coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-02-08 02:36:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

switch win64 intrinsics to loadu / storeu for unaligned memory.

Browse Source 
BUG=372
TESTED=untested
R=brucedawson@google.com, harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30729004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1124 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2014-10-16 23:46:48 +00:00
parent e737688603
commit 3dbaaf0032
3 changed files with 7 additions and 7 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: 1123 Version: 1124
License: BSD License: BSD
License File: LICENSE License File: LICENSE

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 1123 #define LIBYUV_VERSION 1124
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

10
source/row_win.cc
View File

@ -89,8 +89,8 @@ void I422ToARGBRow_SSSE3(const uint8* y_buf,
xmm1 = _mm_cvtsi32_si128(*(uint32*)(u_buf + offset)); xmm1 = _mm_cvtsi32_si128(*(uint32*)(u_buf + offset));
xmm0 = _mm_unpacklo_epi8(xmm0, xmm1); xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);
xmm0 = _mm_unpacklo_epi16(xmm0, xmm0); xmm0 = _mm_unpacklo_epi16(xmm0, xmm0);
xmm1 = _mm_load_si128(&xmm0); xmm1 = _mm_loadu_si128(&xmm0);
xmm2 = _mm_load_si128(&xmm0); xmm2 = _mm_loadu_si128(&xmm0);
xmm0 = _mm_maddubs_epi16(xmm0, *(__m128i*)kUVToB); xmm0 = _mm_maddubs_epi16(xmm0, *(__m128i*)kUVToB);
xmm1 = _mm_maddubs_epi16(xmm1, *(__m128i*)kUVToG); xmm1 = _mm_maddubs_epi16(xmm1, *(__m128i*)kUVToG);
xmm2 = _mm_maddubs_epi16(xmm2, *(__m128i*)kUVToR); xmm2 = _mm_maddubs_epi16(xmm2, *(__m128i*)kUVToR);
@ -112,12 +112,12 @@ void I422ToARGBRow_SSSE3(const uint8* y_buf,
xmm2 = _mm_packus_epi16(xmm2, xmm2); xmm2 = _mm_packus_epi16(xmm2, xmm2);
xmm0 = _mm_unpacklo_epi8(xmm0, xmm1); xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);
xmm2 = _mm_unpacklo_epi8(xmm2, xmm5); xmm2 = _mm_unpacklo_epi8(xmm2, xmm5);
xmm1 = _mm_load_si128(&xmm0); xmm1 = _mm_loadu_si128(&xmm0);
xmm0 = _mm_unpacklo_epi16(xmm0, xmm2); xmm0 = _mm_unpacklo_epi16(xmm0, xmm2);
xmm1 = _mm_unpackhi_epi16(xmm1, xmm2); xmm1 = _mm_unpackhi_epi16(xmm1, xmm2);
_mm_store_si128((__m128i *)dst_argb, xmm0); _mm_storeu_si128((__m128i *)dst_argb, xmm0);
_mm_store_si128((__m128i *)(dst_argb + 16), xmm1); _mm_storeu_si128((__m128i *)(dst_argb + 16), xmm1);
y_buf += 8; y_buf += 8;
u_buf += 4; u_buf += 4;