Recomputed JPeg coefficients normalized to 128. Apply to ARGBGray function reusing YJ function/coefficients and rounding.
BUG=201
TESTED=Gray unittest improved
Review URL: https://webrtc-codereview.appspot.com/1269006
git-svn-id: http://libyuv.googlecode.com/svn/trunk@629 16f28f9a-4ce2-e073-06de-1de4eb20be90
diff --git a/source/row_posix.cc b/source/row_posix.cc
index 4f722c7..74ae032 100644
--- a/source/row_posix.cc
+++ b/source/row_posix.cc
@@ -37,17 +37,25 @@
// JPeg full range.
CONST vec8 kARGBToYJ = {
- 15, 74, 38, 0, 15, 74, 38, 0, 15, 74, 38, 0, 15, 74, 38, 0
+ 15, 75, 38, 0, 15, 75, 38, 0, 15, 75, 38, 0, 15, 75, 38, 0
};
CONST vec8 kARGBToU = {
112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38, 0
};
+CONST vec8 kARGBToUJ = {
+ 127, -84, -43, 0, 127, -84, -43, 0, 127, -84, -43, 0, 127, -84, -43, 0
+};
+
CONST vec8 kARGBToV = {
-18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0,
};
+CONST vec8 kARGBToVJ = {
+ -20, -107, 127, 0, -20, -107, 127, 0, -20, -107, 127, 0, -20, -107, 127, 0
+};
+
// Constants for BGRA
CONST vec8 kBGRAToY = {
0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13, 0, 33, 65, 13
@@ -100,6 +108,10 @@
128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u
};
+CONST uvec16 kAddUVJ128 = {
+ 0x8080u, 0x8080u, 0x8080u, 0x8080u, 0x8080u, 0x8080u, 0x8080u, 0x8080u
+};
+
// Shuffle table for converting RGB24 to ARGB.
CONST uvec8 kShuffleMaskRGB24ToARGB = {
0u, 1u, 2u, 12u, 3u, 4u, 5u, 13u, 6u, 7u, 8u, 14u, 9u, 10u, 11u, 15u
@@ -830,6 +842,69 @@
);
}
+// TODO(fbarchard): Share code with ARGBToUVRow_SSSE3.
+void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb,
+ uint8* dst_u, uint8* dst_v, int width) {
+ asm volatile (
+ "movdqa %0,%%xmm4 \n"
+ "movdqa %1,%%xmm3 \n"
+ "movdqa %2,%%xmm5 \n"
+ :
+ : "m"(kARGBToUJ), // %0
+ "m"(kARGBToVJ), // %1
+ "m"(kAddUVJ128) // %2
+ );
+ asm volatile (
+ "sub %1,%2 \n"
+ ".p2align 4 \n"
+ "1: \n"
+ "movdqa (%0),%%xmm0 \n"
+ "movdqa 0x10(%0),%%xmm1 \n"
+ "movdqa 0x20(%0),%%xmm2 \n"
+ "movdqa 0x30(%0),%%xmm6 \n"
+ "pavgb (%0,%4,1),%%xmm0 \n"
+ "pavgb 0x10(%0,%4,1),%%xmm1 \n"
+ "pavgb 0x20(%0,%4,1),%%xmm2 \n"
+ "pavgb 0x30(%0,%4,1),%%xmm6 \n"
+ "lea 0x40(%0),%0 \n"
+ "movdqa %%xmm0,%%xmm7 \n"
+ "shufps $0x88,%%xmm1,%%xmm0 \n"
+ "shufps $0xdd,%%xmm1,%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm0 \n"
+ "movdqa %%xmm2,%%xmm7 \n"
+ "shufps $0x88,%%xmm6,%%xmm2 \n"
+ "shufps $0xdd,%%xmm6,%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm2 \n"
+ "movdqa %%xmm0,%%xmm1 \n"
+ "movdqa %%xmm2,%%xmm6 \n"
+ "pmaddubsw %%xmm4,%%xmm0 \n"
+ "pmaddubsw %%xmm4,%%xmm2 \n"
+ "pmaddubsw %%xmm3,%%xmm1 \n"
+ "pmaddubsw %%xmm3,%%xmm6 \n"
+ "phaddw %%xmm2,%%xmm0 \n"
+ "phaddw %%xmm6,%%xmm1 \n"
+ "paddw %%xmm5,%%xmm0 \n"
+ "paddw %%xmm5,%%xmm1 \n"
+ "psraw $0x8,%%xmm0 \n"
+ "psraw $0x8,%%xmm1 \n"
+ "packsswb %%xmm1,%%xmm0 \n"
+ "sub $0x10,%3 \n"
+ "movlps %%xmm0,(%1) \n"
+ "movhps %%xmm0,(%1,%2,1) \n"
+ "lea 0x8(%1),%1 \n"
+ "jg 1b \n"
+ : "+r"(src_argb0), // %0
+ "+r"(dst_u), // %1
+ "+r"(dst_v), // %2
+ "+rm"(width) // %3
+ : "r"(static_cast<intptr_t>(src_stride_argb))
+ : "memory", "cc"
+#if defined(__SSE2__)
+ , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
+#endif
+ );
+}
+
void ARGBToUVRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb,
uint8* dst_u, uint8* dst_v, int width) {
asm volatile (
@@ -895,6 +970,72 @@
);
}
+void ARGBToUVJRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb,
+ uint8* dst_u, uint8* dst_v, int width) {
+ asm volatile (
+ "movdqa %0,%%xmm4 \n"
+ "movdqa %1,%%xmm3 \n"
+ "movdqa %2,%%xmm5 \n"
+ :
+ : "m"(kARGBToUJ), // %0
+ "m"(kARGBToVJ), // %1
+ "m"(kAddUVJ128) // %2
+ );
+ asm volatile (
+ "sub %1,%2 \n"
+ ".p2align 4 \n"
+ "1: \n"
+ "movdqu (%0),%%xmm0 \n"
+ "movdqu 0x10(%0),%%xmm1 \n"
+ "movdqu 0x20(%0),%%xmm2 \n"
+ "movdqu 0x30(%0),%%xmm6 \n"
+ "movdqu (%0,%4,1),%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm0 \n"
+ "movdqu 0x10(%0,%4,1),%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm1 \n"
+ "movdqu 0x20(%0,%4,1),%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm2 \n"
+ "movdqu 0x30(%0,%4,1),%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm6 \n"
+ "lea 0x40(%0),%0 \n"
+ "movdqa %%xmm0,%%xmm7 \n"
+ "shufps $0x88,%%xmm1,%%xmm0 \n"
+ "shufps $0xdd,%%xmm1,%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm0 \n"
+ "movdqa %%xmm2,%%xmm7 \n"
+ "shufps $0x88,%%xmm6,%%xmm2 \n"
+ "shufps $0xdd,%%xmm6,%%xmm7 \n"
+ "pavgb %%xmm7,%%xmm2 \n"
+ "movdqa %%xmm0,%%xmm1 \n"
+ "movdqa %%xmm2,%%xmm6 \n"
+ "pmaddubsw %%xmm4,%%xmm0 \n"
+ "pmaddubsw %%xmm4,%%xmm2 \n"
+ "pmaddubsw %%xmm3,%%xmm1 \n"
+ "pmaddubsw %%xmm3,%%xmm6 \n"
+ "phaddw %%xmm2,%%xmm0 \n"
+ "phaddw %%xmm6,%%xmm1 \n"
+ "paddw %%xmm5,%%xmm0 \n"
+ "paddw %%xmm5,%%xmm1 \n"
+ "psraw $0x8,%%xmm0 \n"
+ "psraw $0x8,%%xmm1 \n"
+ "packsswb %%xmm1,%%xmm0 \n"
+ "sub $0x10,%3 \n"
+ "movlps %%xmm0,(%1) \n"
+ "movhps %%xmm0,(%1,%2,1) \n"
+ "lea 0x8(%1),%1 \n"
+ "jg 1b \n"
+ : "+r"(src_argb0), // %0
+ "+r"(dst_u), // %1
+ "+r"(dst_v), // %2
+ "+rm"(width) // %3
+ : "r"(static_cast<intptr_t>(src_stride_argb))
+ : "memory", "cc"
+#if defined(__SSE2__)
+ , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
+#endif
+ );
+}
+
void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v,
int width) {
asm volatile (
@@ -3764,15 +3905,11 @@
#endif // HAS_ARGBUNATTENUATEROW_SSE2
#ifdef HAS_ARGBGRAYROW_SSSE3
-// Constant for ARGB color to gray scale. 0.11 * B + 0.59 * G + 0.30 * R
-CONST vec8 kARGBToGray = {
- 14, 76, 38, 0, 14, 76, 38, 0, 14, 76, 38, 0, 14, 76, 38, 0
-};
-
// Convert 8 ARGB pixels (64 bytes) to 8 Gray ARGB pixels
void ARGBGrayRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
asm volatile (
"movdqa %3,%%xmm4 \n"
+ "movdqa %4,%%xmm5 \n"
"sub %0,%1 \n"
// 8 pixel loop.
@@ -3783,6 +3920,7 @@
"pmaddubsw %%xmm4,%%xmm0 \n"
"pmaddubsw %%xmm4,%%xmm1 \n"
"phaddw %%xmm1,%%xmm0 \n"
+ "paddw %%xmm5,%%xmm0 \n"
"psrlw $0x7,%%xmm0 \n"
"packuswb %%xmm0,%%xmm0 \n"
"movdqa (%0),%%xmm2 \n"
@@ -3805,10 +3943,11 @@
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+r"(width) // %2
- : "m"(kARGBToGray) // %3
+ : "m"(kARGBToYJ), // %3
+ "m"(kAddYJ64) // %4
: "memory", "cc"
#if defined(__SSE2__)
- , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
+ , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
#endif
);
}