coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Reenable AVX2 scaling with bug fix for any width

Browse Source 
BUG=376
TESTED=unittest on scale functions
R=brucedawson@google.com, harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30759004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1135 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2014-10-22 01:15:20 +00:00
parent 4165437c3e
commit af6f25245e
5 changed files with 51 additions and 53 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: 1134 Version: 1135
License: BSD License: BSD
License File: LICENSE License File: LICENSE

3
include/libyuv/row.h
View File

@ -201,8 +201,7 @@ extern "C" {
#define HAS_ARGBTOYJROW_AVX2 #define HAS_ARGBTOYJROW_AVX2
#define HAS_ARGBTOYROW_AVX2 #define HAS_ARGBTOYROW_AVX2
#define HAS_I422TOARGBROW_AVX2 #define HAS_I422TOARGBROW_AVX2
// TODO(fbarchard): fix bug #376. #define HAS_INTERPOLATEROW_AVX2
// #define HAS_INTERPOLATEROW_AVX2
#define HAS_MERGEUVROW_AVX2 #define HAS_MERGEUVROW_AVX2
#define HAS_MIRRORROW_AVX2 #define HAS_MIRRORROW_AVX2
#define HAS_SPLITUVROW_AVX2 #define HAS_SPLITUVROW_AVX2

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 1134 #define LIBYUV_VERSION 1135
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

16
source/row_any.cc
View File

@ -581,24 +581,20 @@ YANY(ARGBShuffleRow_Any_NEON, ARGBShuffleRow_NEON,
} }
#ifdef HAS_INTERPOLATEROW_AVX2 #ifdef HAS_INTERPOLATEROW_AVX2
NANY(InterpolateRow_Any_AVX2, InterpolateRow_AVX2, NANY(InterpolateRow_Any_AVX2, InterpolateRow_AVX2, InterpolateRow_C, 1, 1, 31)
InterpolateRow_C, 1, 1, 32)
#endif #endif
#ifdef HAS_INTERPOLATEROW_SSSE3 #ifdef HAS_INTERPOLATEROW_SSSE3
NANY(InterpolateRow_Any_SSSE3, InterpolateRow_SSSE3, NANY(InterpolateRow_Any_SSSE3, InterpolateRow_SSSE3, InterpolateRow_C, 1, 1, 15)
InterpolateRow_C, 1, 1, 15)
#endif #endif
#ifdef HAS_INTERPOLATEROW_SSE2 #ifdef HAS_INTERPOLATEROW_SSE2
NANY(InterpolateRow_Any_SSE2, InterpolateRow_SSE2, NANY(InterpolateRow_Any_SSE2, InterpolateRow_SSE2, InterpolateRow_C, 1, 1, 15)
InterpolateRow_C, 1, 1, 15)
#endif #endif
#ifdef HAS_INTERPOLATEROW_NEON #ifdef HAS_INTERPOLATEROW_NEON
NANY(InterpolateRow_Any_NEON, InterpolateRow_NEON, NANY(InterpolateRow_Any_NEON, InterpolateRow_NEON, InterpolateRow_C, 1, 1, 15)
InterpolateRow_C, 1, 1, 15)
#endif #endif
#ifdef HAS_INTERPOLATEROW_MIPS_DSPR2 #ifdef HAS_INTERPOLATEROW_MIPS_DSPR2
NANY(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2, NANY(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2, InterpolateRow_C,
InterpolateRow_C, 1, 1, 3) 1, 1, 3)
#endif #endif
#undef NANY #undef NANY

81
source/row_win.cc
View File

@ -4972,11 +4972,11 @@ void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride,
#endif // HAS_ARGBAFFINEROW_SSE2 #endif // HAS_ARGBAFFINEROW_SSE2
#ifdef HAS_INTERPOLATEROW_AVX2 #ifdef HAS_INTERPOLATEROW_AVX2
// Bilinear filter 16x2 -> 16x1 // Bilinear filter 32x2 -> 32x1
__declspec(naked) __declspec(align(16)) __declspec(naked) __declspec(align(16))
void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr, void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width, ptrdiff_t src_stride, int dst_width,
int source_y_fraction) { int source_y_fraction) {
__asm { __asm {
push esi push esi
push edi push edi
@ -5023,45 +5023,48 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr,
jg xloop jg xloop
jmp xloop99 jmp xloop99
// Blend 25 / 75. // Blend 25 / 75.
align 4 align 4
xloop25: xloop25:
vmovdqu ymm0, [esi] vmovdqu ymm0, [esi]
vpavgb ymm0, ymm0, [esi + edx] vmovdqu ymm1, [esi + edx]
vpavgb ymm0, ymm0, [esi + edx] vpavgb ymm0, ymm0, ymm1
sub ecx, 32 vpavgb ymm0, ymm0, ymm1
vmovdqu [esi + edi], ymm0 sub ecx, 32
lea esi, [esi + 32] vmovdqu [esi + edi], ymm0
jg xloop25 lea esi, [esi + 32]
jmp xloop99 jg xloop25
jmp xloop99
// Blend 50 / 50. // Blend 50 / 50.
align 4 align 4
xloop50: xloop50:
vmovdqu ymm0, [esi] vmovdqu ymm0, [esi]
vpavgb ymm0, ymm0, [esi + edx] vmovdqu ymm1, [esi + edx]
sub ecx, 32 vpavgb ymm0, ymm0, ymm1
vmovdqu [esi + edi], ymm0 sub ecx, 32
lea esi, [esi + 32] vmovdqu [esi + edi], ymm0
jg xloop50 lea esi, [esi + 32]
jmp xloop99 jg xloop50
jmp xloop99
// Blend 75 / 25. // Blend 75 / 25.
align 4 align 4
xloop75: xloop75:
vmovdqu ymm0, [esi + edx] vmovdqu ymm1, [esi]
vpavgb ymm0, ymm0, [esi] vmovdqu ymm0, [esi + edx]
vpavgb ymm0, ymm0, [esi] vpavgb ymm0, ymm0, ymm1
sub ecx, 32 vpavgb ymm0, ymm0, ymm1
vmovdqu [esi + edi], ymm0 sub ecx, 32
lea esi, [esi + 32] vmovdqu [esi + edi], ymm0
jg xloop75 lea esi, [esi + 32]
jmp xloop99 jg xloop75
jmp xloop99
// Blend 100 / 0 - Copy row unchanged. // Blend 100 / 0 - Copy row unchanged.
align 4 align 4
xloop100: xloop100:
rep movsb rep movsb
xloop99: xloop99:
pop edi pop edi