coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-02-16 23:29:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

ARGBShuffle_SSE2 ported to GCC and NaCL, and HalfRow_SSE2 ported to NaCL.

Browse Source 
BUG=271
TESTED=ABGRToARGB on linux
R=johannkoenig@google.com, nfullagar@google.com

Review URL: https://webrtc-codereview.appspot.com/2362004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@808 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2013-10-08 00:15:34 +00:00
parent 212a1a5000
commit 8b0cdb4a6e
5 changed files with 156 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: 807 Version: 808
License: BSD License: BSD
License File: LICENSE License File: LICENSE

17
include/libyuv/row.h
View File

@ -64,24 +64,26 @@ extern "C" {
#define HAS_SOBELYROW_SSSE3 #define HAS_SOBELYROW_SSSE3
// Conversions: // Conversions:
#define HAS_ABGRTOYROW_SSSE3
#define HAS_ARGBTOUV444ROW_SSSE3
#define HAS_ARGBTOUVJROW_SSSE3
#define HAS_ABGRTOUVROW_SSSE3 #define HAS_ABGRTOUVROW_SSSE3
#define HAS_ABGRTOYROW_SSSE3
#define HAS_ARGBSHUFFLEROW_SSE2
#define HAS_ARGBSHUFFLEROW_SSSE3 #define HAS_ARGBSHUFFLEROW_SSSE3
#define HAS_ARGBTOBAYERROW_SSSE3 #define HAS_ARGBTOBAYERROW_SSSE3
#define HAS_ARGBTOUV422ROW_SSSE3
#define HAS_ARGBTOUV444ROW_SSSE3
#define HAS_ARGBTOUVJROW_SSSE3
#define HAS_ARGBTOYJROW_SSSE3 #define HAS_ARGBTOYJROW_SSSE3
#define HAS_ARGBTOYROW_SSSE3 #define HAS_ARGBTOYROW_SSSE3
#define HAS_ARGBTOUV422ROW_SSSE3
#define HAS_BGRATOYROW_SSSE3
#define HAS_BGRATOUVROW_SSSE3 #define HAS_BGRATOUVROW_SSSE3
#define HAS_BGRATOYROW_SSSE3
#define HAS_COPYROW_ERMS #define HAS_COPYROW_ERMS
#define HAS_COPYROW_SSE2 #define HAS_COPYROW_SSE2
#define HAS_COPYROW_X86 #define HAS_COPYROW_X86
#define HAS_FIXEDDIV_X86 #define HAS_FIXEDDIV_X86
#define HAS_HALFROW_SSE2
#define HAS_I400TOARGBROW_SSE2 #define HAS_I400TOARGBROW_SSE2
#define HAS_RGBATOYROW_SSSE3
#define HAS_RGBATOUVROW_SSSE3 #define HAS_RGBATOUVROW_SSSE3
#define HAS_RGBATOYROW_SSSE3
#define HAS_SETROW_X86 #define HAS_SETROW_X86
#define HAS_UYVYTOUVROW_SSE2 #define HAS_UYVYTOUVROW_SSE2
#define HAS_UYVYTOYROW_SSE2 #define HAS_UYVYTOYROW_SSE2
@ -103,7 +105,6 @@ extern "C" {
#define HAS_ARGBTORAWROW_SSSE3 #define HAS_ARGBTORAWROW_SSSE3
#define HAS_ARGBTORGB24ROW_SSSE3 #define HAS_ARGBTORGB24ROW_SSSE3
#define HAS_ARGBTORGB565ROW_SSE2 #define HAS_ARGBTORGB565ROW_SSE2
#define HAS_HALFROW_SSE2
#define HAS_I411TOARGBROW_SSSE3 #define HAS_I411TOARGBROW_SSSE3
#define HAS_I422TOABGRROW_SSSE3 #define HAS_I422TOABGRROW_SSSE3
#define HAS_I422TOARGB1555ROW_SSSE3 #define HAS_I422TOARGB1555ROW_SSSE3
@ -163,8 +164,6 @@ extern "C" {
// The following are Windows only: // The following are Windows only:
// TODO(fbarchard): Port to gcc. // TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) #if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
#define HAS_ARGBSHUFFLEROW_SSE2
// Effects: // Effects:
// TODO(fbarchard): Optimize and enable // TODO(fbarchard): Optimize and enable
// #define HAS_ARGBLUMACOLORTABLEROW_SSSE3 // #define HAS_ARGBLUMACOLORTABLEROW_SSSE3

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 807 #define LIBYUV_VERSION 808
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

143
source/row_posix.cc
View File

@ -40,6 +40,9 @@ extern "C" {
#define MEMOPMEM(opcode, reg, offset, base, index, scale) \ #define MEMOPMEM(opcode, reg, offset, base, index, scale) \
"lea " #offset "(%q" #base ",%q" #index "," #scale "),%%r14d\n" \ "lea " #offset "(%q" #base ",%q" #index "," #scale "),%%r14d\n" \
#opcode " %%" #reg ",(%%r15,%%r14)\n" #opcode " %%" #reg ",(%%r15,%%r14)\n"
#define MEMOP(opcode, offset, base, index, scale) \
"lea " #offset "(%q" #base ",%q" #index "," #scale "),%%r14d\n" \
#opcode " (%%r15,%%r14)"
#define BUNDLEALIGN ".p2align 5 \n" #define BUNDLEALIGN ".p2align 5 \n"
#else #else
#define MEMACCESS(base) "(%" #base ")" #define MEMACCESS(base) "(%" #base ")"
@ -53,6 +56,8 @@ extern "C" {
#opcode " " #offset "(%" #base ",%" #index "," #scale "),%%" #reg "\n" #opcode " " #offset "(%" #base ",%" #index "," #scale "),%%" #reg "\n"
#define MEMOPMEM(opcode, reg, offset, base, index, scale) \ #define MEMOPMEM(opcode, reg, offset, base, index, scale) \
#opcode " %%" #reg ","#offset "(%" #base ",%" #index "," #scale ")\n" #opcode " %%" #reg ","#offset "(%" #base ",%" #index "," #scale ")\n"
#define MEMOP(opcode, offset, base, index, scale) \
#opcode " " #offset "(%" #base ",%" #index "," #scale ")"
#define BUNDLEALIGN #define BUNDLEALIGN
#endif #endif
@ -5625,11 +5630,11 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
"sub %0,%1 \n" "sub %0,%1 \n"
".p2align 4 \n" ".p2align 4 \n"
"1: \n" "1: \n"
"movdqa (%0),%%xmm0 \n" "movdqa "MEMACCESS(0)",%%xmm0 \n"
"pavgb (%0,%3),%%xmm0 \n" MEMOPREG(pavgb,0x00,0,3,1,xmm0) // pavgb (%0,%3),%%xmm0
"sub $0x10,%2 \n" "sub $0x10,%2 \n"
"movdqa %%xmm0,(%0,%1) \n" MEMOPMEM(movdqa,xmm0,0x00,0,1,1) // movdqa %%xmm0,(%0,%1)
"lea 0x10(%0),%0 \n" "lea "MEMLEA(0x10,0)",%0 \n"
"jg 1b \n" "jg 1b \n"
: "+r"(src_uv), // %0 : "+r"(src_uv), // %0
"+r"(dst_uv), // %1 "+r"(dst_uv), // %1
@ -5736,9 +5741,7 @@ void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb,
void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) { const uint8* shuffler, int pix) {
asm volatile ( asm volatile (
"vmovdqa "MEMACCESS(3)",%%xmm5 \n" "vbroadcastf128 "MEMACCESS(3)",%%ymm5 \n"
"vpermq $0x44,%%ymm5,%%ymm5 \n"
".p2align 4 \n" ".p2align 4 \n"
"1: \n" "1: \n"
"vmovdqu "MEMACCESS(0)",%%ymm0 \n" "vmovdqu "MEMACCESS(0)",%%ymm0 \n"
@ -5763,6 +5766,132 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
} }
#endif // HAS_ARGBSHUFFLEROW_AVX2 #endif // HAS_ARGBSHUFFLEROW_AVX2
#ifdef HAS_ARGBSHUFFLEROW_SSE2
// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
const uint8* shuffler, int pix) {
uintptr_t pixel_temp = 0u;
asm volatile (
"pxor %%xmm5,%%xmm5 \n"
"mov "MEMACCESS(4)",%k2 \n"
"cmp $0x3000102,%k2 \n"
"je 3012f \n"
"cmp $0x10203,%k2 \n"
"je 123f \n"
"cmp $0x30201,%k2 \n"
"je 321f \n"
"cmp $0x2010003,%k2 \n"
"je 2103f \n"
"1: \n"
BUNDLEALIGN
"movzb "MEMACCESS(4)",%2 \n"
MEMOP(movzb,0x00,0,2,1)",%2 \n" // movzb (%0,%2,1),%2
"mov %b2,"MEMACCESS(1)" \n"
"movzb "MEMACCESS2(0x1,4)",%2 \n"
MEMOP(movzb,0x00,0,2,1)",%2 \n" // movzb (%0,%2,1),%2
"mov %b2,"MEMACCESS2(0x1,1)" \n"
BUNDLEALIGN
"movzb "MEMACCESS2(0x2,4)",%2 \n"
MEMOP(movzb,0x00,0,2,1)",%2 \n" // movzb (%0,%2,1),%2
"mov %b2,"MEMACCESS2(0x2,1)" \n"
"movzb "MEMACCESS2(0x3,4)",%2 \n"
MEMOP(movzb,0x00,0,2,1)",%2 \n" // movzb (%0,%2,1),%2
"mov %b2,"MEMACCESS2(0x3,1)" \n"
"lea "MEMLEA(0x4,0)",%0 \n"
"lea "MEMLEA(0x4,1)",%1 \n"
"sub $0x1,%3 \n"
"jg 1b \n"
"jmp 99f \n"
".p2align 4 \n"
"123: \n"
"movdqu "MEMACCESS(0)",%%xmm0 \n"
"lea "MEMLEA(0x10,0)",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x1b,%%xmm0,%%xmm0 \n"
"pshuflw $0x1b,%%xmm0,%%xmm0 \n"
"pshufhw $0x1b,%%xmm1,%%xmm1 \n"
"pshuflw $0x1b,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"sub $0x4,%3 \n"
"movdqu %%xmm0,"MEMACCESS(1)" \n"
"lea "MEMLEA(0x10,1)",%1 \n"
"jg 123b \n"
"jmp 99f \n"
".p2align 4 \n"
"321: \n"
"movdqu "MEMACCESS(0)",%%xmm0 \n"
"lea "MEMLEA(0x10,0)",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x39,%%xmm0,%%xmm0 \n"
"pshuflw $0x39,%%xmm0,%%xmm0 \n"
"pshufhw $0x39,%%xmm1,%%xmm1 \n"
"pshuflw $0x39,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"sub $0x4,%3 \n"
"movdqu %%xmm0,"MEMACCESS(1)" \n"
"lea "MEMLEA(0x10,1)",%1 \n"
"jg 321b \n"
"jmp 99f \n"
".p2align 4 \n"
"2103: \n"
"movdqu "MEMACCESS(0)",%%xmm0 \n"
"lea "MEMLEA(0x10,0)",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0x93,%%xmm0,%%xmm0 \n"
"pshuflw $0x93,%%xmm0,%%xmm0 \n"
"pshufhw $0x93,%%xmm1,%%xmm1 \n"
"pshuflw $0x93,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"sub $0x4,%3 \n"
"movdqu %%xmm0,"MEMACCESS(1)" \n"
"lea "MEMLEA(0x10,1)",%1 \n"
"jg 2103b \n"
"jmp 99f \n"
".p2align 4 \n"
"3012: \n"
"movdqu "MEMACCESS(0)",%%xmm0 \n"
"lea "MEMLEA(0x10,0)",%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm5,%%xmm0 \n"
"punpckhbw %%xmm5,%%xmm1 \n"
"pshufhw $0xc6,%%xmm0,%%xmm0 \n"
"pshuflw $0xc6,%%xmm0,%%xmm0 \n"
"pshufhw $0xc6,%%xmm1,%%xmm1 \n"
"pshuflw $0xc6,%%xmm1,%%xmm1 \n"
"packuswb %%xmm1,%%xmm0 \n"
"sub $0x4,%3 \n"
"movdqu %%xmm0,"MEMACCESS(1)" \n"
"lea "MEMLEA(0x10,1)",%1 \n"
"jg 3012b \n"
"99: \n"
: "+r"(src_argb), // %0
"+r"(dst_argb), // %1
"+d"(pixel_temp), // %2
"+r"(pix) // %3
: "r"(shuffler) // %4
: "memory", "cc"
#if defined(__native_client__) && defined(__x86_64__)
, "r14"
#endif
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm5"
#endif
);
}
#endif // HAS_ARGBSHUFFLEROW_SSE2
#ifdef HAS_I422TOYUY2ROW_SSE2 #ifdef HAS_I422TOYUY2ROW_SSE2
void I422ToYUY2Row_SSE2(const uint8* src_y, void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u, const uint8* src_u,

20
source/row_win.cc
View File

@ -6627,7 +6627,7 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
ret ret
} }
} }
#endif #endif // HAS_ARGBSHUFFLEROW_AVX2
__declspec(naked) __declspec(align(16)) __declspec(naked) __declspec(align(16))
void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb, void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
@ -6639,7 +6639,7 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
mov edx, [esp + 8 + 8] // dst_argb mov edx, [esp + 8 + 8] // dst_argb
mov esi, [esp + 8 + 12] // shuffler mov esi, [esp + 8 + 12] // shuffler
mov ecx, [esp + 8 + 16] // pix mov ecx, [esp + 8 + 16] // pix
pxor xmm7, xmm7 pxor xmm5, xmm5
mov ebx, [esi] // shuffler mov ebx, [esi] // shuffler
cmp ebx, 0x03000102 cmp ebx, 0x03000102
@ -6676,8 +6676,8 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
movdqu xmm0, [eax] movdqu xmm0, [eax]
lea eax, [eax + 16] lea eax, [eax + 16]
movdqa xmm1, xmm0 movdqa xmm1, xmm0
punpcklbw xmm0, xmm7 punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm7 punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 01Bh // 1B = 00011011 = 0x0123 = BGRAToARGB pshufhw xmm0, xmm0, 01Bh // 1B = 00011011 = 0x0123 = BGRAToARGB
pshuflw xmm0, xmm0, 01Bh pshuflw xmm0, xmm0, 01Bh
pshufhw xmm1, xmm1, 01Bh pshufhw xmm1, xmm1, 01Bh
@ -6694,8 +6694,8 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
movdqu xmm0, [eax] movdqu xmm0, [eax]
lea eax, [eax + 16] lea eax, [eax + 16]
movdqa xmm1, xmm0 movdqa xmm1, xmm0
punpcklbw xmm0, xmm7 punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm7 punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 039h // 39 = 00111001 = 0x0321 = RGBAToARGB pshufhw xmm0, xmm0, 039h // 39 = 00111001 = 0x0321 = RGBAToARGB
pshuflw xmm0, xmm0, 039h pshuflw xmm0, xmm0, 039h
pshufhw xmm1, xmm1, 039h pshufhw xmm1, xmm1, 039h
@ -6712,8 +6712,8 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
movdqu xmm0, [eax] movdqu xmm0, [eax]
lea eax, [eax + 16] lea eax, [eax + 16]
movdqa xmm1, xmm0 movdqa xmm1, xmm0
punpcklbw xmm0, xmm7 punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm7 punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 093h // 93 = 10010011 = 0x2103 = ARGBToRGBA pshufhw xmm0, xmm0, 093h // 93 = 10010011 = 0x2103 = ARGBToRGBA
pshuflw xmm0, xmm0, 093h pshuflw xmm0, xmm0, 093h
pshufhw xmm1, xmm1, 093h pshufhw xmm1, xmm1, 093h
@ -6730,8 +6730,8 @@ void ARGBShuffleRow_SSE2(const uint8* src_argb, uint8* dst_argb,
movdqu xmm0, [eax] movdqu xmm0, [eax]
lea eax, [eax + 16] lea eax, [eax + 16]
movdqa xmm1, xmm0 movdqa xmm1, xmm0
punpcklbw xmm0, xmm7 punpcklbw xmm0, xmm5
punpckhbw xmm1, xmm7 punpckhbw xmm1, xmm5
pshufhw xmm0, xmm0, 0C6h // C6 = 11000110 = 0x3012 = ABGRToARGB pshufhw xmm0, xmm0, 0C6h // C6 = 11000110 = 0x3012 = ABGRToARGB
pshuflw xmm0, xmm0, 0C6h pshuflw xmm0, xmm0, 0C6h
pshufhw xmm1, xmm1, 0C6h pshufhw xmm1, xmm1, 0C6h