Skip to content

L
libyuv
Commit 080a3164 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

port yuv chroma improvements to gcc. YUV to RGB is more accurate using a… 

port yuv chroma improvements to gcc.  YUV to RGB is more accurate using a negative matrix.  2% slower but half as much error.
BUG=324
TESTED=try bots
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/41629004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1249 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent d12a0871
Show whitespace changes
Inline Side-by-side
Showing with 53 additions and 42 deletions
README.chromium
View file @ 080a3164
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 1248 Version: 1249
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
include/libyuv/version.h
View file @ 080a3164
...@@ -11,6 +11,6 @@ ...@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1248 #define LIBYUV_VERSION 1249
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/row_posix.cc
View file @ 080a3164
...@@ -1518,21 +1518,21 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, ...@@ -1518,21 +1518,21 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba,
} }
#ifdef HAS_I422TOARGBROW_SSSE3 #ifdef HAS_I422TOARGBROW_SSSE3
#define YG 19071 /* round(1.164 * 64 * 256) */
#define YGB 1192 /* round(1.164 * 64 * 16) */
#define UB 127 /* min(63,(int8)round(2.018 * 64)) */
#define UG -25 /* (int8)round(-0.391 * 64 - 0.5) */
#define UR 0
#define VB 0 // YUV to RGB conversion constants.
#define VG -52 /* (int8)round(-0.813 * 64 - 0.5) */ #define YG 19071 /* round(1.164 * 64 * 256) */
#define VR 102 /* (int8)round(1.596 * 64 + 0.5) */ #define YGB 1197 /* 1.164 * 64 * 16 - adjusted for even error distribution */
#define UB -128 /* -min(128, round(2.018 * 64)) */
// Bias #define UBB -16385 /* approx -round(2.018 * 64 * 128) */
#define BB (UB * 128 + VB * 128 + YGB) #define UG 25 /* -round(-0.391 * 64) */
#define BG (UG * 128 + VG * 128 + YGB) #define UGB 3200 /* -round(-0.391 * 64) * 128 */
#define BR (UR * 128 + VR * 128 + YGB) #define VG 52 /* -round(-0.813 * 64) */
#define VGB 6656 /* -round(-0.813 * 64) * 128 */
#define VR -102 /* -round(1.596 * 64) */
#define VRB -13056 /* -round(1.596 * 64) * 128 */
#define BB (UBB - YGB)
#define BG (UGB + VGB - YGB)
#define BR ( VRB - YGB)
struct { struct {
vec8 kUVToB; // 0 vec8 kUVToB; // 0
...@@ -1546,16 +1546,16 @@ struct { ...@@ -1546,16 +1546,16 @@ struct {
vec8 kVUToG; // 128 vec8 kVUToG; // 128
vec8 kVUToR; // 144 vec8 kVUToR; // 144
} static SIMD_ALIGNED(kYuvConstants) = { } static SIMD_ALIGNED(kYuvConstants) = {
{ UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB }, { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 },
{ UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG },
{ UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR }, { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR },
{ BB, BB, BB, BB, BB, BB, BB, BB }, { BB, BB, BB, BB, BB, BB, BB, BB },
{ BG, BG, BG, BG, BG, BG, BG, BG }, { BG, BG, BG, BG, BG, BG, BG, BG },
{ BR, BR, BR, BR, BR, BR, BR, BR }, { BR, BR, BR, BR, BR, BR, BR, BR },
{ YG, YG, YG, YG, YG, YG, YG, YG }, { YG, YG, YG, YG, YG, YG, YG, YG },
{ VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB }, { 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB },
{ VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG }, { VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG },
{ VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR } { VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0 }
}; };
// Read 8 UV from 411 // Read 8 UV from 411
...@@ -1592,12 +1592,16 @@ struct { ...@@ -1592,12 +1592,16 @@ struct {
#define YUVTORGB \ #define YUVTORGB \
"movdqa %%xmm0,%%xmm1 \n" \ "movdqa %%xmm0,%%xmm1 \n" \
"movdqa %%xmm0,%%xmm2 \n" \ "movdqa %%xmm0,%%xmm2 \n" \
"pmaddubsw " MEMACCESS([kYuvConstants]) ",%%xmm0 \n" \ "movdqa %%xmm0,%%xmm3 \n" \
"pmaddubsw " MEMACCESS2(16, [kYuvConstants]) ",%%xmm1 \n" \ "movdqa " MEMACCESS2(48, [kYuvConstants]) ",%%xmm0 \n" \
"pmaddubsw " MEMACCESS2(32, [kYuvConstants]) ",%%xmm2 \n" \ "pmaddubsw " MEMACCESS([kYuvConstants]) ",%%xmm1 \n" \
"psubw " MEMACCESS2(48, [kYuvConstants]) ",%%xmm0 \n" \ "psubw %%xmm1,%%xmm0 \n" \
"psubw " MEMACCESS2(64, [kYuvConstants]) ",%%xmm1 \n" \ "movdqa " MEMACCESS2(64, [kYuvConstants]) ",%%xmm1 \n" \
"psubw " MEMACCESS2(80, [kYuvConstants]) ",%%xmm2 \n" \ "pmaddubsw " MEMACCESS2(16, [kYuvConstants]) ",%%xmm2 \n" \
"psubw %%xmm2,%%xmm1 \n" \
"movdqa " MEMACCESS2(80, [kYuvConstants]) ",%%xmm2 \n" \
"pmaddubsw " MEMACCESS2(32, [kYuvConstants]) ",%%xmm3 \n" \
"psubw %%xmm3,%%xmm2 \n" \
"movq " MEMACCESS([y_buf]) ",%%xmm3 \n" \ "movq " MEMACCESS([y_buf]) ",%%xmm3 \n" \
"lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \ "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \
"punpcklbw %%xmm3,%%xmm3 \n" \ "punpcklbw %%xmm3,%%xmm3 \n" \
...@@ -1616,12 +1620,16 @@ struct { ...@@ -1616,12 +1620,16 @@ struct {
#define YVUTORGB \ #define YVUTORGB \
"movdqa %%xmm0,%%xmm1 \n" \ "movdqa %%xmm0,%%xmm1 \n" \
"movdqa %%xmm0,%%xmm2 \n" \ "movdqa %%xmm0,%%xmm2 \n" \
"pmaddubsw " MEMACCESS2(112, [kYuvConstants]) ",%%xmm0 \n" \ "movdqa %%xmm0,%%xmm3 \n" \
"pmaddubsw " MEMACCESS2(128, [kYuvConstants]) ",%%xmm1 \n" \ "movdqa " MEMACCESS2(48, [kYuvConstants]) ",%%xmm0 \n" \
"pmaddubsw " MEMACCESS2(144, [kYuvConstants]) ",%%xmm2 \n" \ "pmaddubsw " MEMACCESS2(112, [kYuvConstants]) ",%%xmm1 \n" \
"psubw " MEMACCESS2(48, [kYuvConstants]) ",%%xmm0 \n" \ "psubw %%xmm1,%%xmm0 \n" \
"psubw " MEMACCESS2(64, [kYuvConstants]) ",%%xmm1 \n" \ "movdqa " MEMACCESS2(64, [kYuvConstants]) ",%%xmm1 \n" \
"psubw " MEMACCESS2(80, [kYuvConstants]) ",%%xmm2 \n" \ "pmaddubsw " MEMACCESS2(128, [kYuvConstants]) ",%%xmm2 \n" \
"psubw %%xmm2,%%xmm1 \n" \
"movdqa " MEMACCESS2(80, [kYuvConstants]) ",%%xmm2 \n" \
"pmaddubsw " MEMACCESS2(144, [kYuvConstants]) ",%%xmm3 \n" \
"psubw %%xmm3,%%xmm2 \n" \
"movq " MEMACCESS([y_buf]) ",%%xmm3 \n" \ "movq " MEMACCESS([y_buf]) ",%%xmm3 \n" \
"lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \ "lea " MEMLEA(0x8, [y_buf]) ",%[y_buf] \n" \
"punpcklbw %%xmm3,%%xmm3 \n" \ "punpcklbw %%xmm3,%%xmm3 \n" \
...@@ -2043,12 +2051,12 @@ struct { ...@@ -2043,12 +2051,12 @@ struct {
lvec16 kUVBiasR_AVX; // 160 lvec16 kUVBiasR_AVX; // 160
lvec16 kYToRgb_AVX; // 192 lvec16 kYToRgb_AVX; // 192
} static SIMD_ALIGNED(kYuvConstants_AVX) = { } static SIMD_ALIGNED(kYuvConstants_AVX) = {
{ UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0,
UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB, UB, VB }, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 },
{ UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG,
UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG },
{ UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR,
UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR, UR, VR }, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR },
{ BB, BB, BB, BB, BB, BB, BB, BB, { BB, BB, BB, BB, BB, BB, BB, BB,
BB, BB, BB, BB, BB, BB, BB, BB }, BB, BB, BB, BB, BB, BB, BB, BB },
{ BG, BG, BG, BG, BG, BG, BG, BG, { BG, BG, BG, BG, BG, BG, BG, BG,
...@@ -2074,9 +2082,12 @@ struct { ...@@ -2074,9 +2082,12 @@ struct {
"vpmaddubsw " MEMACCESS2(64, [kYuvConstants]) ",%%ymm0,%%ymm2 \n" \ "vpmaddubsw " MEMACCESS2(64, [kYuvConstants]) ",%%ymm0,%%ymm2 \n" \
"vpmaddubsw " MEMACCESS2(32, [kYuvConstants]) ",%%ymm0,%%ymm1 \n" \ "vpmaddubsw " MEMACCESS2(32, [kYuvConstants]) ",%%ymm0,%%ymm1 \n" \
"vpmaddubsw " MEMACCESS([kYuvConstants]) ",%%ymm0,%%ymm0 \n" \ "vpmaddubsw " MEMACCESS([kYuvConstants]) ",%%ymm0,%%ymm0 \n" \
"vpsubw " MEMACCESS2(160, [kYuvConstants]) ",%%ymm2,%%ymm2 \n" \ "vmovdqu " MEMACCESS2(160, [kYuvConstants]) ",%%ymm3 \n" \
"vpsubw " MEMACCESS2(128, [kYuvConstants]) ",%%ymm1,%%ymm1 \n" \ "vpsubw %%ymm3,%%ymm2,%%ymm2 \n" \
"vpsubw " MEMACCESS2(96, [kYuvConstants]) ",%%ymm0,%%ymm0 \n" \ "vmovdqu " MEMACCESS2(128, [kYuvConstants]) ",%%ymm2 \n" \
"vpsubw %%ymm2,%%ymm1,%%ymm1 \n" \
"vmovdqu " MEMACCESS2(96, [kYuvConstants]) ",%%ymm1 \n" \
"vpsubw %%ymm1,%%ymm0,%%ymm0 \n" \
"vmovdqu " MEMACCESS([y_buf]) ",%%xmm3 \n" \ "vmovdqu " MEMACCESS([y_buf]) ",%%xmm3 \n" \
"lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" \ "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" \
"vpermq $0xd8,%%ymm3,%%ymm3 \n" \ "vpermq $0xd8,%%ymm3,%%ymm3 \n" \
...@@ -2285,10 +2296,10 @@ void YToARGBRow_SSE2(const uint8* y_buf, ...@@ -2285,10 +2296,10 @@ void YToARGBRow_SSE2(const uint8* y_buf,
asm volatile ( asm volatile (
"pcmpeqb %%xmm4,%%xmm4 \n" "pcmpeqb %%xmm4,%%xmm4 \n"
"pslld $0x18,%%xmm4 \n" "pslld $0x18,%%xmm4 \n"
"mov $0x04a804a8,%%eax \n" "mov $0x04ad04ad,%%eax \n" // 04ad = 1197 = 1.164 * 16
"movd %%eax,%%xmm3 \n" "movd %%eax,%%xmm3 \n"
"pshufd $0x0,%%xmm3,%%xmm3 \n" "pshufd $0x0,%%xmm3,%%xmm3 \n"
"mov $0x4a7f4a7f,%%eax \n" "mov $0x4a7f4a7f,%%eax \n" // 4a7f = 19071 = 1.164
"movd %%eax,%%xmm2 \n" "movd %%eax,%%xmm2 \n"
"pshufd $0x0,%%xmm2,%%xmm2 \n" "pshufd $0x0,%%xmm2,%%xmm2 \n"
LABELALIGN LABELALIGN
......
source/row_win.cc
View file @ 080a3164
...@@ -2324,7 +2324,7 @@ void YToARGBRow_SSE2(const uint8* y_buf, ...@@ -2324,7 +2324,7 @@ void YToARGBRow_SSE2(const uint8* y_buf,
lea eax, [eax + 8] lea eax, [eax + 8]
punpcklbw xmm0, xmm0 // Y.Y punpcklbw xmm0, xmm0 // Y.Y
pmulhuw xmm0, xmm2 pmulhuw xmm0, xmm2
psubusw xmm0, xmm3 // TODO(fbarchard): round 0.5 psubusw xmm0, xmm3
psrlw xmm0, 6 psrlw xmm0, 6
packuswb xmm0, xmm0 // G packuswb xmm0, xmm0 // G
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment