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

1 files changed, 21 insertions, 0 deletions

diff --git a/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp b/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp
index aa193d4955..6ff0884cec 100644
--- a/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp
+++ b/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp
@@ -21,6 +21,10 @@
 #include "VertexLoader_Position.h"
 #include "NativeVertexWriter.h"
 
+#if _M_SSE >= 301
+#include <tmmintrin.h>
+#endif
+
 extern float posScale;
 extern TVtxAttr *pVtxAttr;
 
@@ -146,16 +150,33 @@ inline void Pos_ReadIndex_Short(int Index)
 	VertexManager::s_pCurBufferPointer += 12;
 }
 
+#if _M_SSE >= 0x301
+static const __m128i kMaskSwap32_3 = _mm_set_epi32(0xFFFFFFFFL, 0x08090A0BL, 0x04050607L, 0x00010203L);
+static const __m128i kMaskSwap32_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x04050607L, 0x00010203L);
+#endif
+
 template<bool three>
 inline void Pos_ReadIndex_Float(int Index)
 {
 	const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (Index * arraystrides[ARRAY_POSITION]));
+
+#if _M_SSE >= 0x301
+
+	const __m128i a = _mm_loadu_si128((__m128i*)pData);
+	__m128i b = _mm_shuffle_epi8(a, three ? kMaskSwap32_3 : kMaskSwap32_2);
+	_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, b);
+
+#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;
+
+#endif
+
 	LOG_VTX();
 	VertexManager::s_pCurBufferPointer += 12;
 }