Vertex Loader: SSE implementations of more position/texcoord/normal formats

~35-45% faster NFS:HP2, possibly other vertex-bound games.

7 files changed, 298 insertions, 105 deletions

diff --git a/Source/Core/VideoBackends/Software/SWVertexLoader.cpp b/Source/Core/VideoBackends/Software/SWVertexLoader.cpp
index c712115282..0612a743da 100644
--- a/Source/Core/VideoBackends/Software/SWVertexLoader.cpp
+++ b/Source/Core/VideoBackends/Software/SWVertexLoader.cpp
@@ -42,15 +42,15 @@ void SWVertexLoader::SetFormat(u8 attributeIndex, u8 primitiveType)
 {
 	m_CurrentVat = &g_main_cp_state.vtx_attr[attributeIndex];
 
-	posScale = 1.0f / float(1 << m_CurrentVat->g0.PosFrac);
-	tcScale[0] = 1.0f / float(1 << m_CurrentVat->g0.Tex0Frac);
-	tcScale[1] = 1.0f / float(1 << m_CurrentVat->g1.Tex1Frac);
-	tcScale[2] = 1.0f / float(1 << m_CurrentVat->g1.Tex2Frac);
-	tcScale[3] = 1.0f / float(1 << m_CurrentVat->g1.Tex3Frac);
-	tcScale[4] = 1.0f / float(1 << m_CurrentVat->g2.Tex4Frac);
-	tcScale[5] = 1.0f / float(1 << m_CurrentVat->g2.Tex5Frac);
-	tcScale[6] = 1.0f / float(1 << m_CurrentVat->g2.Tex6Frac);
-	tcScale[7] = 1.0f / float(1 << m_CurrentVat->g2.Tex7Frac);
+	posScale[0] = posScale[1] = posScale[2] = posScale[3] = 1.0f / float(1 << m_CurrentVat->g0.PosFrac);
+	tcScale[0][0] = tcScale[0][1] = 1.0f / float(1 << m_CurrentVat->g0.Tex0Frac);
+	tcScale[1][0] = tcScale[1][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex1Frac);
+	tcScale[2][0] = tcScale[2][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex2Frac);
+	tcScale[3][0] = tcScale[3][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex3Frac);
+	tcScale[4][0] = tcScale[4][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex4Frac);
+	tcScale[5][0] = tcScale[5][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex5Frac);
+	tcScale[6][0] = tcScale[6][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex6Frac);
+	tcScale[7][0] = tcScale[7][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex7Frac);
 
 	//TexMtx
 	const u64 tmDesc[8] = {
diff --git a/Source/Core/VideoCommon/VertexLoader.cpp b/Source/Core/VideoCommon/VertexLoader.cpp
index 373dfae16b..80a891ff26 100644
--- a/Source/Core/VideoCommon/VertexLoader.cpp
+++ b/Source/Core/VideoCommon/VertexLoader.cpp
@@ -41,8 +41,9 @@ static int s_texmtxread = 0;
 int tcIndex;
 int colIndex;
 int colElements[2];
-float posScale;
-float tcScale[8];
+// Duplicated (4x and 2x respectively) and used in SSE code in the vertex loader JIT
+GC_ALIGNED128(float posScale[4]);
+GC_ALIGNED64(float tcScale[8][2]);
 
 static const float fractionTable[32] = {
 	1.0f / (1U << 0), 1.0f / (1U << 1), 1.0f / (1U << 2), 1.0f / (1U << 3),
@@ -65,10 +66,8 @@ static void LOADERDECL PosMtx_ReadDirect_UByte()
 
 static void LOADERDECL PosMtx_Write()
 {
-	DataWrite<u8>(s_curposmtx);
-	DataWrite<u8>(0);
-	DataWrite<u8>(0);
-	DataWrite<u8>(0);
+	// u8, 0, 0, 0
+	DataWrite<u32>(s_curposmtx);
 }
 
 static void LOADERDECL TexMtx_ReadDirect_UByte()
@@ -451,10 +450,10 @@ void VertexLoader::SetupRunVertices(const VAT& vat, int primitive, int const cou
 	m_VtxAttr.texCoord[6].Frac = vat.g2.Tex6Frac;
 	m_VtxAttr.texCoord[7].Frac = vat.g2.Tex7Frac;
 
-	posScale = fractionTable[m_VtxAttr.PosFrac];
+	posScale[0] = posScale[1] = posScale[2] = posScale[3] = fractionTable[m_VtxAttr.PosFrac];
 	if (m_native_components & VB_HAS_UVALL)
 		for (int i = 0; i < 8; i++)
-			tcScale[i] = fractionTable[m_VtxAttr.texCoord[i].Frac];
+			tcScale[i][0] = tcScale[i][1] = fractionTable[m_VtxAttr.texCoord[i].Frac];
 	for (int i = 0; i < 2; i++)
 		colElements[i] = m_VtxAttr.color[i].Elements;
 
diff --git a/Source/Core/VideoCommon/VertexLoader.h b/Source/Core/VideoCommon/VertexLoader.h
index c81be52556..71159b60d7 100644
--- a/Source/Core/VideoCommon/VertexLoader.h
+++ b/Source/Core/VideoCommon/VertexLoader.h
@@ -19,6 +19,13 @@
 #include "VideoCommon/DataReader.h"
 #include "VideoCommon/NativeVertexFormat.h"
 
+#if _M_SSE >= 0x401
+#include <smmintrin.h>
+#include <emmintrin.h>
+#elif _M_SSE >= 0x301 && !(defined __GNUC__ && !defined __SSSE3__)
+#include <tmmintrin.h>
+#endif
+
 #ifdef _M_X86
 #define USE_VERTEX_LOADER_JIT
 #endif
@@ -27,8 +34,8 @@
 extern int tcIndex;
 extern int colIndex;
 extern int colElements[2];
-extern float posScale;
-extern float tcScale[8];
+GC_ALIGNED128(extern float posScale[4]);
+GC_ALIGNED64(extern float tcScale[8][2]);
 
 class VertexLoaderUID
 {
@@ -155,3 +162,61 @@ private:
 	void WriteSetVariable(int bits, void *address, Gen::OpArg dest);
 #endif
 };
+
+#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);
+static const __m128i kMaskSwap16to32l_3 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFF0405L, 0xFFFF0203L, 0xFFFF0001L);
+static const __m128i kMaskSwap16to32l_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFF0203L, 0xFFFF0001L);
+static const __m128i kMaskSwap16to32h_3 = _mm_set_epi32(0xFFFFFFFFL, 0x0405FFFFL, 0x0203FFFFL, 0x0001FFFFL);
+static const __m128i kMaskSwap16to32h_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x0203FFFFL, 0x0001FFFFL);
+static const __m128i kMask8to32l_3 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFF02L, 0xFFFFFF01L, 0xFFFFFF00L);
+static const __m128i kMask8to32l_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFF01L, 0xFFFFFF00L);
+static const __m128i kMask8to32h_3 = _mm_set_epi32(0xFFFFFFFFL, 0x02FFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL);
+static const __m128i kMask8to32h_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL);
+
+template <typename T, bool threeIn, bool threeOut>
+__forceinline void Vertex_Read_SSSE3(const T* pData, __m128 scale)
+{
+	__m128i coords, mask;
+
+	int loadBytes = sizeof(T) * (2 + threeIn);
+	if (loadBytes > 8)
+		coords = _mm_loadu_si128((__m128i*)pData);
+	else if (loadBytes > 4)
+		coords = _mm_loadl_epi64((__m128i*)pData);
+	else
+		coords = _mm_cvtsi32_si128(*(u32*)pData);
+
+	// Float case (no scaling)
+	if (sizeof(T) == 4)
+	{
+		coords = _mm_shuffle_epi8(coords, threeIn ? kMaskSwap32_3 : kMaskSwap32_2);
+		if (threeOut)
+			_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, coords);
+		else
+			_mm_storel_epi64((__m128i*)VertexManager::s_pCurBufferPointer, coords);
+	}
+	else
+	{
+		// Byte swap, unpack, and move to high bytes for sign extend.
+		if (std::is_unsigned<T>::value)
+			mask = sizeof(T) == 2 ? (threeIn ? kMaskSwap16to32l_3 : kMaskSwap16to32l_2) : (threeIn ? kMask8to32l_3 : kMask8to32l_2);
+		else
+			mask = sizeof(T) == 2 ? (threeIn ? kMaskSwap16to32h_3 : kMaskSwap16to32h_2) : (threeIn ? kMask8to32h_3 : kMask8to32h_2);
+		coords = _mm_shuffle_epi8(coords, mask);
+
+		// Sign extend
+		if (std::is_signed<T>::value)
+			coords = _mm_srai_epi32(coords, 32 - sizeof(T) * 8);
+
+		__m128 out = _mm_mul_ps(_mm_cvtepi32_ps(coords), scale);
+		if (threeOut)
+			_mm_storeu_ps((float*)VertexManager::s_pCurBufferPointer, out);
+		else
+			_mm_storel_pi((__m64*)VertexManager::s_pCurBufferPointer, out);
+	}
+
+	VertexManager::s_pCurBufferPointer += sizeof(float) * (2 + threeOut);
+}
+#endif
\ No newline at end of file
diff --git a/Source/Core/VideoCommon/VertexLoader_Normal.cpp b/Source/Core/VideoCommon/VertexLoader_Normal.cpp
index 3d58592d70..8ff5a767e7 100644
--- a/Source/Core/VideoCommon/VertexLoader_Normal.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_Normal.cpp
@@ -3,7 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <cmath>
-#include <limits>
+#include <type_traits>
 
 #include "Common/CommonTypes.h"
 #include "Common/CPUDetect.h"
@@ -13,13 +13,6 @@
 #include "VideoCommon/VertexManagerBase.h"
 #include "VideoCommon/VideoCommon.h"
 
-#if _M_SSE >= 0x401
-#include <smmintrin.h>
-#include <emmintrin.h>
-#elif _M_SSE >= 0x301 && !(defined __GNUC__ && !defined __SSSE3__)
-#include <tmmintrin.h>
-#endif
-
 // warning: mapping buffer should be disabled to use this
 #define LOG_NORM()  // PRIM_LOG("norm: %f %f %f, ", ((float*)VertexManager::s_pCurBufferPointer)[-3], ((float*)VertexManager::s_pCurBufferPointer)[-2], ((float*)VertexManager::s_pCurBufferPointer)[-1]);
 
@@ -37,7 +30,7 @@ __forceinline float FracAdjust(T val)
 	//auto const U16FRAC = 1.f / (1u << 15);
 
 	// TODO: is this right?
-	return val / float(1u << (sizeof(T) * 8 - std::numeric_limits<T>::is_signed - 1));
+	return val / float(1u << (sizeof(T) * 8 - std::is_signed<T>::value - 1));
 }
 
 template <>
@@ -76,11 +69,11 @@ struct Normal_Direct
 template <typename I, typename T, int N, int Offset>
 __forceinline void Normal_Index_Offset()
 {
-	static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
 
 	auto const index = DataRead<I>();
 	auto const data = reinterpret_cast<const T*>(cached_arraybases[ARRAY_NORMAL]
-		+ (index * g_main_cp_state.array_strides[ARRAY_NORMAL]) + sizeof(T) * 3 * Offset);
+	                + (index * g_main_cp_state.array_strides[ARRAY_NORMAL]) + sizeof(T) * 3 * Offset);
 	ReadIndirect<T, N * 3>(data);
 }
 
@@ -108,6 +101,63 @@ struct Normal_Index_Indices3
 	static const int size = sizeof(I) * 3;
 };
 
+#if _M_SSE >= 0x301
+template <typename T, int N>
+struct Normal_Direct_SSSE3
+{
+	static void LOADERDECL function()
+	{
+		const T* pData = reinterpret_cast<const T*>(DataGetPosition());
+		const float frac = 1. / float(1u << (sizeof(T) * 8 - std::is_signed<T>::value - 1));
+		const __m128 scale = _mm_set_ps(frac, frac, frac, frac);
+		for (int i = 0; i < N; i++, pData += 3)
+			Vertex_Read_SSSE3<T, true, true>(pData, scale);
+		DataSkip<N * 3 * sizeof(T)>();
+	}
+
+	static const int size = sizeof(T) * N * 3;
+};
+
+template <typename I, typename T, int N, int Offset>
+__forceinline void Normal_Index_Offset_SSSE3()
+{
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
+
+	auto const index = DataRead<I>();
+	const T* pData = (const T*)(cached_arraybases[ARRAY_NORMAL]
+	               + (index * g_main_cp_state.array_strides[ARRAY_NORMAL]) + sizeof(T) * 3 * Offset);
+	const float frac = 1. / float(1u << (sizeof(T) * 8 - std::is_signed<T>::value - 1));
+	const __m128 scale = _mm_set_ps(frac, frac, frac, frac);
+	for (int i = 0; i < N; i++, pData += 3)
+		Vertex_Read_SSSE3<T, true, true>(pData, scale);
+}
+
+template <typename I, typename T, int N>
+struct Normal_Index_SSSE3
+{
+	static void LOADERDECL function()
+	{
+		Normal_Index_Offset_SSSE3<I, T, N, 0>();
+	}
+
+	static const int size = sizeof(I);
+};
+
+template <typename I, typename T>
+struct Normal_Index_Indices3_SSSE3
+{
+	static void LOADERDECL function()
+	{
+		Normal_Index_Offset_SSSE3<I, T, 1, 0>();
+		Normal_Index_Offset_SSSE3<I, T, 1, 1>();
+		Normal_Index_Offset_SSSE3<I, T, 1, 2>();
+	}
+
+	static const int size = sizeof(I) * 3;
+};
+
+#endif
+
 }
 
 void VertexLoader_Normal::Init()
@@ -180,6 +230,77 @@ void VertexLoader_Normal::Init()
 	m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_USHORT] = Normal_Index_Indices3<u16, u16>();
 	m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_SHORT]  = Normal_Index_Indices3<u16, s16>();
 	m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_FLOAT]  = Normal_Index_Indices3<u16, float>();
+
+#if _M_SSE >= 0x301
+	if (cpu_info.bSSSE3)
+	{
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT][FORMAT_UBYTE] = Normal_Direct_SSSE3<u8, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT][FORMAT_BYTE] = Normal_Direct_SSSE3<s8, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT][FORMAT_USHORT] = Normal_Direct_SSSE3<u16, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT][FORMAT_SHORT] = Normal_Direct_SSSE3<s16, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT][FORMAT_FLOAT] = Normal_Direct_SSSE3<float, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT3][FORMAT_UBYTE] = Normal_Direct_SSSE3<u8, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT3][FORMAT_BYTE] = Normal_Direct_SSSE3<s8, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT3][FORMAT_USHORT] = Normal_Direct_SSSE3<u16, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT3][FORMAT_SHORT] = Normal_Direct_SSSE3<s16, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES1][NRM_NBT3][FORMAT_FLOAT] = Normal_Direct_SSSE3<float, 3>();
+
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT][FORMAT_UBYTE] = Normal_Direct_SSSE3<u8, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT][FORMAT_BYTE] = Normal_Direct_SSSE3<s8, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT][FORMAT_USHORT] = Normal_Direct_SSSE3<u16, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT][FORMAT_SHORT] = Normal_Direct_SSSE3<s16, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT][FORMAT_FLOAT] = Normal_Direct_SSSE3<float, 1>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT3][FORMAT_UBYTE] = Normal_Direct_SSSE3<u8, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT3][FORMAT_BYTE] = Normal_Direct_SSSE3<s8, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT3][FORMAT_USHORT] = Normal_Direct_SSSE3<u16, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT3][FORMAT_SHORT] = Normal_Direct_SSSE3<s16, 3>();
+		m_Table[NRM_DIRECT][NRM_INDICES3][NRM_NBT3][FORMAT_FLOAT] = Normal_Direct_SSSE3<float, 3>();
+
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT][FORMAT_UBYTE] = Normal_Index_SSSE3<u8, u8, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT][FORMAT_BYTE] = Normal_Index_SSSE3<u8, s8, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT][FORMAT_USHORT] = Normal_Index_SSSE3<u8, u16, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT][FORMAT_SHORT] = Normal_Index_SSSE3<u8, s16, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT][FORMAT_FLOAT] = Normal_Index_SSSE3<u8, float, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT3][FORMAT_UBYTE] = Normal_Index_SSSE3<u8, u8, 3>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT3][FORMAT_BYTE] = Normal_Index_SSSE3<u8, s8, 3>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT3][FORMAT_USHORT] = Normal_Index_SSSE3<u8, u16, 3>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT3][FORMAT_SHORT] = Normal_Index_SSSE3<u8, s16, 3>();
+		m_Table[NRM_INDEX8][NRM_INDICES1][NRM_NBT3][FORMAT_FLOAT] = Normal_Index_SSSE3<u8, float, 3>();
+
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT][FORMAT_UBYTE] = Normal_Index_SSSE3<u8, u8, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT][FORMAT_BYTE] = Normal_Index_SSSE3<u8, s8, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT][FORMAT_USHORT] = Normal_Index_SSSE3<u8, u16, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT][FORMAT_SHORT] = Normal_Index_SSSE3<u8, s16, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT][FORMAT_FLOAT] = Normal_Index_SSSE3<u8, float, 1>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT3][FORMAT_UBYTE] = Normal_Index_Indices3_SSSE3<u8, u8>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT3][FORMAT_BYTE] = Normal_Index_Indices3_SSSE3<u8, s8>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT3][FORMAT_USHORT] = Normal_Index_Indices3_SSSE3<u8, u16>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT3][FORMAT_SHORT] = Normal_Index_Indices3_SSSE3<u8, s16>();
+		m_Table[NRM_INDEX8][NRM_INDICES3][NRM_NBT3][FORMAT_FLOAT] = Normal_Index_Indices3_SSSE3<u8, float>();
+
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT][FORMAT_UBYTE] = Normal_Index_SSSE3<u16, u8, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT][FORMAT_BYTE] = Normal_Index_SSSE3<u16, s8, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT][FORMAT_USHORT] = Normal_Index_SSSE3<u16, u16, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT][FORMAT_SHORT] = Normal_Index_SSSE3<u16, s16, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT][FORMAT_FLOAT] = Normal_Index_SSSE3<u16, float, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT3][FORMAT_UBYTE] = Normal_Index_SSSE3<u16, u8, 3>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT3][FORMAT_BYTE] = Normal_Index_SSSE3<u16, s8, 3>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT3][FORMAT_USHORT] = Normal_Index_SSSE3<u16, u16, 3>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT3][FORMAT_SHORT] = Normal_Index_SSSE3<u16, s16, 3>();
+		m_Table[NRM_INDEX16][NRM_INDICES1][NRM_NBT3][FORMAT_FLOAT] = Normal_Index_SSSE3<u16, float, 3>();
+
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT][FORMAT_UBYTE] = Normal_Index_SSSE3<u16, u8, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT][FORMAT_BYTE] = Normal_Index_SSSE3<u16, s8, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT][FORMAT_USHORT] = Normal_Index_SSSE3<u16, u16, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT][FORMAT_SHORT] = Normal_Index_SSSE3<u16, s16, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT][FORMAT_FLOAT] = Normal_Index_SSSE3<u16, float, 1>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_UBYTE] = Normal_Index_Indices3_SSSE3<u16, u8>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_BYTE] = Normal_Index_Indices3_SSSE3<u16, s8>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_USHORT] = Normal_Index_Indices3_SSSE3<u16, u16>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_SHORT] = Normal_Index_Indices3_SSSE3<u16, s16>();
+		m_Table[NRM_INDEX16][NRM_INDICES3][NRM_NBT3][FORMAT_FLOAT] = Normal_Index_Indices3_SSSE3<u16, float>();
+	}
+#endif
 }
 
 unsigned int VertexLoader_Normal::GetSize(u64 _type,
diff --git a/Source/Core/VideoCommon/VertexLoader_Position.cpp b/Source/Core/VideoCommon/VertexLoader_Position.cpp
index a38d429d58..6c50dd1560 100644
--- a/Source/Core/VideoCommon/VertexLoader_Position.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_Position.cpp
@@ -2,7 +2,7 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
-#include <limits>
+#include <type_traits>
 
 #include "Common/CommonTypes.h"
 #include "Common/CPUDetect.h"
@@ -74,7 +74,7 @@ template <typename T, int N>
 void LOADERDECL Pos_ReadDirect()
 {
 	static_assert(N <= 3, "N > 3 is not sane!");
-	auto const scale = posScale;
+	auto const scale = posScale[0];
 	DataWriter dst;
 	DataReader src;
 
@@ -87,12 +87,12 @@ void LOADERDECL Pos_ReadDirect()
 template <typename I, typename T, int N>
 void LOADERDECL Pos_ReadIndex()
 {
-	static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
 	static_assert(N <= 3, "N > 3 is not sane!");
 
 	auto const index = DataRead<I>();
 	auto const data = reinterpret_cast<const T*>(cached_arraybases[ARRAY_POSITION] + (index * g_main_cp_state.array_strides[ARRAY_POSITION]));
-	auto const scale = posScale;
+	auto const scale = posScale[0];
 	DataWriter dst;
 
 	for (int i = 0; i < 3; ++i)
@@ -102,18 +102,22 @@ void LOADERDECL Pos_ReadIndex()
 }
 
 #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);
+template <typename T, bool three>
+void LOADERDECL Pos_ReadDirect_SSSE3()
+{
+	const T* pData = reinterpret_cast<const T*>(DataGetPosition());
+	Vertex_Read_SSSE3<T, three, true>(pData, *(__m128*)posScale);
+	DataSkip<(2 + three) * sizeof(T)>();
+	LOG_VTX();
+}
 
-template <typename I, bool three>
-void LOADERDECL Pos_ReadIndex_Float_SSSE3()
+template <typename I, typename T, bool three>
+void LOADERDECL Pos_ReadIndex_SSSE3()
 {
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
 	auto const index = DataRead<I>();
-	const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (index * g_main_cp_state.array_strides[ARRAY_POSITION]));
-	GC_ALIGNED128(const __m128i a = _mm_loadu_si128((__m128i*)pData));
-	GC_ALIGNED128(__m128i b = _mm_shuffle_epi8(a, three ? kMaskSwap32_3 : kMaskSwap32_2));
-	_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, b);
-	VertexManager::s_pCurBufferPointer += sizeof(float) * 3;
+	const T* pData = (const T*)(cached_arraybases[ARRAY_POSITION] + (index * g_main_cp_state.array_strides[ARRAY_POSITION]));
+	Vertex_Read_SSSE3<T, three, true>(pData, *(__m128*)posScale);
 	LOG_VTX();
 }
 #endif
@@ -169,15 +173,39 @@ void VertexLoader_Position::Init()
 {
 
 #if _M_SSE >= 0x301
-
 	if (cpu_info.bSSSE3)
 	{
-		tableReadPosition[2][4][0] = Pos_ReadIndex_Float_SSSE3<u8, false>;
-		tableReadPosition[2][4][1] = Pos_ReadIndex_Float_SSSE3<u8, true>;
-		tableReadPosition[3][4][0] = Pos_ReadIndex_Float_SSSE3<u16, false>;
-		tableReadPosition[3][4][1] = Pos_ReadIndex_Float_SSSE3<u16, true>;
+		tableReadPosition[1][0][0] = Pos_ReadDirect_SSSE3<u8, false>;
+		tableReadPosition[1][0][1] = Pos_ReadDirect_SSSE3<u8, true>;
+		tableReadPosition[1][1][0] = Pos_ReadDirect_SSSE3<s8, false>;
+		tableReadPosition[1][1][1] = Pos_ReadDirect_SSSE3<s8, true>;
+		tableReadPosition[1][2][0] = Pos_ReadDirect_SSSE3<u16, false>;
+		tableReadPosition[1][2][1] = Pos_ReadDirect_SSSE3<u16, true>;
+		tableReadPosition[1][3][0] = Pos_ReadDirect_SSSE3<s16, false>;
+		tableReadPosition[1][3][1] = Pos_ReadDirect_SSSE3<s16, true>;
+		tableReadPosition[1][4][0] = Pos_ReadDirect_SSSE3<float, false>;
+		tableReadPosition[1][4][1] = Pos_ReadDirect_SSSE3<float, true>;
+		tableReadPosition[2][0][0] = Pos_ReadIndex_SSSE3<u8, u8, false>;
+		tableReadPosition[2][0][1] = Pos_ReadIndex_SSSE3<u8, u8, true>;
+		tableReadPosition[3][0][0] = Pos_ReadIndex_SSSE3<u16, u8, false>;
+		tableReadPosition[3][0][1] = Pos_ReadIndex_SSSE3<u16, u8, true>;
+		tableReadPosition[2][1][0] = Pos_ReadIndex_SSSE3<u8, s8, false>;
+		tableReadPosition[2][1][1] = Pos_ReadIndex_SSSE3<u8, s8, true>;
+		tableReadPosition[3][1][0] = Pos_ReadIndex_SSSE3<u16, s8, false>;
+		tableReadPosition[3][1][1] = Pos_ReadIndex_SSSE3<u16, s8, true>;
+		tableReadPosition[2][2][0] = Pos_ReadIndex_SSSE3<u8, u16, false>;
+		tableReadPosition[2][2][1] = Pos_ReadIndex_SSSE3<u8, u16, true>;
+		tableReadPosition[3][2][0] = Pos_ReadIndex_SSSE3<u16, u16, false>;
+		tableReadPosition[3][2][1] = Pos_ReadIndex_SSSE3<u16, u16, true>;
+		tableReadPosition[2][3][0] = Pos_ReadIndex_SSSE3<u8, s16, false>;
+		tableReadPosition[2][3][1] = Pos_ReadIndex_SSSE3<u8, s16, true>;
+		tableReadPosition[3][3][0] = Pos_ReadIndex_SSSE3<u16, s16, false>;
+		tableReadPosition[3][3][1] = Pos_ReadIndex_SSSE3<u16, s16, true>;
+		tableReadPosition[2][4][0] = Pos_ReadIndex_SSSE3<u8, float, false>;
+		tableReadPosition[2][4][1] = Pos_ReadIndex_SSSE3<u8, float, true>;
+		tableReadPosition[3][4][0] = Pos_ReadIndex_SSSE3<u16, float, false>;
+		tableReadPosition[3][4][1] = Pos_ReadIndex_SSSE3<u16, float, true>;
 	}
-
 #endif
 
 }
diff --git a/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp b/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
index 25114ac82d..8acebd9213 100644
--- a/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
@@ -2,6 +2,8 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
+#include <type_traits>
+
 #include "Common/CommonTypes.h"
 #include "Common/CPUDetect.h"
 
@@ -10,13 +12,6 @@
 #include "VideoCommon/VertexManagerBase.h"
 #include "VideoCommon/VideoCommon.h"
 
-
-#if _M_SSE >= 0x401
-#include <smmintrin.h>
-#elif _M_SSE >= 0x301 && !(defined __GNUC__ && !defined __SSSE3__)
-#include <tmmintrin.h>
-#endif
-
 template <int N>
 void LOG_TEX();
 
@@ -54,7 +49,7 @@ float TCScale(float val, float scale)
 template <typename T, int N>
 void LOADERDECL TexCoord_ReadDirect()
 {
-	auto const scale = tcScale[tcIndex];
+	auto const scale = tcScale[tcIndex][0];
 	DataWriter dst;
 	DataReader src;
 
@@ -69,12 +64,12 @@ void LOADERDECL TexCoord_ReadDirect()
 template <typename I, typename T, int N>
 void LOADERDECL TexCoord_ReadIndex()
 {
-	static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
 
 	auto const index = DataRead<I>();
 	auto const data = reinterpret_cast<const T*>(cached_arraybases[ARRAY_TEXCOORD0 + tcIndex]
-		+ (index * g_main_cp_state.array_strides[ARRAY_TEXCOORD0 + tcIndex]));
-	auto const scale = tcScale[tcIndex];
+	                + (index * g_main_cp_state.array_strides[ARRAY_TEXCOORD0 + tcIndex]));
+	auto const scale = tcScale[tcIndex][0];
 	DataWriter dst;
 
 	for (int i = 0; i != N; ++i)
@@ -84,44 +79,27 @@ void LOADERDECL TexCoord_ReadIndex()
 	++tcIndex;
 }
 
-#if _M_SSE >= 0x401
-static const __m128i kMaskSwap16_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0x02030001L);
-
-template <typename I>
-void LOADERDECL TexCoord_ReadIndex_Short2_SSE4()
+#if _M_SSE >= 0x301
+template <typename T>
+void LOADERDECL TexCoord_ReadDirect2_SSSE3()
 {
-	static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
-
-	// Heavy in ZWW
-	auto const index = DataRead<I>();
-	const s32 *pData = (const s32*)(cached_arraybases[ARRAY_TEXCOORD0+tcIndex] + (index * g_main_cp_state.array_strides[ARRAY_TEXCOORD0+tcIndex]));
-	const __m128i a = _mm_cvtsi32_si128(*pData);
-	const __m128i b = _mm_shuffle_epi8(a, kMaskSwap16_2);
-	const __m128i c = _mm_cvtepi16_epi32(b);
-	const __m128 d = _mm_cvtepi32_ps(c);
-	const __m128 e = _mm_load1_ps(&tcScale[tcIndex]);
-	const __m128 f = _mm_mul_ps(d, e);
-	_mm_storeu_ps((float*)VertexManager::s_pCurBufferPointer, f);
-	VertexManager::s_pCurBufferPointer += sizeof(float) * 2;
+	const T* pData = reinterpret_cast<const T*>(DataGetPosition());
+	__m128 scale = _mm_castsi128_ps(_mm_loadl_epi64((__m128i*)tcScale[tcIndex]));
+	Vertex_Read_SSSE3<T, false, false>(pData, scale);
+	DataSkip<2 * sizeof(T)>();
 	LOG_TEX<2>();
 	tcIndex++;
 }
-#endif
 
-#if _M_SSE >= 0x301
-static const __m128i kMaskSwap32 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x04050607L, 0x00010203L);
-
-template <typename I>
-void LOADERDECL TexCoord_ReadIndex_Float2_SSSE3()
+template <typename I, typename T>
+void LOADERDECL TexCoord_ReadIndex2_SSSE3()
 {
-	static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
+	static_assert(std::is_unsigned<I>::value, "Only unsigned I is sane!");
 
 	auto const index = DataRead<I>();
-	const u32 *pData = (const u32 *)(cached_arraybases[ARRAY_TEXCOORD0+tcIndex] + (index * g_main_cp_state.array_strides[ARRAY_TEXCOORD0+tcIndex]));
-	GC_ALIGNED128(const __m128i a = _mm_loadl_epi64((__m128i*)pData));
-	GC_ALIGNED128(const __m128i b = _mm_shuffle_epi8(a, kMaskSwap32));
-	_mm_storel_epi64((__m128i*)VertexManager::s_pCurBufferPointer, b);
-	VertexManager::s_pCurBufferPointer += sizeof(float) * 2;
+	const T* pData = (const T*)(cached_arraybases[ARRAY_TEXCOORD0 + tcIndex] + (index * g_main_cp_state.array_strides[ARRAY_TEXCOORD0 + tcIndex]));
+	__m128 scale = _mm_castsi128_ps(_mm_loadl_epi64((__m128i*)tcScale[tcIndex]));
+	Vertex_Read_SSSE3<T, false, false>(pData, scale);
 	LOG_TEX<2>();
 	tcIndex++;
 }
@@ -177,23 +155,24 @@ void VertexLoader_TextCoord::Init()
 {
 
 #if _M_SSE >= 0x301
-
 	if (cpu_info.bSSSE3)
 	{
-		tableReadTexCoord[2][4][1] = TexCoord_ReadIndex_Float2_SSSE3<u8>;
-		tableReadTexCoord[3][4][1] = TexCoord_ReadIndex_Float2_SSSE3<u16>;
-	}
-
-#endif
-
-#if _M_SSE >= 0x401
-
-	if (cpu_info.bSSE4_1)
-	{
-		tableReadTexCoord[2][3][1] = TexCoord_ReadIndex_Short2_SSE4<u8>;
-		tableReadTexCoord[3][3][1] = TexCoord_ReadIndex_Short2_SSE4<u16>;
+		tableReadTexCoord[1][0][1] = TexCoord_ReadDirect2_SSSE3<u8>;
+		tableReadTexCoord[1][1][1] = TexCoord_ReadDirect2_SSSE3<s8>;
+		tableReadTexCoord[1][2][1] = TexCoord_ReadDirect2_SSSE3<u16>;
+		tableReadTexCoord[1][3][1] = TexCoord_ReadDirect2_SSSE3<s16>;
+		tableReadTexCoord[1][4][1] = TexCoord_ReadDirect2_SSSE3<float>;
+		tableReadTexCoord[2][0][1] = TexCoord_ReadIndex2_SSSE3<u8, u8>;
+		tableReadTexCoord[3][0][1] = TexCoord_ReadIndex2_SSSE3<u16, u8>;
+		tableReadTexCoord[2][1][1] = TexCoord_ReadIndex2_SSSE3<u8, s8>;
+		tableReadTexCoord[3][1][1] = TexCoord_ReadIndex2_SSSE3<u16, s8>;
+		tableReadTexCoord[2][2][1] = TexCoord_ReadIndex2_SSSE3<u8, u16>;
+		tableReadTexCoord[3][2][1] = TexCoord_ReadIndex2_SSSE3<u16, u16>;
+		tableReadTexCoord[2][3][1] = TexCoord_ReadIndex2_SSSE3<u8, s16>;
+		tableReadTexCoord[3][3][1] = TexCoord_ReadIndex2_SSSE3<u16, s16>;
+		tableReadTexCoord[2][4][1] = TexCoord_ReadIndex2_SSSE3<u8, float>;
+		tableReadTexCoord[3][4][1] = TexCoord_ReadIndex2_SSSE3<u16, float>;
 	}
-
 #endif
 
 }
diff --git a/Source/Core/VideoCommon/VertexManagerBase.cpp b/Source/Core/VideoCommon/VertexManagerBase.cpp
index 7a18ba435b..dcd8780ede 100644
--- a/Source/Core/VideoCommon/VertexManagerBase.cpp
+++ b/Source/Core/VideoCommon/VertexManagerBase.cpp
@@ -53,7 +53,8 @@ u32 VertexManager::GetRemainingSize()
 
 void VertexManager::PrepareForAdditionalData(int primitive, u32 count, u32 stride)
 {
-	u32 const needed_vertex_bytes = count * stride;
+	// The SSE vertex loader can write up to 4 bytes past the end
+	u32 const needed_vertex_bytes = count * stride + 4;
 
 	// We can't merge different kinds of primitives, so we have to flush here
 	if (current_primitive_type != primitive_from_gx[primitive])