,    const35_1       // mm5 = 66 xx cd xx 33 xx 9a xx
-    movq        mm6,    const35_2       // mm6 = 9a xx 33 xx cd xx 66 xx
-
-    movq        mm4,    round_values     // mm4 = 80 xx 80 xx 80 xx 80 xx
-    pxor        mm7,    mm7             // clear mm7
-
-    horiz_line_3_5_loop:
-
-    mov        eax,    DWORD PTR [esi] // eax = 00 01 02 03
-    mov        ebx,    eax
-
-    and         ebx,    0xffff00        // ebx = xx 01 02 xx
-    mov         ecx,    eax             // ecx = 00 01 02 03
-
-    and         eax,    0xffff0000      // eax = xx xx 02 03
-    xor         ecx,    eax             // ecx = 00 01 xx xx
-
-    shr         ebx,    8               // ebx = 01 02 xx xx
-    or          eax,    ebx             // eax = 01 02 02 03
-
-    shl         ebx,    16              // ebx = xx xx 01 02
-    movd        mm1,    eax             // mm1 = 01 02 02 03 xx xx xx xx
-
-    or          ebx,    ecx             // ebx = 00 01 01 02
-    punpcklbw   mm1,    mm7             // mm1 = 01 xx 02 xx 02 xx 03 xx
-
-    movd        mm0,    ebx             // mm0 = 00 01 01 02
-    pmullw      mm1,    mm6             //
-
-    punpcklbw   mm0,    mm7             // mm0 = 00 xx 01 xx 01 xx 02 xx
-    pmullw      mm0,    mm5             //
-
-    mov         [edi],  ebx             // writeoutput 00 xx xx xx
-    add         esi,    3
-
-    add         edi,    5
-    paddw       mm0,    mm1
-
-    paddw       mm0,    mm4
-    psrlw       mm0,    8
-
-    cmp         esi,    edx
-    packuswb    mm0,    mm7
-
-    movd        DWORD Ptr [edi-4], mm0
-    jl          horiz_line_3_5_loop
-
-// Exit:
-    mov         eax,    DWORD PTR [esi] // eax = 00 01 02 03
-    mov         ebx,    eax
-
-    and         ebx,    0xffff00        // ebx = xx 01 02 xx
-    mov         ecx,    eax             // ecx = 00 01 02 03
-
-    and         eax,    0xffff0000      // eax = xx xx 02 03
-    xor         ecx,    eax             // ecx = 00 01 xx xx
-
-    shr         ebx,    8               // ebx = 01 02 xx xx
-    or          eax,    ebx             // eax = 01 02 02 03
-
-    shl         eax,    8               // eax = xx 01 02 02
-    and         eax,    0xffff0000      // eax = xx xx 02 02
-
-    or          eax,    ebx             // eax = 01 02 02 02
-
-    shl         ebx,    16              // ebx = xx xx 01 02
-    movd        mm1,    eax             // mm1 = 01 02 02 02 xx xx xx xx
-
-    or          ebx,    ecx             // ebx = 00 01 01 02
-    punpcklbw   mm1,    mm7             // mm1 = 01 xx 02 xx 02 xx 02 xx
-
-    movd        mm0,    ebx             // mm0 = 00 01 01 02
-    pmullw      mm1,    mm6             //
-
-    punpcklbw   mm0,    mm7             // mm0 = 00 xx 01 xx 01 xx 02 xx
-    pmullw      mm0,    mm5             //
-
-    mov         [edi],  ebx             // writeoutput 00 xx xx xx
-    paddw       mm0,    mm1
-
-    paddw       mm0,    mm4
-    psrlw       mm0,    8
-
-    packuswb    mm0,    mm7
-    movd        DWORD Ptr [edi+1], mm0
-
-    pop ebx
-
-  }
-
-}
-
-
-/****************************************************************************
- *
- *  ROUTINE       : horizontal_line_4_5_scale_mmx
- *
- *  INPUTS        : const unsigned char *source :
- *                  unsigned int source_width    :
- *                  unsigned char *dest         :
- *                  unsigned int dest_width      :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 4 to 5 up-scaling of a horizontal line of pixels.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void horizontal_line_4_5_scale_mmx
-(
-  const unsigned char *source,
-  unsigned int source_width,
-  unsigned char *dest,
-  unsigned int dest_width
-) {
-  (void)dest_width;
-
-  __asm {
-
-    mov         esi,    source
-    mov         edi,    dest
-
-    mov         ecx,    source_width
-    lea         edx,    [esi+ecx-8];
-
-    movq        mm5,    const45_1       // mm5 = 33 xx 66 xx 9a xx cd xx
-    movq        mm6,    const45_2       // mm6 = cd xx 9a xx 66 xx 33 xx
-
-    movq        mm4,    round_values     // mm4 = 80 xx 80 xx 80 xx 80 xx
-    pxor        mm7,    mm7             // clear mm7
-
-    horiz_line_4_5_loop:
-
-    movq        mm0,    QWORD PTR [esi]           // mm0 = 00 01 02 03 04 05 06 07
-    movq        mm1,    QWORD PTR [esi+1];        // mm1 = 01 02 03 04 05 06 07 08
-
-    movq        mm2,    mm0             // mm2 = 00 01 02 03 04 05 06 07
-    movq        mm3,    mm1             // mm3 = 01 02 03 04 05 06 07 08
-
-    movd        DWORD PTR [edi],  mm0             // write output 00 xx xx xx
-    punpcklbw   mm0,    mm7             // mm0 = 00 xx 01 xx 02 xx 03 xx
-
-    punpcklbw   mm1,    mm7             // mm1 = 01 xx 02 xx 03 xx 04 xx
-    pmullw      mm0,    mm5             // 00* 51 01*102 02*154 03*205
-
-    pmullw      mm1,    mm6             // 01*205 02*154 03*102 04* 51
-    punpckhbw   mm2,    mm7             // mm2 = 04 xx 05 xx 06 xx 07 xx
-
-    movd        DWORD PTR [edi+5], mm2            // write ouput 05 xx xx xx
-    pmullw      mm2,    mm5             // 04* 51 05*102 06*154 07*205
-
-    punpckhbw   mm3,    mm7             // mm3 = 05 xx 06 xx 07 xx 08 xx
-    pmullw      mm3,    mm6             // 05*205 06*154 07*102 08* 51
-
-    paddw       mm0,    mm1             // added round values
-    paddw       mm0,    mm4
-
-    psrlw       mm0,    8               // output: 01 xx 02 xx 03 xx 04 xx
-    packuswb    mm0,    mm7
-
-    movd        DWORD PTR [edi+1], mm0  // write output 01 02 03 04
-    add         edi,    10
-
-    add         esi,    8
-    paddw       mm2,    mm3             //
-
-    paddw       mm2,    mm4             // added round values
-    cmp         esi,    edx
-
-    psrlw       mm2,    8
-    packuswb    mm2,    mm7
-
-    movd        DWORD PTR [edi-4], mm2 // writeoutput 06 07 08 09
-    jl         horiz_line_4_5_loop
-
-// Exit:
-    movq        mm0,    [esi]           // mm0 = 00 01 02 03 04 05 06 07
-    movq        mm1,    mm0             // mm1 = 00 01 02 03 04 05 06 07
-
-    movq        mm2,    mm0             // mm2 = 00 01 02 03 04 05 06 07
-    psrlq       mm1,    8               // mm1 = 01 02 03 04 05 06 07 00
-
-    movq        mm3,    mask45          // mm3 = 00 00 00 00 00 00 ff 00
-    pand        mm3,    mm1             // mm3 = 00 00 00 00 00 00 07 00
-
-    psllq       mm3,    8               // mm3 = 00 00 00 00 00 00 00 07
-    por         mm1,    mm3             // mm1 = 01 02 03 04 05 06 07 07
-
-    movq        mm3,    mm1
-
-    movd        DWORD PTR [edi],  mm0   // write output 00 xx xx xx
-    punpcklbw   mm0,    mm7             // mm0 = 00 xx 01 xx 02 xx 03 xx
-
-    punpcklbw   mm1,    mm7             // mm1 = 01 xx 02 xx 03 xx 04 xx
-    pmullw      mm0,    mm5             // 00* 51 01*102 02*154 03*205
-
-    pmullw      mm1,    mm6             // 01*205 02*154 03*102 04* 51
-    punpckhbw   mm2,    mm7             // mm2 = 04 xx 05 xx 06 xx 07 xx
-
-    movd        DWORD PTR [edi+5], mm2  // write ouput 05 xx xx xx
-    pmullw      mm2,    mm5             // 04* 51 05*102 06*154 07*205
-
-    punpckhbw   mm3,    mm7             // mm3 = 05 xx 06 xx 07 xx 08 xx
-    pmullw      mm3,    mm6             // 05*205 06*154 07*102 07* 51
-
-    paddw       mm0,    mm1             // added round values
-    paddw       mm0,    mm4
-
-    psrlw       mm0,    8               // output: 01 xx 02 xx 03 xx 04 xx
-    packuswb    mm0,    mm7             // 01 02 03 04 xx xx xx xx
-
-    movd        DWORD PTR [edi+1], mm0  // write output 01 02 03 04
-    paddw       mm2,    mm3             //
-
-    paddw       mm2,    mm4             // added round values
-    psrlw       mm2,    8
-
-    packuswb    mm2,    mm7
-    movd        DWORD PTR [edi+6], mm2  // writeoutput 06 07 08 09
-
-
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vertical_band_4_5_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 4 to 5 up-scaling of a 4 pixel high band of pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has a "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void vertical_band_4_5_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    lea         edi,    [esi+ecx*2]             // tow lines below
-    add         edi,    ecx                     // three lines below
-
-    pxor        mm7,    mm7                     // clear out mm7
-    mov         edx,    dest_width               // Loop counter
-
-    vs_4_5_loop:
-
-    movq        mm0,    QWORD ptr [esi]         // src[0];
-    movq        mm1,    QWORD ptr [esi+ecx]     // src[1];
-
-    movq        mm2,    mm0                     // Make a copy
-    punpcklbw   mm0,    mm7                     // unpack low to word
-
-    movq        mm5,    one_fifth
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm0,    mm5                     // a * 1/5
-
-    movq        mm3,    mm1                     // make a copy
-    punpcklbw   mm1,    mm7                     // unpack low to word
-
-    pmullw      mm2,    mm5                     // a * 1/5
-    movq        mm6,    four_fifths               // constan
-
-    movq        mm4,    mm1                     // copy of low b
-    pmullw      mm4,    mm6                     // b * 4/5
-
-    punpckhbw   mm3,    mm7                     // unpack high to word
-    movq        mm5,    mm3                     // copy of high b
-
-    pmullw      mm5,    mm6                     // b * 4/5
-    paddw       mm0,    mm4                     // a * 1/5 + b * 4/5
-
-    paddw       mm2,    mm5                     // a * 1/5 + b * 4/5
-    paddw       mm0,    round_values             // + 128
-
-    paddw       mm2,    round_values             // + 128
-    psrlw       mm0,    8
-
-    psrlw       mm2,    8
-    packuswb    mm0,    mm2                     // des [1]
-
-    movq        QWORD ptr [esi+ecx], mm0        // write des[1]
-    movq        mm0,    [esi+ecx*2]             // mm0 = src[2]
-
-    // mm1, mm3 --- Src[1]
-    // mm0 --- Src[2]
-    // mm7 for unpacking
-
-    movq        mm5,    two_fifths
-    movq        mm2,    mm0                     // make a copy
-
-    pmullw      mm1,    mm5                     // b * 2/5
-    movq        mm6,    three_fifths
-
-
-    punpcklbw   mm0,    mm7                     // unpack low to word
-    pmullw      mm3,    mm5                     // b * 2/5
-
-    movq        mm4,    mm0                     // make copy of c
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm4,    mm6                     // c * 3/5
-    movq        mm5,    mm2
-
-    pmullw      mm5,    mm6                     // c * 3/5
-    paddw       mm1,    mm4                     // b * 2/5 + c * 3/5
-
-    paddw       mm3,    mm5                     // b * 2/5 + c * 3/5
-    paddw       mm1,    round_values             // + 128
-
-    paddw       mm3,    round_values             // + 128
-    psrlw       mm1,    8
-
-    psrlw       mm3,    8
-    packuswb    mm1,    mm3                     // des[2]
-
-    movq        QWORD ptr [esi+ecx*2], mm1      // write des[2]
-    movq        mm1,    [edi]                   // mm1=Src[3];
-
-    // mm0, mm2 --- Src[2]
-    // mm1 --- Src[3]
-    // mm6 --- 3/5
-    // mm7 for unpacking
-
-    pmullw      mm0,    mm6                     // c * 3/5
-    movq        mm5,    two_fifths               // mm5 = 2/5
-
-    movq        mm3,    mm1                     // make a copy
-    pmullw      mm2,    mm6                     // c * 3/5
-
-    punpcklbw   mm1,    mm7                     // unpack low
-    movq        mm4,    mm1                     // make a copy
-
-    punpckhbw   mm3,    mm7                     // unpack high
-    pmullw      mm4,    mm5                     // d * 2/5
-
-    movq        mm6,    mm3                     // make a copy
-    pmullw      mm6,    mm5                     // d * 2/5
-
-    paddw       mm0,    mm4                     // c * 3/5 + d * 2/5
-    paddw       mm2,    mm6                     // c * 3/5 + d * 2/5
-
-    paddw       mm0,    round_values             // + 128
-    paddw       mm2,    round_values             // + 128
-
-    psrlw       mm0,    8
-    psrlw       mm2,    8
-
-    packuswb    mm0,    mm2                     // des[3]
-    movq        QWORD ptr [edi], mm0            // write des[3]
-
-    //  mm1, mm3 --- Src[3]
-    //  mm7 -- cleared for unpacking
-
-    movq        mm0,    [edi+ecx*2]             // mm0, Src[0] of the next group
-
-    movq        mm5,    four_fifths              // mm5 = 4/5
-    pmullw      mm1,    mm5                     // d * 4/5
-
-    movq        mm6,    one_fifth                // mm6 = 1/5
-    movq        mm2,    mm0                     // make a copy
-
-    pmullw      mm3,    mm5                     // d * 4/5
-    punpcklbw   mm0,    mm7                     // unpack low
-
-    pmullw      mm0,    mm6                     // an * 1/5
-    punpckhbw   mm2,    mm7                     // unpack high
-
-    paddw       mm1,    mm0                     // d * 4/5 + an * 1/5
-    pmullw      mm2,    mm6                     // an * 1/5
-
-    paddw       mm3,    mm2                     // d * 4/5 + an * 1/5
-    paddw       mm1,    round_values             // + 128
-
-    paddw       mm3,    round_values             // + 128
-    psrlw       mm1,    8
-
-    psrlw       mm3,    8
-    packuswb    mm1,    mm3                     // des[4]
-
-    movq        QWORD ptr [edi+ecx], mm1        // write des[4]
-
-    add         edi,    8
-    add         esi,    8
-
-    sub         edx,    8
-    jg         vs_4_5_loop
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : last_vertical_band_4_5_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : None
- *
- *  FUNCTION      : 4 to 5 up-scaling of the last 4-pixel high band in an image.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has an "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void last_vertical_band_4_5_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    lea         edi,    [esi+ecx*2]             // tow lines below
-    add         edi,    ecx                     // three lines below
-
-    pxor        mm7,    mm7                     // clear out mm7
-    mov         edx,    dest_width               // Loop counter
-
-    last_vs_4_5_loop:
-
-    movq        mm0,    QWORD ptr [esi]         // src[0];
-    movq        mm1,    QWORD ptr [esi+ecx]     // src[1];
-
-    movq        mm2,    mm0                     // Make a copy
-    punpcklbw   mm0,    mm7                     // unpack low to word
-
-    movq        mm5,    one_fifth
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm0,    mm5                     // a * 1/5
-
-    movq        mm3,    mm1                     // make a copy
-    punpcklbw   mm1,    mm7                     // unpack low to word
-
-    pmullw      mm2,    mm5                     // a * 1/5
-    movq        mm6,    four_fifths               // constan
-
-    movq        mm4,    mm1                     // copy of low b
-    pmullw      mm4,    mm6                     // b * 4/5
-
-    punpckhbw   mm3,    mm7                     // unpack high to word
-    movq        mm5,    mm3                     // copy of high b
-
-    pmullw      mm5,    mm6                     // b * 4/5
-    paddw       mm0,    mm4                     // a * 1/5 + b * 4/5
-
-    paddw       mm2,    mm5                     // a * 1/5 + b * 4/5
-    paddw       mm0,    round_values             // + 128
-
-    paddw       mm2,    round_values             // + 128
-    psrlw       mm0,    8
-
-    psrlw       mm2,    8
-    packuswb    mm0,    mm2                     // des [1]
-
-    movq        QWORD ptr [esi+ecx], mm0        // write des[1]
-    movq        mm0,    [esi+ecx*2]             // mm0 = src[2]
-
-    // mm1, mm3 --- Src[1]
-    // mm0 --- Src[2]
-    // mm7 for unpacking
-
-    movq        mm5,    two_fifths
-    movq        mm2,    mm0                     // make a copy
-
-    pmullw      mm1,    mm5                     // b * 2/5
-    movq        mm6,    three_fifths
-
-
-    punpcklbw   mm0,    mm7                     // unpack low to word
-    pmullw      mm3,    mm5                     // b * 2/5
-
-    movq        mm4,    mm0                     // make copy of c
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm4,    mm6                     // c * 3/5
-    movq        mm5,    mm2
-
-    pmullw      mm5,    mm6                     // c * 3/5
-    paddw       mm1,    mm4                     // b * 2/5 + c * 3/5
-
-    paddw       mm3,    mm5                     // b * 2/5 + c * 3/5
-    paddw       mm1,    round_values             // + 128
-
-    paddw       mm3,    round_values             // + 128
-    psrlw       mm1,    8
-
-    psrlw       mm3,    8
-    packuswb    mm1,    mm3                     // des[2]
-
-    movq        QWORD ptr [esi+ecx*2], mm1      // write des[2]
-    movq        mm1,    [edi]                   // mm1=Src[3];
-
-    movq        QWORD ptr [edi+ecx], mm1        // write des[4];
-
-    // mm0, mm2 --- Src[2]
-    // mm1 --- Src[3]
-    // mm6 --- 3/5
-    // mm7 for unpacking
-
-    pmullw      mm0,    mm6                     // c * 3/5
-    movq        mm5,    two_fifths               // mm5 = 2/5
-
-    movq        mm3,    mm1                     // make a copy
-    pmullw      mm2,    mm6                     // c * 3/5
-
-    punpcklbw   mm1,    mm7                     // unpack low
-    movq        mm4,    mm1                     // make a copy
-
-    punpckhbw   mm3,    mm7                     // unpack high
-    pmullw      mm4,    mm5                     // d * 2/5
-
-    movq        mm6,    mm3                     // make a copy
-    pmullw      mm6,    mm5                     // d * 2/5
-
-    paddw       mm0,    mm4                     // c * 3/5 + d * 2/5
-    paddw       mm2,    mm6                     // c * 3/5 + d * 2/5
-
-    paddw       mm0,    round_values             // + 128
-    paddw       mm2,    round_values             // + 128
-
-    psrlw       mm0,    8
-    psrlw       mm2,    8
-
-    packuswb    mm0,    mm2                     // des[3]
-    movq        QWORD ptr [edi], mm0            // write des[3]
-
-    //  mm1, mm3 --- Src[3]
-    //  mm7 -- cleared for unpacking
-    add         edi,    8
-    add         esi,    8
-
-    sub         edx,    8
-    jg          last_vs_4_5_loop
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vertical_band_3_5_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 3 to 5 up-scaling of a 3-pixel high band of pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has an "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void vertical_band_3_5_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    lea         edi,    [esi+ecx*2]             // tow lines below
-    add         edi,    ecx                     // three lines below
-
-    pxor        mm7,    mm7                     // clear out mm7
-    mov         edx,    dest_width               // Loop counter
-
-    vs_3_5_loop:
-
-    movq        mm0,    QWORD ptr [esi]         // src[0];
-    movq        mm1,    QWORD ptr [esi+ecx]     // src[1];
-
-    movq        mm2,    mm0                     // Make a copy
-    punpcklbw   mm0,    mm7                     // unpack low to word
-
-    movq        mm5,    two_fifths               // mm5 = 2/5
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm0,    mm5                     // a * 2/5
-
-    movq        mm3,    mm1                     // make a copy
-    punpcklbw   mm1,    mm7                     // unpack low to word
-
-    pmullw      mm2,    mm5                     // a * 2/5
-    movq        mm6,    three_fifths             // mm6 = 3/5
-
-    movq        mm4,    mm1                     // copy of low b
-    pmullw      mm4,    mm6                     // b * 3/5
-
-    punpckhbw   mm3,    mm7                     // unpack high to word
-    movq        mm5,    mm3                     // copy of high b
-
-    pmullw      mm5,    mm6                     // b * 3/5
-    paddw       mm0,    mm4                     // a * 2/5 + b * 3/5
-
-    paddw       mm2,    mm5                     // a * 2/5 + b * 3/5
-    paddw       mm0,    round_values             // + 128
-
-    paddw       mm2,    round_values             // + 128
-    psrlw       mm0,    8
-
-    psrlw       mm2,    8
-    packuswb    mm0,    mm2                     // des [1]
-
-    movq        QWORD ptr [esi+ecx], mm0        // write des[1]
-    movq        mm0,    [esi+ecx*2]             // mm0 = src[2]
-
-    // mm1, mm3 --- Src[1]
-    // mm0 --- Src[2]
-    // mm7 for unpacking
-
-    movq        mm4,    mm1                     // b low
-    pmullw      mm1,    four_fifths              // b * 4/5 low
-
-    movq        mm5,    mm3                     // b high
-    pmullw      mm3,    four_fifths              // b * 4/5 high
-
-    movq        mm2,    mm0                     // c
-    pmullw      mm4,    one_fifth                // b * 1/5
-
-    punpcklbw   mm0,    mm7                     // c low
-    pmullw      mm5,    one_fifth                // b * 1/5
-
-    movq        mm6,    mm0                     // make copy of c low
-    punpckhbw   mm2,    mm7                     // c high
-
-    pmullw      mm6,    one_fifth                // c * 1/5 low
-    movq        mm7,    mm2                     // make copy of c high
-
-    pmullw      mm7,    one_fifth                // c * 1/5 high
-    paddw       mm1,    mm6                     // b * 4/5 + c * 1/5 low
-
-    paddw       mm3,    mm7                     // b * 4/5 + c * 1/5 high
-    movq        mm6,    mm0                     // make copy of c low
-
-    pmullw      mm6,    four_fifths              // c * 4/5 low
-    movq        mm7,    mm2                     // make copy of c high
-
-    pmullw      mm7,    four_fifths              // c * 4/5 high
-
-    paddw       mm4,    mm6                     // b * 1/5 + c * 4/5 low
-    paddw       mm5,    mm7                     // b * 1/5 + c * 4/5 high
-
-    paddw       mm1,    round_values             // + 128
-    paddw       mm3,    round_values             // + 128
-
-    psrlw       mm1,    8
-    psrlw       mm3,    8
-
-    packuswb    mm1,    mm3                     // des[2]
-    movq        QWORD ptr [esi+ecx*2], mm1      // write des[2]
-
-    paddw       mm4,    round_values             // + 128
-    paddw       mm5,    round_values             // + 128
-
-    psrlw       mm4,    8
-    psrlw       mm5,    8
-
-    packuswb    mm4,    mm5                     // des[3]
-    movq        QWORD ptr [edi], mm4            // write des[3]
-
-    //  mm0, mm2 --- Src[3]
-
-    pxor        mm7,    mm7                     // clear mm7 for unpacking
-    movq        mm1,    [edi+ecx*2]             // mm1 = Src[0] of the next group
-
-    movq        mm5,    three_fifths             // mm5 = 3/5
-    pmullw      mm0,    mm5                     // d * 3/5
-
-    movq        mm6,    two_fifths                // mm6 = 2/5
-    movq        mm3,    mm1                     // make a copy
-
-    pmullw      mm2,    mm5                     // d * 3/5
-    punpcklbw   mm1,    mm7                     // unpack low
-
-    pmullw      mm1,    mm6                     // an * 2/5
-    punpckhbw   mm3,    mm7                     // unpack high
-
-    paddw       mm0,    mm1                     // d * 3/5 + an * 2/5
-    pmullw      mm3,    mm6                     // an * 2/5
-
-    paddw       mm2,    mm3                     // d * 3/5 + an * 2/5
-    paddw       mm0,    round_values             // + 128
-
-    paddw       mm2,    round_values             // + 128
-    psrlw       mm0,    8
-
-    psrlw       mm2,    8
-    packuswb    mm0,    mm2                     // des[4]
-
-    movq        QWORD ptr [edi+ecx], mm0        // write des[4]
-
-    add         edi,    8
-    add         esi,    8
-
-    sub         edx,    8
-    jg          vs_3_5_loop
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : last_vertical_band_3_5_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 3 to 5 up-scaling of a 3-pixel high band of pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has an "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void last_vertical_band_3_5_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    lea         edi,    [esi+ecx*2]             // tow lines below
-    add         edi,    ecx                     // three lines below
-
-    pxor        mm7,    mm7                     // clear out mm7
-    mov         edx,    dest_width               // Loop counter
-
-
-    last_vs_3_5_loop:
-
-    movq        mm0,    QWORD ptr [esi]         // src[0];
-    movq        mm1,    QWORD ptr [esi+ecx]     // src[1];
-
-    movq        mm2,    mm0                     // Make a copy
-    punpcklbw   mm0,    mm7                     // unpack low to word
-
-    movq        mm5,    two_fifths               // mm5 = 2/5
-    punpckhbw   mm2,    mm7                     // unpack high to word
-
-    pmullw      mm0,    mm5                     // a * 2/5
-
-    movq        mm3,    mm1                     // make a copy
-    punpcklbw   mm1,    mm7                     // unpack low to word
-
-    pmullw      mm2,    mm5                     // a * 2/5
-    movq        mm6,    three_fifths             // mm6 = 3/5
-
-    movq        mm4,    mm1                     // copy of low b
-    pmullw      mm4,    mm6                     // b * 3/5
-
-    punpckhbw   mm3,    mm7                     // unpack high to word
-    movq        mm5,    mm3                     // copy of high b
-
-    pmullw      mm5,    mm6                     // b * 3/5
-    paddw       mm0,    mm4                     // a * 2/5 + b * 3/5
-
-    paddw       mm2,    mm5                     // a * 2/5 + b * 3/5
-    paddw       mm0,    round_values             // + 128
-
-    paddw       mm2,    round_values             // + 128
-    psrlw       mm0,    8
-
-    psrlw       mm2,    8
-    packuswb    mm0,    mm2                     // des [1]
-
-    movq        QWORD ptr [esi+ecx], mm0        // write des[1]
-    movq        mm0,    [esi+ecx*2]             // mm0 = src[2]
-
-
-
-    // mm1, mm3 --- Src[1]
-    // mm0 --- Src[2]
-    // mm7 for unpacking
-
-    movq        mm4,    mm1                     // b low
-    pmullw      mm1,    four_fifths              // b * 4/5 low
-
-    movq        QWORD ptr [edi+ecx], mm0        // write des[4]
-
-    movq        mm5,    mm3                     // b high
-    pmullw      mm3,    four_fifths              // b * 4/5 high
-
-    movq        mm2,    mm0                     // c
-    pmullw      mm4,    one_fifth                // b * 1/5
-
-    punpcklbw   mm0,    mm7                     // c low
-    pmullw      mm5,    one_fifth                // b * 1/5
-
-    movq        mm6,    mm0                     // make copy of c low
-    punpckhbw   mm2,    mm7                     // c high
-
-    pmullw      mm6,    one_fifth                // c * 1/5 low
-    movq        mm7,    mm2                     // make copy of c high
-
-    pmullw      mm7,    one_fifth                // c * 1/5 high
-    paddw       mm1,    mm6                     // b * 4/5 + c * 1/5 low
-
-    paddw       mm3,    mm7                     // b * 4/5 + c * 1/5 high
-    movq        mm6,    mm0                     // make copy of c low
-
-    pmullw      mm6,    four_fifths              // c * 4/5 low
-    movq        mm7,    mm2                     // make copy of c high
-
-    pmullw      mm7,    four_fifths              // c * 4/5 high
-
-    paddw       mm4,    mm6                     // b * 1/5 + c * 4/5 low
-    paddw       mm5,    mm7                     // b * 1/5 + c * 4/5 high
-
-    paddw       mm1,    round_values             // + 128
-    paddw       mm3,    round_values             // + 128
-
-    psrlw       mm1,    8
-    psrlw       mm3,    8
-
-    packuswb    mm1,    mm3                     // des[2]
-    movq        QWORD ptr [esi+ecx*2], mm1      // write des[2]
-
-    paddw       mm4,    round_values             // + 128
-    paddw       mm5,    round_values             // + 128
-
-    psrlw       mm4,    8
-    psrlw       mm5,    8
-
-    packuswb    mm4,    mm5                     // des[3]
-    movq        QWORD ptr [edi], mm4            // write des[3]
-
-    //  mm0, mm2 --- Src[3]
-
-    add         edi,    8
-    add         esi,    8
-
-    sub         edx,    8
-    jg          last_vs_3_5_loop
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : vertical_band_1_2_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 1 to 2 up-scaling of a band of pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has an "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void vertical_band_1_2_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    pxor        mm7,    mm7                     // clear out mm7
-    mov         edx,    dest_width               // Loop counter
-
-    vs_1_2_loop:
-
-    movq        mm0,    [esi]                   // get Src[0]
-    movq        mm1,    [esi + ecx * 2]         // get Src[1]
-
-    movq        mm2,    mm0                     // make copy before unpack
-    movq        mm3,    mm1                     // make copy before unpack
-
-    punpcklbw   mm0,    mm7                     // low Src[0]
-    movq        mm6,    four_ones                // mm6= 1, 1, 1, 1
-
-    punpcklbw   mm1,    mm7                     // low Src[1]
-    paddw       mm0,    mm1                     // low (a + b)
-
-    punpckhbw   mm2,    mm7                     // high Src[0]
-    paddw       mm0,    mm6                     // low (a + b + 1)
-
-    punpckhbw   mm3,    mm7
-    paddw       mm2,    mm3                     // high (a + b )
-
-    psraw       mm0,    1                       // low (a + b +1 )/2
-    paddw       mm2,    mm6                     // high (a + b + 1)
-
-    psraw       mm2,    1                       // high (a + b + 1)/2
-    packuswb    mm0,    mm2                     // pack results
-
-    movq        [esi+ecx], mm0                  // write out eight bytes
-    add         esi,    8
-
-    sub         edx,    8
-    jg          vs_1_2_loop
-  }
-
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : last_vertical_band_1_2_scale_mmx
- *
- *  INPUTS        : unsigned char *dest    :
- *                  unsigned int dest_pitch :
- *                  unsigned int dest_width :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 1 to 2 up-scaling of band of pixels.
- *
- *  SPECIAL NOTES : The routine uses the first line of the band below
- *                  the current band. The function also has an "C" only
- *                  version.
- *
- ****************************************************************************/
-static
-void last_vertical_band_1_2_scale_mmx
-(
-  unsigned char *dest,
-  unsigned int dest_pitch,
-  unsigned int dest_width
-) {
-  __asm {
-    mov         esi,    dest                    // Get the source and destination pointer
-    mov         ecx,    dest_pitch               // Get the pitch size
-
-    mov         edx,    dest_width               // Loop counter
-
-    last_vs_1_2_loop:
-
-    movq        mm0,    [esi]                   // get Src[0]
-    movq        [esi+ecx], mm0                  // write out eight bytes
-
-    add         esi,    8
-    sub         edx,    8
-
-    jg         last_vs_1_2_loop
-  }
-}
-
-/****************************************************************************
- *
- *  ROUTINE       : horizontal_line_1_2_scale
- *
- *  INPUTS        : const unsigned char *source :
- *                  unsigned int source_width    :
- *                  unsigned char *dest         :
- *                  unsigned int dest_width      :
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : 1 to 2 up-scaling of a horizontal line of pixels.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-static
-void horizontal_line_1_2_scale_mmx
-(
-  const unsigned char *source,
-  unsigned int source_width,
-  unsigned char *dest,
-  unsigned int dest_width
-) {
-  (void) dest_width;
-
-  __asm {
-    mov         esi,    source
-    mov         edi,    dest
-
-    pxor        mm7,    mm7
-    movq        mm6,    four_ones
-
-    mov         ecx,    source_width
-
-    hs_1_2_loop:
-
-    movq        mm0,    [esi]
-    movq        mm1,    [esi+1]
-
-    movq        mm2,    mm0
-    movq        mm3,    mm1
-
-    movq        mm4,    mm0
-    punpcklbw   mm0,    mm7
-
-    punpcklbw   mm1,    mm7
-    paddw       mm0,    mm1
-
-    paddw       mm0,    mm6
-    punpckhbw   mm2,    mm7
-
-    punpckhbw   mm3,    mm7
-    paddw       mm2,    mm3
-
-    paddw       mm2,    mm6
-    psraw       mm0,    1
-
-    psraw       mm2,    1
-    packuswb    mm0,    mm2
-
-    movq        mm2,    mm4
-    punpcklbw   mm2,    mm0
-
-    movq        [edi],  mm2
-    punpckhbw   mm4,    mm0
-
-    movq        [edi+8], mm4
-    add         esi,    8
-
-    add         edi,    16
-    sub         ecx,    8
-
-    cmp         ecx,    8
-    jg          hs_1_2_loop
-
-// last eight pixel
-
-    movq        mm0,    [esi]
-    movq        mm1,    mm0
-
-    movq        mm2,    mm0
-    movq        mm3,    mm1
-
-    psrlq       mm1,    8
-    psrlq       mm3,    56
-
-    psllq       mm3,    56
-    por         mm1,    mm3
-
-    movq        mm3,    mm1
-    movq        mm4,    mm0
-
-    punpcklbw   mm0,    mm7
-    punpcklbw   mm1,    mm7
-
-    paddw       mm0,    mm1
-    paddw       mm0,    mm6
-
-    punpckhbw   mm2,    mm7
-    punpckhbw   mm3,    mm7
-
-    paddw       mm2,    mm3
-    paddw       mm2,    mm6
-
-    psraw       mm0,    1
-    psraw       mm2,    1
-
-    packuswb    mm0,    mm2
-    movq        mm2,    mm4
-
-    punpcklbw   mm2,    mm0
-    movq        [edi],  mm2
-
-    punpckhbw   mm4,    mm0
-    movq        [edi+8], mm4
-  }
-}
-
-
-
-
-
 __declspec(align(16)) const static unsigned short const54_2[] = {  0,  64, 128, 192 };
 __declspec(align(16)) const static unsigned short const54_1[] = {256, 192, 128,  64 };
 
@@ -1685,25 +515,6 @@
 
 void
 register_mmxscalers(void) {
-  vp8_horizontal_line_1_2_scale        = horizontal_line_1_2_scale_mmx;
-  vp8_vertical_band_1_2_scale          = vertical_band_1_2_scale_mmx;
-  vp8_last_vertical_band_1_2_scale      = last_vertical_band_1_2_scale_mmx;
-  vp8_horizontal_line_3_5_scale        = horizontal_line_3_5_scale_mmx;
-  vp8_vertical_band_3_5_scale          = vertical_band_3_5_scale_mmx;
-  vp8_last_vertical_band_3_5_scale      = last_vertical_band_3_5_scale_mmx;
-  vp8_horizontal_line_4_5_scale        = horizontal_line_4_5_scale_mmx;
-  vp8_vertical_band_4_5_scale          = vertical_band_4_5_scale_mmx;
-  vp8_last_vertical_band_4_5_scale      = last_vertical_band_4_5_scale_mmx;
-
-  vp8_horizontal_line_3_4_scale        = vp8cx_horizontal_line_3_4_scale_c;
-  vp8_vertical_band_3_4_scale          = vp8cx_vertical_band_3_4_scale_c;
-  vp8_last_vertical_band_3_4_scale      = vp8cx_last_vertical_band_3_4_scale_c;
-  vp8_horizontal_line_2_3_scale        = vp8cx_horizontal_line_2_3_scale_c;
-  vp8_vertical_band_2_3_scale          = vp8cx_vertical_band_2_3_scale_c;
-  vp8_last_vertical_band_2_3_scale      = vp8cx_last_vertical_band_2_3_scale_c;
-
-
-
   vp8_vertical_band_5_4_scale           = vertical_band_5_4_scale_mmx;
   vp8_vertical_band_5_3_scale           = vertical_band_5_3_scale_mmx;
   vp8_vertical_band_2_1_scale           = vertical_band_2_1_scale_mmx;
@@ -1711,8 +522,4 @@
   vp8_horizontal_line_2_1_scale         = horizontal_line_2_1_scale_mmx;
   vp8_horizontal_line_5_3_scale         = horizontal_line_5_3_scale_mmx;
   vp8_horizontal_line_5_4_scale         = horizontal_line_5_4_scale_mmx;
-
-
-
-
 }
diff --git a/libvpx/vpx_scale/win32/scalesystemdependent.c b/libvpx/vpx_scale/win32/scalesystemdependent.c
deleted file mode 100644
index 98913d1..0000000
--- a/libvpx/vpx_scale/win32/scalesystemdependent.c
+++ /dev/null
@@ -1,87 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-
-/****************************************************************************
-*
-*   Module Title :     system_dependent.c
-*
-*   Description  :     Miscellaneous system dependent functions
-*
-****************************************************************************/
-
-/****************************************************************************
-*  Header Files
-****************************************************************************/
-#include "vpx_scale/vpxscale.h"
-#include "cpuidlib.h"
-
-/****************************************************************************
-*  Imports
-*****************************************************************************/
-extern void register_generic_scalers(void);
-extern void register_mmxscalers(void);
-
-/****************************************************************************
- *
- *  ROUTINE       : post_proc_machine_specific_config
- *
- *  INPUTS        : UINT32 Version : Codec version number.
- *
- *  OUTPUTS       : None.
- *
- *  RETURNS       : void
- *
- *  FUNCTION      : Checks for machine specifc features such as MMX support
- *                  sets appropriate flags and function pointers.
- *
- *  SPECIAL NOTES : None.
- *
- ****************************************************************************/
-void
-vp8_scale_machine_specific_config(void) {
-  // If MMX supported then set to use MMX versions of functions else
-  // use original 'C' versions.
-  int mmx_enabled;
-  int xmm_enabled;
-  int wmt_enabled;
-
-  vpx_get_processor_flags(&mmx_enabled, &xmm_enabled, &wmt_enabled);
-
-  if (mmx_enabled || xmm_enabled || wmt_enabled) {
-    register_mmxscalers();
-  } else {
-    vp8_horizontal_line_1_2_scale        = vp8cx_horizontal_line_1_2_scale_c;
-    vp8_vertical_band_1_2_scale          = vp8cx_vertical_band_1_2_scale_c;
-    vp8_last_vertical_band_1_2_scale      = vp8cx_last_vertical_band_1_2_scale_c;
-    vp8_horizontal_line_3_5_scale        = vp8cx_horizontal_line_3_5_scale_c;
-    vp8_vertical_band_3_5_scale          = vp8cx_vertical_band_3_5_scale_c;
-    vp8_last_vertical_band_3_5_scale      = vp8cx_last_vertical_band_3_5_scale_c;
-    vp8_horizontal_line_3_4_scale        = vp8cx_horizontal_line_3_4_scale_c;
-    vp8_vertical_band_3_4_scale          = vp8cx_vertical_band_3_4_scale_c;
-    vp8_last_vertical_band_3_4_scale      = vp8cx_last_vertical_band_3_4_scale_c;
-    vp8_horizontal_line_2_3_scale        = vp8cx_horizontal_line_2_3_scale_c;
-    vp8_vertical_band_2_3_scale          = vp8cx_vertical_band_2_3_scale_c;
-    vp8_last_vertical_band_2_3_scale      = vp8cx_last_vertical_band_2_3_scale_c;
-    vp8_horizontal_line_4_5_scale        = vp8cx_horizontal_line_4_5_scale_c;
-    vp8_vertical_band_4_5_scale          = vp8cx_vertical_band_4_5_scale_c;
-    vp8_last_vertical_band_4_5_scale      = vp8cx_last_vertical_band_4_5_scale_c;
-
-
-    vp8_vertical_band_5_4_scale           = vp8cx_vertical_band_5_4_scale_c;
-    vp8_vertical_band_5_3_scale           = vp8cx_vertical_band_5_3_scale_c;
-    vp8_vertical_band_2_1_scale           = vp8cx_vertical_band_2_1_scale_c;
-    vp8_vertical_band_2_1_scale_i         = vp8cx_vertical_band_2_1_scale_i_c;
-    vp8_horizontal_line_2_1_scale         = vp8cx_horizontal_line_2_1_scale_c;
-    vp8_horizontal_line_5_3_scale         = vp8cx_horizontal_line_5_3_scale_c;
-    vp8_horizontal_line_5_4_scale         = vp8cx_horizontal_line_5_4_scale_c;
-
-  }
-}
diff --git a/libvpx/vpx_scale/yv12config.h b/libvpx/vpx_scale/yv12config.h
index 6a8a1fc..7b8bd85 100644
--- a/libvpx/vpx_scale/yv12config.h
+++ b/libvpx/vpx_scale/yv12config.h
@@ -8,16 +8,17 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
 #ifndef YV12_CONFIG_H
 #define YV12_CONFIG_H
+
 #ifdef __cplusplus
-extern "C"
-{
+extern "C" {
 #endif
 
+#include "vpx/vpx_integer.h"
+
 #define VP8BORDERINPIXELS       32
-#define VP9BORDERINPIXELS       64
+#define VP9BORDERINPIXELS       96
 #define VP9_INTERP_EXTEND        4
 
   /*************************************
@@ -41,6 +42,8 @@
   typedef struct yv12_buffer_config {
     int   y_width;
     int   y_height;
+    int   y_crop_width;
+    int   y_crop_height;
     int   y_stride;
     /*    int   yinternal_width; */
 
@@ -49,11 +52,17 @@
     int   uv_stride;
     /*    int   uvinternal_width; */
 
-    unsigned char *y_buffer;
-    unsigned char *u_buffer;
-    unsigned char *v_buffer;
+    int   alpha_width;
+    int   alpha_height;
+    int   alpha_stride;
 
-    unsigned char *buffer_alloc;
+    uint8_t *y_buffer;
+    uint8_t *u_buffer;
+    uint8_t *v_buffer;
+    uint8_t *alpha_buffer;
+
+    uint8_t *buffer_alloc;
+    int buffer_alloc_sz;
     int border;
     int frame_size;
     YUV_TYPE clrtype;
@@ -62,12 +71,22 @@
     int flags;
   } YV12_BUFFER_CONFIG;
 
-  int vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, int border);
+  int vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                                  int width, int height, int border);
+  int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                                    int width, int height, int border);
   int vp8_yv12_de_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf);
 
+  int vp9_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                             int width, int height, int ss_x, int ss_y,
+                             int border);
+  int vp9_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
+                               int width, int height, int ss_x, int ss_y,
+                               int border);
+  int vp9_free_frame_buffer(YV12_BUFFER_CONFIG *ybf);
+
 #ifdef __cplusplus
 }
 #endif
 
-
-#endif /*YV12_CONFIG_H*/
+#endif  // YV12_CONFIG_H
diff --git a/libvpx/vpxdec.c b/libvpx/vpxdec.c
index 9b728bf..513d7bd 100644
--- a/libvpx/vpxdec.c
+++ b/libvpx/vpxdec.c
@@ -12,6 +12,7 @@
 /* This is a simple program that reads ivf files and decodes them
  * using the new interface. Decoded frames are output as YV12 raw.
  */
+#include <assert.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdarg.h>
@@ -22,7 +23,7 @@
 #include "vpx_config.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx_ports/vpx_timer.h"
-#if CONFIG_VP8_DECODER
+#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
 #include "vpx/vp8dx.h"
 #endif
 #if CONFIG_MD5
@@ -30,6 +31,7 @@
 #endif
 #include "tools_common.h"
 #include "nestegg/include/nestegg/nestegg.h"
+#include "third_party/libyuv/include/libyuv/scale.h"
 
 #if CONFIG_OS_SUPPORT
 #if defined(_MSC_VER)
@@ -49,175 +51,175 @@
 static const char *exec_name;
 
 #define VP8_FOURCC (0x00385056)
-static const struct
-{
-    char const *name;
-    vpx_codec_iface_t *iface;
-    unsigned int             fourcc;
-    unsigned int             fourcc_mask;
-} ifaces[] =
-{
+#define VP9_FOURCC (0x00395056)
+static const struct {
+  char const *name;
+  const vpx_codec_iface_t *(*iface)(void);
+  unsigned int             fourcc;
+  unsigned int             fourcc_mask;
+} ifaces[] = {
 #if CONFIG_VP8_DECODER
-    {"vp8",  &vpx_codec_vp8_dx_algo,   VP8_FOURCC, 0x00FFFFFF},
+  {"vp8",  vpx_codec_vp8_dx,   VP8_FOURCC, 0x00FFFFFF},
+#endif
+#if CONFIG_VP9_DECODER
+  {"vp9",  vpx_codec_vp9_dx,   VP9_FOURCC, 0x00FFFFFF},
 #endif
 };
 
 #include "args.h"
+static const arg_def_t looparg = ARG_DEF(NULL, "loops", 1,
+                                          "Number of times to decode the file");
 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
-                                  "Codec to use");
+                                          "Codec to use");
 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
-                                  "Output raw YV12 frames");
+                                          "Output raw YV12 frames");
 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
-                                  "Output raw I420 frames");
+                                          "Output raw I420 frames");
 static const arg_def_t flipuvarg = ARG_DEF(NULL, "flipuv", 0,
-                                   "Flip the chroma planes in the output");
+                                           "Flip the chroma planes in the output");
 static const arg_def_t noblitarg = ARG_DEF(NULL, "noblit", 0,
-                                   "Don't process the decoded frames");
+                                           "Don't process the decoded frames");
 static const arg_def_t progressarg = ARG_DEF(NULL, "progress", 0,
-                                     "Show progress after each frame decodes");
+                                             "Show progress after each frame decodes");
 static const arg_def_t limitarg = ARG_DEF(NULL, "limit", 1,
-                                  "Stop decoding after n frames");
+                                          "Stop decoding after n frames");
+static const arg_def_t skiparg = ARG_DEF(NULL, "skip", 1,
+                                         "Skip the first n input frames");
 static const arg_def_t postprocarg = ARG_DEF(NULL, "postproc", 0,
-                                     "Postprocess decoded frames");
+                                             "Postprocess decoded frames");
 static const arg_def_t summaryarg = ARG_DEF(NULL, "summary", 0,
-                                    "Show timing summary");
+                                            "Show timing summary");
 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
-                                    "Output file name pattern (see below)");
+                                            "Output file name pattern (see below)");
 static const arg_def_t threadsarg = ARG_DEF("t", "threads", 1,
-                                    "Max threads to use");
+                                            "Max threads to use");
 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
-                                  "Show version string");
+                                            "Show version string");
 static const arg_def_t error_concealment = ARG_DEF(NULL, "error-concealment", 0,
-                                       "Enable decoder error-concealment");
+                                                   "Enable decoder error-concealment");
+static const arg_def_t scalearg = ARG_DEF("S", "scale", 0,
+                                            "Scale output frames uniformly");
 
 
 #if CONFIG_MD5
 static const arg_def_t md5arg = ARG_DEF(NULL, "md5", 0,
                                         "Compute the MD5 sum of the decoded frame");
 #endif
-static const arg_def_t *all_args[] =
-{
-    &codecarg, &use_yv12, &use_i420, &flipuvarg, &noblitarg,
-    &progressarg, &limitarg, &postprocarg, &summaryarg, &outputfile,
-    &threadsarg, &verbosearg,
+static const arg_def_t *all_args[] = {
+  &codecarg, &use_yv12, &use_i420, &flipuvarg, &noblitarg,
+  &progressarg, &limitarg, &skiparg, &postprocarg, &summaryarg, &outputfile,
+  &threadsarg, &verbosearg, &scalearg,
 #if CONFIG_MD5
-    &md5arg,
+  &md5arg,
 #endif
-    &error_concealment,
-    NULL
+  &error_concealment,
+  NULL
 };
 
 #if CONFIG_VP8_DECODER
 static const arg_def_t addnoise_level = ARG_DEF(NULL, "noise-level", 1,
-                                        "Enable VP8 postproc add noise");
+                                                "Enable VP8 postproc add noise");
 static const arg_def_t deblock = ARG_DEF(NULL, "deblock", 0,
-                                 "Enable VP8 deblocking");
+                                         "Enable VP8 deblocking");
 static const arg_def_t demacroblock_level = ARG_DEF(NULL, "demacroblock-level", 1,
-        "Enable VP8 demacroblocking, w/ level");
+                                                    "Enable VP8 demacroblocking, w/ level");
 static const arg_def_t pp_debug_info = ARG_DEF(NULL, "pp-debug-info", 1,
-                                       "Enable VP8 visible debug info");
+                                               "Enable VP8 visible debug info");
 static const arg_def_t pp_disp_ref_frame = ARG_DEF(NULL, "pp-dbg-ref-frame", 1,
-                                       "Display only selected reference frame per macro block");
+                                                   "Display only selected reference frame per macro block");
 static const arg_def_t pp_disp_mb_modes = ARG_DEF(NULL, "pp-dbg-mb-modes", 1,
-                                       "Display only selected macro block modes");
+                                                  "Display only selected macro block modes");
 static const arg_def_t pp_disp_b_modes = ARG_DEF(NULL, "pp-dbg-b-modes", 1,
-                                       "Display only selected block modes");
+                                                 "Display only selected block modes");
 static const arg_def_t pp_disp_mvs = ARG_DEF(NULL, "pp-dbg-mvs", 1,
-                                       "Draw only selected motion vectors");
+                                             "Draw only selected motion vectors");
 static const arg_def_t mfqe = ARG_DEF(NULL, "mfqe", 0,
-                                       "Enable multiframe quality enhancement");
+                                      "Enable multiframe quality enhancement");
 
-static const arg_def_t *vp8_pp_args[] =
-{
-    &addnoise_level, &deblock, &demacroblock_level, &pp_debug_info,
-    &pp_disp_ref_frame, &pp_disp_mb_modes, &pp_disp_b_modes, &pp_disp_mvs, &mfqe,
-    NULL
+static const arg_def_t *vp8_pp_args[] = {
+  &addnoise_level, &deblock, &demacroblock_level, &pp_debug_info,
+  &pp_disp_ref_frame, &pp_disp_mb_modes, &pp_disp_b_modes, &pp_disp_mvs, &mfqe,
+  NULL
 };
 #endif
 
-static void usage_exit()
-{
-    int i;
+static void usage_exit() {
+  int i;
 
-    fprintf(stderr, "Usage: %s <options> filename\n\n"
-            "Options:\n", exec_name);
-    arg_show_usage(stderr, all_args);
+  fprintf(stderr, "Usage: %s <options> filename\n\n"
+          "Options:\n", exec_name);
+  arg_show_usage(stderr, all_args);
 #if CONFIG_VP8_DECODER
-    fprintf(stderr, "\nVP8 Postprocessing Options:\n");
-    arg_show_usage(stderr, vp8_pp_args);
+  fprintf(stderr, "\nVP8 Postprocessing Options:\n");
+  arg_show_usage(stderr, vp8_pp_args);
 #endif
-    fprintf(stderr,
-            "\nOutput File Patterns:\n\n"
-            "  The -o argument specifies the name of the file(s) to "
-            "write to. If the\n  argument does not include any escape "
-            "characters, the output will be\n  written to a single file. "
-            "Otherwise, the filename will be calculated by\n  expanding "
-            "the following escape characters:\n");
-    fprintf(stderr,
-            "\n\t%%w   - Frame width"
-            "\n\t%%h   - Frame height"
-            "\n\t%%<n> - Frame number, zero padded to <n> places (1..9)"
-            "\n\n  Pattern arguments are only supported in conjunction "
-            "with the --yv12 and\n  --i420 options. If the -o option is "
-            "not specified, the output will be\n  directed to stdout.\n"
-            );
-    fprintf(stderr, "\nIncluded decoders:\n\n");
+  fprintf(stderr,
+          "\nOutput File Patterns:\n\n"
+          "  The -o argument specifies the name of the file(s) to "
+          "write to. If the\n  argument does not include any escape "
+          "characters, the output will be\n  written to a single file. "
+          "Otherwise, the filename will be calculated by\n  expanding "
+          "the following escape characters:\n");
+  fprintf(stderr,
+          "\n\t%%w   - Frame width"
+          "\n\t%%h   - Frame height"
+          "\n\t%%<n> - Frame number, zero padded to <n> places (1..9)"
+          "\n\n  Pattern arguments are only supported in conjunction "
+          "with the --yv12 and\n  --i420 options. If the -o option is "
+          "not specified, the output will be\n  directed to stdout.\n"
+         );
+  fprintf(stderr, "\nIncluded decoders:\n\n");
 
-    for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
-        fprintf(stderr, "    %-6s - %s\n",
-                ifaces[i].name,
-                vpx_codec_iface_name(ifaces[i].iface));
+  for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
+    fprintf(stderr, "    %-6s - %s\n",
+            ifaces[i].name,
+            vpx_codec_iface_name(ifaces[i].iface()));
 
-    exit(EXIT_FAILURE);
+  exit(EXIT_FAILURE);
 }
 
-void die(const char *fmt, ...)
-{
-    va_list ap;
-    va_start(ap, fmt);
-    vfprintf(stderr, fmt, ap);
-    fprintf(stderr, "\n");
-    usage_exit();
+void die(const char *fmt, ...) {
+  va_list ap;
+  va_start(ap, fmt);
+  vfprintf(stderr, fmt, ap);
+  fprintf(stderr, "\n");
+  usage_exit();
 }
 
-static unsigned int mem_get_le16(const void *vmem)
-{
-    unsigned int  val;
-    const unsigned char *mem = (const unsigned char *)vmem;
+static unsigned int mem_get_le16(const void *vmem) {
+  unsigned int  val;
+  const unsigned char *mem = (const unsigned char *)vmem;
 
-    val = mem[1] << 8;
-    val |= mem[0];
-    return val;
+  val = mem[1] << 8;
+  val |= mem[0];
+  return val;
 }
 
-static unsigned int mem_get_le32(const void *vmem)
-{
-    unsigned int  val;
-    const unsigned char *mem = (const unsigned char *)vmem;
+static unsigned int mem_get_le32(const void *vmem) {
+  unsigned int  val;
+  const unsigned char *mem = (const unsigned char *)vmem;
 
-    val = mem[3] << 24;
-    val |= mem[2] << 16;
-    val |= mem[1] << 8;
-    val |= mem[0];
-    return val;
+  val = mem[3] << 24;
+  val |= mem[2] << 16;
+  val |= mem[1] << 8;
+  val |= mem[0];
+  return val;
 }
 
-enum file_kind
-{
-    RAW_FILE,
-    IVF_FILE,
-    WEBM_FILE
+enum file_kind {
+  RAW_FILE,
+  IVF_FILE,
+  WEBM_FILE
 };
 
-struct input_ctx
-{
-    enum file_kind  kind;
-    FILE           *infile;
-    nestegg        *nestegg_ctx;
-    nestegg_packet *pkt;
-    unsigned int    chunk;
-    unsigned int    chunks;
-    unsigned int    video_track;
+struct input_ctx {
+  enum file_kind  kind;
+  FILE           *infile;
+  nestegg        *nestegg_ctx;
+  nestegg_packet *pkt;
+  unsigned int    chunk;
+  unsigned int    chunks;
+  unsigned int    video_track;
 };
 
 #define IVF_FRAME_HDR_SZ (sizeof(uint32_t) + sizeof(uint64_t))
@@ -225,163 +227,136 @@
 static int read_frame(struct input_ctx      *input,
                       uint8_t               **buf,
                       size_t                *buf_sz,
-                      size_t                *buf_alloc_sz)
-{
-    char            raw_hdr[IVF_FRAME_HDR_SZ];
-    size_t          new_buf_sz;
-    FILE           *infile = input->infile;
-    enum file_kind  kind = input->kind;
-    if(kind == WEBM_FILE)
-    {
-        if(input->chunk >= input->chunks)
-        {
-            unsigned int track;
+                      size_t                *buf_alloc_sz) {
+  char            raw_hdr[IVF_FRAME_HDR_SZ];
+  size_t          new_buf_sz;
+  FILE           *infile = input->infile;
+  enum file_kind  kind = input->kind;
+  if (kind == WEBM_FILE) {
+    if (input->chunk >= input->chunks) {
+      unsigned int track;
 
-            do
-            {
-                /* End of this packet, get another. */
-                if(input->pkt)
-                    nestegg_free_packet(input->pkt);
+      do {
+        /* End of this packet, get another. */
+        if (input->pkt)
+          nestegg_free_packet(input->pkt);
 
-                if(nestegg_read_packet(input->nestegg_ctx, &input->pkt) <= 0
-                   || nestegg_packet_track(input->pkt, &track))
-                    return 1;
+        if (nestegg_read_packet(input->nestegg_ctx, &input->pkt) <= 0
+            || nestegg_packet_track(input->pkt, &track))
+          return 1;
 
-            } while(track != input->video_track);
+      } while (track != input->video_track);
 
-            if(nestegg_packet_count(input->pkt, &input->chunks))
-                return 1;
-            input->chunk = 0;
-        }
-
-        if(nestegg_packet_data(input->pkt, input->chunk, buf, buf_sz))
-            return 1;
-        input->chunk++;
-
-        return 0;
-    }
-    /* For both the raw and ivf formats, the frame size is the first 4 bytes
-     * of the frame header. We just need to special case on the header
-     * size.
-     */
-    else if (fread(raw_hdr, kind==IVF_FILE
-                   ? IVF_FRAME_HDR_SZ : RAW_FRAME_HDR_SZ, 1, infile) != 1)
-    {
-        if (!feof(infile))
-            fprintf(stderr, "Failed to read frame size\n");
-
-        new_buf_sz = 0;
-    }
-    else
-    {
-        new_buf_sz = mem_get_le32(raw_hdr);
-
-        if (new_buf_sz > 256 * 1024 * 1024)
-        {
-            fprintf(stderr, "Error: Read invalid frame size (%u)\n",
-                    (unsigned int)new_buf_sz);
-            new_buf_sz = 0;
-        }
-
-        if (kind == RAW_FILE && new_buf_sz > 256 * 1024)
-            fprintf(stderr, "Warning: Read invalid frame size (%u)"
-                    " - not a raw file?\n", (unsigned int)new_buf_sz);
-
-        if (new_buf_sz > *buf_alloc_sz)
-        {
-            uint8_t *new_buf = realloc(*buf, 2 * new_buf_sz);
-
-            if (new_buf)
-            {
-                *buf = new_buf;
-                *buf_alloc_sz = 2 * new_buf_sz;
-            }
-            else
-            {
-                fprintf(stderr, "Failed to allocate compressed data buffer\n");
-                new_buf_sz = 0;
-            }
-        }
+      if (nestegg_packet_count(input->pkt, &input->chunks))
+        return 1;
+      input->chunk = 0;
     }
 
-    *buf_sz = new_buf_sz;
+    if (nestegg_packet_data(input->pkt, input->chunk, buf, buf_sz))
+      return 1;
+    input->chunk++;
 
+    return 0;
+  }
+  /* For both the raw and ivf formats, the frame size is the first 4 bytes
+   * of the frame header. We just need to special case on the header
+   * size.
+   */
+  else if (fread(raw_hdr, kind == IVF_FILE
+                 ? IVF_FRAME_HDR_SZ : RAW_FRAME_HDR_SZ, 1, infile) != 1) {
     if (!feof(infile))
-    {
-        if (fread(*buf, 1, *buf_sz, infile) != *buf_sz)
-        {
-            fprintf(stderr, "Failed to read full frame\n");
-            return 1;
-        }
+      fprintf(stderr, "Failed to read frame size\n");
 
-        return 0;
+    new_buf_sz = 0;
+  } else {
+    new_buf_sz = mem_get_le32(raw_hdr);
+
+    if (new_buf_sz > 256 * 1024 * 1024) {
+      fprintf(stderr, "Error: Read invalid frame size (%u)\n",
+              (unsigned int)new_buf_sz);
+      new_buf_sz = 0;
     }
 
-    return 1;
+    if (kind == RAW_FILE && new_buf_sz > 256 * 1024)
+      fprintf(stderr, "Warning: Read invalid frame size (%u)"
+              " - not a raw file?\n", (unsigned int)new_buf_sz);
+
+    if (new_buf_sz > *buf_alloc_sz) {
+      uint8_t *new_buf = realloc(*buf, 2 * new_buf_sz);
+
+      if (new_buf) {
+        *buf = new_buf;
+        *buf_alloc_sz = 2 * new_buf_sz;
+      } else {
+        fprintf(stderr, "Failed to allocate compressed data buffer\n");
+        new_buf_sz = 0;
+      }
+    }
+  }
+
+  *buf_sz = new_buf_sz;
+
+  if (!feof(infile)) {
+    if (fread(*buf, 1, *buf_sz, infile) != *buf_sz) {
+      fprintf(stderr, "Failed to read full frame\n");
+      return 1;
+    }
+
+    return 0;
+  }
+
+  return 1;
 }
 
-void *out_open(const char *out_fn, int do_md5)
-{
-    void *out = NULL;
+void *out_open(const char *out_fn, int do_md5) {
+  void *out = NULL;
 
-    if (do_md5)
-    {
+  if (do_md5) {
 #if CONFIG_MD5
-        MD5Context *md5_ctx = out = malloc(sizeof(MD5Context));
-        (void)out_fn;
-        MD5Init(md5_ctx);
+    MD5Context *md5_ctx = out = malloc(sizeof(MD5Context));
+    (void)out_fn;
+    MD5Init(md5_ctx);
 #endif
-    }
-    else
-    {
-        FILE *outfile = out = strcmp("-", out_fn) ? fopen(out_fn, "wb")
-                                                  : set_binary_mode(stdout);
+  } else {
+    FILE *outfile = out = strcmp("-", out_fn) ? fopen(out_fn, "wb")
+                          : set_binary_mode(stdout);
 
-        if (!outfile)
-        {
-            fprintf(stderr, "Failed to output file");
-            exit(EXIT_FAILURE);
-        }
+    if (!outfile) {
+      fprintf(stderr, "Failed to output file");
+      exit(EXIT_FAILURE);
     }
+  }
 
-    return out;
+  return out;
 }
 
-void out_put(void *out, const uint8_t *buf, unsigned int len, int do_md5)
-{
-    if (do_md5)
-    {
+void out_put(void *out, const uint8_t *buf, unsigned int len, int do_md5) {
+  if (do_md5) {
 #if CONFIG_MD5
-        MD5Update(out, buf, len);
+    MD5Update(out, buf, len);
 #endif
-    }
-    else
-    {
-        (void) fwrite(buf, 1, len, out);
-    }
+  } else {
+    (void) fwrite(buf, 1, len, out);
+  }
 }
 
-void out_close(void *out, const char *out_fn, int do_md5)
-{
-    if (do_md5)
-    {
+void out_close(void *out, const char *out_fn, int do_md5) {
+  if (do_md5) {
 #if CONFIG_MD5
-        uint8_t md5[16];
-        int i;
+    uint8_t md5[16];
+    int i;
 
-        MD5Final(md5, out);
-        free(out);
+    MD5Final(md5, out);
+    free(out);
 
-        for (i = 0; i < 16; i++)
-            printf("%02x", md5[i]);
+    for (i = 0; i < 16; i++)
+      printf("%02x", md5[i]);
 
-        printf("  %s\n", out_fn);
+    printf("  %s\n", out_fn);
 #endif
-    }
-    else
-    {
-        fclose(out);
-    }
+  } else {
+    fclose(out);
+  }
 }
 
 unsigned int file_is_ivf(FILE *infile,
@@ -389,56 +364,50 @@
                          unsigned int *width,
                          unsigned int *height,
                          unsigned int *fps_den,
-                         unsigned int *fps_num)
-{
-    char raw_hdr[32];
-    int is_ivf = 0;
+                         unsigned int *fps_num) {
+  char raw_hdr[32];
+  int is_ivf = 0;
 
-    if (fread(raw_hdr, 1, 32, infile) == 32)
-    {
-        if (raw_hdr[0] == 'D' && raw_hdr[1] == 'K'
-            && raw_hdr[2] == 'I' && raw_hdr[3] == 'F')
-        {
-            is_ivf = 1;
+  if (fread(raw_hdr, 1, 32, infile) == 32) {
+    if (raw_hdr[0] == 'D' && raw_hdr[1] == 'K'
+        && raw_hdr[2] == 'I' && raw_hdr[3] == 'F') {
+      is_ivf = 1;
 
-            if (mem_get_le16(raw_hdr + 4) != 0)
-                fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
-                        " decode properly.");
+      if (mem_get_le16(raw_hdr + 4) != 0)
+        fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
+                " decode properly.");
 
-            *fourcc = mem_get_le32(raw_hdr + 8);
-            *width = mem_get_le16(raw_hdr + 12);
-            *height = mem_get_le16(raw_hdr + 14);
-            *fps_num = mem_get_le32(raw_hdr + 16);
-            *fps_den = mem_get_le32(raw_hdr + 20);
+      *fourcc = mem_get_le32(raw_hdr + 8);
+      *width = mem_get_le16(raw_hdr + 12);
+      *height = mem_get_le16(raw_hdr + 14);
+      *fps_num = mem_get_le32(raw_hdr + 16);
+      *fps_den = mem_get_le32(raw_hdr + 20);
 
-            /* Some versions of vpxenc used 1/(2*fps) for the timebase, so
-             * we can guess the framerate using only the timebase in this
-             * case. Other files would require reading ahead to guess the
-             * timebase, like we do for webm.
-             */
-            if(*fps_num < 1000)
-            {
-                /* Correct for the factor of 2 applied to the timebase in the
-                 * encoder.
-                 */
-                if(*fps_num&1)*fps_den<<=1;
-                else *fps_num>>=1;
-            }
-            else
-            {
-                /* Don't know FPS for sure, and don't have readahead code
-                 * (yet?), so just default to 30fps.
-                 */
-                *fps_num = 30;
-                *fps_den = 1;
-            }
-        }
+      /* Some versions of vpxenc used 1/(2*fps) for the timebase, so
+       * we can guess the framerate using only the timebase in this
+       * case. Other files would require reading ahead to guess the
+       * timebase, like we do for webm.
+       */
+      if (*fps_num < 1000) {
+        /* Correct for the factor of 2 applied to the timebase in the
+         * encoder.
+         */
+        if (*fps_num & 1)*fps_den <<= 1;
+        else *fps_num >>= 1;
+      } else {
+        /* Don't know FPS for sure, and don't have readahead code
+         * (yet?), so just default to 30fps.
+         */
+        *fps_num = 30;
+        *fps_den = 1;
+      }
     }
+  }
 
-    if (!is_ivf)
-        rewind(infile);
+  if (!is_ivf)
+    rewind(infile);
 
-    return is_ivf;
+  return is_ivf;
 }
 
 
@@ -447,126 +416,121 @@
                          unsigned int *width,
                          unsigned int *height,
                          unsigned int *fps_den,
-                         unsigned int *fps_num)
-{
-    unsigned char buf[32];
-    int is_raw = 0;
-    vpx_codec_stream_info_t si;
+                         unsigned int *fps_num) {
+  unsigned char buf[32];
+  int is_raw = 0;
+  vpx_codec_stream_info_t si;
 
-    si.sz = sizeof(si);
+  si.sz = sizeof(si);
 
-    if (fread(buf, 1, 32, infile) == 32)
-    {
-        int i;
+  if (fread(buf, 1, 32, infile) == 32) {
+    int i;
 
-        if(mem_get_le32(buf) < 256 * 1024 * 1024)
-            for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
-                if(!vpx_codec_peek_stream_info(ifaces[i].iface,
-                                               buf + 4, 32 - 4, &si))
-                {
-                    is_raw = 1;
-                    *fourcc = ifaces[i].fourcc;
-                    *width = si.w;
-                    *height = si.h;
-                    *fps_num = 30;
-                    *fps_den = 1;
-                    break;
-                }
-    }
+    if (mem_get_le32(buf) < 256 * 1024 * 1024)
+      for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
+        if (!vpx_codec_peek_stream_info(ifaces[i].iface(),
+                                        buf + 4, 32 - 4, &si)) {
+          is_raw = 1;
+          *fourcc = ifaces[i].fourcc;
+          *width = si.w;
+          *height = si.h;
+          *fps_num = 30;
+          *fps_den = 1;
+          break;
+        }
+  }
 
-    rewind(infile);
-    return is_raw;
+  rewind(infile);
+  return is_raw;
 }
 
 
 static int
-nestegg_read_cb(void *buffer, size_t length, void *userdata)
-{
-    FILE *f = userdata;
+nestegg_read_cb(void *buffer, size_t length, void *userdata) {
+  FILE *f = userdata;
 
-    if(fread(buffer, 1, length, f) < length)
-    {
-        if (ferror(f))
-            return -1;
-        if (feof(f))
-            return 0;
-    }
-    return 1;
+  if (fread(buffer, 1, length, f) < length) {
+    if (ferror(f))
+      return -1;
+    if (feof(f))
+      return 0;
+  }
+  return 1;
 }
 
 
 static int
-nestegg_seek_cb(int64_t offset, int whence, void * userdata)
-{
-    switch(whence) {
-        case NESTEGG_SEEK_SET: whence = SEEK_SET; break;
-        case NESTEGG_SEEK_CUR: whence = SEEK_CUR; break;
-        case NESTEGG_SEEK_END: whence = SEEK_END; break;
-    };
-    return fseek(userdata, (long)offset, whence)? -1 : 0;
+nestegg_seek_cb(int64_t offset, int whence, void *userdata) {
+  switch (whence) {
+    case NESTEGG_SEEK_SET:
+      whence = SEEK_SET;
+      break;
+    case NESTEGG_SEEK_CUR:
+      whence = SEEK_CUR;
+      break;
+    case NESTEGG_SEEK_END:
+      whence = SEEK_END;
+      break;
+  };
+  return fseek(userdata, (long)offset, whence) ? -1 : 0;
 }
 
 
 static int64_t
-nestegg_tell_cb(void * userdata)
-{
-    return ftell(userdata);
+nestegg_tell_cb(void *userdata) {
+  return ftell(userdata);
 }
 
 
 static void
-nestegg_log_cb(nestegg * context, unsigned int severity, char const * format,
-               ...)
-{
-    va_list ap;
+nestegg_log_cb(nestegg *context, unsigned int severity, char const *format,
+               ...) {
+  va_list ap;
 
-    va_start(ap, format);
-    vfprintf(stderr, format, ap);
-    fprintf(stderr, "\n");
-    va_end(ap);
+  va_start(ap, format);
+  vfprintf(stderr, format, ap);
+  fprintf(stderr, "\n");
+  va_end(ap);
 }
 
 
 static int
 webm_guess_framerate(struct input_ctx *input,
                      unsigned int     *fps_den,
-                     unsigned int     *fps_num)
-{
-    unsigned int i;
-    uint64_t     tstamp=0;
+                     unsigned int     *fps_num) {
+  unsigned int i;
+  uint64_t     tstamp = 0;
 
-    /* Guess the framerate. Read up to 1 second, or 50 video packets,
-     * whichever comes first.
-     */
-    for(i=0; tstamp < 1000000000 && i < 50;)
-    {
-        nestegg_packet * pkt;
-        unsigned int track;
+  /* Guess the framerate. Read up to 1 second, or 50 video packets,
+   * whichever comes first.
+   */
+  for (i = 0; tstamp < 1000000000 && i < 50;) {
+    nestegg_packet *pkt;
+    unsigned int track;
 
-        if(nestegg_read_packet(input->nestegg_ctx, &pkt) <= 0)
-            break;
+    if (nestegg_read_packet(input->nestegg_ctx, &pkt) <= 0)
+      break;
 
-        nestegg_packet_track(pkt, &track);
-        if(track == input->video_track)
-        {
-            nestegg_packet_tstamp(pkt, &tstamp);
-            i++;
-        }
-
-        nestegg_free_packet(pkt);
+    nestegg_packet_track(pkt, &track);
+    if (track == input->video_track) {
+      nestegg_packet_tstamp(pkt, &tstamp);
+      i++;
     }
 
-    if(nestegg_track_seek(input->nestegg_ctx, input->video_track, 0))
-        goto fail;
+    nestegg_free_packet(pkt);
+  }
 
-    *fps_num = (i - 1) * 1000000;
-    *fps_den = (unsigned int)(tstamp / 1000);
-    return 0;
+  if (nestegg_track_seek(input->nestegg_ctx, input->video_track, 0))
+    goto fail;
+
+  *fps_num = (i - 1) * 1000000;
+  *fps_den = (unsigned int)(tstamp / 1000);
+  return 0;
 fail:
-    nestegg_destroy(input->nestegg_ctx);
-    input->nestegg_ctx = NULL;
-    rewind(input->infile);
-    return 1;
+  nestegg_destroy(input->nestegg_ctx);
+  input->nestegg_ctx = NULL;
+  rewind(input->infile);
+  return 1;
 }
 
 
@@ -576,586 +540,641 @@
              unsigned int     *width,
              unsigned int     *height,
              unsigned int     *fps_den,
-             unsigned int     *fps_num)
-{
-    unsigned int i, n;
-    int          track_type = -1;
+             unsigned int     *fps_num) {
+  unsigned int i, n;
+  int          track_type = -1;
+  int          codec_id;
 
-    nestegg_io io = {nestegg_read_cb, nestegg_seek_cb, nestegg_tell_cb, 0};
-    nestegg_video_params params;
+  nestegg_io io = {nestegg_read_cb, nestegg_seek_cb, nestegg_tell_cb, 0};
+  nestegg_video_params params;
 
-    io.userdata = input->infile;
-    if(nestegg_init(&input->nestegg_ctx, io, NULL))
-        goto fail;
+  io.userdata = input->infile;
+  if (nestegg_init(&input->nestegg_ctx, io, NULL))
+    goto fail;
 
-    if(nestegg_track_count(input->nestegg_ctx, &n))
-        goto fail;
+  if (nestegg_track_count(input->nestegg_ctx, &n))
+    goto fail;
 
-    for(i=0; i<n; i++)
-    {
-        track_type = nestegg_track_type(input->nestegg_ctx, i);
+  for (i = 0; i < n; i++) {
+    track_type = nestegg_track_type(input->nestegg_ctx, i);
 
-        if(track_type == NESTEGG_TRACK_VIDEO)
-            break;
-        else if(track_type < 0)
-            goto fail;
-    }
+    if (track_type == NESTEGG_TRACK_VIDEO)
+      break;
+    else if (track_type < 0)
+      goto fail;
+  }
 
-    if(nestegg_track_codec_id(input->nestegg_ctx, i) != NESTEGG_CODEC_VP8)
-    {
-        fprintf(stderr, "Not VP8 video, quitting.\n");
-        exit(1);
-    }
-
-    input->video_track = i;
-
-    if(nestegg_track_video_params(input->nestegg_ctx, i, &params))
-        goto fail;
-
-    *fps_den = 0;
-    *fps_num = 0;
+  codec_id = nestegg_track_codec_id(input->nestegg_ctx, i);
+  if (codec_id == NESTEGG_CODEC_VP8) {
     *fourcc = VP8_FOURCC;
-    *width = params.width;
-    *height = params.height;
-    return 1;
+  } else if (codec_id == NESTEGG_CODEC_VP9) {
+    *fourcc = VP9_FOURCC;
+  } else {
+    fprintf(stderr, "Not VPx video, quitting.\n");
+    exit(1);
+  }
+
+  input->video_track = i;
+
+  if (nestegg_track_video_params(input->nestegg_ctx, i, &params))
+    goto fail;
+
+  *fps_den = 0;
+  *fps_num = 0;
+  *width = params.width;
+  *height = params.height;
+  return 1;
 fail:
-    input->nestegg_ctx = NULL;
-    rewind(input->infile);
-    return 0;
+  input->nestegg_ctx = NULL;
+  rewind(input->infile);
+  return 0;
 }
 
 
-void show_progress(int frame_in, int frame_out, unsigned long dx_time)
-{
-    fprintf(stderr, "%d decoded frames/%d showed frames in %lu us (%.2f fps)\r",
-            frame_in, frame_out, dx_time,
-            (float)frame_out * 1000000.0 / (float)dx_time);
+void show_progress(int frame_in, int frame_out, unsigned long dx_time) {
+  fprintf(stderr, "%d decoded frames/%d showed frames in %lu us (%.2f fps)\r",
+          frame_in, frame_out, dx_time,
+          (float)frame_out * 1000000.0 / (float)dx_time);
 }
 
 
 void generate_filename(const char *pattern, char *out, size_t q_len,
                        unsigned int d_w, unsigned int d_h,
-                       unsigned int frame_in)
-{
-    const char *p = pattern;
-    char *q = out;
+                       unsigned int frame_in) {
+  const char *p = pattern;
+  char *q = out;
 
-    do
-    {
-        char *next_pat = strchr(p, '%');
+  do {
+    char *next_pat = strchr(p, '%');
 
-        if(p == next_pat)
-        {
-            size_t pat_len;
+    if (p == next_pat) {
+      size_t pat_len;
 
-            /* parse the pattern */
-            q[q_len - 1] = '\0';
-            switch(p[1])
-            {
-            case 'w': snprintf(q, q_len - 1, "%d", d_w); break;
-            case 'h': snprintf(q, q_len - 1, "%d", d_h); break;
-            case '1': snprintf(q, q_len - 1, "%d", frame_in); break;
-            case '2': snprintf(q, q_len - 1, "%02d", frame_in); break;
-            case '3': snprintf(q, q_len - 1, "%03d", frame_in); break;
-            case '4': snprintf(q, q_len - 1, "%04d", frame_in); break;
-            case '5': snprintf(q, q_len - 1, "%05d", frame_in); break;
-            case '6': snprintf(q, q_len - 1, "%06d", frame_in); break;
-            case '7': snprintf(q, q_len - 1, "%07d", frame_in); break;
-            case '8': snprintf(q, q_len - 1, "%08d", frame_in); break;
-            case '9': snprintf(q, q_len - 1, "%09d", frame_in); break;
-            default:
-                die("Unrecognized pattern %%%c\n", p[1]);
-            }
+      /* parse the pattern */
+      q[q_len - 1] = '\0';
+      switch (p[1]) {
+        case 'w':
+          snprintf(q, q_len - 1, "%d", d_w);
+          break;
+        case 'h':
+          snprintf(q, q_len - 1, "%d", d_h);
+          break;
+        case '1':
+          snprintf(q, q_len - 1, "%d", frame_in);
+          break;
+        case '2':
+          snprintf(q, q_len - 1, "%02d", frame_in);
+          break;
+        case '3':
+          snprintf(q, q_len - 1, "%03d", frame_in);
+          break;
+        case '4':
+          snprintf(q, q_len - 1, "%04d", frame_in);
+          break;
+        case '5':
+          snprintf(q, q_len - 1, "%05d", frame_in);
+          break;
+        case '6':
+          snprintf(q, q_len - 1, "%06d", frame_in);
+          break;
+        case '7':
+          snprintf(q, q_len - 1, "%07d", frame_in);
+          break;
+        case '8':
+          snprintf(q, q_len - 1, "%08d", frame_in);
+          break;
+        case '9':
+          snprintf(q, q_len - 1, "%09d", frame_in);
+          break;
+        default:
+          die("Unrecognized pattern %%%c\n", p[1]);
+      }
 
-            pat_len = strlen(q);
-            if(pat_len >= q_len - 1)
-                die("Output filename too long.\n");
-            q += pat_len;
-            p += 2;
-            q_len -= pat_len;
-        }
-        else
-        {
-            size_t copy_len;
+      pat_len = strlen(q);
+      if (pat_len >= q_len - 1)
+        die("Output filename too long.\n");
+      q += pat_len;
+      p += 2;
+      q_len -= pat_len;
+    } else {
+      size_t copy_len;
 
-            /* copy the next segment */
-            if(!next_pat)
-                copy_len = strlen(p);
-            else
-                copy_len = next_pat - p;
+      /* copy the next segment */
+      if (!next_pat)
+        copy_len = strlen(p);
+      else
+        copy_len = next_pat - p;
 
-            if(copy_len >= q_len - 1)
-                die("Output filename too long.\n");
+      if (copy_len >= q_len - 1)
+        die("Output filename too long.\n");
 
-            memcpy(q, p, copy_len);
-            q[copy_len] = '\0';
-            q += copy_len;
-            p += copy_len;
-            q_len -= copy_len;
-        }
-    } while(*p);
+      memcpy(q, p, copy_len);
+      q[copy_len] = '\0';
+      q += copy_len;
+      p += copy_len;
+      q_len -= copy_len;
+    }
+  } while (*p);
 }
 
 
-int main(int argc, const char **argv_)
-{
-    vpx_codec_ctx_t          decoder;
-    char                  *fn = NULL;
-    int                    i;
-    uint8_t               *buf = NULL;
-    size_t                 buf_sz = 0, buf_alloc_sz = 0;
-    FILE                  *infile;
-    int                    frame_in = 0, frame_out = 0, flipuv = 0, noblit = 0, do_md5 = 0, progress = 0;
-    int                    stop_after = 0, postproc = 0, summary = 0, quiet = 1;
-    int                    ec_enabled = 0;
-    vpx_codec_iface_t       *iface = NULL;
-    unsigned int           fourcc;
-    unsigned long          dx_time = 0;
-    struct arg               arg;
-    char                   **argv, **argi, **argj;
-    const char             *outfile_pattern = 0;
-    char                    outfile[PATH_MAX];
-    int                     single_file;
-    int                     use_y4m = 1;
-    unsigned int            width;
-    unsigned int            height;
-    unsigned int            fps_den;
-    unsigned int            fps_num;
-    void                   *out = NULL;
-    vpx_codec_dec_cfg_t     cfg = {0};
+int main_loop(int argc, const char **argv_) {
+  vpx_codec_ctx_t          decoder;
+  char                  *fn = NULL;
+  int                    i;
+  uint8_t               *buf = NULL;
+  size_t                 buf_sz = 0, buf_alloc_sz = 0;
+  FILE                  *infile;
+  int                    frame_in = 0, frame_out = 0, flipuv = 0, noblit = 0, do_md5 = 0, progress = 0;
+  int                    stop_after = 0, postproc = 0, summary = 0, quiet = 1;
+  int                    arg_skip = 0;
+  int                    ec_enabled = 0;
+  vpx_codec_iface_t       *iface = NULL;
+  unsigned int           fourcc;
+  unsigned long          dx_time = 0;
+  struct arg               arg;
+  char                   **argv, **argi, **argj;
+  const char             *outfile_pattern = 0;
+  char                    outfile[PATH_MAX];
+  int                     single_file;
+  int                     use_y4m = 1;
+  unsigned int            width;
+  unsigned int            height;
+  unsigned int            fps_den;
+  unsigned int            fps_num;
+  void                   *out = NULL;
+  vpx_codec_dec_cfg_t     cfg = {0};
 #if CONFIG_VP8_DECODER
-    vp8_postproc_cfg_t      vp8_pp_cfg = {0};
-    int                     vp8_dbg_color_ref_frame = 0;
-    int                     vp8_dbg_color_mb_modes = 0;
-    int                     vp8_dbg_color_b_modes = 0;
-    int                     vp8_dbg_display_mv = 0;
+  vp8_postproc_cfg_t      vp8_pp_cfg = {0};
+  int                     vp8_dbg_color_ref_frame = 0;
+  int                     vp8_dbg_color_mb_modes = 0;
+  int                     vp8_dbg_color_b_modes = 0;
+  int                     vp8_dbg_display_mv = 0;
 #endif
-    struct input_ctx        input = {0};
-    int                     frames_corrupted = 0;
-    int                     dec_flags = 0;
+  struct input_ctx        input = {0};
+  int                     frames_corrupted = 0;
+  int                     dec_flags = 0;
+  int                     do_scale = 0;
+  int                     stream_w = 0, stream_h = 0;
+  vpx_image_t             *scaled_img = NULL;
+  int                     frame_avail, got_data;
 
-    /* Parse command line */
-    exec_name = argv_[0];
-    argv = argv_dup(argc - 1, argv_ + 1);
+  /* Parse command line */
+  exec_name = argv_[0];
+  argv = argv_dup(argc - 1, argv_ + 1);
 
-    for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
-    {
-        memset(&arg, 0, sizeof(arg));
-        arg.argv_step = 1;
+  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
+    memset(&arg, 0, sizeof(arg));
+    arg.argv_step = 1;
 
-        if (arg_match(&arg, &codecarg, argi))
-        {
-            int j, k = -1;
+    if (arg_match(&arg, &codecarg, argi)) {
+      int j, k = -1;
 
-            for (j = 0; j < sizeof(ifaces) / sizeof(ifaces[0]); j++)
-                if (!strcmp(ifaces[j].name, arg.val))
-                    k = j;
+      for (j = 0; j < sizeof(ifaces) / sizeof(ifaces[0]); j++)
+        if (!strcmp(ifaces[j].name, arg.val))
+          k = j;
 
-            if (k >= 0)
-                iface = ifaces[k].iface;
-            else
-                die("Error: Unrecognized argument (%s) to --codec\n",
-                    arg.val);
-        }
-        else if (arg_match(&arg, &outputfile, argi))
-            outfile_pattern = arg.val;
-        else if (arg_match(&arg, &use_yv12, argi))
-        {
-            use_y4m = 0;
-            flipuv = 1;
-        }
-        else if (arg_match(&arg, &use_i420, argi))
-        {
-            use_y4m = 0;
-            flipuv = 0;
-        }
-        else if (arg_match(&arg, &flipuvarg, argi))
-            flipuv = 1;
-        else if (arg_match(&arg, &noblitarg, argi))
-            noblit = 1;
-        else if (arg_match(&arg, &progressarg, argi))
-            progress = 1;
-        else if (arg_match(&arg, &limitarg, argi))
-            stop_after = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &postprocarg, argi))
-            postproc = 1;
-        else if (arg_match(&arg, &md5arg, argi))
-            do_md5 = 1;
-        else if (arg_match(&arg, &summaryarg, argi))
-            summary = 1;
-        else if (arg_match(&arg, &threadsarg, argi))
-            cfg.threads = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &verbosearg, argi))
-            quiet = 0;
+      if (k >= 0)
+        iface = ifaces[k].iface();
+      else
+        die("Error: Unrecognized argument (%s) to --codec\n",
+            arg.val);
+    } else if (arg_match(&arg, &looparg, argi)) {
+      // no-op
+    } else if (arg_match(&arg, &outputfile, argi))
+      outfile_pattern = arg.val;
+    else if (arg_match(&arg, &use_yv12, argi)) {
+      use_y4m = 0;
+      flipuv = 1;
+    } else if (arg_match(&arg, &use_i420, argi)) {
+      use_y4m = 0;
+      flipuv = 0;
+    } else if (arg_match(&arg, &flipuvarg, argi))
+      flipuv = 1;
+    else if (arg_match(&arg, &noblitarg, argi))
+      noblit = 1;
+    else if (arg_match(&arg, &progressarg, argi))
+      progress = 1;
+    else if (arg_match(&arg, &limitarg, argi))
+      stop_after = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &skiparg, argi))
+      arg_skip = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &postprocarg, argi))
+      postproc = 1;
+    else if (arg_match(&arg, &md5arg, argi))
+      do_md5 = 1;
+    else if (arg_match(&arg, &summaryarg, argi))
+      summary = 1;
+    else if (arg_match(&arg, &threadsarg, argi))
+      cfg.threads = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &verbosearg, argi))
+      quiet = 0;
+    else if (arg_match(&arg, &scalearg, argi))
+      do_scale = 1;
 
 #if CONFIG_VP8_DECODER
-        else if (arg_match(&arg, &addnoise_level, argi))
-        {
-            postproc = 1;
-            vp8_pp_cfg.post_proc_flag |= VP8_ADDNOISE;
-            vp8_pp_cfg.noise_level = arg_parse_uint(&arg);
-        }
-        else if (arg_match(&arg, &demacroblock_level, argi))
-        {
-            postproc = 1;
-            vp8_pp_cfg.post_proc_flag |= VP8_DEMACROBLOCK;
-            vp8_pp_cfg.deblocking_level = arg_parse_uint(&arg);
-        }
-        else if (arg_match(&arg, &deblock, argi))
-        {
-            postproc = 1;
-            vp8_pp_cfg.post_proc_flag |= VP8_DEBLOCK;
-        }
-        else if (arg_match(&arg, &mfqe, argi))
-        {
-            postproc = 1;
-            vp8_pp_cfg.post_proc_flag |= VP8_MFQE;
-        }
-        else if (arg_match(&arg, &pp_debug_info, argi))
-        {
-            unsigned int level = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &addnoise_level, argi)) {
+      postproc = 1;
+      vp8_pp_cfg.post_proc_flag |= VP8_ADDNOISE;
+      vp8_pp_cfg.noise_level = arg_parse_uint(&arg);
+    } else if (arg_match(&arg, &demacroblock_level, argi)) {
+      postproc = 1;
+      vp8_pp_cfg.post_proc_flag |= VP8_DEMACROBLOCK;
+      vp8_pp_cfg.deblocking_level = arg_parse_uint(&arg);
+    } else if (arg_match(&arg, &deblock, argi)) {
+      postproc = 1;
+      vp8_pp_cfg.post_proc_flag |= VP8_DEBLOCK;
+    } else if (arg_match(&arg, &mfqe, argi)) {
+      postproc = 1;
+      vp8_pp_cfg.post_proc_flag |= VP8_MFQE;
+    } else if (arg_match(&arg, &pp_debug_info, argi)) {
+      unsigned int level = arg_parse_uint(&arg);
 
-            postproc = 1;
-            vp8_pp_cfg.post_proc_flag &= ~0x7;
+      postproc = 1;
+      vp8_pp_cfg.post_proc_flag &= ~0x7;
 
-            if (level)
-                vp8_pp_cfg.post_proc_flag |= level;
-        }
-        else if (arg_match(&arg, &pp_disp_ref_frame, argi))
-        {
-            unsigned int flags = arg_parse_int(&arg);
-            if (flags)
-            {
-                postproc = 1;
-                vp8_dbg_color_ref_frame = flags;
-            }
-        }
-        else if (arg_match(&arg, &pp_disp_mb_modes, argi))
-        {
-            unsigned int flags = arg_parse_int(&arg);
-            if (flags)
-            {
-                postproc = 1;
-                vp8_dbg_color_mb_modes = flags;
-            }
-        }
-        else if (arg_match(&arg, &pp_disp_b_modes, argi))
-        {
-            unsigned int flags = arg_parse_int(&arg);
-            if (flags)
-            {
-                postproc = 1;
-                vp8_dbg_color_b_modes = flags;
-            }
-        }
-        else if (arg_match(&arg, &pp_disp_mvs, argi))
-        {
-            unsigned int flags = arg_parse_int(&arg);
-            if (flags)
-            {
-                postproc = 1;
-                vp8_dbg_display_mv = flags;
-            }
-        }
-        else if (arg_match(&arg, &error_concealment, argi))
-        {
-            ec_enabled = 1;
-        }
+      if (level)
+        vp8_pp_cfg.post_proc_flag |= level;
+    } else if (arg_match(&arg, &pp_disp_ref_frame, argi)) {
+      unsigned int flags = arg_parse_int(&arg);
+      if (flags) {
+        postproc = 1;
+        vp8_dbg_color_ref_frame = flags;
+      }
+    } else if (arg_match(&arg, &pp_disp_mb_modes, argi)) {
+      unsigned int flags = arg_parse_int(&arg);
+      if (flags) {
+        postproc = 1;
+        vp8_dbg_color_mb_modes = flags;
+      }
+    } else if (arg_match(&arg, &pp_disp_b_modes, argi)) {
+      unsigned int flags = arg_parse_int(&arg);
+      if (flags) {
+        postproc = 1;
+        vp8_dbg_color_b_modes = flags;
+      }
+    } else if (arg_match(&arg, &pp_disp_mvs, argi)) {
+      unsigned int flags = arg_parse_int(&arg);
+      if (flags) {
+        postproc = 1;
+        vp8_dbg_display_mv = flags;
+      }
+    } else if (arg_match(&arg, &error_concealment, argi)) {
+      ec_enabled = 1;
+    }
 
 #endif
-        else
-            argj++;
-    }
-
-    /* Check for unrecognized options */
-    for (argi = argv; *argi; argi++)
-        if (argi[0][0] == '-' && strlen(argi[0]) > 1)
-            die("Error: Unrecognized option %s\n", *argi);
-
-    /* Handle non-option arguments */
-    fn = argv[0];
-
-    if (!fn)
-        usage_exit();
-
-    /* Open file */
-    infile = strcmp(fn, "-") ? fopen(fn, "rb") : set_binary_mode(stdin);
-
-    if (!infile)
-    {
-        fprintf(stderr, "Failed to open file '%s'",
-                strcmp(fn, "-") ? fn : "stdin");
-        return EXIT_FAILURE;
-    }
-#if CONFIG_OS_SUPPORT
-    /* Make sure we don't dump to the terminal, unless forced to with -o - */
-    if(!outfile_pattern && isatty(fileno(stdout)) && !do_md5 && !noblit)
-    {
-        fprintf(stderr,
-                "Not dumping raw video to your terminal. Use '-o -' to "
-                "override.\n");
-        return EXIT_FAILURE;
-    }
-#endif
-    input.infile = infile;
-    if(file_is_ivf(infile, &fourcc, &width, &height, &fps_den,
-                   &fps_num))
-        input.kind = IVF_FILE;
-    else if(file_is_webm(&input, &fourcc, &width, &height, &fps_den, &fps_num))
-        input.kind = WEBM_FILE;
-    else if(file_is_raw(infile, &fourcc, &width, &height, &fps_den, &fps_num))
-        input.kind = RAW_FILE;
     else
-    {
-        fprintf(stderr, "Unrecognized input file type.\n");
+      argj++;
+  }
+
+  /* Check for unrecognized options */
+  for (argi = argv; *argi; argi++)
+    if (argi[0][0] == '-' && strlen(argi[0]) > 1)
+      die("Error: Unrecognized option %s\n", *argi);
+
+  /* Handle non-option arguments */
+  fn = argv[0];
+
+  if (!fn)
+    usage_exit();
+
+  /* Open file */
+  infile = strcmp(fn, "-") ? fopen(fn, "rb") : set_binary_mode(stdin);
+
+  if (!infile) {
+    fprintf(stderr, "Failed to open file '%s'",
+            strcmp(fn, "-") ? fn : "stdin");
+    return EXIT_FAILURE;
+  }
+#if CONFIG_OS_SUPPORT
+  /* Make sure we don't dump to the terminal, unless forced to with -o - */
+  if (!outfile_pattern && isatty(fileno(stdout)) && !do_md5 && !noblit) {
+    fprintf(stderr,
+            "Not dumping raw video to your terminal. Use '-o -' to "
+            "override.\n");
+    return EXIT_FAILURE;
+  }
+#endif
+  input.infile = infile;
+  if (file_is_ivf(infile, &fourcc, &width, &height, &fps_den,
+                  &fps_num))
+    input.kind = IVF_FILE;
+  else if (file_is_webm(&input, &fourcc, &width, &height, &fps_den, &fps_num))
+    input.kind = WEBM_FILE;
+  else if (file_is_raw(infile, &fourcc, &width, &height, &fps_den, &fps_num))
+    input.kind = RAW_FILE;
+  else {
+    fprintf(stderr, "Unrecognized input file type.\n");
+    return EXIT_FAILURE;
+  }
+
+  /* If the output file is not set or doesn't have a sequence number in
+   * it, then we only open it once.
+   */
+  outfile_pattern = outfile_pattern ? outfile_pattern : "-";
+  single_file = 1;
+  {
+    const char *p = outfile_pattern;
+    do {
+      p = strchr(p, '%');
+      if (p && p[1] >= '1' && p[1] <= '9') {
+        /* pattern contains sequence number, so it's not unique. */
+        single_file = 0;
+        break;
+      }
+      if (p)
+        p++;
+    } while (p);
+  }
+
+  if (single_file && !noblit) {
+    generate_filename(outfile_pattern, outfile, sizeof(outfile) - 1,
+                      width, height, 0);
+    out = out_open(outfile, do_md5);
+  }
+
+  if (use_y4m && !noblit) {
+    char buffer[128];
+
+    if (!single_file) {
+      fprintf(stderr, "YUV4MPEG2 not supported with output patterns,"
+              " try --i420 or --yv12.\n");
+      return EXIT_FAILURE;
+    }
+
+    if (input.kind == WEBM_FILE)
+      if (webm_guess_framerate(&input, &fps_den, &fps_num)) {
+        fprintf(stderr, "Failed to guess framerate -- error parsing "
+                "webm file?\n");
         return EXIT_FAILURE;
+      }
+
+
+    /*Note: We can't output an aspect ratio here because IVF doesn't
+       store one, and neither does VP8.
+      That will have to wait until these tools support WebM natively.*/
+    snprintf(buffer, sizeof(buffer), "YUV4MPEG2 W%u H%u F%u:%u I%c ",
+             width, height, fps_num, fps_den, 'p');
+    out_put(out, (unsigned char *)buffer,
+            (unsigned int)strlen(buffer), do_md5);
+  }
+
+  /* Try to determine the codec from the fourcc. */
+  for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
+    if ((fourcc & ifaces[i].fourcc_mask) == ifaces[i].fourcc) {
+      vpx_codec_iface_t  *ivf_iface = ifaces[i].iface();
+
+      if (iface && iface != ivf_iface)
+        fprintf(stderr, "Notice -- IVF header indicates codec: %s\n",
+                ifaces[i].name);
+      else
+        iface = ivf_iface;
+
+      break;
     }
 
-    /* If the output file is not set or doesn't have a sequence number in
-     * it, then we only open it once.
-     */
-    outfile_pattern = outfile_pattern ? outfile_pattern : "-";
-    single_file = 1;
-    {
-        const char *p = outfile_pattern;
-        do
-        {
-            p = strchr(p, '%');
-            if(p && p[1] >= '1' && p[1] <= '9')
-            {
-                /* pattern contains sequence number, so it's not unique. */
-                single_file = 0;
-                break;
-            }
-            if(p)
-                p++;
-        } while(p);
-    }
+  dec_flags = (postproc ? VPX_CODEC_USE_POSTPROC : 0) |
+              (ec_enabled ? VPX_CODEC_USE_ERROR_CONCEALMENT : 0);
+  if (vpx_codec_dec_init(&decoder, iface ? iface :  ifaces[0].iface(), &cfg,
+                         dec_flags)) {
+    fprintf(stderr, "Failed to initialize decoder: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 
-    if(single_file && !noblit)
-    {
-        generate_filename(outfile_pattern, outfile, sizeof(outfile)-1,
-                          width, height, 0);
-        out = out_open(outfile, do_md5);
-    }
-
-    if (use_y4m && !noblit)
-    {
-        char buffer[128];
-        if (!single_file)
-        {
-            fprintf(stderr, "YUV4MPEG2 not supported with output patterns,"
-                            " try --i420 or --yv12.\n");
-            return EXIT_FAILURE;
-        }
-
-        if(input.kind == WEBM_FILE)
-            if(webm_guess_framerate(&input, &fps_den, &fps_num))
-            {
-                fprintf(stderr, "Failed to guess framerate -- error parsing "
-                                "webm file?\n");
-                return EXIT_FAILURE;
-            }
-
-
-        /*Note: We can't output an aspect ratio here because IVF doesn't
-           store one, and neither does VP8.
-          That will have to wait until these tools support WebM natively.*/
-        sprintf(buffer, "YUV4MPEG2 C%s W%u H%u F%u:%u I%c\n",
-                "420jpeg", width, height, fps_num, fps_den, 'p');
-        out_put(out, (unsigned char *)buffer,
-                (unsigned int)strlen(buffer), do_md5);
-    }
-
-    /* Try to determine the codec from the fourcc. */
-    for (i = 0; i < sizeof(ifaces) / sizeof(ifaces[0]); i++)
-        if ((fourcc & ifaces[i].fourcc_mask) == ifaces[i].fourcc)
-        {
-            vpx_codec_iface_t  *ivf_iface = ifaces[i].iface;
-
-            if (iface && iface != ivf_iface)
-                fprintf(stderr, "Notice -- IVF header indicates codec: %s\n",
-                        ifaces[i].name);
-            else
-                iface = ivf_iface;
-
-            break;
-        }
-
-    dec_flags = (postproc ? VPX_CODEC_USE_POSTPROC : 0) |
-                (ec_enabled ? VPX_CODEC_USE_ERROR_CONCEALMENT : 0);
-    if (vpx_codec_dec_init(&decoder, iface ? iface :  ifaces[0].iface, &cfg,
-                           dec_flags))
-    {
-        fprintf(stderr, "Failed to initialize decoder: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
-
-    if (!quiet)
-        fprintf(stderr, "%s\n", decoder.name);
+  if (!quiet)
+    fprintf(stderr, "%s\n", decoder.name);
 
 #if CONFIG_VP8_DECODER
 
-    if (vp8_pp_cfg.post_proc_flag
-        && vpx_codec_control(&decoder, VP8_SET_POSTPROC, &vp8_pp_cfg))
-    {
-        fprintf(stderr, "Failed to configure postproc: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
+  if (vp8_pp_cfg.post_proc_flag
+      && vpx_codec_control(&decoder, VP8_SET_POSTPROC, &vp8_pp_cfg)) {
+    fprintf(stderr, "Failed to configure postproc: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 
-    if (vp8_dbg_color_ref_frame
-        && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_REF_FRAME, vp8_dbg_color_ref_frame))
-    {
-        fprintf(stderr, "Failed to configure reference block visualizer: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
+  if (vp8_dbg_color_ref_frame
+      && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_REF_FRAME, vp8_dbg_color_ref_frame)) {
+    fprintf(stderr, "Failed to configure reference block visualizer: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 
-    if (vp8_dbg_color_mb_modes
-        && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_MB_MODES, vp8_dbg_color_mb_modes))
-    {
-        fprintf(stderr, "Failed to configure macro block visualizer: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
+  if (vp8_dbg_color_mb_modes
+      && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_MB_MODES, vp8_dbg_color_mb_modes)) {
+    fprintf(stderr, "Failed to configure macro block visualizer: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 
-    if (vp8_dbg_color_b_modes
-        && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_B_MODES, vp8_dbg_color_b_modes))
-    {
-        fprintf(stderr, "Failed to configure block visualizer: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
+  if (vp8_dbg_color_b_modes
+      && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_B_MODES, vp8_dbg_color_b_modes)) {
+    fprintf(stderr, "Failed to configure block visualizer: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 
-    if (vp8_dbg_display_mv
-        && vpx_codec_control(&decoder, VP8_SET_DBG_DISPLAY_MV, vp8_dbg_display_mv))
-    {
-        fprintf(stderr, "Failed to configure motion vector visualizer: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
-    }
+  if (vp8_dbg_display_mv
+      && vpx_codec_control(&decoder, VP8_SET_DBG_DISPLAY_MV, vp8_dbg_display_mv)) {
+    fprintf(stderr, "Failed to configure motion vector visualizer: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
 #endif
 
-    /* Decode file */
-    while (!read_frame(&input, &buf, &buf_sz, &buf_alloc_sz))
-    {
-        vpx_codec_iter_t  iter = NULL;
-        vpx_image_t    *img;
-        struct vpx_usec_timer timer;
-        int                   corrupted;
+
+  if(arg_skip)
+    fprintf(stderr, "Skiping first %d frames.\n", arg_skip);
+  while (arg_skip) {
+    if (read_frame(&input, &buf, &buf_sz, &buf_alloc_sz))
+      break;
+    arg_skip--;
+  }
+
+  frame_avail = 1;
+  got_data = 0;
+
+  /* Decode file */
+  while (frame_avail || got_data) {
+    vpx_codec_iter_t  iter = NULL;
+    vpx_image_t    *img;
+    struct vpx_usec_timer timer;
+    int                   corrupted;
+
+    frame_avail = 0;
+    if (!stop_after || frame_in < stop_after) {
+      if(!read_frame(&input, &buf, &buf_sz, &buf_alloc_sz)) {
+        frame_avail = 1;
+        frame_in++;
 
         vpx_usec_timer_start(&timer);
 
-        if (vpx_codec_decode(&decoder, buf, (unsigned int)buf_sz, NULL, 0))
-        {
-            const char *detail = vpx_codec_error_detail(&decoder);
-            fprintf(stderr, "Failed to decode frame: %s\n", vpx_codec_error(&decoder));
+        if (vpx_codec_decode(&decoder, buf, (unsigned int)buf_sz, NULL, 0)) {
+          const char *detail = vpx_codec_error_detail(&decoder);
+          fprintf(stderr, "Failed to decode frame: %s\n",
+                  vpx_codec_error(&decoder));
 
-            if (detail)
-                fprintf(stderr, "  Additional information: %s\n", detail);
-
-            goto fail;
+          if (detail)
+            fprintf(stderr, "  Additional information: %s\n", detail);
+          goto fail;
         }
 
         vpx_usec_timer_mark(&timer);
         dx_time += (unsigned int)vpx_usec_timer_elapsed(&timer);
-
-        ++frame_in;
-
-        if (vpx_codec_control(&decoder, VP8D_GET_FRAME_CORRUPTED, &corrupted))
-        {
-            fprintf(stderr, "Failed VP8_GET_FRAME_CORRUPTED: %s\n",
-                    vpx_codec_error(&decoder));
-            goto fail;
-        }
-        frames_corrupted += corrupted;
-
-        vpx_usec_timer_start(&timer);
-
-        if ((img = vpx_codec_get_frame(&decoder, &iter)))
-            ++frame_out;
-
-        vpx_usec_timer_mark(&timer);
-        dx_time += (unsigned int)vpx_usec_timer_elapsed(&timer);
-
-        if (progress)
-            show_progress(frame_in, frame_out, dx_time);
-
-        if (!noblit)
-        {
-            if (img)
-            {
-                unsigned int y;
-                char out_fn[PATH_MAX];
-                uint8_t *buf;
-
-                if (!single_file)
-                {
-                    size_t len = sizeof(out_fn)-1;
-
-                    out_fn[len] = '\0';
-                    generate_filename(outfile_pattern, out_fn, len-1,
-                                      img->d_w, img->d_h, frame_in);
-                    out = out_open(out_fn, do_md5);
-                }
-                else if(use_y4m)
-                    out_put(out, (unsigned char *)"FRAME\n", 6, do_md5);
-
-                buf = img->planes[VPX_PLANE_Y];
-
-                for (y = 0; y < img->d_h; y++)
-                {
-                    out_put(out, buf, img->d_w, do_md5);
-                    buf += img->stride[VPX_PLANE_Y];
-                }
-
-                buf = img->planes[flipuv?VPX_PLANE_V:VPX_PLANE_U];
-
-                for (y = 0; y < (1 + img->d_h) / 2; y++)
-                {
-                    out_put(out, buf, (1 + img->d_w) / 2, do_md5);
-                    buf += img->stride[VPX_PLANE_U];
-                }
-
-                buf = img->planes[flipuv?VPX_PLANE_U:VPX_PLANE_V];
-
-                for (y = 0; y < (1 + img->d_h) / 2; y++)
-                {
-                    out_put(out, buf, (1 + img->d_w) / 2, do_md5);
-                    buf += img->stride[VPX_PLANE_V];
-                }
-
-                if (!single_file)
-                    out_close(out, out_fn, do_md5);
-            }
-        }
-
-        if (stop_after && frame_in >= stop_after)
-            break;
+      }
     }
 
-    if (summary || progress)
-    {
-        show_progress(frame_in, frame_out, dx_time);
-        fprintf(stderr, "\n");
+    vpx_usec_timer_start(&timer);
+
+    got_data = 0;
+    if ((img = vpx_codec_get_frame(&decoder, &iter))) {
+      ++frame_out;
+      got_data = 1;
     }
 
-    if (frames_corrupted)
-        fprintf(stderr, "WARNING: %d frames corrupted.\n",frames_corrupted);
+    vpx_usec_timer_mark(&timer);
+    dx_time += (unsigned int)vpx_usec_timer_elapsed(&timer);
+
+    if (vpx_codec_control(&decoder, VP8D_GET_FRAME_CORRUPTED, &corrupted)) {
+      fprintf(stderr, "Failed VP8_GET_FRAME_CORRUPTED: %s\n",
+              vpx_codec_error(&decoder));
+      goto fail;
+    }
+    frames_corrupted += corrupted;
+
+    if (progress)
+      show_progress(frame_in, frame_out, dx_time);
+
+    if (!noblit) {
+      if (frame_out == 1 && img && use_y4m) {
+        /* Write out the color format to terminate the header line */
+        const char *color =
+            img->fmt == VPX_IMG_FMT_444A ? "C444alpha\n" :
+            img->fmt == VPX_IMG_FMT_I444 ? "C444\n" :
+            img->fmt == VPX_IMG_FMT_I422 ? "C422\n" :
+            "C420jpeg\n";
+
+        out_put(out, (const unsigned char*)color, strlen(color), do_md5);
+      }
+
+      if (do_scale) {
+        if (img && frame_out == 1) {
+          stream_w = img->d_w;
+          stream_h = img->d_h;
+          scaled_img = vpx_img_alloc(NULL, VPX_IMG_FMT_I420,
+                                     stream_w, stream_h, 16);
+        }
+        if (img && (img->d_w != stream_w || img->d_h != stream_h)) {
+          assert(img->fmt == VPX_IMG_FMT_I420);
+          I420Scale(img->planes[VPX_PLANE_Y], img->stride[VPX_PLANE_Y],
+                    img->planes[VPX_PLANE_U], img->stride[VPX_PLANE_U],
+                    img->planes[VPX_PLANE_V], img->stride[VPX_PLANE_V],
+                    img->d_w, img->d_h,
+                    scaled_img->planes[VPX_PLANE_Y],
+                    scaled_img->stride[VPX_PLANE_Y],
+                    scaled_img->planes[VPX_PLANE_U],
+                    scaled_img->stride[VPX_PLANE_U],
+                    scaled_img->planes[VPX_PLANE_V],
+                    scaled_img->stride[VPX_PLANE_V],
+                    stream_w, stream_h,
+                    kFilterBox);
+          img = scaled_img;
+        }
+      }
+
+      if (img) {
+        unsigned int y;
+        char out_fn[PATH_MAX];
+        uint8_t *buf;
+        unsigned int c_w =
+            img->x_chroma_shift ? (1 + img->d_w) >> img->x_chroma_shift
+                                : img->d_w;
+        unsigned int c_h =
+            img->y_chroma_shift ? (1 + img->d_h) >> img->y_chroma_shift
+                                : img->d_h;
+
+        if (!single_file) {
+          size_t len = sizeof(out_fn) - 1;
+
+          out_fn[len] = '\0';
+          generate_filename(outfile_pattern, out_fn, len - 1,
+                            img->d_w, img->d_h, frame_in);
+          out = out_open(out_fn, do_md5);
+        } else if (use_y4m)
+          out_put(out, (unsigned char *)"FRAME\n", 6, do_md5);
+
+        buf = img->planes[VPX_PLANE_Y];
+
+        for (y = 0; y < img->d_h; y++) {
+          out_put(out, buf, img->d_w, do_md5);
+          buf += img->stride[VPX_PLANE_Y];
+        }
+
+        buf = img->planes[flipuv ? VPX_PLANE_V : VPX_PLANE_U];
+
+        for (y = 0; y < c_h; y++) {
+          out_put(out, buf, c_w, do_md5);
+          buf += img->stride[VPX_PLANE_U];
+        }
+
+        buf = img->planes[flipuv ? VPX_PLANE_U : VPX_PLANE_V];
+
+        for (y = 0; y < c_h; y++) {
+          out_put(out, buf, c_w, do_md5);
+          buf += img->stride[VPX_PLANE_V];
+        }
+
+        if (!single_file)
+          out_close(out, out_fn, do_md5);
+      }
+    }
+
+    if (stop_after && frame_in >= stop_after)
+      break;
+  }
+
+  if (summary || progress) {
+    show_progress(frame_in, frame_out, dx_time);
+    fprintf(stderr, "\n");
+  }
+
+  if (frames_corrupted)
+    fprintf(stderr, "WARNING: %d frames corrupted.\n", frames_corrupted);
 
 fail:
 
-    if (vpx_codec_destroy(&decoder))
-    {
-        fprintf(stderr, "Failed to destroy decoder: %s\n", vpx_codec_error(&decoder));
-        return EXIT_FAILURE;
+  if (vpx_codec_destroy(&decoder)) {
+    fprintf(stderr, "Failed to destroy decoder: %s\n", vpx_codec_error(&decoder));
+    return EXIT_FAILURE;
+  }
+
+  if (single_file && !noblit)
+    out_close(out, outfile, do_md5);
+
+  if (input.nestegg_ctx)
+    nestegg_destroy(input.nestegg_ctx);
+  if (input.kind != WEBM_FILE)
+    free(buf);
+  fclose(infile);
+  free(argv);
+
+  return frames_corrupted ? EXIT_FAILURE : EXIT_SUCCESS;
+}
+
+int main(int argc, const char **argv_) {
+  unsigned int loops = 1, i;
+  char **argv, **argi, **argj;
+  struct arg arg;
+  int error = 0;
+
+  argv = argv_dup(argc - 1, argv_ + 1);
+  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
+    memset(&arg, 0, sizeof(arg));
+    arg.argv_step = 1;
+
+    if (arg_match(&arg, &looparg, argi)) {
+      loops = arg_parse_uint(&arg);
+      break;
     }
-
-    if (single_file && !noblit)
-        out_close(out, outfile, do_md5);
-
-    if(input.nestegg_ctx)
-        nestegg_destroy(input.nestegg_ctx);
-    if(input.kind != WEBM_FILE)
-        free(buf);
-    fclose(infile);
-    free(argv);
-
-    return frames_corrupted ? EXIT_FAILURE : EXIT_SUCCESS;
+  }
+  free(argv);
+  for (i = 0; !error && i < loops; i++)
+    error = main_loop(argc, argv_);
+  return error;
 }
diff --git a/libvpx/vpxenc.c b/libvpx/vpxenc.c
index 7449e6c..a60b84d 100644
--- a/libvpx/vpxenc.c
+++ b/libvpx/vpxenc.c
@@ -8,11 +8,9 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include "vpx_config.h"
 
-/* This is a simple program that encodes YV12 files and generates ivf
- * files using the new interface.
- */
-#if defined(_WIN32) || !CONFIG_OS_SUPPORT
+#if defined(_WIN32) || defined(__OS2__) || !CONFIG_OS_SUPPORT
 #define USE_POSIX_MMAP 0
 #else
 #define USE_POSIX_MMAP 1
@@ -25,6 +23,9 @@
 #include <limits.h>
 #include <assert.h>
 #include "vpx/vpx_encoder.h"
+#if CONFIG_DECODERS
+#include "vpx/vpx_decoder.h"
+#endif
 #if USE_POSIX_MMAP
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -32,13 +33,21 @@
 #include <fcntl.h>
 #include <unistd.h>
 #endif
+
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
 #include "vpx/vp8cx.h"
+#endif
+#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
+#include "vpx/vp8dx.h"
+#endif
+
 #include "vpx_ports/mem_ops.h"
 #include "vpx_ports/vpx_timer.h"
 #include "tools_common.h"
 #include "y4minput.h"
 #include "libmkv/EbmlWriter.h"
 #include "libmkv/EbmlIDs.h"
+#include "third_party/libyuv/include/libyuv/scale.h"
 
 /* Need special handling of these functions on Windows */
 #if defined(_MSC_VER)
@@ -66,643 +75,610 @@
 
 /* Swallow warnings about unused results of fread/fwrite */
 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb,
-                         FILE *stream)
-{
-    return fread(ptr, size, nmemb, stream);
+                         FILE *stream) {
+  return fread(ptr, size, nmemb, stream);
 }
 #define fread wrap_fread
 
 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
-                          FILE *stream)
-{
-    return fwrite(ptr, size, nmemb, stream);
+                          FILE *stream) {
+  return fwrite(ptr, size, nmemb, stream);
 }
 #define fwrite wrap_fwrite
 
 
 static const char *exec_name;
 
-static const struct codec_item
-{
-    char const              *name;
-    vpx_codec_iface_t       *iface;
-    unsigned int             fourcc;
-} codecs[] =
-{
-#if CONFIG_VP8_ENCODER
-    {"vp8",  &vpx_codec_vp8_cx_algo, 0x30385056},
+#define VP8_FOURCC (0x30385056)
+#define VP9_FOURCC (0x30395056)
+static const struct codec_item {
+  char const              *name;
+  const vpx_codec_iface_t *(*iface)(void);
+  const vpx_codec_iface_t *(*dx_iface)(void);
+  unsigned int             fourcc;
+} codecs[] = {
+#if CONFIG_VP8_ENCODER && CONFIG_VP8_DECODER
+  {"vp8", &vpx_codec_vp8_cx, &vpx_codec_vp8_dx, VP8_FOURCC},
+#elif CONFIG_VP8_ENCODER && !CONFIG_VP8_DECODER
+  {"vp8", &vpx_codec_vp8_cx, NULL, VP8_FOURCC},
+#endif
+#if CONFIG_VP9_ENCODER && CONFIG_VP9_DECODER
+  {"vp9", &vpx_codec_vp9_cx, &vpx_codec_vp9_dx, VP9_FOURCC},
+#elif CONFIG_VP9_ENCODER && !CONFIG_VP9_DECODER
+  {"vp9", &vpx_codec_vp9_cx, NULL, VP9_FOURCC},
 #endif
 };
 
 static void usage_exit();
 
 #define LOG_ERROR(label) do \
-{\
+  {\
     const char *l=label;\
     va_list ap;\
     va_start(ap, fmt);\
     if(l)\
-        fprintf(stderr, "%s: ", l);\
+      fprintf(stderr, "%s: ", l);\
     vfprintf(stderr, fmt, ap);\
     fprintf(stderr, "\n");\
     va_end(ap);\
-} while(0)
+  } while(0)
 
-void die(const char *fmt, ...)
-{
-    LOG_ERROR(NULL);
-    usage_exit();
+void die(const char *fmt, ...) {
+  LOG_ERROR(NULL);
+  usage_exit();
 }
 
 
-void fatal(const char *fmt, ...)
-{
-    LOG_ERROR("Fatal");
-    exit(EXIT_FAILURE);
+void fatal(const char *fmt, ...) {
+  LOG_ERROR("Fatal");
+  exit(EXIT_FAILURE);
 }
 
 
-void warn(const char *fmt, ...)
-{
-    LOG_ERROR("Warning");
+void warn(const char *fmt, ...) {
+  LOG_ERROR("Warning");
 }
 
 
-static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...)
-{
-    va_list ap;
+static void warn_or_exit_on_errorv(vpx_codec_ctx_t *ctx, int fatal,
+                                   const char *s, va_list ap) {
+  if (ctx->err) {
+    const char *detail = vpx_codec_error_detail(ctx);
 
-    va_start(ap, s);
-    if (ctx->err)
-    {
-        const char *detail = vpx_codec_error_detail(ctx);
+    vfprintf(stderr, s, ap);
+    fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
 
-        vfprintf(stderr, s, ap);
-        fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
+    if (detail)
+      fprintf(stderr, "    %s\n", detail);
 
-        if (detail)
-            fprintf(stderr, "    %s\n", detail);
+    if (fatal)
+      exit(EXIT_FAILURE);
+  }
+}
 
-        exit(EXIT_FAILURE);
-    }
+static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...) {
+  va_list ap;
+
+  va_start(ap, s);
+  warn_or_exit_on_errorv(ctx, 1, s, ap);
+  va_end(ap);
+}
+
+static void warn_or_exit_on_error(vpx_codec_ctx_t *ctx, int fatal,
+                                  const char *s, ...) {
+  va_list ap;
+
+  va_start(ap, s);
+  warn_or_exit_on_errorv(ctx, fatal, s, ap);
+  va_end(ap);
 }
 
 /* This structure is used to abstract the different ways of handling
  * first pass statistics.
  */
-typedef struct
-{
-    vpx_fixed_buf_t buf;
-    int             pass;
-    FILE           *file;
-    char           *buf_ptr;
-    size_t          buf_alloc_sz;
+typedef struct {
+  vpx_fixed_buf_t buf;
+  int             pass;
+  FILE           *file;
+  char           *buf_ptr;
+  size_t          buf_alloc_sz;
 } stats_io_t;
 
-int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
-{
-    int res;
+int stats_open_file(stats_io_t *stats, const char *fpf, int pass) {
+  int res;
 
-    stats->pass = pass;
+  stats->pass = pass;
 
-    if (pass == 0)
-    {
-        stats->file = fopen(fpf, "wb");
-        stats->buf.sz = 0;
-        stats->buf.buf = NULL,
-                   res = (stats->file != NULL);
-    }
-    else
-    {
+  if (pass == 0) {
+    stats->file = fopen(fpf, "wb");
+    stats->buf.sz = 0;
+    stats->buf.buf = NULL,
+               res = (stats->file != NULL);
+  } else {
 #if 0
 #elif USE_POSIX_MMAP
-        struct stat stat_buf;
-        int fd;
+    struct stat stat_buf;
+    int fd;
 
-        fd = open(fpf, O_RDONLY);
-        stats->file = fdopen(fd, "rb");
-        fstat(fd, &stat_buf);
-        stats->buf.sz = stat_buf.st_size;
-        stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
-                              fd, 0);
-        res = (stats->buf.buf != NULL);
-#else
-        size_t nbytes;
-
-        stats->file = fopen(fpf, "rb");
-
-        if (fseek(stats->file, 0, SEEK_END))
-            fatal("First-pass stats file must be seekable!");
-
-        stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
-        rewind(stats->file);
-
-        stats->buf.buf = malloc(stats->buf_alloc_sz);
-
-        if (!stats->buf.buf)
-            fatal("Failed to allocate first-pass stats buffer (%lu bytes)",
-                  (unsigned long)stats->buf_alloc_sz);
-
-        nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
-        res = (nbytes == stats->buf.sz);
-#endif
-    }
-
-    return res;
-}
-
-int stats_open_mem(stats_io_t *stats, int pass)
-{
-    int res;
-    stats->pass = pass;
-
-    if (!pass)
-    {
-        stats->buf.sz = 0;
-        stats->buf_alloc_sz = 64 * 1024;
-        stats->buf.buf = malloc(stats->buf_alloc_sz);
-    }
-
-    stats->buf_ptr = stats->buf.buf;
+    fd = open(fpf, O_RDONLY);
+    stats->file = fdopen(fd, "rb");
+    fstat(fd, &stat_buf);
+    stats->buf.sz = stat_buf.st_size;
+    stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
+                          fd, 0);
     res = (stats->buf.buf != NULL);
-    return res;
+#else
+    size_t nbytes;
+
+    stats->file = fopen(fpf, "rb");
+
+    if (fseek(stats->file, 0, SEEK_END))
+      fatal("First-pass stats file must be seekable!");
+
+    stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
+    rewind(stats->file);
+
+    stats->buf.buf = malloc(stats->buf_alloc_sz);
+
+    if (!stats->buf.buf)
+      fatal("Failed to allocate first-pass stats buffer (%lu bytes)",
+            (unsigned long)stats->buf_alloc_sz);
+
+    nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
+    res = (nbytes == stats->buf.sz);
+#endif
+  }
+
+  return res;
+}
+
+int stats_open_mem(stats_io_t *stats, int pass) {
+  int res;
+  stats->pass = pass;
+
+  if (!pass) {
+    stats->buf.sz = 0;
+    stats->buf_alloc_sz = 64 * 1024;
+    stats->buf.buf = malloc(stats->buf_alloc_sz);
+  }
+
+  stats->buf_ptr = stats->buf.buf;
+  res = (stats->buf.buf != NULL);
+  return res;
 }
 
 
-void stats_close(stats_io_t *stats, int last_pass)
-{
-    if (stats->file)
-    {
-        if (stats->pass == last_pass)
-        {
+void stats_close(stats_io_t *stats, int last_pass) {
+  if (stats->file) {
+    if (stats->pass == last_pass) {
 #if 0
 #elif USE_POSIX_MMAP
-            munmap(stats->buf.buf, stats->buf.sz);
+      munmap(stats->buf.buf, stats->buf.sz);
 #else
-            free(stats->buf.buf);
+      free(stats->buf.buf);
 #endif
-        }
+    }
 
-        fclose(stats->file);
-        stats->file = NULL;
-    }
-    else
-    {
-        if (stats->pass == last_pass)
-            free(stats->buf.buf);
-    }
+    fclose(stats->file);
+    stats->file = NULL;
+  } else {
+    if (stats->pass == last_pass)
+      free(stats->buf.buf);
+  }
 }
 
-void stats_write(stats_io_t *stats, const void *pkt, size_t len)
-{
-    if (stats->file)
-    {
-        (void) fwrite(pkt, 1, len, stats->file);
-    }
-    else
-    {
-        if (stats->buf.sz + len > stats->buf_alloc_sz)
-        {
-            size_t  new_sz = stats->buf_alloc_sz + 64 * 1024;
-            char   *new_ptr = realloc(stats->buf.buf, new_sz);
+void stats_write(stats_io_t *stats, const void *pkt, size_t len) {
+  if (stats->file) {
+    (void) fwrite(pkt, 1, len, stats->file);
+  } else {
+    if (stats->buf.sz + len > stats->buf_alloc_sz) {
+      size_t  new_sz = stats->buf_alloc_sz + 64 * 1024;
+      char   *new_ptr = realloc(stats->buf.buf, new_sz);
 
-            if (new_ptr)
-            {
-                stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
-                stats->buf.buf = new_ptr;
-                stats->buf_alloc_sz = new_sz;
-            }
-            else
-                fatal("Failed to realloc firstpass stats buffer.");
-        }
-
-        memcpy(stats->buf_ptr, pkt, len);
-        stats->buf.sz += len;
-        stats->buf_ptr += len;
+      if (new_ptr) {
+        stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
+        stats->buf.buf = new_ptr;
+        stats->buf_alloc_sz = new_sz;
+      } else
+        fatal("Failed to realloc firstpass stats buffer.");
     }
+
+    memcpy(stats->buf_ptr, pkt, len);
+    stats->buf.sz += len;
+    stats->buf_ptr += len;
+  }
 }
 
-vpx_fixed_buf_t stats_get(stats_io_t *stats)
-{
-    return stats->buf;
+vpx_fixed_buf_t stats_get(stats_io_t *stats) {
+  return stats->buf;
 }
 
 /* Stereo 3D packed frame format */
-typedef enum stereo_format
-{
-    STEREO_FORMAT_MONO       = 0,
-    STEREO_FORMAT_LEFT_RIGHT = 1,
-    STEREO_FORMAT_BOTTOM_TOP = 2,
-    STEREO_FORMAT_TOP_BOTTOM = 3,
-    STEREO_FORMAT_RIGHT_LEFT = 11
+typedef enum stereo_format {
+  STEREO_FORMAT_MONO       = 0,
+  STEREO_FORMAT_LEFT_RIGHT = 1,
+  STEREO_FORMAT_BOTTOM_TOP = 2,
+  STEREO_FORMAT_TOP_BOTTOM = 3,
+  STEREO_FORMAT_RIGHT_LEFT = 11
 } stereo_format_t;
 
-enum video_file_type
-{
-    FILE_TYPE_RAW,
-    FILE_TYPE_IVF,
-    FILE_TYPE_Y4M
+enum video_file_type {
+  FILE_TYPE_RAW,
+  FILE_TYPE_IVF,
+  FILE_TYPE_Y4M
 };
 
 struct detect_buffer {
-    char buf[4];
-    size_t buf_read;
-    size_t position;
+  char buf[4];
+  size_t buf_read;
+  size_t position;
 };
 
 
-struct input_state
-{
-    char                 *fn;
-    FILE                 *file;
-    y4m_input             y4m;
-    struct detect_buffer  detect;
-    enum video_file_type  file_type;
-    unsigned int          w;
-    unsigned int          h;
-    struct vpx_rational   framerate;
-    int                   use_i420;
+struct input_state {
+  char                 *fn;
+  FILE                 *file;
+  off_t                 length;
+  y4m_input             y4m;
+  struct detect_buffer  detect;
+  enum video_file_type  file_type;
+  unsigned int          w;
+  unsigned int          h;
+  struct vpx_rational   framerate;
+  int                   use_i420;
+  int                   only_i420;
 };
 
 
 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
-static int read_frame(struct input_state *input, vpx_image_t *img)
-{
-    FILE *f = input->file;
-    enum video_file_type file_type = input->file_type;
-    y4m_input *y4m = &input->y4m;
-    struct detect_buffer *detect = &input->detect;
-    int plane = 0;
-    int shortread = 0;
+static int read_frame(struct input_state *input, vpx_image_t *img) {
+  FILE *f = input->file;
+  enum video_file_type file_type = input->file_type;
+  y4m_input *y4m = &input->y4m;
+  struct detect_buffer *detect = &input->detect;
+  int plane = 0;
+  int shortread = 0;
 
-    if (file_type == FILE_TYPE_Y4M)
-    {
-        if (y4m_input_fetch_frame(y4m, f, img) < 1)
-           return 0;
-    }
-    else
-    {
-        if (file_type == FILE_TYPE_IVF)
-        {
-            char junk[IVF_FRAME_HDR_SZ];
+  if (file_type == FILE_TYPE_Y4M) {
+    if (y4m_input_fetch_frame(y4m, f, img) < 1)
+      return 0;
+  } else {
+    if (file_type == FILE_TYPE_IVF) {
+      char junk[IVF_FRAME_HDR_SZ];
 
-            /* Skip the frame header. We know how big the frame should be. See
-             * write_ivf_frame_header() for documentation on the frame header
-             * layout.
-             */
-            (void) fread(junk, 1, IVF_FRAME_HDR_SZ, f);
-        }
-
-        for (plane = 0; plane < 3; plane++)
-        {
-            unsigned char *ptr;
-            int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
-            int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
-            int r;
-
-            /* Determine the correct plane based on the image format. The for-loop
-             * always counts in Y,U,V order, but this may not match the order of
-             * the data on disk.
-             */
-            switch (plane)
-            {
-            case 1:
-                ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
-                break;
-            case 2:
-                ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
-                break;
-            default:
-                ptr = img->planes[plane];
-            }
-
-            for (r = 0; r < h; r++)
-            {
-                size_t needed = w;
-                size_t buf_position = 0;
-                const size_t left = detect->buf_read - detect->position;
-                if (left > 0)
-                {
-                    const size_t more = (left < needed) ? left : needed;
-                    memcpy(ptr, detect->buf + detect->position, more);
-                    buf_position = more;
-                    needed -= more;
-                    detect->position += more;
-                }
-                if (needed > 0)
-                {
-                    shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
-                }
-
-                ptr += img->stride[plane];
-            }
-        }
+      /* Skip the frame header. We know how big the frame should be. See
+       * write_ivf_frame_header() for documentation on the frame header
+       * layout.
+       */
+      (void) fread(junk, 1, IVF_FRAME_HDR_SZ, f);
     }
 
-    return !shortread;
+    for (plane = 0; plane < 3; plane++) {
+      unsigned char *ptr;
+      int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
+      int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
+      int r;
+
+      /* Determine the correct plane based on the image format. The for-loop
+       * always counts in Y,U,V order, but this may not match the order of
+       * the data on disk.
+       */
+      switch (plane) {
+        case 1:
+          ptr = img->planes[img->fmt == VPX_IMG_FMT_YV12 ? VPX_PLANE_V : VPX_PLANE_U];
+          break;
+        case 2:
+          ptr = img->planes[img->fmt == VPX_IMG_FMT_YV12 ? VPX_PLANE_U : VPX_PLANE_V];
+          break;
+        default:
+          ptr = img->planes[plane];
+      }
+
+      for (r = 0; r < h; r++) {
+        size_t needed = w;
+        size_t buf_position = 0;
+        const size_t left = detect->buf_read - detect->position;
+        if (left > 0) {
+          const size_t more = (left < needed) ? left : needed;
+          memcpy(ptr, detect->buf + detect->position, more);
+          buf_position = more;
+          needed -= more;
+          detect->position += more;
+        }
+        if (needed > 0) {
+          shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
+        }
+
+        ptr += img->stride[plane];
+      }
+    }
+  }
+
+  return !shortread;
 }
 
 
 unsigned int file_is_y4m(FILE      *infile,
                          y4m_input *y4m,
-                         char       detect[4])
-{
-    if(memcmp(detect, "YUV4", 4) == 0)
-    {
-        return 1;
-    }
-    return 0;
+                         char       detect[4]) {
+  if (memcmp(detect, "YUV4", 4) == 0) {
+    return 1;
+  }
+  return 0;
 }
 
 #define IVF_FILE_HDR_SZ (32)
 unsigned int file_is_ivf(struct input_state *input,
-                         unsigned int *fourcc)
-{
-    char raw_hdr[IVF_FILE_HDR_SZ];
-    int is_ivf = 0;
-    FILE *infile = input->file;
-    unsigned int *width = &input->w;
-    unsigned int *height = &input->h;
-    struct detect_buffer *detect = &input->detect;
+                         unsigned int *fourcc) {
+  char raw_hdr[IVF_FILE_HDR_SZ];
+  int is_ivf = 0;
+  FILE *infile = input->file;
+  unsigned int *width = &input->w;
+  unsigned int *height = &input->h;
+  struct detect_buffer *detect = &input->detect;
 
-    if(memcmp(detect->buf, "DKIF", 4) != 0)
-        return 0;
+  if (memcmp(detect->buf, "DKIF", 4) != 0)
+    return 0;
 
-    /* See write_ivf_file_header() for more documentation on the file header
-     * layout.
-     */
-    if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
-        == IVF_FILE_HDR_SZ - 4)
+  /* See write_ivf_file_header() for more documentation on the file header
+   * layout.
+   */
+  if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
+      == IVF_FILE_HDR_SZ - 4) {
     {
-        {
-            is_ivf = 1;
+      is_ivf = 1;
 
-            if (mem_get_le16(raw_hdr + 4) != 0)
-                warn("Unrecognized IVF version! This file may not decode "
-                     "properly.");
+      if (mem_get_le16(raw_hdr + 4) != 0)
+        warn("Unrecognized IVF version! This file may not decode "
+             "properly.");
 
-            *fourcc = mem_get_le32(raw_hdr + 8);
-        }
+      *fourcc = mem_get_le32(raw_hdr + 8);
     }
+  }
 
-    if (is_ivf)
-    {
-        *width = mem_get_le16(raw_hdr + 12);
-        *height = mem_get_le16(raw_hdr + 14);
-        detect->position = 4;
-    }
+  if (is_ivf) {
+    *width = mem_get_le16(raw_hdr + 12);
+    *height = mem_get_le16(raw_hdr + 14);
+    detect->position = 4;
+  }
 
-    return is_ivf;
+  return is_ivf;
 }
 
 
 static void write_ivf_file_header(FILE *outfile,
                                   const vpx_codec_enc_cfg_t *cfg,
                                   unsigned int fourcc,
-                                  int frame_cnt)
-{
-    char header[32];
+                                  int frame_cnt) {
+  char header[32];
 
-    if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
-        return;
+  if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
+    return;
 
-    header[0] = 'D';
-    header[1] = 'K';
-    header[2] = 'I';
-    header[3] = 'F';
-    mem_put_le16(header + 4,  0);                 /* version */
-    mem_put_le16(header + 6,  32);                /* headersize */
-    mem_put_le32(header + 8,  fourcc);            /* headersize */
-    mem_put_le16(header + 12, cfg->g_w);          /* width */
-    mem_put_le16(header + 14, cfg->g_h);          /* height */
-    mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
-    mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
-    mem_put_le32(header + 24, frame_cnt);         /* length */
-    mem_put_le32(header + 28, 0);                 /* unused */
+  header[0] = 'D';
+  header[1] = 'K';
+  header[2] = 'I';
+  header[3] = 'F';
+  mem_put_le16(header + 4,  0);                 /* version */
+  mem_put_le16(header + 6,  32);                /* headersize */
+  mem_put_le32(header + 8,  fourcc);            /* headersize */
+  mem_put_le16(header + 12, cfg->g_w);          /* width */
+  mem_put_le16(header + 14, cfg->g_h);          /* height */
+  mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
+  mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
+  mem_put_le32(header + 24, frame_cnt);         /* length */
+  mem_put_le32(header + 28, 0);                 /* unused */
 
-    (void) fwrite(header, 1, 32, outfile);
+  (void) fwrite(header, 1, 32, outfile);
 }
 
 
 static void write_ivf_frame_header(FILE *outfile,
-                                   const vpx_codec_cx_pkt_t *pkt)
-{
-    char             header[12];
-    vpx_codec_pts_t  pts;
+                                   const vpx_codec_cx_pkt_t *pkt) {
+  char             header[12];
+  vpx_codec_pts_t  pts;
 
-    if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
-        return;
+  if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
+    return;
 
-    pts = pkt->data.frame.pts;
-    mem_put_le32(header, (int)pkt->data.frame.sz);
-    mem_put_le32(header + 4, pts & 0xFFFFFFFF);
-    mem_put_le32(header + 8, pts >> 32);
+  pts = pkt->data.frame.pts;
+  mem_put_le32(header, (int)pkt->data.frame.sz);
+  mem_put_le32(header + 4, pts & 0xFFFFFFFF);
+  mem_put_le32(header + 8, pts >> 32);
 
-    (void) fwrite(header, 1, 12, outfile);
+  (void) fwrite(header, 1, 12, outfile);
 }
 
-static void write_ivf_frame_size(FILE *outfile, size_t size)
-{
-    char             header[4];
-    mem_put_le32(header, (int)size);
-    (void) fwrite(header, 1, 4, outfile);
+static void write_ivf_frame_size(FILE *outfile, size_t size) {
+  char             header[4];
+  mem_put_le32(header, (int)size);
+  (void) fwrite(header, 1, 4, outfile);
 }
 
 
 typedef off_t EbmlLoc;
 
 
-struct cue_entry
-{
-    unsigned int time;
-    uint64_t     loc;
+struct cue_entry {
+  unsigned int time;
+  uint64_t     loc;
 };
 
 
-struct EbmlGlobal
-{
-    int debug;
+struct EbmlGlobal {
+  int debug;
 
-    FILE    *stream;
-    int64_t last_pts_ms;
-    vpx_rational_t  framerate;
+  FILE    *stream;
+  int64_t last_pts_ms;
+  vpx_rational_t  framerate;
 
-    /* These pointers are to the start of an element */
-    off_t    position_reference;
-    off_t    seek_info_pos;
-    off_t    segment_info_pos;
-    off_t    track_pos;
-    off_t    cue_pos;
-    off_t    cluster_pos;
+  /* These pointers are to the start of an element */
+  off_t    position_reference;
+  off_t    seek_info_pos;
+  off_t    segment_info_pos;
+  off_t    track_pos;
+  off_t    cue_pos;
+  off_t    cluster_pos;
 
-    /* This pointer is to a specific element to be serialized */
-    off_t    track_id_pos;
+  /* This pointer is to a specific element to be serialized */
+  off_t    track_id_pos;
 
-    /* These pointers are to the size field of the element */
-    EbmlLoc  startSegment;
-    EbmlLoc  startCluster;
+  /* These pointers are to the size field of the element */
+  EbmlLoc  startSegment;
+  EbmlLoc  startCluster;
 
-    uint32_t cluster_timecode;
-    int      cluster_open;
+  uint32_t cluster_timecode;
+  int      cluster_open;
 
-    struct cue_entry *cue_list;
-    unsigned int      cues;
+  struct cue_entry *cue_list;
+  unsigned int      cues;
 
 };
 
 
-void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
-{
-    (void) fwrite(buffer_in, 1, len, glob->stream);
+void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len) {
+  (void) fwrite(buffer_in, 1, len, glob->stream);
 }
 
 #define WRITE_BUFFER(s) \
-for(i = len-1; i>=0; i--)\
-{ \
+  for(i = len-1; i>=0; i--)\
+  { \
     x = (char)(*(const s *)buffer_in >> (i * CHAR_BIT)); \
     Ebml_Write(glob, &x, 1); \
-}
-void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len)
-{
-    char x;
-    int i;
+  }
+void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len) {
+  char x;
+  int i;
 
-    /* buffer_size:
-     * 1 - int8_t;
-     * 2 - int16_t;
-     * 3 - int32_t;
-     * 4 - int64_t;
-     */
-    switch (buffer_size)
-    {
-        case 1:
-            WRITE_BUFFER(int8_t)
-            break;
-        case 2:
-            WRITE_BUFFER(int16_t)
-            break;
-        case 4:
-            WRITE_BUFFER(int32_t)
-            break;
-        case 8:
-            WRITE_BUFFER(int64_t)
-            break;
-        default:
-            break;
-    }
+  /* buffer_size:
+   * 1 - int8_t;
+   * 2 - int16_t;
+   * 3 - int32_t;
+   * 4 - int64_t;
+   */
+  switch (buffer_size) {
+    case 1:
+      WRITE_BUFFER(int8_t)
+      break;
+    case 2:
+      WRITE_BUFFER(int16_t)
+      break;
+    case 4:
+      WRITE_BUFFER(int32_t)
+      break;
+    case 8:
+      WRITE_BUFFER(int64_t)
+      break;
+    default:
+      break;
+  }
 }
 #undef WRITE_BUFFER
 
 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit
  * one, but not a 32 bit one.
  */
-static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
-{
-    unsigned char sizeSerialized = 4 | 0x80;
-    Ebml_WriteID(glob, class_id);
-    Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
-    Ebml_Serialize(glob, &ui, sizeof(ui), 4);
+static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui) {
+  unsigned char sizeSerialized = 4 | 0x80;
+  Ebml_WriteID(glob, class_id);
+  Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
+  Ebml_Serialize(glob, &ui, sizeof(ui), 4);
 }
 
 
 static void
 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
-                          unsigned long class_id)
-{
-    /* todo this is always taking 8 bytes, this may need later optimization */
-    /* this is a key that says length unknown */
-    uint64_t unknownLen = LITERALU64(0x01FFFFFF, 0xFFFFFFFF);
+                     unsigned long class_id) {
+  /* todo this is always taking 8 bytes, this may need later optimization */
+  /* this is a key that says length unknown */
+  uint64_t unknownLen = LITERALU64(0x01FFFFFF, 0xFFFFFFFF);
 
-    Ebml_WriteID(glob, class_id);
-    *ebmlLoc = ftello(glob->stream);
-    Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
+  Ebml_WriteID(glob, class_id);
+  *ebmlLoc = ftello(glob->stream);
+  Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
 }
 
 static void
-Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
-{
-    off_t pos;
-    uint64_t size;
+Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc) {
+  off_t pos;
+  uint64_t size;
 
-    /* Save the current stream pointer */
-    pos = ftello(glob->stream);
+  /* Save the current stream pointer */
+  pos = ftello(glob->stream);
 
-    /* Calculate the size of this element */
-    size = pos - *ebmlLoc - 8;
-    size |= LITERALU64(0x01000000,0x00000000);
+  /* Calculate the size of this element */
+  size = pos - *ebmlLoc - 8;
+  size |= LITERALU64(0x01000000, 0x00000000);
 
-    /* Seek back to the beginning of the element and write the new size */
-    fseeko(glob->stream, *ebmlLoc, SEEK_SET);
-    Ebml_Serialize(glob, &size, sizeof(size), 8);
+  /* Seek back to the beginning of the element and write the new size */
+  fseeko(glob->stream, *ebmlLoc, SEEK_SET);
+  Ebml_Serialize(glob, &size, sizeof(size), 8);
 
-    /* Reset the stream pointer */
-    fseeko(glob->stream, pos, SEEK_SET);
+  /* Reset the stream pointer */
+  fseeko(glob->stream, pos, SEEK_SET);
 }
 
 
 static void
-write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
-{
-    uint64_t offset = pos - ebml->position_reference;
+write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos) {
+  uint64_t offset = pos - ebml->position_reference;
+  EbmlLoc start;
+  Ebml_StartSubElement(ebml, &start, Seek);
+  Ebml_SerializeBinary(ebml, SeekID, id);
+  Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
+  Ebml_EndSubElement(ebml, &start);
+}
+
+
+static void
+write_webm_seek_info(EbmlGlobal *ebml) {
+
+  off_t pos;
+
+  /* Save the current stream pointer */
+  pos = ftello(ebml->stream);
+
+  if (ebml->seek_info_pos)
+    fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
+  else
+    ebml->seek_info_pos = pos;
+
+  {
     EbmlLoc start;
-    Ebml_StartSubElement(ebml, &start, Seek);
-    Ebml_SerializeBinary(ebml, SeekID, id);
-    Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
+
+    Ebml_StartSubElement(ebml, &start, SeekHead);
+    write_webm_seek_element(ebml, Tracks, ebml->track_pos);
+    write_webm_seek_element(ebml, Cues,   ebml->cue_pos);
+    write_webm_seek_element(ebml, Info,   ebml->segment_info_pos);
     Ebml_EndSubElement(ebml, &start);
-}
+  }
+  {
+    /* segment info */
+    EbmlLoc startInfo;
+    uint64_t frame_time;
+    char version_string[64];
 
-
-static void
-write_webm_seek_info(EbmlGlobal *ebml)
-{
-
-    off_t pos;
-
-    /* Save the current stream pointer */
-    pos = ftello(ebml->stream);
-
-    if(ebml->seek_info_pos)
-        fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
-    else
-        ebml->seek_info_pos = pos;
-
-    {
-        EbmlLoc start;
-
-        Ebml_StartSubElement(ebml, &start, SeekHead);
-        write_webm_seek_element(ebml, Tracks, ebml->track_pos);
-        write_webm_seek_element(ebml, Cues,   ebml->cue_pos);
-        write_webm_seek_element(ebml, Info,   ebml->segment_info_pos);
-        Ebml_EndSubElement(ebml, &start);
+    /* Assemble version string */
+    if (ebml->debug)
+      strcpy(version_string, "vpxenc");
+    else {
+      strcpy(version_string, "vpxenc ");
+      strncat(version_string,
+              vpx_codec_version_str(),
+              sizeof(version_string) - 1 - strlen(version_string));
     }
-    {
-        /* segment info */
-        EbmlLoc startInfo;
-        uint64_t frame_time;
-        char version_string[64];
 
-        /* Assemble version string */
-        if(ebml->debug)
-            strcpy(version_string, "vpxenc");
-        else
-        {
-            strcpy(version_string, "vpxenc ");
-            strncat(version_string,
-                    vpx_codec_version_str(),
-                    sizeof(version_string) - 1 - strlen(version_string));
-        }
-
-        frame_time = (uint64_t)1000 * ebml->framerate.den
-                     / ebml->framerate.num;
-        ebml->segment_info_pos = ftello(ebml->stream);
-        Ebml_StartSubElement(ebml, &startInfo, Info);
-        Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
-        Ebml_SerializeFloat(ebml, Segment_Duration,
-                            (double)(ebml->last_pts_ms + frame_time));
-        Ebml_SerializeString(ebml, 0x4D80, version_string);
-        Ebml_SerializeString(ebml, 0x5741, version_string);
-        Ebml_EndSubElement(ebml, &startInfo);
-    }
+    frame_time = (uint64_t)1000 * ebml->framerate.den
+                 / ebml->framerate.num;
+    ebml->segment_info_pos = ftello(ebml->stream);
+    Ebml_StartSubElement(ebml, &startInfo, Info);
+    Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
+    Ebml_SerializeFloat(ebml, Segment_Duration,
+                        (double)(ebml->last_pts_ms + frame_time));
+    Ebml_SerializeString(ebml, 0x4D80, version_string);
+    Ebml_SerializeString(ebml, 0x5741, version_string);
+    Ebml_EndSubElement(ebml, &startInfo);
+  }
 }
 
 
@@ -710,1858 +686,2108 @@
 write_webm_file_header(EbmlGlobal                *glob,
                        const vpx_codec_enc_cfg_t *cfg,
                        const struct vpx_rational *fps,
-                       stereo_format_t            stereo_fmt)
-{
+                       stereo_format_t            stereo_fmt,
+                       unsigned int               fourcc) {
+  {
+    EbmlLoc start;
+    Ebml_StartSubElement(glob, &start, EBML);
+    Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
+    Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1);
+    Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4);
+    Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8);
+    Ebml_SerializeString(glob, DocType, "webm");
+    Ebml_SerializeUnsigned(glob, DocTypeVersion, 2);
+    Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2);
+    Ebml_EndSubElement(glob, &start);
+  }
+  {
+    Ebml_StartSubElement(glob, &glob->startSegment, Segment);
+    glob->position_reference = ftello(glob->stream);
+    glob->framerate = *fps;
+    write_webm_seek_info(glob);
+
     {
+      EbmlLoc trackStart;
+      glob->track_pos = ftello(glob->stream);
+      Ebml_StartSubElement(glob, &trackStart, Tracks);
+      {
+        unsigned int trackNumber = 1;
+        uint64_t     trackID = 0;
+
         EbmlLoc start;
-        Ebml_StartSubElement(glob, &start, EBML);
-        Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
-        Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1);
-        Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4);
-        Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8);
-        Ebml_SerializeString(glob, DocType, "webm");
-        Ebml_SerializeUnsigned(glob, DocTypeVersion, 2);
-        Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2);
-        Ebml_EndSubElement(glob, &start);
-    }
-    {
-        Ebml_StartSubElement(glob, &glob->startSegment, Segment);
-        glob->position_reference = ftello(glob->stream);
-        glob->framerate = *fps;
-        write_webm_seek_info(glob);
-
+        Ebml_StartSubElement(glob, &start, TrackEntry);
+        Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
+        glob->track_id_pos = ftello(glob->stream);
+        Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
+        Ebml_SerializeUnsigned(glob, TrackType, 1);
+        Ebml_SerializeString(glob, CodecID,
+                             fourcc == VP8_FOURCC ? "V_VP8" : "V_VP9");
         {
-            EbmlLoc trackStart;
-            glob->track_pos = ftello(glob->stream);
-            Ebml_StartSubElement(glob, &trackStart, Tracks);
-            {
-                unsigned int trackNumber = 1;
-                uint64_t     trackID = 0;
+          unsigned int pixelWidth = cfg->g_w;
+          unsigned int pixelHeight = cfg->g_h;
+          float        frameRate   = (float)fps->num / (float)fps->den;
 
-                EbmlLoc start;
-                Ebml_StartSubElement(glob, &start, TrackEntry);
-                Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
-                glob->track_id_pos = ftello(glob->stream);
-                Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
-                Ebml_SerializeUnsigned(glob, TrackType, 1);
-                Ebml_SerializeString(glob, CodecID, "V_VP8");
-                {
-                    unsigned int pixelWidth = cfg->g_w;
-                    unsigned int pixelHeight = cfg->g_h;
-                    float        frameRate   = (float)fps->num/(float)fps->den;
-
-                    EbmlLoc videoStart;
-                    Ebml_StartSubElement(glob, &videoStart, Video);
-                    Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
-                    Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
-                    Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
-                    Ebml_SerializeFloat(glob, FrameRate, frameRate);
-                    Ebml_EndSubElement(glob, &videoStart);
-                }
-                Ebml_EndSubElement(glob, &start); /* Track Entry */
-            }
-            Ebml_EndSubElement(glob, &trackStart);
+          EbmlLoc videoStart;
+          Ebml_StartSubElement(glob, &videoStart, Video);
+          Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
+          Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
+          Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
+          Ebml_SerializeFloat(glob, FrameRate, frameRate);
+          Ebml_EndSubElement(glob, &videoStart);
         }
-        /* segment element is open */
+        Ebml_EndSubElement(glob, &start); /* Track Entry */
+      }
+      Ebml_EndSubElement(glob, &trackStart);
     }
+    /* segment element is open */
+  }
 }
 
 
 static void
 write_webm_block(EbmlGlobal                *glob,
                  const vpx_codec_enc_cfg_t *cfg,
-                 const vpx_codec_cx_pkt_t  *pkt)
-{
-    unsigned long  block_length;
-    unsigned char  track_number;
-    unsigned short block_timecode = 0;
-    unsigned char  flags;
-    int64_t        pts_ms;
-    int            start_cluster = 0, is_keyframe;
+                 const vpx_codec_cx_pkt_t  *pkt) {
+  unsigned long  block_length;
+  unsigned char  track_number;
+  unsigned short block_timecode = 0;
+  unsigned char  flags;
+  int64_t        pts_ms;
+  int            start_cluster = 0, is_keyframe;
 
-    /* Calculate the PTS of this frame in milliseconds */
-    pts_ms = pkt->data.frame.pts * 1000
-             * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
-    if(pts_ms <= glob->last_pts_ms)
-        pts_ms = glob->last_pts_ms + 1;
-    glob->last_pts_ms = pts_ms;
+  /* Calculate the PTS of this frame in milliseconds */
+  pts_ms = pkt->data.frame.pts * 1000
+           * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
+  if (pts_ms <= glob->last_pts_ms)
+    pts_ms = glob->last_pts_ms + 1;
+  glob->last_pts_ms = pts_ms;
 
-    /* Calculate the relative time of this block */
-    if(pts_ms - glob->cluster_timecode > SHRT_MAX)
-        start_cluster = 1;
-    else
-        block_timecode = (unsigned short)pts_ms - glob->cluster_timecode;
+  /* Calculate the relative time of this block */
+  if (pts_ms - glob->cluster_timecode > SHRT_MAX)
+    start_cluster = 1;
+  else
+    block_timecode = (unsigned short)pts_ms - glob->cluster_timecode;
 
-    is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
-    if(start_cluster || is_keyframe)
-    {
-        if(glob->cluster_open)
-            Ebml_EndSubElement(glob, &glob->startCluster);
+  is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
+  if (start_cluster || is_keyframe) {
+    if (glob->cluster_open)
+      Ebml_EndSubElement(glob, &glob->startCluster);
 
-        /* Open the new cluster */
-        block_timecode = 0;
-        glob->cluster_open = 1;
-        glob->cluster_timecode = (uint32_t)pts_ms;
-        glob->cluster_pos = ftello(glob->stream);
-        Ebml_StartSubElement(glob, &glob->startCluster, Cluster); /* cluster */
-        Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
+    /* Open the new cluster */
+    block_timecode = 0;
+    glob->cluster_open = 1;
+    glob->cluster_timecode = (uint32_t)pts_ms;
+    glob->cluster_pos = ftello(glob->stream);
+    Ebml_StartSubElement(glob, &glob->startCluster, Cluster); /* cluster */
+    Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
 
-        /* Save a cue point if this is a keyframe. */
-        if(is_keyframe)
-        {
-            struct cue_entry *cue, *new_cue_list;
+    /* Save a cue point if this is a keyframe. */
+    if (is_keyframe) {
+      struct cue_entry *cue, *new_cue_list;
 
-            new_cue_list = realloc(glob->cue_list,
-                                   (glob->cues+1) * sizeof(struct cue_entry));
-            if(new_cue_list)
-                glob->cue_list = new_cue_list;
-            else
-                fatal("Failed to realloc cue list.");
+      new_cue_list = realloc(glob->cue_list,
+                             (glob->cues + 1) * sizeof(struct cue_entry));
+      if (new_cue_list)
+        glob->cue_list = new_cue_list;
+      else
+        fatal("Failed to realloc cue list.");
 
-            cue = &glob->cue_list[glob->cues];
-            cue->time = glob->cluster_timecode;
-            cue->loc = glob->cluster_pos;
-            glob->cues++;
-        }
+      cue = &glob->cue_list[glob->cues];
+      cue->time = glob->cluster_timecode;
+      cue->loc = glob->cluster_pos;
+      glob->cues++;
     }
+  }
 
-    /* Write the Simple Block */
-    Ebml_WriteID(glob, SimpleBlock);
+  /* Write the Simple Block */
+  Ebml_WriteID(glob, SimpleBlock);
 
-    block_length = (unsigned long)pkt->data.frame.sz + 4;
-    block_length |= 0x10000000;
-    Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
+  block_length = (unsigned long)pkt->data.frame.sz + 4;
+  block_length |= 0x10000000;
+  Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
 
-    track_number = 1;
-    track_number |= 0x80;
-    Ebml_Write(glob, &track_number, 1);
+  track_number = 1;
+  track_number |= 0x80;
+  Ebml_Write(glob, &track_number, 1);
 
-    Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
+  Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
 
-    flags = 0;
-    if(is_keyframe)
-        flags |= 0x80;
-    if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
-        flags |= 0x08;
-    Ebml_Write(glob, &flags, 1);
+  flags = 0;
+  if (is_keyframe)
+    flags |= 0x80;
+  if (pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
+    flags |= 0x08;
+  Ebml_Write(glob, &flags, 1);
 
-    Ebml_Write(glob, pkt->data.frame.buf, (unsigned long)pkt->data.frame.sz);
+  Ebml_Write(glob, pkt->data.frame.buf, (unsigned long)pkt->data.frame.sz);
 }
 
 
 static void
-write_webm_file_footer(EbmlGlobal *glob, long hash)
-{
+write_webm_file_footer(EbmlGlobal *glob, long hash) {
 
-    if(glob->cluster_open)
-        Ebml_EndSubElement(glob, &glob->startCluster);
+  if (glob->cluster_open)
+    Ebml_EndSubElement(glob, &glob->startCluster);
 
-    {
+  {
+    EbmlLoc start;
+    unsigned int i;
+
+    glob->cue_pos = ftello(glob->stream);
+    Ebml_StartSubElement(glob, &start, Cues);
+    for (i = 0; i < glob->cues; i++) {
+      struct cue_entry *cue = &glob->cue_list[i];
+      EbmlLoc start;
+
+      Ebml_StartSubElement(glob, &start, CuePoint);
+      {
         EbmlLoc start;
-        unsigned int i;
 
-        glob->cue_pos = ftello(glob->stream);
-        Ebml_StartSubElement(glob, &start, Cues);
-        for(i=0; i<glob->cues; i++)
-        {
-            struct cue_entry *cue = &glob->cue_list[i];
-            EbmlLoc start;
+        Ebml_SerializeUnsigned(glob, CueTime, cue->time);
 
-            Ebml_StartSubElement(glob, &start, CuePoint);
-            {
-                EbmlLoc start;
-
-                Ebml_SerializeUnsigned(glob, CueTime, cue->time);
-
-                Ebml_StartSubElement(glob, &start, CueTrackPositions);
-                Ebml_SerializeUnsigned(glob, CueTrack, 1);
-                Ebml_SerializeUnsigned64(glob, CueClusterPosition,
-                                         cue->loc - glob->position_reference);
-                Ebml_EndSubElement(glob, &start);
-            }
-            Ebml_EndSubElement(glob, &start);
-        }
+        Ebml_StartSubElement(glob, &start, CueTrackPositions);
+        Ebml_SerializeUnsigned(glob, CueTrack, 1);
+        Ebml_SerializeUnsigned64(glob, CueClusterPosition,
+                                 cue->loc - glob->position_reference);
         Ebml_EndSubElement(glob, &start);
+      }
+      Ebml_EndSubElement(glob, &start);
     }
+    Ebml_EndSubElement(glob, &start);
+  }
 
-    Ebml_EndSubElement(glob, &glob->startSegment);
+  Ebml_EndSubElement(glob, &glob->startSegment);
 
-    /* Patch up the seek info block */
-    write_webm_seek_info(glob);
+  /* Patch up the seek info block */
+  write_webm_seek_info(glob);
 
-    /* Patch up the track id */
-    fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
-    Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
+  /* Patch up the track id */
+  fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
+  Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
 
-    fseeko(glob->stream, 0, SEEK_END);
+  fseeko(glob->stream, 0, SEEK_END);
 }
 
 
 /* Murmur hash derived from public domain reference implementation at
- *   http://sites.google.com/site/murmurhash/
+ *   http:// sites.google.com/site/murmurhash/
  */
-static unsigned int murmur ( const void * key, int len, unsigned int seed )
-{
-    const unsigned int m = 0x5bd1e995;
-    const int r = 24;
+static unsigned int murmur(const void *key, int len, unsigned int seed) {
+  const unsigned int m = 0x5bd1e995;
+  const int r = 24;
 
-    unsigned int h = seed ^ len;
+  unsigned int h = seed ^ len;
 
-    const unsigned char * data = (const unsigned char *)key;
+  const unsigned char *data = (const unsigned char *)key;
 
-    while(len >= 4)
-    {
-        unsigned int k;
+  while (len >= 4) {
+    unsigned int k;
 
-        k  = data[0];
-        k |= data[1] << 8;
-        k |= data[2] << 16;
-        k |= data[3] << 24;
+    k  = data[0];
+    k |= data[1] << 8;
+    k |= data[2] << 16;
+    k |= data[3] << 24;
 
-        k *= m;
-        k ^= k >> r;
-        k *= m;
+    k *= m;
+    k ^= k >> r;
+    k *= m;
 
-        h *= m;
-        h ^= k;
-
-        data += 4;
-        len -= 4;
-    }
-
-    switch(len)
-    {
-    case 3: h ^= data[2] << 16;
-    case 2: h ^= data[1] << 8;
-    case 1: h ^= data[0];
-            h *= m;
-    };
-
-    h ^= h >> 13;
     h *= m;
-    h ^= h >> 15;
+    h ^= k;
 
-    return h;
+    data += 4;
+    len -= 4;
+  }
+
+  switch (len) {
+    case 3:
+      h ^= data[2] << 16;
+    case 2:
+      h ^= data[1] << 8;
+    case 1:
+      h ^= data[0];
+      h *= m;
+  };
+
+  h ^= h >> 13;
+  h *= m;
+  h ^= h >> 15;
+
+  return h;
 }
 
 #include "math.h"
+#define MAX_PSNR 100
+static double vp8_mse2psnr(double Samples, double Peak, double Mse) {
+  double psnr;
 
-static double vp8_mse2psnr(double Samples, double Peak, double Mse)
-{
-    double psnr;
+  if ((double)Mse > 0.0)
+    psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
+  else
+    psnr = MAX_PSNR;      /* Limit to prevent / 0 */
 
-    if ((double)Mse > 0.0)
-        psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
-    else
-        psnr = 60;      /* Limit to prevent / 0 */
+  if (psnr > MAX_PSNR)
+    psnr = MAX_PSNR;
 
-    if (psnr > 60)
-        psnr = 60;
-
-    return psnr;
+  return psnr;
 }
 
 
 #include "args.h"
 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
-        "Debug mode (makes output deterministic)");
+                                           "Debug mode (makes output deterministic)");
 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
-        "Output filename");
+                                            "Output filename");
 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
-                                  "Input file is YV12 ");
+                                          "Input file is YV12 ");
 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
-                                  "Input file is I420 (default)");
+                                          "Input file is I420 (default)");
 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
-                                  "Codec to use");
+                                          "Codec to use");
 static const arg_def_t passes           = ARG_DEF("p", "passes", 1,
-        "Number of passes (1/2)");
+                                                  "Number of passes (1/2)");
 static const arg_def_t pass_arg         = ARG_DEF(NULL, "pass", 1,
-        "Pass to execute (1/2)");
+                                                  "Pass to execute (1/2)");
 static const arg_def_t fpf_name         = ARG_DEF(NULL, "fpf", 1,
-        "First pass statistics file name");
+                                                  "First pass statistics file name");
 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
                                        "Stop encoding after n input frames");
+static const arg_def_t skip = ARG_DEF(NULL, "skip", 1,
+                                      "Skip the first n input frames");
 static const arg_def_t deadline         = ARG_DEF("d", "deadline", 1,
-        "Deadline per frame (usec)");
+                                                  "Deadline per frame (usec)");
 static const arg_def_t best_dl          = ARG_DEF(NULL, "best", 0,
-        "Use Best Quality Deadline");
+                                                  "Use Best Quality Deadline");
 static const arg_def_t good_dl          = ARG_DEF(NULL, "good", 0,
-        "Use Good Quality Deadline");
+                                                  "Use Good Quality Deadline");
 static const arg_def_t rt_dl            = ARG_DEF(NULL, "rt", 0,
-        "Use Realtime Quality Deadline");
+                                                  "Use Realtime Quality Deadline");
 static const arg_def_t quietarg         = ARG_DEF("q", "quiet", 0,
-        "Do not print encode progress");
+                                                  "Do not print encode progress");
 static const arg_def_t verbosearg       = ARG_DEF("v", "verbose", 0,
-        "Show encoder parameters");
+                                                  "Show encoder parameters");
 static const arg_def_t psnrarg          = ARG_DEF(NULL, "psnr", 0,
-        "Show PSNR in status line");
+                                                  "Show PSNR in status line");
+enum TestDecodeFatality {
+  TEST_DECODE_OFF,
+  TEST_DECODE_FATAL,
+  TEST_DECODE_WARN,
+};
+static const struct arg_enum_list test_decode_enum[] = {
+  {"off",   TEST_DECODE_OFF},
+  {"fatal", TEST_DECODE_FATAL},
+  {"warn",  TEST_DECODE_WARN},
+  {NULL, 0}
+};
+static const arg_def_t recontest = ARG_DEF_ENUM(NULL, "test-decode", 1,
+                                                "Test encode/decode mismatch",
+                                                test_decode_enum);
 static const arg_def_t framerate        = ARG_DEF(NULL, "fps", 1,
-        "Stream frame rate (rate/scale)");
+                                                  "Stream frame rate (rate/scale)");
 static const arg_def_t use_ivf          = ARG_DEF(NULL, "ivf", 0,
-        "Output IVF (default is WebM)");
+                                                  "Output IVF (default is WebM)");
 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0,
-        "Makes encoder output partitions. Requires IVF output!");
+                                          "Makes encoder output partitions. Requires IVF output!");
 static const arg_def_t q_hist_n         = ARG_DEF(NULL, "q-hist", 1,
-        "Show quantizer histogram (n-buckets)");
+                                                  "Show quantizer histogram (n-buckets)");
 static const arg_def_t rate_hist_n         = ARG_DEF(NULL, "rate-hist", 1,
-        "Show rate histogram (n-buckets)");
-static const arg_def_t *main_args[] =
-{
-    &debugmode,
-    &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
-    &best_dl, &good_dl, &rt_dl,
-    &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n,
-    NULL
+                                                     "Show rate histogram (n-buckets)");
+static const arg_def_t *main_args[] = {
+  &debugmode,
+  &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &skip,
+  &deadline, &best_dl, &good_dl, &rt_dl,
+  &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n,
+  NULL
 };
 
 static const arg_def_t usage            = ARG_DEF("u", "usage", 1,
-        "Usage profile number to use");
+                                                  "Usage profile number to use");
 static const arg_def_t threads          = ARG_DEF("t", "threads", 1,
-        "Max number of threads to use");
+                                                  "Max number of threads to use");
 static const arg_def_t profile          = ARG_DEF(NULL, "profile", 1,
-        "Bitstream profile number to use");
+                                                  "Bitstream profile number to use");
 static const arg_def_t width            = ARG_DEF("w", "width", 1,
-        "Frame width");
+                                                  "Frame width");
 static const arg_def_t height           = ARG_DEF("h", "height", 1,
-        "Frame height");
+                                                  "Frame height");
 static const struct arg_enum_list stereo_mode_enum[] = {
-    {"mono"      , STEREO_FORMAT_MONO},
-    {"left-right", STEREO_FORMAT_LEFT_RIGHT},
-    {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
-    {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
-    {"right-left", STEREO_FORMAT_RIGHT_LEFT},
-    {NULL, 0}
+  {"mono", STEREO_FORMAT_MONO},
+  {"left-right", STEREO_FORMAT_LEFT_RIGHT},
+  {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
+  {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
+  {"right-left", STEREO_FORMAT_RIGHT_LEFT},
+  {NULL, 0}
 };
 static const arg_def_t stereo_mode      = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
-        "Stereo 3D video format", stereo_mode_enum);
+                                                       "Stereo 3D video format", stereo_mode_enum);
 static const arg_def_t timebase         = ARG_DEF(NULL, "timebase", 1,
-        "Output timestamp precision (fractional seconds)");
+                                                  "Output timestamp precision (fractional seconds)");
 static const arg_def_t error_resilient  = ARG_DEF(NULL, "error-resilient", 1,
-        "Enable error resiliency features");
+                                                  "Enable error resiliency features");
 static const arg_def_t lag_in_frames    = ARG_DEF(NULL, "lag-in-frames", 1,
-        "Max number of frames to lag");
+                                                  "Max number of frames to lag");
 
-static const arg_def_t *global_args[] =
-{
-    &use_yv12, &use_i420, &usage, &threads, &profile,
-    &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
-    &lag_in_frames, NULL
+static const arg_def_t *global_args[] = {
+  &use_yv12, &use_i420, &usage, &threads, &profile,
+  &width, &height, &stereo_mode, &timebase, &framerate,
+  &error_resilient,
+  &lag_in_frames, NULL
 };
 
 static const arg_def_t dropframe_thresh   = ARG_DEF(NULL, "drop-frame", 1,
-        "Temporal resampling threshold (buf %)");
+                                                    "Temporal resampling threshold (buf %)");
 static const arg_def_t resize_allowed     = ARG_DEF(NULL, "resize-allowed", 1,
-        "Spatial resampling enabled (bool)");
+                                                    "Spatial resampling enabled (bool)");
 static const arg_def_t resize_up_thresh   = ARG_DEF(NULL, "resize-up", 1,
-        "Upscale threshold (buf %)");
+                                                    "Upscale threshold (buf %)");
 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
-        "Downscale threshold (buf %)");
+                                                    "Downscale threshold (buf %)");
 static const struct arg_enum_list end_usage_enum[] = {
-    {"vbr", VPX_VBR},
-    {"cbr", VPX_CBR},
-    {"cq",  VPX_CQ},
-    {NULL, 0}
+  {"vbr", VPX_VBR},
+  {"cbr", VPX_CBR},
+  {"cq",  VPX_CQ},
+  {NULL, 0}
 };
 static const arg_def_t end_usage          = ARG_DEF_ENUM(NULL, "end-usage", 1,
-        "Rate control mode", end_usage_enum);
+                                                         "Rate control mode", end_usage_enum);
 static const arg_def_t target_bitrate     = ARG_DEF(NULL, "target-bitrate", 1,
-        "Bitrate (kbps)");
+                                                    "Bitrate (kbps)");
 static const arg_def_t min_quantizer      = ARG_DEF(NULL, "min-q", 1,
-        "Minimum (best) quantizer");
+                                                    "Minimum (best) quantizer");
 static const arg_def_t max_quantizer      = ARG_DEF(NULL, "max-q", 1,
-        "Maximum (worst) quantizer");
+                                                    "Maximum (worst) quantizer");
 static const arg_def_t undershoot_pct     = ARG_DEF(NULL, "undershoot-pct", 1,
-        "Datarate undershoot (min) target (%)");
+                                                    "Datarate undershoot (min) target (%)");
 static const arg_def_t overshoot_pct      = ARG_DEF(NULL, "overshoot-pct", 1,
-        "Datarate overshoot (max) target (%)");
+                                                    "Datarate overshoot (max) target (%)");
 static const arg_def_t buf_sz             = ARG_DEF(NULL, "buf-sz", 1,
-        "Client buffer size (ms)");
+                                                    "Client buffer size (ms)");
 static const arg_def_t buf_initial_sz     = ARG_DEF(NULL, "buf-initial-sz", 1,
-        "Client initial buffer size (ms)");
+                                                    "Client initial buffer size (ms)");
 static const arg_def_t buf_optimal_sz     = ARG_DEF(NULL, "buf-optimal-sz", 1,
-        "Client optimal buffer size (ms)");
-static const arg_def_t *rc_args[] =
-{
-    &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
-    &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
-    &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
-    NULL
+                                                    "Client optimal buffer size (ms)");
+static const arg_def_t *rc_args[] = {
+  &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
+  &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
+  &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
+  NULL
 };
 
 
 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
-                                  "CBR/VBR bias (0=CBR, 100=VBR)");
+                                          "CBR/VBR bias (0=CBR, 100=VBR)");
 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
-                                        "GOP min bitrate (% of target)");
+                                                "GOP min bitrate (% of target)");
 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
-                                        "GOP max bitrate (% of target)");
-static const arg_def_t *rc_twopass_args[] =
-{
-    &bias_pct, &minsection_pct, &maxsection_pct, NULL
+                                                "GOP max bitrate (% of target)");
+static const arg_def_t *rc_twopass_args[] = {
+  &bias_pct, &minsection_pct, &maxsection_pct, NULL
 };
 
 
 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
-                                     "Minimum keyframe interval (frames)");
+                                             "Minimum keyframe interval (frames)");
 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
-                                     "Maximum keyframe interval (frames)");
+                                             "Maximum keyframe interval (frames)");
 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
-                                     "Disable keyframe placement");
-static const arg_def_t *kf_args[] =
-{
-    &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
+                                             "Disable keyframe placement");
+static const arg_def_t *kf_args[] = {
+  &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
 };
 
 
-#if CONFIG_VP8_ENCODER
 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
-                                    "Noise sensitivity (frames to blur)");
+                                            "Noise sensitivity (frames to blur)");
 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
-                                   "Filter sharpness (0-7)");
+                                           "Filter sharpness (0-7)");
 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
-                                       "Motion detection threshold");
-#endif
-
-#if CONFIG_VP8_ENCODER
+                                               "Motion detection threshold");
 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
-                                  "CPU Used (-16..16)");
-#endif
-
-
-#if CONFIG_VP8_ENCODER
+                                          "CPU Used (-16..16)");
 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
                                      "Number of token partitions to use, log2");
+static const arg_def_t tile_cols = ARG_DEF(NULL, "tile-columns", 1,
+                                         "Number of tile columns to use, log2");
+static const arg_def_t tile_rows = ARG_DEF(NULL, "tile-rows", 1,
+                                           "Number of tile rows to use, log2");
 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
-                                     "Enable automatic alt reference frames");
+                                             "Enable automatic alt reference frames");
 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
-                                        "AltRef Max Frames");
+                                                "AltRef Max Frames");
 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
-                                       "AltRef Strength");
+                                               "AltRef Strength");
 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
-                                   "AltRef Type");
+                                           "AltRef Type");
 static const struct arg_enum_list tuning_enum[] = {
-    {"psnr", VP8_TUNE_PSNR},
-    {"ssim", VP8_TUNE_SSIM},
-    {NULL, 0}
+  {"psnr", VP8_TUNE_PSNR},
+  {"ssim", VP8_TUNE_SSIM},
+  {NULL, 0}
 };
 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
-                                   "Material to favor", tuning_enum);
+                                                "Material to favor", tuning_enum);
 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
-                                   "Constrained Quality Level");
+                                          "Constrained Quality Level");
 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
-        "Max I-frame bitrate (pct)");
+                                                    "Max I-frame bitrate (pct)");
+static const arg_def_t lossless = ARG_DEF(NULL, "lossless", 1, "Lossless mode");
+#if CONFIG_VP9_ENCODER
+static const arg_def_t frame_parallel_decoding  = ARG_DEF(
+    NULL, "frame-parallel", 1, "Enable frame parallel decodability features");
+#endif
 
-static const arg_def_t *vp8_args[] =
-{
-    &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
-    &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
-    &tune_ssim, &cq_level, &max_intra_rate_pct, NULL
+#if CONFIG_VP8_ENCODER
+static const arg_def_t *vp8_args[] = {
+  &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
+  &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
+  &tune_ssim, &cq_level, &max_intra_rate_pct,
+  NULL
 };
-static const int vp8_arg_ctrl_map[] =
-{
-    VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
-    VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
-    VP8E_SET_TOKEN_PARTITIONS,
-    VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
-    VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0
+static const int vp8_arg_ctrl_map[] = {
+  VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
+  VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
+  VP8E_SET_TOKEN_PARTITIONS,
+  VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
+  VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
+  0
+};
+#endif
+
+#if CONFIG_VP9_ENCODER
+static const arg_def_t *vp9_args[] = {
+  &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
+  &tile_cols, &tile_rows, &arnr_maxframes, &arnr_strength, &arnr_type,
+  &tune_ssim, &cq_level, &max_intra_rate_pct, &lossless,
+  &frame_parallel_decoding,
+  NULL
+};
+static const int vp9_arg_ctrl_map[] = {
+  VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
+  VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
+  VP9E_SET_TILE_COLUMNS, VP9E_SET_TILE_ROWS,
+  VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
+  VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
+  VP9E_SET_LOSSLESS, VP9E_SET_FRAME_PARALLEL_DECODING,
+  0
 };
 #endif
 
 static const arg_def_t *no_args[] = { NULL };
 
-static void usage_exit()
-{
-    int i;
+static void usage_exit() {
+  int i;
 
-    fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
-            exec_name);
+  fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
+          exec_name);
 
-    fprintf(stderr, "\nOptions:\n");
-    arg_show_usage(stdout, main_args);
-    fprintf(stderr, "\nEncoder Global Options:\n");
-    arg_show_usage(stdout, global_args);
-    fprintf(stderr, "\nRate Control Options:\n");
-    arg_show_usage(stdout, rc_args);
-    fprintf(stderr, "\nTwopass Rate Control Options:\n");
-    arg_show_usage(stdout, rc_twopass_args);
-    fprintf(stderr, "\nKeyframe Placement Options:\n");
-    arg_show_usage(stdout, kf_args);
+  fprintf(stderr, "\nOptions:\n");
+  arg_show_usage(stdout, main_args);
+  fprintf(stderr, "\nEncoder Global Options:\n");
+  arg_show_usage(stdout, global_args);
+  fprintf(stderr, "\nRate Control Options:\n");
+  arg_show_usage(stdout, rc_args);
+  fprintf(stderr, "\nTwopass Rate Control Options:\n");
+  arg_show_usage(stdout, rc_twopass_args);
+  fprintf(stderr, "\nKeyframe Placement Options:\n");
+  arg_show_usage(stdout, kf_args);
 #if CONFIG_VP8_ENCODER
-    fprintf(stderr, "\nVP8 Specific Options:\n");
-    arg_show_usage(stdout, vp8_args);
+  fprintf(stderr, "\nVP8 Specific Options:\n");
+  arg_show_usage(stdout, vp8_args);
 #endif
-    fprintf(stderr, "\nStream timebase (--timebase):\n"
-            "  The desired precision of timestamps in the output, expressed\n"
-            "  in fractional seconds. Default is 1/1000.\n");
-    fprintf(stderr, "\n"
-           "Included encoders:\n"
-           "\n");
+#if CONFIG_VP9_ENCODER
+  fprintf(stderr, "\nVP9 Specific Options:\n");
+  arg_show_usage(stdout, vp9_args);
+#endif
+  fprintf(stderr, "\nStream timebase (--timebase):\n"
+          "  The desired precision of timestamps in the output, expressed\n"
+          "  in fractional seconds. Default is 1/1000.\n");
+  fprintf(stderr, "\n"
+          "Included encoders:\n"
+          "\n");
 
-    for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
-        fprintf(stderr, "    %-6s - %s\n",
-               codecs[i].name,
-               vpx_codec_iface_name(codecs[i].iface));
+  for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
+    fprintf(stderr, "    %-6s - %s\n",
+            codecs[i].name,
+            vpx_codec_iface_name(codecs[i].iface()));
 
-    exit(EXIT_FAILURE);
+  exit(EXIT_FAILURE);
 }
 
 
 #define HIST_BAR_MAX 40
-struct hist_bucket
-{
-    int low, high, count;
+struct hist_bucket {
+  int low, high, count;
 };
 
 
 static int merge_hist_buckets(struct hist_bucket *bucket,
                               int *buckets_,
-                              int max_buckets)
-{
-    int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
-    int buckets = *buckets_;
-    int i;
+                              int max_buckets) {
+  int small_bucket = 0, merge_bucket = INT_MAX, big_bucket = 0;
+  int buckets = *buckets_;
+  int i;
 
-    /* Find the extrema for this list of buckets */
-    big_bucket = small_bucket = 0;
-    for(i=0; i < buckets; i++)
-    {
-        if(bucket[i].count < bucket[small_bucket].count)
-            small_bucket = i;
-        if(bucket[i].count > bucket[big_bucket].count)
-            big_bucket = i;
+  /* Find the extrema for this list of buckets */
+  big_bucket = small_bucket = 0;
+  for (i = 0; i < buckets; i++) {
+    if (bucket[i].count < bucket[small_bucket].count)
+      small_bucket = i;
+    if (bucket[i].count > bucket[big_bucket].count)
+      big_bucket = i;
+  }
+
+  /* If we have too many buckets, merge the smallest with an adjacent
+   * bucket.
+   */
+  while (buckets > max_buckets) {
+    int last_bucket = buckets - 1;
+
+    /* merge the small bucket with an adjacent one. */
+    if (small_bucket == 0)
+      merge_bucket = 1;
+    else if (small_bucket == last_bucket)
+      merge_bucket = last_bucket - 1;
+    else if (bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
+      merge_bucket = small_bucket - 1;
+    else
+      merge_bucket = small_bucket + 1;
+
+    assert(abs(merge_bucket - small_bucket) <= 1);
+    assert(small_bucket < buckets);
+    assert(big_bucket < buckets);
+    assert(merge_bucket < buckets);
+
+    if (merge_bucket < small_bucket) {
+      bucket[merge_bucket].high = bucket[small_bucket].high;
+      bucket[merge_bucket].count += bucket[small_bucket].count;
+    } else {
+      bucket[small_bucket].high = bucket[merge_bucket].high;
+      bucket[small_bucket].count += bucket[merge_bucket].count;
+      merge_bucket = small_bucket;
     }
 
-    /* If we have too many buckets, merge the smallest with an adjacent
-     * bucket.
+    assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
+
+    buckets--;
+
+    /* Remove the merge_bucket from the list, and find the new small
+     * and big buckets while we're at it
      */
-    while(buckets > max_buckets)
-    {
-        int last_bucket = buckets - 1;
+    big_bucket = small_bucket = 0;
+    for (i = 0; i < buckets; i++) {
+      if (i > merge_bucket)
+        bucket[i] = bucket[i + 1];
 
-        /* merge the small bucket with an adjacent one. */
-        if(small_bucket == 0)
-            merge_bucket = 1;
-        else if(small_bucket == last_bucket)
-            merge_bucket = last_bucket - 1;
-        else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
-            merge_bucket = small_bucket - 1;
-        else
-            merge_bucket = small_bucket + 1;
-
-        assert(abs(merge_bucket - small_bucket) <= 1);
-        assert(small_bucket < buckets);
-        assert(big_bucket < buckets);
-        assert(merge_bucket < buckets);
-
-        if(merge_bucket < small_bucket)
-        {
-            bucket[merge_bucket].high = bucket[small_bucket].high;
-            bucket[merge_bucket].count += bucket[small_bucket].count;
-        }
-        else
-        {
-            bucket[small_bucket].high = bucket[merge_bucket].high;
-            bucket[small_bucket].count += bucket[merge_bucket].count;
-            merge_bucket = small_bucket;
-        }
-
-        assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
-
-        buckets--;
-
-        /* Remove the merge_bucket from the list, and find the new small
-         * and big buckets while we're at it
-         */
-        big_bucket = small_bucket = 0;
-        for(i=0; i < buckets; i++)
-        {
-            if(i > merge_bucket)
-                bucket[i] = bucket[i+1];
-
-            if(bucket[i].count < bucket[small_bucket].count)
-                small_bucket = i;
-            if(bucket[i].count > bucket[big_bucket].count)
-                big_bucket = i;
-        }
-
+      if (bucket[i].count < bucket[small_bucket].count)
+        small_bucket = i;
+      if (bucket[i].count > bucket[big_bucket].count)
+        big_bucket = i;
     }
 
-    *buckets_ = buckets;
-    return bucket[big_bucket].count;
+  }
+
+  *buckets_ = buckets;
+  return bucket[big_bucket].count;
 }
 
 
 static void show_histogram(const struct hist_bucket *bucket,
                            int                       buckets,
                            int                       total,
-                           int                       scale)
-{
-    const char *pat1, *pat2;
-    int i;
+                           int                       scale) {
+  const char *pat1, *pat2;
+  int i;
 
-    switch((int)(log(bucket[buckets-1].high)/log(10))+1)
-    {
-        case 1:
-        case 2:
-            pat1 = "%4d %2s: ";
-            pat2 = "%4d-%2d: ";
-            break;
-        case 3:
-            pat1 = "%5d %3s: ";
-            pat2 = "%5d-%3d: ";
-            break;
-        case 4:
-            pat1 = "%6d %4s: ";
-            pat2 = "%6d-%4d: ";
-            break;
-        case 5:
-            pat1 = "%7d %5s: ";
-            pat2 = "%7d-%5d: ";
-            break;
-        case 6:
-            pat1 = "%8d %6s: ";
-            pat2 = "%8d-%6d: ";
-            break;
-        case 7:
-            pat1 = "%9d %7s: ";
-            pat2 = "%9d-%7d: ";
-            break;
-        default:
-            pat1 = "%12d %10s: ";
-            pat2 = "%12d-%10d: ";
-            break;
-    }
+  switch ((int)(log(bucket[buckets - 1].high) / log(10)) + 1) {
+    case 1:
+    case 2:
+      pat1 = "%4d %2s: ";
+      pat2 = "%4d-%2d: ";
+      break;
+    case 3:
+      pat1 = "%5d %3s: ";
+      pat2 = "%5d-%3d: ";
+      break;
+    case 4:
+      pat1 = "%6d %4s: ";
+      pat2 = "%6d-%4d: ";
+      break;
+    case 5:
+      pat1 = "%7d %5s: ";
+      pat2 = "%7d-%5d: ";
+      break;
+    case 6:
+      pat1 = "%8d %6s: ";
+      pat2 = "%8d-%6d: ";
+      break;
+    case 7:
+      pat1 = "%9d %7s: ";
+      pat2 = "%9d-%7d: ";
+      break;
+    default:
+      pat1 = "%12d %10s: ";
+      pat2 = "%12d-%10d: ";
+      break;
+  }
 
-    for(i=0; i<buckets; i++)
-    {
-        int len;
-        int j;
-        float pct;
+  for (i = 0; i < buckets; i++) {
+    int len;
+    int j;
+    float pct;
 
-        pct = (float)(100.0 * bucket[i].count / total);
-        len = HIST_BAR_MAX * bucket[i].count / scale;
-        if(len < 1)
-            len = 1;
-        assert(len <= HIST_BAR_MAX);
+    pct = (float)(100.0 * bucket[i].count / total);
+    len = HIST_BAR_MAX * bucket[i].count / scale;
+    if (len < 1)
+      len = 1;
+    assert(len <= HIST_BAR_MAX);
 
-        if(bucket[i].low == bucket[i].high)
-            fprintf(stderr, pat1, bucket[i].low, "");
-        else
-            fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
+    if (bucket[i].low == bucket[i].high)
+      fprintf(stderr, pat1, bucket[i].low, "");
+    else
+      fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
 
-        for(j=0; j<HIST_BAR_MAX; j++)
-            fprintf(stderr, j<len?"=":" ");
-        fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
-    }
+    for (j = 0; j < HIST_BAR_MAX; j++)
+      fprintf(stderr, j < len ? "=" : " ");
+    fprintf(stderr, "\t%5d (%6.2f%%)\n", bucket[i].count, pct);
+  }
 }
 
 
-static void show_q_histogram(const int counts[64], int max_buckets)
-{
-    struct hist_bucket bucket[64];
-    int buckets = 0;
-    int total = 0;
-    int scale;
-    int i;
+static void show_q_histogram(const int counts[64], int max_buckets) {
+  struct hist_bucket bucket[64];
+  int buckets = 0;
+  int total = 0;
+  int scale;
+  int i;
 
 
-    for(i=0; i<64; i++)
-    {
-        if(counts[i])
-        {
-            bucket[buckets].low = bucket[buckets].high = i;
-            bucket[buckets].count = counts[i];
-            buckets++;
-            total += counts[i];
-        }
+  for (i = 0; i < 64; i++) {
+    if (counts[i]) {
+      bucket[buckets].low = bucket[buckets].high = i;
+      bucket[buckets].count = counts[i];
+      buckets++;
+      total += counts[i];
     }
+  }
 
-    fprintf(stderr, "\nQuantizer Selection:\n");
-    scale = merge_hist_buckets(bucket, &buckets, max_buckets);
-    show_histogram(bucket, buckets, total, scale);
+  fprintf(stderr, "\nQuantizer Selection:\n");
+  scale = merge_hist_buckets(bucket, &buckets, max_buckets);
+  show_histogram(bucket, buckets, total, scale);
 }
 
 
 #define RATE_BINS (100)
-struct rate_hist
-{
-    int64_t            *pts;
-    int                *sz;
-    int                 samples;
-    int                 frames;
-    struct hist_bucket  bucket[RATE_BINS];
-    int                 total;
+struct rate_hist {
+  int64_t            *pts;
+  int                *sz;
+  int                 samples;
+  int                 frames;
+  struct hist_bucket  bucket[RATE_BINS];
+  int                 total;
 };
 
 
 static void init_rate_histogram(struct rate_hist          *hist,
                                 const vpx_codec_enc_cfg_t *cfg,
-                                const vpx_rational_t      *fps)
-{
-    int i;
+                                const vpx_rational_t      *fps) {
+  int i;
 
-    /* Determine the number of samples in the buffer. Use the file's framerate
-     * to determine the number of frames in rc_buf_sz milliseconds, with an
-     * adjustment (5/4) to account for alt-refs
-     */
-    hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
+  /* Determine the number of samples in the buffer. Use the file's framerate
+   * to determine the number of frames in rc_buf_sz milliseconds, with an
+   * adjustment (5/4) to account for alt-refs
+   */
+  hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
 
-    /* prevent division by zero */
-    if (hist->samples == 0)
-      hist->samples=1;
+  /* prevent division by zero */
+  if (hist->samples == 0)
+    hist->samples = 1;
 
-    hist->pts = calloc(hist->samples, sizeof(*hist->pts));
-    hist->sz = calloc(hist->samples, sizeof(*hist->sz));
-    for(i=0; i<RATE_BINS; i++)
-    {
-        hist->bucket[i].low = INT_MAX;
-        hist->bucket[i].high = 0;
-        hist->bucket[i].count = 0;
-    }
+  hist->pts = calloc(hist->samples, sizeof(*hist->pts));
+  hist->sz = calloc(hist->samples, sizeof(*hist->sz));
+  for (i = 0; i < RATE_BINS; i++) {
+    hist->bucket[i].low = INT_MAX;
+    hist->bucket[i].high = 0;
+    hist->bucket[i].count = 0;
+  }
 }
 
 
-static void destroy_rate_histogram(struct rate_hist *hist)
-{
-    free(hist->pts);
-    free(hist->sz);
+static void destroy_rate_histogram(struct rate_hist *hist) {
+  free(hist->pts);
+  free(hist->sz);
 }
 
 
 static void update_rate_histogram(struct rate_hist          *hist,
                                   const vpx_codec_enc_cfg_t *cfg,
-                                  const vpx_codec_cx_pkt_t  *pkt)
-{
-    int i, idx;
-    int64_t now, then, sum_sz = 0, avg_bitrate;
+                                  const vpx_codec_cx_pkt_t  *pkt) {
+  int i, idx;
+  int64_t now, then, sum_sz = 0, avg_bitrate;
 
-    now = pkt->data.frame.pts * 1000
-          * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
+  now = pkt->data.frame.pts * 1000
+        * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
 
-    idx = hist->frames++ % hist->samples;
-    hist->pts[idx] = now;
-    hist->sz[idx] = (int)pkt->data.frame.sz;
+  idx = hist->frames++ % hist->samples;
+  hist->pts[idx] = now;
+  hist->sz[idx] = (int)pkt->data.frame.sz;
 
-    if(now < cfg->rc_buf_initial_sz)
-        return;
+  if (now < cfg->rc_buf_initial_sz)
+    return;
 
-    then = now;
+  then = now;
 
-    /* Sum the size over the past rc_buf_sz ms */
-    for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
-    {
-        int i_idx = (i-1) % hist->samples;
+  /* Sum the size over the past rc_buf_sz ms */
+  for (i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) {
+    int i_idx = (i - 1) % hist->samples;
 
-        then = hist->pts[i_idx];
-        if(now - then > cfg->rc_buf_sz)
-            break;
-        sum_sz += hist->sz[i_idx];
-    }
+    then = hist->pts[i_idx];
+    if (now - then > cfg->rc_buf_sz)
+      break;
+    sum_sz += hist->sz[i_idx];
+  }
 
-    if (now == then)
-        return;
+  if (now == then)
+    return;
 
-    avg_bitrate = sum_sz * 8 * 1000 / (now - then);
-    idx = (int)(avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000));
-    if(idx < 0)
-        idx = 0;
-    if(idx > RATE_BINS-1)
-        idx = RATE_BINS-1;
-    if(hist->bucket[idx].low > avg_bitrate)
-        hist->bucket[idx].low = (int)avg_bitrate;
-    if(hist->bucket[idx].high < avg_bitrate)
-        hist->bucket[idx].high = (int)avg_bitrate;
-    hist->bucket[idx].count++;
-    hist->total++;
+  avg_bitrate = sum_sz * 8 * 1000 / (now - then);
+  idx = (int)(avg_bitrate * (RATE_BINS / 2) / (cfg->rc_target_bitrate * 1000));
+  if (idx < 0)
+    idx = 0;
+  if (idx > RATE_BINS - 1)
+    idx = RATE_BINS - 1;
+  if (hist->bucket[idx].low > avg_bitrate)
+    hist->bucket[idx].low = (int)avg_bitrate;
+  if (hist->bucket[idx].high < avg_bitrate)
+    hist->bucket[idx].high = (int)avg_bitrate;
+  hist->bucket[idx].count++;
+  hist->total++;
 }
 
 
 static void show_rate_histogram(struct rate_hist          *hist,
                                 const vpx_codec_enc_cfg_t *cfg,
-                                int                        max_buckets)
-{
-    int i, scale;
-    int buckets = 0;
+                                int                        max_buckets) {
+  int i, scale;
+  int buckets = 0;
 
-    for(i = 0; i < RATE_BINS; i++)
-    {
-        if(hist->bucket[i].low == INT_MAX)
-            continue;
-        hist->bucket[buckets++] = hist->bucket[i];
-    }
+  for (i = 0; i < RATE_BINS; i++) {
+    if (hist->bucket[i].low == INT_MAX)
+      continue;
+    hist->bucket[buckets++] = hist->bucket[i];
+  }
 
-    fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
-    scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
-    show_histogram(hist->bucket, buckets, hist->total, scale);
+  fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
+  scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
+  show_histogram(hist->bucket, buckets, hist->total, scale);
 }
 
-#define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
-#define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
+#define mmin(a, b)  ((a) < (b) ? (a) : (b))
+static void find_mismatch(vpx_image_t *img1, vpx_image_t *img2,
+                          int yloc[4], int uloc[4], int vloc[4]) {
+  const unsigned int bsize = 64;
+  const unsigned int bsizey = bsize >> img1->y_chroma_shift;
+  const unsigned int bsizex = bsize >> img1->x_chroma_shift;
+  const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
+  const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
+  unsigned int match = 1;
+  unsigned int i, j;
+  yloc[0] = yloc[1] = yloc[2] = yloc[3] = -1;
+  for (i = 0, match = 1; match && i < img1->d_h; i += bsize) {
+    for (j = 0; match && j < img1->d_w; j += bsize) {
+      int k, l;
+      int si = mmin(i + bsize, img1->d_h) - i;
+      int sj = mmin(j + bsize, img1->d_w) - j;
+      for (k = 0; match && k < si; k++)
+        for (l = 0; match && l < sj; l++) {
+          if (*(img1->planes[VPX_PLANE_Y] +
+                (i + k) * img1->stride[VPX_PLANE_Y] + j + l) !=
+              *(img2->planes[VPX_PLANE_Y] +
+                (i + k) * img2->stride[VPX_PLANE_Y] + j + l)) {
+            yloc[0] = i + k;
+            yloc[1] = j + l;
+            yloc[2] = *(img1->planes[VPX_PLANE_Y] +
+                        (i + k) * img1->stride[VPX_PLANE_Y] + j + l);
+            yloc[3] = *(img2->planes[VPX_PLANE_Y] +
+                        (i + k) * img2->stride[VPX_PLANE_Y] + j + l);
+            match = 0;
+            break;
+          }
+        }
+    }
+  }
 
+  uloc[0] = uloc[1] = uloc[2] = uloc[3] = -1;
+  for (i = 0, match = 1; match && i < c_h; i += bsizey) {
+    for (j = 0; match && j < c_w; j += bsizex) {
+      int k, l;
+      int si = mmin(i + bsizey, c_h - i);
+      int sj = mmin(j + bsizex, c_w - j);
+      for (k = 0; match && k < si; k++)
+        for (l = 0; match && l < sj; l++) {
+          if (*(img1->planes[VPX_PLANE_U] +
+                (i + k) * img1->stride[VPX_PLANE_U] + j + l) !=
+              *(img2->planes[VPX_PLANE_U] +
+                (i + k) * img2->stride[VPX_PLANE_U] + j + l)) {
+            uloc[0] = i + k;
+            uloc[1] = j + l;
+            uloc[2] = *(img1->planes[VPX_PLANE_U] +
+                        (i + k) * img1->stride[VPX_PLANE_U] + j + l);
+            uloc[3] = *(img2->planes[VPX_PLANE_U] +
+                        (i + k) * img2->stride[VPX_PLANE_V] + j + l);
+            match = 0;
+            break;
+          }
+        }
+    }
+  }
+  vloc[0] = vloc[1] = vloc[2] = vloc[3] = -1;
+  for (i = 0, match = 1; match && i < c_h; i += bsizey) {
+    for (j = 0; match && j < c_w; j += bsizex) {
+      int k, l;
+      int si = mmin(i + bsizey, c_h - i);
+      int sj = mmin(j + bsizex, c_w - j);
+      for (k = 0; match && k < si; k++)
+        for (l = 0; match && l < sj; l++) {
+          if (*(img1->planes[VPX_PLANE_V] +
+                (i + k) * img1->stride[VPX_PLANE_V] + j + l) !=
+              *(img2->planes[VPX_PLANE_V] +
+                (i + k) * img2->stride[VPX_PLANE_V] + j + l)) {
+            vloc[0] = i + k;
+            vloc[1] = j + l;
+            vloc[2] = *(img1->planes[VPX_PLANE_V] +
+                        (i + k) * img1->stride[VPX_PLANE_V] + j + l);
+            vloc[3] = *(img2->planes[VPX_PLANE_V] +
+                        (i + k) * img2->stride[VPX_PLANE_V] + j + l);
+            match = 0;
+            break;
+          }
+        }
+    }
+  }
+}
+
+static int compare_img(vpx_image_t *img1, vpx_image_t *img2)
+{
+  const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
+  const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
+  int match = 1;
+  unsigned int i;
+
+  match &= (img1->fmt == img2->fmt);
+  match &= (img1->w == img2->w);
+  match &= (img1->h == img2->h);
+
+  for (i = 0; i < img1->d_h; i++)
+    match &= (memcmp(img1->planes[VPX_PLANE_Y]+i*img1->stride[VPX_PLANE_Y],
+                     img2->planes[VPX_PLANE_Y]+i*img2->stride[VPX_PLANE_Y],
+                     img1->d_w) == 0);
+
+  for (i = 0; i < c_h; i++)
+    match &= (memcmp(img1->planes[VPX_PLANE_U]+i*img1->stride[VPX_PLANE_U],
+                     img2->planes[VPX_PLANE_U]+i*img2->stride[VPX_PLANE_U],
+                     c_w) == 0);
+
+  for (i = 0; i < c_h; i++)
+    match &= (memcmp(img1->planes[VPX_PLANE_V]+i*img1->stride[VPX_PLANE_U],
+                     img2->planes[VPX_PLANE_V]+i*img2->stride[VPX_PLANE_U],
+                     c_w) == 0);
+
+  return match;
+}
+
+
+#define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
+#define MAX(x,y) ((x)>(y)?(x):(y))
+#if CONFIG_VP8_ENCODER && !CONFIG_VP9_ENCODER
+#define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
+#elif !CONFIG_VP8_ENCODER && CONFIG_VP9_ENCODER
+#define ARG_CTRL_CNT_MAX NELEMENTS(vp9_arg_ctrl_map)
+#else
+#define ARG_CTRL_CNT_MAX MAX(NELEMENTS(vp8_arg_ctrl_map), \
+                             NELEMENTS(vp9_arg_ctrl_map))
+#endif
 
 /* Configuration elements common to all streams */
-struct global_config
-{
-    const struct codec_item  *codec;
-    int                       passes;
-    int                       pass;
-    int                       usage;
-    int                       deadline;
-    int                       use_i420;
-    int                       quiet;
-    int                       verbose;
-    int                       limit;
-    int                       show_psnr;
-    int                       have_framerate;
-    struct vpx_rational       framerate;
-    int                       out_part;
-    int                       debug;
-    int                       show_q_hist_buckets;
-    int                       show_rate_hist_buckets;
+struct global_config {
+  const struct codec_item  *codec;
+  int                       passes;
+  int                       pass;
+  int                       usage;
+  int                       deadline;
+  int                       use_i420;
+  int                       quiet;
+  int                       verbose;
+  int                       limit;
+  int                       skip_frames;
+  int                       show_psnr;
+  enum TestDecodeFatality   test_decode;
+  int                       have_framerate;
+  struct vpx_rational       framerate;
+  int                       out_part;
+  int                       debug;
+  int                       show_q_hist_buckets;
+  int                       show_rate_hist_buckets;
 };
 
 
 /* Per-stream configuration */
-struct stream_config
-{
-    struct vpx_codec_enc_cfg  cfg;
-    const char               *out_fn;
-    const char               *stats_fn;
-    stereo_format_t           stereo_fmt;
-    int                       arg_ctrls[ARG_CTRL_CNT_MAX][2];
-    int                       arg_ctrl_cnt;
-    int                       write_webm;
-    int                       have_kf_max_dist;
+struct stream_config {
+  struct vpx_codec_enc_cfg  cfg;
+  const char               *out_fn;
+  const char               *stats_fn;
+  stereo_format_t           stereo_fmt;
+  int                       arg_ctrls[ARG_CTRL_CNT_MAX][2];
+  int                       arg_ctrl_cnt;
+  int                       write_webm;
+  int                       have_kf_max_dist;
 };
 
 
-struct stream_state
-{
-    int                       index;
-    struct stream_state      *next;
-    struct stream_config      config;
-    FILE                     *file;
-    struct rate_hist          rate_hist;
-    EbmlGlobal                ebml;
-    uint32_t                  hash;
-    uint64_t                  psnr_sse_total;
-    uint64_t                  psnr_samples_total;
-    double                    psnr_totals[4];
-    int                       psnr_count;
-    int                       counts[64];
-    vpx_codec_ctx_t           encoder;
-    unsigned int              frames_out;
-    uint64_t                  cx_time;
-    size_t                    nbytes;
-    stats_io_t                stats;
+struct stream_state {
+  int                       index;
+  struct stream_state      *next;
+  struct stream_config      config;
+  FILE                     *file;
+  struct rate_hist          rate_hist;
+  EbmlGlobal                ebml;
+  uint32_t                  hash;
+  uint64_t                  psnr_sse_total;
+  uint64_t                  psnr_samples_total;
+  double                    psnr_totals[4];
+  int                       psnr_count;
+  int                       counts[64];
+  vpx_codec_ctx_t           encoder;
+  unsigned int              frames_out;
+  uint64_t                  cx_time;
+  size_t                    nbytes;
+  stats_io_t                stats;
+  struct vpx_image         *img;
+  vpx_codec_ctx_t           decoder;
+  int                       mismatch_seen;
 };
 
 
 void validate_positive_rational(const char          *msg,
-                                struct vpx_rational *rat)
-{
-    if (rat->den < 0)
-    {
-        rat->num *= -1;
-        rat->den *= -1;
-    }
+                                struct vpx_rational *rat) {
+  if (rat->den < 0) {
+    rat->num *= -1;
+    rat->den *= -1;
+  }
 
-    if (rat->num < 0)
-        die("Error: %s must be positive\n", msg);
+  if (rat->num < 0)
+    die("Error: %s must be positive\n", msg);
 
-    if (!rat->den)
-        die("Error: %s has zero denominator\n", msg);
+  if (!rat->den)
+    die("Error: %s has zero denominator\n", msg);
 }
 
 
-static void parse_global_config(struct global_config *global, char **argv)
-{
-    char       **argi, **argj;
-    struct arg   arg;
+static void parse_global_config(struct global_config *global, char **argv) {
+  char       **argi, **argj;
+  struct arg   arg;
 
-    /* Initialize default parameters */
-    memset(global, 0, sizeof(*global));
-    global->codec = codecs;
-    global->passes = 1;
-    global->use_i420 = 1;
+  /* Initialize default parameters */
+  memset(global, 0, sizeof(*global));
+  global->codec = codecs;
+  global->passes = 1;
+  global->use_i420 = 1;
 
-    for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
-    {
-        arg.argv_step = 1;
+  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
+    arg.argv_step = 1;
 
-        if (arg_match(&arg, &codecarg, argi))
-        {
-            int j, k = -1;
+    if (arg_match(&arg, &codecarg, argi)) {
+      int j, k = -1;
 
-            for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
-                if (!strcmp(codecs[j].name, arg.val))
-                    k = j;
+      for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
+        if (!strcmp(codecs[j].name, arg.val))
+          k = j;
 
-            if (k >= 0)
-                global->codec = codecs + k;
-            else
-                die("Error: Unrecognized argument (%s) to --codec\n",
-                    arg.val);
+      if (k >= 0)
+        global->codec = codecs + k;
+      else
+        die("Error: Unrecognized argument (%s) to --codec\n",
+            arg.val);
 
-        }
-        else if (arg_match(&arg, &passes, argi))
-        {
-            global->passes = arg_parse_uint(&arg);
+    } else if (arg_match(&arg, &passes, argi)) {
+      global->passes = arg_parse_uint(&arg);
 
-            if (global->passes < 1 || global->passes > 2)
-                die("Error: Invalid number of passes (%d)\n", global->passes);
-        }
-        else if (arg_match(&arg, &pass_arg, argi))
-        {
-            global->pass = arg_parse_uint(&arg);
+      if (global->passes < 1 || global->passes > 2)
+        die("Error: Invalid number of passes (%d)\n", global->passes);
+    } else if (arg_match(&arg, &pass_arg, argi)) {
+      global->pass = arg_parse_uint(&arg);
 
-            if (global->pass < 1 || global->pass > 2)
-                die("Error: Invalid pass selected (%d)\n",
-                    global->pass);
-        }
-        else if (arg_match(&arg, &usage, argi))
-            global->usage = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &deadline, argi))
-            global->deadline = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &best_dl, argi))
-            global->deadline = VPX_DL_BEST_QUALITY;
-        else if (arg_match(&arg, &good_dl, argi))
-            global->deadline = VPX_DL_GOOD_QUALITY;
-        else if (arg_match(&arg, &rt_dl, argi))
-            global->deadline = VPX_DL_REALTIME;
-        else if (arg_match(&arg, &use_yv12, argi))
-            global->use_i420 = 0;
-        else if (arg_match(&arg, &use_i420, argi))
-            global->use_i420 = 1;
-        else if (arg_match(&arg, &quietarg, argi))
-            global->quiet = 1;
-        else if (arg_match(&arg, &verbosearg, argi))
-            global->verbose = 1;
-        else if (arg_match(&arg, &limit, argi))
-            global->limit = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &psnrarg, argi))
-            global->show_psnr = 1;
-        else if (arg_match(&arg, &framerate, argi))
-        {
-            global->framerate = arg_parse_rational(&arg);
-            validate_positive_rational(arg.name, &global->framerate);
-            global->have_framerate = 1;
-        }
-        else if (arg_match(&arg,&out_part, argi))
-            global->out_part = 1;
-        else if (arg_match(&arg, &debugmode, argi))
-            global->debug = 1;
-        else if (arg_match(&arg, &q_hist_n, argi))
-            global->show_q_hist_buckets = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &rate_hist_n, argi))
-            global->show_rate_hist_buckets = arg_parse_uint(&arg);
-        else
-            argj++;
-    }
-
-    /* Validate global config */
-
-    if (global->pass)
-    {
-        /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
-        if (global->pass > global->passes)
-        {
-            warn("Assuming --pass=%d implies --passes=%d\n",
-                 global->pass, global->pass);
-            global->passes = global->pass;
-        }
-    }
-}
-
-
-void open_input_file(struct input_state *input)
-{
-    unsigned int fourcc;
-
-    /* Parse certain options from the input file, if possible */
-    input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb")
-                                         : set_binary_mode(stdin);
-
-    if (!input->file)
-        fatal("Failed to open input file");
-
-    /* For RAW input sources, these bytes will applied on the first frame
-     *  in read_frame().
-     */
-    input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
-    input->detect.position = 0;
-
-    if (input->detect.buf_read == 4
-        && file_is_y4m(input->file, &input->y4m, input->detect.buf))
-    {
-        if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4) >= 0)
-        {
-            input->file_type = FILE_TYPE_Y4M;
-            input->w = input->y4m.pic_w;
-            input->h = input->y4m.pic_h;
-            input->framerate.num = input->y4m.fps_n;
-            input->framerate.den = input->y4m.fps_d;
-            input->use_i420 = 0;
-        }
-        else
-            fatal("Unsupported Y4M stream.");
-    }
-    else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc))
-    {
-        input->file_type = FILE_TYPE_IVF;
-        switch (fourcc)
-        {
-        case 0x32315659:
-            input->use_i420 = 0;
-            break;
-        case 0x30323449:
-            input->use_i420 = 1;
-            break;
-        default:
-            fatal("Unsupported fourcc (%08x) in IVF", fourcc);
-        }
-    }
+      if (global->pass < 1 || global->pass > 2)
+        die("Error: Invalid pass selected (%d)\n",
+            global->pass);
+    } else if (arg_match(&arg, &usage, argi))
+      global->usage = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &deadline, argi))
+      global->deadline = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &best_dl, argi))
+      global->deadline = VPX_DL_BEST_QUALITY;
+    else if (arg_match(&arg, &good_dl, argi))
+      global->deadline = VPX_DL_GOOD_QUALITY;
+    else if (arg_match(&arg, &rt_dl, argi))
+      global->deadline = VPX_DL_REALTIME;
+    else if (arg_match(&arg, &use_yv12, argi))
+      global->use_i420 = 0;
+    else if (arg_match(&arg, &use_i420, argi))
+      global->use_i420 = 1;
+    else if (arg_match(&arg, &quietarg, argi))
+      global->quiet = 1;
+    else if (arg_match(&arg, &verbosearg, argi))
+      global->verbose = 1;
+    else if (arg_match(&arg, &limit, argi))
+      global->limit = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &skip, argi))
+      global->skip_frames = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &psnrarg, argi))
+      global->show_psnr = 1;
+    else if (arg_match(&arg, &recontest, argi))
+      global->test_decode = arg_parse_enum_or_int(&arg);
+    else if (arg_match(&arg, &framerate, argi)) {
+      global->framerate = arg_parse_rational(&arg);
+      validate_positive_rational(arg.name, &global->framerate);
+      global->have_framerate = 1;
+    } else if (arg_match(&arg, &out_part, argi))
+      global->out_part = 1;
+    else if (arg_match(&arg, &debugmode, argi))
+      global->debug = 1;
+    else if (arg_match(&arg, &q_hist_n, argi))
+      global->show_q_hist_buckets = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &rate_hist_n, argi))
+      global->show_rate_hist_buckets = arg_parse_uint(&arg);
     else
-    {
-        input->file_type = FILE_TYPE_RAW;
+      argj++;
+  }
+
+  /* Validate global config */
+
+  if (global->pass) {
+    /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
+    if (global->pass > global->passes) {
+      warn("Assuming --pass=%d implies --passes=%d\n",
+           global->pass, global->pass);
+      global->passes = global->pass;
     }
+  }
 }
 
 
-static void close_input_file(struct input_state *input)
-{
-    fclose(input->file);
-    if (input->file_type == FILE_TYPE_Y4M)
-        y4m_input_close(&input->y4m);
+void open_input_file(struct input_state *input) {
+  unsigned int fourcc;
+
+  /* Parse certain options from the input file, if possible */
+  input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb")
+                : set_binary_mode(stdin);
+
+  if (!input->file)
+    fatal("Failed to open input file");
+
+  if (!fseeko(input->file, 0, SEEK_END)) {
+    /* Input file is seekable. Figure out how long it is, so we can get
+     * progress info.
+     */
+    input->length = ftello(input->file);
+    rewind(input->file);
+  }
+
+  /* For RAW input sources, these bytes will applied on the first frame
+   *  in read_frame().
+   */
+  input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
+  input->detect.position = 0;
+
+  if (input->detect.buf_read == 4
+      && file_is_y4m(input->file, &input->y4m, input->detect.buf)) {
+    if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
+                       input->only_i420) >= 0) {
+      input->file_type = FILE_TYPE_Y4M;
+      input->w = input->y4m.pic_w;
+      input->h = input->y4m.pic_h;
+      input->framerate.num = input->y4m.fps_n;
+      input->framerate.den = input->y4m.fps_d;
+      input->use_i420 = 0;
+    } else
+      fatal("Unsupported Y4M stream.");
+  } else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc)) {
+    input->file_type = FILE_TYPE_IVF;
+    switch (fourcc) {
+      case 0x32315659:
+        input->use_i420 = 0;
+        break;
+      case 0x30323449:
+        input->use_i420 = 1;
+        break;
+      default:
+        fatal("Unsupported fourcc (%08x) in IVF", fourcc);
+    }
+  } else {
+    input->file_type = FILE_TYPE_RAW;
+  }
+}
+
+
+static void close_input_file(struct input_state *input) {
+  fclose(input->file);
+  if (input->file_type == FILE_TYPE_Y4M)
+    y4m_input_close(&input->y4m);
 }
 
 static struct stream_state *new_stream(struct global_config *global,
-                                       struct stream_state  *prev)
-{
-    struct stream_state *stream;
+                                       struct stream_state  *prev) {
+  struct stream_state *stream;
 
-    stream = calloc(1, sizeof(*stream));
-    if(!stream)
-        fatal("Failed to allocate new stream.");
-    if(prev)
-    {
-        memcpy(stream, prev, sizeof(*stream));
-        stream->index++;
-        prev->next = stream;
-    }
-    else
-    {
-        vpx_codec_err_t  res;
+  stream = calloc(1, sizeof(*stream));
+  if (!stream)
+    fatal("Failed to allocate new stream.");
+  if (prev) {
+    memcpy(stream, prev, sizeof(*stream));
+    stream->index++;
+    prev->next = stream;
+  } else {
+    vpx_codec_err_t  res;
 
-        /* Populate encoder configuration */
-        res = vpx_codec_enc_config_default(global->codec->iface,
-                                           &stream->config.cfg,
-                                           global->usage);
-        if (res)
-            fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
+    /* Populate encoder configuration */
+    res = vpx_codec_enc_config_default(global->codec->iface(),
+                                       &stream->config.cfg,
+                                       global->usage);
+    if (res)
+      fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
 
-        /* Change the default timebase to a high enough value so that the
-         * encoder will always create strictly increasing timestamps.
-         */
-        stream->config.cfg.g_timebase.den = 1000;
+    /* Change the default timebase to a high enough value so that the
+     * encoder will always create strictly increasing timestamps.
+     */
+    stream->config.cfg.g_timebase.den = 1000;
 
-        /* Never use the library's default resolution, require it be parsed
-         * from the file or set on the command line.
-         */
-        stream->config.cfg.g_w = 0;
-        stream->config.cfg.g_h = 0;
+    /* Never use the library's default resolution, require it be parsed
+     * from the file or set on the command line.
+     */
+    stream->config.cfg.g_w = 0;
+    stream->config.cfg.g_h = 0;
 
-        /* Initialize remaining stream parameters */
-        stream->config.stereo_fmt = STEREO_FORMAT_MONO;
-        stream->config.write_webm = 1;
-        stream->ebml.last_pts_ms = -1;
+    /* Initialize remaining stream parameters */
+    stream->config.stereo_fmt = STEREO_FORMAT_MONO;
+    stream->config.write_webm = 1;
+    stream->ebml.last_pts_ms = -1;
 
-        /* Allows removal of the application version from the EBML tags */
-        stream->ebml.debug = global->debug;
-    }
+    /* Allows removal of the application version from the EBML tags */
+    stream->ebml.debug = global->debug;
+  }
 
-    /* Output files must be specified for each stream */
-    stream->config.out_fn = NULL;
+  /* Output files must be specified for each stream */
+  stream->config.out_fn = NULL;
 
-    stream->next = NULL;
-    return stream;
+  stream->next = NULL;
+  return stream;
 }
 
 
 static int parse_stream_params(struct global_config *global,
                                struct stream_state  *stream,
-                               char **argv)
-{
-    char                   **argi, **argj;
-    struct arg               arg;
-    static const arg_def_t **ctrl_args = no_args;
-    static const int        *ctrl_args_map = NULL;
-    struct stream_config    *config = &stream->config;
-    int                      eos_mark_found = 0;
+                               char **argv) {
+  char                   **argi, **argj;
+  struct arg               arg;
+  static const arg_def_t **ctrl_args = no_args;
+  static const int        *ctrl_args_map = NULL;
+  struct stream_config    *config = &stream->config;
+  int                      eos_mark_found = 0;
 
-    /* Handle codec specific options */
-    if (global->codec->iface == &vpx_codec_vp8_cx_algo)
-    {
-        ctrl_args = vp8_args;
-        ctrl_args_map = vp8_arg_ctrl_map;
+  /* Handle codec specific options */
+  if (0) {
+#if CONFIG_VP8_ENCODER
+  } else if (global->codec->iface == vpx_codec_vp8_cx) {
+    ctrl_args = vp8_args;
+    ctrl_args_map = vp8_arg_ctrl_map;
+#endif
+#if CONFIG_VP9_ENCODER
+  } else if (global->codec->iface == vpx_codec_vp9_cx) {
+    ctrl_args = vp9_args;
+    ctrl_args_map = vp9_arg_ctrl_map;
+#endif
+  }
+
+  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
+    arg.argv_step = 1;
+
+    /* Once we've found an end-of-stream marker (--) we want to continue
+     * shifting arguments but not consuming them.
+     */
+    if (eos_mark_found) {
+      argj++;
+      continue;
+    } else if (!strcmp(*argj, "--")) {
+      eos_mark_found = 1;
+      continue;
     }
 
-    for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
-    {
-        arg.argv_step = 1;
+    if (0);
+    else if (arg_match(&arg, &outputfile, argi))
+      config->out_fn = arg.val;
+    else if (arg_match(&arg, &fpf_name, argi))
+      config->stats_fn = arg.val;
+    else if (arg_match(&arg, &use_ivf, argi))
+      config->write_webm = 0;
+    else if (arg_match(&arg, &threads, argi))
+      config->cfg.g_threads = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &profile, argi))
+      config->cfg.g_profile = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &width, argi))
+      config->cfg.g_w = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &height, argi))
+      config->cfg.g_h = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &stereo_mode, argi))
+      config->stereo_fmt = arg_parse_enum_or_int(&arg);
+    else if (arg_match(&arg, &timebase, argi)) {
+      config->cfg.g_timebase = arg_parse_rational(&arg);
+      validate_positive_rational(arg.name, &config->cfg.g_timebase);
+    } else if (arg_match(&arg, &error_resilient, argi))
+      config->cfg.g_error_resilient = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &lag_in_frames, argi))
+      config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &dropframe_thresh, argi))
+      config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &resize_allowed, argi))
+      config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &resize_up_thresh, argi))
+      config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &resize_down_thresh, argi))
+      config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &end_usage, argi))
+      config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
+    else if (arg_match(&arg, &target_bitrate, argi))
+      config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &min_quantizer, argi))
+      config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &max_quantizer, argi))
+      config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &undershoot_pct, argi))
+      config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &overshoot_pct, argi))
+      config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &buf_sz, argi))
+      config->cfg.rc_buf_sz = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &buf_initial_sz, argi))
+      config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &buf_optimal_sz, argi))
+      config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &bias_pct, argi)) {
+      config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
 
-        /* Once we've found an end-of-stream marker (--) we want to continue
-         * shifting arguments but not consuming them.
-         */
-        if (eos_mark_found)
-        {
-            argj++;
-            continue;
+      if (global->passes < 2)
+        warn("option %s ignored in one-pass mode.\n", arg.name);
+    } else if (arg_match(&arg, &minsection_pct, argi)) {
+      config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
+
+      if (global->passes < 2)
+        warn("option %s ignored in one-pass mode.\n", arg.name);
+    } else if (arg_match(&arg, &maxsection_pct, argi)) {
+      config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
+
+      if (global->passes < 2)
+        warn("option %s ignored in one-pass mode.\n", arg.name);
+    } else if (arg_match(&arg, &kf_min_dist, argi))
+      config->cfg.kf_min_dist = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &kf_max_dist, argi)) {
+      config->cfg.kf_max_dist = arg_parse_uint(&arg);
+      config->have_kf_max_dist = 1;
+    } else if (arg_match(&arg, &kf_disabled, argi))
+      config->cfg.kf_mode = VPX_KF_DISABLED;
+    else {
+      int i, match = 0;
+
+      for (i = 0; ctrl_args[i]; i++) {
+        if (arg_match(&arg, ctrl_args[i], argi)) {
+          int j;
+          match = 1;
+
+          /* Point either to the next free element or the first
+          * instance of this control.
+          */
+          for (j = 0; j < config->arg_ctrl_cnt; j++)
+            if (config->arg_ctrls[j][0] == ctrl_args_map[i])
+              break;
+
+          /* Update/insert */
+          assert(j < ARG_CTRL_CNT_MAX);
+          if (j < ARG_CTRL_CNT_MAX) {
+            config->arg_ctrls[j][0] = ctrl_args_map[i];
+            config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
+            if (j == config->arg_ctrl_cnt)
+              config->arg_ctrl_cnt++;
+          }
+
         }
-        else if (!strcmp(*argj, "--"))
-        {
-            eos_mark_found = 1;
-            continue;
-        }
+      }
 
-        if (0);
-        else if (arg_match(&arg, &outputfile, argi))
-            config->out_fn = arg.val;
-        else if (arg_match(&arg, &fpf_name, argi))
-            config->stats_fn = arg.val;
-        else if (arg_match(&arg, &use_ivf, argi))
-            config->write_webm = 0;
-        else if (arg_match(&arg, &threads, argi))
-            config->cfg.g_threads = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &profile, argi))
-            config->cfg.g_profile = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &width, argi))
-            config->cfg.g_w = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &height, argi))
-            config->cfg.g_h = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &stereo_mode, argi))
-            config->stereo_fmt = arg_parse_enum_or_int(&arg);
-        else if (arg_match(&arg, &timebase, argi))
-        {
-            config->cfg.g_timebase = arg_parse_rational(&arg);
-            validate_positive_rational(arg.name, &config->cfg.g_timebase);
-        }
-        else if (arg_match(&arg, &error_resilient, argi))
-            config->cfg.g_error_resilient = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &lag_in_frames, argi))
-            config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &dropframe_thresh, argi))
-            config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &resize_allowed, argi))
-            config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &resize_up_thresh, argi))
-            config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &resize_down_thresh, argi))
-            config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &end_usage, argi))
-            config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
-        else if (arg_match(&arg, &target_bitrate, argi))
-            config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &min_quantizer, argi))
-            config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &max_quantizer, argi))
-            config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &undershoot_pct, argi))
-            config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &overshoot_pct, argi))
-            config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &buf_sz, argi))
-            config->cfg.rc_buf_sz = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &buf_initial_sz, argi))
-            config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &buf_optimal_sz, argi))
-            config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &bias_pct, argi))
-        {
-            config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
-
-            if (global->passes < 2)
-                warn("option %s ignored in one-pass mode.\n", arg.name);
-        }
-        else if (arg_match(&arg, &minsection_pct, argi))
-        {
-            config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
-
-            if (global->passes < 2)
-                warn("option %s ignored in one-pass mode.\n", arg.name);
-        }
-        else if (arg_match(&arg, &maxsection_pct, argi))
-        {
-            config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
-
-            if (global->passes < 2)
-                warn("option %s ignored in one-pass mode.\n", arg.name);
-        }
-        else if (arg_match(&arg, &kf_min_dist, argi))
-            config->cfg.kf_min_dist = arg_parse_uint(&arg);
-        else if (arg_match(&arg, &kf_max_dist, argi))
-        {
-            config->cfg.kf_max_dist = arg_parse_uint(&arg);
-            config->have_kf_max_dist = 1;
-        }
-        else if (arg_match(&arg, &kf_disabled, argi))
-            config->cfg.kf_mode = VPX_KF_DISABLED;
-        else
-        {
-            int i, match = 0;
-
-            for (i = 0; ctrl_args[i]; i++)
-            {
-                if (arg_match(&arg, ctrl_args[i], argi))
-                {
-                    int j;
-                    match = 1;
-
-                    /* Point either to the next free element or the first
-                    * instance of this control.
-                    */
-                    for(j=0; j<config->arg_ctrl_cnt; j++)
-                        if(config->arg_ctrls[j][0] == ctrl_args_map[i])
-                            break;
-
-                    /* Update/insert */
-                    assert(j < ARG_CTRL_CNT_MAX);
-                    if (j < ARG_CTRL_CNT_MAX)
-                    {
-                        config->arg_ctrls[j][0] = ctrl_args_map[i];
-                        config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
-                        if(j == config->arg_ctrl_cnt)
-                            config->arg_ctrl_cnt++;
-                    }
-
-                }
-            }
-
-            if (!match)
-                argj++;
-        }
+      if (!match)
+        argj++;
     }
+  }
 
-    return eos_mark_found;
+  return eos_mark_found;
 }
 
 
 #define FOREACH_STREAM(func)\
-do\
-{\
+  do\
+  {\
     struct stream_state  *stream;\
-\
+    \
     for(stream = streams; stream; stream = stream->next)\
-        func;\
-}while(0)
+      func;\
+  }while(0)
 
 
-static void validate_stream_config(struct stream_state *stream)
-{
-    struct stream_state *streami;
+static void validate_stream_config(struct stream_state *stream) {
+  struct stream_state *streami;
 
-    if(!stream->config.cfg.g_w || !stream->config.cfg.g_h)
-        fatal("Stream %d: Specify stream dimensions with --width (-w) "
-              " and --height (-h)", stream->index);
+  if (!stream->config.cfg.g_w || !stream->config.cfg.g_h)
+    fatal("Stream %d: Specify stream dimensions with --width (-w) "
+          " and --height (-h)", stream->index);
 
-    for(streami = stream; streami; streami = streami->next)
-    {
-        /* All streams require output files */
-        if(!streami->config.out_fn)
-            fatal("Stream %d: Output file is required (specify with -o)",
-                  streami->index);
+  for (streami = stream; streami; streami = streami->next) {
+    /* All streams require output files */
+    if (!streami->config.out_fn)
+      fatal("Stream %d: Output file is required (specify with -o)",
+            streami->index);
 
-        /* Check for two streams outputting to the same file */
-        if(streami != stream)
-        {
-            const char *a = stream->config.out_fn;
-            const char *b = streami->config.out_fn;
-            if(!strcmp(a,b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
-                fatal("Stream %d: duplicate output file (from stream %d)",
-                      streami->index, stream->index);
-        }
-
-        /* Check for two streams sharing a stats file. */
-        if(streami != stream)
-        {
-            const char *a = stream->config.stats_fn;
-            const char *b = streami->config.stats_fn;
-            if(a && b && !strcmp(a,b))
-                fatal("Stream %d: duplicate stats file (from stream %d)",
-                      streami->index, stream->index);
-        }
+    /* Check for two streams outputting to the same file */
+    if (streami != stream) {
+      const char *a = stream->config.out_fn;
+      const char *b = streami->config.out_fn;
+      if (!strcmp(a, b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
+        fatal("Stream %d: duplicate output file (from stream %d)",
+              streami->index, stream->index);
     }
+
+    /* Check for two streams sharing a stats file. */
+    if (streami != stream) {
+      const char *a = stream->config.stats_fn;
+      const char *b = streami->config.stats_fn;
+      if (a && b && !strcmp(a, b))
+        fatal("Stream %d: duplicate stats file (from stream %d)",
+              streami->index, stream->index);
+    }
+  }
 }
 
 
 static void set_stream_dimensions(struct stream_state *stream,
                                   unsigned int w,
-                                  unsigned int h)
-{
-    if ((stream->config.cfg.g_w && stream->config.cfg.g_w != w)
-        ||(stream->config.cfg.g_h && stream->config.cfg.g_h != h))
-        fatal("Stream %d: Resizing not yet supported", stream->index);
-    stream->config.cfg.g_w = w;
-    stream->config.cfg.g_h = h;
+                                  unsigned int h) {
+  if (!stream->config.cfg.g_w) {
+    if (!stream->config.cfg.g_h)
+      stream->config.cfg.g_w = w;
+    else
+      stream->config.cfg.g_w = w * stream->config.cfg.g_h / h;
+  }
+  if (!stream->config.cfg.g_h) {
+    stream->config.cfg.g_h = h * stream->config.cfg.g_w / w;
+  }
 }
 
 
 static void set_default_kf_interval(struct stream_state  *stream,
-                                    struct global_config *global)
-{
-    /* Use a max keyframe interval of 5 seconds, if none was
-     * specified on the command line.
-     */
-    if (!stream->config.have_kf_max_dist)
-    {
-        double framerate = (double)global->framerate.num/global->framerate.den;
-        if (framerate > 0.0)
-            stream->config.cfg.kf_max_dist = (unsigned int)(5.0*framerate);
-    }
+                                    struct global_config *global) {
+  /* Use a max keyframe interval of 5 seconds, if none was
+   * specified on the command line.
+   */
+  if (!stream->config.have_kf_max_dist) {
+    double framerate = (double)global->framerate.num / global->framerate.den;
+    if (framerate > 0.0)
+      stream->config.cfg.kf_max_dist = (unsigned int)(5.0 * framerate);
+  }
 }
 
 
 static void show_stream_config(struct stream_state  *stream,
                                struct global_config *global,
-                               struct input_state   *input)
-{
+                               struct input_state   *input) {
 
 #define SHOW(field) \
-    fprintf(stderr, "    %-28s = %d\n", #field, stream->config.cfg.field)
+  fprintf(stderr, "    %-28s = %d\n", #field, stream->config.cfg.field)
 
-    if(stream->index == 0)
-    {
-        fprintf(stderr, "Codec: %s\n",
-                vpx_codec_iface_name(global->codec->iface));
-        fprintf(stderr, "Source file: %s Format: %s\n", input->fn,
-                input->use_i420 ? "I420" : "YV12");
-    }
-    if(stream->next || stream->index)
-        fprintf(stderr, "\nStream Index: %d\n", stream->index);
-    fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
-    fprintf(stderr, "Encoder parameters:\n");
+  if (stream->index == 0) {
+    fprintf(stderr, "Codec: %s\n",
+            vpx_codec_iface_name(global->codec->iface()));
+    fprintf(stderr, "Source file: %s Format: %s\n", input->fn,
+            input->use_i420 ? "I420" : "YV12");
+  }
+  if (stream->next || stream->index)
+    fprintf(stderr, "\nStream Index: %d\n", stream->index);
+  fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
+  fprintf(stderr, "Encoder parameters:\n");
 
-    SHOW(g_usage);
-    SHOW(g_threads);
-    SHOW(g_profile);
-    SHOW(g_w);
-    SHOW(g_h);
-    SHOW(g_timebase.num);
-    SHOW(g_timebase.den);
-    SHOW(g_error_resilient);
-    SHOW(g_pass);
-    SHOW(g_lag_in_frames);
-    SHOW(rc_dropframe_thresh);
-    SHOW(rc_resize_allowed);
-    SHOW(rc_resize_up_thresh);
-    SHOW(rc_resize_down_thresh);
-    SHOW(rc_end_usage);
-    SHOW(rc_target_bitrate);
-    SHOW(rc_min_quantizer);
-    SHOW(rc_max_quantizer);
-    SHOW(rc_undershoot_pct);
-    SHOW(rc_overshoot_pct);
-    SHOW(rc_buf_sz);
-    SHOW(rc_buf_initial_sz);
-    SHOW(rc_buf_optimal_sz);
-    SHOW(rc_2pass_vbr_bias_pct);
-    SHOW(rc_2pass_vbr_minsection_pct);
-    SHOW(rc_2pass_vbr_maxsection_pct);
-    SHOW(kf_mode);
-    SHOW(kf_min_dist);
-    SHOW(kf_max_dist);
+  SHOW(g_usage);
+  SHOW(g_threads);
+  SHOW(g_profile);
+  SHOW(g_w);
+  SHOW(g_h);
+  SHOW(g_timebase.num);
+  SHOW(g_timebase.den);
+  SHOW(g_error_resilient);
+  SHOW(g_pass);
+  SHOW(g_lag_in_frames);
+  SHOW(rc_dropframe_thresh);
+  SHOW(rc_resize_allowed);
+  SHOW(rc_resize_up_thresh);
+  SHOW(rc_resize_down_thresh);
+  SHOW(rc_end_usage);
+  SHOW(rc_target_bitrate);
+  SHOW(rc_min_quantizer);
+  SHOW(rc_max_quantizer);
+  SHOW(rc_undershoot_pct);
+  SHOW(rc_overshoot_pct);
+  SHOW(rc_buf_sz);
+  SHOW(rc_buf_initial_sz);
+  SHOW(rc_buf_optimal_sz);
+  SHOW(rc_2pass_vbr_bias_pct);
+  SHOW(rc_2pass_vbr_minsection_pct);
+  SHOW(rc_2pass_vbr_maxsection_pct);
+  SHOW(kf_mode);
+  SHOW(kf_min_dist);
+  SHOW(kf_max_dist);
 }
 
 
 static void open_output_file(struct stream_state *stream,
-                             struct global_config *global)
-{
-    const char *fn = stream->config.out_fn;
+                             struct global_config *global) {
+  const char *fn = stream->config.out_fn;
 
-    stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
+  stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
 
-    if (!stream->file)
-        fatal("Failed to open output file");
+  if (!stream->file)
+    fatal("Failed to open output file");
 
-    if(stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
-        fatal("WebM output to pipes not supported.");
+  if (stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
+    fatal("WebM output to pipes not supported.");
 
-    if(stream->config.write_webm)
-    {
-        stream->ebml.stream = stream->file;
-        write_webm_file_header(&stream->ebml, &stream->config.cfg,
-                               &global->framerate,
-                               stream->config.stereo_fmt);
-    }
-    else
-        write_ivf_file_header(stream->file, &stream->config.cfg,
-                              global->codec->fourcc, 0);
+  if (stream->config.write_webm) {
+    stream->ebml.stream = stream->file;
+    write_webm_file_header(&stream->ebml, &stream->config.cfg,
+                           &global->framerate,
+                           stream->config.stereo_fmt,
+                           global->codec->fourcc);
+  } else
+    write_ivf_file_header(stream->file, &stream->config.cfg,
+                          global->codec->fourcc, 0);
 }
 
 
 static void close_output_file(struct stream_state *stream,
-                              unsigned int         fourcc)
-{
-    if(stream->config.write_webm)
-    {
-        write_webm_file_footer(&stream->ebml, stream->hash);
-        free(stream->ebml.cue_list);
-        stream->ebml.cue_list = NULL;
-    }
-    else
-    {
-        if (!fseek(stream->file, 0, SEEK_SET))
-            write_ivf_file_header(stream->file, &stream->config.cfg,
-                                  fourcc,
-                                  stream->frames_out);
-    }
+                              unsigned int         fourcc) {
+  if (stream->config.write_webm) {
+    write_webm_file_footer(&stream->ebml, stream->hash);
+    free(stream->ebml.cue_list);
+    stream->ebml.cue_list = NULL;
+  } else {
+    if (!fseek(stream->file, 0, SEEK_SET))
+      write_ivf_file_header(stream->file, &stream->config.cfg,
+                            fourcc,
+                            stream->frames_out);
+  }
 
-    fclose(stream->file);
+  fclose(stream->file);
 }
 
 
 static void setup_pass(struct stream_state  *stream,
                        struct global_config *global,
-                       int                   pass)
-{
-    if (stream->config.stats_fn)
-    {
-        if (!stats_open_file(&stream->stats, stream->config.stats_fn,
-                             pass))
-            fatal("Failed to open statistics store");
-    }
-    else
-    {
-        if (!stats_open_mem(&stream->stats, pass))
-            fatal("Failed to open statistics store");
-    }
+                       int                   pass) {
+  if (stream->config.stats_fn) {
+    if (!stats_open_file(&stream->stats, stream->config.stats_fn,
+                         pass))
+      fatal("Failed to open statistics store");
+  } else {
+    if (!stats_open_mem(&stream->stats, pass))
+      fatal("Failed to open statistics store");
+  }
 
-    stream->config.cfg.g_pass = global->passes == 2
-        ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
-        : VPX_RC_ONE_PASS;
-    if (pass)
-        stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
+  stream->config.cfg.g_pass = global->passes == 2
+                              ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
+                            : VPX_RC_ONE_PASS;
+  if (pass)
+    stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
 
-    stream->cx_time = 0;
-    stream->nbytes = 0;
-    stream->frames_out = 0;
+  stream->cx_time = 0;
+  stream->nbytes = 0;
+  stream->frames_out = 0;
 }
 
 
 static void initialize_encoder(struct stream_state  *stream,
-                               struct global_config *global)
-{
-    int i;
-    int flags = 0;
+                               struct global_config *global) {
+  int i;
+  int flags = 0;
 
-    flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
-    flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
+  flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
+  flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
 
-    /* Construct Encoder Context */
-    vpx_codec_enc_init(&stream->encoder, global->codec->iface,
-                        &stream->config.cfg, flags);
-    ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
+  /* Construct Encoder Context */
+  vpx_codec_enc_init(&stream->encoder, global->codec->iface(),
+                     &stream->config.cfg, flags);
+  ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
 
-    /* Note that we bypass the vpx_codec_control wrapper macro because
-     * we're being clever to store the control IDs in an array. Real
-     * applications will want to make use of the enumerations directly
-     */
-    for (i = 0; i < stream->config.arg_ctrl_cnt; i++)
-    {
-        int ctrl = stream->config.arg_ctrls[i][0];
-        int value = stream->config.arg_ctrls[i][1];
-        if (vpx_codec_control_(&stream->encoder, ctrl, value))
-            fprintf(stderr, "Error: Tried to set control %d = %d\n",
-                    ctrl, value);
+  /* Note that we bypass the vpx_codec_control wrapper macro because
+   * we're being clever to store the control IDs in an array. Real
+   * applications will want to make use of the enumerations directly
+   */
+  for (i = 0; i < stream->config.arg_ctrl_cnt; i++) {
+    int ctrl = stream->config.arg_ctrls[i][0];
+    int value = stream->config.arg_ctrls[i][1];
+    if (vpx_codec_control_(&stream->encoder, ctrl, value))
+      fprintf(stderr, "Error: Tried to set control %d = %d\n",
+              ctrl, value);
 
-        ctx_exit_on_error(&stream->encoder, "Failed to control codec");
-    }
+    ctx_exit_on_error(&stream->encoder, "Failed to control codec");
+  }
+
+#if CONFIG_DECODERS
+  if (global->test_decode != TEST_DECODE_OFF) {
+    vpx_codec_dec_init(&stream->decoder, global->codec->dx_iface(), NULL, 0);
+  }
+#endif
 }
 
 
 static void encode_frame(struct stream_state  *stream,
                          struct global_config *global,
                          struct vpx_image     *img,
-                         unsigned int          frames_in)
-{
-    vpx_codec_pts_t frame_start, next_frame_start;
-    struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
-    struct vpx_usec_timer timer;
+                         unsigned int          frames_in) {
+  vpx_codec_pts_t frame_start, next_frame_start;
+  struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
+  struct vpx_usec_timer timer;
 
-    frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
-                  * global->framerate.den)
-                  / cfg->g_timebase.num / global->framerate.num;
-    next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
-                        * global->framerate.den)
-                        / cfg->g_timebase.num / global->framerate.num;
-    vpx_usec_timer_start(&timer);
-    vpx_codec_encode(&stream->encoder, img, frame_start,
-                     (unsigned long)(next_frame_start - frame_start),
-                     0, global->deadline);
-    vpx_usec_timer_mark(&timer);
-    stream->cx_time += vpx_usec_timer_elapsed(&timer);
-    ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
-                      stream->index);
+  frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
+                 * global->framerate.den)
+                / cfg->g_timebase.num / global->framerate.num;
+  next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
+                      * global->framerate.den)
+                     / cfg->g_timebase.num / global->framerate.num;
+
+  /* Scale if necessary */
+  if (img && (img->d_w != cfg->g_w || img->d_h != cfg->g_h)) {
+    if (!stream->img)
+      stream->img = vpx_img_alloc(NULL, VPX_IMG_FMT_I420,
+                                  cfg->g_w, cfg->g_h, 16);
+    I420Scale(img->planes[VPX_PLANE_Y], img->stride[VPX_PLANE_Y],
+              img->planes[VPX_PLANE_U], img->stride[VPX_PLANE_U],
+              img->planes[VPX_PLANE_V], img->stride[VPX_PLANE_V],
+              img->d_w, img->d_h,
+              stream->img->planes[VPX_PLANE_Y],
+              stream->img->stride[VPX_PLANE_Y],
+              stream->img->planes[VPX_PLANE_U],
+              stream->img->stride[VPX_PLANE_U],
+              stream->img->planes[VPX_PLANE_V],
+              stream->img->stride[VPX_PLANE_V],
+              stream->img->d_w, stream->img->d_h,
+              kFilterBox);
+
+    img = stream->img;
+  }
+
+  vpx_usec_timer_start(&timer);
+  vpx_codec_encode(&stream->encoder, img, frame_start,
+                   (unsigned long)(next_frame_start - frame_start),
+                   0, global->deadline);
+  vpx_usec_timer_mark(&timer);
+  stream->cx_time += vpx_usec_timer_elapsed(&timer);
+  ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
+                    stream->index);
 }
 
 
-static void update_quantizer_histogram(struct stream_state *stream)
-{
-    if(stream->config.cfg.g_pass != VPX_RC_FIRST_PASS)
-    {
-        int q;
+static void update_quantizer_histogram(struct stream_state *stream) {
+  if (stream->config.cfg.g_pass != VPX_RC_FIRST_PASS) {
+    int q;
 
-        vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
-        ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
-        stream->counts[q]++;
-    }
+    vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
+    ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
+    stream->counts[q]++;
+  }
 }
 
 
 static void get_cx_data(struct stream_state  *stream,
                         struct global_config *global,
-                        int                  *got_data)
-{
-    const vpx_codec_cx_pkt_t *pkt;
-    const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
-    vpx_codec_iter_t iter = NULL;
+                        int                  *got_data) {
+  const vpx_codec_cx_pkt_t *pkt;
+  const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
+  vpx_codec_iter_t iter = NULL;
 
-    while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter)))
-    {
-        static size_t fsize = 0;
-        static off_t ivf_header_pos = 0;
+  *got_data = 0;
+  while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter))) {
+    static size_t fsize = 0;
+    static off_t ivf_header_pos = 0;
+
+    switch (pkt->kind) {
+      case VPX_CODEC_CX_FRAME_PKT:
+        if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
+          stream->frames_out++;
+        }
+        if (!global->quiet)
+          fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz);
+
+        update_rate_histogram(&stream->rate_hist, cfg, pkt);
+        if (stream->config.write_webm) {
+          /* Update the hash */
+          if (!stream->ebml.debug)
+            stream->hash = murmur(pkt->data.frame.buf,
+                                  (int)pkt->data.frame.sz,
+                                  stream->hash);
+
+          write_webm_block(&stream->ebml, cfg, pkt);
+        } else {
+          if (pkt->data.frame.partition_id <= 0) {
+            ivf_header_pos = ftello(stream->file);
+            fsize = pkt->data.frame.sz;
+
+            write_ivf_frame_header(stream->file, pkt);
+          } else {
+            fsize += pkt->data.frame.sz;
+
+            if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
+              off_t currpos = ftello(stream->file);
+              fseeko(stream->file, ivf_header_pos, SEEK_SET);
+              write_ivf_frame_size(stream->file, fsize);
+              fseeko(stream->file, currpos, SEEK_SET);
+            }
+          }
+
+          (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
+                        stream->file);
+        }
+        stream->nbytes += pkt->data.raw.sz;
 
         *got_data = 1;
+#if CONFIG_DECODERS
+        if (global->test_decode != TEST_DECODE_OFF && !stream->mismatch_seen) {
+          vpx_codec_decode(&stream->decoder, pkt->data.frame.buf,
+                           pkt->data.frame.sz, NULL, 0);
+          if (stream->decoder.err) {
+            warn_or_exit_on_error(&stream->decoder,
+                                  global->test_decode == TEST_DECODE_FATAL,
+                                  "Failed to decode frame %d in stream %d",
+                                  stream->frames_out + 1, stream->index);
+            stream->mismatch_seen = stream->frames_out + 1;
+          }
+        }
+#endif
+        break;
+      case VPX_CODEC_STATS_PKT:
+        stream->frames_out++;
+        stats_write(&stream->stats,
+                    pkt->data.twopass_stats.buf,
+                    pkt->data.twopass_stats.sz);
+        stream->nbytes += pkt->data.raw.sz;
+        break;
+      case VPX_CODEC_PSNR_PKT:
 
-        switch (pkt->kind)
-        {
-        case VPX_CODEC_CX_FRAME_PKT:
-            if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
-            {
-                stream->frames_out++;
-            }
+        if (global->show_psnr) {
+          int i;
+
+          stream->psnr_sse_total += pkt->data.psnr.sse[0];
+          stream->psnr_samples_total += pkt->data.psnr.samples[0];
+          for (i = 0; i < 4; i++) {
             if (!global->quiet)
-                fprintf(stderr, " %6luF",
-                        (unsigned long)pkt->data.frame.sz);
-
-            update_rate_histogram(&stream->rate_hist, cfg, pkt);
-            if(stream->config.write_webm)
-            {
-                /* Update the hash */
-                if(!stream->ebml.debug)
-                    stream->hash = murmur(pkt->data.frame.buf,
-                                          (int)pkt->data.frame.sz,
-                                          stream->hash);
-
-                write_webm_block(&stream->ebml, cfg, pkt);
-            }
-            else
-            {
-                if (pkt->data.frame.partition_id <= 0)
-                {
-                    ivf_header_pos = ftello(stream->file);
-                    fsize = pkt->data.frame.sz;
-
-                    write_ivf_frame_header(stream->file, pkt);
-                }
-                else
-                {
-                    fsize += pkt->data.frame.sz;
-
-                    if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
-                    {
-                        off_t currpos = ftello(stream->file);
-                        fseeko(stream->file, ivf_header_pos, SEEK_SET);
-                        write_ivf_frame_size(stream->file, fsize);
-                        fseeko(stream->file, currpos, SEEK_SET);
-                    }
-                }
-
-                (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
-                              stream->file);
-            }
-            stream->nbytes += pkt->data.raw.sz;
-            break;
-        case VPX_CODEC_STATS_PKT:
-            stream->frames_out++;
-            fprintf(stderr, " %6luS",
-                   (unsigned long)pkt->data.twopass_stats.sz);
-            stats_write(&stream->stats,
-                        pkt->data.twopass_stats.buf,
-                        pkt->data.twopass_stats.sz);
-            stream->nbytes += pkt->data.raw.sz;
-            break;
-        case VPX_CODEC_PSNR_PKT:
-
-            if (global->show_psnr)
-            {
-                int i;
-
-                stream->psnr_sse_total += pkt->data.psnr.sse[0];
-                stream->psnr_samples_total += pkt->data.psnr.samples[0];
-                for (i = 0; i < 4; i++)
-                {
-                    if (!global->quiet)
-                        fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
-                    stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
-                }
-                stream->psnr_count++;
-            }
-
-            break;
-        default:
-            break;
+              fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
+            stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
+          }
+          stream->psnr_count++;
         }
+
+        break;
+      default:
+        break;
     }
+  }
 }
 
 
-static void show_psnr(struct stream_state  *stream)
-{
-    int i;
-    double ovpsnr;
+static void show_psnr(struct stream_state  *stream) {
+  int i;
+  double ovpsnr;
 
-    if (!stream->psnr_count)
-        return;
+  if (!stream->psnr_count)
+    return;
 
-    fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
-    ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0,
-                          (double)stream->psnr_sse_total);
-    fprintf(stderr, " %.3f", ovpsnr);
+  fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
+  ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0,
+                        (double)stream->psnr_sse_total);
+  fprintf(stderr, " %.3f", ovpsnr);
 
-    for (i = 0; i < 4; i++)
-    {
-        fprintf(stderr, " %.3f", stream->psnr_totals[i]/stream->psnr_count);
-    }
-    fprintf(stderr, "\n");
+  for (i = 0; i < 4; i++) {
+    fprintf(stderr, " %.3f", stream->psnr_totals[i] / stream->psnr_count);
+  }
+  fprintf(stderr, "\n");
 }
 
 
-float usec_to_fps(uint64_t usec, unsigned int frames)
-{
-    return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0);
+static float usec_to_fps(uint64_t usec, unsigned int frames) {
+  return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0);
 }
 
 
-int main(int argc, const char **argv_)
-{
-    int                    pass;
-    vpx_image_t            raw;
-    int                    frame_avail, got_data;
+static void test_decode(struct stream_state  *stream,
+                        enum TestDecodeFatality fatal,
+                        const struct codec_item *codec) {
+  vpx_image_t enc_img, dec_img;
 
-    struct input_state       input = {0};
-    struct global_config     global;
-    struct stream_state     *streams = NULL;
-    char                   **argv, **argi;
-    unsigned long            cx_time = 0;
-    int                      stream_cnt = 0;
+  if (stream->mismatch_seen)
+    return;
 
-    exec_name = argv_[0];
+  /* Get the internal reference frame */
+  if (codec->fourcc == VP8_FOURCC) {
+    struct vpx_ref_frame ref_enc, ref_dec;
+    int width, height;
 
-    if (argc < 3)
-        usage_exit();
+    width = (stream->config.cfg.g_w + 15) & ~15;
+    height = (stream->config.cfg.g_h + 15) & ~15;
+    vpx_img_alloc(&ref_enc.img, VPX_IMG_FMT_I420, width, height, 1);
+    enc_img = ref_enc.img;
+    vpx_img_alloc(&ref_dec.img, VPX_IMG_FMT_I420, width, height, 1);
+    dec_img = ref_dec.img;
 
-    /* Setup default input stream settings */
-    input.framerate.num = 30;
-    input.framerate.den = 1;
-    input.use_i420 = 1;
+    ref_enc.frame_type = VP8_LAST_FRAME;
+    ref_dec.frame_type = VP8_LAST_FRAME;
+    vpx_codec_control(&stream->encoder, VP8_COPY_REFERENCE, &ref_enc);
+    vpx_codec_control(&stream->decoder, VP8_COPY_REFERENCE, &ref_dec);
+  } else {
+    struct vp9_ref_frame ref;
 
-    /* First parse the global configuration values, because we want to apply
-     * other parameters on top of the default configuration provided by the
-     * codec.
+    ref.idx = 0;
+    vpx_codec_control(&stream->encoder, VP9_GET_REFERENCE, &ref);
+    enc_img = ref.img;
+    vpx_codec_control(&stream->decoder, VP9_GET_REFERENCE, &ref);
+    dec_img = ref.img;
+  }
+  ctx_exit_on_error(&stream->encoder, "Failed to get encoder reference frame");
+  ctx_exit_on_error(&stream->decoder, "Failed to get decoder reference frame");
+
+  if (!compare_img(&enc_img, &dec_img)) {
+    int y[4], u[4], v[4];
+    find_mismatch(&enc_img, &dec_img, y, u, v);
+    stream->decoder.err = 1;
+    warn_or_exit_on_error(&stream->decoder, fatal == TEST_DECODE_FATAL,
+                          "Stream %d: Encode/decode mismatch on frame %d at"
+                          " Y[%d, %d] {%d/%d},"
+                          " U[%d, %d] {%d/%d},"
+                          " V[%d, %d] {%d/%d}",
+                          stream->index, stream->frames_out,
+                          y[0], y[1], y[2], y[3],
+                          u[0], u[1], u[2], u[3],
+                          v[0], v[1], v[2], v[3]);
+    stream->mismatch_seen = stream->frames_out;
+  }
+
+  vpx_img_free(&enc_img);
+  vpx_img_free(&dec_img);
+}
+
+
+static void print_time(const char *label, int64_t etl) {
+  int hours, mins, secs;
+
+  if (etl >= 0) {
+    hours = etl / 3600;
+    etl -= hours * 3600;
+    mins = etl / 60;
+    etl -= mins * 60;
+    secs = etl;
+
+    fprintf(stderr, "[%3s %2d:%02d:%02d] ",
+            label, hours, mins, secs);
+  } else {
+    fprintf(stderr, "[%3s  unknown] ", label);
+  }
+}
+
+int main(int argc, const char **argv_) {
+  int                    pass;
+  vpx_image_t            raw;
+  int                    frame_avail, got_data;
+
+  struct input_state       input = {0};
+  struct global_config     global;
+  struct stream_state     *streams = NULL;
+  char                   **argv, **argi;
+  uint64_t                 cx_time = 0;
+  int                      stream_cnt = 0;
+  int                      res = 0;
+
+  exec_name = argv_[0];
+
+  if (argc < 3)
+    usage_exit();
+
+  /* Setup default input stream settings */
+  input.framerate.num = 30;
+  input.framerate.den = 1;
+  input.use_i420 = 1;
+  input.only_i420 = 1;
+
+  /* First parse the global configuration values, because we want to apply
+   * other parameters on top of the default configuration provided by the
+   * codec.
+   */
+  argv = argv_dup(argc - 1, argv_ + 1);
+  parse_global_config(&global, argv);
+
+  {
+    /* Now parse each stream's parameters. Using a local scope here
+     * due to the use of 'stream' as loop variable in FOREACH_STREAM
+     * loops
      */
-    argv = argv_dup(argc - 1, argv_ + 1);
-    parse_global_config(&global, argv);
+    struct stream_state *stream = NULL;
 
-    {
-        /* Now parse each stream's parameters. Using a local scope here
-         * due to the use of 'stream' as loop variable in FOREACH_STREAM
-         * loops
-         */
-        struct stream_state *stream = NULL;
+    do {
+      stream = new_stream(&global, stream);
+      stream_cnt++;
+      if (!streams)
+        streams = stream;
+    } while (parse_stream_params(&global, stream, argv));
+  }
 
-        do
-        {
-            stream = new_stream(&global, stream);
-            stream_cnt++;
-            if(!streams)
-                streams = stream;
-        } while(parse_stream_params(&global, stream, argv));
+  /* Check for unrecognized options */
+  for (argi = argv; *argi; argi++)
+    if (argi[0][0] == '-' && argi[0][1])
+      die("Error: Unrecognized option %s\n", *argi);
+
+  /* Handle non-option arguments */
+  input.fn = argv[0];
+
+  if (!input.fn)
+    usage_exit();
+
+#if CONFIG_NON420
+  /* Decide if other chroma subsamplings than 4:2:0 are supported */
+  if (global.codec->fourcc == VP9_FOURCC)
+    input.only_i420 = 0;
+#endif
+
+  for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++) {
+    int frames_in = 0, seen_frames = 0;
+    int64_t estimated_time_left = -1;
+    int64_t average_rate = -1;
+    off_t lagged_count = 0;
+
+    open_input_file(&input);
+
+    /* If the input file doesn't specify its w/h (raw files), try to get
+     * the data from the first stream's configuration.
+     */
+    if (!input.w || !input.h)
+      FOREACH_STREAM( {
+      if (stream->config.cfg.g_w && stream->config.cfg.g_h) {
+        input.w = stream->config.cfg.g_w;
+        input.h = stream->config.cfg.g_h;
+        break;
+      }
+    });
+
+    /* Update stream configurations from the input file's parameters */
+    if (!input.w || !input.h)
+      fatal("Specify stream dimensions with --width (-w) "
+            " and --height (-h)");
+    FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h));
+    FOREACH_STREAM(validate_stream_config(stream));
+
+    /* Ensure that --passes and --pass are consistent. If --pass is set and
+     * --passes=2, ensure --fpf was set.
+     */
+    if (global.pass && global.passes == 2)
+      FOREACH_STREAM( {
+      if (!stream->config.stats_fn)
+        die("Stream %d: Must specify --fpf when --pass=%d"
+        " and --passes=2\n", stream->index, global.pass);
+    });
+
+    /* Use the frame rate from the file only if none was specified
+     * on the command-line.
+     */
+    if (!global.have_framerate)
+      global.framerate = input.framerate;
+
+    FOREACH_STREAM(set_default_kf_interval(stream, &global));
+
+    /* Show configuration */
+    if (global.verbose && pass == 0)
+      FOREACH_STREAM(show_stream_config(stream, &global, &input));
+
+    if (pass == (global.pass ? global.pass - 1 : 0)) {
+      if (input.file_type == FILE_TYPE_Y4M)
+        /*The Y4M reader does its own allocation.
+          Just initialize this here to avoid problems if we never read any
+           frames.*/
+        memset(&raw, 0, sizeof(raw));
+      else
+        vpx_img_alloc(&raw,
+                      input.use_i420 ? VPX_IMG_FMT_I420
+                      : VPX_IMG_FMT_YV12,
+                      input.w, input.h, 32);
+
+      FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
+                                         &stream->config.cfg,
+                                         &global.framerate));
     }
 
-    /* Check for unrecognized options */
-    for (argi = argv; *argi; argi++)
-        if (argi[0][0] == '-' && argi[0][1])
-            die("Error: Unrecognized option %s\n", *argi);
+    FOREACH_STREAM(open_output_file(stream, &global));
+    FOREACH_STREAM(setup_pass(stream, &global, pass));
+    FOREACH_STREAM(initialize_encoder(stream, &global));
 
-    /* Handle non-option arguments */
-    input.fn = argv[0];
+    frame_avail = 1;
+    got_data = 0;
 
-    if (!input.fn)
-        usage_exit();
+    while (frame_avail || got_data) {
+      struct vpx_usec_timer timer;
 
-    for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++)
-    {
-        int frames_in = 0;
+      if (!global.limit || frames_in < global.limit) {
+        frame_avail = read_frame(&input, &raw);
 
-        open_input_file(&input);
+        if (frame_avail)
+          frames_in++;
+        seen_frames = frames_in > global.skip_frames ?
+                          frames_in - global.skip_frames : 0;
 
-        /* If the input file doesn't specify its w/h (raw files), try to get
-         * the data from the first stream's configuration.
-         */
-        if(!input.w || !input.h)
-            FOREACH_STREAM({
-                if(stream->config.cfg.g_w && stream->config.cfg.g_h)
-                {
-                    input.w = stream->config.cfg.g_w;
-                    input.h = stream->config.cfg.g_h;
-                    break;
-                }
-            });
+        if (!global.quiet) {
+          float fps = usec_to_fps(cx_time, seen_frames);
+          fprintf(stderr, "\rPass %d/%d ", pass + 1, global.passes);
 
-        /* Update stream configurations from the input file's parameters */
-        FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h));
-        FOREACH_STREAM(validate_stream_config(stream));
+          if (stream_cnt == 1)
+            fprintf(stderr,
+                    "frame %4d/%-4d %7"PRId64"B ",
+                    frames_in, streams->frames_out, (int64_t)streams->nbytes);
+          else
+            fprintf(stderr, "frame %4d ", frames_in);
 
-        /* Ensure that --passes and --pass are consistent. If --pass is set and
-         * --passes=2, ensure --fpf was set.
-         */
-        if (global.pass && global.passes == 2)
-            FOREACH_STREAM({
-                if(!stream->config.stats_fn)
-                    die("Stream %d: Must specify --fpf when --pass=%d"
-                        " and --passes=2\n", stream->index, global.pass);
-            });
-
-
-        /* Use the frame rate from the file only if none was specified
-         * on the command-line.
-         */
-        if (!global.have_framerate)
-            global.framerate = input.framerate;
-
-        FOREACH_STREAM(set_default_kf_interval(stream, &global));
-
-        /* Show configuration */
-        if (global.verbose && pass == 0)
-            FOREACH_STREAM(show_stream_config(stream, &global, &input));
-
-        if(pass == (global.pass ? global.pass - 1 : 0)) {
-            if (input.file_type == FILE_TYPE_Y4M)
-                /*The Y4M reader does its own allocation.
-                  Just initialize this here to avoid problems if we never read any
-                   frames.*/
-                memset(&raw, 0, sizeof(raw));
-            else
-                vpx_img_alloc(&raw,
-                              input.use_i420 ? VPX_IMG_FMT_I420
-                                             : VPX_IMG_FMT_YV12,
-                              input.w, input.h, 32);
-
-            FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
-                                               &stream->config.cfg,
-                                               &global.framerate));
+          fprintf(stderr, "%7"PRId64" %s %.2f %s ",
+                  cx_time > 9999999 ? cx_time / 1000 : cx_time,
+                  cx_time > 9999999 ? "ms" : "us",
+                  fps >= 1.0 ? fps : 1000.0 / fps,
+                  fps >= 1.0 ? "fps" : "ms/f");
+          print_time("ETA", estimated_time_left);
+          fprintf(stderr, "\033[K");
         }
 
-        FOREACH_STREAM(open_output_file(stream, &global));
-        FOREACH_STREAM(setup_pass(stream, &global, pass));
-        FOREACH_STREAM(initialize_encoder(stream, &global));
+      } else
+        frame_avail = 0;
 
-        frame_avail = 1;
+      if (frames_in > global.skip_frames) {
+        vpx_usec_timer_start(&timer);
+        FOREACH_STREAM(encode_frame(stream, &global,
+                                    frame_avail ? &raw : NULL,
+                                    frames_in));
+        vpx_usec_timer_mark(&timer);
+        cx_time += vpx_usec_timer_elapsed(&timer);
+
+        FOREACH_STREAM(update_quantizer_histogram(stream));
+
         got_data = 0;
+        FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
 
-        while (frame_avail || got_data)
-        {
-            struct vpx_usec_timer timer;
+        if (!got_data && input.length && !streams->frames_out) {
+          lagged_count = global.limit ? seen_frames : ftello(input.file);
+        } else if (input.length) {
+          int64_t remaining;
+          int64_t rate;
 
-            if (!global.limit || frames_in < global.limit)
-            {
-                frame_avail = read_frame(&input, &raw);
+          if (global.limit) {
+            int frame_in_lagged = (seen_frames - lagged_count) * 1000;
 
-                if (frame_avail)
-                    frames_in++;
+            rate = cx_time ? frame_in_lagged * (int64_t)1000000 / cx_time : 0;
+            remaining = 1000 * (global.limit - global.skip_frames
+                                - seen_frames + lagged_count);
+          } else {
+            off_t input_pos = ftello(input.file);
+            off_t input_pos_lagged = input_pos - lagged_count;
+            int64_t limit = input.length;
 
-                if (!global.quiet)
-                {
-                    if(stream_cnt == 1)
-                        fprintf(stderr,
-                                "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K",
-                                pass + 1, global.passes, frames_in,
-                                streams->frames_out, (int64_t)streams->nbytes);
-                    else
-                        fprintf(stderr,
-                                "\rPass %d/%d frame %4d %7lu %s (%.2f fps)\033[K",
-                                pass + 1, global.passes, frames_in,
-                                cx_time > 9999999 ? cx_time / 1000 : cx_time,
-                                cx_time > 9999999 ? "ms" : "us",
-                                usec_to_fps(cx_time, frames_in));
-                }
+            rate = cx_time ? input_pos_lagged * (int64_t)1000000 / cx_time : 0;
+            remaining = limit - input_pos + lagged_count;
+          }
 
-            }
-            else
-                frame_avail = 0;
-
-            vpx_usec_timer_start(&timer);
-            FOREACH_STREAM(encode_frame(stream, &global,
-                                        frame_avail ? &raw : NULL,
-                                        frames_in));
-            vpx_usec_timer_mark(&timer);
-            cx_time += (unsigned long)vpx_usec_timer_elapsed(&timer);
-
-            FOREACH_STREAM(update_quantizer_histogram(stream));
-
-            got_data = 0;
-            FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
-
-            fflush(stdout);
+          average_rate = (average_rate <= 0)
+              ? rate
+              : (average_rate * 7 + rate) / 8;
+          estimated_time_left = average_rate ? remaining / average_rate : -1;
         }
 
-        if(stream_cnt > 1)
-            fprintf(stderr, "\n");
+        if (got_data && global.test_decode != TEST_DECODE_OFF)
+          FOREACH_STREAM(test_decode(stream, global.test_decode, global.codec));
+      }
 
-        if (!global.quiet)
-            FOREACH_STREAM(fprintf(
-                stderr,
-                "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
-                " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
-                global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
-                frames_in ? (unsigned long)(stream->nbytes * 8 / frames_in) : 0,
-                frames_in ? (int64_t)stream->nbytes * 8
-                            * (int64_t)global.framerate.num / global.framerate.den
-                            / frames_in
-                          : 0,
-                stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
-                stream->cx_time > 9999999 ? "ms" : "us",
-                usec_to_fps(stream->cx_time, frames_in));
-            );
-
-        if (global.show_psnr)
-            FOREACH_STREAM(show_psnr(stream));
-
-        FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
-
-        close_input_file(&input);
-
-        FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
-
-        FOREACH_STREAM(stats_close(&stream->stats, global.passes-1));
-
-        if (global.pass)
-            break;
+      fflush(stdout);
     }
 
-    if (global.show_q_hist_buckets)
-        FOREACH_STREAM(show_q_histogram(stream->counts,
-                                        global.show_q_hist_buckets));
+    if (stream_cnt > 1)
+      fprintf(stderr, "\n");
 
-    if (global.show_rate_hist_buckets)
-        FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
-                                           &stream->config.cfg,
-                                           global.show_rate_hist_buckets));
-    FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
+    if (!global.quiet)
+      FOREACH_STREAM(fprintf(
+                       stderr,
+                       "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
+                       " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
+                       global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
+                       seen_frames ? (unsigned long)(stream->nbytes * 8 / seen_frames) : 0,
+                       seen_frames ? (int64_t)stream->nbytes * 8
+                       * (int64_t)global.framerate.num / global.framerate.den
+                       / seen_frames
+                       : 0,
+                       stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
+                       stream->cx_time > 9999999 ? "ms" : "us",
+                       usec_to_fps(stream->cx_time, seen_frames));
+                    );
 
-    vpx_img_free(&raw);
-    free(argv);
-    free(streams);
-    return EXIT_SUCCESS;
+    if (global.show_psnr)
+      FOREACH_STREAM(show_psnr(stream));
+
+    FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
+
+    if (global.test_decode != TEST_DECODE_OFF) {
+      FOREACH_STREAM(vpx_codec_destroy(&stream->decoder));
+    }
+
+    close_input_file(&input);
+
+    if (global.test_decode == TEST_DECODE_FATAL) {
+      FOREACH_STREAM(res |= stream->mismatch_seen);
+    }
+    FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
+
+    FOREACH_STREAM(stats_close(&stream->stats, global.passes - 1));
+
+    if (global.pass)
+      break;
+  }
+
+  if (global.show_q_hist_buckets)
+    FOREACH_STREAM(show_q_histogram(stream->counts,
+                                    global.show_q_hist_buckets));
+
+  if (global.show_rate_hist_buckets)
+    FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
+                                       &stream->config.cfg,
+                                       global.show_rate_hist_buckets));
+  FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
+
+#if CONFIG_INTERNAL_STATS
+  /* TODO(jkoleszar): This doesn't belong in this executable. Do it for now,
+   * to match some existing utilities.
+   */
+  FOREACH_STREAM({
+    FILE *f = fopen("opsnr.stt", "a");
+    if (stream->mismatch_seen) {
+      fprintf(f, "First mismatch occurred in frame %d\n",
+              stream->mismatch_seen);
+    } else {
+      fprintf(f, "No mismatch detected in recon buffers\n");
+    }
+    fclose(f);
+  });
+#endif
+
+  vpx_img_free(&raw);
+  free(argv);
+  free(streams);
+  return res ? EXIT_FAILURE : EXIT_SUCCESS;
 }
diff --git a/libvpx/y4minput.c b/libvpx/y4minput.c
index ff9ffbc..47f005a 100644
--- a/libvpx/y4minput.c
+++ b/libvpx/y4minput.c
@@ -14,7 +14,7 @@
 #include <string.h>
 #include "y4minput.h"
 
-static int y4m_parse_tags(y4m_input *_y4m,char *_tags){
+static int y4m_parse_tags(y4m_input *_y4m, char *_tags) {
   int   got_w;
   int   got_h;
   int   got_fps;
@@ -23,55 +23,61 @@
   int   got_chroma;
   char *p;
   char *q;
-  got_w=got_h=got_fps=got_interlace=got_par=got_chroma=0;
-  for(p=_tags;;p=q){
+  got_w = got_h = got_fps = got_interlace = got_par = got_chroma = 0;
+  for (p = _tags;; p = q) {
     /*Skip any leading spaces.*/
-    while(*p==' ')p++;
+    while (*p == ' ')p++;
     /*If that's all we have, stop.*/
-    if(p[0]=='\0')break;
+    if (p[0] == '\0')break;
     /*Find the end of this tag.*/
-    for(q=p+1;*q!='\0'&&*q!=' ';q++);
+    for (q = p + 1; *q != '\0' && *q != ' '; q++);
     /*Process the tag.*/
-    switch(p[0]){
-      case 'W':{
-        if(sscanf(p+1,"%d",&_y4m->pic_w)!=1)return -1;
-        got_w=1;
-      }break;
-      case 'H':{
-        if(sscanf(p+1,"%d",&_y4m->pic_h)!=1)return -1;
-        got_h=1;
-      }break;
-      case 'F':{
-        if(sscanf(p+1,"%d:%d",&_y4m->fps_n,&_y4m->fps_d)!=2){
+    switch (p[0]) {
+      case 'W': {
+        if (sscanf(p + 1, "%d", &_y4m->pic_w) != 1)return -1;
+        got_w = 1;
+      }
+      break;
+      case 'H': {
+        if (sscanf(p + 1, "%d", &_y4m->pic_h) != 1)return -1;
+        got_h = 1;
+      }
+      break;
+      case 'F': {
+        if (sscanf(p + 1, "%d:%d", &_y4m->fps_n, &_y4m->fps_d) != 2) {
           return -1;
         }
-        got_fps=1;
-      }break;
-      case 'I':{
-        _y4m->interlace=p[1];
-        got_interlace=1;
-      }break;
-      case 'A':{
-        if(sscanf(p+1,"%d:%d",&_y4m->par_n,&_y4m->par_d)!=2){
+        got_fps = 1;
+      }
+      break;
+      case 'I': {
+        _y4m->interlace = p[1];
+        got_interlace = 1;
+      }
+      break;
+      case 'A': {
+        if (sscanf(p + 1, "%d:%d", &_y4m->par_n, &_y4m->par_d) != 2) {
           return -1;
         }
-        got_par=1;
-      }break;
-      case 'C':{
-        if(q-p>16)return -1;
-        memcpy(_y4m->chroma_type,p+1,q-p-1);
-        _y4m->chroma_type[q-p-1]='\0';
-        got_chroma=1;
-      }break;
+        got_par = 1;
+      }
+      break;
+      case 'C': {
+        if (q - p > 16)return -1;
+        memcpy(_y4m->chroma_type, p + 1, q - p - 1);
+        _y4m->chroma_type[q - p - 1] = '\0';
+        got_chroma = 1;
+      }
+      break;
       /*Ignore unknown tags.*/
     }
   }
-  if(!got_w||!got_h||!got_fps)return -1;
-  if(!got_interlace)_y4m->interlace='?';
-  if(!got_par)_y4m->par_n=_y4m->par_d=0;
+  if (!got_w || !got_h || !got_fps)return -1;
+  if (!got_interlace)_y4m->interlace = '?';
+  if (!got_par)_y4m->par_n = _y4m->par_d = 0;
   /*Chroma-type is not specified in older files, e.g., those generated by
      mplayer.*/
-  if(!got_chroma)strcpy(_y4m->chroma_type,"420");
+  if (!got_chroma)strcpy(_y4m->chroma_type, "420");
   return 0;
 }
 
@@ -145,48 +151,48 @@
    lines, and they are vertically co-sited with the luma samples in both the
    mpeg2 and jpeg cases (thus requiring no vertical resampling).*/
 static void y4m_42xmpeg2_42xjpeg_helper(unsigned char *_dst,
- const unsigned char *_src,int _c_w,int _c_h){
+                                        const unsigned char *_src, int _c_w, int _c_h) {
   int y;
   int x;
-  for(y=0;y<_c_h;y++){
+  for (y = 0; y < _c_h; y++) {
     /*Filter: [4 -17 114 35 -9 1]/128, derived from a 6-tap Lanczos
        window.*/
-    for(x=0;x<OC_MINI(_c_w,2);x++){
-      _dst[x]=(unsigned char)OC_CLAMPI(0,(4*_src[0]-17*_src[OC_MAXI(x-1,0)]+
-       114*_src[x]+35*_src[OC_MINI(x+1,_c_w-1)]-9*_src[OC_MINI(x+2,_c_w-1)]+
-       _src[OC_MINI(x+3,_c_w-1)]+64)>>7,255);
+    for (x = 0; x < OC_MINI(_c_w, 2); x++) {
+      _dst[x] = (unsigned char)OC_CLAMPI(0, (4 * _src[0] - 17 * _src[OC_MAXI(x - 1, 0)] +
+                                             114 * _src[x] + 35 * _src[OC_MINI(x + 1, _c_w - 1)] - 9 * _src[OC_MINI(x + 2, _c_w - 1)] +
+                                             _src[OC_MINI(x + 3, _c_w - 1)] + 64) >> 7, 255);
     }
-    for(;x<_c_w-3;x++){
-      _dst[x]=(unsigned char)OC_CLAMPI(0,(4*_src[x-2]-17*_src[x-1]+
-       114*_src[x]+35*_src[x+1]-9*_src[x+2]+_src[x+3]+64)>>7,255);
+    for (; x < _c_w - 3; x++) {
+      _dst[x] = (unsigned char)OC_CLAMPI(0, (4 * _src[x - 2] - 17 * _src[x - 1] +
+                                             114 * _src[x] + 35 * _src[x + 1] - 9 * _src[x + 2] + _src[x + 3] + 64) >> 7, 255);
     }
-    for(;x<_c_w;x++){
-      _dst[x]=(unsigned char)OC_CLAMPI(0,(4*_src[x-2]-17*_src[x-1]+
-       114*_src[x]+35*_src[OC_MINI(x+1,_c_w-1)]-9*_src[OC_MINI(x+2,_c_w-1)]+
-       _src[_c_w-1]+64)>>7,255);
+    for (; x < _c_w; x++) {
+      _dst[x] = (unsigned char)OC_CLAMPI(0, (4 * _src[x - 2] - 17 * _src[x - 1] +
+                                             114 * _src[x] + 35 * _src[OC_MINI(x + 1, _c_w - 1)] - 9 * _src[OC_MINI(x + 2, _c_w - 1)] +
+                                             _src[_c_w - 1] + 64) >> 7, 255);
     }
-    _dst+=_c_w;
-    _src+=_c_w;
+    _dst += _c_w;
+    _src += _c_w;
   }
 }
 
 /*Handles both 422 and 420mpeg2 to 422jpeg and 420jpeg, respectively.*/
-static void y4m_convert_42xmpeg2_42xjpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_42xmpeg2_42xjpeg(y4m_input *_y4m, unsigned char *_dst,
+                                         unsigned char *_aux) {
   int c_w;
   int c_h;
   int c_sz;
   int pli;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  for(pli=1;pli<3;pli++){
-    y4m_42xmpeg2_42xjpeg_helper(_dst,_aux,c_w,c_h);
-    _dst+=c_sz;
-    _aux+=c_sz;
+  c_w = (_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  for (pli = 1; pli < 3; pli++) {
+    y4m_42xmpeg2_42xjpeg_helper(_dst, _aux, c_w, c_h);
+    _dst += c_sz;
+    _aux += c_sz;
   }
 }
 
@@ -233,8 +239,8 @@
    the chroma plane's resolution) to the right.
   Then we use another filter to move the C_r location down one quarter pixel,
    and the C_b location up one quarter pixel.*/
-static void y4m_convert_42xpaldv_42xjpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_42xpaldv_42xjpeg(y4m_input *_y4m, unsigned char *_dst,
+                                         unsigned char *_aux) {
   unsigned char *tmp;
   int            c_w;
   int            c_h;
@@ -243,69 +249,71 @@
   int            y;
   int            x;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+1)/2;
-  c_h=(_y4m->pic_h+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  c_sz=c_w*c_h;
-  tmp=_aux+2*c_sz;
-  for(pli=1;pli<3;pli++){
+  c_w = (_y4m->pic_w + 1) / 2;
+  c_h = (_y4m->pic_h + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  c_sz = c_w * c_h;
+  tmp = _aux + 2 * c_sz;
+  for (pli = 1; pli < 3; pli++) {
     /*First do the horizontal re-sampling.
       This is the same as the mpeg2 case, except that after the horizontal
        case, we need to apply a second vertical filter.*/
-    y4m_42xmpeg2_42xjpeg_helper(tmp,_aux,c_w,c_h);
-    _aux+=c_sz;
-    switch(pli){
-      case 1:{
+    y4m_42xmpeg2_42xjpeg_helper(tmp, _aux, c_w, c_h);
+    _aux += c_sz;
+    switch (pli) {
+      case 1: {
         /*Slide C_b up a quarter-pel.
           This is the same filter used above, but in the other order.*/
-        for(x=0;x<c_w;x++){
-          for(y=0;y<OC_MINI(c_h,3);y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(tmp[0]
-             -9*tmp[OC_MAXI(y-2,0)*c_w]+35*tmp[OC_MAXI(y-1,0)*c_w]
-             +114*tmp[y*c_w]-17*tmp[OC_MINI(y+1,c_h-1)*c_w]
-             +4*tmp[OC_MINI(y+2,c_h-1)*c_w]+64)>>7,255);
+        for (x = 0; x < c_w; x++) {
+          for (y = 0; y < OC_MINI(c_h, 3); y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (tmp[0]
+                                                         - 9 * tmp[OC_MAXI(y - 2, 0) * c_w] + 35 * tmp[OC_MAXI(y - 1, 0) * c_w]
+                                                         + 114 * tmp[y * c_w] - 17 * tmp[OC_MINI(y + 1, c_h - 1) * c_w]
+                                                         + 4 * tmp[OC_MINI(y + 2, c_h - 1) * c_w] + 64) >> 7, 255);
           }
-          for(;y<c_h-2;y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(tmp[(y-3)*c_w]
-             -9*tmp[(y-2)*c_w]+35*tmp[(y-1)*c_w]+114*tmp[y*c_w]
-             -17*tmp[(y+1)*c_w]+4*tmp[(y+2)*c_w]+64)>>7,255);
+          for (; y < c_h - 2; y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (tmp[(y - 3) * c_w]
+                                                         - 9 * tmp[(y - 2) * c_w] + 35 * tmp[(y - 1) * c_w] + 114 * tmp[y * c_w]
+                                                         - 17 * tmp[(y + 1) * c_w] + 4 * tmp[(y + 2) * c_w] + 64) >> 7, 255);
           }
-          for(;y<c_h;y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(tmp[(y-3)*c_w]
-             -9*tmp[(y-2)*c_w]+35*tmp[(y-1)*c_w]+114*tmp[y*c_w]
-             -17*tmp[OC_MINI(y+1,c_h-1)*c_w]+4*tmp[(c_h-1)*c_w]+64)>>7,255);
+          for (; y < c_h; y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (tmp[(y - 3) * c_w]
+                                                         - 9 * tmp[(y - 2) * c_w] + 35 * tmp[(y - 1) * c_w] + 114 * tmp[y * c_w]
+                                                         - 17 * tmp[OC_MINI(y + 1, c_h - 1) * c_w] + 4 * tmp[(c_h - 1) * c_w] + 64) >> 7, 255);
           }
           _dst++;
           tmp++;
         }
-        _dst+=c_sz-c_w;
-        tmp-=c_w;
-      }break;
-      case 2:{
+        _dst += c_sz - c_w;
+        tmp -= c_w;
+      }
+      break;
+      case 2: {
         /*Slide C_r down a quarter-pel.
           This is the same as the horizontal filter.*/
-        for(x=0;x<c_w;x++){
-          for(y=0;y<OC_MINI(c_h,2);y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(4*tmp[0]
-             -17*tmp[OC_MAXI(y-1,0)*c_w]+114*tmp[y*c_w]
-             +35*tmp[OC_MINI(y+1,c_h-1)*c_w]-9*tmp[OC_MINI(y+2,c_h-1)*c_w]
-             +tmp[OC_MINI(y+3,c_h-1)*c_w]+64)>>7,255);
+        for (x = 0; x < c_w; x++) {
+          for (y = 0; y < OC_MINI(c_h, 2); y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (4 * tmp[0]
+                                                         - 17 * tmp[OC_MAXI(y - 1, 0) * c_w] + 114 * tmp[y * c_w]
+                                                         + 35 * tmp[OC_MINI(y + 1, c_h - 1) * c_w] - 9 * tmp[OC_MINI(y + 2, c_h - 1) * c_w]
+                                                         + tmp[OC_MINI(y + 3, c_h - 1) * c_w] + 64) >> 7, 255);
           }
-          for(;y<c_h-3;y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(4*tmp[(y-2)*c_w]
-             -17*tmp[(y-1)*c_w]+114*tmp[y*c_w]+35*tmp[(y+1)*c_w]
-             -9*tmp[(y+2)*c_w]+tmp[(y+3)*c_w]+64)>>7,255);
+          for (; y < c_h - 3; y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (4 * tmp[(y - 2) * c_w]
+                                                         - 17 * tmp[(y - 1) * c_w] + 114 * tmp[y * c_w] + 35 * tmp[(y + 1) * c_w]
+                                                         - 9 * tmp[(y + 2) * c_w] + tmp[(y + 3) * c_w] + 64) >> 7, 255);
           }
-          for(;y<c_h;y++){
-            _dst[y*c_w]=(unsigned char)OC_CLAMPI(0,(4*tmp[(y-2)*c_w]
-             -17*tmp[(y-1)*c_w]+114*tmp[y*c_w]+35*tmp[OC_MINI(y+1,c_h-1)*c_w]
-             -9*tmp[OC_MINI(y+2,c_h-1)*c_w]+tmp[(c_h-1)*c_w]+64)>>7,255);
+          for (; y < c_h; y++) {
+            _dst[y * c_w] = (unsigned char)OC_CLAMPI(0, (4 * tmp[(y - 2) * c_w]
+                                                         - 17 * tmp[(y - 1) * c_w] + 114 * tmp[y * c_w] + 35 * tmp[OC_MINI(y + 1, c_h - 1) * c_w]
+                                                         - 9 * tmp[OC_MINI(y + 2, c_h - 1) * c_w] + tmp[(c_h - 1) * c_w] + 64) >> 7, 255);
           }
           _dst++;
           tmp++;
         }
-      }break;
+      }
+      break;
     }
     /*For actual interlaced material, this would have to be done separately on
        each field, and the shift amounts would be different.
@@ -320,27 +328,27 @@
 /*Perform vertical filtering to reduce a single plane from 4:2:2 to 4:2:0.
   This is used as a helper by several converation routines.*/
 static void y4m_422jpeg_420jpeg_helper(unsigned char *_dst,
- const unsigned char *_src,int _c_w,int _c_h){
+                                       const unsigned char *_src, int _c_w, int _c_h) {
   int y;
   int x;
   /*Filter: [3 -17 78 78 -17 3]/128, derived from a 6-tap Lanczos window.*/
-  for(x=0;x<_c_w;x++){
-    for(y=0;y<OC_MINI(_c_h,2);y+=2){
-      _dst[(y>>1)*_c_w]=OC_CLAMPI(0,(64*_src[0]
-       +78*_src[OC_MINI(1,_c_h-1)*_c_w]
-       -17*_src[OC_MINI(2,_c_h-1)*_c_w]
-       +3*_src[OC_MINI(3,_c_h-1)*_c_w]+64)>>7,255);
+  for (x = 0; x < _c_w; x++) {
+    for (y = 0; y < OC_MINI(_c_h, 2); y += 2) {
+      _dst[(y >> 1)*_c_w] = OC_CLAMPI(0, (64 * _src[0]
+                                          + 78 * _src[OC_MINI(1, _c_h - 1) * _c_w]
+                                          - 17 * _src[OC_MINI(2, _c_h - 1) * _c_w]
+                                          + 3 * _src[OC_MINI(3, _c_h - 1) * _c_w] + 64) >> 7, 255);
     }
-    for(;y<_c_h-3;y+=2){
-      _dst[(y>>1)*_c_w]=OC_CLAMPI(0,(3*(_src[(y-2)*_c_w]+_src[(y+3)*_c_w])
-       -17*(_src[(y-1)*_c_w]+_src[(y+2)*_c_w])
-       +78*(_src[y*_c_w]+_src[(y+1)*_c_w])+64)>>7,255);
+    for (; y < _c_h - 3; y += 2) {
+      _dst[(y >> 1)*_c_w] = OC_CLAMPI(0, (3 * (_src[(y - 2) * _c_w] + _src[(y + 3) * _c_w])
+                                          - 17 * (_src[(y - 1) * _c_w] + _src[(y + 2) * _c_w])
+                                          + 78 * (_src[y * _c_w] + _src[(y + 1) * _c_w]) + 64) >> 7, 255);
     }
-    for(;y<_c_h;y+=2){
-      _dst[(y>>1)*_c_w]=OC_CLAMPI(0,(3*(_src[(y-2)*_c_w]
-       +_src[(_c_h-1)*_c_w])-17*(_src[(y-1)*_c_w]
-       +_src[OC_MINI(y+2,_c_h-1)*_c_w])
-       +78*(_src[y*_c_w]+_src[OC_MINI(y+1,_c_h-1)*_c_w])+64)>>7,255);
+    for (; y < _c_h; y += 2) {
+      _dst[(y >> 1)*_c_w] = OC_CLAMPI(0, (3 * (_src[(y - 2) * _c_w]
+                                               + _src[(_c_h - 1) * _c_w]) - 17 * (_src[(y - 1) * _c_w]
+                                                                                  + _src[OC_MINI(y + 2, _c_h - 1) * _c_w])
+                                          + 78 * (_src[y * _c_w] + _src[OC_MINI(y + 1, _c_h - 1) * _c_w]) + 64) >> 7, 255);
     }
     _src++;
     _dst++;
@@ -385,8 +393,8 @@
 
   We use a resampling filter to decimate the chroma planes by two in the
    vertical direction.*/
-static void y4m_convert_422jpeg_420jpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_422jpeg_420jpeg(y4m_input *_y4m, unsigned char *_dst,
+                                        unsigned char *_aux) {
   int c_w;
   int c_h;
   int c_sz;
@@ -395,18 +403,18 @@
   int dst_c_sz;
   int pli;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+_y4m->src_c_dec_h-1)/_y4m->src_c_dec_h;
-  c_h=_y4m->pic_h;
-  dst_c_w=(_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  dst_c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  dst_c_sz=dst_c_w*dst_c_h;
-  for(pli=1;pli<3;pli++){
-    y4m_422jpeg_420jpeg_helper(_dst,_aux,c_w,c_h);
-    _aux+=c_sz;
-    _dst+=dst_c_sz;
+  c_w = (_y4m->pic_w + _y4m->src_c_dec_h - 1) / _y4m->src_c_dec_h;
+  c_h = _y4m->pic_h;
+  dst_c_w = (_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  dst_c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  dst_c_sz = dst_c_w * dst_c_h;
+  for (pli = 1; pli < 3; pli++) {
+    y4m_422jpeg_420jpeg_helper(_dst, _aux, c_w, c_h);
+    _aux += c_sz;
+    _dst += dst_c_sz;
   }
 }
 
@@ -450,8 +458,8 @@
    pixel (at the original chroma resolution) to the right.
   Then we use a second resampling filter to decimate the chroma planes by two
    in the vertical direction.*/
-static void y4m_convert_422_420jpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_422_420jpeg(y4m_input *_y4m, unsigned char *_dst,
+                                    unsigned char *_aux) {
   unsigned char *tmp;
   int            c_w;
   int            c_h;
@@ -460,24 +468,24 @@
   int            dst_c_sz;
   int            pli;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+_y4m->src_c_dec_h-1)/_y4m->src_c_dec_h;
-  c_h=_y4m->pic_h;
-  dst_c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  dst_c_sz=c_w*dst_c_h;
-  tmp=_aux+2*c_sz;
-  for(pli=1;pli<3;pli++){
+  c_w = (_y4m->pic_w + _y4m->src_c_dec_h - 1) / _y4m->src_c_dec_h;
+  c_h = _y4m->pic_h;
+  dst_c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  dst_c_sz = c_w * dst_c_h;
+  tmp = _aux + 2 * c_sz;
+  for (pli = 1; pli < 3; pli++) {
     /*In reality, the horizontal and vertical steps could be pipelined, for
        less memory consumption and better cache performance, but we do them
        separately for simplicity.*/
     /*First do horizontal filtering (convert to 422jpeg)*/
-    y4m_42xmpeg2_42xjpeg_helper(tmp,_aux,c_w,c_h);
+    y4m_42xmpeg2_42xjpeg_helper(tmp, _aux, c_w, c_h);
     /*Now do the vertical filtering.*/
-    y4m_422jpeg_420jpeg_helper(_dst,tmp,c_w,c_h);
-    _aux+=c_sz;
-    _dst+=dst_c_sz;
+    y4m_422jpeg_420jpeg_helper(_dst, tmp, c_w, c_h);
+    _aux += c_sz;
+    _dst += dst_c_sz;
   }
 }
 
@@ -522,8 +530,8 @@
    right.
   Then we use another filter to decimate the planes by 2 in the vertical
    direction.*/
-static void y4m_convert_411_420jpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_411_420jpeg(y4m_input *_y4m, unsigned char *_dst,
+                                    unsigned char *_aux) {
   unsigned char *tmp;
   int            c_w;
   int            c_h;
@@ -536,57 +544,57 @@
   int            y;
   int            x;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+_y4m->src_c_dec_h-1)/_y4m->src_c_dec_h;
-  c_h=_y4m->pic_h;
-  dst_c_w=(_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  dst_c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  dst_c_sz=dst_c_w*dst_c_h;
-  tmp_sz=dst_c_w*c_h;
-  tmp=_aux+2*c_sz;
-  for(pli=1;pli<3;pli++){
+  c_w = (_y4m->pic_w + _y4m->src_c_dec_h - 1) / _y4m->src_c_dec_h;
+  c_h = _y4m->pic_h;
+  dst_c_w = (_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  dst_c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  dst_c_sz = dst_c_w * dst_c_h;
+  tmp_sz = dst_c_w * c_h;
+  tmp = _aux + 2 * c_sz;
+  for (pli = 1; pli < 3; pli++) {
     /*In reality, the horizontal and vertical steps could be pipelined, for
        less memory consumption and better cache performance, but we do them
        separately for simplicity.*/
     /*First do horizontal filtering (convert to 422jpeg)*/
-    for(y=0;y<c_h;y++){
+    for (y = 0; y < c_h; y++) {
       /*Filters: [1 110 18 -1]/128 and [-3 50 86 -5]/128, both derived from a
          4-tap Mitchell window.*/
-      for(x=0;x<OC_MINI(c_w,1);x++){
-        tmp[x<<1]=(unsigned char)OC_CLAMPI(0,(111*_aux[0]
-         +18*_aux[OC_MINI(1,c_w-1)]-_aux[OC_MINI(2,c_w-1)]+64)>>7,255);
-        tmp[x<<1|1]=(unsigned char)OC_CLAMPI(0,(47*_aux[0]
-         +86*_aux[OC_MINI(1,c_w-1)]-5*_aux[OC_MINI(2,c_w-1)]+64)>>7,255);
+      for (x = 0; x < OC_MINI(c_w, 1); x++) {
+        tmp[x << 1] = (unsigned char)OC_CLAMPI(0, (111 * _aux[0]
+                                                   + 18 * _aux[OC_MINI(1, c_w - 1)] - _aux[OC_MINI(2, c_w - 1)] + 64) >> 7, 255);
+        tmp[x << 1 | 1] = (unsigned char)OC_CLAMPI(0, (47 * _aux[0]
+                                                       + 86 * _aux[OC_MINI(1, c_w - 1)] - 5 * _aux[OC_MINI(2, c_w - 1)] + 64) >> 7, 255);
       }
-      for(;x<c_w-2;x++){
-        tmp[x<<1]=(unsigned char)OC_CLAMPI(0,(_aux[x-1]+110*_aux[x]
-         +18*_aux[x+1]-_aux[x+2]+64)>>7,255);
-        tmp[x<<1|1]=(unsigned char)OC_CLAMPI(0,(-3*_aux[x-1]+50*_aux[x]
-         +86*_aux[x+1]-5*_aux[x+2]+64)>>7,255);
+      for (; x < c_w - 2; x++) {
+        tmp[x << 1] = (unsigned char)OC_CLAMPI(0, (_aux[x - 1] + 110 * _aux[x]
+                                                   + 18 * _aux[x + 1] - _aux[x + 2] + 64) >> 7, 255);
+        tmp[x << 1 | 1] = (unsigned char)OC_CLAMPI(0, (-3 * _aux[x - 1] + 50 * _aux[x]
+                                                       + 86 * _aux[x + 1] - 5 * _aux[x + 2] + 64) >> 7, 255);
       }
-      for(;x<c_w;x++){
-        tmp[x<<1]=(unsigned char)OC_CLAMPI(0,(_aux[x-1]+110*_aux[x]
-         +18*_aux[OC_MINI(x+1,c_w-1)]-_aux[c_w-1]+64)>>7,255);
-        if((x<<1|1)<dst_c_w){
-          tmp[x<<1|1]=(unsigned char)OC_CLAMPI(0,(-3*_aux[x-1]+50*_aux[x]
-           +86*_aux[OC_MINI(x+1,c_w-1)]-5*_aux[c_w-1]+64)>>7,255);
+      for (; x < c_w; x++) {
+        tmp[x << 1] = (unsigned char)OC_CLAMPI(0, (_aux[x - 1] + 110 * _aux[x]
+                                                   + 18 * _aux[OC_MINI(x + 1, c_w - 1)] - _aux[c_w - 1] + 64) >> 7, 255);
+        if ((x << 1 | 1) < dst_c_w) {
+          tmp[x << 1 | 1] = (unsigned char)OC_CLAMPI(0, (-3 * _aux[x - 1] + 50 * _aux[x]
+                                                         + 86 * _aux[OC_MINI(x + 1, c_w - 1)] - 5 * _aux[c_w - 1] + 64) >> 7, 255);
         }
       }
-      tmp+=dst_c_w;
-      _aux+=c_w;
+      tmp += dst_c_w;
+      _aux += c_w;
     }
-    tmp-=tmp_sz;
+    tmp -= tmp_sz;
     /*Now do the vertical filtering.*/
-    y4m_422jpeg_420jpeg_helper(_dst,tmp,dst_c_w,c_h);
-    _dst+=dst_c_sz;
+    y4m_422jpeg_420jpeg_helper(_dst, tmp, dst_c_w, c_h);
+    _dst += dst_c_sz;
   }
 }
 
 /*Convert 444 to 420jpeg.*/
-static void y4m_convert_444_420jpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_444_420jpeg(y4m_input *_y4m, unsigned char *_dst,
+                                    unsigned char *_aux) {
   unsigned char *tmp;
   int            c_w;
   int            c_h;
@@ -599,218 +607,247 @@
   int            y;
   int            x;
   /*Skip past the luma data.*/
-  _dst+=_y4m->pic_w*_y4m->pic_h;
+  _dst += _y4m->pic_w * _y4m->pic_h;
   /*Compute the size of each chroma plane.*/
-  c_w=(_y4m->pic_w+_y4m->src_c_dec_h-1)/_y4m->src_c_dec_h;
-  c_h=_y4m->pic_h;
-  dst_c_w=(_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  dst_c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  dst_c_sz=dst_c_w*dst_c_h;
-  tmp_sz=dst_c_w*c_h;
-  tmp=_aux+2*c_sz;
-  for(pli=1;pli<3;pli++){
+  c_w = (_y4m->pic_w + _y4m->src_c_dec_h - 1) / _y4m->src_c_dec_h;
+  c_h = _y4m->pic_h;
+  dst_c_w = (_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  dst_c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  dst_c_sz = dst_c_w * dst_c_h;
+  tmp_sz = dst_c_w * c_h;
+  tmp = _aux + 2 * c_sz;
+  for (pli = 1; pli < 3; pli++) {
     /*Filter: [3 -17 78 78 -17 3]/128, derived from a 6-tap Lanczos window.*/
-    for(y=0;y<c_h;y++){
-      for(x=0;x<OC_MINI(c_w,2);x+=2){
-        tmp[x>>1]=OC_CLAMPI(0,(64*_aux[0]+78*_aux[OC_MINI(1,c_w-1)]
-         -17*_aux[OC_MINI(2,c_w-1)]
-         +3*_aux[OC_MINI(3,c_w-1)]+64)>>7,255);
+    for (y = 0; y < c_h; y++) {
+      for (x = 0; x < OC_MINI(c_w, 2); x += 2) {
+        tmp[x >> 1] = OC_CLAMPI(0, (64 * _aux[0] + 78 * _aux[OC_MINI(1, c_w - 1)]
+                                    - 17 * _aux[OC_MINI(2, c_w - 1)]
+                                    + 3 * _aux[OC_MINI(3, c_w - 1)] + 64) >> 7, 255);
       }
-      for(;x<c_w-3;x+=2){
-        tmp[x>>1]=OC_CLAMPI(0,(3*(_aux[x-2]+_aux[x+3])
-         -17*(_aux[x-1]+_aux[x+2])+78*(_aux[x]+_aux[x+1])+64)>>7,255);
+      for (; x < c_w - 3; x += 2) {
+        tmp[x >> 1] = OC_CLAMPI(0, (3 * (_aux[x - 2] + _aux[x + 3])
+                                    - 17 * (_aux[x - 1] + _aux[x + 2]) + 78 * (_aux[x] + _aux[x + 1]) + 64) >> 7, 255);
       }
-      for(;x<c_w;x+=2){
-        tmp[x>>1]=OC_CLAMPI(0,(3*(_aux[x-2]+_aux[c_w-1])-
-         17*(_aux[x-1]+_aux[OC_MINI(x+2,c_w-1)])+
-         78*(_aux[x]+_aux[OC_MINI(x+1,c_w-1)])+64)>>7,255);
+      for (; x < c_w; x += 2) {
+        tmp[x >> 1] = OC_CLAMPI(0, (3 * (_aux[x - 2] + _aux[c_w - 1]) -
+                                    17 * (_aux[x - 1] + _aux[OC_MINI(x + 2, c_w - 1)]) +
+                                    78 * (_aux[x] + _aux[OC_MINI(x + 1, c_w - 1)]) + 64) >> 7, 255);
       }
-      tmp+=dst_c_w;
-      _aux+=c_w;
+      tmp += dst_c_w;
+      _aux += c_w;
     }
-    tmp-=tmp_sz;
+    tmp -= tmp_sz;
     /*Now do the vertical filtering.*/
-    y4m_422jpeg_420jpeg_helper(_dst,tmp,dst_c_w,c_h);
-    _dst+=dst_c_sz;
+    y4m_422jpeg_420jpeg_helper(_dst, tmp, dst_c_w, c_h);
+    _dst += dst_c_sz;
   }
 }
 
 /*The image is padded with empty chroma components at 4:2:0.*/
-static void y4m_convert_mono_420jpeg(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_mono_420jpeg(y4m_input *_y4m, unsigned char *_dst,
+                                     unsigned char *_aux) {
   int c_sz;
-  _dst+=_y4m->pic_w*_y4m->pic_h;
-  c_sz=((_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h)*
-   ((_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v);
-  memset(_dst,128,c_sz*2);
+  _dst += _y4m->pic_w * _y4m->pic_h;
+  c_sz = ((_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h) *
+         ((_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v);
+  memset(_dst, 128, c_sz * 2);
 }
 
 /*No conversion function needed.*/
-static void y4m_convert_null(y4m_input *_y4m,unsigned char *_dst,
- unsigned char *_aux){
+static void y4m_convert_null(y4m_input *_y4m, unsigned char *_dst,
+                             unsigned char *_aux) {
 }
 
-int y4m_input_open(y4m_input *_y4m,FILE *_fin,char *_skip,int _nskip){
+int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
+                   int only_420) {
   char buffer[80];
   int  ret;
   int  i;
   /*Read until newline, or 80 cols, whichever happens first.*/
-  for(i=0;i<79;i++){
-    if(_nskip>0){
-      buffer[i]=*_skip++;
+  for (i = 0; i < 79; i++) {
+    if (_nskip > 0) {
+      buffer[i] = *_skip++;
       _nskip--;
+    } else {
+      ret = (int)fread(buffer + i, 1, 1, _fin);
+      if (ret < 1)return -1;
     }
-    else{
-      ret=(int)fread(buffer+i,1,1,_fin);
-      if(ret<1)return -1;
-    }
-    if(buffer[i]=='\n')break;
+    if (buffer[i] == '\n')break;
   }
   /*We skipped too much header data.*/
-  if(_nskip>0)return -1;
-  if(i==79){
-    fprintf(stderr,"Error parsing header; not a YUV2MPEG2 file?\n");
+  if (_nskip > 0)return -1;
+  if (i == 79) {
+    fprintf(stderr, "Error parsing header; not a YUV2MPEG2 file?\n");
     return -1;
   }
-  buffer[i]='\0';
-  if(memcmp(buffer,"YUV4MPEG",8)){
-    fprintf(stderr,"Incomplete magic for YUV4MPEG file.\n");
+  buffer[i] = '\0';
+  if (memcmp(buffer, "YUV4MPEG", 8)) {
+    fprintf(stderr, "Incomplete magic for YUV4MPEG file.\n");
     return -1;
   }
-  if(buffer[8]!='2'){
-    fprintf(stderr,"Incorrect YUV input file version; YUV4MPEG2 required.\n");
+  if (buffer[8] != '2') {
+    fprintf(stderr, "Incorrect YUV input file version; YUV4MPEG2 required.\n");
   }
-  ret=y4m_parse_tags(_y4m,buffer+5);
-  if(ret<0){
-    fprintf(stderr,"Error parsing YUV4MPEG2 header.\n");
+  ret = y4m_parse_tags(_y4m, buffer + 5);
+  if (ret < 0) {
+    fprintf(stderr, "Error parsing YUV4MPEG2 header.\n");
     return ret;
   }
-  if(_y4m->interlace=='?'){
-    fprintf(stderr,"Warning: Input video interlacing format unknown; "
-     "assuming progressive scan.\n");
-  }
-  else if(_y4m->interlace!='p'){
-    fprintf(stderr,"Input video is interlaced; "
-     "Only progressive scan handled.\n");
+  if (_y4m->interlace == '?') {
+    fprintf(stderr, "Warning: Input video interlacing format unknown; "
+            "assuming progressive scan.\n");
+  } else if (_y4m->interlace != 'p') {
+    fprintf(stderr, "Input video is interlaced; "
+            "Only progressive scan handled.\n");
     return -1;
   }
-  if(strcmp(_y4m->chroma_type,"420")==0||
-   strcmp(_y4m->chroma_type,"420jpeg")==0){
-    _y4m->src_c_dec_h=_y4m->dst_c_dec_h=_y4m->src_c_dec_v=_y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h
-     +2*((_y4m->pic_w+1)/2)*((_y4m->pic_h+1)/2);
+  _y4m->vpx_fmt = VPX_IMG_FMT_I420;
+  _y4m->vpx_bps = 12;
+  if (strcmp(_y4m->chroma_type, "420") == 0 ||
+      strcmp(_y4m->chroma_type, "420jpeg") == 0) {
+    _y4m->src_c_dec_h = _y4m->dst_c_dec_h = _y4m->src_c_dec_v = _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h
+                            + 2 * ((_y4m->pic_w + 1) / 2) * ((_y4m->pic_h + 1) / 2);
     /*Natively supported: no conversion required.*/
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz=0;
-    _y4m->convert=y4m_convert_null;
-  }
-  else if(strcmp(_y4m->chroma_type,"420mpeg2")==0){
-    _y4m->src_c_dec_h=_y4m->dst_c_dec_h=_y4m->src_c_dec_v=_y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
+    _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
+    _y4m->convert = y4m_convert_null;
+  } else if (strcmp(_y4m->chroma_type, "420mpeg2") == 0) {
+    _y4m->src_c_dec_h = _y4m->dst_c_dec_h = _y4m->src_c_dec_v = _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
     /*Chroma filter required: read into the aux buf first.*/
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz=
-     2*((_y4m->pic_w+1)/2)*((_y4m->pic_h+1)/2);
-    _y4m->convert=y4m_convert_42xmpeg2_42xjpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"420paldv")==0){
-    _y4m->src_c_dec_h=_y4m->dst_c_dec_h=_y4m->src_c_dec_v=_y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
+    _y4m->aux_buf_sz = _y4m->aux_buf_read_sz =
+                         2 * ((_y4m->pic_w + 1) / 2) * ((_y4m->pic_h + 1) / 2);
+    _y4m->convert = y4m_convert_42xmpeg2_42xjpeg;
+  } else if (strcmp(_y4m->chroma_type, "420paldv") == 0) {
+    _y4m->src_c_dec_h = _y4m->dst_c_dec_h = _y4m->src_c_dec_v = _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
     /*Chroma filter required: read into the aux buf first.
       We need to make two filter passes, so we need some extra space in the
        aux buffer.*/
-    _y4m->aux_buf_sz=3*((_y4m->pic_w+1)/2)*((_y4m->pic_h+1)/2);
-    _y4m->aux_buf_read_sz=2*((_y4m->pic_w+1)/2)*((_y4m->pic_h+1)/2);
-    _y4m->convert=y4m_convert_42xpaldv_42xjpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"422jpeg")==0){
-    _y4m->src_c_dec_h=_y4m->dst_c_dec_h=2;
-    _y4m->src_c_dec_v=1;
-    _y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
+    _y4m->aux_buf_sz = 3 * ((_y4m->pic_w + 1) / 2) * ((_y4m->pic_h + 1) / 2);
+    _y4m->aux_buf_read_sz = 2 * ((_y4m->pic_w + 1) / 2) * ((_y4m->pic_h + 1) / 2);
+    _y4m->convert = y4m_convert_42xpaldv_42xjpeg;
+  } else if (strcmp(_y4m->chroma_type, "422jpeg") == 0) {
+    _y4m->src_c_dec_h = _y4m->dst_c_dec_h = 2;
+    _y4m->src_c_dec_v = 1;
+    _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
     /*Chroma filter required: read into the aux buf first.*/
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz=2*((_y4m->pic_w+1)/2)*_y4m->pic_h;
-    _y4m->convert=y4m_convert_422jpeg_420jpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"422")==0){
-    _y4m->src_c_dec_h=_y4m->dst_c_dec_h=2;
-    _y4m->src_c_dec_v=1;
-    _y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
+    _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 2 * ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+    _y4m->convert = y4m_convert_422jpeg_420jpeg;
+  } else if (strcmp(_y4m->chroma_type, "422") == 0) {
+    _y4m->src_c_dec_h = 2;
+    _y4m->src_c_dec_v = 1;
+    if (only_420) {
+      _y4m->dst_c_dec_h = 2;
+      _y4m->dst_c_dec_v = 2;
+      _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
+      /*Chroma filter required: read into the aux buf first.
+        We need to make two filter passes, so we need some extra space in the
+         aux buffer.*/
+      _y4m->aux_buf_read_sz = 2 * ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz +
+          ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+      _y4m->convert = y4m_convert_422_420jpeg;
+    } else {
+      _y4m->vpx_fmt = VPX_IMG_FMT_I422;
+      _y4m->vpx_bps = 16;
+      _y4m->dst_c_dec_h = _y4m->src_c_dec_h;
+      _y4m->dst_c_dec_v = _y4m->src_c_dec_v;
+      _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h
+                              + 2 * ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+      /*Natively supported: no conversion required.*/
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
+      _y4m->convert = y4m_convert_null;
+      }
+  } else if (strcmp(_y4m->chroma_type, "411") == 0) {
+    _y4m->src_c_dec_h = 4;
+    _y4m->dst_c_dec_h = 2;
+    _y4m->src_c_dec_v = 1;
+    _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
     /*Chroma filter required: read into the aux buf first.
       We need to make two filter passes, so we need some extra space in the
        aux buffer.*/
-    _y4m->aux_buf_read_sz=2*((_y4m->pic_w+1)/2)*_y4m->pic_h;
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz+((_y4m->pic_w+1)/2)*_y4m->pic_h;
-    _y4m->convert=y4m_convert_422_420jpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"411")==0){
-    _y4m->src_c_dec_h=4;
-    _y4m->dst_c_dec_h=2;
-    _y4m->src_c_dec_v=1;
-    _y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
-    /*Chroma filter required: read into the aux buf first.
-      We need to make two filter passes, so we need some extra space in the
-       aux buffer.*/
-    _y4m->aux_buf_read_sz=2*((_y4m->pic_w+3)/4)*_y4m->pic_h;
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz+((_y4m->pic_w+1)/2)*_y4m->pic_h;
-    _y4m->convert=y4m_convert_411_420jpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"444")==0){
-    _y4m->src_c_dec_h=1;
-    _y4m->dst_c_dec_h=2;
-    _y4m->src_c_dec_v=1;
-    _y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
-    /*Chroma filter required: read into the aux buf first.
-      We need to make two filter passes, so we need some extra space in the
-       aux buffer.*/
-    _y4m->aux_buf_read_sz=2*_y4m->pic_w*_y4m->pic_h;
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz+((_y4m->pic_w+1)/2)*_y4m->pic_h;
-    _y4m->convert=y4m_convert_444_420jpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"444alpha")==0){
-    _y4m->src_c_dec_h=1;
-    _y4m->dst_c_dec_h=2;
-    _y4m->src_c_dec_v=1;
-    _y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
-    /*Chroma filter required: read into the aux buf first.
-      We need to make two filter passes, so we need some extra space in the
-       aux buffer.
-      The extra plane also gets read into the aux buf.
-      It will be discarded.*/
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz=3*_y4m->pic_w*_y4m->pic_h;
-    _y4m->convert=y4m_convert_444_420jpeg;
-  }
-  else if(strcmp(_y4m->chroma_type,"mono")==0){
-    _y4m->src_c_dec_h=_y4m->src_c_dec_v=0;
-    _y4m->dst_c_dec_h=_y4m->dst_c_dec_v=2;
-    _y4m->dst_buf_read_sz=_y4m->pic_w*_y4m->pic_h;
+    _y4m->aux_buf_read_sz = 2 * ((_y4m->pic_w + 3) / 4) * _y4m->pic_h;
+    _y4m->aux_buf_sz = _y4m->aux_buf_read_sz + ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+    _y4m->convert = y4m_convert_411_420jpeg;
+  } else if (strcmp(_y4m->chroma_type, "444") == 0) {
+    _y4m->src_c_dec_h = 1;
+    _y4m->src_c_dec_v = 1;
+    if (only_420) {
+      _y4m->dst_c_dec_h = 2;
+      _y4m->dst_c_dec_v = 2;
+      _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
+      /*Chroma filter required: read into the aux buf first.
+        We need to make two filter passes, so we need some extra space in the
+         aux buffer.*/
+      _y4m->aux_buf_read_sz = 2 * _y4m->pic_w * _y4m->pic_h;
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz +
+          ((_y4m->pic_w + 1) / 2) * _y4m->pic_h;
+      _y4m->convert = y4m_convert_444_420jpeg;
+    } else {
+      _y4m->vpx_fmt = VPX_IMG_FMT_I444;
+      _y4m->vpx_bps = 24;
+      _y4m->dst_c_dec_h = _y4m->src_c_dec_h;
+      _y4m->dst_c_dec_v = _y4m->src_c_dec_v;
+      _y4m->dst_buf_read_sz = 3 * _y4m->pic_w * _y4m->pic_h;
+      /*Natively supported: no conversion required.*/
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
+      _y4m->convert = y4m_convert_null;
+    }
+  } else if (strcmp(_y4m->chroma_type, "444alpha") == 0) {
+    _y4m->src_c_dec_h = 1;
+    _y4m->src_c_dec_v = 1;
+    if (only_420) {
+      _y4m->dst_c_dec_h = 2;
+      _y4m->dst_c_dec_v = 2;
+      _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
+      /*Chroma filter required: read into the aux buf first.
+        We need to make two filter passes, so we need some extra space in the
+         aux buffer.
+        The extra plane also gets read into the aux buf.
+        It will be discarded.*/
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 3 * _y4m->pic_w * _y4m->pic_h;
+      _y4m->convert = y4m_convert_444_420jpeg;
+    } else {
+      _y4m->vpx_fmt = VPX_IMG_FMT_444A;
+      _y4m->vpx_bps = 32;
+      _y4m->dst_c_dec_h = _y4m->src_c_dec_h;
+      _y4m->dst_c_dec_v = _y4m->src_c_dec_v;
+      _y4m->dst_buf_read_sz = 4 * _y4m->pic_w * _y4m->pic_h;
+      /*Natively supported: no conversion required.*/
+      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
+      _y4m->convert = y4m_convert_null;
+    }
+  } else if (strcmp(_y4m->chroma_type, "mono") == 0) {
+    _y4m->src_c_dec_h = _y4m->src_c_dec_v = 0;
+    _y4m->dst_c_dec_h = _y4m->dst_c_dec_v = 2;
+    _y4m->dst_buf_read_sz = _y4m->pic_w * _y4m->pic_h;
     /*No extra space required, but we need to clear the chroma planes.*/
-    _y4m->aux_buf_sz=_y4m->aux_buf_read_sz=0;
-    _y4m->convert=y4m_convert_mono_420jpeg;
-  }
-  else{
-    fprintf(stderr,"Unknown chroma sampling type: %s\n",_y4m->chroma_type);
+    _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
+    _y4m->convert = y4m_convert_mono_420jpeg;
+  } else {
+    fprintf(stderr, "Unknown chroma sampling type: %s\n", _y4m->chroma_type);
     return -1;
   }
   /*The size of the final frame buffers is always computed from the
      destination chroma decimation type.*/
-  _y4m->dst_buf_sz=_y4m->pic_w*_y4m->pic_h
-   +2*((_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h)*
-   ((_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v);
-  _y4m->dst_buf=(unsigned char *)malloc(_y4m->dst_buf_sz);
-  _y4m->aux_buf=(unsigned char *)malloc(_y4m->aux_buf_sz);
+  _y4m->dst_buf_sz = _y4m->pic_w * _y4m->pic_h
+                     + 2 * ((_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h) *
+                     ((_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v);
+  _y4m->dst_buf = (unsigned char *)malloc(_y4m->dst_buf_sz);
+  _y4m->aux_buf = (unsigned char *)malloc(_y4m->aux_buf_sz);
   return 0;
 }
 
-void y4m_input_close(y4m_input *_y4m){
+void y4m_input_close(y4m_input *_y4m) {
   free(_y4m->dst_buf);
   free(_y4m->aux_buf);
 }
 
-int y4m_input_fetch_frame(y4m_input *_y4m,FILE *_fin,vpx_image_t *_img){
+int y4m_input_fetch_frame(y4m_input *_y4m, FILE *_fin, vpx_image_t *_img) {
   char frame[6];
   int  pic_sz;
   int  c_w;
@@ -818,54 +855,55 @@
   int  c_sz;
   int  ret;
   /*Read and skip the frame header.*/
-  ret=(int)fread(frame,1,6,_fin);
-  if(ret<6)return 0;
-  if(memcmp(frame,"FRAME",5)){
-    fprintf(stderr,"Loss of framing in Y4M input data\n");
+  ret = (int)fread(frame, 1, 6, _fin);
+  if (ret < 6)return 0;
+  if (memcmp(frame, "FRAME", 5)) {
+    fprintf(stderr, "Loss of framing in Y4M input data\n");
     return -1;
   }
-  if(frame[5]!='\n'){
+  if (frame[5] != '\n') {
     char c;
     int  j;
-    for(j=0;j<79&&fread(&c,1,1,_fin)&&c!='\n';j++);
-    if(j==79){
-      fprintf(stderr,"Error parsing Y4M frame header\n");
+    for (j = 0; j < 79 && fread(&c, 1, 1, _fin) && c != '\n'; j++);
+    if (j == 79) {
+      fprintf(stderr, "Error parsing Y4M frame header\n");
       return -1;
     }
   }
   /*Read the frame data that needs no conversion.*/
-  if(fread(_y4m->dst_buf,1,_y4m->dst_buf_read_sz,_fin)!=_y4m->dst_buf_read_sz){
-    fprintf(stderr,"Error reading Y4M frame data.\n");
+  if (fread(_y4m->dst_buf, 1, _y4m->dst_buf_read_sz, _fin) != _y4m->dst_buf_read_sz) {
+    fprintf(stderr, "Error reading Y4M frame data.\n");
     return -1;
   }
   /*Read the frame data that does need conversion.*/
-  if(fread(_y4m->aux_buf,1,_y4m->aux_buf_read_sz,_fin)!=_y4m->aux_buf_read_sz){
-    fprintf(stderr,"Error reading Y4M frame data.\n");
+  if (fread(_y4m->aux_buf, 1, _y4m->aux_buf_read_sz, _fin) != _y4m->aux_buf_read_sz) {
+    fprintf(stderr, "Error reading Y4M frame data.\n");
     return -1;
   }
   /*Now convert the just read frame.*/
-  (*_y4m->convert)(_y4m,_y4m->dst_buf,_y4m->aux_buf);
+  (*_y4m->convert)(_y4m, _y4m->dst_buf, _y4m->aux_buf);
   /*Fill in the frame buffer pointers.
     We don't use vpx_img_wrap() because it forces padding for odd picture
      sizes, which would require a separate fread call for every row.*/
-  memset(_img,0,sizeof(*_img));
+  memset(_img, 0, sizeof(*_img));
   /*Y4M has the planes in Y'CbCr order, which libvpx calls Y, U, and V.*/
-  _img->fmt=IMG_FMT_I420;
-  _img->w=_img->d_w=_y4m->pic_w;
-  _img->h=_img->d_h=_y4m->pic_h;
-  /*This is hard-coded to 4:2:0 for now, as that's all VP8 supports.*/
-  _img->x_chroma_shift=1;
-  _img->y_chroma_shift=1;
-  _img->bps=12;
+  _img->fmt = _y4m->vpx_fmt;
+  _img->w = _img->d_w = _y4m->pic_w;
+  _img->h = _img->d_h = _y4m->pic_h;
+  _img->x_chroma_shift = _y4m->dst_c_dec_h >> 1;
+  _img->y_chroma_shift = _y4m->dst_c_dec_v >> 1;
+  _img->bps = _y4m->vpx_bps;
+
   /*Set up the buffer pointers.*/
-  pic_sz=_y4m->pic_w*_y4m->pic_h;
-  c_w=(_y4m->pic_w+_y4m->dst_c_dec_h-1)/_y4m->dst_c_dec_h;
-  c_h=(_y4m->pic_h+_y4m->dst_c_dec_v-1)/_y4m->dst_c_dec_v;
-  c_sz=c_w*c_h;
-  _img->stride[PLANE_Y]=_y4m->pic_w;
-  _img->stride[PLANE_U]=_img->stride[PLANE_V]=c_w;
-  _img->planes[PLANE_Y]=_y4m->dst_buf;
-  _img->planes[PLANE_U]=_y4m->dst_buf+pic_sz;
-  _img->planes[PLANE_V]=_y4m->dst_buf+pic_sz+c_sz;
+  pic_sz = _y4m->pic_w * _y4m->pic_h;
+  c_w = (_y4m->pic_w + _y4m->dst_c_dec_h - 1) / _y4m->dst_c_dec_h;
+  c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
+  c_sz = c_w * c_h;
+  _img->stride[PLANE_Y] = _img->stride[PLANE_ALPHA] = _y4m->pic_w;
+  _img->stride[PLANE_U] = _img->stride[PLANE_V] = c_w;
+  _img->planes[PLANE_Y] = _y4m->dst_buf;
+  _img->planes[PLANE_U] = _y4m->dst_buf + pic_sz;
+  _img->planes[PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
+  _img->planes[PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
   return 1;
 }
diff --git a/libvpx/y4minput.h b/libvpx/y4minput.h
index 1a01bcd..b2a390c 100644
--- a/libvpx/y4minput.h
+++ b/libvpx/y4minput.h
@@ -23,11 +23,11 @@
 
 /*The function used to perform chroma conversion.*/
 typedef void (*y4m_convert_func)(y4m_input *_y4m,
- unsigned char *_dst,unsigned char *_src);
+                                 unsigned char *_dst, unsigned char *_src);
 
 
 
-struct y4m_input{
+struct y4m_input {
   int               pic_w;
   int               pic_h;
   int               fps_n;
@@ -51,10 +51,13 @@
   y4m_convert_func  convert;
   unsigned char    *dst_buf;
   unsigned char    *aux_buf;
+  enum vpx_img_fmt  vpx_fmt;
+  int               vpx_bps;
 };
 
-int y4m_input_open(y4m_input *_y4m,FILE *_fin,char *_skip,int _nskip);
+int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
+                   int only_420);
 void y4m_input_close(y4m_input *_y4m);
-int y4m_input_fetch_frame(y4m_input *_y4m,FILE *_fin,vpx_image_t *img);
+int y4m_input_fetch_frame(y4m_input *_y4m, FILE *_fin, vpx_image_t *img);
 
 #endif
diff --git a/mips-dspr2/libvpx_srcs.txt b/mips-dspr2/libvpx_srcs.txt
index 8b5b727..e74102e 100644
--- a/mips-dspr2/libvpx_srcs.txt
+++ b/mips-dspr2/libvpx_srcs.txt
@@ -4,7 +4,6 @@
 libs.mk
 vp8/common/alloccommon.c
 vp8/common/alloccommon.h
-vp8/common/asm_com_offsets.c
 vp8/common/blockd.c
 vp8/common/blockd.h
 vp8/common/coefupdateprobs.h
@@ -67,8 +66,8 @@
 vp8/common/treecoder.h
 vp8/common/variance_c.c
 vp8/common/variance.h
+vp8/common/vp8_asm_com_offsets.c
 vp8/common/vp8_entropymodedata.h
-vp8/decoder/asm_dec_offsets.c
 vp8/decoder/dboolhuff.c
 vp8/decoder/dboolhuff.h
 vp8/decoder/decodemv.c
@@ -81,7 +80,7 @@
 vp8/decoder/onyxd_int.h
 vp8/decoder/threading.c
 vp8/decoder/treereader.h
-vp8/encoder/asm_enc_offsets.c
+vp8/decoder/vp8_asm_dec_offsets.c
 vp8/encoder/bitstream.c
 vp8/encoder/bitstream.h
 vp8/encoder/block.h
@@ -128,11 +127,91 @@
 vp8/encoder/tokenize.h
 vp8/encoder/treewriter.c
 vp8/encoder/treewriter.h
+vp8/encoder/vp8_asm_enc_offsets.c
 vp8/vp8_common.mk
 vp8/vp8_cx_iface.c
 vp8/vp8cx.mk
 vp8/vp8_dx_iface.c
 vp8/vp8dx.mk
+vp9/common/generic/vp9_systemdependent.c
+vp9/common/vp9_alloccommon.c
+vp9/common/vp9_alloccommon.h
+vp9/common/vp9_asm_com_offsets.c
+vp9/common/vp9_blockd.h
+vp9/common/vp9_common.h
+vp9/common/vp9_convolve.c
+vp9/common/vp9_convolve.h
+vp9/common/vp9_debugmodes.c
+vp9/common/vp9_default_coef_probs.h
+vp9/common/vp9_entropy.c
+vp9/common/vp9_entropy.h
+vp9/common/vp9_entropymode.c
+vp9/common/vp9_entropymode.h
+vp9/common/vp9_entropymv.c
+vp9/common/vp9_entropymv.h
+vp9/common/vp9_enums.h
+vp9/common/vp9_extend.c
+vp9/common/vp9_extend.h
+vp9/common/vp9_filter.c
+vp9/common/vp9_filter.h
+vp9/common/vp9_findnearmv.c
+vp9/common/vp9_findnearmv.h
+vp9/common/vp9_idct.c
+vp9/common/vp9_idct.h
+vp9/common/vp9_loopfilter.c
+vp9/common/vp9_loopfilter_filters.c
+vp9/common/vp9_loopfilter.h
+vp9/common/vp9_mbpitch.c
+vp9/common/vp9_modecont.c
+vp9/common/vp9_modecontext.c
+vp9/common/vp9_modecont.h
+vp9/common/vp9_mv.h
+vp9/common/vp9_mvref_common.c
+vp9/common/vp9_mvref_common.h
+vp9/common/vp9_onyxc_int.h
+vp9/common/vp9_onyx.h
+vp9/common/vp9_ppflags.h
+vp9/common/vp9_pragmas.h
+vp9/common/vp9_pred_common.c
+vp9/common/vp9_pred_common.h
+vp9/common/vp9_quant_common.c
+vp9/common/vp9_quant_common.h
+vp9/common/vp9_reconinter.c
+vp9/common/vp9_reconinter.h
+vp9/common/vp9_reconintra.c
+vp9/common/vp9_reconintra.h
+vp9/common/vp9_rtcd.c
+vp9/common/vp9_rtcd_defs.sh
+vp9/common/vp9_sadmxn.h
+vp9/common/vp9_seg_common.c
+vp9/common/vp9_seg_common.h
+vp9/common/vp9_subpelvar.h
+vp9/common/vp9_systemdependent.h
+vp9/common/vp9_textblit.h
+vp9/common/vp9_tile_common.c
+vp9/common/vp9_tile_common.h
+vp9/common/vp9_treecoder.c
+vp9/common/vp9_treecoder.h
+vp9/decoder/vp9_asm_dec_offsets.c
+vp9/decoder/vp9_dboolhuff.c
+vp9/decoder/vp9_dboolhuff.h
+vp9/decoder/vp9_decodemv.c
+vp9/decoder/vp9_decodemv.h
+vp9/decoder/vp9_decodframe.c
+vp9/decoder/vp9_decodframe.h
+vp9/decoder/vp9_detokenize.c
+vp9/decoder/vp9_detokenize.h
+vp9/decoder/vp9_idct_blk.c
+vp9/decoder/vp9_idct_blk.h
+vp9/decoder/vp9_onyxd.h
+vp9/decoder/vp9_onyxd_if.c
+vp9/decoder/vp9_onyxd_int.h
+vp9/decoder/vp9_read_bit_buffer.h
+vp9/decoder/vp9_treereader.h
+vp9/vp9_common.mk
+vp9/vp9_dx_iface.c
+vp9/vp9dx.mk
+vp9/vp9_iface_common.h
 vpx_config.c
 vpx/internal/vpx_codec_internal.h
 vpx_mem/include/vpx_mem_intrnl.h
@@ -140,17 +219,20 @@
 vpx_mem/vpx_mem.h
 vpx_mem/vpx_mem.mk
 vpx_ports/asm_offsets.h
+vpx_ports/emmintrin_compat.h
 vpx_ports/mem.h
+vpx_ports/vpx_once.h
 vpx_ports/vpx_ports.mk
 vpx_ports/vpx_timer.h
 vpx_scale/generic/gen_scalers.c
-vpx_scale/generic/vpxscale.c
+vpx_scale/generic/vpx_scale.c
 vpx_scale/generic/yv12config.c
 vpx_scale/generic/yv12extend.c
-vpx_scale/generic/yv12extend_generic.h
-vpx_scale/scale_mode.h
-vpx_scale/vpxscale.h
+vpx_scale/vpx_scale_asm_offsets.c
+vpx_scale/vpx_scale.h
 vpx_scale/vpx_scale.mk
+vpx_scale/vpx_scale_rtcd.c
+vpx_scale/vpx_scale_rtcd.sh
 vpx_scale/yv12config.h
 vpx/src/vpx_codec.c
 vpx/src/vpx_decoder.c
diff --git a/mips-dspr2/vpx_rtcd.h b/mips-dspr2/vp8_rtcd.h
similarity index 78%
rename from mips-dspr2/vpx_rtcd.h
rename to mips-dspr2/vp8_rtcd.h
index 4e212f8..b9fc986 100644
--- a/mips-dspr2/vpx_rtcd.h
+++ b/mips-dspr2/vp8_rtcd.h
@@ -1,5 +1,5 @@
-#ifndef VPX_RTCD_
-#define VPX_RTCD_
+#ifndef VP8_RTCD_H_
+#define VP8_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
@@ -7,7 +7,9 @@
 #define RTCD_EXTERN extern
 #endif
 
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -20,6 +22,9 @@
 union int_mv;
 struct yv12_buffer_config;
 
+void vp8_clear_system_state_c();
+#define vp8_clear_system_state vp8_clear_system_state_c
+
 void vp8_dequantize_b_c(struct blockd*, short *dqc);
 #define vp8_dequantize_b vp8_dequantize_b_c
 
@@ -97,7 +102,7 @@
 void vp8_build_intra_predictors_mbuv_s_c(struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride);
 #define vp8_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s_c
 
-void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
 #define vp8_intra4x4_predict vp8_intra4x4_predict_c
 
 void vp8_sixtap_predict16x16_c(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch);
@@ -302,82 +307,7 @@
 int vp8_denoiser_filter_c(struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset);
 #define vp8_denoiser_filter vp8_denoiser_filter_c
 
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_4_5_scale vp8_horizontal_line_4_5_scale_c
-
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_4_5_scale vp8_vertical_band_4_5_scale_c
-
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_4_5_scale vp8_last_vertical_band_4_5_scale_c
-
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_3_scale vp8_horizontal_line_2_3_scale_c
-
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_3_scale vp8_vertical_band_2_3_scale_c
-
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_2_3_scale vp8_last_vertical_band_2_3_scale_c
-
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_5_scale vp8_horizontal_line_3_5_scale_c
-
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_5_scale vp8_vertical_band_3_5_scale_c
-
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_5_scale vp8_last_vertical_band_3_5_scale_c
-
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_4_scale vp8_horizontal_line_3_4_scale_c
-
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_4_scale vp8_vertical_band_3_4_scale_c
-
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_4_scale vp8_last_vertical_band_3_4_scale_c
-
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_1_2_scale vp8_horizontal_line_1_2_scale_c
-
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_1_2_scale vp8_vertical_band_1_2_scale_c
-
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_1_2_scale vp8_last_vertical_band_1_2_scale_c
-
-void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
-
-void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
-
-void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
-
-void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
-
-void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
-
-void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
-#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
-
-void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
-
-void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_y vp8_yv12_copy_y_c
-
-void vpx_rtcd(void);
+void vp8_rtcd(void);
 #include "vpx_config.h"
 
 #ifdef RTCD_C
diff --git a/mips-dspr2/vp9_rtcd.h b/mips-dspr2/vp9_rtcd.h
new file mode 100644
index 0000000..2905eae
--- /dev/null
+++ b/mips-dspr2/vp9_rtcd.h
@@ -0,0 +1,191 @@
+#ifndef VP9_RTCD_H_
+#define VP9_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_16x16 vp9_idct_add_16x16_c
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_8x8 vp9_idct_add_8x8_c
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add vp9_idct_add_c
+
+void vp9_idct_add_32x32_c(int16_t *q, uint8_t *dst, int stride, int eob);
+#define vp9_idct_add_32x32 vp9_idct_add_32x32_c
+
+void vp9_copy_mem16x16_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+void vp9_copy_mem16x16_dspr2(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem16x16 vp9_copy_mem16x16_dspr2
+
+void vp9_copy_mem8x8_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+void vp9_copy_mem8x8_dspr2(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x8 vp9_copy_mem8x8_dspr2
+
+void vp9_copy_mem8x4_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x4 vp9_copy_mem8x4_c
+
+void vp9_build_intra_predictors_c(uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available);
+#define vp9_build_intra_predictors vp9_build_intra_predictors_c
+
+void vp9_build_intra_predictors_sby_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sby_s vp9_build_intra_predictors_sby_s_c
+
+void vp9_build_intra_predictors_sbuv_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sbuv_s vp9_build_intra_predictors_sbuv_s_c
+
+void vp9_intra4x4_predict_c(struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride);
+#define vp9_intra4x4_predict vp9_intra4x4_predict_c
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_8x8 vp9_add_constant_residual_8x8_c
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_16x16 vp9_add_constant_residual_16x16_c
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_32x32 vp9_add_constant_residual_32x32_c
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_vertical_edge_w vp9_mb_lpf_vertical_edge_w_c
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_vertical_edge vp9_mbloop_filter_vertical_edge_c
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_vertical_edge vp9_loop_filter_vertical_edge_c
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_horizontal_edge_w vp9_mb_lpf_horizontal_edge_w_c
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_horizontal_edge vp9_mbloop_filter_horizontal_edge_c
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_horizontal_edge vp9_loop_filter_horizontal_edge_c
+
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_inner vp9_blend_mb_inner_c
+
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_outer vp9_blend_mb_outer_c
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_b vp9_blend_b_c
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8 vp9_convolve8_c
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_horiz vp9_convolve8_horiz_c
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_vert vp9_convolve8_vert_c
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg vp9_convolve8_avg_c
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_horiz vp9_convolve8_avg_horiz_c
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_vert vp9_convolve8_avg_vert_c
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_1_add vp9_short_idct4x4_1_add_c
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_add vp9_short_idct4x4_add_c
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct8x8_add vp9_short_idct8x8_add_c
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_8x8_add vp9_short_idct10_8x8_add_c
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_8x8 vp9_short_idct1_8x8_c
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct16x16_add vp9_short_idct16x16_add_c
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_16x16_add vp9_short_idct10_16x16_add_c
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_16x16 vp9_short_idct1_16x16_c
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct32x32_add vp9_short_idct32x32_add_c
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_32x32 vp9_short_idct1_32x32_c
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_32x32_add vp9_short_idct10_32x32_add_c
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht4x4_add vp9_short_iht4x4_add_c
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht8x8_add vp9_short_iht8x8_add_c
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *output, int pitch, int tx_type);
+#define vp9_short_iht16x16_add vp9_short_iht16x16_add_c
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output);
+#define vp9_idct4_1d vp9_idct4_1d_c
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride);
+#define vp9_dc_only_idct_add vp9_dc_only_idct_add_c
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_1_add vp9_short_iwalsh4x4_1_add_c
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_add vp9_short_iwalsh4x4_add_c
+
+unsigned int vp9_sad32x3_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad32x3 vp9_sad32x3_c
+
+unsigned int vp9_sad3x32_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad3x32 vp9_sad3x32_c
+
+void vp9_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+#if HAVE_DSPR2
+void dsputil_static_init();
+dsputil_static_init();
+#endif
+}
+#endif
+#endif
diff --git a/mips-dspr2/vpx_config.c b/mips-dspr2/vpx_config.c
index 42fc4cb..cf19239 100644
--- a/mips-dspr2/vpx_config.c
+++ b/mips-dspr2/vpx_config.c
@@ -5,5 +5,5 @@
 /* tree. An additional intellectual property rights grant can be found */
 /* in the file PATENTS.  All contributing project authors may */
 /* be found in the AUTHORS file in the root of the source tree. */
-static const char* const cfg = "--force-target=mips32-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-examples --disable-docs --enable-dspr2 --enable-realtime-only";
+static const char* const cfg = "--force-target=mips32-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-vp9-encoder --enable-dspr2 --disable-examples --disable-docs --enable-realtime-only";
 const char *vpx_codec_build_config(void) {return cfg;}
diff --git a/mips-dspr2/vpx_config.h b/mips-dspr2/vpx_config.h
index b03bd7c..0ca4657 100644
--- a/mips-dspr2/vpx_config.h
+++ b/mips-dspr2/vpx_config.h
@@ -9,6 +9,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    
+#define INLINE      __inline__ __attribute__((always_inline))
 #define ARCH_ARM 0
 #define ARCH_MIPS 1
 #define ARCH_X86 0
@@ -34,7 +35,7 @@
 #define HAVE_SYS_MMAN_H 1
 #define HAVE_UNISTD_H 1
 #define CONFIG_EXTERNAL_BUILD 0
-#define CONFIG_INSTALL_DOCS 1
+#define CONFIG_INSTALL_DOCS 0
 #define CONFIG_INSTALL_BINS 1
 #define CONFIG_INSTALL_LIBS 1
 #define CONFIG_INSTALL_SRCS 0
@@ -61,7 +62,10 @@
 #define CONFIG_INTERNAL_STATS 0
 #define CONFIG_VP8_ENCODER 1
 #define CONFIG_VP8_DECODER 1
+#define CONFIG_VP9_ENCODER 0
+#define CONFIG_VP9_DECODER 1
 #define CONFIG_VP8 1
+#define CONFIG_VP9 1
 #define CONFIG_ENCODERS 1
 #define CONFIG_DECODERS 1
 #define CONFIG_STATIC_MSVCRT 0
@@ -77,4 +81,11 @@
 #define CONFIG_UNIT_TESTS 0
 #define CONFIG_MULTI_RES_ENCODING 0
 #define CONFIG_TEMPORAL_DENOISING 1
+#define CONFIG_EXPERIMENTAL 0
+#define CONFIG_DECRYPT 0
+#define CONFIG_ONESHOTQ 0
+#define CONFIG_MULTIPLE_ARF 0
+#define CONFIG_NON420 0
+#define CONFIG_ALPHA 0
+#define CONFIG_BALANCED_COEFTREE 0
 #endif /* VPX_CONFIG_H */
diff --git a/mips-dspr2/vpx_scale_rtcd.h b/mips-dspr2/vpx_scale_rtcd.h
new file mode 100644
index 0000000..7af466a
--- /dev/null
+++ b/mips-dspr2/vpx_scale_rtcd.h
@@ -0,0 +1,58 @@
+#ifndef VPX_SCALE_RTCD_H_
+#define VPX_SCALE_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+struct yv12_buffer_config;
+
+void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
+
+void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
+
+void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
+
+void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
+
+void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
+
+void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
+#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
+
+void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
+
+void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_y vp8_yv12_copy_y_c
+
+void vp9_extend_frame_borders_c(struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y);
+#define vp9_extend_frame_borders vp9_extend_frame_borders_c
+
+void vpx_scale_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+#if HAVE_DSPR2
+void dsputil_static_init();
+dsputil_static_init();
+#endif
+}
+#endif
+#endif
diff --git a/mips-dspr2/vpx_version.h b/mips-dspr2/vpx_version.h
index 663dd49..512851c 100644
--- a/mips-dspr2/vpx_version.h
+++ b/mips-dspr2/vpx_version.h
@@ -1,7 +1,7 @@
 #define VERSION_MAJOR  1
-#define VERSION_MINOR  1
+#define VERSION_MINOR  2
 #define VERSION_PATCH  0
 #define VERSION_EXTRA  ""
 #define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH))
-#define VERSION_STRING_NOSP "v1.1.0"
-#define VERSION_STRING      " v1.1.0"
+#define VERSION_STRING_NOSP "v1.2.0"
+#define VERSION_STRING      " v1.2.0"
diff --git a/mips/.bins b/mips/.bins
deleted file mode 100644
index e69de29..0000000
--- a/mips/.bins
+++ /dev/null
diff --git a/mips/.docs b/mips/.docs
deleted file mode 100644
index e69de29..0000000
--- a/mips/.docs
+++ /dev/null
diff --git a/mips/.libs b/mips/.libs
deleted file mode 100644
index e69de29..0000000
--- a/mips/.libs
+++ /dev/null
diff --git a/mips/libvpx_srcs.txt b/mips/libvpx_srcs.txt
index 5756427..8c1ec80 100644
--- a/mips/libvpx_srcs.txt
+++ b/mips/libvpx_srcs.txt
@@ -4,7 +4,6 @@
 libs.mk
 vp8/common/alloccommon.c
 vp8/common/alloccommon.h
-vp8/common/asm_com_offsets.c
 vp8/common/blockd.c
 vp8/common/blockd.h
 vp8/common/coefupdateprobs.h
@@ -61,8 +60,8 @@
 vp8/common/treecoder.h
 vp8/common/variance_c.c
 vp8/common/variance.h
+vp8/common/vp8_asm_com_offsets.c
 vp8/common/vp8_entropymodedata.h
-vp8/decoder/asm_dec_offsets.c
 vp8/decoder/dboolhuff.c
 vp8/decoder/dboolhuff.h
 vp8/decoder/decodemv.c
@@ -75,7 +74,7 @@
 vp8/decoder/onyxd_int.h
 vp8/decoder/threading.c
 vp8/decoder/treereader.h
-vp8/encoder/asm_enc_offsets.c
+vp8/decoder/vp8_asm_dec_offsets.c
 vp8/encoder/bitstream.c
 vp8/encoder/bitstream.h
 vp8/encoder/block.h
@@ -122,11 +121,91 @@
 vp8/encoder/tokenize.h
 vp8/encoder/treewriter.c
 vp8/encoder/treewriter.h
+vp8/encoder/vp8_asm_enc_offsets.c
 vp8/vp8_common.mk
 vp8/vp8_cx_iface.c
 vp8/vp8cx.mk
 vp8/vp8_dx_iface.c
 vp8/vp8dx.mk
+vp9/common/generic/vp9_systemdependent.c
+vp9/common/vp9_alloccommon.c
+vp9/common/vp9_alloccommon.h
+vp9/common/vp9_asm_com_offsets.c
+vp9/common/vp9_blockd.h
+vp9/common/vp9_common.h
+vp9/common/vp9_convolve.c
+vp9/common/vp9_convolve.h
+vp9/common/vp9_debugmodes.c
+vp9/common/vp9_default_coef_probs.h
+vp9/common/vp9_entropy.c
+vp9/common/vp9_entropy.h
+vp9/common/vp9_entropymode.c
+vp9/common/vp9_entropymode.h
+vp9/common/vp9_entropymv.c
+vp9/common/vp9_entropymv.h
+vp9/common/vp9_enums.h
+vp9/common/vp9_extend.c
+vp9/common/vp9_extend.h
+vp9/common/vp9_filter.c
+vp9/common/vp9_filter.h
+vp9/common/vp9_findnearmv.c
+vp9/common/vp9_findnearmv.h
+vp9/common/vp9_idct.c
+vp9/common/vp9_idct.h
+vp9/common/vp9_loopfilter.c
+vp9/common/vp9_loopfilter_filters.c
+vp9/common/vp9_loopfilter.h
+vp9/common/vp9_mbpitch.c
+vp9/common/vp9_modecont.c
+vp9/common/vp9_modecontext.c
+vp9/common/vp9_modecont.h
+vp9/common/vp9_mv.h
+vp9/common/vp9_mvref_common.c
+vp9/common/vp9_mvref_common.h
+vp9/common/vp9_onyxc_int.h
+vp9/common/vp9_onyx.h
+vp9/common/vp9_ppflags.h
+vp9/common/vp9_pragmas.h
+vp9/common/vp9_pred_common.c
+vp9/common/vp9_pred_common.h
+vp9/common/vp9_quant_common.c
+vp9/common/vp9_quant_common.h
+vp9/common/vp9_reconinter.c
+vp9/common/vp9_reconinter.h
+vp9/common/vp9_reconintra.c
+vp9/common/vp9_reconintra.h
+vp9/common/vp9_rtcd.c
+vp9/common/vp9_rtcd_defs.sh
+vp9/common/vp9_sadmxn.h
+vp9/common/vp9_seg_common.c
+vp9/common/vp9_seg_common.h
+vp9/common/vp9_subpelvar.h
+vp9/common/vp9_systemdependent.h
+vp9/common/vp9_textblit.h
+vp9/common/vp9_tile_common.c
+vp9/common/vp9_tile_common.h
+vp9/common/vp9_treecoder.c
+vp9/common/vp9_treecoder.h
+vp9/decoder/vp9_asm_dec_offsets.c
+vp9/decoder/vp9_dboolhuff.c
+vp9/decoder/vp9_dboolhuff.h
+vp9/decoder/vp9_decodemv.c
+vp9/decoder/vp9_decodemv.h
+vp9/decoder/vp9_decodframe.c
+vp9/decoder/vp9_decodframe.h
+vp9/decoder/vp9_detokenize.c
+vp9/decoder/vp9_detokenize.h
+vp9/decoder/vp9_idct_blk.c
+vp9/decoder/vp9_idct_blk.h
+vp9/decoder/vp9_onyxd.h
+vp9/decoder/vp9_onyxd_if.c
+vp9/decoder/vp9_onyxd_int.h
+vp9/decoder/vp9_read_bit_buffer.h
+vp9/decoder/vp9_treereader.h
+vp9/vp9_common.mk
+vp9/vp9_dx_iface.c
+vp9/vp9dx.mk
+vp9/vp9_iface_common.h
 vpx_config.c
 vpx/internal/vpx_codec_internal.h
 vpx_mem/include/vpx_mem_intrnl.h
@@ -134,17 +213,20 @@
 vpx_mem/vpx_mem.h
 vpx_mem/vpx_mem.mk
 vpx_ports/asm_offsets.h
+vpx_ports/emmintrin_compat.h
 vpx_ports/mem.h
+vpx_ports/vpx_once.h
 vpx_ports/vpx_ports.mk
 vpx_ports/vpx_timer.h
 vpx_scale/generic/gen_scalers.c
-vpx_scale/generic/vpxscale.c
+vpx_scale/generic/vpx_scale.c
 vpx_scale/generic/yv12config.c
 vpx_scale/generic/yv12extend.c
-vpx_scale/generic/yv12extend_generic.h
-vpx_scale/scale_mode.h
-vpx_scale/vpxscale.h
+vpx_scale/vpx_scale_asm_offsets.c
+vpx_scale/vpx_scale.h
 vpx_scale/vpx_scale.mk
+vpx_scale/vpx_scale_rtcd.c
+vpx_scale/vpx_scale_rtcd.sh
 vpx_scale/yv12config.h
 vpx/src/vpx_codec.c
 vpx/src/vpx_decoder.c
diff --git a/mips/vpx_rtcd.h b/mips/vp8_rtcd.h
similarity index 76%
rename from mips/vpx_rtcd.h
rename to mips/vp8_rtcd.h
index fe84d62..5f63677 100644
--- a/mips/vpx_rtcd.h
+++ b/mips/vp8_rtcd.h
@@ -1,5 +1,5 @@
-#ifndef VPX_RTCD_
-#define VPX_RTCD_
+#ifndef VP8_RTCD_H_
+#define VP8_RTCD_H_
 
 #ifdef RTCD_C
 #define RTCD_EXTERN
@@ -7,7 +7,9 @@
 #define RTCD_EXTERN extern
 #endif
 
-#include "vp8/common/blockd.h"
+/*
+ * VP8
+ */
 
 struct blockd;
 struct macroblockd;
@@ -20,6 +22,9 @@
 union int_mv;
 struct yv12_buffer_config;
 
+void vp8_clear_system_state_c();
+#define vp8_clear_system_state vp8_clear_system_state_c
+
 void vp8_dequantize_b_c(struct blockd*, short *dqc);
 #define vp8_dequantize_b vp8_dequantize_b_c
 
@@ -83,7 +88,7 @@
 void vp8_build_intra_predictors_mbuv_s_c(struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride);
 #define vp8_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s_c
 
-void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, B_PREDICTION_MODE b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
+void vp8_intra4x4_predict_c(unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left);
 #define vp8_intra4x4_predict vp8_intra4x4_predict_c
 
 void vp8_sixtap_predict16x16_c(unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch);
@@ -284,82 +289,7 @@
 int vp8_denoiser_filter_c(struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset);
 #define vp8_denoiser_filter vp8_denoiser_filter_c
 
-void vp8_horizontal_line_4_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_4_5_scale vp8_horizontal_line_4_5_scale_c
-
-void vp8_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_4_5_scale vp8_vertical_band_4_5_scale_c
-
-void vp8_last_vertical_band_4_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_4_5_scale vp8_last_vertical_band_4_5_scale_c
-
-void vp8_horizontal_line_2_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_3_scale vp8_horizontal_line_2_3_scale_c
-
-void vp8_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_3_scale vp8_vertical_band_2_3_scale_c
-
-void vp8_last_vertical_band_2_3_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_2_3_scale vp8_last_vertical_band_2_3_scale_c
-
-void vp8_horizontal_line_3_5_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_5_scale vp8_horizontal_line_3_5_scale_c
-
-void vp8_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_5_scale vp8_vertical_band_3_5_scale_c
-
-void vp8_last_vertical_band_3_5_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_5_scale vp8_last_vertical_band_3_5_scale_c
-
-void vp8_horizontal_line_3_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_3_4_scale vp8_horizontal_line_3_4_scale_c
-
-void vp8_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_3_4_scale vp8_vertical_band_3_4_scale_c
-
-void vp8_last_vertical_band_3_4_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_3_4_scale vp8_last_vertical_band_3_4_scale_c
-
-void vp8_horizontal_line_1_2_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_1_2_scale vp8_horizontal_line_1_2_scale_c
-
-void vp8_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_1_2_scale vp8_vertical_band_1_2_scale_c
-
-void vp8_last_vertical_band_1_2_scale_c(unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_last_vertical_band_1_2_scale vp8_last_vertical_band_1_2_scale_c
-
-void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
-
-void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
-
-void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
-
-void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
-
-void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
-#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
-
-void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
-#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
-
-void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
-#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
-
-void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
-
-void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
-#define vp8_yv12_copy_y vp8_yv12_copy_y_c
-
-void vpx_rtcd(void);
+void vp8_rtcd(void);
 #include "vpx_config.h"
 
 #ifdef RTCD_C
diff --git a/mips/vp9_rtcd.h b/mips/vp9_rtcd.h
new file mode 100644
index 0000000..1d7b4d2
--- /dev/null
+++ b/mips/vp9_rtcd.h
@@ -0,0 +1,189 @@
+#ifndef VP9_RTCD_H_
+#define VP9_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+/*
+ * VP9
+ */
+
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_enums.h"
+
+struct loop_filter_info;
+struct macroblockd;
+struct loop_filter_info;
+
+/* Encoder forward decls */
+struct macroblock;
+struct vp9_variance_vtable;
+
+#define DEC_MVCOSTS int *mvjcost, int *mvcost[2]
+union int_mv;
+struct yv12_buffer_config;
+
+void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_16x16 vp9_idct_add_16x16_c
+
+void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add_8x8 vp9_idct_add_8x8_c
+
+void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob);
+#define vp9_idct_add vp9_idct_add_c
+
+void vp9_idct_add_32x32_c(int16_t *q, uint8_t *dst, int stride, int eob);
+#define vp9_idct_add_32x32 vp9_idct_add_32x32_c
+
+void vp9_copy_mem16x16_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem16x16 vp9_copy_mem16x16_c
+
+void vp9_copy_mem8x8_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x8 vp9_copy_mem8x8_c
+
+void vp9_copy_mem8x4_c(const uint8_t *src, int src_pitch, uint8_t *dst, int dst_pitch);
+#define vp9_copy_mem8x4 vp9_copy_mem8x4_c
+
+void vp9_build_intra_predictors_c(uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available);
+#define vp9_build_intra_predictors vp9_build_intra_predictors_c
+
+void vp9_build_intra_predictors_sby_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sby_s vp9_build_intra_predictors_sby_s_c
+
+void vp9_build_intra_predictors_sbuv_s_c(struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize);
+#define vp9_build_intra_predictors_sbuv_s vp9_build_intra_predictors_sbuv_s_c
+
+void vp9_intra4x4_predict_c(struct macroblockd *xd, int block, enum BLOCK_SIZE_TYPE bsize, int b_mode, uint8_t *predictor, int pre_stride);
+#define vp9_intra4x4_predict vp9_intra4x4_predict_c
+
+void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_8x8 vp9_add_constant_residual_8x8_c
+
+void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_16x16 vp9_add_constant_residual_16x16_c
+
+void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest, int stride);
+#define vp9_add_constant_residual_32x32 vp9_add_constant_residual_32x32_c
+
+void vp9_mb_lpf_vertical_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_vertical_edge_w vp9_mb_lpf_vertical_edge_w_c
+
+void vp9_mbloop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_vertical_edge vp9_mbloop_filter_vertical_edge_c
+
+void vp9_loop_filter_vertical_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_vertical_edge vp9_loop_filter_vertical_edge_c
+
+void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh);
+#define vp9_mb_lpf_horizontal_edge_w vp9_mb_lpf_horizontal_edge_w_c
+
+void vp9_mbloop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_mbloop_filter_horizontal_edge vp9_mbloop_filter_horizontal_edge_c
+
+void vp9_loop_filter_horizontal_edge_c(uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count);
+#define vp9_loop_filter_horizontal_edge vp9_loop_filter_horizontal_edge_c
+
+void vp9_blend_mb_inner_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_inner vp9_blend_mb_inner_c
+
+void vp9_blend_mb_outer_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_mb_outer vp9_blend_mb_outer_c
+
+void vp9_blend_b_c(uint8_t *y, uint8_t *u, uint8_t *v, int y1, int u1, int v1, int alpha, int stride);
+#define vp9_blend_b vp9_blend_b_c
+
+void vp9_convolve8_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8 vp9_convolve8_c
+
+void vp9_convolve8_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_horiz vp9_convolve8_horiz_c
+
+void vp9_convolve8_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_vert vp9_convolve8_vert_c
+
+void vp9_convolve8_avg_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg vp9_convolve8_avg_c
+
+void vp9_convolve8_avg_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_horiz vp9_convolve8_avg_horiz_c
+
+void vp9_convolve8_avg_vert_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h);
+#define vp9_convolve8_avg_vert vp9_convolve8_avg_vert_c
+
+void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_1_add vp9_short_idct4x4_1_add_c
+
+void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct4x4_add vp9_short_idct4x4_add_c
+
+void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct8x8_add vp9_short_idct8x8_add_c
+
+void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_8x8_add vp9_short_idct10_8x8_add_c
+
+void vp9_short_idct1_8x8_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_8x8 vp9_short_idct1_8x8_c
+
+void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct16x16_add vp9_short_idct16x16_add_c
+
+void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_16x16_add vp9_short_idct10_16x16_add_c
+
+void vp9_short_idct1_16x16_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_16x16 vp9_short_idct1_16x16_c
+
+void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct32x32_add vp9_short_idct32x32_add_c
+
+void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output);
+#define vp9_short_idct1_32x32 vp9_short_idct1_32x32_c
+
+void vp9_short_idct10_32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_idct10_32x32_add vp9_short_idct10_32x32_add_c
+
+void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht4x4_add vp9_short_iht4x4_add_c
+
+void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride, int tx_type);
+#define vp9_short_iht8x8_add vp9_short_iht8x8_add_c
+
+void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *output, int pitch, int tx_type);
+#define vp9_short_iht16x16_add vp9_short_iht16x16_add_c
+
+void vp9_idct4_1d_c(int16_t *input, int16_t *output);
+#define vp9_idct4_1d vp9_idct4_1d_c
+
+void vp9_dc_only_idct_add_c(int input_dc, uint8_t *pred_ptr, uint8_t *dst_ptr, int pitch, int stride);
+#define vp9_dc_only_idct_add vp9_dc_only_idct_add_c
+
+void vp9_short_iwalsh4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_1_add vp9_short_iwalsh4x4_1_add_c
+
+void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride);
+#define vp9_short_iwalsh4x4_add vp9_short_iwalsh4x4_add_c
+
+unsigned int vp9_sad32x3_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad32x3 vp9_sad32x3_c
+
+unsigned int vp9_sad3x32_c(const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int ref_stride, int max_sad);
+#define vp9_sad3x32 vp9_sad3x32_c
+
+void vp9_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+#if HAVE_DSPR2
+void dsputil_static_init();
+dsputil_static_init();
+#endif
+}
+#endif
+#endif
diff --git a/mips/vpx_config.c b/mips/vpx_config.c
index fa93c2c..84f0e8b 100644
--- a/mips/vpx_config.c
+++ b/mips/vpx_config.c
@@ -5,5 +5,5 @@
 /* tree. An additional intellectual property rights grant can be found */
 /* in the file PATENTS.  All contributing project authors may */
 /* be found in the AUTHORS file in the root of the source tree. */
-static const char* const cfg = "--force-target=mips32-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-examples --disable-docs --enable-realtime-only";
+static const char* const cfg = "--force-target=mips32-android-gcc --disable-runtime-cpu-detect --sdk-path=/usr/local/google/home/johannkoenig/android-ndk --disable-vp9-encoder --disable-examples --disable-docs --enable-realtime-only";
 const char *vpx_codec_build_config(void) {return cfg;}
diff --git a/mips/vpx_config.h b/mips/vpx_config.h
index 9f51b09..49eab1e 100644
--- a/mips/vpx_config.h
+++ b/mips/vpx_config.h
@@ -9,6 +9,7 @@
 #ifndef VPX_CONFIG_H
 #define VPX_CONFIG_H
 #define RESTRICT    
+#define INLINE      __inline__ __attribute__((always_inline))
 #define ARCH_ARM 0
 #define ARCH_MIPS 1
 #define ARCH_X86 0
@@ -34,7 +35,7 @@
 #define HAVE_SYS_MMAN_H 1
 #define HAVE_UNISTD_H 1
 #define CONFIG_EXTERNAL_BUILD 0
-#define CONFIG_INSTALL_DOCS 1
+#define CONFIG_INSTALL_DOCS 0
 #define CONFIG_INSTALL_BINS 1
 #define CONFIG_INSTALL_LIBS 1
 #define CONFIG_INSTALL_SRCS 0
@@ -61,7 +62,10 @@
 #define CONFIG_INTERNAL_STATS 0
 #define CONFIG_VP8_ENCODER 1
 #define CONFIG_VP8_DECODER 1
+#define CONFIG_VP9_ENCODER 0
+#define CONFIG_VP9_DECODER 1
 #define CONFIG_VP8 1
+#define CONFIG_VP9 1
 #define CONFIG_ENCODERS 1
 #define CONFIG_DECODERS 1
 #define CONFIG_STATIC_MSVCRT 0
@@ -77,4 +81,11 @@
 #define CONFIG_UNIT_TESTS 0
 #define CONFIG_MULTI_RES_ENCODING 0
 #define CONFIG_TEMPORAL_DENOISING 1
+#define CONFIG_EXPERIMENTAL 0
+#define CONFIG_DECRYPT 0
+#define CONFIG_ONESHOTQ 0
+#define CONFIG_MULTIPLE_ARF 0
+#define CONFIG_NON420 0
+#define CONFIG_ALPHA 0
+#define CONFIG_BALANCED_COEFTREE 0
 #endif /* VPX_CONFIG_H */
diff --git a/mips/vpx_scale_rtcd.h b/mips/vpx_scale_rtcd.h
new file mode 100644
index 0000000..7af466a
--- /dev/null
+++ b/mips/vpx_scale_rtcd.h
@@ -0,0 +1,58 @@
+#ifndef VPX_SCALE_RTCD_H_
+#define VPX_SCALE_RTCD_H_
+
+#ifdef RTCD_C
+#define RTCD_EXTERN
+#else
+#define RTCD_EXTERN extern
+#endif
+
+struct yv12_buffer_config;
+
+void vp8_horizontal_line_5_4_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_4_scale vp8_horizontal_line_5_4_scale_c
+
+void vp8_vertical_band_5_4_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_4_scale vp8_vertical_band_5_4_scale_c
+
+void vp8_horizontal_line_5_3_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_5_3_scale vp8_horizontal_line_5_3_scale_c
+
+void vp8_vertical_band_5_3_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_5_3_scale vp8_vertical_band_5_3_scale_c
+
+void vp8_horizontal_line_2_1_scale_c(const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width);
+#define vp8_horizontal_line_2_1_scale vp8_horizontal_line_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale vp8_vertical_band_2_1_scale_c
+
+void vp8_vertical_band_2_1_scale_i_c(unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width);
+#define vp8_vertical_band_2_1_scale_i vp8_vertical_band_2_1_scale_i_c
+
+void vp8_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf);
+#define vp8_yv12_extend_frame_borders vp8_yv12_extend_frame_borders_c
+
+void vp8_yv12_copy_frame_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_frame vp8_yv12_copy_frame_c
+
+void vp8_yv12_copy_y_c(struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc);
+#define vp8_yv12_copy_y vp8_yv12_copy_y_c
+
+void vp9_extend_frame_borders_c(struct yv12_buffer_config *ybf, int subsampling_x, int subsampling_y);
+#define vp9_extend_frame_borders vp9_extend_frame_borders_c
+
+void vpx_scale_rtcd(void);
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+static void setup_rtcd_internal(void)
+{
+
+#if HAVE_DSPR2
+void dsputil_static_init();
+dsputil_static_init();
+#endif
+}
+#endif
+#endif
diff --git a/mips/vpx_version.h b/mips/vpx_version.h
index 663dd49..512851c 100644
--- a/mips/vpx_version.h
+++ b/mips/vpx_version.h
@@ -1,7 +1,7 @@
 #define VERSION_MAJOR  1
-#define VERSION_MINOR  1
+#define VERSION_MINOR  2
 #define VERSION_PATCH  0
 #define VERSION_EXTRA  ""
 #define VERSION_PACKED ((VERSION_MAJOR<<16)|(VERSION_MINOR<<8)|(VERSION_PATCH))
-#define VERSION_STRING_NOSP "v1.1.0"
-#define VERSION_STRING      " v1.1.0"
+#define VERSION_STRING_NOSP "v1.2.0"
+#define VERSION_STRING      " v1.2.0"