diff options
| author | nodchip <nodchip@gmail.com> | 2010-02-24 23:58:48 +0000 |
|---|---|---|
| committer | nodchip <nodchip@gmail.com> | 2010-02-24 23:58:48 +0000 |
| commit | 609151c6e88380c799b8d1069387cdbc31907a35 (patch) | |
| tree | f29027c1ce4c515f9a20daa25e336f75eeab400a /Source/Core/VideoCommon/Src/VertexLoader_Position.cpp | |
| parent | 1b43900f30f51b765d599490c5a07a723da28344 (diff) | |
Reverted because of some processor and performance issue. I will develop in a branch about SSSE3/SSE4.1.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5123 8ced0084-cf51-0410-be5f-012b33b47a6e
Diffstat (limited to 'Source/Core/VideoCommon/Src/VertexLoader_Position.cpp')
| -rw-r--r-- | Source/Core/VideoCommon/Src/VertexLoader_Position.cpp | 34 |
1 files changed, 8 insertions, 26 deletions
diff --git a/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp b/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp index c1920a99a9..907a84cca9 100644 --- a/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp +++ b/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp @@ -18,9 +18,7 @@ #ifndef VERTEXLOADER_POSITION_H #define VERTEXLOADER_POSITION_H -#include <nmmintrin.h> #include "Common.h" -#include "CPUDetect.h" #include "VideoCommon.h" #include "VertexLoader.h" #include "VertexLoader_Position.h" @@ -151,34 +149,18 @@ inline void Pos_ReadIndex_Short(int Index) VertexManager::s_pCurBufferPointer += 12; } -static const __m128i kMaskSwap32 = _mm_set_epi32(0xFFFFFFFFL, 0x08090A0BL, 0x04050607L, 0x00010203L); template<bool three> inline void Pos_ReadIndex_Float(int Index) { const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (Index * arraystrides[ARRAY_POSITION])); - - if (cpu_info.bSSE4_1) { - const __m128i a = _mm_loadu_si128((__m128i*)pData); - __m128i b = _mm_shuffle_epi8(a, kMaskSwap32); - if (!three) { - b = _mm_insert_epi32(b, 0, 2); - } - u8* p = VertexManager::s_pCurBufferPointer; - _mm_storeu_si128((__m128i*)p, b); - LOG_VTX(); - p += 12; - VertexManager::s_pCurBufferPointer = p; - - } else { - ((u32*)VertexManager::s_pCurBufferPointer)[0] = Common::swap32(pData[0]); - ((u32*)VertexManager::s_pCurBufferPointer)[1] = Common::swap32(pData[1]); - if (three) - ((u32*)VertexManager::s_pCurBufferPointer)[2] = Common::swap32(pData[2]); - else - ((float*)VertexManager::s_pCurBufferPointer)[2] = 0.0f; - LOG_VTX(); - VertexManager::s_pCurBufferPointer += 12; - } + ((u32*)VertexManager::s_pCurBufferPointer)[0] = Common::swap32(pData[0]); + ((u32*)VertexManager::s_pCurBufferPointer)[1] = Common::swap32(pData[1]); + if (three) + ((u32*)VertexManager::s_pCurBufferPointer)[2] = Common::swap32(pData[2]); + else + ((float*)VertexManager::s_pCurBufferPointer)[2] = 0.0f; + LOG_VTX(); + VertexManager::s_pCurBufferPointer += 12; } // ============================================================================== |