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

Switch to macro for STOREBGRA etc on Posix SSSE3

Browse Source 
BUG=393
TESTED=try bots
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/33339004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1282 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2015-02-13 21:53:16 +00:00
parent 796396b226
commit 8e16c1a341
1 changed files with 38 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
source/row_posix.cc
View File

@ -1537,6 +1537,41 @@ static YuvConstants SIMD_ALIGNED(kYvuConstants) = {
"movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) " \n" \ "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) " \n" \
"lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n"
// Store 8 BGRA values. Assumes XMM5 is zero.
#define STOREBGRA \
"pcmpeqb %%xmm5,%%xmm5 \n" \
"punpcklbw %%xmm0,%%xmm1 \n" \
"punpcklbw %%xmm2,%%xmm5 \n" \
"movdqa %%xmm5,%%xmm0 \n" \
"punpcklwd %%xmm1,%%xmm5 \n" \
"punpckhwd %%xmm1,%%xmm0 \n" \
"movdqu %%xmm5," MEMACCESS([dst_bgra]) " \n" \
"movdqu %%xmm0," MEMACCESS2(0x10,[dst_bgra]) " \n" \
"lea " MEMLEA(0x20,[dst_bgra]) ",%[dst_bgra] \n"
// Store 8 ABGR values. Assumes XMM5 is zero.
#define STOREABGR \
"punpcklbw %%xmm1,%%xmm2 \n" \
"punpcklbw %%xmm5,%%xmm0 \n" \
"movdqa %%xmm2,%%xmm1 \n" \
"punpcklwd %%xmm0,%%xmm2 \n" \
"punpckhwd %%xmm0,%%xmm1 \n" \
"movdqu %%xmm2," MEMACCESS([dst_abgr]) " \n" \
"movdqu %%xmm1," MEMACCESS2(0x10,[dst_abgr]) " \n" \
"lea " MEMLEA(0x20,[dst_abgr]) ",%[dst_abgr] \n"
// Store 8 RGBA values. Assumes XMM5 is zero.
#define STORERGBA \
"pcmpeqb %%xmm5,%%xmm5 \n" \
"punpcklbw %%xmm2,%%xmm1 \n" \
"punpcklbw %%xmm0,%%xmm5 \n" \
"movdqa %%xmm5,%%xmm0 \n" \
"punpcklwd %%xmm1,%%xmm5 \n" \
"punpckhwd %%xmm1,%%xmm0 \n" \
"movdqu %%xmm5," MEMACCESS([dst_rgba]) " \n" \
"movdqu %%xmm0," MEMACCESS2(0x10,[dst_rgba]) " \n" \
"lea " MEMLEA(0x20,[dst_rgba]) ",%[dst_rgba] \n"
void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf, void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf,
const uint8* u_buf, const uint8* u_buf,
const uint8* v_buf, const uint8* v_buf,
@ -1762,15 +1797,7 @@ void OMITFP I422ToBGRARow_SSSE3(const uint8* y_buf,
"1: \n" "1: \n"
READYUV422 READYUV422
YUVTORGB(kYuvConstants) YUVTORGB(kYuvConstants)
"pcmpeqb %%xmm5,%%xmm5 \n" STOREBGRA
"punpcklbw %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm2,%%xmm5 \n"
"movdqa %%xmm5,%%xmm0 \n"
"punpcklwd %%xmm1,%%xmm5 \n"
"punpckhwd %%xmm1,%%xmm0 \n"
"movdqu %%xmm5," MEMACCESS([dst_bgra]) "\n"
"movdqu %%xmm0," MEMACCESS2(0x10,[dst_bgra]) "\n"
"lea " MEMLEA(0x20,[dst_bgra]) ",%[dst_bgra] \n"
"sub $0x8,%[width] \n" "sub $0x8,%[width] \n"
"jg 1b \n" "jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf] : [y_buf]"+r"(y_buf), // %[y_buf]
@ -1796,14 +1823,7 @@ void OMITFP I422ToABGRRow_SSSE3(const uint8* y_buf,
"1: \n" "1: \n"
READYUV422 READYUV422
YUVTORGB(kYuvConstants) YUVTORGB(kYuvConstants)
"punpcklbw %%xmm1,%%xmm2 \n" STOREABGR
"punpcklbw %%xmm5,%%xmm0 \n"
"movdqa %%xmm2,%%xmm1 \n"
"punpcklwd %%xmm0,%%xmm2 \n"
"punpckhwd %%xmm0,%%xmm1 \n"
"movdqu %%xmm2," MEMACCESS([dst_abgr]) "\n"
"movdqu %%xmm1," MEMACCESS2(0x10,[dst_abgr]) "\n"
"lea " MEMLEA(0x20,[dst_abgr]) ",%[dst_abgr] \n"
"sub $0x8,%[width] \n" "sub $0x8,%[width] \n"
"jg 1b \n" "jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf] : [y_buf]"+r"(y_buf), // %[y_buf]
@ -1829,15 +1849,7 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf,
"1: \n" "1: \n"
READYUV422 READYUV422
YUVTORGB(kYuvConstants) YUVTORGB(kYuvConstants)
"pcmpeqb %%xmm5,%%xmm5 \n" STORERGBA
"punpcklbw %%xmm2,%%xmm1 \n"
"punpcklbw %%xmm0,%%xmm5 \n"
"movdqa %%xmm5,%%xmm0 \n"
"punpcklwd %%xmm1,%%xmm5 \n"
"punpckhwd %%xmm1,%%xmm0 \n"
"movdqu %%xmm5," MEMACCESS([dst_rgba]) "\n"
"movdqu %%xmm0," MEMACCESS2(0x10,[dst_rgba]) "\n"
"lea " MEMLEA(0x20,[dst_rgba]) ",%[dst_rgba] \n"
"sub $0x8,%[width] \n" "sub $0x8,%[width] \n"
"jg 1b \n" "jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf] : [y_buf]"+r"(y_buf), // %[y_buf]