coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-02-09 11:16:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enable SSE version of I420ToRAW/RGB24 for Linux/Mac/ChromeOS x86

Browse Source 
BUG=116
TEST=xcodebuild/Release/libyuv_unittest --gtest_filter=*I420To*R*Opt
Review URL: https://webrtc-codereview.appspot.com/863015

git-svn-id: http://libyuv.googlecode.com/svn/trunk@402 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2012-10-11 01:25:46 +00:00
parent 8d37dd5c20
commit 4de0c439aa
5 changed files with 241 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 401 Version: 402
License: BSD License: BSD
License File: LICENSE License File: LICENSE

12
include/libyuv/row.h
View File

@ -36,6 +36,7 @@ extern "C" {
(defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)) (defined(_M_IX86) || defined(__x86_64__) || defined(__i386__))
// Conversions. // Conversions.
#define HAS_ABGRTOARGBROW_SSSE3 #define HAS_ABGRTOARGBROW_SSSE3
#define HAS_ABGRTOARGBROW_SSSE3
#define HAS_ABGRTOUVROW_SSSE3 #define HAS_ABGRTOUVROW_SSSE3
#define HAS_ABGRTOYROW_SSSE3 #define HAS_ABGRTOYROW_SSSE3
#define HAS_ARGB1555TOARGBROW_SSE2 #define HAS_ARGB1555TOARGBROW_SSE2
@ -60,6 +61,8 @@ extern "C" {
#define HAS_I422TOABGRROW_SSSE3 #define HAS_I422TOABGRROW_SSSE3
#define HAS_I422TOARGBROW_SSSE3 #define HAS_I422TOARGBROW_SSSE3
#define HAS_I422TOBGRAROW_SSSE3 #define HAS_I422TOBGRAROW_SSSE3
#define HAS_I422TORAWROW_SSSE3
#define HAS_I422TORGB24ROW_SSSE3
#define HAS_I422TORGBAROW_SSSE3 #define HAS_I422TORGBAROW_SSSE3
#define HAS_I444TOARGBROW_SSSE3 #define HAS_I444TOARGBROW_SSSE3
#define HAS_MIRRORROW_SSSE3 #define HAS_MIRRORROW_SSSE3
@ -69,6 +72,9 @@ extern "C" {
#define HAS_RAWTOARGBROW_SSSE3 #define HAS_RAWTOARGBROW_SSSE3
#define HAS_RGB24TOARGBROW_SSSE3 #define HAS_RGB24TOARGBROW_SSSE3
#define HAS_RGB565TOARGBROW_SSE2 #define HAS_RGB565TOARGBROW_SSE2
#define HAS_RGBATOARGBROW_SSSE3
#define HAS_RGBATOUVROW_SSSE3
#define HAS_RGBATOYROW_SSSE3
#define HAS_SETROW_X86 #define HAS_SETROW_X86
#define HAS_SPLITUV_SSE2 #define HAS_SPLITUV_SSE2
#define HAS_UYVYTOUV422ROW_SSE2 #define HAS_UYVYTOUV422ROW_SSE2
@ -97,13 +103,7 @@ extern "C" {
// The following are Windows only. TODO(fbarchard): Port to gcc. // The following are Windows only. TODO(fbarchard): Port to gcc.
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86) #if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
#define HAS_ABGRTOARGBROW_SSSE3
#define HAS_ARGBCOLORTABLEROW_X86 #define HAS_ARGBCOLORTABLEROW_X86
#define HAS_RGBATOARGBROW_SSSE3
#define HAS_RGBATOUVROW_SSSE3
#define HAS_RGBATOYROW_SSSE3
#define HAS_I422TORGB24ROW_SSSE3
#define HAS_I422TORAWROW_SSSE3
#endif #endif
// The following are disabled when SSSE3 is available: // The following are disabled when SSSE3 is available:

2
include/libyuv/version.h
View File

@ -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 401 #define LIBYUV_VERSION 402
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

250
source/row_posix.cc
View File

@ -69,6 +69,19 @@ CONST vec8 kABGRToV = {
112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0, 112, -94, -18, 0
}; };
// Constants for RGBA.
CONST vec8 kRGBAToY = {
0, 13, 65, 33, 0, 13, 65, 33, 0, 13, 65, 33, 0, 13, 65, 33
};
CONST vec8 kRGBAToU = {
0, 112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38, 0, 112, -74, -38
};
CONST vec8 kRGBAToV = {
0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112
};
CONST uvec8 kAddY16 = { CONST uvec8 kAddY16 = {
16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u
}; };
@ -118,7 +131,7 @@ CONST uvec8 kShuffleMaskARGBToRAW = {
2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u 2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u
}; };
// Shuffle table for converting ARGBToRGB24 for I420ToRGB24. First 8 + next 4 // Shuffle table for converting ARGBToRGB24 for I422ToRGB24. First 8 + next 4
CONST uvec8 kShuffleMaskARGBToRGB24_0 = { CONST uvec8 kShuffleMaskARGBToRGB24_0 = {
0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u 0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u
}; };
@ -1155,6 +1168,80 @@ void ABGRToYRow_Unaligned_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix) {
); );
} }
void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) {
asm volatile (
"movdqa %4,%%xmm5 \n"
"movdqa %3,%%xmm4 \n"
".p2align 4 \n"
"1: \n"
"movdqa (%0),%%xmm0 \n"
"movdqa 0x10(%0),%%xmm1 \n"
"movdqa 0x20(%0),%%xmm2 \n"
"movdqa 0x30(%0),%%xmm3 \n"
"pmaddubsw %%xmm4,%%xmm0 \n"
"pmaddubsw %%xmm4,%%xmm1 \n"
"pmaddubsw %%xmm4,%%xmm2 \n"
"pmaddubsw %%xmm4,%%xmm3 \n"
"lea 0x40(%0),%0 \n"
"phaddw %%xmm1,%%xmm0 \n"
"phaddw %%xmm3,%%xmm2 \n"
"psrlw $0x7,%%xmm0 \n"
"psrlw $0x7,%%xmm2 \n"
"packuswb %%xmm2,%%xmm0 \n"
"paddb %%xmm5,%%xmm0 \n"
"sub $0x10,%2 \n"
"movdqa %%xmm0,(%1) \n"
"lea 0x10(%1),%1 \n"
"jg 1b \n"
: "+r"(src_rgba), // %0
"+r"(dst_y), // %1
"+r"(pix) // %2
: "m"(kRGBAToY), // %3
"m"(kAddY16) // %4
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
#endif
);
}
void RGBAToYRow_Unaligned_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) {
asm volatile (
"movdqa %4,%%xmm5 \n"
"movdqa %3,%%xmm4 \n"
".p2align 4 \n"
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu 0x10(%0),%%xmm1 \n"
"movdqu 0x20(%0),%%xmm2 \n"
"movdqu 0x30(%0),%%xmm3 \n"
"pmaddubsw %%xmm4,%%xmm0 \n"
"pmaddubsw %%xmm4,%%xmm1 \n"
"pmaddubsw %%xmm4,%%xmm2 \n"
"pmaddubsw %%xmm4,%%xmm3 \n"
"lea 0x40(%0),%0 \n"
"phaddw %%xmm1,%%xmm0 \n"
"phaddw %%xmm3,%%xmm2 \n"
"psrlw $0x7,%%xmm0 \n"
"psrlw $0x7,%%xmm2 \n"
"packuswb %%xmm2,%%xmm0 \n"
"paddb %%xmm5,%%xmm0 \n"
"sub $0x10,%2 \n"
"movdqu %%xmm0,(%1) \n"
"lea 0x10(%1),%1 \n"
"jg 1b \n"
: "+r"(src_rgba), // %0
"+r"(dst_y), // %1
"+r"(pix) // %2
: "m"(kRGBAToY), // %3
"m"(kAddY16) // %4
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
#endif
);
}
void ABGRToUVRow_SSSE3(const uint8* src_abgr0, int src_stride_abgr, void ABGRToUVRow_SSSE3(const uint8* src_abgr0, int src_stride_abgr,
uint8* dst_u, uint8* dst_v, int width) { uint8* dst_u, uint8* dst_v, int width) {
asm volatile ( asm volatile (
@ -1280,6 +1367,132 @@ void ABGRToUVRow_Unaligned_SSSE3(const uint8* src_abgr0, int src_stride_abgr,
#endif #endif
); );
} }
void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba,
uint8* dst_u, uint8* dst_v, int width) {
asm volatile (
"movdqa %0,%%xmm4 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm5 \n"
:
: "m"(kRGBAToU), // %0
"m"(kRGBAToV), // %1
"m"(kAddUV128) // %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"
"psraw $0x8,%%xmm0 \n"
"psraw $0x8,%%xmm1 \n"
"packsswb %%xmm1,%%xmm0 \n"
"paddb %%xmm5,%%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_rgba0), // %0
"+r"(dst_u), // %1
"+r"(dst_v), // %2
"+rm"(width) // %3
: "r"(static_cast<intptr_t>(src_stride_rgba))
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
#endif
);
}
void RGBAToUVRow_Unaligned_SSSE3(const uint8* src_rgba0, int src_stride_rgba,
uint8* dst_u, uint8* dst_v, int width) {
asm volatile (
"movdqa %0,%%xmm4 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm5 \n"
:
: "m"(kRGBAToU), // %0
"m"(kRGBAToV), // %1
"m"(kAddUV128) // %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"
"psraw $0x8,%%xmm0 \n"
"psraw $0x8,%%xmm1 \n"
"packsswb %%xmm1,%%xmm0 \n"
"paddb %%xmm5,%%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_rgba0), // %0
"+r"(dst_u), // %1
"+r"(dst_v), // %2
"+rm"(width) // %3
: "r"(static_cast<intptr_t>(src_stride_rgba))
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
#endif
);
}
#endif // HAS_ARGBTOYROW_SSSE3 #endif // HAS_ARGBTOYROW_SSSE3
#ifdef HAS_I422TOARGBROW_SSSE3 #ifdef HAS_I422TOARGBROW_SSSE3
@ -1449,16 +1662,16 @@ void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
// fpic 32 bit gcc 4.2 on OSX runs out of GPR regs. // fpic 32 bit gcc 4.2 on OSX runs out of GPR regs.
#ifdef __APPLE__ #ifdef __APPLE__
asm volatile ( asm volatile (
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm5 \n" "movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm6 \n" "movdqa %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
:: [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24), :: [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
[kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0)); [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24));
#endif #endif
asm volatile ( asm volatile (
#ifndef __APPLE__ #ifndef __APPLE__
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm5 \n" "movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm6 \n" "movdqa %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
#endif #endif
"sub %[u_buf],%[v_buf] \n" "sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n" "pxor %%xmm4,%%xmm4 \n"
@ -1486,8 +1699,8 @@ void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
[width]"+rm"(width) // %[width] [width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB) : [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#ifndef __APPLE__ #ifndef __APPLE__
, [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24), , [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
[kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0) [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24)
#endif #endif
: "memory", "cc" : "memory", "cc"
#if defined(__SSE2__) #if defined(__SSE2__)
@ -1501,18 +1714,19 @@ void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf,
const uint8* v_buf, const uint8* v_buf,
uint8* raw_buf, uint8* raw_buf,
int width) { int width) {
// fpic 32 bit gcc 4.2 on OSX runs out of GPR regs.
#ifdef __APPLE__ #ifdef __APPLE__
asm volatile ( asm volatile (
"movdqa %[kShuffleMaskARGBToRAW],%%xmm5 \n" "movdqa %[kShuffleMaskARGBToRAW_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm6 \n" "movdqa %[kShuffleMaskARGBToRAW],%%xmm6 \n"
:: [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW), :: [kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0),
[kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0)); [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW));
#endif #endif
asm volatile ( asm volatile (
#ifndef __APPLE__ #ifndef __APPLE__
"movdqa %[kShuffleMaskARGBToRAW],%%xmm5 \n" "movdqa %[kShuffleMaskARGBToRAW_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm6 \n" "movdqa %[kShuffleMaskARGBToRAW],%%xmm6 \n"
#endif #endif
"sub %[u_buf],%[v_buf] \n" "sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n" "pxor %%xmm4,%%xmm4 \n"
@ -1540,12 +1754,12 @@ void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf,
[width]"+rm"(width) // %[width] [width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB) : [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#ifndef __APPLE__ #ifndef __APPLE__
, [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW), , [kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0),
[kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0) [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW)
#endif #endif
: "memory", "cc" : "memory", "cc"
#if defined(__SSE2__) #if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
#endif #endif
); );
} }

3
source/row_win.cc
View File

@ -18,7 +18,6 @@ extern "C" {
// This module is for Visual C x86. // This module is for Visual C x86.
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86) #if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
// TODO(fbarchard): I420ToRGB24, I420ToRAW
#ifdef HAS_ARGBTOYROW_SSSE3 #ifdef HAS_ARGBTOYROW_SSSE3
// Constants for ARGB. // Constants for ARGB.
@ -122,7 +121,7 @@ static const uvec8 kShuffleMaskARGBToRAW = {
2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u 2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u
}; };
// Shuffle table for converting ARGBToRGB24 for I420ToRGB24. First 8 + next 4 // Shuffle table for converting ARGBToRGB24 for I422ToRGB24. First 8 + next 4
static const uvec8 kShuffleMaskARGBToRGB24_0 = { static const uvec8 kShuffleMaskARGBToRGB24_0 = {
0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u 0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u
}; };