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

Use vmovd to avoid switch to sse mode

Browse Source 
BUG=none
TEST=c:\intelsde\sde -hsw -- out\release\libyuv_unittest.exe --gtest_filter=*Psnr*
Review URL: https://webrtc-codereview.appspot.com/1097013

git-svn-id: http://libyuv.googlecode.com/svn/trunk@573 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2013-02-14 19:03:30 +00:00
parent f3ad618d40
commit 408e574366
4 changed files with 25 additions and 11 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: 572
Version: 573
License: BSD
License File: LICENSE

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 572
#define LIBYUV_VERSION 573
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

15
source/compare.cc
View File

@ -102,7 +102,9 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
}
#endif
#if defined(HAS_SUMSQUAREERROR_AVX2)
bool clear = false;
if (TestCpuFlag(kCpuHasAVX2)) {
clear = true;
// Note only used for multiples of 32 so count is not checked.
SumSquareError = SumSquareError_AVX2;
}
@ -130,6 +132,12 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
if (remainder) {
sse += SumSquareError_C(src_a, src_b, remainder);
}
#if defined(HAS_SUMSQUAREERROR_AVX2)
if (clear) {
__asm vzeroupper;
}
#endif
return sse;
}
@ -157,7 +165,9 @@ uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
}
#endif
#if defined(HAS_SUMSQUAREERROR_AVX2)
bool clear = false;
if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 32)) {
clear = true;
SumSquareError = SumSquareError_AVX2;
}
#endif
@ -168,6 +178,11 @@ uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
src_b += stride_b;
}
#if defined(HAS_SUMSQUAREERROR_AVX2)
if (clear) {
__asm vzeroupper;
}
#endif
return sse;
}

17
source/compare_win.cc
View File

@ -47,9 +47,9 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
paddd xmm0, xmm2
jg wloop
pshufd xmm1, xmm0, 0EEh
pshufd xmm1, xmm0, 0xee
paddd xmm0, xmm1
pshufd xmm1, xmm0, 01h
pshufd xmm1, xmm0, 0x01
paddd xmm0, xmm1
movd eax, xmm0
ret
@ -67,7 +67,7 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
vpxor ymm0, ymm0, ymm0 // sum
vpxor ymm5, ymm5, ymm5 // for unpack.
vpxor ymm5, ymm5, ymm5 // constant 0 for unpck
sub edx, eax
align 16
@ -92,9 +92,8 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
vpshufd ymm1, ymm0, 0x01 // 1 + 0 both lanes.
vpaddd ymm0, ymm0, ymm1
vpermq ymm1, ymm0, 0x02 // high + low lane.
vpaddd ymm4, ymm0, ymm1
vzeroupper // TODO(fbarchard): Remove.
movd eax, xmm4
vpaddd ymm0, ymm0, ymm1
vmovd eax, xmm0
ret
}
}
@ -173,14 +172,14 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
sub ecx, 16
paddd xmm1, xmm3
pshufd xmm2, xmm1, 14 // upper 2 dwords
pshufd xmm2, xmm1, 0x0e // upper 2 dwords
paddd xmm1, xmm2
pshufd xmm2, xmm1, 1
pshufd xmm2, xmm1, 0x01
paddd xmm1, xmm2
paddd xmm0, xmm1
jg wloop
movd eax, xmm0 // return hash
movd eax, xmm0 // return hash
ret
}
}