coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-08 01:36:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

ARGBToYRow_AVX2 and ARGBToYJRow_AVX2 ported to GCC.

Browse Source 
BUG=269
TESTED=try bots
R=brucedawson@google.com, tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/30299004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1184 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2014-12-02 22:00:08 +00:00
parent d0bfd10147
commit b036cf700b
4 changed files with 96 additions and 5 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
URL: http://code.google.com/p/libyuv/
Version: 1183
Version: 1184
License: BSD
License File: LICENSE

4
include/libyuv/row.h
View File

@ -203,6 +203,8 @@ extern "C" {
#define HAS_MERGEUVROW_AVX2
#define HAS_MIRRORROW_AVX2
#define HAS_ARGBMIRRORROW_AVX2
#define HAS_ARGBTOYROW_AVX2
#define HAS_ARGBTOYJROW_AVX2
// Effects:
#define HAS_ARGBADDROW_AVX2
@ -216,8 +218,6 @@ extern "C" {
// TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined(VISUALC_HAS_AVX2)
#define HAS_ARGBTOUVROW_AVX2
#define HAS_ARGBTOYJROW_AVX2
#define HAS_ARGBTOYROW_AVX2
#define HAS_INTERPOLATEROW_AVX2
#endif // defined(VISUALC_HAS_AVX2)

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1183
#define LIBYUV_VERSION 1184
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

93
source/row_posix.cc
View File

@ -706,8 +706,8 @@ void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int pix) {
#ifdef HAS_ARGBTOYROW_SSSE3
void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) {
asm volatile (
"movdqa %4,%%xmm5 \n"
"movdqa %3,%%xmm4 \n"
"movdqa %4,%%xmm5 \n"
LABELALIGN
"1: \n"
"movdqu " MEMACCESS(0) ",%%xmm0 \n"
@ -782,6 +782,97 @@ void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) {
}
#endif // HAS_ARGBTOYJROW_SSSE3
#ifdef HAS_ARGBTOYROW_AVX2
// vpermd for vphaddw + vpackuswb vpermd.
static const lvec32 kPermdARGBToY_AVX = {
0, 4, 1, 5, 2, 6, 3, 7
};
void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) {
asm volatile (
"vbroadcastf128 %3,%%ymm4 \n"
"vbroadcastf128 %4,%%ymm5 \n"
"vmovdqu %5,%%ymm6 \n"
LABELALIGN
"1: \n"
"vmovdqu " MEMACCESS(0) ",%%ymm0 \n"
"vmovdqu " MEMACCESS2(0x10,0) ",%%ymm1 \n"
"vmovdqu " MEMACCESS2(0x20,0) ",%%ymm2 \n"
"vmovdqu " MEMACCESS2(0x30,0) ",%%ymm3 \n"
"vpmaddubsw %%ymm4,%%ymm0,%%ymm0 \n"
"vpmaddubsw %%ymm4,%%ymm1,%%ymm1 \n"
"vpmaddubsw %%ymm4,%%ymm2,%%ymm2 \n"
"vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n"
"lea " MEMLEA(0x80,0) ",%0 \n"
"vphaddw %%ymm1,%%ymm0,%%ymm0 \n"
"vphaddw %%ymm3,%%ymm2,%%ymm2 \n"
"vpsrlw $0x7,%%ymm0,%%ymm0 \n"
"vpsrlw $0x7,%%ymm2,%%ymm2 \n"
"vpackuswb %%ymm2,%%ymm0,%%ymm0 \n"
"vpermd %%ymm0,%%ymm6,%%ymm0 \n"
"vpaddb %%ymm5,%%ymm0,%%ymm0 \n"
"vmovdqu %%ymm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x20,1) ",%1 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_argb), // %0
"+r"(dst_y), // %1
"+r"(pix) // %2
: "m"(kARGBToY), // %3
"m"(kAddY16), // %4
"m"(kPermdARGBToY_AVX) // %5
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
#endif
);
}
#endif // HAS_ARGBTOYROW_AVX2
#ifdef HAS_ARGBTOYJROW_AVX2
void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) {
asm volatile (
"vbroadcastf128 %3,%%ymm4 \n"
"vbroadcastf128 %4,%%ymm5 \n"
"vmovdqu %5,%%ymm6 \n"
LABELALIGN
"1: \n"
"vmovdqu " MEMACCESS(0) ",%%ymm0 \n"
"vmovdqu " MEMACCESS2(0x10,0) ",%%ymm1 \n"
"vmovdqu " MEMACCESS2(0x20,0) ",%%ymm2 \n"
"vmovdqu " MEMACCESS2(0x30,0) ",%%ymm3 \n"
"vpmaddubsw %%ymm4,%%ymm0,%%ymm0 \n"
"vpmaddubsw %%ymm4,%%ymm1,%%ymm1 \n"
"vpmaddubsw %%ymm4,%%ymm2,%%ymm2 \n"
"vpmaddubsw %%ymm4,%%ymm3,%%ymm3 \n"
"lea " MEMLEA(0x80,0) ",%0 \n"
"vphaddw %%ymm1,%%ymm0,%%ymm0 \n"
"vphaddw %%ymm3,%%ymm2,%%ymm2 \n"
"vpsrlw $0x7,%%ymm0,%%ymm0 \n"
"vpsrlw $0x7,%%ymm2,%%ymm2 \n"
"vpackuswb %%ymm2,%%ymm0,%%ymm0 \n"
"vpermd %%ymm0,%%ymm6,%%ymm0 \n"
"vpaddb %%ymm5,%%ymm0,%%ymm0 \n"
"vmovdqu %%ymm0," MEMACCESS(1) " \n"
"lea " MEMLEA(0x20,1) ",%1 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src_argb), // %0
"+r"(dst_y), // %1
"+r"(pix) // %2
: "m"(kARGBToYJ), // %3
"m"(kAddY16), // %4
"m"(kPermdARGBToY_AVX) // %5
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
#endif
);
}
#endif // HAS_ARGBTOYJROW_AVX2
#ifdef HAS_ARGBTOUVROW_SSSE3
// TODO(fbarchard): pass xmm constants to single block of assembly.
// fpic on GCC 4.2 for OSX runs out of GPR registers. "m" effectively takes