Commit 2f11976b authored by fbarchard@google.com's avatar fbarchard@google.com

munged addressing for copyrows

BUG=none
TEST=none
Review URL: http://webrtc-codereview.appspot.com/333001

git-svn-id: http://libyuv.googlecode.com/svn/trunk@109 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent e7ae4e10
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 108 Version: 109
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
...@@ -134,19 +134,19 @@ static void SplitUV_C(const uint8* src_uv, ...@@ -134,19 +134,19 @@ static void SplitUV_C(const uint8* src_uv,
#if defined(_M_IX86) && !defined(YUV_DISABLE_ASM) #if defined(_M_IX86) && !defined(YUV_DISABLE_ASM)
#define HAS_COPYROW_SSE2 #define HAS_COPYROW_SSE2
__declspec(naked) __declspec(naked)
void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
__asm { __asm {
mov eax, [esp + 4] // src mov eax, [esp + 4] // src
mov edx, [esp + 8] // dst mov edx, [esp + 8] // dst
mov ecx, [esp + 12] // count mov ecx, [esp + 12] // count
sub edx, eax
convertloop: convertloop:
movdqa xmm0, [eax] movdqa xmm0, [eax]
movdqa xmm1, [eax + 16] movdqa xmm1, [eax + 16]
movdqa [eax + edx], xmm0
movdqa [eax + edx + 16], xmm1
lea eax, [eax + 32] lea eax, [eax + 32]
movdqa [edx], xmm0
movdqa [edx + 16], xmm1
lea edx, [edx + 32]
sub ecx, 32 sub ecx, 32
ja convertloop ja convertloop
ret ret
...@@ -173,13 +173,13 @@ void CopyRow_X86(const uint8* src, uint8* dst, int count) { ...@@ -173,13 +173,13 @@ void CopyRow_X86(const uint8* src, uint8* dst, int count) {
#define HAS_COPYROW_SSE2 #define HAS_COPYROW_SSE2
void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
asm volatile ( asm volatile (
"sub %0,%1 \n"
"1: \n" "1: \n"
"movdqa (%0),%%xmm0 \n" "movdqa (%0),%%xmm0 \n"
"movdqa 0x10(%0),%%xmm1 \n" "movdqa 0x10(%0),%%xmm1 \n"
"movdqa %%xmm0,(%0,%1) \n"
"movdqa %%xmm1,0x10(%0,%1) \n"
"lea 0x20(%0),%0 \n" "lea 0x20(%0),%0 \n"
"movdqa %%xmm0,(%1) \n"
"movdqa %%xmm1,0x10(%1) \n"
"lea 0x20(%1),%1 \n"
"sub $0x20,%2 \n" "sub $0x20,%2 \n"
"ja 1b \n" "ja 1b \n"
: "+r"(src), // %0 : "+r"(src), // %0
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment