diff --git a/jccolor.c b/jccolor.c
index 2407456..f20ed9d 100644
--- a/jccolor.c
+++ b/jccolor.c
@@ -3,6 +3,7 @@
  *
  * Copyright (C) 1991-1996, Thomas G. Lane.
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright 2009 D. R. Commander
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -148,10 +149,10 @@
     outptr2 = output_buf[2][output_row];
     output_row++;
     for (col = 0; col < num_cols; col++) {
-      r = GETJSAMPLE(inptr[RGB_RED]);
-      g = GETJSAMPLE(inptr[RGB_GREEN]);
-      b = GETJSAMPLE(inptr[RGB_BLUE]);
-      inptr += RGB_PIXELSIZE;
+      r = GETJSAMPLE(inptr[rgb_red[cinfo->in_color_space]]);
+      g = GETJSAMPLE(inptr[rgb_green[cinfo->in_color_space]]);
+      b = GETJSAMPLE(inptr[rgb_blue[cinfo->in_color_space]]);
+      inptr += rgb_pixelsize[cinfo->in_color_space];
       /* If the inputs are 0..MAXJSAMPLE, the outputs of these equations
        * must be too; we do not need an explicit range-limiting operation.
        * Hence the value being shifted is never negative, and we don't
@@ -202,10 +203,10 @@
     outptr = output_buf[0][output_row];
     output_row++;
     for (col = 0; col < num_cols; col++) {
-      r = GETJSAMPLE(inptr[RGB_RED]);
-      g = GETJSAMPLE(inptr[RGB_GREEN]);
-      b = GETJSAMPLE(inptr[RGB_BLUE]);
-      inptr += RGB_PIXELSIZE;
+      r = GETJSAMPLE(inptr[rgb_red[cinfo->in_color_space]]);
+      g = GETJSAMPLE(inptr[rgb_green[cinfo->in_color_space]]);
+      b = GETJSAMPLE(inptr[rgb_blue[cinfo->in_color_space]]);
+      inptr += rgb_pixelsize[cinfo->in_color_space];
       /* Y */
       outptr[col] = (JSAMPLE)
 		((ctab[r+R_Y_OFF] + ctab[g+G_Y_OFF] + ctab[b+B_Y_OFF])
@@ -376,6 +377,16 @@
     break;
 #endif /* else share code with YCbCr */
 
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+    if (cinfo->input_components != rgb_pixelsize[cinfo->in_color_space])
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
   case JCS_YCbCr:
     if (cinfo->input_components != 3)
       ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
@@ -400,7 +411,13 @@
       ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
     if (cinfo->in_color_space == JCS_GRAYSCALE)
       cconvert->pub.color_convert = grayscale_convert;
-    else if (cinfo->in_color_space == JCS_RGB) {
+    else if (cinfo->in_color_space == JCS_RGB ||
+             cinfo->in_color_space == JCS_EXT_RGB ||
+             cinfo->in_color_space == JCS_EXT_RGBX ||
+             cinfo->in_color_space == JCS_EXT_BGR ||
+             cinfo->in_color_space == JCS_EXT_BGRX ||
+             cinfo->in_color_space == JCS_EXT_XBGR ||
+             cinfo->in_color_space == JCS_EXT_XRGB) {
       cconvert->pub.start_pass = rgb_ycc_start;
       cconvert->pub.color_convert = rgb_gray_convert;
     } else if (cinfo->in_color_space == JCS_YCbCr)
@@ -410,9 +427,16 @@
     break;
 
   case JCS_RGB:
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
     if (cinfo->num_components != 3)
       ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
-    if (cinfo->in_color_space == JCS_RGB && RGB_PIXELSIZE == 3)
+    if (cinfo->in_color_space == cinfo->jpeg_color_space &&
+      rgb_pixelsize[cinfo->in_color_space] == 3)
       cconvert->pub.color_convert = null_convert;
     else
       ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
@@ -421,7 +445,13 @@
   case JCS_YCbCr:
     if (cinfo->num_components != 3)
       ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
-    if (cinfo->in_color_space == JCS_RGB) {
+    if (cinfo->in_color_space == JCS_RGB ||
+        cinfo->in_color_space == JCS_EXT_RGB ||
+        cinfo->in_color_space == JCS_EXT_RGBX ||
+        cinfo->in_color_space == JCS_EXT_BGR ||
+        cinfo->in_color_space == JCS_EXT_BGRX ||
+        cinfo->in_color_space == JCS_EXT_XBGR ||
+        cinfo->in_color_space == JCS_EXT_XRGB) {
       if (jsimd_can_rgb_ycc())
         cconvert->pub.color_convert = jsimd_rgb_ycc_convert;
       else {
diff --git a/jcparam.c b/jcparam.c
index 6fc48f5..2954f63 100644
--- a/jcparam.c
+++ b/jcparam.c
@@ -363,6 +363,12 @@
     jpeg_set_colorspace(cinfo, JCS_GRAYSCALE);
     break;
   case JCS_RGB:
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
     jpeg_set_colorspace(cinfo, JCS_YCbCr);
     break;
   case JCS_YCbCr:
diff --git a/jdcolor.c b/jdcolor.c
index d2d1483..e1dd3ff 100644
--- a/jdcolor.c
+++ b/jdcolor.c
@@ -148,12 +148,12 @@
       cb = GETJSAMPLE(inptr1[col]);
       cr = GETJSAMPLE(inptr2[col]);
       /* Range-limiting is essential due to noise introduced by DCT losses. */
-      outptr[RGB_RED] =   range_limit[y + Crrtab[cr]];
-      outptr[RGB_GREEN] = range_limit[y +
+      outptr[rgb_red[cinfo->out_color_space]] =   range_limit[y + Crrtab[cr]];
+      outptr[rgb_green[cinfo->out_color_space]] = range_limit[y +
 			      ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
 						 SCALEBITS))];
-      outptr[RGB_BLUE] =  range_limit[y + Cbbtab[cb]];
-      outptr += RGB_PIXELSIZE;
+      outptr[rgb_blue[cinfo->out_color_space]] =  range_limit[y + Cbbtab[cb]];
+      outptr += rgb_pixelsize[cinfo->out_color_space];
     }
   }
 }
@@ -229,8 +229,10 @@
     outptr = *output_buf++;
     for (col = 0; col < num_cols; col++) {
       /* We can dispense with GETJSAMPLE() here */
-      outptr[RGB_RED] = outptr[RGB_GREEN] = outptr[RGB_BLUE] = inptr[col];
-      outptr += RGB_PIXELSIZE;
+      outptr[rgb_red[cinfo->out_color_space]] =
+          outptr[rgb_green[cinfo->out_color_space]] =
+          outptr[rgb_blue[cinfo->out_color_space]] = inptr[col];
+      outptr += rgb_pixelsize[cinfo->out_color_space];
     }
   }
 }
@@ -358,7 +360,13 @@
     break;
 
   case JCS_RGB:
-    cinfo->out_color_components = RGB_PIXELSIZE;
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+    cinfo->out_color_components = rgb_pixelsize[cinfo->out_color_space];
     if (cinfo->jpeg_color_space == JCS_YCbCr) {
       if (jsimd_can_ycc_rgb())
         cconvert->pub.color_convert = jsimd_ycc_rgb_convert;
@@ -368,7 +376,8 @@
       }
     } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) {
       cconvert->pub.color_convert = gray_rgb_convert;
-    } else if (cinfo->jpeg_color_space == JCS_RGB && RGB_PIXELSIZE == 3) {
+    } else if (cinfo->jpeg_color_space == cinfo->out_color_space &&
+      rgb_pixelsize[cinfo->out_color_space] == 3) {
       cconvert->pub.color_convert = null_convert;
     } else
       ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
diff --git a/jdmerge.c b/jdmerge.c
index 7742941..7b1a079 100644
--- a/jdmerge.c
+++ b/jdmerge.c
@@ -257,15 +257,15 @@
     cblue = Cbbtab[cb];
     /* Fetch 2 Y values and emit 2 pixels */
     y  = GETJSAMPLE(*inptr0++);
-    outptr[RGB_RED] =   range_limit[y + cred];
-    outptr[RGB_GREEN] = range_limit[y + cgreen];
-    outptr[RGB_BLUE] =  range_limit[y + cblue];
-    outptr += RGB_PIXELSIZE;
+    outptr[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
+    outptr += rgb_pixelsize[cinfo->out_color_space];
     y  = GETJSAMPLE(*inptr0++);
-    outptr[RGB_RED] =   range_limit[y + cred];
-    outptr[RGB_GREEN] = range_limit[y + cgreen];
-    outptr[RGB_BLUE] =  range_limit[y + cblue];
-    outptr += RGB_PIXELSIZE;
+    outptr[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
+    outptr += rgb_pixelsize[cinfo->out_color_space];
   }
   /* If image width is odd, do the last output column separately */
   if (cinfo->output_width & 1) {
@@ -275,9 +275,9 @@
     cgreen = (int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
     cblue = Cbbtab[cb];
     y  = GETJSAMPLE(*inptr0);
-    outptr[RGB_RED] =   range_limit[y + cred];
-    outptr[RGB_GREEN] = range_limit[y + cgreen];
-    outptr[RGB_BLUE] =  range_limit[y + cblue];
+    outptr[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
   }
 }
 
@@ -321,24 +321,24 @@
     cblue = Cbbtab[cb];
     /* Fetch 4 Y values and emit 4 pixels */
     y  = GETJSAMPLE(*inptr00++);
-    outptr0[RGB_RED] =   range_limit[y + cred];
-    outptr0[RGB_GREEN] = range_limit[y + cgreen];
-    outptr0[RGB_BLUE] =  range_limit[y + cblue];
+    outptr0[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr0[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr0[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
     outptr0 += RGB_PIXELSIZE;
     y  = GETJSAMPLE(*inptr00++);
-    outptr0[RGB_RED] =   range_limit[y + cred];
-    outptr0[RGB_GREEN] = range_limit[y + cgreen];
-    outptr0[RGB_BLUE] =  range_limit[y + cblue];
+    outptr0[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr0[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr0[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
     outptr0 += RGB_PIXELSIZE;
     y  = GETJSAMPLE(*inptr01++);
-    outptr1[RGB_RED] =   range_limit[y + cred];
-    outptr1[RGB_GREEN] = range_limit[y + cgreen];
-    outptr1[RGB_BLUE] =  range_limit[y + cblue];
+    outptr1[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr1[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr1[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
     outptr1 += RGB_PIXELSIZE;
     y  = GETJSAMPLE(*inptr01++);
-    outptr1[RGB_RED] =   range_limit[y + cred];
-    outptr1[RGB_GREEN] = range_limit[y + cgreen];
-    outptr1[RGB_BLUE] =  range_limit[y + cblue];
+    outptr1[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr1[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr1[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
     outptr1 += RGB_PIXELSIZE;
   }
   /* If image width is odd, do the last output column separately */
@@ -349,13 +349,13 @@
     cgreen = (int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS);
     cblue = Cbbtab[cb];
     y  = GETJSAMPLE(*inptr00);
-    outptr0[RGB_RED] =   range_limit[y + cred];
-    outptr0[RGB_GREEN] = range_limit[y + cgreen];
-    outptr0[RGB_BLUE] =  range_limit[y + cblue];
+    outptr0[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr0[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr0[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
     y  = GETJSAMPLE(*inptr01);
-    outptr1[RGB_RED] =   range_limit[y + cred];
-    outptr1[RGB_GREEN] = range_limit[y + cgreen];
-    outptr1[RGB_BLUE] =  range_limit[y + cblue];
+    outptr1[rgb_red[cinfo->out_color_space]] =   range_limit[y + cred];
+    outptr1[rgb_green[cinfo->out_color_space]] = range_limit[y + cgreen];
+    outptr1[rgb_blue[cinfo->out_color_space]] =  range_limit[y + cblue];
   }
 }
 
diff --git a/jmorecfg.h b/jmorecfg.h
index 18670ec..0e7fb72 100644
--- a/jmorecfg.h
+++ b/jmorecfg.h
@@ -2,6 +2,7 @@
  * jmorecfg.h
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Copyright (C) 2009, D. R. Commander.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -316,6 +317,23 @@
 #define RGB_BLUE	2	/* Offset of Blue */
 #define RGB_PIXELSIZE	3	/* JSAMPLEs per RGB scanline element */
 
+#define JPEG_NUMCS 12
+
+static const int rgb_red[JPEG_NUMCS] = {
+	-1, -1, RGB_RED, -1, -1, -1, 0, 0, 2, 2, 3, 1
+};
+
+static const int rgb_green[JPEG_NUMCS] = {
+	-1, -1, RGB_GREEN, -1, -1, -1, 1, 1, 1, 1, 2, 2
+};
+
+static const int rgb_blue[JPEG_NUMCS] = {
+	-1, -1, RGB_BLUE, -1, -1, -1, 2, 2, 0, 0, 1, 3
+};
+
+static const int rgb_pixelsize[JPEG_NUMCS] = {
+	-1, -1, RGB_PIXELSIZE, -1, -1, -1, 3, 4, 3, 4, 4, 4
+};
 
 /* Definitions for speed-related optimizations. */
 
diff --git a/jpeglib.h b/jpeglib.h
index d1be8dd..b28ff8d 100644
--- a/jpeglib.h
+++ b/jpeglib.h
@@ -203,13 +203,22 @@
 
 /* Known color spaces. */
 
+#define JCS_EXTENSIONS 1
+
 typedef enum {
 	JCS_UNKNOWN,		/* error/unspecified */
 	JCS_GRAYSCALE,		/* monochrome */
-	JCS_RGB,		/* red/green/blue */
+	JCS_RGB,		/* red/green/blue as specified by the RGB_RED, RGB_GREEN,
+                 RGB_BLUE, and RGB_PIXELSIZE macros */
 	JCS_YCbCr,		/* Y/Cb/Cr (also known as YUV) */
 	JCS_CMYK,		/* C/M/Y/K */
-	JCS_YCCK		/* Y/Cb/Cr/K */
+	JCS_YCCK,		/* Y/Cb/Cr/K */
+	JCS_EXT_RGB,		/* red/green/blue */
+	JCS_EXT_RGBX,		/* red/green/blue/x */
+	JCS_EXT_BGR,		/* blue/green/red */
+	JCS_EXT_BGRX,		/* blue/green/red/x */
+	JCS_EXT_XBGR,		/* x/blue/green/red */
+	JCS_EXT_XRGB,		/* x/red/green/blue */
 } J_COLOR_SPACE;
 
 /* DCT/IDCT algorithm options. */
diff --git a/jquant1.c b/jquant1.c
index b2f96aa..eac65a3 100644
--- a/jquant1.c
+++ b/jquant1.c
@@ -193,7 +193,10 @@
   int total_colors, iroot, i, j;
   boolean changed;
   long temp;
-  static const int RGB_order[3] = { RGB_GREEN, RGB_RED, RGB_BLUE };
+  int RGB_order[3] = { RGB_GREEN, RGB_RED, RGB_BLUE };
+  RGB_order[0] = rgb_green[cinfo->out_color_space];
+  RGB_order[1] = rgb_red[cinfo->out_color_space];
+  RGB_order[2] = rgb_blue[cinfo->out_color_space];
 
   /* We can allocate at least the nc'th root of max_colors per component. */
   /* Compute floor(nc'th root of max_colors). */
diff --git a/jquant2.c b/jquant2.c
index af601e3..9ce3ff3 100644
--- a/jquant2.c
+++ b/jquant2.c
@@ -74,29 +74,10 @@
 #define G_SCALE 3		/* scale G distances by this much */
 #define B_SCALE 1		/* and B by this much */
 
-/* Relabel R/G/B as components 0/1/2, respecting the RGB ordering defined
- * in jmorecfg.h.  As the code stands, it will do the right thing for R,G,B
- * and B,G,R orders.  If you define some other weird order in jmorecfg.h,
- * you'll get compile errors until you extend this logic.  In that case
- * you'll probably want to tweak the histogram sizes too.
- */
-
-#if RGB_RED == 0
-#define C0_SCALE R_SCALE
-#endif
-#if RGB_BLUE == 0
-#define C0_SCALE B_SCALE
-#endif
-#if RGB_GREEN == 1
-#define C1_SCALE G_SCALE
-#endif
-#if RGB_RED == 2
-#define C2_SCALE R_SCALE
-#endif
-#if RGB_BLUE == 2
-#define C2_SCALE B_SCALE
-#endif
-
+static const int c_scales[3]={R_SCALE, G_SCALE, B_SCALE};
+#define C0_SCALE c_scales[rgb_red[cinfo->out_color_space]]
+#define C1_SCALE c_scales[rgb_green[cinfo->out_color_space]]
+#define C2_SCALE c_scales[rgb_blue[cinfo->out_color_space]]
 
 /*
  * First we have the histogram data structure and routines for creating it.
@@ -454,15 +435,16 @@
     /* We want to break any ties in favor of green, then red, blue last.
      * This code does the right thing for R,G,B or B,G,R color orders only.
      */
-#if RGB_RED == 0
-    cmax = c1; n = 1;
-    if (c0 > cmax) { cmax = c0; n = 0; }
-    if (c2 > cmax) { n = 2; }
-#else
-    cmax = c1; n = 1;
-    if (c2 > cmax) { cmax = c2; n = 2; }
-    if (c0 > cmax) { n = 0; }
-#endif
+    if (rgb_red[cinfo->out_color_space] == 0) {
+      cmax = c1; n = 1;
+      if (c0 > cmax) { cmax = c0; n = 0; }
+      if (c2 > cmax) { n = 2; }
+    }
+    else {
+      cmax = c1; n = 1;
+      if (c2 > cmax) { cmax = c2; n = 2; }
+      if (c0 > cmax) { n = 0; }
+    }
     /* Choose split point along selected axis, and update box bounds.
      * Current algorithm: split at halfway point.
      * (Since the box has been shrunk to minimum volume,
diff --git a/jsimd.c b/jsimd.c
index 7f81dda..2c49a6b 100644
--- a/jsimd.c
+++ b/jsimd.c
@@ -2,6 +2,7 @@
  * jsimd.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright 2009 D. R. Commander
  * 
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -107,12 +108,45 @@
                        JDIMENSION output_row, int num_rows)
 {
 #ifdef WITH_SIMD
+  void (*sse2fct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+  void (*mmxfct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+  switch(cinfo->in_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_extrgb_ycc_convert_sse2;
+      mmxfct=jsimd_extrgb_ycc_convert_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_extrgbx_ycc_convert_sse2;
+      mmxfct=jsimd_extrgbx_ycc_convert_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_extbgr_ycc_convert_sse2;
+      mmxfct=jsimd_extbgr_ycc_convert_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_extbgrx_ycc_convert_sse2;
+      mmxfct=jsimd_extbgrx_ycc_convert_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_extxbgr_ycc_convert_sse2;
+      mmxfct=jsimd_extxbgr_ycc_convert_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_extxrgb_ycc_convert_sse2;
+      mmxfct=jsimd_extxrgb_ycc_convert_mmx;
+      break;
+    default:
+      sse2fct=jsimd_rgb_ycc_convert_sse2;
+      mmxfct=jsimd_rgb_ycc_convert_mmx;
+      break;
+  }
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
-    jsimd_rgb_ycc_convert_sse2(cinfo->image_width, input_buf,
+    sse2fct(cinfo->image_width, input_buf,
         output_buf, output_row, num_rows);
   else if (simd_support & JSIMD_MMX)
-    jsimd_rgb_ycc_convert_mmx(cinfo->image_width, input_buf,
+    mmxfct(cinfo->image_width, input_buf,
         output_buf, output_row, num_rows);
 #endif
 }
@@ -123,12 +157,45 @@
                        JSAMPARRAY output_buf, int num_rows)
 {
 #ifdef WITH_SIMD
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_ycc_extrgb_convert_sse2;
+      mmxfct=jsimd_ycc_extrgb_convert_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_ycc_extrgbx_convert_sse2;
+      mmxfct=jsimd_ycc_extrgbx_convert_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_ycc_extbgr_convert_sse2;
+      mmxfct=jsimd_ycc_extbgr_convert_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_ycc_extbgrx_convert_sse2;
+      mmxfct=jsimd_ycc_extbgrx_convert_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_ycc_extxbgr_convert_sse2;
+      mmxfct=jsimd_ycc_extxbgr_convert_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_ycc_extxrgb_convert_sse2;
+      mmxfct=jsimd_ycc_extxrgb_convert_mmx;
+      break;
+    default:
+      sse2fct=jsimd_ycc_rgb_convert_sse2;
+      mmxfct=jsimd_ycc_rgb_convert_mmx;
+      break;
+  }
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
-    jsimd_ycc_rgb_convert_sse2(cinfo->output_width, input_buf,
+    sse2fct(cinfo->output_width, input_buf,
         input_row, output_buf, num_rows);
   else if (simd_support & JSIMD_MMX)
-    jsimd_ycc_rgb_convert_mmx(cinfo->output_width, input_buf,
+    mmxfct(cinfo->output_width, input_buf,
         input_row, output_buf, num_rows);
 #endif
 }
diff --git a/simd/Makefile.am b/simd/Makefile.am
index ea53fee..2282804 100644
--- a/simd/Makefile.am
+++ b/simd/Makefile.am
@@ -19,6 +19,11 @@
 	jiss2red.asm jiss2int.asm jiss2fst.asm \
 	jcqnts2f.asm jiss2flt.asm
 
+jccolmmx.lo: jcclrmmx.asm
+jccolss2.lo: jcclrss2.asm
+jdcolmmx.lo: jdclrmmx.asm
+jdcolss2.lo: jdclrss2.asm
+
 .asm.lo:
 	$(LIBTOOL) --mode=compile --tag NASM ./nasm_lt.sh $(NASM) $(NAFLAGS) $< -o $@
 
diff --git a/simd/jcclrmmx.asm b/simd/jcclrmmx.asm
new file mode 100644
index 0000000..9cbae72
--- /dev/null
+++ b/simd/jcclrmmx.asm
@@ -0,0 +1,474 @@
+;
+; jcclrmmx.asm - colorspace conversion (MMX)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_mmx (JDIMENSION img_width,
+;                           JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+;                           JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b)	(b)+8			; JDIMENSION img_width
+%define input_buf(b)	(b)+12		; JSAMPARRAY input_buf
+%define output_buf(b)	(b)+16		; JSAMPIMAGE output_buf
+%define output_row(b)	(b)+20		; JDIMENSION output_row
+%define num_rows(b)	(b)+24		; int num_rows
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
+%define WK_NUM		8
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+	global	EXTN(jsimd_rgb_ycc_convert_mmx)
+
+EXTN(jsimd_rgb_ycc_convert_mmx):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [img_width(eax)]	; num_cols
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	esi, JSAMPIMAGE [output_buf(eax)]
+	mov	ecx, JDIMENSION [output_row(eax)]
+	mov	edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
+	lea	edi, [edi+ecx*SIZEOF_JSAMPROW]
+	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+	pop	ecx
+
+	mov	esi, JSAMPARRAY [input_buf(eax)]
+	mov	eax, INT [num_rows(eax)]
+	test	eax,eax
+	jle	near .return
+	alignx	16,7
+.rowloop:
+	pushpic	eax
+	push	edx
+	push	ebx
+	push	edi
+	push	esi
+	push	ecx			; col
+
+	mov	esi, JSAMPROW [esi]	; inptr
+	mov	edi, JSAMPROW [edi]	; outptr0
+	mov	ebx, JSAMPROW [ebx]	; outptr1
+	mov	edx, JSAMPROW [edx]	; outptr2
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+
+	cmp	ecx, byte SIZEOF_MMWORD
+	jae	short .columnloop
+	alignx	16,7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+	push	eax
+	push	edx
+	lea	ecx,[ecx+ecx*2]		; imul ecx,RGB_PIXELSIZE
+	test	cl, SIZEOF_BYTE
+	jz	short .column_ld2
+	sub	ecx, byte SIZEOF_BYTE
+	xor	eax,eax
+	mov	al, BYTE [esi+ecx]
+.column_ld2:
+	test	cl, SIZEOF_WORD
+	jz	short .column_ld4
+	sub	ecx, byte SIZEOF_WORD
+	xor	edx,edx
+	mov	dx, WORD [esi+ecx]
+	shl	eax, WORD_BIT
+	or	eax,edx
+.column_ld4:
+	movd	mmA,eax
+	pop	edx
+	pop	eax
+	test	cl, SIZEOF_DWORD
+	jz	short .column_ld8
+	sub	ecx, byte SIZEOF_DWORD
+	movd	mmG, DWORD [esi+ecx]
+	psllq	mmA, DWORD_BIT
+	por	mmA,mmG
+.column_ld8:
+	test	cl, SIZEOF_MMWORD
+	jz	short .column_ld16
+	movq	mmG,mmA
+	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
+	mov	ecx, SIZEOF_MMWORD
+	jmp	short .rgb_ycc_cnv
+.column_ld16:
+	test	cl, 2*SIZEOF_MMWORD
+	mov	ecx, SIZEOF_MMWORD
+	jz	short .rgb_ycc_cnv
+	movq	mmF,mmA
+	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
+	movq	mmG, MMWORD [esi+1*SIZEOF_MMWORD]
+	jmp	short .rgb_ycc_cnv
+	alignx	16,7
+
+.columnloop:
+	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
+	movq	mmG, MMWORD [esi+1*SIZEOF_MMWORD]
+	movq	mmF, MMWORD [esi+2*SIZEOF_MMWORD]
+
+.rgb_ycc_cnv:
+	; mmA=(00 10 20 01 11 21 02 12)
+	; mmG=(22 03 13 23 04 14 24 05)
+	; mmF=(15 25 06 16 26 07 17 27)
+
+	movq      mmD,mmA
+	psllq     mmA,4*BYTE_BIT	; mmA=(-- -- -- -- 00 10 20 01)
+	psrlq     mmD,4*BYTE_BIT	; mmD=(11 21 02 12 -- -- -- --)
+
+	punpckhbw mmA,mmG		; mmA=(00 04 10 14 20 24 01 05)
+	psllq     mmG,4*BYTE_BIT	; mmG=(-- -- -- -- 22 03 13 23)
+
+	punpcklbw mmD,mmF		; mmD=(11 15 21 25 02 06 12 16)
+	punpckhbw mmG,mmF		; mmG=(22 26 03 07 13 17 23 27)
+
+	movq      mmE,mmA
+	psllq     mmA,4*BYTE_BIT	; mmA=(-- -- -- -- 00 04 10 14)
+	psrlq     mmE,4*BYTE_BIT	; mmE=(20 24 01 05 -- -- -- --)
+
+	punpckhbw mmA,mmD		; mmA=(00 02 04 06 10 12 14 16)
+	psllq     mmD,4*BYTE_BIT	; mmD=(-- -- -- -- 11 15 21 25)
+
+	punpcklbw mmE,mmG		; mmE=(20 22 24 26 01 03 05 07)
+	punpckhbw mmD,mmG		; mmD=(11 13 15 17 21 23 25 27)
+
+	pxor      mmH,mmH
+
+	movq      mmC,mmA
+	punpcklbw mmA,mmH		; mmA=(00 02 04 06)
+	punpckhbw mmC,mmH		; mmC=(10 12 14 16)
+
+	movq      mmB,mmE
+	punpcklbw mmE,mmH		; mmE=(20 22 24 26)
+	punpckhbw mmB,mmH		; mmB=(01 03 05 07)
+
+	movq      mmF,mmD
+	punpcklbw mmD,mmH		; mmD=(11 13 15 17)
+	punpckhbw mmF,mmH		; mmF=(21 23 25 27)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+	test	cl, SIZEOF_MMWORD/8
+	jz	short .column_ld2
+	sub	ecx, byte SIZEOF_MMWORD/8
+	movd	mmA, DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+	test	cl, SIZEOF_MMWORD/4
+	jz	short .column_ld4
+	sub	ecx, byte SIZEOF_MMWORD/4
+	movq	mmF,mmA
+	movq	mmA, MMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld4:
+	test	cl, SIZEOF_MMWORD/2
+	mov	ecx, SIZEOF_MMWORD
+	jz	short .rgb_ycc_cnv
+	movq	mmD,mmA
+	movq	mmC,mmF
+	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
+	movq	mmF, MMWORD [esi+1*SIZEOF_MMWORD]
+	jmp	short .rgb_ycc_cnv
+	alignx	16,7
+
+.columnloop:
+	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
+	movq	mmF, MMWORD [esi+1*SIZEOF_MMWORD]
+	movq	mmD, MMWORD [esi+2*SIZEOF_MMWORD]
+	movq	mmC, MMWORD [esi+3*SIZEOF_MMWORD]
+
+.rgb_ycc_cnv:
+	; mmA=(00 10 20 30 01 11 21 31)
+	; mmF=(02 12 22 32 03 13 23 33)
+	; mmD=(04 14 24 34 05 15 25 35)
+	; mmC=(06 16 26 36 07 17 27 37)
+
+	movq      mmB,mmA
+	punpcklbw mmA,mmF		; mmA=(00 02 10 12 20 22 30 32)
+	punpckhbw mmB,mmF		; mmB=(01 03 11 13 21 23 31 33)
+
+	movq      mmG,mmD
+	punpcklbw mmD,mmC		; mmD=(04 06 14 16 24 26 34 36)
+	punpckhbw mmG,mmC		; mmG=(05 07 15 17 25 27 35 37)
+
+	movq      mmE,mmA
+	punpcklwd mmA,mmD		; mmA=(00 02 04 06 10 12 14 16)
+	punpckhwd mmE,mmD		; mmE=(20 22 24 26 30 32 34 36)
+
+	movq      mmH,mmB
+	punpcklwd mmB,mmG		; mmB=(01 03 05 07 11 13 15 17)
+	punpckhwd mmH,mmG		; mmH=(21 23 25 27 31 33 35 37)
+
+	pxor      mmF,mmF
+
+	movq      mmC,mmA
+	punpcklbw mmA,mmF		; mmA=(00 02 04 06)
+	punpckhbw mmC,mmF		; mmC=(10 12 14 16)
+
+	movq      mmD,mmB
+	punpcklbw mmB,mmF		; mmB=(01 03 05 07)
+	punpckhbw mmD,mmF		; mmD=(11 13 15 17)
+
+	movq      mmG,mmE
+	punpcklbw mmE,mmF		; mmE=(20 22 24 26)
+	punpckhbw mmG,mmF		; mmG=(30 32 34 36)
+
+	punpcklbw mmF,mmH
+	punpckhbw mmH,mmH
+	psrlw     mmF,BYTE_BIT		; mmF=(21 23 25 27)
+	psrlw     mmH,BYTE_BIT		; mmH=(31 33 35 37)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+	; mm0=(R0 R2 R4 R6)=RE, mm2=(G0 G2 G4 G6)=GE, mm4=(B0 B2 B4 B6)=BE
+	; mm1=(R1 R3 R5 R7)=RO, mm3=(G1 G3 G5 G7)=GO, mm5=(B1 B3 B5 B7)=BO
+
+	; (Original)
+	; Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+	;
+	; (This implementation)
+	; Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+	movq      MMWORD [wk(0)], mm0	; wk(0)=RE
+	movq      MMWORD [wk(1)], mm1	; wk(1)=RO
+	movq      MMWORD [wk(2)], mm4	; wk(2)=BE
+	movq      MMWORD [wk(3)], mm5	; wk(3)=BO
+
+	movq      mm6,mm1
+	punpcklwd mm1,mm3
+	punpckhwd mm6,mm3
+	movq      mm7,mm1
+	movq      mm4,mm6
+	pmaddwd   mm1,[GOTOFF(eax,PW_F0299_F0337)] ; mm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+	pmaddwd   mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+	pmaddwd   mm7,[GOTOFF(eax,PW_MF016_MF033)] ; mm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+	pmaddwd   mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+	movq      MMWORD [wk(4)], mm1	; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+	movq      MMWORD [wk(5)], mm6	; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+	pxor      mm1,mm1
+	pxor      mm6,mm6
+	punpcklwd mm1,mm5		; mm1=BOL
+	punpckhwd mm6,mm5		; mm6=BOH
+	psrld     mm1,1			; mm1=BOL*FIX(0.500)
+	psrld     mm6,1			; mm6=BOH*FIX(0.500)
+
+	movq      mm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm5=[PD_ONEHALFM1_CJ]
+
+	paddd     mm7,mm1
+	paddd     mm4,mm6
+	paddd     mm7,mm5
+	paddd     mm4,mm5
+	psrld     mm7,SCALEBITS		; mm7=CbOL
+	psrld     mm4,SCALEBITS		; mm4=CbOH
+	packssdw  mm7,mm4		; mm7=CbO
+
+	movq      mm1, MMWORD [wk(2)]	; mm1=BE
+
+	movq      mm6,mm0
+	punpcklwd mm0,mm2
+	punpckhwd mm6,mm2
+	movq      mm5,mm0
+	movq      mm4,mm6
+	pmaddwd   mm0,[GOTOFF(eax,PW_F0299_F0337)] ; mm0=REL*FIX(0.299)+GEL*FIX(0.337)
+	pmaddwd   mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=REH*FIX(0.299)+GEH*FIX(0.337)
+	pmaddwd   mm5,[GOTOFF(eax,PW_MF016_MF033)] ; mm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+	pmaddwd   mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+	movq      MMWORD [wk(6)], mm0	; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+	movq      MMWORD [wk(7)], mm6	; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+	pxor      mm0,mm0
+	pxor      mm6,mm6
+	punpcklwd mm0,mm1		; mm0=BEL
+	punpckhwd mm6,mm1		; mm6=BEH
+	psrld     mm0,1			; mm0=BEL*FIX(0.500)
+	psrld     mm6,1			; mm6=BEH*FIX(0.500)
+
+	movq      mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
+
+	paddd     mm5,mm0
+	paddd     mm4,mm6
+	paddd     mm5,mm1
+	paddd     mm4,mm1
+	psrld     mm5,SCALEBITS		; mm5=CbEL
+	psrld     mm4,SCALEBITS		; mm4=CbEH
+	packssdw  mm5,mm4		; mm5=CbE
+
+	psllw     mm7,BYTE_BIT
+	por       mm5,mm7		; mm5=Cb
+	movq      MMWORD [ebx], mm5	; Save Cb
+
+	movq      mm0, MMWORD [wk(3)]	; mm0=BO
+	movq      mm6, MMWORD [wk(2)]	; mm6=BE
+	movq      mm1, MMWORD [wk(1)]	; mm1=RO
+
+	movq      mm4,mm0
+	punpcklwd mm0,mm3
+	punpckhwd mm4,mm3
+	movq      mm7,mm0
+	movq      mm5,mm4
+	pmaddwd   mm0,[GOTOFF(eax,PW_F0114_F0250)] ; mm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+	pmaddwd   mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+	pmaddwd   mm7,[GOTOFF(eax,PW_MF008_MF041)] ; mm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+	pmaddwd   mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+	movq      mm3,[GOTOFF(eax,PD_ONEHALF)]	; mm3=[PD_ONEHALF]
+
+	paddd     mm0, MMWORD [wk(4)]
+	paddd     mm4, MMWORD [wk(5)]
+	paddd     mm0,mm3
+	paddd     mm4,mm3
+	psrld     mm0,SCALEBITS		; mm0=YOL
+	psrld     mm4,SCALEBITS		; mm4=YOH
+	packssdw  mm0,mm4		; mm0=YO
+
+	pxor      mm3,mm3
+	pxor      mm4,mm4
+	punpcklwd mm3,mm1		; mm3=ROL
+	punpckhwd mm4,mm1		; mm4=ROH
+	psrld     mm3,1			; mm3=ROL*FIX(0.500)
+	psrld     mm4,1			; mm4=ROH*FIX(0.500)
+
+	movq      mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
+
+	paddd     mm7,mm3
+	paddd     mm5,mm4
+	paddd     mm7,mm1
+	paddd     mm5,mm1
+	psrld     mm7,SCALEBITS		; mm7=CrOL
+	psrld     mm5,SCALEBITS		; mm5=CrOH
+	packssdw  mm7,mm5		; mm7=CrO
+
+	movq      mm3, MMWORD [wk(0)]	; mm3=RE
+
+	movq      mm4,mm6
+	punpcklwd mm6,mm2
+	punpckhwd mm4,mm2
+	movq      mm1,mm6
+	movq      mm5,mm4
+	pmaddwd   mm6,[GOTOFF(eax,PW_F0114_F0250)] ; mm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+	pmaddwd   mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+	pmaddwd   mm1,[GOTOFF(eax,PW_MF008_MF041)] ; mm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+	pmaddwd   mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+	movq      mm2,[GOTOFF(eax,PD_ONEHALF)]	; mm2=[PD_ONEHALF]
+
+	paddd     mm6, MMWORD [wk(6)]
+	paddd     mm4, MMWORD [wk(7)]
+	paddd     mm6,mm2
+	paddd     mm4,mm2
+	psrld     mm6,SCALEBITS		; mm6=YEL
+	psrld     mm4,SCALEBITS		; mm4=YEH
+	packssdw  mm6,mm4		; mm6=YE
+
+	psllw     mm0,BYTE_BIT
+	por       mm6,mm0		; mm6=Y
+	movq      MMWORD [edi], mm6	; Save Y
+
+	pxor      mm2,mm2
+	pxor      mm4,mm4
+	punpcklwd mm2,mm3		; mm2=REL
+	punpckhwd mm4,mm3		; mm4=REH
+	psrld     mm2,1			; mm2=REL*FIX(0.500)
+	psrld     mm4,1			; mm4=REH*FIX(0.500)
+
+	movq      mm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm0=[PD_ONEHALFM1_CJ]
+
+	paddd     mm1,mm2
+	paddd     mm5,mm4
+	paddd     mm1,mm0
+	paddd     mm5,mm0
+	psrld     mm1,SCALEBITS		; mm1=CrEL
+	psrld     mm5,SCALEBITS		; mm5=CrEH
+	packssdw  mm1,mm5		; mm1=CrE
+
+	psllw     mm7,BYTE_BIT
+	por       mm1,mm7		; mm1=Cr
+	movq      MMWORD [edx], mm1	; Save Cr
+
+	sub	ecx, byte SIZEOF_MMWORD
+	add	esi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; inptr
+	add	edi, byte SIZEOF_MMWORD			; outptr0
+	add	ebx, byte SIZEOF_MMWORD			; outptr1
+	add	edx, byte SIZEOF_MMWORD			; outptr2
+	cmp	ecx, byte SIZEOF_MMWORD
+	jae	near .columnloop
+	test	ecx,ecx
+	jnz	near .column_ld1
+
+	pop	ecx			; col
+	pop	esi
+	pop	edi
+	pop	ebx
+	pop	edx
+	poppic	eax
+
+	add	esi, byte SIZEOF_JSAMPROW	; input_buf
+	add	edi, byte SIZEOF_JSAMPROW
+	add	ebx, byte SIZEOF_JSAMPROW
+	add	edx, byte SIZEOF_JSAMPROW
+	dec	eax				; num_rows
+	jg	near .rowloop
+
+	emms		; empty MMX state
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
diff --git a/simd/jcclrss2.asm b/simd/jcclrss2.asm
new file mode 100644
index 0000000..10b11f9
--- /dev/null
+++ b/simd/jcclrss2.asm
@@ -0,0 +1,500 @@
+;
+; jcclrss2.asm - colorspace conversion (SSE2)
+;
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
+;                             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+;                             JDIMENSION output_row, int num_rows);
+;
+
+%define img_width(b)	(b)+8			; JDIMENSION img_width
+%define input_buf(b)	(b)+12		; JSAMPARRAY input_buf
+%define output_buf(b)	(b)+16		; JSAMPIMAGE output_buf
+%define output_row(b)	(b)+20		; JDIMENSION output_row
+%define num_rows(b)	(b)+24		; int num_rows
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
+%define WK_NUM		8
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+
+	global	EXTN(jsimd_rgb_ycc_convert_sse2)
+
+EXTN(jsimd_rgb_ycc_convert_sse2):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [img_width(eax)]
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	esi, JSAMPIMAGE [output_buf(eax)]
+	mov	ecx, JDIMENSION [output_row(eax)]
+	mov	edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
+	lea	edi, [edi+ecx*SIZEOF_JSAMPROW]
+	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+	pop	ecx
+
+	mov	esi, JSAMPARRAY [input_buf(eax)]
+	mov	eax, INT [num_rows(eax)]
+	test	eax,eax
+	jle	near .return
+	alignx	16,7
+.rowloop:
+	pushpic	eax
+	push	edx
+	push	ebx
+	push	edi
+	push	esi
+	push	ecx			; col
+
+	mov	esi, JSAMPROW [esi]	; inptr
+	mov	edi, JSAMPROW [edi]	; outptr0
+	mov	ebx, JSAMPROW [ebx]	; outptr1
+	mov	edx, JSAMPROW [edx]	; outptr2
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+
+	cmp	ecx, byte SIZEOF_XMMWORD
+	jae	near .columnloop
+	alignx	16,7
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+.column_ld1:
+	push	eax
+	push	edx
+	lea	ecx,[ecx+ecx*2]		; imul ecx,RGB_PIXELSIZE
+	test	cl, SIZEOF_BYTE
+	jz	short .column_ld2
+	sub	ecx, byte SIZEOF_BYTE
+	movzx	eax, BYTE [esi+ecx]
+.column_ld2:
+	test	cl, SIZEOF_WORD
+	jz	short .column_ld4
+	sub	ecx, byte SIZEOF_WORD
+	movzx	edx, WORD [esi+ecx]
+	shl	eax, WORD_BIT
+	or	eax,edx
+.column_ld4:
+	movd	xmmA,eax
+	pop	edx
+	pop	eax
+	test	cl, SIZEOF_DWORD
+	jz	short .column_ld8
+	sub	ecx, byte SIZEOF_DWORD
+	movd	xmmF, XMM_DWORD [esi+ecx]
+	pslldq	xmmA, SIZEOF_DWORD
+	por	xmmA,xmmF
+.column_ld8:
+	test	cl, SIZEOF_MMWORD
+	jz	short .column_ld16
+	sub	ecx, byte SIZEOF_MMWORD
+	movq	xmmB, XMM_MMWORD [esi+ecx]
+	pslldq	xmmA, SIZEOF_MMWORD
+	por	xmmA,xmmB
+.column_ld16:
+	test	cl, SIZEOF_XMMWORD
+	jz	short .column_ld32
+	movdqa	xmmF,xmmA
+	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+	mov	ecx, SIZEOF_XMMWORD
+	jmp	short .rgb_ycc_cnv
+.column_ld32:
+	test	cl, 2*SIZEOF_XMMWORD
+	mov	ecx, SIZEOF_XMMWORD
+	jz	short .rgb_ycc_cnv
+	movdqa	xmmB,xmmA
+	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+	movdqu	xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+	jmp	short .rgb_ycc_cnv
+	alignx	16,7
+
+.columnloop:
+	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+	movdqu	xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
+	movdqu	xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+	; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+	; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+	movdqa    xmmG,xmmA
+	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
+	psrldq    xmmG,8	; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
+
+	punpckhbw xmmA,xmmF	; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
+	pslldq    xmmF,8	; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
+
+	punpcklbw xmmG,xmmB	; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
+	punpckhbw xmmF,xmmB	; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
+
+	movdqa    xmmD,xmmA
+	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
+	psrldq    xmmD,8	; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
+
+	punpckhbw xmmA,xmmG	; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
+	pslldq    xmmG,8	; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
+
+	punpcklbw xmmD,xmmF	; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
+	punpckhbw xmmG,xmmF	; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
+
+	movdqa    xmmE,xmmA
+	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
+	psrldq    xmmE,8	; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
+
+	punpckhbw xmmA,xmmD	; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+	pslldq    xmmD,8	; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
+
+	punpcklbw xmmE,xmmG	; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
+	punpckhbw xmmD,xmmG	; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
+
+	pxor      xmmH,xmmH
+
+	movdqa    xmmC,xmmA
+	punpcklbw xmmA,xmmH	; xmmA=(00 02 04 06 08 0A 0C 0E)
+	punpckhbw xmmC,xmmH	; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+	movdqa    xmmB,xmmE
+	punpcklbw xmmE,xmmH	; xmmE=(20 22 24 26 28 2A 2C 2E)
+	punpckhbw xmmB,xmmH	; xmmB=(01 03 05 07 09 0B 0D 0F)
+
+	movdqa    xmmF,xmmD
+	punpcklbw xmmD,xmmH	; xmmD=(11 13 15 17 19 1B 1D 1F)
+	punpckhbw xmmF,xmmH	; xmmF=(21 23 25 27 29 2B 2D 2F)
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+.column_ld1:
+	test	cl, SIZEOF_XMMWORD/16
+	jz	short .column_ld2
+	sub	ecx, byte SIZEOF_XMMWORD/16
+	movd	xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld2:
+	test	cl, SIZEOF_XMMWORD/8
+	jz	short .column_ld4
+	sub	ecx, byte SIZEOF_XMMWORD/8
+	movq	xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
+	pslldq	xmmA, SIZEOF_MMWORD
+	por	xmmA,xmmE
+.column_ld4:
+	test	cl, SIZEOF_XMMWORD/4
+	jz	short .column_ld8
+	sub	ecx, byte SIZEOF_XMMWORD/4
+	movdqa	xmmE,xmmA
+	movdqu	xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
+.column_ld8:
+	test	cl, SIZEOF_XMMWORD/2
+	mov	ecx, SIZEOF_XMMWORD
+	jz	short .rgb_ycc_cnv
+	movdqa	xmmF,xmmA
+	movdqa	xmmH,xmmE
+	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+	movdqu	xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+	jmp	short .rgb_ycc_cnv
+	alignx	16,7
+
+.columnloop:
+	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
+	movdqu	xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
+	movdqu	xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
+	movdqu	xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
+
+.rgb_ycc_cnv:
+	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+	; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+	; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+	movdqa    xmmD,xmmA
+	punpcklbw xmmA,xmmE	; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
+	punpckhbw xmmD,xmmE	; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
+
+	movdqa    xmmC,xmmF
+	punpcklbw xmmF,xmmH	; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
+	punpckhbw xmmC,xmmH	; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
+
+	movdqa    xmmB,xmmA
+	punpcklwd xmmA,xmmF	; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
+	punpckhwd xmmB,xmmF	; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
+
+	movdqa    xmmG,xmmD
+	punpcklwd xmmD,xmmC	; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
+	punpckhwd xmmG,xmmC	; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
+
+	movdqa    xmmE,xmmA
+	punpcklbw xmmA,xmmD	; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
+	punpckhbw xmmE,xmmD	; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
+
+	movdqa    xmmH,xmmB
+	punpcklbw xmmB,xmmG	; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
+	punpckhbw xmmH,xmmG	; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
+
+	pxor      xmmF,xmmF
+
+	movdqa    xmmC,xmmA
+	punpcklbw xmmA,xmmF	; xmmA=(00 02 04 06 08 0A 0C 0E)
+	punpckhbw xmmC,xmmF	; xmmC=(10 12 14 16 18 1A 1C 1E)
+
+	movdqa    xmmD,xmmB
+	punpcklbw xmmB,xmmF	; xmmB=(01 03 05 07 09 0B 0D 0F)
+	punpckhbw xmmD,xmmF	; xmmD=(11 13 15 17 19 1B 1D 1F)
+
+	movdqa    xmmG,xmmE
+	punpcklbw xmmE,xmmF	; xmmE=(20 22 24 26 28 2A 2C 2E)
+	punpckhbw xmmG,xmmF	; xmmG=(30 32 34 36 38 3A 3C 3E)
+
+	punpcklbw xmmF,xmmH
+	punpckhbw xmmH,xmmH
+	psrlw     xmmF,BYTE_BIT	; xmmF=(21 23 25 27 29 2B 2D 2F)
+	psrlw     xmmH,BYTE_BIT	; xmmH=(31 33 35 37 39 3B 3D 3F)
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+	; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
+	; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
+
+	; (Original)
+	; Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+	;
+	; (This implementation)
+	; Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
+	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
+	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
+
+	movdqa    XMMWORD [wk(0)], xmm0	; wk(0)=RE
+	movdqa    XMMWORD [wk(1)], xmm1	; wk(1)=RO
+	movdqa    XMMWORD [wk(2)], xmm4	; wk(2)=BE
+	movdqa    XMMWORD [wk(3)], xmm5	; wk(3)=BO
+
+	movdqa    xmm6,xmm1
+	punpcklwd xmm1,xmm3
+	punpckhwd xmm6,xmm3
+	movdqa    xmm7,xmm1
+	movdqa    xmm4,xmm6
+	pmaddwd   xmm1,[GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
+	pmaddwd   xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
+	pmaddwd   xmm7,[GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
+	pmaddwd   xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
+
+	movdqa    XMMWORD [wk(4)], xmm1	; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
+	movdqa    XMMWORD [wk(5)], xmm6	; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
+
+	pxor      xmm1,xmm1
+	pxor      xmm6,xmm6
+	punpcklwd xmm1,xmm5		; xmm1=BOL
+	punpckhwd xmm6,xmm5		; xmm6=BOH
+	psrld     xmm1,1		; xmm1=BOL*FIX(0.500)
+	psrld     xmm6,1		; xmm6=BOH*FIX(0.500)
+
+	movdqa    xmm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ]
+
+	paddd     xmm7,xmm1
+	paddd     xmm4,xmm6
+	paddd     xmm7,xmm5
+	paddd     xmm4,xmm5
+	psrld     xmm7,SCALEBITS	; xmm7=CbOL
+	psrld     xmm4,SCALEBITS	; xmm4=CbOH
+	packssdw  xmm7,xmm4		; xmm7=CbO
+
+	movdqa    xmm1, XMMWORD [wk(2)]	; xmm1=BE
+
+	movdqa    xmm6,xmm0
+	punpcklwd xmm0,xmm2
+	punpckhwd xmm6,xmm2
+	movdqa    xmm5,xmm0
+	movdqa    xmm4,xmm6
+	pmaddwd   xmm0,[GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
+	pmaddwd   xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
+	pmaddwd   xmm5,[GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
+	pmaddwd   xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
+
+	movdqa    XMMWORD [wk(6)], xmm0	; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
+	movdqa    XMMWORD [wk(7)], xmm6	; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
+
+	pxor      xmm0,xmm0
+	pxor      xmm6,xmm6
+	punpcklwd xmm0,xmm1		; xmm0=BEL
+	punpckhwd xmm6,xmm1		; xmm6=BEH
+	psrld     xmm0,1		; xmm0=BEL*FIX(0.500)
+	psrld     xmm6,1		; xmm6=BEH*FIX(0.500)
+
+	movdqa    xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
+
+	paddd     xmm5,xmm0
+	paddd     xmm4,xmm6
+	paddd     xmm5,xmm1
+	paddd     xmm4,xmm1
+	psrld     xmm5,SCALEBITS	; xmm5=CbEL
+	psrld     xmm4,SCALEBITS	; xmm4=CbEH
+	packssdw  xmm5,xmm4		; xmm5=CbE
+
+	psllw     xmm7,BYTE_BIT
+	por       xmm5,xmm7		; xmm5=Cb
+	movdqa    XMMWORD [ebx], xmm5	; Save Cb
+
+	movdqa    xmm0, XMMWORD [wk(3)]	; xmm0=BO
+	movdqa    xmm6, XMMWORD [wk(2)]	; xmm6=BE
+	movdqa    xmm1, XMMWORD [wk(1)]	; xmm1=RO
+
+	movdqa    xmm4,xmm0
+	punpcklwd xmm0,xmm3
+	punpckhwd xmm4,xmm3
+	movdqa    xmm7,xmm0
+	movdqa    xmm5,xmm4
+	pmaddwd   xmm0,[GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
+	pmaddwd   xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
+	pmaddwd   xmm7,[GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
+	pmaddwd   xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
+
+	movdqa    xmm3,[GOTOFF(eax,PD_ONEHALF)]	; xmm3=[PD_ONEHALF]
+
+	paddd     xmm0, XMMWORD [wk(4)]
+	paddd     xmm4, XMMWORD [wk(5)]
+	paddd     xmm0,xmm3
+	paddd     xmm4,xmm3
+	psrld     xmm0,SCALEBITS	; xmm0=YOL
+	psrld     xmm4,SCALEBITS	; xmm4=YOH
+	packssdw  xmm0,xmm4		; xmm0=YO
+
+	pxor      xmm3,xmm3
+	pxor      xmm4,xmm4
+	punpcklwd xmm3,xmm1		; xmm3=ROL
+	punpckhwd xmm4,xmm1		; xmm4=ROH
+	psrld     xmm3,1		; xmm3=ROL*FIX(0.500)
+	psrld     xmm4,1		; xmm4=ROH*FIX(0.500)
+
+	movdqa    xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
+
+	paddd     xmm7,xmm3
+	paddd     xmm5,xmm4
+	paddd     xmm7,xmm1
+	paddd     xmm5,xmm1
+	psrld     xmm7,SCALEBITS	; xmm7=CrOL
+	psrld     xmm5,SCALEBITS	; xmm5=CrOH
+	packssdw  xmm7,xmm5		; xmm7=CrO
+
+	movdqa    xmm3, XMMWORD [wk(0)]	; xmm3=RE
+
+	movdqa    xmm4,xmm6
+	punpcklwd xmm6,xmm2
+	punpckhwd xmm4,xmm2
+	movdqa    xmm1,xmm6
+	movdqa    xmm5,xmm4
+	pmaddwd   xmm6,[GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
+	pmaddwd   xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
+	pmaddwd   xmm1,[GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
+	pmaddwd   xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
+
+	movdqa    xmm2,[GOTOFF(eax,PD_ONEHALF)]	; xmm2=[PD_ONEHALF]
+
+	paddd     xmm6, XMMWORD [wk(6)]
+	paddd     xmm4, XMMWORD [wk(7)]
+	paddd     xmm6,xmm2
+	paddd     xmm4,xmm2
+	psrld     xmm6,SCALEBITS	; xmm6=YEL
+	psrld     xmm4,SCALEBITS	; xmm4=YEH
+	packssdw  xmm6,xmm4		; xmm6=YE
+
+	psllw     xmm0,BYTE_BIT
+	por       xmm6,xmm0		; xmm6=Y
+	movdqa    XMMWORD [edi], xmm6	; Save Y
+
+	pxor      xmm2,xmm2
+	pxor      xmm4,xmm4
+	punpcklwd xmm2,xmm3		; xmm2=REL
+	punpckhwd xmm4,xmm3		; xmm4=REH
+	psrld     xmm2,1		; xmm2=REL*FIX(0.500)
+	psrld     xmm4,1		; xmm4=REH*FIX(0.500)
+
+	movdqa    xmm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ]
+
+	paddd     xmm1,xmm2
+	paddd     xmm5,xmm4
+	paddd     xmm1,xmm0
+	paddd     xmm5,xmm0
+	psrld     xmm1,SCALEBITS	; xmm1=CrEL
+	psrld     xmm5,SCALEBITS	; xmm5=CrEH
+	packssdw  xmm1,xmm5		; xmm1=CrE
+
+	psllw     xmm7,BYTE_BIT
+	por       xmm1,xmm7		; xmm1=Cr
+	movdqa    XMMWORD [edx], xmm1	; Save Cr
+
+	sub	ecx, byte SIZEOF_XMMWORD
+	add	esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; inptr
+	add	edi, byte SIZEOF_XMMWORD		; outptr0
+	add	ebx, byte SIZEOF_XMMWORD		; outptr1
+	add	edx, byte SIZEOF_XMMWORD		; outptr2
+	cmp	ecx, byte SIZEOF_XMMWORD
+	jae	near .columnloop
+	test	ecx,ecx
+	jnz	near .column_ld1
+
+	pop	ecx			; col
+	pop	esi
+	pop	edi
+	pop	ebx
+	pop	edx
+	poppic	eax
+
+	add	esi, byte SIZEOF_JSAMPROW	; input_buf
+	add	edi, byte SIZEOF_JSAMPROW
+	add	ebx, byte SIZEOF_JSAMPROW
+	add	edx, byte SIZEOF_JSAMPROW
+	dec	eax				; num_rows
+	jg	near .rowloop
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
diff --git a/simd/jccolmmx.asm b/simd/jccolmmx.asm
index 340af00..8352fc3 100644
--- a/simd/jccolmmx.asm
+++ b/simd/jccolmmx.asm
@@ -2,6 +2,7 @@
 ; jccolmmx.asm - colorspace conversion (MMX)
 ;
 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009 D. R. Commander
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
@@ -51,458 +52,70 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_ycc_convert_mmx (JDIMENSION img_width,
-;                           JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-;                           JDIMENSION output_row, int num_rows);
-;
+%include "jcclrmmx.asm"
 
-%define img_width(b)	(b)+8			; JDIMENSION img_width
-%define input_buf(b)	(b)+12		; JSAMPARRAY input_buf
-%define output_buf(b)	(b)+16		; JSAMPIMAGE output_buf
-%define output_row(b)	(b)+20		; JDIMENSION output_row
-%define num_rows(b)	(b)+24		; int num_rows
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_rgb_ycc_convert_mmx jsimd_extrgb_ycc_convert_mmx
+%include "jcclrmmx.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
-%define WK_NUM		8
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_mmx jsimd_extrgbx_ycc_convert_mmx
+%include "jcclrmmx.asm"
 
-	align	16
-	global	EXTN(jsimd_rgb_ycc_convert_mmx)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_rgb_ycc_convert_mmx jsimd_extbgr_ycc_convert_mmx
+%include "jcclrmmx.asm"
 
-EXTN(jsimd_rgb_ycc_convert_mmx):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_mmx jsimd_extbgrx_ycc_convert_mmx
+%include "jcclrmmx.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_mmx jsimd_extxbgr_ycc_convert_mmx
+%include "jcclrmmx.asm"
 
-	mov	ecx, JDIMENSION [img_width(eax)]	; num_cols
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	esi, JSAMPIMAGE [output_buf(eax)]
-	mov	ecx, JDIMENSION [output_row(eax)]
-	mov	edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
-	lea	edi, [edi+ecx*SIZEOF_JSAMPROW]
-	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
-	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
-
-	pop	ecx
-
-	mov	esi, JSAMPARRAY [input_buf(eax)]
-	mov	eax, INT [num_rows(eax)]
-	test	eax,eax
-	jle	near .return
-	alignx	16,7
-.rowloop:
-	pushpic	eax
-	push	edx
-	push	ebx
-	push	edi
-	push	esi
-	push	ecx			; col
-
-	mov	esi, JSAMPROW [esi]	; inptr
-	mov	edi, JSAMPROW [edi]	; outptr0
-	mov	ebx, JSAMPROW [ebx]	; outptr1
-	mov	edx, JSAMPROW [edx]	; outptr2
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-
-	cmp	ecx, byte SIZEOF_MMWORD
-	jae	short .columnloop
-	alignx	16,7
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
-	push	eax
-	push	edx
-	lea	ecx,[ecx+ecx*2]		; imul ecx,RGB_PIXELSIZE
-	test	cl, SIZEOF_BYTE
-	jz	short .column_ld2
-	sub	ecx, byte SIZEOF_BYTE
-	xor	eax,eax
-	mov	al, BYTE [esi+ecx]
-.column_ld2:
-	test	cl, SIZEOF_WORD
-	jz	short .column_ld4
-	sub	ecx, byte SIZEOF_WORD
-	xor	edx,edx
-	mov	dx, WORD [esi+ecx]
-	shl	eax, WORD_BIT
-	or	eax,edx
-.column_ld4:
-	movd	mmA,eax
-	pop	edx
-	pop	eax
-	test	cl, SIZEOF_DWORD
-	jz	short .column_ld8
-	sub	ecx, byte SIZEOF_DWORD
-	movd	mmG, DWORD [esi+ecx]
-	psllq	mmA, DWORD_BIT
-	por	mmA,mmG
-.column_ld8:
-	test	cl, SIZEOF_MMWORD
-	jz	short .column_ld16
-	movq	mmG,mmA
-	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
-	mov	ecx, SIZEOF_MMWORD
-	jmp	short .rgb_ycc_cnv
-.column_ld16:
-	test	cl, 2*SIZEOF_MMWORD
-	mov	ecx, SIZEOF_MMWORD
-	jz	short .rgb_ycc_cnv
-	movq	mmF,mmA
-	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
-	movq	mmG, MMWORD [esi+1*SIZEOF_MMWORD]
-	jmp	short .rgb_ycc_cnv
-	alignx	16,7
-
-.columnloop:
-	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
-	movq	mmG, MMWORD [esi+1*SIZEOF_MMWORD]
-	movq	mmF, MMWORD [esi+2*SIZEOF_MMWORD]
-
-.rgb_ycc_cnv:
-	; mmA=(00 10 20 01 11 21 02 12)
-	; mmG=(22 03 13 23 04 14 24 05)
-	; mmF=(15 25 06 16 26 07 17 27)
-
-	movq      mmD,mmA
-	psllq     mmA,4*BYTE_BIT	; mmA=(-- -- -- -- 00 10 20 01)
-	psrlq     mmD,4*BYTE_BIT	; mmD=(11 21 02 12 -- -- -- --)
-
-	punpckhbw mmA,mmG		; mmA=(00 04 10 14 20 24 01 05)
-	psllq     mmG,4*BYTE_BIT	; mmG=(-- -- -- -- 22 03 13 23)
-
-	punpcklbw mmD,mmF		; mmD=(11 15 21 25 02 06 12 16)
-	punpckhbw mmG,mmF		; mmG=(22 26 03 07 13 17 23 27)
-
-	movq      mmE,mmA
-	psllq     mmA,4*BYTE_BIT	; mmA=(-- -- -- -- 00 04 10 14)
-	psrlq     mmE,4*BYTE_BIT	; mmE=(20 24 01 05 -- -- -- --)
-
-	punpckhbw mmA,mmD		; mmA=(00 02 04 06 10 12 14 16)
-	psllq     mmD,4*BYTE_BIT	; mmD=(-- -- -- -- 11 15 21 25)
-
-	punpcklbw mmE,mmG		; mmE=(20 22 24 26 01 03 05 07)
-	punpckhbw mmD,mmG		; mmD=(11 13 15 17 21 23 25 27)
-
-	pxor      mmH,mmH
-
-	movq      mmC,mmA
-	punpcklbw mmA,mmH		; mmA=(00 02 04 06)
-	punpckhbw mmC,mmH		; mmC=(10 12 14 16)
-
-	movq      mmB,mmE
-	punpcklbw mmE,mmH		; mmE=(20 22 24 26)
-	punpckhbw mmB,mmH		; mmB=(01 03 05 07)
-
-	movq      mmF,mmD
-	punpcklbw mmD,mmH		; mmD=(11 13 15 17)
-	punpckhbw mmF,mmH		; mmF=(21 23 25 27)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
-	test	cl, SIZEOF_MMWORD/8
-	jz	short .column_ld2
-	sub	ecx, byte SIZEOF_MMWORD/8
-	movd	mmA, DWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld2:
-	test	cl, SIZEOF_MMWORD/4
-	jz	short .column_ld4
-	sub	ecx, byte SIZEOF_MMWORD/4
-	movq	mmF,mmA
-	movq	mmA, MMWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld4:
-	test	cl, SIZEOF_MMWORD/2
-	mov	ecx, SIZEOF_MMWORD
-	jz	short .rgb_ycc_cnv
-	movq	mmD,mmA
-	movq	mmC,mmF
-	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
-	movq	mmF, MMWORD [esi+1*SIZEOF_MMWORD]
-	jmp	short .rgb_ycc_cnv
-	alignx	16,7
-
-.columnloop:
-	movq	mmA, MMWORD [esi+0*SIZEOF_MMWORD]
-	movq	mmF, MMWORD [esi+1*SIZEOF_MMWORD]
-	movq	mmD, MMWORD [esi+2*SIZEOF_MMWORD]
-	movq	mmC, MMWORD [esi+3*SIZEOF_MMWORD]
-
-.rgb_ycc_cnv:
-	; mmA=(00 10 20 30 01 11 21 31)
-	; mmF=(02 12 22 32 03 13 23 33)
-	; mmD=(04 14 24 34 05 15 25 35)
-	; mmC=(06 16 26 36 07 17 27 37)
-
-	movq      mmB,mmA
-	punpcklbw mmA,mmF		; mmA=(00 02 10 12 20 22 30 32)
-	punpckhbw mmB,mmF		; mmB=(01 03 11 13 21 23 31 33)
-
-	movq      mmG,mmD
-	punpcklbw mmD,mmC		; mmD=(04 06 14 16 24 26 34 36)
-	punpckhbw mmG,mmC		; mmG=(05 07 15 17 25 27 35 37)
-
-	movq      mmE,mmA
-	punpcklwd mmA,mmD		; mmA=(00 02 04 06 10 12 14 16)
-	punpckhwd mmE,mmD		; mmE=(20 22 24 26 30 32 34 36)
-
-	movq      mmH,mmB
-	punpcklwd mmB,mmG		; mmB=(01 03 05 07 11 13 15 17)
-	punpckhwd mmH,mmG		; mmH=(21 23 25 27 31 33 35 37)
-
-	pxor      mmF,mmF
-
-	movq      mmC,mmA
-	punpcklbw mmA,mmF		; mmA=(00 02 04 06)
-	punpckhbw mmC,mmF		; mmC=(10 12 14 16)
-
-	movq      mmD,mmB
-	punpcklbw mmB,mmF		; mmB=(01 03 05 07)
-	punpckhbw mmD,mmF		; mmD=(11 13 15 17)
-
-	movq      mmG,mmE
-	punpcklbw mmE,mmF		; mmE=(20 22 24 26)
-	punpckhbw mmG,mmF		; mmG=(30 32 34 36)
-
-	punpcklbw mmF,mmH
-	punpckhbw mmH,mmH
-	psrlw     mmF,BYTE_BIT		; mmF=(21 23 25 27)
-	psrlw     mmH,BYTE_BIT		; mmH=(31 33 35 37)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-	; mm0=(R0 R2 R4 R6)=RE, mm2=(G0 G2 G4 G6)=GE, mm4=(B0 B2 B4 B6)=BE
-	; mm1=(R1 R3 R5 R7)=RO, mm3=(G1 G3 G5 G7)=GO, mm5=(B1 B3 B5 B7)=BO
-
-	; (Original)
-	; Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
-	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
-	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-	;
-	; (This implementation)
-	; Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
-	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
-	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-
-	movq      MMWORD [wk(0)], mm0	; wk(0)=RE
-	movq      MMWORD [wk(1)], mm1	; wk(1)=RO
-	movq      MMWORD [wk(2)], mm4	; wk(2)=BE
-	movq      MMWORD [wk(3)], mm5	; wk(3)=BO
-
-	movq      mm6,mm1
-	punpcklwd mm1,mm3
-	punpckhwd mm6,mm3
-	movq      mm7,mm1
-	movq      mm4,mm6
-	pmaddwd   mm1,[GOTOFF(eax,PW_F0299_F0337)] ; mm1=ROL*FIX(0.299)+GOL*FIX(0.337)
-	pmaddwd   mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=ROH*FIX(0.299)+GOH*FIX(0.337)
-	pmaddwd   mm7,[GOTOFF(eax,PW_MF016_MF033)] ; mm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
-	pmaddwd   mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
-
-	movq      MMWORD [wk(4)], mm1	; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
-	movq      MMWORD [wk(5)], mm6	; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
-
-	pxor      mm1,mm1
-	pxor      mm6,mm6
-	punpcklwd mm1,mm5		; mm1=BOL
-	punpckhwd mm6,mm5		; mm6=BOH
-	psrld     mm1,1			; mm1=BOL*FIX(0.500)
-	psrld     mm6,1			; mm6=BOH*FIX(0.500)
-
-	movq      mm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm5=[PD_ONEHALFM1_CJ]
-
-	paddd     mm7,mm1
-	paddd     mm4,mm6
-	paddd     mm7,mm5
-	paddd     mm4,mm5
-	psrld     mm7,SCALEBITS		; mm7=CbOL
-	psrld     mm4,SCALEBITS		; mm4=CbOH
-	packssdw  mm7,mm4		; mm7=CbO
-
-	movq      mm1, MMWORD [wk(2)]	; mm1=BE
-
-	movq      mm6,mm0
-	punpcklwd mm0,mm2
-	punpckhwd mm6,mm2
-	movq      mm5,mm0
-	movq      mm4,mm6
-	pmaddwd   mm0,[GOTOFF(eax,PW_F0299_F0337)] ; mm0=REL*FIX(0.299)+GEL*FIX(0.337)
-	pmaddwd   mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=REH*FIX(0.299)+GEH*FIX(0.337)
-	pmaddwd   mm5,[GOTOFF(eax,PW_MF016_MF033)] ; mm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
-	pmaddwd   mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
-
-	movq      MMWORD [wk(6)], mm0	; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
-	movq      MMWORD [wk(7)], mm6	; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
-
-	pxor      mm0,mm0
-	pxor      mm6,mm6
-	punpcklwd mm0,mm1		; mm0=BEL
-	punpckhwd mm6,mm1		; mm6=BEH
-	psrld     mm0,1			; mm0=BEL*FIX(0.500)
-	psrld     mm6,1			; mm6=BEH*FIX(0.500)
-
-	movq      mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
-
-	paddd     mm5,mm0
-	paddd     mm4,mm6
-	paddd     mm5,mm1
-	paddd     mm4,mm1
-	psrld     mm5,SCALEBITS		; mm5=CbEL
-	psrld     mm4,SCALEBITS		; mm4=CbEH
-	packssdw  mm5,mm4		; mm5=CbE
-
-	psllw     mm7,BYTE_BIT
-	por       mm5,mm7		; mm5=Cb
-	movq      MMWORD [ebx], mm5	; Save Cb
-
-	movq      mm0, MMWORD [wk(3)]	; mm0=BO
-	movq      mm6, MMWORD [wk(2)]	; mm6=BE
-	movq      mm1, MMWORD [wk(1)]	; mm1=RO
-
-	movq      mm4,mm0
-	punpcklwd mm0,mm3
-	punpckhwd mm4,mm3
-	movq      mm7,mm0
-	movq      mm5,mm4
-	pmaddwd   mm0,[GOTOFF(eax,PW_F0114_F0250)] ; mm0=BOL*FIX(0.114)+GOL*FIX(0.250)
-	pmaddwd   mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BOH*FIX(0.114)+GOH*FIX(0.250)
-	pmaddwd   mm7,[GOTOFF(eax,PW_MF008_MF041)] ; mm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
-	pmaddwd   mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
-
-	movq      mm3,[GOTOFF(eax,PD_ONEHALF)]	; mm3=[PD_ONEHALF]
-
-	paddd     mm0, MMWORD [wk(4)]
-	paddd     mm4, MMWORD [wk(5)]
-	paddd     mm0,mm3
-	paddd     mm4,mm3
-	psrld     mm0,SCALEBITS		; mm0=YOL
-	psrld     mm4,SCALEBITS		; mm4=YOH
-	packssdw  mm0,mm4		; mm0=YO
-
-	pxor      mm3,mm3
-	pxor      mm4,mm4
-	punpcklwd mm3,mm1		; mm3=ROL
-	punpckhwd mm4,mm1		; mm4=ROH
-	psrld     mm3,1			; mm3=ROL*FIX(0.500)
-	psrld     mm4,1			; mm4=ROH*FIX(0.500)
-
-	movq      mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
-
-	paddd     mm7,mm3
-	paddd     mm5,mm4
-	paddd     mm7,mm1
-	paddd     mm5,mm1
-	psrld     mm7,SCALEBITS		; mm7=CrOL
-	psrld     mm5,SCALEBITS		; mm5=CrOH
-	packssdw  mm7,mm5		; mm7=CrO
-
-	movq      mm3, MMWORD [wk(0)]	; mm3=RE
-
-	movq      mm4,mm6
-	punpcklwd mm6,mm2
-	punpckhwd mm4,mm2
-	movq      mm1,mm6
-	movq      mm5,mm4
-	pmaddwd   mm6,[GOTOFF(eax,PW_F0114_F0250)] ; mm6=BEL*FIX(0.114)+GEL*FIX(0.250)
-	pmaddwd   mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BEH*FIX(0.114)+GEH*FIX(0.250)
-	pmaddwd   mm1,[GOTOFF(eax,PW_MF008_MF041)] ; mm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
-	pmaddwd   mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
-
-	movq      mm2,[GOTOFF(eax,PD_ONEHALF)]	; mm2=[PD_ONEHALF]
-
-	paddd     mm6, MMWORD [wk(6)]
-	paddd     mm4, MMWORD [wk(7)]
-	paddd     mm6,mm2
-	paddd     mm4,mm2
-	psrld     mm6,SCALEBITS		; mm6=YEL
-	psrld     mm4,SCALEBITS		; mm4=YEH
-	packssdw  mm6,mm4		; mm6=YE
-
-	psllw     mm0,BYTE_BIT
-	por       mm6,mm0		; mm6=Y
-	movq      MMWORD [edi], mm6	; Save Y
-
-	pxor      mm2,mm2
-	pxor      mm4,mm4
-	punpcklwd mm2,mm3		; mm2=REL
-	punpckhwd mm4,mm3		; mm4=REH
-	psrld     mm2,1			; mm2=REL*FIX(0.500)
-	psrld     mm4,1			; mm4=REH*FIX(0.500)
-
-	movq      mm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm0=[PD_ONEHALFM1_CJ]
-
-	paddd     mm1,mm2
-	paddd     mm5,mm4
-	paddd     mm1,mm0
-	paddd     mm5,mm0
-	psrld     mm1,SCALEBITS		; mm1=CrEL
-	psrld     mm5,SCALEBITS		; mm5=CrEH
-	packssdw  mm1,mm5		; mm1=CrE
-
-	psllw     mm7,BYTE_BIT
-	por       mm1,mm7		; mm1=Cr
-	movq      MMWORD [edx], mm1	; Save Cr
-
-	sub	ecx, byte SIZEOF_MMWORD
-	add	esi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; inptr
-	add	edi, byte SIZEOF_MMWORD			; outptr0
-	add	ebx, byte SIZEOF_MMWORD			; outptr1
-	add	edx, byte SIZEOF_MMWORD			; outptr2
-	cmp	ecx, byte SIZEOF_MMWORD
-	jae	near .columnloop
-	test	ecx,ecx
-	jnz	near .column_ld1
-
-	pop	ecx			; col
-	pop	esi
-	pop	edi
-	pop	ebx
-	pop	edx
-	poppic	eax
-
-	add	esi, byte SIZEOF_JSAMPROW	; input_buf
-	add	edi, byte SIZEOF_JSAMPROW
-	add	ebx, byte SIZEOF_JSAMPROW
-	add	edx, byte SIZEOF_JSAMPROW
-	dec	eax				; num_rows
-	jg	near .rowloop
-
-	emms		; empty MMX state
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_mmx jsimd_extxrgb_ycc_convert_mmx
+%include "jcclrmmx.asm"
diff --git a/simd/jccolss2.asm b/simd/jccolss2.asm
index 4d06bb8..2d7bb57 100644
--- a/simd/jccolss2.asm
+++ b/simd/jccolss2.asm
@@ -3,6 +3,7 @@
 ;
 ; x86 SIMD extension for IJG JPEG library
 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; Copyright (C) 2009, D. R. Commander.
 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
 ;
 ; This file should be assembled with NASM (Netwide Assembler),
@@ -48,486 +49,70 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
-;                             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
-;                             JDIMENSION output_row, int num_rows);
-;
+%include "jcclrss2.asm"
 
-%define img_width(b)	(b)+8			; JDIMENSION img_width
-%define input_buf(b)	(b)+12		; JSAMPARRAY input_buf
-%define output_buf(b)	(b)+16		; JSAMPIMAGE output_buf
-%define output_row(b)	(b)+20		; JDIMENSION output_row
-%define num_rows(b)	(b)+24		; int num_rows
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgb_ycc_convert_sse2
+%include "jcclrss2.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
-%define WK_NUM		8
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgbx_ycc_convert_sse2
+%include "jcclrss2.asm"
 
-	align	16
-	global	EXTN(jsimd_rgb_ycc_convert_sse2)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgr_ycc_convert_sse2
+%include "jcclrss2.asm"
 
-EXTN(jsimd_rgb_ycc_convert_sse2):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgrx_ycc_convert_sse2
+%include "jcclrss2.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extxbgr_ycc_convert_sse2
+%include "jcclrss2.asm"
 
-	mov	ecx, JDIMENSION [img_width(eax)]
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	esi, JSAMPIMAGE [output_buf(eax)]
-	mov	ecx, JDIMENSION [output_row(eax)]
-	mov	edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
-	lea	edi, [edi+ecx*SIZEOF_JSAMPROW]
-	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
-	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
-
-	pop	ecx
-
-	mov	esi, JSAMPARRAY [input_buf(eax)]
-	mov	eax, INT [num_rows(eax)]
-	test	eax,eax
-	jle	near .return
-	alignx	16,7
-.rowloop:
-	pushpic	eax
-	push	edx
-	push	ebx
-	push	edi
-	push	esi
-	push	ecx			; col
-
-	mov	esi, JSAMPROW [esi]	; inptr
-	mov	edi, JSAMPROW [edi]	; outptr0
-	mov	ebx, JSAMPROW [ebx]	; outptr1
-	mov	edx, JSAMPROW [edx]	; outptr2
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-
-	cmp	ecx, byte SIZEOF_XMMWORD
-	jae	near .columnloop
-	alignx	16,7
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-.column_ld1:
-	push	eax
-	push	edx
-	lea	ecx,[ecx+ecx*2]		; imul ecx,RGB_PIXELSIZE
-	test	cl, SIZEOF_BYTE
-	jz	short .column_ld2
-	sub	ecx, byte SIZEOF_BYTE
-	movzx	eax, BYTE [esi+ecx]
-.column_ld2:
-	test	cl, SIZEOF_WORD
-	jz	short .column_ld4
-	sub	ecx, byte SIZEOF_WORD
-	movzx	edx, WORD [esi+ecx]
-	shl	eax, WORD_BIT
-	or	eax,edx
-.column_ld4:
-	movd	xmmA,eax
-	pop	edx
-	pop	eax
-	test	cl, SIZEOF_DWORD
-	jz	short .column_ld8
-	sub	ecx, byte SIZEOF_DWORD
-	movd	xmmF, XMM_DWORD [esi+ecx]
-	pslldq	xmmA, SIZEOF_DWORD
-	por	xmmA,xmmF
-.column_ld8:
-	test	cl, SIZEOF_MMWORD
-	jz	short .column_ld16
-	sub	ecx, byte SIZEOF_MMWORD
-	movq	xmmB, XMM_MMWORD [esi+ecx]
-	pslldq	xmmA, SIZEOF_MMWORD
-	por	xmmA,xmmB
-.column_ld16:
-	test	cl, SIZEOF_XMMWORD
-	jz	short .column_ld32
-	movdqa	xmmF,xmmA
-	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
-	mov	ecx, SIZEOF_XMMWORD
-	jmp	short .rgb_ycc_cnv
-.column_ld32:
-	test	cl, 2*SIZEOF_XMMWORD
-	mov	ecx, SIZEOF_XMMWORD
-	jz	short .rgb_ycc_cnv
-	movdqa	xmmB,xmmA
-	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
-	movdqu	xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
-	jmp	short .rgb_ycc_cnv
-	alignx	16,7
-
-.columnloop:
-	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
-	movdqu	xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
-	movdqu	xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
-	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
-	; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
-	; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
-	movdqa    xmmG,xmmA
-	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
-	psrldq    xmmG,8	; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
-
-	punpckhbw xmmA,xmmF	; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
-	pslldq    xmmF,8	; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
-
-	punpcklbw xmmG,xmmB	; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
-	punpckhbw xmmF,xmmB	; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
-
-	movdqa    xmmD,xmmA
-	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
-	psrldq    xmmD,8	; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
-
-	punpckhbw xmmA,xmmG	; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
-	pslldq    xmmG,8	; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
-
-	punpcklbw xmmD,xmmF	; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
-	punpckhbw xmmG,xmmF	; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
-
-	movdqa    xmmE,xmmA
-	pslldq    xmmA,8	; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
-	psrldq    xmmE,8	; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
-
-	punpckhbw xmmA,xmmD	; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
-	pslldq    xmmD,8	; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
-
-	punpcklbw xmmE,xmmG	; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
-	punpckhbw xmmD,xmmG	; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
-
-	pxor      xmmH,xmmH
-
-	movdqa    xmmC,xmmA
-	punpcklbw xmmA,xmmH	; xmmA=(00 02 04 06 08 0A 0C 0E)
-	punpckhbw xmmC,xmmH	; xmmC=(10 12 14 16 18 1A 1C 1E)
-
-	movdqa    xmmB,xmmE
-	punpcklbw xmmE,xmmH	; xmmE=(20 22 24 26 28 2A 2C 2E)
-	punpckhbw xmmB,xmmH	; xmmB=(01 03 05 07 09 0B 0D 0F)
-
-	movdqa    xmmF,xmmD
-	punpcklbw xmmD,xmmH	; xmmD=(11 13 15 17 19 1B 1D 1F)
-	punpckhbw xmmF,xmmH	; xmmF=(21 23 25 27 29 2B 2D 2F)
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-.column_ld1:
-	test	cl, SIZEOF_XMMWORD/16
-	jz	short .column_ld2
-	sub	ecx, byte SIZEOF_XMMWORD/16
-	movd	xmmA, DWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld2:
-	test	cl, SIZEOF_XMMWORD/8
-	jz	short .column_ld4
-	sub	ecx, byte SIZEOF_XMMWORD/8
-	movq	xmmE, MMWORD [esi+ecx*RGB_PIXELSIZE]
-	pslldq	xmmA, SIZEOF_MMWORD
-	por	xmmA,xmmE
-.column_ld4:
-	test	cl, SIZEOF_XMMWORD/4
-	jz	short .column_ld8
-	sub	ecx, byte SIZEOF_XMMWORD/4
-	movdqa	xmmE,xmmA
-	movdqu	xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
-.column_ld8:
-	test	cl, SIZEOF_XMMWORD/2
-	mov	ecx, SIZEOF_XMMWORD
-	jz	short .rgb_ycc_cnv
-	movdqa	xmmF,xmmA
-	movdqa	xmmH,xmmE
-	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
-	movdqu	xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
-	jmp	short .rgb_ycc_cnv
-	alignx	16,7
-
-.columnloop:
-	movdqu	xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
-	movdqu	xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
-	movdqu	xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
-	movdqu	xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
-
-.rgb_ycc_cnv:
-	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
-	; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
-	; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
-	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
-	movdqa    xmmD,xmmA
-	punpcklbw xmmA,xmmE	; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
-	punpckhbw xmmD,xmmE	; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
-
-	movdqa    xmmC,xmmF
-	punpcklbw xmmF,xmmH	; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
-	punpckhbw xmmC,xmmH	; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
-
-	movdqa    xmmB,xmmA
-	punpcklwd xmmA,xmmF	; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
-	punpckhwd xmmB,xmmF	; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
-
-	movdqa    xmmG,xmmD
-	punpcklwd xmmD,xmmC	; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
-	punpckhwd xmmG,xmmC	; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
-
-	movdqa    xmmE,xmmA
-	punpcklbw xmmA,xmmD	; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
-	punpckhbw xmmE,xmmD	; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
-
-	movdqa    xmmH,xmmB
-	punpcklbw xmmB,xmmG	; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
-	punpckhbw xmmH,xmmG	; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
-
-	pxor      xmmF,xmmF
-
-	movdqa    xmmC,xmmA
-	punpcklbw xmmA,xmmF	; xmmA=(00 02 04 06 08 0A 0C 0E)
-	punpckhbw xmmC,xmmF	; xmmC=(10 12 14 16 18 1A 1C 1E)
-
-	movdqa    xmmD,xmmB
-	punpcklbw xmmB,xmmF	; xmmB=(01 03 05 07 09 0B 0D 0F)
-	punpckhbw xmmD,xmmF	; xmmD=(11 13 15 17 19 1B 1D 1F)
-
-	movdqa    xmmG,xmmE
-	punpcklbw xmmE,xmmF	; xmmE=(20 22 24 26 28 2A 2C 2E)
-	punpckhbw xmmG,xmmF	; xmmG=(30 32 34 36 38 3A 3C 3E)
-
-	punpcklbw xmmF,xmmH
-	punpckhbw xmmH,xmmH
-	psrlw     xmmF,BYTE_BIT	; xmmF=(21 23 25 27 29 2B 2D 2F)
-	psrlw     xmmH,BYTE_BIT	; xmmH=(31 33 35 37 39 3B 3D 3F)
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-	; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
-	; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
-
-	; (Original)
-	; Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
-	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
-	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-	;
-	; (This implementation)
-	; Y  =  0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
-	; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
-	; Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
-
-	movdqa    XMMWORD [wk(0)], xmm0	; wk(0)=RE
-	movdqa    XMMWORD [wk(1)], xmm1	; wk(1)=RO
-	movdqa    XMMWORD [wk(2)], xmm4	; wk(2)=BE
-	movdqa    XMMWORD [wk(3)], xmm5	; wk(3)=BO
-
-	movdqa    xmm6,xmm1
-	punpcklwd xmm1,xmm3
-	punpckhwd xmm6,xmm3
-	movdqa    xmm7,xmm1
-	movdqa    xmm4,xmm6
-	pmaddwd   xmm1,[GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
-	pmaddwd   xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
-	pmaddwd   xmm7,[GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
-	pmaddwd   xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
-
-	movdqa    XMMWORD [wk(4)], xmm1	; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
-	movdqa    XMMWORD [wk(5)], xmm6	; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
-
-	pxor      xmm1,xmm1
-	pxor      xmm6,xmm6
-	punpcklwd xmm1,xmm5		; xmm1=BOL
-	punpckhwd xmm6,xmm5		; xmm6=BOH
-	psrld     xmm1,1		; xmm1=BOL*FIX(0.500)
-	psrld     xmm6,1		; xmm6=BOH*FIX(0.500)
-
-	movdqa    xmm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ]
-
-	paddd     xmm7,xmm1
-	paddd     xmm4,xmm6
-	paddd     xmm7,xmm5
-	paddd     xmm4,xmm5
-	psrld     xmm7,SCALEBITS	; xmm7=CbOL
-	psrld     xmm4,SCALEBITS	; xmm4=CbOH
-	packssdw  xmm7,xmm4		; xmm7=CbO
-
-	movdqa    xmm1, XMMWORD [wk(2)]	; xmm1=BE
-
-	movdqa    xmm6,xmm0
-	punpcklwd xmm0,xmm2
-	punpckhwd xmm6,xmm2
-	movdqa    xmm5,xmm0
-	movdqa    xmm4,xmm6
-	pmaddwd   xmm0,[GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
-	pmaddwd   xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
-	pmaddwd   xmm5,[GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
-	pmaddwd   xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
-
-	movdqa    XMMWORD [wk(6)], xmm0	; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
-	movdqa    XMMWORD [wk(7)], xmm6	; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
-
-	pxor      xmm0,xmm0
-	pxor      xmm6,xmm6
-	punpcklwd xmm0,xmm1		; xmm0=BEL
-	punpckhwd xmm6,xmm1		; xmm6=BEH
-	psrld     xmm0,1		; xmm0=BEL*FIX(0.500)
-	psrld     xmm6,1		; xmm6=BEH*FIX(0.500)
-
-	movdqa    xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
-
-	paddd     xmm5,xmm0
-	paddd     xmm4,xmm6
-	paddd     xmm5,xmm1
-	paddd     xmm4,xmm1
-	psrld     xmm5,SCALEBITS	; xmm5=CbEL
-	psrld     xmm4,SCALEBITS	; xmm4=CbEH
-	packssdw  xmm5,xmm4		; xmm5=CbE
-
-	psllw     xmm7,BYTE_BIT
-	por       xmm5,xmm7		; xmm5=Cb
-	movdqa    XMMWORD [ebx], xmm5	; Save Cb
-
-	movdqa    xmm0, XMMWORD [wk(3)]	; xmm0=BO
-	movdqa    xmm6, XMMWORD [wk(2)]	; xmm6=BE
-	movdqa    xmm1, XMMWORD [wk(1)]	; xmm1=RO
-
-	movdqa    xmm4,xmm0
-	punpcklwd xmm0,xmm3
-	punpckhwd xmm4,xmm3
-	movdqa    xmm7,xmm0
-	movdqa    xmm5,xmm4
-	pmaddwd   xmm0,[GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
-	pmaddwd   xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
-	pmaddwd   xmm7,[GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
-	pmaddwd   xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
-
-	movdqa    xmm3,[GOTOFF(eax,PD_ONEHALF)]	; xmm3=[PD_ONEHALF]
-
-	paddd     xmm0, XMMWORD [wk(4)]
-	paddd     xmm4, XMMWORD [wk(5)]
-	paddd     xmm0,xmm3
-	paddd     xmm4,xmm3
-	psrld     xmm0,SCALEBITS	; xmm0=YOL
-	psrld     xmm4,SCALEBITS	; xmm4=YOH
-	packssdw  xmm0,xmm4		; xmm0=YO
-
-	pxor      xmm3,xmm3
-	pxor      xmm4,xmm4
-	punpcklwd xmm3,xmm1		; xmm3=ROL
-	punpckhwd xmm4,xmm1		; xmm4=ROH
-	psrld     xmm3,1		; xmm3=ROL*FIX(0.500)
-	psrld     xmm4,1		; xmm4=ROH*FIX(0.500)
-
-	movdqa    xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
-
-	paddd     xmm7,xmm3
-	paddd     xmm5,xmm4
-	paddd     xmm7,xmm1
-	paddd     xmm5,xmm1
-	psrld     xmm7,SCALEBITS	; xmm7=CrOL
-	psrld     xmm5,SCALEBITS	; xmm5=CrOH
-	packssdw  xmm7,xmm5		; xmm7=CrO
-
-	movdqa    xmm3, XMMWORD [wk(0)]	; xmm3=RE
-
-	movdqa    xmm4,xmm6
-	punpcklwd xmm6,xmm2
-	punpckhwd xmm4,xmm2
-	movdqa    xmm1,xmm6
-	movdqa    xmm5,xmm4
-	pmaddwd   xmm6,[GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
-	pmaddwd   xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
-	pmaddwd   xmm1,[GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
-	pmaddwd   xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
-
-	movdqa    xmm2,[GOTOFF(eax,PD_ONEHALF)]	; xmm2=[PD_ONEHALF]
-
-	paddd     xmm6, XMMWORD [wk(6)]
-	paddd     xmm4, XMMWORD [wk(7)]
-	paddd     xmm6,xmm2
-	paddd     xmm4,xmm2
-	psrld     xmm6,SCALEBITS	; xmm6=YEL
-	psrld     xmm4,SCALEBITS	; xmm4=YEH
-	packssdw  xmm6,xmm4		; xmm6=YE
-
-	psllw     xmm0,BYTE_BIT
-	por       xmm6,xmm0		; xmm6=Y
-	movdqa    XMMWORD [edi], xmm6	; Save Y
-
-	pxor      xmm2,xmm2
-	pxor      xmm4,xmm4
-	punpcklwd xmm2,xmm3		; xmm2=REL
-	punpckhwd xmm4,xmm3		; xmm4=REH
-	psrld     xmm2,1		; xmm2=REL*FIX(0.500)
-	psrld     xmm4,1		; xmm4=REH*FIX(0.500)
-
-	movdqa    xmm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ]
-
-	paddd     xmm1,xmm2
-	paddd     xmm5,xmm4
-	paddd     xmm1,xmm0
-	paddd     xmm5,xmm0
-	psrld     xmm1,SCALEBITS	; xmm1=CrEL
-	psrld     xmm5,SCALEBITS	; xmm5=CrEH
-	packssdw  xmm1,xmm5		; xmm1=CrE
-
-	psllw     xmm7,BYTE_BIT
-	por       xmm1,xmm7		; xmm1=Cr
-	movdqa    XMMWORD [edx], xmm1	; Save Cr
-
-	sub	ecx, byte SIZEOF_XMMWORD
-	add	esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; inptr
-	add	edi, byte SIZEOF_XMMWORD		; outptr0
-	add	ebx, byte SIZEOF_XMMWORD		; outptr1
-	add	edx, byte SIZEOF_XMMWORD		; outptr2
-	cmp	ecx, byte SIZEOF_XMMWORD
-	jae	near .columnloop
-	test	ecx,ecx
-	jnz	near .column_ld1
-
-	pop	ecx			; col
-	pop	esi
-	pop	edi
-	pop	ebx
-	pop	edx
-	poppic	eax
-
-	add	esi, byte SIZEOF_JSAMPROW	; input_buf
-	add	edi, byte SIZEOF_JSAMPROW
-	add	ebx, byte SIZEOF_JSAMPROW
-	add	edx, byte SIZEOF_JSAMPROW
-	dec	eax				; num_rows
-	jg	near .rowloop
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_rgb_ycc_convert_sse2 jsimd_extxrgb_ycc_convert_sse2
+%include "jcclrss2.asm"
diff --git a/simd/jdclrmmx.asm b/simd/jdclrmmx.asm
new file mode 100644
index 0000000..506500e
--- /dev/null
+++ b/simd/jdclrmmx.asm
@@ -0,0 +1,402 @@
+;
+; jdclrmmx.asm - colorspace conversion (MMX)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_mmx (JDIMENSION out_width,
+;                            JSAMPIMAGE input_buf, JDIMENSION input_row,
+;                            JSAMPARRAY output_buf, int num_rows)
+;
+
+%define out_width(b)	(b)+8			; JDIMENSION out_width
+%define input_buf(b)	(b)+12		; JSAMPIMAGE input_buf
+%define input_row(b)	(b)+16		; JDIMENSION input_row
+%define output_buf(b)	(b)+20		; JSAMPARRAY output_buf
+%define num_rows(b)	(b)+24		; int num_rows
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
+%define WK_NUM		2
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+	global	EXTN(jsimd_ycc_rgb_convert_mmx)
+
+EXTN(jsimd_ycc_rgb_convert_mmx):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [out_width(eax)]	; num_cols
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	edi, JSAMPIMAGE [input_buf(eax)]
+	mov	ecx, JDIMENSION [input_row(eax)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
+	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+	pop	ecx
+
+	mov	edi, JSAMPARRAY [output_buf(eax)]
+	mov	eax, INT [num_rows(eax)]
+	test	eax,eax
+	jle	near .return
+	alignx	16,7
+.rowloop:
+	push	eax
+	push	edi
+	push	edx
+	push	ebx
+	push	esi
+	push	ecx			; col
+
+	mov	esi, JSAMPROW [esi]	; inptr0
+	mov	ebx, JSAMPROW [ebx]	; inptr1
+	mov	edx, JSAMPROW [edx]	; inptr2
+	mov	edi, JSAMPROW [edi]	; outptr
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+	alignx	16,7
+.columnloop:
+
+	movq	mm5, MMWORD [ebx]	; mm5=Cb(01234567)
+	movq	mm1, MMWORD [edx]	; mm1=Cr(01234567)
+
+	pcmpeqw	mm4,mm4
+	pcmpeqw	mm7,mm7
+	psrlw	mm4,BYTE_BIT
+	psllw	mm7,7			; mm7={0xFF80 0xFF80 0xFF80 0xFF80}
+	movq	mm0,mm4			; mm0=mm4={0xFF 0x00 0xFF 0x00 ..}
+
+	pand	mm4,mm5			; mm4=Cb(0246)=CbE
+	psrlw	mm5,BYTE_BIT		; mm5=Cb(1357)=CbO
+	pand	mm0,mm1			; mm0=Cr(0246)=CrE
+	psrlw	mm1,BYTE_BIT		; mm1=Cr(1357)=CrO
+
+	paddw	mm4,mm7
+	paddw	mm5,mm7
+	paddw	mm0,mm7
+	paddw	mm1,mm7
+
+	; (Original)
+	; R = Y                + 1.40200 * Cr
+	; G = Y - 0.34414 * Cb - 0.71414 * Cr
+	; B = Y + 1.77200 * Cb
+	;
+	; (This implementation)
+	; R = Y                + 0.40200 * Cr + Cr
+	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+	; B = Y - 0.22800 * Cb + Cb + Cb
+
+	movq	mm2,mm4			; mm2=CbE
+	movq	mm3,mm5			; mm3=CbO
+	paddw	mm4,mm4			; mm4=2*CbE
+	paddw	mm5,mm5			; mm5=2*CbO
+	movq	mm6,mm0			; mm6=CrE
+	movq	mm7,mm1			; mm7=CrO
+	paddw	mm0,mm0			; mm0=2*CrE
+	paddw	mm1,mm1			; mm1=2*CrO
+
+	pmulhw	mm4,[GOTOFF(eax,PW_MF0228)]	; mm4=(2*CbE * -FIX(0.22800))
+	pmulhw	mm5,[GOTOFF(eax,PW_MF0228)]	; mm5=(2*CbO * -FIX(0.22800))
+	pmulhw	mm0,[GOTOFF(eax,PW_F0402)]	; mm0=(2*CrE * FIX(0.40200))
+	pmulhw	mm1,[GOTOFF(eax,PW_F0402)]	; mm1=(2*CrO * FIX(0.40200))
+
+	paddw	mm4,[GOTOFF(eax,PW_ONE)]
+	paddw	mm5,[GOTOFF(eax,PW_ONE)]
+	psraw	mm4,1			; mm4=(CbE * -FIX(0.22800))
+	psraw	mm5,1			; mm5=(CbO * -FIX(0.22800))
+	paddw	mm0,[GOTOFF(eax,PW_ONE)]
+	paddw	mm1,[GOTOFF(eax,PW_ONE)]
+	psraw	mm0,1			; mm0=(CrE * FIX(0.40200))
+	psraw	mm1,1			; mm1=(CrO * FIX(0.40200))
+
+	paddw	mm4,mm2
+	paddw	mm5,mm3
+	paddw	mm4,mm2			; mm4=(CbE * FIX(1.77200))=(B-Y)E
+	paddw	mm5,mm3			; mm5=(CbO * FIX(1.77200))=(B-Y)O
+	paddw	mm0,mm6			; mm0=(CrE * FIX(1.40200))=(R-Y)E
+	paddw	mm1,mm7			; mm1=(CrO * FIX(1.40200))=(R-Y)O
+
+	movq	MMWORD [wk(0)], mm4	; wk(0)=(B-Y)E
+	movq	MMWORD [wk(1)], mm5	; wk(1)=(B-Y)O
+
+	movq      mm4,mm2
+	movq      mm5,mm3
+	punpcklwd mm2,mm6
+	punpckhwd mm4,mm6
+	pmaddwd   mm2,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   mm4,[GOTOFF(eax,PW_MF0344_F0285)]
+	punpcklwd mm3,mm7
+	punpckhwd mm5,mm7
+	pmaddwd   mm3,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   mm5,[GOTOFF(eax,PW_MF0344_F0285)]
+
+	paddd     mm2,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     mm4,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     mm2,SCALEBITS
+	psrad     mm4,SCALEBITS
+	paddd     mm3,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     mm5,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     mm3,SCALEBITS
+	psrad     mm5,SCALEBITS
+
+	packssdw  mm2,mm4	; mm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+	packssdw  mm3,mm5	; mm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+	psubw     mm2,mm6	; mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+	psubw     mm3,mm7	; mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+	movq      mm5, MMWORD [esi]	; mm5=Y(01234567)
+
+	pcmpeqw   mm4,mm4
+	psrlw     mm4,BYTE_BIT		; mm4={0xFF 0x00 0xFF 0x00 ..}
+	pand      mm4,mm5		; mm4=Y(0246)=YE
+	psrlw     mm5,BYTE_BIT		; mm5=Y(1357)=YO
+
+	paddw     mm0,mm4		; mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6)
+	paddw     mm1,mm5		; mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7)
+	packuswb  mm0,mm0		; mm0=(R0 R2 R4 R6 ** ** ** **)
+	packuswb  mm1,mm1		; mm1=(R1 R3 R5 R7 ** ** ** **)
+
+	paddw     mm2,mm4		; mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6)
+	paddw     mm3,mm5		; mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7)
+	packuswb  mm2,mm2		; mm2=(G0 G2 G4 G6 ** ** ** **)
+	packuswb  mm3,mm3		; mm3=(G1 G3 G5 G7 ** ** ** **)
+
+	paddw     mm4, MMWORD [wk(0)]	; mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6)
+	paddw     mm5, MMWORD [wk(1)]	; mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7)
+	packuswb  mm4,mm4		; mm4=(B0 B2 B4 B6 ** ** ** **)
+	packuswb  mm5,mm5		; mm5=(B1 B3 B5 B7 ** ** ** **)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
+	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
+	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
+	; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **)
+
+	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
+	punpcklbw mmE,mmB		; mmE=(20 01 22 03 24 05 26 07)
+	punpcklbw mmD,mmF		; mmD=(11 21 13 23 15 25 17 27)
+
+	movq      mmG,mmA
+	movq      mmH,mmA
+	punpcklwd mmA,mmE		; mmA=(00 10 20 01 02 12 22 03)
+	punpckhwd mmG,mmE		; mmG=(04 14 24 05 06 16 26 07)
+
+	psrlq     mmH,2*BYTE_BIT	; mmH=(02 12 04 14 06 16 -- --)
+	psrlq     mmE,2*BYTE_BIT	; mmE=(22 03 24 05 26 07 -- --)
+
+	movq      mmC,mmD
+	movq      mmB,mmD
+	punpcklwd mmD,mmH		; mmD=(11 21 02 12 13 23 04 14)
+	punpckhwd mmC,mmH		; mmC=(15 25 06 16 17 27 -- --)
+
+	psrlq     mmB,2*BYTE_BIT	; mmB=(13 23 15 25 17 27 -- --)
+
+	movq      mmF,mmE
+	punpcklwd mmE,mmB		; mmE=(22 03 13 23 24 05 15 25)
+	punpckhwd mmF,mmB		; mmF=(26 07 17 27 -- -- -- --)
+
+	punpckldq mmA,mmD		; mmA=(00 10 20 01 11 21 02 12)
+	punpckldq mmE,mmG		; mmE=(22 03 13 23 04 14 24 05)
+	punpckldq mmC,mmF		; mmC=(15 25 06 16 26 07 17 27)
+
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st16
+
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
+	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
+
+	sub	ecx, byte SIZEOF_MMWORD
+	jz	short .nextrow
+
+	add	esi, byte SIZEOF_MMWORD			; inptr0
+	add	ebx, byte SIZEOF_MMWORD			; inptr1
+	add	edx, byte SIZEOF_MMWORD			; inptr2
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st16:
+	lea	ecx, [ecx+ecx*2]	; imul ecx, RGB_PIXELSIZE
+	cmp	ecx, byte 2*SIZEOF_MMWORD
+	jb	short .column_st8
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
+	movq	mmA,mmC
+	sub	ecx, byte 2*SIZEOF_MMWORD
+	add	edi, byte 2*SIZEOF_MMWORD
+	jmp	short .column_st4
+.column_st8:
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st4
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	mmA,mmE
+	sub	ecx, byte SIZEOF_MMWORD
+	add	edi, byte SIZEOF_MMWORD
+.column_st4:
+	movd	eax,mmA
+	cmp	ecx, byte SIZEOF_DWORD
+	jb	short .column_st2
+	mov	DWORD [edi+0*SIZEOF_DWORD], eax
+	psrlq	mmA,DWORD_BIT
+	movd	eax,mmA
+	sub	ecx, byte SIZEOF_DWORD
+	add	edi, byte SIZEOF_DWORD
+.column_st2:
+	cmp	ecx, byte SIZEOF_WORD
+	jb	short .column_st1
+	mov	WORD [edi+0*SIZEOF_WORD], ax
+	shr	eax,WORD_BIT
+	sub	ecx, byte SIZEOF_WORD
+	add	edi, byte SIZEOF_WORD
+.column_st1:
+	cmp	ecx, byte SIZEOF_BYTE
+	jb	short .nextrow
+	mov	BYTE [edi+0*SIZEOF_BYTE], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+	pcmpeqb   mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
+	pcmpeqb   mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
+%else
+	pxor      mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
+	pxor      mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
+%endif
+	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
+	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
+	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
+	; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **)
+
+	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
+	punpcklbw mmE,mmG		; mmE=(20 30 22 32 24 34 26 36)
+	punpcklbw mmB,mmD		; mmB=(01 11 03 13 05 15 07 17)
+	punpcklbw mmF,mmH		; mmF=(21 31 23 33 25 35 27 37)
+
+	movq      mmC,mmA
+	punpcklwd mmA,mmE		; mmA=(00 10 20 30 02 12 22 32)
+	punpckhwd mmC,mmE		; mmC=(04 14 24 34 06 16 26 36)
+	movq      mmG,mmB
+	punpcklwd mmB,mmF		; mmB=(01 11 21 31 03 13 23 33)
+	punpckhwd mmG,mmF		; mmG=(05 15 25 35 07 17 27 37)
+
+	movq      mmD,mmA
+	punpckldq mmA,mmB		; mmA=(00 10 20 30 01 11 21 31)
+	punpckhdq mmD,mmB		; mmD=(02 12 22 32 03 13 23 33)
+	movq      mmH,mmC
+	punpckldq mmC,mmG		; mmC=(04 14 24 34 05 15 25 35)
+	punpckhdq mmH,mmG		; mmH=(06 16 26 36 07 17 27 37)
+
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st16
+
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
+	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
+	movq	MMWORD [edi+3*SIZEOF_MMWORD], mmH
+
+	sub	ecx, byte SIZEOF_MMWORD
+	jz	short .nextrow
+
+	add	esi, byte SIZEOF_MMWORD			; inptr0
+	add	ebx, byte SIZEOF_MMWORD			; inptr1
+	add	edx, byte SIZEOF_MMWORD			; inptr2
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st16:
+	cmp	ecx, byte SIZEOF_MMWORD/2
+	jb	short .column_st8
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
+	movq	mmA,mmC
+	movq	mmD,mmH
+	sub	ecx, byte SIZEOF_MMWORD/2
+	add	edi, byte 2*SIZEOF_MMWORD
+.column_st8:
+	cmp	ecx, byte SIZEOF_MMWORD/4
+	jb	short .column_st4
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	mmA,mmD
+	sub	ecx, byte SIZEOF_MMWORD/4
+	add	edi, byte 1*SIZEOF_MMWORD
+.column_st4:
+	cmp	ecx, byte SIZEOF_MMWORD/8
+	jb	short .nextrow
+	movd	DWORD [edi+0*SIZEOF_DWORD], mmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+	alignx	16,7
+
+.nextrow:
+	pop	ecx
+	pop	esi
+	pop	ebx
+	pop	edx
+	pop	edi
+	pop	eax
+
+	add	esi, byte SIZEOF_JSAMPROW
+	add	ebx, byte SIZEOF_JSAMPROW
+	add	edx, byte SIZEOF_JSAMPROW
+	add	edi, byte SIZEOF_JSAMPROW	; output_buf
+	dec	eax				; num_rows
+	jg	near .rowloop
+
+	emms		; empty MMX state
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
diff --git a/simd/jdclrss2.asm b/simd/jdclrss2.asm
new file mode 100644
index 0000000..b3ef195
--- /dev/null
+++ b/simd/jdclrss2.asm
@@ -0,0 +1,500 @@
+;
+; jdclrss2.asm - colorspace conversion (SSE2)
+;
+; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+;
+; Based on
+; x86 SIMD extension for IJG JPEG library
+; Copyright (C) 1999-2006, MIYASAKA Masaru.
+; For conditions of distribution and use, see copyright notice in jsimdext.inc
+;
+; This file should be assembled with NASM (Netwide Assembler),
+; can *not* be assembled with Microsoft's MASM or any compatible
+; assembler (including Borland's Turbo Assembler).
+; NASM is available from http://nasm.sourceforge.net/ or
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
+;                             JSAMPIMAGE input_buf, JDIMENSION input_row,
+;                             JSAMPARRAY output_buf, int num_rows)
+;
+
+%define out_width(b)	(b)+8			; JDIMENSION out_width
+%define input_buf(b)	(b)+12		; JSAMPIMAGE input_buf
+%define input_row(b)	(b)+16		; JDIMENSION input_row
+%define output_buf(b)	(b)+20		; JSAMPARRAY output_buf
+%define num_rows(b)	(b)+24		; int num_rows
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
+%define WK_NUM		2
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+	global	EXTN(jsimd_ycc_rgb_convert_sse2)
+
+EXTN(jsimd_ycc_rgb_convert_sse2):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [out_width(eax)]	; num_cols
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	edi, JSAMPIMAGE [input_buf(eax)]
+	mov	ecx, JDIMENSION [input_row(eax)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
+	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
+	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
+
+	pop	ecx
+
+	mov	edi, JSAMPARRAY [output_buf(eax)]
+	mov	eax, INT [num_rows(eax)]
+	test	eax,eax
+	jle	near .return
+	alignx	16,7
+.rowloop:
+	push	eax
+	push	edi
+	push	edx
+	push	ebx
+	push	esi
+	push	ecx			; col
+
+	mov	esi, JSAMPROW [esi]	; inptr0
+	mov	ebx, JSAMPROW [ebx]	; inptr1
+	mov	edx, JSAMPROW [edx]	; inptr2
+	mov	edi, JSAMPROW [edi]	; outptr
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+	alignx	16,7
+.columnloop:
+
+	movdqa	xmm5, XMMWORD [ebx]	; xmm5=Cb(0123456789ABCDEF)
+	movdqa	xmm1, XMMWORD [edx]	; xmm1=Cr(0123456789ABCDEF)
+
+	pcmpeqw	xmm4,xmm4
+	pcmpeqw	xmm7,xmm7
+	psrlw	xmm4,BYTE_BIT
+	psllw	xmm7,7			; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+	movdqa	xmm0,xmm4		; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
+
+	pand	xmm4,xmm5		; xmm4=Cb(02468ACE)=CbE
+	psrlw	xmm5,BYTE_BIT		; xmm5=Cb(13579BDF)=CbO
+	pand	xmm0,xmm1		; xmm0=Cr(02468ACE)=CrE
+	psrlw	xmm1,BYTE_BIT		; xmm1=Cr(13579BDF)=CrO
+
+	paddw	xmm4,xmm7
+	paddw	xmm5,xmm7
+	paddw	xmm0,xmm7
+	paddw	xmm1,xmm7
+
+	; (Original)
+	; R = Y                + 1.40200 * Cr
+	; G = Y - 0.34414 * Cb - 0.71414 * Cr
+	; B = Y + 1.77200 * Cb
+	;
+	; (This implementation)
+	; R = Y                + 0.40200 * Cr + Cr
+	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+	; B = Y - 0.22800 * Cb + Cb + Cb
+
+	movdqa	xmm2,xmm4		; xmm2=CbE
+	movdqa	xmm3,xmm5		; xmm3=CbO
+	paddw	xmm4,xmm4		; xmm4=2*CbE
+	paddw	xmm5,xmm5		; xmm5=2*CbO
+	movdqa	xmm6,xmm0		; xmm6=CrE
+	movdqa	xmm7,xmm1		; xmm7=CrO
+	paddw	xmm0,xmm0		; xmm0=2*CrE
+	paddw	xmm1,xmm1		; xmm1=2*CrO
+
+	pmulhw	xmm4,[GOTOFF(eax,PW_MF0228)]	; xmm4=(2*CbE * -FIX(0.22800))
+	pmulhw	xmm5,[GOTOFF(eax,PW_MF0228)]	; xmm5=(2*CbO * -FIX(0.22800))
+	pmulhw	xmm0,[GOTOFF(eax,PW_F0402)]	; xmm0=(2*CrE * FIX(0.40200))
+	pmulhw	xmm1,[GOTOFF(eax,PW_F0402)]	; xmm1=(2*CrO * FIX(0.40200))
+
+	paddw	xmm4,[GOTOFF(eax,PW_ONE)]
+	paddw	xmm5,[GOTOFF(eax,PW_ONE)]
+	psraw	xmm4,1			; xmm4=(CbE * -FIX(0.22800))
+	psraw	xmm5,1			; xmm5=(CbO * -FIX(0.22800))
+	paddw	xmm0,[GOTOFF(eax,PW_ONE)]
+	paddw	xmm1,[GOTOFF(eax,PW_ONE)]
+	psraw	xmm0,1			; xmm0=(CrE * FIX(0.40200))
+	psraw	xmm1,1			; xmm1=(CrO * FIX(0.40200))
+
+	paddw	xmm4,xmm2
+	paddw	xmm5,xmm3
+	paddw	xmm4,xmm2		; xmm4=(CbE * FIX(1.77200))=(B-Y)E
+	paddw	xmm5,xmm3		; xmm5=(CbO * FIX(1.77200))=(B-Y)O
+	paddw	xmm0,xmm6		; xmm0=(CrE * FIX(1.40200))=(R-Y)E
+	paddw	xmm1,xmm7		; xmm1=(CrO * FIX(1.40200))=(R-Y)O
+
+	movdqa	XMMWORD [wk(0)], xmm4	; wk(0)=(B-Y)E
+	movdqa	XMMWORD [wk(1)], xmm5	; wk(1)=(B-Y)O
+
+	movdqa    xmm4,xmm2
+	movdqa    xmm5,xmm3
+	punpcklwd xmm2,xmm6
+	punpckhwd xmm4,xmm6
+	pmaddwd   xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   xmm4,[GOTOFF(eax,PW_MF0344_F0285)]
+	punpcklwd xmm3,xmm7
+	punpckhwd xmm5,xmm7
+	pmaddwd   xmm3,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
+
+	paddd     xmm2,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     xmm4,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     xmm2,SCALEBITS
+	psrad     xmm4,SCALEBITS
+	paddd     xmm3,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     xmm5,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     xmm3,SCALEBITS
+	psrad     xmm5,SCALEBITS
+
+	packssdw  xmm2,xmm4	; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+	packssdw  xmm3,xmm5	; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+	psubw     xmm2,xmm6	; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+	psubw     xmm3,xmm7	; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+	movdqa    xmm5, XMMWORD [esi]	; xmm5=Y(0123456789ABCDEF)
+
+	pcmpeqw   xmm4,xmm4
+	psrlw     xmm4,BYTE_BIT		; xmm4={0xFF 0x00 0xFF 0x00 ..}
+	pand      xmm4,xmm5		; xmm4=Y(02468ACE)=YE
+	psrlw     xmm5,BYTE_BIT		; xmm5=Y(13579BDF)=YO
+
+	paddw     xmm0,xmm4		; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
+	paddw     xmm1,xmm5		; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
+	packuswb  xmm0,xmm0		; xmm0=R(02468ACE********)
+	packuswb  xmm1,xmm1		; xmm1=R(13579BDF********)
+
+	paddw     xmm2,xmm4		; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
+	paddw     xmm3,xmm5		; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
+	packuswb  xmm2,xmm2		; xmm2=G(02468ACE********)
+	packuswb  xmm3,xmm3		; xmm3=G(13579BDF********)
+
+	paddw     xmm4, XMMWORD [wk(0)]	; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
+	paddw     xmm5, XMMWORD [wk(1)]	; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
+	packuswb  xmm4,xmm4		; xmm4=B(02468ACE********)
+	packuswb  xmm5,xmm5		; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+	; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+	punpcklbw xmmE,xmmB	; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+	punpcklbw xmmD,xmmF	; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+	movdqa    xmmG,xmmA
+	movdqa    xmmH,xmmA
+	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+	punpckhwd xmmG,xmmE	; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+	psrldq    xmmH,2	; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+	psrldq    xmmE,2	; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+	movdqa    xmmC,xmmD
+	movdqa    xmmB,xmmD
+	punpcklwd xmmD,xmmH	; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+	punpckhwd xmmC,xmmH	; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+	psrldq    xmmB,2	; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+	movdqa    xmmF,xmmE
+	punpcklwd xmmE,xmmB	; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+	punpckhwd xmmF,xmmB	; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+	pshufd    xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+	movdqa    xmmB,xmmE
+	punpckldq xmmA,xmmD	; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+	punpckldq xmmE,xmmH	; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+	punpckhdq xmmD,xmmB	; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+	pshufd    xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+	movdqa    xmmB,xmmF
+	punpckldq xmmG,xmmC	; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+	punpckldq xmmF,xmmH	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+	punpckhdq xmmC,xmmB	; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+	punpcklqdq xmmA,xmmE	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+	punpcklqdq xmmD,xmmG	; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+	punpcklqdq xmmF,xmmC	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+	cmp	ecx, byte SIZEOF_XMMWORD
+	jb	short .column_st32
+
+	test	edi, SIZEOF_XMMWORD-1
+	jnz	short .out1
+	; --(aligned)-------------------
+	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
+	jmp	short .out0
+.out1:	; --(unaligned)-----------------
+	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
+	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [edi], xmmF
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+.out0:
+	sub	ecx, byte SIZEOF_XMMWORD
+	jz	near .nextrow
+
+	add	esi, byte SIZEOF_XMMWORD	; inptr0
+	add	ebx, byte SIZEOF_XMMWORD	; inptr1
+	add	edx, byte SIZEOF_XMMWORD	; inptr2
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st32:
+	pcmpeqb	xmmH,xmmH			; xmmH=(all 1's)
+	lea	ecx, [ecx+ecx*2]		; imul ecx, RGB_PIXELSIZE
+	cmp	ecx, byte 2*SIZEOF_XMMWORD
+	jb	short .column_st16
+	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmF
+	sub	ecx, byte 2*SIZEOF_XMMWORD
+	jmp	short .column_st15
+.column_st16:
+	cmp	ecx, byte SIZEOF_XMMWORD
+	jb	short .column_st15
+	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmD
+	sub	ecx, byte SIZEOF_XMMWORD
+.column_st15:
+	mov	eax,ecx
+	xor	ecx, byte 0x0F
+	shl	ecx, 2
+	movd	xmmB,ecx
+	psrlq	xmmH,4
+	pcmpeqb	xmmE,xmmE
+	psrlq	xmmH,xmmB
+	psrlq	xmmE,xmmB
+	punpcklbw xmmE,xmmH
+	; ----------------
+	mov	ecx,edi
+	and	ecx, byte SIZEOF_XMMWORD-1
+	jz	short .adj0
+	add	eax,ecx
+	cmp	eax, byte SIZEOF_XMMWORD
+	ja	short .adj0
+	and	edi, byte (-SIZEOF_XMMWORD)	; align to 16-byte boundary
+	shl	ecx, 3			; pslldq xmmA,ecx & pslldq xmmE,ecx
+	movdqa	xmmG,xmmA
+	movdqa	xmmC,xmmE
+	pslldq	xmmA, SIZEOF_XMMWORD/2
+	pslldq	xmmE, SIZEOF_XMMWORD/2
+	movd	xmmD,ecx
+	sub	ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
+	jb	short .adj1
+	movd	xmmF,ecx
+	psllq	xmmA,xmmF
+	psllq	xmmE,xmmF
+	jmp	short .adj0
+.adj1:	neg	ecx
+	movd	xmmF,ecx
+	psrlq	xmmA,xmmF
+	psrlq	xmmE,xmmF
+	psllq	xmmG,xmmD
+	psllq	xmmC,xmmD
+	por	xmmA,xmmG
+	por	xmmE,xmmC
+.adj0:	; ----------------
+	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+	pcmpeqb   xmm6,xmm6		; xmm6=XE=X(02468ACE********)
+	pcmpeqb   xmm7,xmm7		; xmm7=XO=X(13579BDF********)
+%else
+	pxor      xmm6,xmm6		; xmm6=XE=X(02468ACE********)
+	pxor      xmm7,xmm7		; xmm7=XO=X(13579BDF********)
+%endif
+	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+	; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+	punpcklbw xmmE,xmmG	; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+	punpcklbw xmmB,xmmD	; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+	punpcklbw xmmF,xmmH	; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+	movdqa    xmmC,xmmA
+	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+	punpckhwd xmmC,xmmE	; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+	movdqa    xmmG,xmmB
+	punpcklwd xmmB,xmmF	; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+	punpckhwd xmmG,xmmF	; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+	movdqa    xmmD,xmmA
+	punpckldq xmmA,xmmB	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+	punpckhdq xmmD,xmmB	; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+	movdqa    xmmH,xmmC
+	punpckldq xmmC,xmmG	; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+	punpckhdq xmmH,xmmG	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+	cmp	ecx, byte SIZEOF_XMMWORD
+	jb	short .column_st32
+
+	test	edi, SIZEOF_XMMWORD-1
+	jnz	short .out1
+	; --(aligned)-------------------
+	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+	movntdq	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
+	jmp	short .out0
+.out1:	; --(unaligned)-----------------
+	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
+	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [edi], xmmC
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [edi], xmmH
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+.out0:
+	sub	ecx, byte SIZEOF_XMMWORD
+	jz	near .nextrow
+
+	add	esi, byte SIZEOF_XMMWORD	; inptr0
+	add	ebx, byte SIZEOF_XMMWORD	; inptr1
+	add	edx, byte SIZEOF_XMMWORD	; inptr2
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st32:
+	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
+	cmp	ecx, byte SIZEOF_XMMWORD/2
+	jb	short .column_st16
+	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmC
+	movdqa	xmmD,xmmH
+	sub	ecx, byte SIZEOF_XMMWORD/2
+.column_st16:
+	cmp	ecx, byte SIZEOF_XMMWORD/4
+	jb	short .column_st15
+	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmD
+	sub	ecx, byte SIZEOF_XMMWORD/4
+.column_st15:
+	cmp	ecx, byte SIZEOF_XMMWORD/16
+	jb	short .nextrow
+	mov	eax,ecx
+	xor	ecx, byte 0x03
+	inc	ecx
+	shl	ecx, 4
+	movd	xmmF,ecx
+	psrlq	xmmE,xmmF
+	punpcklbw xmmE,xmmE
+	; ----------------
+	mov	ecx,edi
+	and	ecx, byte SIZEOF_XMMWORD-1
+	jz	short .adj0
+	lea	eax, [ecx+eax*4]	; RGB_PIXELSIZE
+	cmp	eax, byte SIZEOF_XMMWORD
+	ja	short .adj0
+	and	edi, byte (-SIZEOF_XMMWORD)	; align to 16-byte boundary
+	shl	ecx, 3			; pslldq xmmA,ecx & pslldq xmmE,ecx
+	movdqa	xmmB,xmmA
+	movdqa	xmmG,xmmE
+	pslldq	xmmA, SIZEOF_XMMWORD/2
+	pslldq	xmmE, SIZEOF_XMMWORD/2
+	movd	xmmC,ecx
+	sub	ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
+	jb	short .adj1
+	movd	xmmH,ecx
+	psllq	xmmA,xmmH
+	psllq	xmmE,xmmH
+	jmp	short .adj0
+.adj1:	neg	ecx
+	movd	xmmH,ecx
+	psrlq	xmmA,xmmH
+	psrlq	xmmE,xmmH
+	psllq	xmmB,xmmC
+	psllq	xmmG,xmmC
+	por	xmmA,xmmB
+	por	xmmE,xmmG
+.adj0:	; ----------------
+	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+	alignx	16,7
+
+.nextrow:
+	pop	ecx
+	pop	esi
+	pop	ebx
+	pop	edx
+	pop	edi
+	pop	eax
+
+	add	esi, byte SIZEOF_JSAMPROW
+	add	ebx, byte SIZEOF_JSAMPROW
+	add	edx, byte SIZEOF_JSAMPROW
+	add	edi, byte SIZEOF_JSAMPROW	; output_buf
+	dec	eax				; num_rows
+	jg	near .rowloop
+
+	sfence		; flush the write buffer
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
diff --git a/simd/jdcolmmx.asm b/simd/jdcolmmx.asm
index 742c9ce..f34387b 100644
--- a/simd/jdcolmmx.asm
+++ b/simd/jdcolmmx.asm
@@ -2,6 +2,7 @@
 ; jdcolmmx.asm - colorspace conversion (MMX)
 ;
 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009 D. R. Commander
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
@@ -48,386 +49,70 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_ycc_rgb_convert_mmx (JDIMENSION out_width,
-;                            JSAMPIMAGE input_buf, JDIMENSION input_row,
-;                            JSAMPARRAY output_buf, int num_rows)
-;
+%include "jdclrmmx.asm"
 
-%define out_width(b)	(b)+8			; JDIMENSION out_width
-%define input_buf(b)	(b)+12		; JSAMPIMAGE input_buf
-%define input_row(b)	(b)+16		; JDIMENSION input_row
-%define output_buf(b)	(b)+20		; JSAMPARRAY output_buf
-%define num_rows(b)	(b)+24		; int num_rows
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extrgb_convert_mmx
+%include "jdclrmmx.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
-%define WK_NUM		2
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extrgbx_convert_mmx
+%include "jdclrmmx.asm"
 
-	align	16
-	global	EXTN(jsimd_ycc_rgb_convert_mmx)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extbgr_convert_mmx
+%include "jdclrmmx.asm"
 
-EXTN(jsimd_ycc_rgb_convert_mmx):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extbgrx_convert_mmx
+%include "jdclrmmx.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extxbgr_convert_mmx
+%include "jdclrmmx.asm"
 
-	mov	ecx, JDIMENSION [out_width(eax)]	; num_cols
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	edi, JSAMPIMAGE [input_buf(eax)]
-	mov	ecx, JDIMENSION [input_row(eax)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
-	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
-	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
-
-	pop	ecx
-
-	mov	edi, JSAMPARRAY [output_buf(eax)]
-	mov	eax, INT [num_rows(eax)]
-	test	eax,eax
-	jle	near .return
-	alignx	16,7
-.rowloop:
-	push	eax
-	push	edi
-	push	edx
-	push	ebx
-	push	esi
-	push	ecx			; col
-
-	mov	esi, JSAMPROW [esi]	; inptr0
-	mov	ebx, JSAMPROW [ebx]	; inptr1
-	mov	edx, JSAMPROW [edx]	; inptr2
-	mov	edi, JSAMPROW [edi]	; outptr
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-	alignx	16,7
-.columnloop:
-
-	movq	mm5, MMWORD [ebx]	; mm5=Cb(01234567)
-	movq	mm1, MMWORD [edx]	; mm1=Cr(01234567)
-
-	pcmpeqw	mm4,mm4
-	pcmpeqw	mm7,mm7
-	psrlw	mm4,BYTE_BIT
-	psllw	mm7,7			; mm7={0xFF80 0xFF80 0xFF80 0xFF80}
-	movq	mm0,mm4			; mm0=mm4={0xFF 0x00 0xFF 0x00 ..}
-
-	pand	mm4,mm5			; mm4=Cb(0246)=CbE
-	psrlw	mm5,BYTE_BIT		; mm5=Cb(1357)=CbO
-	pand	mm0,mm1			; mm0=Cr(0246)=CrE
-	psrlw	mm1,BYTE_BIT		; mm1=Cr(1357)=CrO
-
-	paddw	mm4,mm7
-	paddw	mm5,mm7
-	paddw	mm0,mm7
-	paddw	mm1,mm7
-
-	; (Original)
-	; R = Y                + 1.40200 * Cr
-	; G = Y - 0.34414 * Cb - 0.71414 * Cr
-	; B = Y + 1.77200 * Cb
-	;
-	; (This implementation)
-	; R = Y                + 0.40200 * Cr + Cr
-	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
-	; B = Y - 0.22800 * Cb + Cb + Cb
-
-	movq	mm2,mm4			; mm2=CbE
-	movq	mm3,mm5			; mm3=CbO
-	paddw	mm4,mm4			; mm4=2*CbE
-	paddw	mm5,mm5			; mm5=2*CbO
-	movq	mm6,mm0			; mm6=CrE
-	movq	mm7,mm1			; mm7=CrO
-	paddw	mm0,mm0			; mm0=2*CrE
-	paddw	mm1,mm1			; mm1=2*CrO
-
-	pmulhw	mm4,[GOTOFF(eax,PW_MF0228)]	; mm4=(2*CbE * -FIX(0.22800))
-	pmulhw	mm5,[GOTOFF(eax,PW_MF0228)]	; mm5=(2*CbO * -FIX(0.22800))
-	pmulhw	mm0,[GOTOFF(eax,PW_F0402)]	; mm0=(2*CrE * FIX(0.40200))
-	pmulhw	mm1,[GOTOFF(eax,PW_F0402)]	; mm1=(2*CrO * FIX(0.40200))
-
-	paddw	mm4,[GOTOFF(eax,PW_ONE)]
-	paddw	mm5,[GOTOFF(eax,PW_ONE)]
-	psraw	mm4,1			; mm4=(CbE * -FIX(0.22800))
-	psraw	mm5,1			; mm5=(CbO * -FIX(0.22800))
-	paddw	mm0,[GOTOFF(eax,PW_ONE)]
-	paddw	mm1,[GOTOFF(eax,PW_ONE)]
-	psraw	mm0,1			; mm0=(CrE * FIX(0.40200))
-	psraw	mm1,1			; mm1=(CrO * FIX(0.40200))
-
-	paddw	mm4,mm2
-	paddw	mm5,mm3
-	paddw	mm4,mm2			; mm4=(CbE * FIX(1.77200))=(B-Y)E
-	paddw	mm5,mm3			; mm5=(CbO * FIX(1.77200))=(B-Y)O
-	paddw	mm0,mm6			; mm0=(CrE * FIX(1.40200))=(R-Y)E
-	paddw	mm1,mm7			; mm1=(CrO * FIX(1.40200))=(R-Y)O
-
-	movq	MMWORD [wk(0)], mm4	; wk(0)=(B-Y)E
-	movq	MMWORD [wk(1)], mm5	; wk(1)=(B-Y)O
-
-	movq      mm4,mm2
-	movq      mm5,mm3
-	punpcklwd mm2,mm6
-	punpckhwd mm4,mm6
-	pmaddwd   mm2,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   mm4,[GOTOFF(eax,PW_MF0344_F0285)]
-	punpcklwd mm3,mm7
-	punpckhwd mm5,mm7
-	pmaddwd   mm3,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   mm5,[GOTOFF(eax,PW_MF0344_F0285)]
-
-	paddd     mm2,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     mm4,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     mm2,SCALEBITS
-	psrad     mm4,SCALEBITS
-	paddd     mm3,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     mm5,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     mm3,SCALEBITS
-	psrad     mm5,SCALEBITS
-
-	packssdw  mm2,mm4	; mm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
-	packssdw  mm3,mm5	; mm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
-	psubw     mm2,mm6	; mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
-	psubw     mm3,mm7	; mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
-
-	movq      mm5, MMWORD [esi]	; mm5=Y(01234567)
-
-	pcmpeqw   mm4,mm4
-	psrlw     mm4,BYTE_BIT		; mm4={0xFF 0x00 0xFF 0x00 ..}
-	pand      mm4,mm5		; mm4=Y(0246)=YE
-	psrlw     mm5,BYTE_BIT		; mm5=Y(1357)=YO
-
-	paddw     mm0,mm4		; mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6)
-	paddw     mm1,mm5		; mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7)
-	packuswb  mm0,mm0		; mm0=(R0 R2 R4 R6 ** ** ** **)
-	packuswb  mm1,mm1		; mm1=(R1 R3 R5 R7 ** ** ** **)
-
-	paddw     mm2,mm4		; mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6)
-	paddw     mm3,mm5		; mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7)
-	packuswb  mm2,mm2		; mm2=(G0 G2 G4 G6 ** ** ** **)
-	packuswb  mm3,mm3		; mm3=(G1 G3 G5 G7 ** ** ** **)
-
-	paddw     mm4, MMWORD [wk(0)]	; mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6)
-	paddw     mm5, MMWORD [wk(1)]	; mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7)
-	packuswb  mm4,mm4		; mm4=(B0 B2 B4 B6 ** ** ** **)
-	packuswb  mm5,mm5		; mm5=(B1 B3 B5 B7 ** ** ** **)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
-	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
-	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
-	; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **)
-
-	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
-	punpcklbw mmE,mmB		; mmE=(20 01 22 03 24 05 26 07)
-	punpcklbw mmD,mmF		; mmD=(11 21 13 23 15 25 17 27)
-
-	movq      mmG,mmA
-	movq      mmH,mmA
-	punpcklwd mmA,mmE		; mmA=(00 10 20 01 02 12 22 03)
-	punpckhwd mmG,mmE		; mmG=(04 14 24 05 06 16 26 07)
-
-	psrlq     mmH,2*BYTE_BIT	; mmH=(02 12 04 14 06 16 -- --)
-	psrlq     mmE,2*BYTE_BIT	; mmE=(22 03 24 05 26 07 -- --)
-
-	movq      mmC,mmD
-	movq      mmB,mmD
-	punpcklwd mmD,mmH		; mmD=(11 21 02 12 13 23 04 14)
-	punpckhwd mmC,mmH		; mmC=(15 25 06 16 17 27 -- --)
-
-	psrlq     mmB,2*BYTE_BIT	; mmB=(13 23 15 25 17 27 -- --)
-
-	movq      mmF,mmE
-	punpcklwd mmE,mmB		; mmE=(22 03 13 23 24 05 15 25)
-	punpckhwd mmF,mmB		; mmF=(26 07 17 27 -- -- -- --)
-
-	punpckldq mmA,mmD		; mmA=(00 10 20 01 11 21 02 12)
-	punpckldq mmE,mmG		; mmE=(22 03 13 23 04 14 24 05)
-	punpckldq mmC,mmF		; mmC=(15 25 06 16 26 07 17 27)
-
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st16
-
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
-	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
-
-	sub	ecx, byte SIZEOF_MMWORD
-	jz	short .nextrow
-
-	add	esi, byte SIZEOF_MMWORD			; inptr0
-	add	ebx, byte SIZEOF_MMWORD			; inptr1
-	add	edx, byte SIZEOF_MMWORD			; inptr2
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st16:
-	lea	ecx, [ecx+ecx*2]	; imul ecx, RGB_PIXELSIZE
-	cmp	ecx, byte 2*SIZEOF_MMWORD
-	jb	short .column_st8
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
-	movq	mmA,mmC
-	sub	ecx, byte 2*SIZEOF_MMWORD
-	add	edi, byte 2*SIZEOF_MMWORD
-	jmp	short .column_st4
-.column_st8:
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st4
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	mmA,mmE
-	sub	ecx, byte SIZEOF_MMWORD
-	add	edi, byte SIZEOF_MMWORD
-.column_st4:
-	movd	eax,mmA
-	cmp	ecx, byte SIZEOF_DWORD
-	jb	short .column_st2
-	mov	DWORD [edi+0*SIZEOF_DWORD], eax
-	psrlq	mmA,DWORD_BIT
-	movd	eax,mmA
-	sub	ecx, byte SIZEOF_DWORD
-	add	edi, byte SIZEOF_DWORD
-.column_st2:
-	cmp	ecx, byte SIZEOF_WORD
-	jb	short .column_st1
-	mov	WORD [edi+0*SIZEOF_WORD], ax
-	shr	eax,WORD_BIT
-	sub	ecx, byte SIZEOF_WORD
-	add	edi, byte SIZEOF_WORD
-.column_st1:
-	cmp	ecx, byte SIZEOF_BYTE
-	jb	short .nextrow
-	mov	BYTE [edi+0*SIZEOF_BYTE], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
-	pcmpeqb   mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
-	pcmpeqb   mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
-%else
-	pxor      mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
-	pxor      mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
-%endif
-	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
-	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
-	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
-	; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **)
-
-	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
-	punpcklbw mmE,mmG		; mmE=(20 30 22 32 24 34 26 36)
-	punpcklbw mmB,mmD		; mmB=(01 11 03 13 05 15 07 17)
-	punpcklbw mmF,mmH		; mmF=(21 31 23 33 25 35 27 37)
-
-	movq      mmC,mmA
-	punpcklwd mmA,mmE		; mmA=(00 10 20 30 02 12 22 32)
-	punpckhwd mmC,mmE		; mmC=(04 14 24 34 06 16 26 36)
-	movq      mmG,mmB
-	punpcklwd mmB,mmF		; mmB=(01 11 21 31 03 13 23 33)
-	punpckhwd mmG,mmF		; mmG=(05 15 25 35 07 17 27 37)
-
-	movq      mmD,mmA
-	punpckldq mmA,mmB		; mmA=(00 10 20 30 01 11 21 31)
-	punpckhdq mmD,mmB		; mmD=(02 12 22 32 03 13 23 33)
-	movq      mmH,mmC
-	punpckldq mmC,mmG		; mmC=(04 14 24 34 05 15 25 35)
-	punpckhdq mmH,mmG		; mmH=(06 16 26 36 07 17 27 37)
-
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st16
-
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
-	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
-	movq	MMWORD [edi+3*SIZEOF_MMWORD], mmH
-
-	sub	ecx, byte SIZEOF_MMWORD
-	jz	short .nextrow
-
-	add	esi, byte SIZEOF_MMWORD			; inptr0
-	add	ebx, byte SIZEOF_MMWORD			; inptr1
-	add	edx, byte SIZEOF_MMWORD			; inptr2
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st16:
-	cmp	ecx, byte SIZEOF_MMWORD/2
-	jb	short .column_st8
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
-	movq	mmA,mmC
-	movq	mmD,mmH
-	sub	ecx, byte SIZEOF_MMWORD/2
-	add	edi, byte 2*SIZEOF_MMWORD
-.column_st8:
-	cmp	ecx, byte SIZEOF_MMWORD/4
-	jb	short .column_st4
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	mmA,mmD
-	sub	ecx, byte SIZEOF_MMWORD/4
-	add	edi, byte 1*SIZEOF_MMWORD
-.column_st4:
-	cmp	ecx, byte SIZEOF_MMWORD/8
-	jb	short .nextrow
-	movd	DWORD [edi+0*SIZEOF_DWORD], mmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-	alignx	16,7
-
-.nextrow:
-	pop	ecx
-	pop	esi
-	pop	ebx
-	pop	edx
-	pop	edi
-	pop	eax
-
-	add	esi, byte SIZEOF_JSAMPROW
-	add	ebx, byte SIZEOF_JSAMPROW
-	add	edx, byte SIZEOF_JSAMPROW
-	add	edi, byte SIZEOF_JSAMPROW	; output_buf
-	dec	eax				; num_rows
-	jg	near .rowloop
-
-	emms		; empty MMX state
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extxrgb_convert_mmx
+%include "jdclrmmx.asm"
diff --git a/simd/jdcolss2.asm b/simd/jdcolss2.asm
index 6f74cc6..b13c944 100644
--- a/simd/jdcolss2.asm
+++ b/simd/jdcolss2.asm
@@ -2,6 +2,7 @@
 ; jdcolss2.asm - colorspace conversion (SSE2)
 ;
 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009 D. R. Commander
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
@@ -48,484 +49,70 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Convert some rows of samples to the output colorspace.
-;
-; GLOBAL(void)
-; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
-;                             JSAMPIMAGE input_buf, JDIMENSION input_row,
-;                             JSAMPARRAY output_buf, int num_rows)
-;
+%include "jdclrss2.asm"
 
-%define out_width(b)	(b)+8			; JDIMENSION out_width
-%define input_buf(b)	(b)+12		; JSAMPIMAGE input_buf
-%define input_row(b)	(b)+16		; JDIMENSION input_row
-%define output_buf(b)	(b)+20		; JSAMPARRAY output_buf
-%define num_rows(b)	(b)+24		; int num_rows
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgb_convert_sse2
+%include "jdclrss2.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
-%define WK_NUM		2
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgbx_convert_sse2
+%include "jdclrss2.asm"
 
-	align	16
-	global	EXTN(jsimd_ycc_rgb_convert_sse2)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgr_convert_sse2
+%include "jdclrss2.asm"
 
-EXTN(jsimd_ycc_rgb_convert_sse2):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgrx_convert_sse2
+%include "jdclrss2.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxbgr_convert_sse2
+%include "jdclrss2.asm"
 
-	mov	ecx, JDIMENSION [out_width(eax)]	; num_cols
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	edi, JSAMPIMAGE [input_buf(eax)]
-	mov	ecx, JDIMENSION [input_row(eax)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
-	lea	ebx, [ebx+ecx*SIZEOF_JSAMPROW]
-	lea	edx, [edx+ecx*SIZEOF_JSAMPROW]
-
-	pop	ecx
-
-	mov	edi, JSAMPARRAY [output_buf(eax)]
-	mov	eax, INT [num_rows(eax)]
-	test	eax,eax
-	jle	near .return
-	alignx	16,7
-.rowloop:
-	push	eax
-	push	edi
-	push	edx
-	push	ebx
-	push	esi
-	push	ecx			; col
-
-	mov	esi, JSAMPROW [esi]	; inptr0
-	mov	ebx, JSAMPROW [ebx]	; inptr1
-	mov	edx, JSAMPROW [edx]	; inptr2
-	mov	edi, JSAMPROW [edi]	; outptr
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-	alignx	16,7
-.columnloop:
-
-	movdqa	xmm5, XMMWORD [ebx]	; xmm5=Cb(0123456789ABCDEF)
-	movdqa	xmm1, XMMWORD [edx]	; xmm1=Cr(0123456789ABCDEF)
-
-	pcmpeqw	xmm4,xmm4
-	pcmpeqw	xmm7,xmm7
-	psrlw	xmm4,BYTE_BIT
-	psllw	xmm7,7			; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-	movdqa	xmm0,xmm4		; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
-
-	pand	xmm4,xmm5		; xmm4=Cb(02468ACE)=CbE
-	psrlw	xmm5,BYTE_BIT		; xmm5=Cb(13579BDF)=CbO
-	pand	xmm0,xmm1		; xmm0=Cr(02468ACE)=CrE
-	psrlw	xmm1,BYTE_BIT		; xmm1=Cr(13579BDF)=CrO
-
-	paddw	xmm4,xmm7
-	paddw	xmm5,xmm7
-	paddw	xmm0,xmm7
-	paddw	xmm1,xmm7
-
-	; (Original)
-	; R = Y                + 1.40200 * Cr
-	; G = Y - 0.34414 * Cb - 0.71414 * Cr
-	; B = Y + 1.77200 * Cb
-	;
-	; (This implementation)
-	; R = Y                + 0.40200 * Cr + Cr
-	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
-	; B = Y - 0.22800 * Cb + Cb + Cb
-
-	movdqa	xmm2,xmm4		; xmm2=CbE
-	movdqa	xmm3,xmm5		; xmm3=CbO
-	paddw	xmm4,xmm4		; xmm4=2*CbE
-	paddw	xmm5,xmm5		; xmm5=2*CbO
-	movdqa	xmm6,xmm0		; xmm6=CrE
-	movdqa	xmm7,xmm1		; xmm7=CrO
-	paddw	xmm0,xmm0		; xmm0=2*CrE
-	paddw	xmm1,xmm1		; xmm1=2*CrO
-
-	pmulhw	xmm4,[GOTOFF(eax,PW_MF0228)]	; xmm4=(2*CbE * -FIX(0.22800))
-	pmulhw	xmm5,[GOTOFF(eax,PW_MF0228)]	; xmm5=(2*CbO * -FIX(0.22800))
-	pmulhw	xmm0,[GOTOFF(eax,PW_F0402)]	; xmm0=(2*CrE * FIX(0.40200))
-	pmulhw	xmm1,[GOTOFF(eax,PW_F0402)]	; xmm1=(2*CrO * FIX(0.40200))
-
-	paddw	xmm4,[GOTOFF(eax,PW_ONE)]
-	paddw	xmm5,[GOTOFF(eax,PW_ONE)]
-	psraw	xmm4,1			; xmm4=(CbE * -FIX(0.22800))
-	psraw	xmm5,1			; xmm5=(CbO * -FIX(0.22800))
-	paddw	xmm0,[GOTOFF(eax,PW_ONE)]
-	paddw	xmm1,[GOTOFF(eax,PW_ONE)]
-	psraw	xmm0,1			; xmm0=(CrE * FIX(0.40200))
-	psraw	xmm1,1			; xmm1=(CrO * FIX(0.40200))
-
-	paddw	xmm4,xmm2
-	paddw	xmm5,xmm3
-	paddw	xmm4,xmm2		; xmm4=(CbE * FIX(1.77200))=(B-Y)E
-	paddw	xmm5,xmm3		; xmm5=(CbO * FIX(1.77200))=(B-Y)O
-	paddw	xmm0,xmm6		; xmm0=(CrE * FIX(1.40200))=(R-Y)E
-	paddw	xmm1,xmm7		; xmm1=(CrO * FIX(1.40200))=(R-Y)O
-
-	movdqa	XMMWORD [wk(0)], xmm4	; wk(0)=(B-Y)E
-	movdqa	XMMWORD [wk(1)], xmm5	; wk(1)=(B-Y)O
-
-	movdqa    xmm4,xmm2
-	movdqa    xmm5,xmm3
-	punpcklwd xmm2,xmm6
-	punpckhwd xmm4,xmm6
-	pmaddwd   xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   xmm4,[GOTOFF(eax,PW_MF0344_F0285)]
-	punpcklwd xmm3,xmm7
-	punpckhwd xmm5,xmm7
-	pmaddwd   xmm3,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
-
-	paddd     xmm2,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     xmm4,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     xmm2,SCALEBITS
-	psrad     xmm4,SCALEBITS
-	paddd     xmm3,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     xmm5,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     xmm3,SCALEBITS
-	psrad     xmm5,SCALEBITS
-
-	packssdw  xmm2,xmm4	; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
-	packssdw  xmm3,xmm5	; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
-	psubw     xmm2,xmm6	; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
-	psubw     xmm3,xmm7	; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
-
-	movdqa    xmm5, XMMWORD [esi]	; xmm5=Y(0123456789ABCDEF)
-
-	pcmpeqw   xmm4,xmm4
-	psrlw     xmm4,BYTE_BIT		; xmm4={0xFF 0x00 0xFF 0x00 ..}
-	pand      xmm4,xmm5		; xmm4=Y(02468ACE)=YE
-	psrlw     xmm5,BYTE_BIT		; xmm5=Y(13579BDF)=YO
-
-	paddw     xmm0,xmm4		; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
-	paddw     xmm1,xmm5		; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
-	packuswb  xmm0,xmm0		; xmm0=R(02468ACE********)
-	packuswb  xmm1,xmm1		; xmm1=R(13579BDF********)
-
-	paddw     xmm2,xmm4		; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
-	paddw     xmm3,xmm5		; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
-	packuswb  xmm2,xmm2		; xmm2=G(02468ACE********)
-	packuswb  xmm3,xmm3		; xmm3=G(13579BDF********)
-
-	paddw     xmm4, XMMWORD [wk(0)]	; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
-	paddw     xmm5, XMMWORD [wk(1)]	; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
-	packuswb  xmm4,xmm4		; xmm4=B(02468ACE********)
-	packuswb  xmm5,xmm5		; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
-	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
-	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
-	; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
-	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
-	punpcklbw xmmE,xmmB	; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
-	punpcklbw xmmD,xmmF	; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
-	movdqa    xmmG,xmmA
-	movdqa    xmmH,xmmA
-	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
-	punpckhwd xmmG,xmmE	; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
-	psrldq    xmmH,2	; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
-	psrldq    xmmE,2	; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
-	movdqa    xmmC,xmmD
-	movdqa    xmmB,xmmD
-	punpcklwd xmmD,xmmH	; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
-	punpckhwd xmmC,xmmH	; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
-	psrldq    xmmB,2	; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
-	movdqa    xmmF,xmmE
-	punpcklwd xmmE,xmmB	; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
-	punpckhwd xmmF,xmmB	; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
-	pshufd    xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
-	movdqa    xmmB,xmmE
-	punpckldq xmmA,xmmD	; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
-	punpckldq xmmE,xmmH	; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
-	punpckhdq xmmD,xmmB	; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
-	pshufd    xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
-	movdqa    xmmB,xmmF
-	punpckldq xmmG,xmmC	; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
-	punpckldq xmmF,xmmH	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
-	punpckhdq xmmC,xmmB	; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
-	punpcklqdq xmmA,xmmE	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
-	punpcklqdq xmmD,xmmG	; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
-	punpcklqdq xmmF,xmmC	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
-	cmp	ecx, byte SIZEOF_XMMWORD
-	jb	short .column_st32
-
-	test	edi, SIZEOF_XMMWORD-1
-	jnz	short .out1
-	; --(aligned)-------------------
-	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
-	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
-	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
-	jmp	short .out0
-.out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [edi], xmmF
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-.out0:
-	sub	ecx, byte SIZEOF_XMMWORD
-	jz	near .nextrow
-
-	add	esi, byte SIZEOF_XMMWORD	; inptr0
-	add	ebx, byte SIZEOF_XMMWORD	; inptr1
-	add	edx, byte SIZEOF_XMMWORD	; inptr2
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st32:
-	pcmpeqb	xmmH,xmmH			; xmmH=(all 1's)
-	lea	ecx, [ecx+ecx*2]		; imul ecx, RGB_PIXELSIZE
-	cmp	ecx, byte 2*SIZEOF_XMMWORD
-	jb	short .column_st16
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmF
-	sub	ecx, byte 2*SIZEOF_XMMWORD
-	jmp	short .column_st15
-.column_st16:
-	cmp	ecx, byte SIZEOF_XMMWORD
-	jb	short .column_st15
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmD
-	sub	ecx, byte SIZEOF_XMMWORD
-.column_st15:
-	mov	eax,ecx
-	xor	ecx, byte 0x0F
-	shl	ecx, 2
-	movd	xmmB,ecx
-	psrlq	xmmH,4
-	pcmpeqb	xmmE,xmmE
-	psrlq	xmmH,xmmB
-	psrlq	xmmE,xmmB
-	punpcklbw xmmE,xmmH
-	; ----------------
-	mov	ecx,edi
-	and	ecx, byte SIZEOF_XMMWORD-1
-	jz	short .adj0
-	add	eax,ecx
-	cmp	eax, byte SIZEOF_XMMWORD
-	ja	short .adj0
-	and	edi, byte (-SIZEOF_XMMWORD)	; align to 16-byte boundary
-	shl	ecx, 3			; pslldq xmmA,ecx & pslldq xmmE,ecx
-	movdqa	xmmG,xmmA
-	movdqa	xmmC,xmmE
-	pslldq	xmmA, SIZEOF_XMMWORD/2
-	pslldq	xmmE, SIZEOF_XMMWORD/2
-	movd	xmmD,ecx
-	sub	ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
-	jb	short .adj1
-	movd	xmmF,ecx
-	psllq	xmmA,xmmF
-	psllq	xmmE,xmmF
-	jmp	short .adj0
-.adj1:	neg	ecx
-	movd	xmmF,ecx
-	psrlq	xmmA,xmmF
-	psrlq	xmmE,xmmF
-	psllq	xmmG,xmmD
-	psllq	xmmC,xmmD
-	por	xmmA,xmmG
-	por	xmmE,xmmC
-.adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
-	pcmpeqb   xmm6,xmm6		; xmm6=XE=X(02468ACE********)
-	pcmpeqb   xmm7,xmm7		; xmm7=XO=X(13579BDF********)
-%else
-	pxor      xmm6,xmm6		; xmm6=XE=X(02468ACE********)
-	pxor      xmm7,xmm7		; xmm7=XO=X(13579BDF********)
-%endif
-	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
-	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
-	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
-	; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
-	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
-	punpcklbw xmmE,xmmG	; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
-	punpcklbw xmmB,xmmD	; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
-	punpcklbw xmmF,xmmH	; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
-	movdqa    xmmC,xmmA
-	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
-	punpckhwd xmmC,xmmE	; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
-	movdqa    xmmG,xmmB
-	punpcklwd xmmB,xmmF	; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
-	punpckhwd xmmG,xmmF	; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
-	movdqa    xmmD,xmmA
-	punpckldq xmmA,xmmB	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
-	punpckhdq xmmD,xmmB	; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
-	movdqa    xmmH,xmmC
-	punpckldq xmmC,xmmG	; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
-	punpckhdq xmmH,xmmG	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
-	cmp	ecx, byte SIZEOF_XMMWORD
-	jb	short .column_st32
-
-	test	edi, SIZEOF_XMMWORD-1
-	jnz	short .out1
-	; --(aligned)-------------------
-	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
-	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
-	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
-	movntdq	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
-	jmp	short .out0
-.out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [edi], xmmC
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [edi], xmmH
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-.out0:
-	sub	ecx, byte SIZEOF_XMMWORD
-	jz	near .nextrow
-
-	add	esi, byte SIZEOF_XMMWORD	; inptr0
-	add	ebx, byte SIZEOF_XMMWORD	; inptr1
-	add	edx, byte SIZEOF_XMMWORD	; inptr2
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st32:
-	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
-	cmp	ecx, byte SIZEOF_XMMWORD/2
-	jb	short .column_st16
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmC
-	movdqa	xmmD,xmmH
-	sub	ecx, byte SIZEOF_XMMWORD/2
-.column_st16:
-	cmp	ecx, byte SIZEOF_XMMWORD/4
-	jb	short .column_st15
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmD
-	sub	ecx, byte SIZEOF_XMMWORD/4
-.column_st15:
-	cmp	ecx, byte SIZEOF_XMMWORD/16
-	jb	short .nextrow
-	mov	eax,ecx
-	xor	ecx, byte 0x03
-	inc	ecx
-	shl	ecx, 4
-	movd	xmmF,ecx
-	psrlq	xmmE,xmmF
-	punpcklbw xmmE,xmmE
-	; ----------------
-	mov	ecx,edi
-	and	ecx, byte SIZEOF_XMMWORD-1
-	jz	short .adj0
-	lea	eax, [ecx+eax*4]	; RGB_PIXELSIZE
-	cmp	eax, byte SIZEOF_XMMWORD
-	ja	short .adj0
-	and	edi, byte (-SIZEOF_XMMWORD)	; align to 16-byte boundary
-	shl	ecx, 3			; pslldq xmmA,ecx & pslldq xmmE,ecx
-	movdqa	xmmB,xmmA
-	movdqa	xmmG,xmmE
-	pslldq	xmmA, SIZEOF_XMMWORD/2
-	pslldq	xmmE, SIZEOF_XMMWORD/2
-	movd	xmmC,ecx
-	sub	ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
-	jb	short .adj1
-	movd	xmmH,ecx
-	psllq	xmmA,xmmH
-	psllq	xmmE,xmmH
-	jmp	short .adj0
-.adj1:	neg	ecx
-	movd	xmmH,ecx
-	psrlq	xmmA,xmmH
-	psrlq	xmmE,xmmH
-	psllq	xmmB,xmmC
-	psllq	xmmG,xmmC
-	por	xmmA,xmmB
-	por	xmmE,xmmG
-.adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-	alignx	16,7
-
-.nextrow:
-	pop	ecx
-	pop	esi
-	pop	ebx
-	pop	edx
-	pop	edi
-	pop	eax
-
-	add	esi, byte SIZEOF_JSAMPROW
-	add	ebx, byte SIZEOF_JSAMPROW
-	add	edx, byte SIZEOF_JSAMPROW
-	add	edi, byte SIZEOF_JSAMPROW	; output_buf
-	dec	eax				; num_rows
-	jg	near .rowloop
-
-	sfence		; flush the write buffer
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxrgb_convert_sse2
+%include "jdclrss2.asm"
diff --git a/simd/jsimd.h b/simd/jsimd.h
index 371586e..a5b432f 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -21,11 +21,35 @@
 #ifdef NEED_SHORT_EXTERNAL_NAMES
 #define jpeg_simd_cpu_support                 jSiCpuSupport
 #define jsimd_rgb_ycc_convert_mmx             jSRGBYCCM
+#define jsimd_extrgb_ycc_convert_mmx          jSEXTRGBYCCM
+#define jsimd_extrgbx_ycc_convert_mmx         jSEXTRGBXYCCM
+#define jsimd_extbgr_ycc_convert_mmx          jSEXTBGRYCCM
+#define jsimd_extbgrx_ycc_convert_mmx         jSEXTBGRXYCCM
+#define jsimd_extxbgr_ycc_convert_mmx         jSEXTXBGRYCCM
+#define jsimd_extxrgb_ycc_convert_mmx         jSEXTXRGBYCCM
 #define jsimd_ycc_rgb_convert_mmx             jSYCCRGBM
+#define jsimd_ycc_extrgb_convert_mmx          jSYCCEXTRGBM
+#define jsimd_ycc_extrgbx_convert_mmx         jSYCCEXTRGBXM
+#define jsimd_ycc_extbgr_convert_mmx          jSYCCEXTBGRM
+#define jsimd_ycc_extbgrx_convert_mmx         jSYCCEXTBGRXM
+#define jsimd_ycc_extxbgr_convert_mmx         jSYCCEXTXBGRM
+#define jsimd_ycc_extxrgb_convert_mmx         jSYCCEXTXRGBM
 #define jconst_rgb_ycc_convert_sse2           jSCRGBYCCS2
 #define jsimd_rgb_ycc_convert_sse2            jSRGBYCCS2
+#define jsimd_extrgb_ycc_convert_sse2         jSEXTRGBYCCS2
+#define jsimd_extrgbx_ycc_convert_sse2        jSEXTRGBXYCCS2
+#define jsimd_extbgr_ycc_convert_sse2         jSEXTBGRYCCS2
+#define jsimd_extbgrx_ycc_convert_sse2        jSEXTBGRXYCCS2
+#define jsimd_extxbgr_ycc_convert_sse2        jSEXTXBGRYCCS2
+#define jsimd_extxrgb_ycc_convert_sse2        jSEXTXRGBYCCS2
 #define jconst_ycc_rgb_convert_sse2           jSCYCCRGBS2
 #define jsimd_ycc_rgb_convert_sse2            jSYCCRGBS2
+#define jsimd_ycc_extrgb_convert_sse2         jSYCCEXTRGBS2
+#define jsimd_ycc_extrgbx_convert_sse2        jSYCCEXTRGBXS2
+#define jsimd_ycc_extbgr_convert_sse2         jSYCCEXTBGRS2
+#define jsimd_ycc_extbgrx_convert_sse2        jSYCCEXTBGRXS2
+#define jsimd_ycc_extxbgr_convert_sse2        jSYCCEXTXBGRS2
+#define jsimd_ycc_extxrgb_convert_sse2        jSYCCEXTXRGBS2
 #define jsimd_h2v2_downsample_mmx             jSDnH2V2M
 #define jsimd_h2v1_downsample_mmx             jSDnH2V1M
 #define jsimd_h2v2_downsample_sse2            jSDnH2V2S2
@@ -89,21 +113,119 @@
         JPP((JDIMENSION img_width,
              JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
              JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extrgb_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extrgbx_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extbgr_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extbgrx_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extxbgr_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extxrgb_ycc_convert_mmx
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+
 EXTERN(void) jsimd_ycc_rgb_convert_mmx
         JPP((JDIMENSION out_width,
              JSAMPIMAGE input_buf, JDIMENSION input_row,
              JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extrgb_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extrgbx_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extbgr_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extbgrx_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extxbgr_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extxrgb_convert_mmx
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
 
 extern const int jconst_rgb_ycc_convert_sse2[];
 EXTERN(void) jsimd_rgb_ycc_convert_sse2
         JPP((JDIMENSION img_width,
              JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
              JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extrgb_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extrgbx_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extbgr_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extbgrx_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extxbgr_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+EXTERN(void) jsimd_extxrgb_ycc_convert_sse2
+        JPP((JDIMENSION img_width,
+             JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+             JDIMENSION output_row, int num_rows));
+
 extern const int jconst_ycc_rgb_convert_sse2[];
 EXTERN(void) jsimd_ycc_rgb_convert_sse2
         JPP((JDIMENSION out_width,
              JSAMPIMAGE input_buf, JDIMENSION input_row,
              JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extrgb_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extrgbx_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extbgr_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extbgrx_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extxbgr_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
+EXTERN(void) jsimd_ycc_extxrgb_convert_sse2
+        JPP((JDIMENSION out_width,
+             JSAMPIMAGE input_buf, JDIMENSION input_row,
+             JSAMPARRAY output_buf, int num_rows));
 
 /* SIMD Downsample */
 EXTERN(void) jsimd_h2v2_downsample_mmx