coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-13 15:09:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

Change 24 bit conversions to use single asm block instead of 2, but with memory counter

Browse Source 
BUG=389,378
TESTED=out\release\libyuv_unittest --gtest_catch_exceptions=0 --gtest_filter=*420ToRGB24_Opt | sortms
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/29359004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1214 16f28f9a-4ce2-e073-06de-1de4eb20be90
This commit is contained in:
fbarchard@google.com 2014-12-30 21:35:35 +00:00
parent 5304aaaed1
commit 16338ba85f
3 changed files with 19 additions and 34 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: 1211
Version: 1214
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 1211
#define LIBYUV_VERSION 1214
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT

49
source/row_posix.cc
View File

@ -1683,25 +1683,15 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf,
);
}
// TODO(fbarchard): Consider putting masks into constants.
void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* dst_rgb24,
int width) {
// fpic 32 bit gcc 4.2 on OSX runs out of GPR regs.
#if defined(__i386__)
asm volatile (
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
:: [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
[kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24));
#endif
asm volatile (
#if !defined(__i386__)
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
#endif
"sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n"
LABELALIGN
@ -1719,18 +1709,21 @@ void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
"movq %%xmm0," MEMACCESS([dst_rgb24]) "\n"
"movdqu %%xmm1," MEMACCESS2(0x8,[dst_rgb24]) "\n"
"lea " MEMLEA(0x18,[dst_rgb24]) ",%[dst_rgb24] \n"
"sub $0x8,%[width] \n"
"subl $0x8,%[width] \n"
"jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf]
[u_buf]"+r"(u_buf), // %[u_buf]
[v_buf]"+r"(v_buf), // %[v_buf]
[dst_rgb24]"+r"(dst_rgb24), // %[dst_rgb24]
// TODO(fbarchard): Make width a register for 32 bit.
#if defined(__APPLE__) && defined(__i386__))
[width]"+m"(width) // %[width]
#else
[width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#if !defined(__i386__)
, [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
[kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24)
#endif
: [kYuvConstants]"r"(&kYuvConstants.kUVToB),
[kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
[kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24)
: "memory", "cc"
#if defined(__native_client__) && defined(__x86_64__)
, "r14"
@ -1746,20 +1739,9 @@ void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf,
const uint8* v_buf,
uint8* dst_raw,
int width) {
// fpic 32 bit gcc 4.2 on OSX runs out of GPR regs.
#if defined(__i386__)
asm volatile (
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW],%%xmm6 \n"
:: [kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0),
[kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW));
#endif
asm volatile (
#if !defined(__i386__)
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW],%%xmm6 \n"
#endif
"sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n"
LABELALIGN
@ -1777,18 +1759,21 @@ void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf,
"movq %%xmm0," MEMACCESS([dst_raw]) " \n"
"movdqu %%xmm1," MEMACCESS2(0x8,[dst_raw]) "\n"
"lea " MEMLEA(0x18,[dst_raw]) ",%[dst_raw] \n"
"sub $0x8,%[width] \n"
"subl $0x8,%[width] \n"
"jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf]
[u_buf]"+r"(u_buf), // %[u_buf]
[v_buf]"+r"(v_buf), // %[v_buf]
[dst_raw]"+r"(dst_raw), // %[dst_raw]
// TODO(fbarchard): Make width a register for 32 bit.
#if defined(__APPLE__) && defined(__i386__))
[width]"+m"(width) // %[width]
#else
[width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#if !defined(__i386__)
, [kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0),
[kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW)
#endif
: [kYuvConstants]"r"(&kYuvConstants.kUVToB),
[kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0),
[kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW)
: "memory", "cc"
#if defined(__native_client__) && defined(__x86_64__)
, "r14"