ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Align stack variables on a 16-bytes boundary in SSSE3 functions 

Fixes issue 4450. Thanks to pholklore1 for his patch.
This commit is contained in:
Pierre Bourdon 2011-10-08 17:33:21 +02:00
parent 2f254585bf
commit 59072adc32
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Source/Core/VideoCommon/Src/VertexLoader_Position.cpp
View File

@ -179,8 +179,8 @@ void Pos_ReadIndex_Float_SSSE3(int Index)
if(Index < MaxSize) if(Index < MaxSize)
{ {
const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (Index * arraystrides[ARRAY_POSITION])); const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (Index * arraystrides[ARRAY_POSITION]));
const __m128i a = _mm_loadu_si128((__m128i*)pData); GC_ALIGNED128(const __m128i a = _mm_loadu_si128((__m128i*)pData));
__m128i b = _mm_shuffle_epi8(a, three ? kMaskSwap32_3 : kMaskSwap32_2); GC_ALIGNED128(__m128i b = _mm_shuffle_epi8(a, three ? kMaskSwap32_3 : kMaskSwap32_2));
_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, b); _mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, b);
LOG_VTX(); LOG_VTX();
VertexManager::s_pCurBufferPointer += 12; VertexManager::s_pCurBufferPointer += 12;

4
Source/Core/VideoCommon/Src/VertexLoader_TextCoord.cpp
View File

@ -353,8 +353,8 @@ void LOADERDECL TexCoord_ReadIndex16_Float2_SSSE3()
{ {
u16 Index = DataReadU16(); u16 Index = DataReadU16();
const u32 *pData = (const u32 *)(cached_arraybases[ARRAY_TEXCOORD0+tcIndex] + (Index * arraystrides[ARRAY_TEXCOORD0+tcIndex])); const u32 *pData = (const u32 *)(cached_arraybases[ARRAY_TEXCOORD0+tcIndex] + (Index * arraystrides[ARRAY_TEXCOORD0+tcIndex]));
const __m128i a = _mm_loadl_epi64((__m128i*)pData); GC_ALIGNED128(const __m128i a = _mm_loadl_epi64((__m128i*)pData));
const __m128i b = _mm_shuffle_epi8(a, kMaskSwap32); GC_ALIGNED128(const __m128i b = _mm_shuffle_epi8(a, kMaskSwap32));
u8* p = VertexManager::s_pCurBufferPointer; u8* p = VertexManager::s_pCurBufferPointer;
_mm_storel_epi64((__m128i*)p, b); _mm_storel_epi64((__m128i*)p, b);
LOG_TEX2(); LOG_TEX2();