diff options
| author | nodchip <nodchip@gmail.com> | 2010-04-09 03:02:12 +0000 |
|---|---|---|
| committer | nodchip <nodchip@gmail.com> | 2010-04-09 03:02:12 +0000 |
| commit | 6136c94de594713375d2df4dc2b3d36c55699e76 (patch) | |
| tree | 0d945214845fd66c70f3ebac484d698ed77a2c62 /Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp | |
| parent | 73caf37bcab99de6b9c87d1a35738df31a6465cd (diff) | |
VideoCommon: merged SSSE3/SSE4.1 codes. Added some additional SSSE3/SSE4.1 codes which will be used in "The Legend of Zelda: Twilight Princess".
These codes don't work unless "_M_SSE=0x301", for SSSE3, or "_M_SSE=0x401", for SSE4.1, is defined as a preprocessor definition.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5300 8ced0084-cf51-0410-be5f-012b33b47a6e
Diffstat (limited to 'Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp')
| -rw-r--r-- | Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp b/Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp index bf38ef1e80..4cee83d9f6 100644 --- a/Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp +++ b/Source/Core/VideoCommon/Src/VertexLoader_Normal.cpp @@ -21,6 +21,10 @@ #include "VertexLoader_Normal.h" #include "NativeVertexWriter.h" +#if _M_SSE >= 0x301 +#include <tmmintrin.h> +#endif + #define LOG_NORM8() // PRIM_LOG("norm: %f %f %f, ", ((s8*)VertexManager::s_pCurBufferPointer)[-3]/127.0f, ((s8*)VertexManager::s_pCurBufferPointer)[-2]/127.0f, ((s8*)VertexManager::s_pCurBufferPointer)[-1]/127.0f); #define LOG_NORM16() // PRIM_LOG("norm: %f %f %f, ", ((s16*)VertexManager::s_pCurBufferPointer)[-3]/32767.0f, ((s16*)VertexManager::s_pCurBufferPointer)[-2]/32767.0f, ((s16*)VertexManager::s_pCurBufferPointer)[-1]/32767.0f); #define LOG_NORMF() // PRIM_LOG("norm: %f %f %f, ", ((float*)VertexManager::s_pCurBufferPointer)[-3], ((float*)VertexManager::s_pCurBufferPointer)[-2], ((float*)VertexManager::s_pCurBufferPointer)[-1]); @@ -411,14 +415,30 @@ void LOADERDECL VertexLoader_Normal::Normal_Index16_Byte_Expand16() LOG_NORM16(); } +#if _M_SSE >= 0x301 +static const __m128i kMaskSwap16_3 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x0FFFF0405L, 0x02030001L); +#endif + void LOADERDECL VertexLoader_Normal::Normal_Index16_Short() { u16 Index = DataReadU16(); const u16* pData = (const u16 *)(cached_arraybases[ARRAY_NORMAL] + (Index * arraystrides[ARRAY_NORMAL])); + +#if _M_SSE >= 0x301 + + __m128i a = _mm_loadl_epi64((__m128i*)pData); + __m128i b = _mm_shuffle_epi8(a, kMaskSwap16_3); + _mm_storel_epi64((__m128i*)VertexManager::s_pCurBufferPointer, b); + +#else + ((u16*)VertexManager::s_pCurBufferPointer)[0] = Common::swap16(pData[0]); ((u16*)VertexManager::s_pCurBufferPointer)[1] = Common::swap16(pData[1]); ((u16*)VertexManager::s_pCurBufferPointer)[2] = Common::swap16(pData[2]); ((u16*)VertexManager::s_pCurBufferPointer)[3] = 0; + +#endif + VertexManager::s_pCurBufferPointer += 8; LOG_NORM16(); } |