Merge pull request #1887 from Tilka/vertex_loader_jit

VertexLoader: rewrite x64 JIT

17 files changed, 597 insertions, 221 deletions

diff --git a/Source/Core/VideoCommon/CMakeLists.txt b/Source/Core/VideoCommon/CMakeLists.txt
index 2101a719c2..ceb83564f1 100644
--- a/Source/Core/VideoCommon/CMakeLists.txt
+++ b/Source/Core/VideoCommon/CMakeLists.txt
@@ -45,7 +45,7 @@ set(SRCS	BoundingBox.cpp
 set(LIBS core png)
 
 if(_M_X86)
-	set(SRCS ${SRCS}	TextureDecoder_x64.cpp)
+	set(SRCS ${SRCS}	TextureDecoder_x64.cpp VertexLoaderX64.cpp)
 else()
 	set(SRCS ${SRCS}	TextureDecoder_Generic.cpp)
 endif()
diff --git a/Source/Core/VideoCommon/CPMemory.h b/Source/Core/VideoCommon/CPMemory.h
index 124c30a04a..51e1ccc74f 100644
--- a/Source/Core/VideoCommon/CPMemory.h
+++ b/Source/Core/VideoCommon/CPMemory.h
@@ -20,10 +20,12 @@ enum
 // Vertex components
 enum
 {
-	NOT_PRESENT = 0,
-	DIRECT      = 1,
-	INDEX8      = 2,
-	INDEX16     = 3,
+	NOT_PRESENT  = 0,
+	DIRECT       = 1,
+	INDEX8       = 2,
+	INDEX16      = 3,
+
+	MASK_INDEXED = 2,
 };
 
 enum
diff --git a/Source/Core/VideoCommon/DataReader.h b/Source/Core/VideoCommon/DataReader.h
index f7e030739f..b27d513781 100644
--- a/Source/Core/VideoCommon/DataReader.h
+++ b/Source/Core/VideoCommon/DataReader.h
@@ -15,9 +15,9 @@ public:
 	__forceinline DataReader(u8* src, u8* _end)
 	: buffer(src), end(_end) {}
 
-	__forceinline void WritePointer(u8** src)
+	__forceinline u8* GetPointer()
 	{
-		*src = buffer;
+		return buffer;
 	}
 
 	__forceinline u8* operator=(u8* src)
diff --git a/Source/Core/VideoCommon/OpcodeDecoding.cpp b/Source/Core/VideoCommon/OpcodeDecoding.cpp
index 34b821605f..0c1083b9c2 100644
--- a/Source/Core/VideoCommon/OpcodeDecoding.cpp
+++ b/Source/Core/VideoCommon/OpcodeDecoding.cpp
@@ -137,7 +137,7 @@ u8* OpcodeDecoder_Run(DataReader src, u32* cycles, bool in_display_list)
 	u8* opcodeStart;
 	while (true)
 	{
-		src.WritePointer(&opcodeStart);
+		opcodeStart = src.GetPointer();
 
 		if (!src.size())
 			goto end;
@@ -301,7 +301,7 @@ u8* OpcodeDecoder_Run(DataReader src, u32* cycles, bool in_display_list)
 		if (!is_preprocess && g_bRecordFifoData && cmd_byte != GX_CMD_CALL_DL)
 		{
 			u8* opcodeEnd;
-			src.WritePointer(&opcodeEnd);
+			opcodeEnd = src.GetPointer();
 			FifoRecorder::GetInstance().WriteGPCommand(opcodeStart, u32(opcodeEnd - opcodeStart));
 		}
 	}
diff --git a/Source/Core/VideoCommon/VertexLoader.cpp b/Source/Core/VideoCommon/VertexLoader.cpp
index e2a2073ccf..734ea9a4a6 100644
--- a/Source/Core/VideoCommon/VertexLoader.cpp
+++ b/Source/Core/VideoCommon/VertexLoader.cpp
@@ -4,8 +4,6 @@
 
 #include "Common/CommonTypes.h"
 #include "Common/MemoryUtil.h"
-#include "Common/x64ABI.h"
-#include "Common/x64Emitter.h"
 
 #include "Core/Host.h"
 
@@ -21,9 +19,6 @@
 #include "VideoCommon/VideoCommon.h"
 #include "VideoCommon/VideoConfig.h"
 
-
-#define COMPILED_CODE_SIZE 4096
-
 #ifndef _WIN32
 	#undef inline
 	#define inline
@@ -33,9 +28,6 @@
 u8* g_video_buffer_read_ptr;
 u8* g_vertex_manager_write_ptr;
 
-using namespace Gen;
-
-
 void* VertexLoader::operator new (size_t size)
 {
 	return AllocateAlignedMemory(size, 16);
@@ -48,14 +40,9 @@ void VertexLoader::operator delete (void *p)
 
 static void LOADERDECL PosMtx_ReadDirect_UByte(VertexLoader* loader)
 {
-	BoundingBox::posMtxIdx = loader->m_curposmtx = DataReadU8() & 0x3f;
-	PRIM_LOG("posmtx: %d, ", loader->m_curposmtx);
-}
-
-static void LOADERDECL PosMtx_Write(VertexLoader* loader)
-{
-	// u8, 0, 0, 0
-	DataWrite<u32>(loader->m_curposmtx);
+	u8 posmtx = BoundingBox::posMtxIdx = DataReadU8() & 0x3f;
+	DataWrite<u32>(posmtx);
+	PRIM_LOG("posmtx: %d, ", posmtx);
 }
 
 static void LOADERDECL TexMtx_ReadDirect_UByte(VertexLoader* loader)
@@ -77,18 +64,16 @@ static void LOADERDECL TexMtx_Write_Float2(VertexLoader* loader)
 	DataWrite(float(loader->m_curtexmtx[loader->m_texmtxwrite++]));
 }
 
-static void LOADERDECL TexMtx_Write_Float4(VertexLoader* loader)
+static void LOADERDECL TexMtx_Write_Float3(VertexLoader* loader)
 {
 #if _M_SSE >= 0x200
 	__m128 output = _mm_cvtsi32_ss(_mm_castsi128_ps(_mm_setzero_si128()), loader->m_curtexmtx[loader->m_texmtxwrite++]);
 	_mm_storeu_ps((float*)g_vertex_manager_write_ptr, _mm_shuffle_ps(output, output, 0x45 /* 1, 1, 0, 1 */));
-	g_vertex_manager_write_ptr += sizeof(float) * 4;
+	g_vertex_manager_write_ptr += sizeof(float) * 3;
 #else
 	DataWrite(0.f);
 	DataWrite(0.f);
 	DataWrite(float(loader->m_curtexmtx[loader->m_texmtxwrite++]));
-	// Just to fill out with 0.
-	DataWrite(0.f);
 #endif
 }
 
@@ -106,19 +91,11 @@ static void LOADERDECL SkipVertex(VertexLoader* loader)
 VertexLoader::VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr)
 : VertexLoaderBase(vtx_desc, vtx_attr)
 {
-	m_compiledCode = nullptr;
 	VertexLoader_Normal::Init();
 	VertexLoader_Position::Init();
 	VertexLoader_TextCoord::Init();
 
-	#ifdef USE_VERTEX_LOADER_JIT
-	AllocCodeSpace(COMPILED_CODE_SIZE);
 	CompileVertexTranslator();
-	WriteProtect();
-	#else
-	m_numPipelineStages = 0;
-	CompileVertexTranslator();
-	#endif
 
 	// generate frac factors
 	m_posScale[0] = m_posScale[1] = m_posScale[2] = m_posScale[3] = 1.0f / (1U << m_VtxAttr.PosFrac);
@@ -129,55 +106,13 @@ VertexLoader::VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr)
 		m_colElements[i] = m_VtxAttr.color[i].Elements;
 }
 
-VertexLoader::~VertexLoader()
-{
-	#ifdef USE_VERTEX_LOADER_JIT
-	FreeCodeSpace();
-	#endif
-}
-
 void VertexLoader::CompileVertexTranslator()
 {
 	m_VertexSize = 0;
 	const TVtxAttr &vtx_attr = m_VtxAttr;
 
-#ifdef USE_VERTEX_LOADER_JIT
-	if (m_compiledCode)
-		PanicAlert("Trying to recompile a vertex translator");
-
-	m_compiledCode = GetCodePtr();
-	// We only use RAX (caller saved) and RBX (callee saved).
-	ABI_PushRegistersAndAdjustStack({RBX, RBP}, 8);
-
-	// save count
-	MOV(64, R(RBX), R(ABI_PARAM1));
-
-	// save loader
-	MOV(64, R(RBP), R(ABI_PARAM2));
-
-	// Start loop here
-	const u8 *loop_start = GetCodePtr();
-
-	// Reset component counters if present in vertex format only.
-	if (m_VtxDesc.Tex0Coord || m_VtxDesc.Tex1Coord || m_VtxDesc.Tex2Coord || m_VtxDesc.Tex3Coord ||
-		m_VtxDesc.Tex4Coord || m_VtxDesc.Tex5Coord || m_VtxDesc.Tex6Coord || m_VtxDesc.Tex7Coord)
-	{
-		WriteSetVariable(32, &m_tcIndex, Imm32(0));
-	}
-	if (m_VtxDesc.Color0 || m_VtxDesc.Color1)
-	{
-		WriteSetVariable(32, &m_colIndex, Imm32(0));
-	}
-	if (m_VtxDesc.Tex0MatIdx || m_VtxDesc.Tex1MatIdx || m_VtxDesc.Tex2MatIdx || m_VtxDesc.Tex3MatIdx ||
-		m_VtxDesc.Tex4MatIdx || m_VtxDesc.Tex5MatIdx || m_VtxDesc.Tex6MatIdx || m_VtxDesc.Tex7MatIdx)
-	{
-		WriteSetVariable(32, &m_texmtxwrite, Imm32(0));
-		WriteSetVariable(32, &m_texmtxread, Imm32(0));
-	}
-#else
 	// Reset pipeline
 	m_numPipelineStages = 0;
-#endif
 
 	// Get the pointer to this vertex's buffer data for the bounding box
 	if (!g_ActiveConfig.backend_info.bSupportsBBox)
@@ -195,16 +130,22 @@ void VertexLoader::CompileVertexTranslator()
 
 	// Position in pc vertex format.
 	int nat_offset = 0;
-	memset(&m_native_vtx_decl, 0, sizeof(m_native_vtx_decl));
 
 	// Position Matrix Index
 	if (m_VtxDesc.PosMatIdx)
 	{
 		WriteCall(PosMtx_ReadDirect_UByte);
 		components |= VB_HAS_POSMTXIDX;
+		m_native_vtx_decl.posmtx.components = 4;
+		m_native_vtx_decl.posmtx.enable = true;
+		m_native_vtx_decl.posmtx.offset = nat_offset;
+		m_native_vtx_decl.posmtx.type = VAR_UNSIGNED_BYTE;
+		m_native_vtx_decl.posmtx.integer = true;
+		nat_offset += 4;
 		m_VertexSize += 1;
 	}
 
+
 	if (m_VtxDesc.Tex0MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX0; WriteCall(TexMtx_ReadDirect_UByte); }
 	if (m_VtxDesc.Tex1MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX1; WriteCall(TexMtx_ReadDirect_UByte); }
 	if (m_VtxDesc.Tex2MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX2; WriteCall(TexMtx_ReadDirect_UByte); }
@@ -325,11 +266,7 @@ void VertexLoader::CompileVertexTranslator()
 		const int format = m_VtxAttr.texCoord[i].Format;
 		const int elements = m_VtxAttr.texCoord[i].Elements;
 
-		if (tc[i] == NOT_PRESENT)
-		{
-			components &= ~(VB_HAS_UV0 << i);
-		}
-		else
+		if (tc[i] != NOT_PRESENT)
 		{
 			_assert_msg_(VIDEO, DIRECT <= tc[i] && tc[i] <= INDEX16, "Invalid texture coordinates!\n(tc[i] = %d)", (u32)tc[i]);
 			_assert_msg_(VIDEO, FORMAT_UBYTE <= format && format <= FORMAT_FLOAT, "Invalid texture coordinates format!\n(format = %d)", format);
@@ -353,9 +290,9 @@ void VertexLoader::CompileVertexTranslator()
 			else
 			{
 				components |= VB_HAS_UV0 << i; // have to include since using now
-				m_native_vtx_decl.texcoords[i].components = 4;
-				nat_offset += 16; // still include the texture coordinate, but this time as 6 + 2 bytes
-				WriteCall(TexMtx_Write_Float4);
+				m_native_vtx_decl.texcoords[i].components = 3;
+				nat_offset += 12;
+				WriteCall(TexMtx_Write_Float3);
 			}
 		}
 		else
@@ -393,17 +330,6 @@ void VertexLoader::CompileVertexTranslator()
 	if (!g_ActiveConfig.backend_info.bSupportsBBox)
 		WriteCall(BoundingBox::Update);
 
-	if (m_VtxDesc.PosMatIdx)
-	{
-		WriteCall(PosMtx_Write);
-		m_native_vtx_decl.posmtx.components = 4;
-		m_native_vtx_decl.posmtx.enable = true;
-		m_native_vtx_decl.posmtx.offset = nat_offset;
-		m_native_vtx_decl.posmtx.type = VAR_UNSIGNED_BYTE;
-		m_native_vtx_decl.posmtx.integer = true;
-		nat_offset += 4;
-	}
-
 	// indexed position formats may skip a the vertex
 	if (m_VtxDesc.Position & 2)
 	{
@@ -412,49 +338,17 @@ void VertexLoader::CompileVertexTranslator()
 
 	m_native_components = components;
 	m_native_vtx_decl.stride = nat_offset;
-
-#ifdef USE_VERTEX_LOADER_JIT
-	// End loop here
-	SUB(64, R(RBX), Imm8(1));
-
-	J_CC(CC_NZ, loop_start);
-	ABI_PopRegistersAndAdjustStack({RBX, RBP}, 8);
-	RET();
-#endif
 }
 
 void VertexLoader::WriteCall(TPipelineFunction func)
 {
-#ifdef USE_VERTEX_LOADER_JIT
-	MOV(64, R(ABI_PARAM1), R(RBP));
-	ABI_CallFunction((const void*)func);
-#else
 	m_PipelineStages[m_numPipelineStages++] = func;
-#endif
-}
-// ARMTODO: This should be done in a better way
-#ifndef _M_GENERIC
-void VertexLoader::WriteGetVariable(int bits, OpArg dest, void *address)
-{
-#ifdef USE_VERTEX_LOADER_JIT
-	MOV(64, R(RAX), Imm64((u64)address));
-	MOV(bits, dest, MatR(RAX));
-#endif
-}
-
-void VertexLoader::WriteSetVariable(int bits, void *address, OpArg value)
-{
-#ifdef USE_VERTEX_LOADER_JIT
-	MOV(64, R(RAX), Imm64((u64)address));
-	MOV(bits, MatR(RAX), value);
-#endif
 }
-#endif
 
 int VertexLoader::RunVertices(int primitive, int count, DataReader src, DataReader dst)
 {
-	dst.WritePointer(&g_vertex_manager_write_ptr);
-	src.WritePointer(&g_video_buffer_read_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
+	g_video_buffer_read_ptr = src.GetPointer();
 
 	m_numLoadedVertices += count;
 	m_skippedVertices = 0;
@@ -463,12 +357,6 @@ int VertexLoader::RunVertices(int primitive, int count, DataReader src, DataRead
 	if (!g_ActiveConfig.backend_info.bSupportsBBox)
 		BoundingBox::Prepare(m_vat, primitive, m_VtxDesc, m_native_vtx_decl);
 
-#ifdef USE_VERTEX_LOADER_JIT
-	if (count > 0)
-	{
-		((void (*)(int, VertexLoader* loader))(void*)m_compiledCode)(count, this);
-	}
-#else
 	for (int s = 0; s < count; s++)
 	{
 		m_tcIndex = 0;
@@ -478,7 +366,6 @@ int VertexLoader::RunVertices(int primitive, int count, DataReader src, DataRead
 			m_PipelineStages[i](this);
 		PRIM_LOG("\n");
 	}
-#endif
 
 	return count - m_skippedVertices;
 }
diff --git a/Source/Core/VideoCommon/VertexLoader.h b/Source/Core/VideoCommon/VertexLoader.h
index 4c883a3242..a7acd5dba1 100644
--- a/Source/Core/VideoCommon/VertexLoader.h
+++ b/Source/Core/VideoCommon/VertexLoader.h
@@ -11,7 +11,6 @@
 #include <string>
 
 #include "Common/CommonTypes.h"
-#include "Common/x64Emitter.h"
 
 #include "VideoCommon/CPMemory.h"
 #include "VideoCommon/DataReader.h"
@@ -26,10 +25,6 @@
 #include <tmmintrin.h>
 #endif
 
-#ifdef _M_X86
-#define USE_VERTEX_LOADER_JIT
-#endif
-
 #ifdef WIN32
 #define LOADERDECL __cdecl
 #else
@@ -39,12 +34,7 @@
 class VertexLoader;
 typedef void (LOADERDECL *TPipelineFunction)(VertexLoader* loader);
 
-// ARMTODO: This should be done in a better way
-#ifndef _M_GENERIC
-class VertexLoader : public Gen::X64CodeBlock, public VertexLoaderBase
-#else
 class VertexLoader : public VertexLoaderBase
-#endif
 {
 public:
 	// This class need a 16 byte alignment. As this is broken on
@@ -53,7 +43,6 @@ public:
 	void operator delete (void *p);
 
 	VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr);
-	~VertexLoader();
 
 	int RunVertices(int primitive, int count, DataReader src, DataReader dst) override;
 	std::string GetName() const override { return "OldLoader"; }
@@ -69,7 +58,6 @@ public:
 
 	// Matrix components are first in GC format but later in PC format - we need to store it temporarily
 	// when decoding each vertex.
-	u8 m_curposmtx;
 	u8 m_curtexmtx[8];
 	int m_texmtxwrite;
 	int m_texmtxread;
@@ -77,22 +65,13 @@ public:
 	int m_skippedVertices;
 
 private:
-#ifndef USE_VERTEX_LOADER_JIT
 	// Pipeline.
 	TPipelineFunction m_PipelineStages[64];  // TODO - figure out real max. it's lower.
 	int m_numPipelineStages;
-#endif
 
 	void CompileVertexTranslator();
 
 	void WriteCall(TPipelineFunction);
-
-#ifndef _M_GENERIC
-	void WriteGetVariable(int bits, Gen::OpArg dest, void *address);
-	void WriteSetVariable(int bits, void *address, Gen::OpArg dest);
-#endif
-
-	const u8 *m_compiledCode;
 };
 
 #if _M_SSE >= 0x301
diff --git a/Source/Core/VideoCommon/VertexLoaderBase.cpp b/Source/Core/VideoCommon/VertexLoaderBase.cpp
index ed56db0655..815e162689 100644
--- a/Source/Core/VideoCommon/VertexLoaderBase.cpp
+++ b/Source/Core/VideoCommon/VertexLoaderBase.cpp
@@ -2,6 +2,7 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
+#include <cinttypes>
 #include <vector>
 
 #include "Common/StringUtil.h"
@@ -9,11 +10,17 @@
 #include "VideoCommon/VertexLoader.h"
 #include "VideoCommon/VertexLoaderBase.h"
 
+#ifdef _M_X86_64
+#include "VideoCommon/VertexLoaderX64.h"
+#endif
+
 VertexLoaderBase::VertexLoaderBase(const TVtxDesc &vtx_desc, const VAT &vtx_attr)
 {
 	m_numLoadedVertices = 0;
 	m_VertexSize = 0;
 	m_native_vertex_format = nullptr;
+	m_native_components = 0;
+	memset(&m_native_vtx_decl, 0, sizeof(m_native_vtx_decl));
 
 	SetVAT(vtx_attr);
 	m_VtxDesc = vtx_desc;
@@ -131,15 +138,34 @@ class VertexLoaderTester : public VertexLoaderBase
 {
 public:
 	VertexLoaderTester(VertexLoaderBase* _a, VertexLoaderBase* _b, const TVtxDesc& vtx_desc, const VAT& vtx_attr)
-	: VertexLoaderBase(vtx_desc, vtx_attr)
+	: VertexLoaderBase(vtx_desc, vtx_attr), a(_a), b(_b)
 	{
-		a = _a;
-		b = _b;
 		m_initialized = a && b && a->IsInitialized() && b->IsInitialized();
-		m_initialized = m_initialized && (a->m_VertexSize == b->m_VertexSize);
-		m_initialized = m_initialized && (a->m_native_vtx_decl.stride == b->m_native_vtx_decl.stride);
+		bool can_test = a->m_VertexSize == b->m_VertexSize &&
+		                a->m_native_components == b->m_native_components &&
+		                a->m_native_vtx_decl.stride == b->m_native_vtx_decl.stride;
+
+		if (m_initialized)
+		{
+			if (can_test)
+			{
+				m_VertexSize = a->m_VertexSize;
+				m_native_components = a->m_native_components;
+				memcpy(&m_native_vtx_decl, &a->m_native_vtx_decl, sizeof(PortableVertexDeclaration));
+			}
+			else
+			{
+				ERROR_LOG(VIDEO, "Can't compare vertex loaders that expect different vertex formats!");
+				ERROR_LOG(VIDEO, "a: m_VertexSize %d, m_native_components 0x%08x, stride %d\n",
+				                 a->m_VertexSize, a->m_native_components, a->m_native_vtx_decl.stride);
+				ERROR_LOG(VIDEO, "b: m_VertexSize %d, m_native_components 0x%08x, stride %d\n",
+				                 b->m_VertexSize, b->m_native_components, b->m_native_vtx_decl.stride);
+			}
+		}
+
+		m_initialized &= can_test;
 	}
-	~VertexLoaderTester()
+	~VertexLoaderTester() override
 	{
 		delete a;
 		delete b;
@@ -147,21 +173,22 @@ public:
 
 	int RunVertices(int primitive, int count, DataReader src, DataReader dst) override
 	{
-		buffer_a.resize(count * a->m_native_vtx_decl.stride);
-		buffer_b.resize(count * b->m_native_vtx_decl.stride);
+		buffer_a.resize(count * a->m_native_vtx_decl.stride + 4);
+		buffer_b.resize(count * b->m_native_vtx_decl.stride + 4);
 
 		int count_a = a->RunVertices(primitive, count, src, DataReader(buffer_a.data(), buffer_a.data()+buffer_a.size()));
 		int count_b = b->RunVertices(primitive, count, src, DataReader(buffer_b.data(), buffer_b.data()+buffer_b.size()));
 
 		if (count_a != count_b)
-			ERROR_LOG(VIDEO, "Both vertexloaders have loaded a different amount of vertices.");
+			ERROR_LOG(VIDEO, "The two vertex loaders have loaded a different amount of vertices (a: %d, b: %d).", count_a, count_b);
 
-		if (memcmp(buffer_a.data(), buffer_b.data(), std::min(count_a, count_b)))
-			ERROR_LOG(VIDEO, "Both vertexloaders have loaded different data.");
+		if (memcmp(buffer_a.data(), buffer_b.data(), std::min(count_a, count_b) * m_native_vtx_decl.stride))
+			ERROR_LOG(VIDEO, "The two vertex loaders have loaded different data "
+			                 "(guru meditation 0x%016" PRIx64 ", 0x%08x, 0x%08x, 0x%08x).",
+			                 m_VtxDesc.Hex, m_vat.g0.Hex, m_vat.g1.Hex, m_vat.g2.Hex);
 
-		u8* dstptr;
-		dst.WritePointer(&dstptr);
-		memcpy(dstptr, buffer_a.data(), count_a);
+		memcpy(dst.GetPointer(), buffer_a.data(), count_a * m_native_vtx_decl.stride);
+		m_numLoadedVertices += count;
 		return count_a;
 	}
 	std::string GetName() const override { return "CompareLoader"; }
@@ -177,15 +204,22 @@ VertexLoaderBase* VertexLoaderBase::CreateVertexLoader(const TVtxDesc& vtx_desc,
 {
 	VertexLoaderBase* loader;
 
-#if 0
+//#define COMPARE_VERTEXLOADERS
+
+#if defined(COMPARE_VERTEXLOADERS) && defined(_M_X86_64)
 	// first try: Any new VertexLoader vs the old one
 	loader = new VertexLoaderTester(
 			new VertexLoader(vtx_desc, vtx_attr), // the software one
-			new VertexLoader(vtx_desc, vtx_attr), // the new one to compare
+			new VertexLoaderX64(vtx_desc, vtx_attr), // the new one to compare
 			vtx_desc, vtx_attr);
 	if (loader->IsInitialized())
 		return loader;
 	delete loader;
+#elif defined(_M_X86_64)
+	loader = new VertexLoaderX64(vtx_desc, vtx_attr);
+	if (loader->IsInitialized())
+		return loader;
+	delete loader;
 #endif
 
 	// last try: The old VertexLoader
diff --git a/Source/Core/VideoCommon/VertexLoaderBase.h b/Source/Core/VideoCommon/VertexLoaderBase.h
index 7da2d866f9..c5a11d2e63 100644
--- a/Source/Core/VideoCommon/VertexLoaderBase.h
+++ b/Source/Core/VideoCommon/VertexLoaderBase.h
@@ -72,7 +72,7 @@ class VertexLoaderBase
 {
 public:
 	static VertexLoaderBase* CreateVertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr);
-	virtual ~VertexLoaderBase() {};
+	virtual ~VertexLoaderBase() {}
 
 	virtual int RunVertices(int primitive, int count, DataReader src, DataReader dst) = 0;
 
diff --git a/Source/Core/VideoCommon/VertexLoaderManager.cpp b/Source/Core/VideoCommon/VertexLoaderManager.cpp
index b95e286c1e..786cb45a96 100644
--- a/Source/Core/VideoCommon/VertexLoaderManager.cpp
+++ b/Source/Core/VideoCommon/VertexLoaderManager.cpp
@@ -121,11 +121,10 @@ static VertexLoaderBase* RefreshLoader(int vtx_attr_group, bool preprocess = fal
 		{
 			// search for a cached native vertex format
 			const PortableVertexDeclaration& format = loader->m_native_vtx_decl;
-			auto& native = s_native_vertex_map[format];
+			std::unique_ptr<NativeVertexFormat>& native = s_native_vertex_map[format];
 			if (!native)
 			{
-				auto raw_pointer = g_vertex_manager->CreateNativeVertexFormat();
-				native = std::unique_ptr<NativeVertexFormat>(raw_pointer);
+				native.reset(g_vertex_manager->CreateNativeVertexFormat());
 				native->Initialize(format);
 				native->m_components = loader->m_native_components;
 			}
diff --git a/Source/Core/VideoCommon/VertexLoaderX64.cpp b/Source/Core/VideoCommon/VertexLoaderX64.cpp
new file mode 100644
index 0000000000..c9b435bae3
--- /dev/null
+++ b/Source/Core/VideoCommon/VertexLoaderX64.cpp
@@ -0,0 +1,456 @@
+#ifdef _MSC_VER
+#include <intrin.h>
+#else
+#include <x86intrin.h>
+#endif
+
+#include "Common/CPUDetect.h"
+#include "Common/JitRegister.h"
+#include "Common/x64ABI.h"
+#include "VideoCommon/VertexLoaderX64.h"
+
+using namespace Gen;
+
+static const X64Reg src_reg = ABI_PARAM1;
+static const X64Reg dst_reg = ABI_PARAM2;
+static const X64Reg scratch1 = RAX;
+static const X64Reg scratch2 = ABI_PARAM3;
+static const X64Reg scratch3 = ABI_PARAM4;
+static const X64Reg count_reg = R10;
+static const X64Reg skipped_reg = R11;
+
+VertexLoaderX64::VertexLoaderX64(const TVtxDesc& vtx_desc, const VAT& vtx_att): VertexLoaderBase(vtx_desc, vtx_att)
+{
+	if (!IsInitialized())
+		return;
+
+	AllocCodeSpace(4096);
+	ClearCodeSpace();
+	GenerateVertexLoader();
+	WriteProtect();
+
+	std::string name;
+	AppendToString(&name);
+	JitRegister::Register(region, (u32)(GetCodePtr() - region), name.c_str());
+}
+
+OpArg VertexLoaderX64::GetVertexAddr(int array, u64 attribute)
+{
+	OpArg data = MDisp(src_reg, m_src_ofs);
+	if (attribute & MASK_INDEXED)
+	{
+		if (attribute == INDEX8)
+		{
+			MOVZX(64, 8, scratch1, data);
+			m_src_ofs += 1;
+		}
+		else
+		{
+			MOV(16, R(scratch1), data);
+			m_src_ofs += 2;
+			BSWAP(16, scratch1);
+			MOVZX(64, 16, scratch1, R(scratch1));
+		}
+		if (array == ARRAY_POSITION)
+		{
+			CMP(attribute == INDEX8 ? 8 : 16, R(scratch1), Imm8(-1));
+			m_skip_vertex = J_CC(CC_E, true);
+		}
+		// TODO: Move cached_arraybases into CPState and use MDisp() relative to a constant register loaded with &g_main_cp_state.
+		IMUL(32, scratch1, M(&g_main_cp_state.array_strides[array]));
+		MOV(64, R(scratch2), M(&cached_arraybases[array]));
+		return MRegSum(scratch1, scratch2);
+	}
+	else
+	{
+		return data;
+	}
+}
+
+int VertexLoaderX64::ReadVertex(OpArg data, u64 attribute, int format, int count_in, int count_out, u8 scaling_exponent, AttributeFormat* native_format)
+{
+	static const __m128i shuffle_lut[5][3] = {
+		{_mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFF00L),  // 1x u8
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFF01L, 0xFFFFFF00L),  // 2x u8
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFF02L, 0xFFFFFF01L, 0xFFFFFF00L)}, // 3x u8
+		{_mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0x00FFFFFFL),  // 1x s8
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL),  // 2x s8
+		 _mm_set_epi32(0xFFFFFFFFL, 0x02FFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL)}, // 3x s8
+		{_mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFF0001L),  // 1x u16
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFF0203L, 0xFFFF0001L),  // 2x u16
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFF0405L, 0xFFFF0203L, 0xFFFF0001L)}, // 3x u16
+		{_mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0x0001FFFFL),  // 1x s16
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x0203FFFFL, 0x0001FFFFL),  // 2x s16
+		 _mm_set_epi32(0xFFFFFFFFL, 0x0405FFFFL, 0x0203FFFFL, 0x0001FFFFL)}, // 3x s16
+		{_mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFFFFL, 0x00010203L),  // 1x float
+		 _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x04050607L, 0x00010203L),  // 2x float
+		 _mm_set_epi32(0xFFFFFFFFL, 0x08090A0BL, 0x04050607L, 0x00010203L)}, // 3x float
+	};
+	static const __m128 scale_factors[32] = {
+		_mm_set_ps1(1./(1u<< 0)), _mm_set_ps1(1./(1u<< 1)), _mm_set_ps1(1./(1u<< 2)), _mm_set_ps1(1./(1u<< 3)),
+		_mm_set_ps1(1./(1u<< 4)), _mm_set_ps1(1./(1u<< 5)), _mm_set_ps1(1./(1u<< 6)), _mm_set_ps1(1./(1u<< 7)),
+		_mm_set_ps1(1./(1u<< 8)), _mm_set_ps1(1./(1u<< 9)), _mm_set_ps1(1./(1u<<10)), _mm_set_ps1(1./(1u<<11)),
+		_mm_set_ps1(1./(1u<<12)), _mm_set_ps1(1./(1u<<13)), _mm_set_ps1(1./(1u<<14)), _mm_set_ps1(1./(1u<<15)),
+		_mm_set_ps1(1./(1u<<16)), _mm_set_ps1(1./(1u<<17)), _mm_set_ps1(1./(1u<<18)), _mm_set_ps1(1./(1u<<19)),
+		_mm_set_ps1(1./(1u<<20)), _mm_set_ps1(1./(1u<<21)), _mm_set_ps1(1./(1u<<22)), _mm_set_ps1(1./(1u<<23)),
+		_mm_set_ps1(1./(1u<<24)), _mm_set_ps1(1./(1u<<25)), _mm_set_ps1(1./(1u<<26)), _mm_set_ps1(1./(1u<<27)),
+		_mm_set_ps1(1./(1u<<28)), _mm_set_ps1(1./(1u<<29)), _mm_set_ps1(1./(1u<<30)), _mm_set_ps1(1./(1u<<31)),
+	};
+
+	X64Reg coords = XMM0;
+
+	int elem_size = 1 << (format / 2);
+	int load_bytes  = elem_size * count_in;
+	if (load_bytes >= 8)
+		MOVDQU(coords, data);
+	else if (load_bytes >= 4)
+		MOVQ_xmm(coords, data);
+	else
+		MOVD_xmm(coords, data);
+
+	PSHUFB(coords, M(&shuffle_lut[format][count_in - 1]));
+
+	if (format != FORMAT_FLOAT)
+	{
+		// Sign extend
+		if (format == FORMAT_BYTE)
+			PSRAD(coords, 24);
+		if (format == FORMAT_SHORT)
+			PSRAD(coords, 16);
+
+		CVTDQ2PS(coords, R(coords));
+
+		if (scaling_exponent)
+			MULPS(coords, M(&scale_factors[scaling_exponent]));
+	}
+
+	OpArg dest = MDisp(dst_reg, m_dst_ofs);
+	switch (count_out)
+	{
+		case 1: MOVSS(dest, coords); break;
+		case 2: MOVLPS(dest, coords); break;
+		case 3: MOVUPS(dest, coords); break;
+	}
+
+	native_format->components = count_out;
+	native_format->enable = true;
+	native_format->offset = m_dst_ofs;
+	native_format->type = VAR_FLOAT;
+	native_format->integer = false;
+	m_dst_ofs += sizeof(float) * count_out;
+
+	if (attribute == DIRECT)
+		m_src_ofs += load_bytes;
+
+	return load_bytes;
+}
+
+void VertexLoaderX64::ReadColor(OpArg data, u64 attribute, int format, int elements)
+{
+	int load_bytes = 0;
+	switch (format)
+	{
+		case FORMAT_24B_888:
+		case FORMAT_32B_888x:
+		case FORMAT_32B_8888:
+			MOV(32, R(scratch1), data);
+			// See VertexLoader_Color.cpp for a comment on this condition.
+			if (format != FORMAT_32B_8888 || !elements)
+				OR(32, R(scratch1), Imm32(0xFF000000));
+			MOV(32, MDisp(dst_reg, m_dst_ofs), R(scratch1));
+			load_bytes = 3 + (format != FORMAT_24B_888);
+			break;
+
+		case FORMAT_16B_565:
+			//                   RRRRRGGG GGGBBBBB
+			// AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
+			LoadAndSwap(16, scratch1, data);
+			if (cpu_info.bBMI1 && cpu_info.bBMI2)
+			{
+				MOV(32, R(scratch2), Imm32(0x07C3F7C0));
+				PDEP(32, scratch3, scratch1, R(scratch2));
+
+				MOV(32, R(scratch2), Imm32(0xF8FCF800));
+				PDEP(32, scratch1, scratch1, R(scratch2));
+				ANDN(32, scratch2, scratch2, R(scratch3));
+
+				OR(32, R(scratch1), R(scratch2));
+			}
+			else
+			{
+				MOV(32, R(scratch3), R(scratch1));
+				SHL(32, R(scratch1), Imm8(16));
+				AND(32, R(scratch1), Imm32(0xF8000000));
+
+				MOV(32, R(scratch2), R(scratch3));
+				SHL(32, R(scratch2), Imm8(13));
+				AND(32, R(scratch2), Imm32(0x00FC0000));
+				OR(32, R(scratch1), R(scratch2));
+
+				SHL(32, R(scratch3), Imm8(11));
+				AND(32, R(scratch3), Imm32(0x0000F800));
+				OR(32, R(scratch1), R(scratch3));
+
+				MOV(32, R(scratch2), R(scratch1));
+				SHR(32, R(scratch1), Imm8(5));
+				AND(32, R(scratch1), Imm32(0x07000700));
+				OR(32, R(scratch1), R(scratch2));
+
+				SHR(32, R(scratch2), Imm8(6));
+				AND(32, R(scratch2), Imm32(0x00030000));
+				OR(32, R(scratch1), R(scratch2));
+			}
+
+			OR(32, R(scratch1), Imm32(0x000000FF));
+			SwapAndStore(32, MDisp(dst_reg, m_dst_ofs), scratch1);
+			load_bytes = 2;
+			break;
+
+		case FORMAT_16B_4444:
+			//                   RRRRGGGG BBBBAAAA
+			// AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
+			LoadAndSwap(16, scratch1, data);
+			if (cpu_info.bBMI2)
+			{
+				MOV(32, R(scratch3), Imm32(0x0F0F0F0F));
+				PDEP(32, scratch2, scratch1, R(scratch3));
+				MOV(32, R(scratch3), Imm32(0xF0F0F0F0));
+				PDEP(32, scratch1, scratch1, R(scratch3));
+			}
+			else
+			{
+				MOV(32, R(scratch3), R(scratch1));
+				SHL(32, R(scratch1), Imm8(12));
+				AND(32, R(scratch1), Imm32(0x0F000000));
+				MOV(32, R(scratch2), R(scratch1));
+
+				MOV(32, R(scratch1), R(scratch3));
+				SHL(32, R(scratch1), Imm8(8));
+				AND(32, R(scratch1), Imm32(0x000F0000));
+				OR(32, R(scratch2), R(scratch1));
+
+				MOV(32, R(scratch1), R(scratch3));
+				SHL(32, R(scratch1), Imm8(4));
+				AND(32, R(scratch1), Imm32(0x00000F00));
+				OR(32, R(scratch2), R(scratch1));
+
+				AND(32, R(scratch3), Imm8(0x0F));
+				OR(32, R(scratch2), R(scratch3));
+
+				MOV(32, R(scratch1), R(scratch2));
+				SHL(32, R(scratch1), Imm8(4));
+			}
+			OR(32, R(scratch1), R(scratch2));
+			SwapAndStore(32, MDisp(dst_reg, m_dst_ofs), scratch1);
+			load_bytes = 2;
+			break;
+
+		case FORMAT_24B_6666:
+			//          RRRRRRGG GGGGBBBB BBAAAAAA
+			// AAAAAAAA BBBBBBBB GGGGGGGG RRRRRRRR
+			data.offset -= 1;
+			LoadAndSwap(32, scratch1, data);
+			if (cpu_info.bBMI2)
+			{
+				MOV(32, R(scratch2), Imm32(0xFCFCFCFC));
+				PDEP(32, scratch1, scratch1, R(scratch2));
+				MOV(32, R(scratch2), R(scratch1));
+			}
+			else
+			{
+				MOV(32, R(scratch3), R(scratch1));
+				SHL(32, R(scratch1), Imm8(8));
+				AND(32, R(scratch1), Imm32(0xFC000000));
+				MOV(32, R(scratch2), R(scratch1));
+
+				MOV(32, R(scratch1), R(scratch3));
+				SHL(32, R(scratch1), Imm8(6));
+				AND(32, R(scratch1), Imm32(0x00FC0000));
+				OR(32, R(scratch2), R(scratch1));
+
+				MOV(32, R(scratch1), R(scratch3));
+				SHL(32, R(scratch1), Imm8(4));
+				AND(32, R(scratch1), Imm32(0x0000FC00));
+				OR(32, R(scratch2), R(scratch1));
+
+				SHL(32, R(scratch3), Imm8(2));
+				AND(32, R(scratch3), Imm32(0x000000FC));
+				OR(32, R(scratch2), R(scratch3));
+
+				MOV(32, R(scratch1), R(scratch2));
+			}
+
+			SHR(32, R(scratch1), Imm8(6));
+			AND(32, R(scratch1), Imm32(0x03030303));
+			OR(32, R(scratch1), R(scratch2));
+
+			SwapAndStore(32, MDisp(dst_reg, m_dst_ofs), scratch1);
+			load_bytes = 3;
+			break;
+	}
+	if (attribute == DIRECT)
+		m_src_ofs += load_bytes;
+}
+
+void VertexLoaderX64::GenerateVertexLoader()
+{
+	// Backup count since we're going to count it down.
+	PUSH(32, R(ABI_PARAM3));
+
+	// We use ABI_PARAM3 for scratch2.
+	MOV(32, R(count_reg), R(ABI_PARAM3));
+
+	if (m_VtxDesc.Position & MASK_INDEXED)
+		XOR(32, R(skipped_reg), R(skipped_reg));
+
+	// TODO: load constants into registers outside the main loop
+
+	const u8* loop_start = GetCodePtr();
+
+	if (m_VtxDesc.PosMatIdx)
+	{
+		MOVZX(32, 8, scratch1, MDisp(src_reg, m_src_ofs));
+		AND(32, R(scratch1), Imm8(0x3F));
+		MOV(32, MDisp(dst_reg, m_dst_ofs), R(scratch1));
+		m_native_components |= VB_HAS_POSMTXIDX;
+		m_native_vtx_decl.posmtx.components = 4;
+		m_native_vtx_decl.posmtx.enable = true;
+		m_native_vtx_decl.posmtx.offset = m_dst_ofs;
+		m_native_vtx_decl.posmtx.type = VAR_UNSIGNED_BYTE;
+		m_native_vtx_decl.posmtx.integer = true;
+		m_src_ofs += sizeof(u8);
+		m_dst_ofs += sizeof(u32);
+	}
+
+	u32 texmatidx_ofs[8];
+	const u64 tm[8] = {
+		m_VtxDesc.Tex0MatIdx, m_VtxDesc.Tex1MatIdx, m_VtxDesc.Tex2MatIdx, m_VtxDesc.Tex3MatIdx,
+		m_VtxDesc.Tex4MatIdx, m_VtxDesc.Tex5MatIdx, m_VtxDesc.Tex6MatIdx, m_VtxDesc.Tex7MatIdx,
+	};
+	for (int i = 0; i < 8; i++)
+	{
+		if (tm[i])
+			texmatidx_ofs[i] = m_src_ofs++;
+	}
+
+	OpArg data = GetVertexAddr(ARRAY_POSITION, m_VtxDesc.Position);
+	ReadVertex(data, m_VtxDesc.Position, m_VtxAttr.PosFormat, m_VtxAttr.PosElements + 2, 3, m_VtxAttr.PosFrac, &m_native_vtx_decl.position);
+
+	if (m_VtxDesc.Normal)
+	{
+		static const u8 map[8] = {7, 6, 15, 14};
+		u8 scaling_exponent = map[m_VtxAttr.NormalFormat];
+
+		for (int i = 0; i < (m_VtxAttr.NormalElements ? 3 : 1); i++)
+		{
+			if (!i || m_VtxAttr.NormalIndex3)
+			{
+				data = GetVertexAddr(ARRAY_NORMAL, m_VtxDesc.Normal);
+				int elem_size = 1 << (m_VtxAttr.NormalFormat / 2);
+				data.offset += i * elem_size * 3;
+			}
+			data.offset += ReadVertex(data, m_VtxDesc.Normal, m_VtxAttr.NormalFormat, 3, 3, scaling_exponent, &m_native_vtx_decl.normals[i]);
+		}
+
+		m_native_components |= VB_HAS_NRM0;
+		if (m_VtxAttr.NormalElements)
+			m_native_components |= VB_HAS_NRM1 | VB_HAS_NRM2;
+	}
+
+	const u64 col[2] = {m_VtxDesc.Color0, m_VtxDesc.Color1};
+	for (int i = 0; i < 2; i++)
+	{
+		if (col[i])
+		{
+			data = GetVertexAddr(ARRAY_COLOR + i, col[i]);
+			ReadColor(data, col[i], m_VtxAttr.color[i].Comp, m_VtxAttr.color[i].Elements);
+			m_native_components |= VB_HAS_COL0 << i;
+			m_native_vtx_decl.colors[i].components = 4;
+			m_native_vtx_decl.colors[i].enable = true;
+			m_native_vtx_decl.colors[i].offset = m_dst_ofs;
+			m_native_vtx_decl.colors[i].type = VAR_UNSIGNED_BYTE;
+			m_native_vtx_decl.colors[i].integer = false;
+			m_dst_ofs += 4;
+		}
+	}
+
+	const u64 tc[8] = {
+		m_VtxDesc.Tex0Coord, m_VtxDesc.Tex1Coord, m_VtxDesc.Tex2Coord, m_VtxDesc.Tex3Coord,
+		m_VtxDesc.Tex4Coord, m_VtxDesc.Tex5Coord, m_VtxDesc.Tex6Coord, m_VtxDesc.Tex7Coord,
+	};
+	for (int i = 0; i < 8; i++)
+	{
+		int elements = m_VtxAttr.texCoord[i].Elements + 1;
+		if (tc[i])
+		{
+			data = GetVertexAddr(ARRAY_TEXCOORD0 + i, tc[i]);
+			u8 scaling_exponent = m_VtxAttr.texCoord[i].Frac;
+			ReadVertex(data, tc[i], m_VtxAttr.texCoord[i].Format, elements, tm[i] ? 2 : elements, scaling_exponent, &m_native_vtx_decl.texcoords[i]);
+			m_native_components |= VB_HAS_UV0 << i;
+		}
+		if (tm[i])
+		{
+			m_native_components |= VB_HAS_TEXMTXIDX0 << i;
+			m_native_vtx_decl.texcoords[i].components = 3;
+			m_native_vtx_decl.texcoords[i].enable = true;
+			m_native_vtx_decl.texcoords[i].type = VAR_FLOAT;
+			m_native_vtx_decl.texcoords[i].integer = false;
+			MOVZX(64, 8, scratch1, MDisp(src_reg, texmatidx_ofs[i]));
+			if (tc[i])
+			{
+				CVTSI2SS(XMM0, R(scratch1));
+				MOVSS(MDisp(dst_reg, m_dst_ofs), XMM0);
+				m_dst_ofs += sizeof(float);
+			}
+			else
+			{
+				m_native_vtx_decl.texcoords[i].offset = m_dst_ofs;
+				PXOR(XMM0, R(XMM0));
+				CVTSI2SS(XMM0, R(scratch1));
+				SHUFPS(XMM0, R(XMM0), 0x45);
+				MOVUPS(MDisp(dst_reg, m_dst_ofs), XMM0);
+				m_dst_ofs += sizeof(float) * 3;
+			}
+		}
+	}
+
+	// Prepare for the next vertex.
+	ADD(64, R(dst_reg), Imm32(m_dst_ofs));
+	const u8* cont = GetCodePtr();
+	ADD(64, R(src_reg), Imm32(m_src_ofs));
+
+	SUB(32, R(count_reg), Imm8(1));
+	J_CC(CC_NZ, loop_start);
+
+	// Get the original count.
+	POP(32, R(ABI_RETURN));
+
+	if (m_VtxDesc.Position & MASK_INDEXED)
+	{
+		SUB(32, R(ABI_RETURN), R(skipped_reg));
+		RET();
+
+		SetJumpTarget(m_skip_vertex);
+		ADD(32, R(skipped_reg), Imm8(1));
+		JMP(cont);
+	}
+	else
+	{
+		RET();
+	}
+
+	m_VertexSize = m_src_ofs;
+	m_native_vtx_decl.stride = m_dst_ofs;
+}
+
+bool VertexLoaderX64::IsInitialized()
+{
+	// Uses PSHUFB.
+	return cpu_info.bSSSE3;
+}
+
+int VertexLoaderX64::RunVertices(int primitive, int count, DataReader src, DataReader dst)
+{
+	m_numLoadedVertices += count;
+	return ((int (*)(u8* src, u8* dst, int count))region)(src.GetPointer(), dst.GetPointer(), count);
+}
diff --git a/Source/Core/VideoCommon/VertexLoaderX64.h b/Source/Core/VideoCommon/VertexLoaderX64.h
new file mode 100644
index 0000000000..3cf9d7c8ba
--- /dev/null
+++ b/Source/Core/VideoCommon/VertexLoaderX64.h
@@ -0,0 +1,22 @@
+#include "Common/x64Emitter.h"
+#include "VideoCommon/VertexLoaderBase.h"
+
+class VertexLoaderX64 : public VertexLoaderBase, public Gen::X64CodeBlock
+{
+public:
+	VertexLoaderX64(const TVtxDesc& vtx_desc, const VAT& vtx_att);
+
+protected:
+	std::string GetName() const override { return "VertexLoaderX64"; }
+	bool IsInitialized() override;
+	int RunVertices(int primitive, int count, DataReader src, DataReader dst) override;
+
+private:
+	u32 m_src_ofs = 0;
+	u32 m_dst_ofs = 0;
+	Gen::FixupBranch m_skip_vertex;
+	Gen::OpArg GetVertexAddr(int array, u64 attribute);
+	int ReadVertex(Gen::OpArg data, u64 attribute, int format, int count_in, int count_out, u8 scaling_exponent, AttributeFormat* native_format);
+	void ReadColor(Gen::OpArg data, u64 attribute, int format, int elements);
+	void GenerateVertexLoader();
+};
diff --git a/Source/Core/VideoCommon/VertexLoader_Color.cpp b/Source/Core/VideoCommon/VertexLoader_Color.cpp
index 58b1b96f31..5f2a8106c7 100644
--- a/Source/Core/VideoCommon/VertexLoader_Color.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_Color.cpp
@@ -10,10 +10,6 @@
 #include "VideoCommon/VertexManagerBase.h"
 #include "VideoCommon/VideoCommon.h"
 
-#define RSHIFT 0
-#define GSHIFT 8
-#define BSHIFT 16
-#define ASHIFT 24
 #define AMASK 0xFF000000
 
 __forceinline void _SetCol(VertexLoader* loader, u32 val)
@@ -24,13 +20,14 @@ __forceinline void _SetCol(VertexLoader* loader, u32 val)
 
 //color comes in format BARG in 16 bits
 //BARG -> AABBGGRR
-__forceinline void _SetCol4444(VertexLoader* loader, u16 val)
-{
-	u32 col = (val & 0xF0);             // col  = 000000R0;
-	col |=    (val & 0xF ) << 12;       // col |= 0000G000;
-	col |= (((u32)val) & 0xF000) << 8;  // col |= 00B00000;
-	col |= (((u32)val) & 0x0F00) << 20; // col |= A0000000;
-	col |= col >> 4;                    // col =  A0B0G0R0 | 0A0B0G0R;
+__forceinline void _SetCol4444(VertexLoader* loader, u16 val_)
+{
+	u32 col, val = val_;
+	col  =  val & 0x00F0;        // col  = 000000R0;
+	col |= (val & 0x000F) << 12; // col |= 0000G000;
+	col |= (val & 0xF000) << 8;  // col |= 00B00000;
+	col |= (val & 0x0F00) << 20; // col |= A0000000;
+	col |= col >> 4;             // col  = A0B0G0R0 | 0A0B0G0R;
 	_SetCol(loader, col);
 }
 
@@ -38,9 +35,9 @@ __forceinline void _SetCol4444(VertexLoader* loader, u16 val)
 //RRRRRRGG GGGGBBBB BBAAAAAA
 __forceinline void _SetCol6666(VertexLoader* loader, u32 val)
 {
-	u32 col = (val >> 16) & 0xFC;
-	col |=    (val >>  2) & 0xFC00;
-	col |=    (val << 12) & 0xFC0000;
+	u32 col = (val >> 16) & 0x000000FC;
+	col |=    (val >>  2) & 0x0000FC00;
+	col |=    (val << 12) & 0x00FC0000;
 	col |=    (val << 26) & 0xFC000000;
 	col |=    (col >>  6) & 0x03030303;
 	_SetCol(loader, col);
@@ -48,13 +45,14 @@ __forceinline void _SetCol6666(VertexLoader* loader, u32 val)
 
 //color comes in RGB
 //RRRRRGGG GGGBBBBB
-__forceinline void _SetCol565(VertexLoader* loader, u16 val)
-{
-	u32 col =   (val  >>  8) & 0xF8;
-	col |=      (val  <<  5) & 0xFC00;
-	col |=(((u32)val) << 19) & 0xF80000;
-	col |= (col >> 5) & 0x070007;
-	col |= (col >> 6) & 0x000300;
+__forceinline void _SetCol565(VertexLoader* loader, u16 val_)
+{
+	u32 col, val = val_;
+	col  = (val >>  8) & 0x0000F8;
+	col |= (val <<  5) & 0x00FC00;
+	col |= (val << 19) & 0xF80000;
+	col |= (col >>  5) & 0x070007;
+	col |= (col >>  6) & 0x000300;
 	_SetCol(loader, col | AMASK);
 }
 
@@ -96,11 +94,6 @@ void LOADERDECL Color_ReadDirect_24b_6666(VertexLoader* loader)
 }
 // F|RES: i am not 100 percent sure, but the colElements seems to be important for rendering only
 // at least it fixes mario party 4
-//
-//  if (colElements[colIndex])
-//  else
-//      col |= 0xFF<<ASHIFT;
-//
 void LOADERDECL Color_ReadDirect_32b_8888(VertexLoader* loader)
 {
 	// TODO (mb2): check this
@@ -108,7 +101,7 @@ void LOADERDECL Color_ReadDirect_32b_8888(VertexLoader* loader)
 
 	// "kill" the alpha
 	if (!loader->m_colElements[loader->m_colIndex])
-		col |= 0xFF << ASHIFT;
+		col |= AMASK;
 
 	_SetCol(loader, col);
 }
diff --git a/Source/Core/VideoCommon/VertexLoader_Normal.cpp b/Source/Core/VideoCommon/VertexLoader_Normal.cpp
index 70081ead11..0d794d9944 100644
--- a/Source/Core/VideoCommon/VertexLoader_Normal.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_Normal.cpp
@@ -50,7 +50,7 @@ __forceinline void ReadIndirect(const T* data)
 		dst.Write(FracAdjust(Common::FromBigEndian(data[i])));
 	}
 
-	dst.WritePointer(&g_vertex_manager_write_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
 	LOG_NORM();
 }
 
@@ -110,7 +110,7 @@ struct Normal_Direct_SSSE3
 	{
 		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);
+		const __m128 scale = _mm_set_ps1(frac);
 		for (int i = 0; i < N; i++, pData += 3)
 			Vertex_Read_SSSE3<T, true, true>(pData, scale);
 		DataSkip<N * 3 * sizeof(T)>();
@@ -128,7 +128,7 @@ __forceinline void Normal_Index_Offset_SSSE3()
 	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);
+	const __m128 scale = _mm_set_ps1(frac);
 	for (int i = 0; i < N; i++, pData += 3)
 		Vertex_Read_SSSE3<T, true, true>(pData, scale);
 }
diff --git a/Source/Core/VideoCommon/VertexLoader_Position.cpp b/Source/Core/VideoCommon/VertexLoader_Position.cpp
index 419c041b5b..3313c12b5a 100644
--- a/Source/Core/VideoCommon/VertexLoader_Position.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_Position.cpp
@@ -81,8 +81,8 @@ void LOADERDECL Pos_ReadDirect(VertexLoader* loader)
 	for (int i = 0; i < 3; ++i)
 		dst.Write(i<N ? PosScale(src.Read<T>(), scale) : 0.f);
 
-	dst.WritePointer(&g_vertex_manager_write_ptr);
-	src.WritePointer(&g_video_buffer_read_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
+	g_video_buffer_read_ptr = src.GetPointer();
 	LOG_VTX();
 }
 
@@ -101,7 +101,7 @@ void LOADERDECL Pos_ReadIndex(VertexLoader* loader)
 	for (int i = 0; i < 3; ++i)
 		dst.Write(i<N ? PosScale(Common::FromBigEndian(data[i]), scale) : 0.f);
 
-	dst.WritePointer(&g_vertex_manager_write_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
 	LOG_VTX();
 }
 
diff --git a/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp b/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
index 4a858c84f9..6e807a46e8 100644
--- a/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
+++ b/Source/Core/VideoCommon/VertexLoader_TextCoord.cpp
@@ -56,8 +56,8 @@ void LOADERDECL TexCoord_ReadDirect(VertexLoader* loader)
 	for (int i = 0; i != N; ++i)
 		dst.Write(TCScale(src.Read<T>(), scale));
 
-	dst.WritePointer(&g_vertex_manager_write_ptr);
-	src.WritePointer(&g_video_buffer_read_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
+	g_video_buffer_read_ptr = src.GetPointer();
 	LOG_TEX<N>();
 
 	++loader->m_tcIndex;
@@ -77,7 +77,7 @@ void LOADERDECL TexCoord_ReadIndex(VertexLoader* loader)
 	for (int i = 0; i != N; ++i)
 		dst.Write(TCScale(Common::FromBigEndian(data[i]), scale));
 
-	dst.WritePointer(&g_vertex_manager_write_ptr);
+	g_vertex_manager_write_ptr = dst.GetPointer();
 	LOG_TEX<N>();
 	++loader->m_tcIndex;
 }
@@ -166,14 +166,14 @@ void VertexLoader_TextCoord::Init()
 		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][0][1] = TexCoord_ReadIndex2_SSSE3<u16, u8>;
+		tableReadTexCoord[3][1][1] = TexCoord_ReadIndex2_SSSE3<u16, s8>;
+		tableReadTexCoord[3][2][1] = TexCoord_ReadIndex2_SSSE3<u16, u16>;
+		tableReadTexCoord[3][3][1] = TexCoord_ReadIndex2_SSSE3<u16, s16>;
 		tableReadTexCoord[3][4][1] = TexCoord_ReadIndex2_SSSE3<u16, float>;
 	}
 #endif
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj b/Source/Core/VideoCommon/VideoCommon.vcxproj
index b4db374e28..f76c59e8ba 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj
@@ -66,6 +66,7 @@
     <ClCompile Include="TextureConversionShader.cpp" />
     <ClCompile Include="VertexLoader.cpp" />
     <ClCompile Include="VertexLoaderBase.cpp" />
+    <ClCompile Include="VertexLoaderX64.cpp" />
     <ClCompile Include="VertexLoaderManager.cpp" />
     <ClCompile Include="VertexLoader_Color.cpp" />
     <ClCompile Include="VertexLoader_Normal.cpp" />
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
index d00ee5518c..eca9287439 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
@@ -122,6 +122,9 @@
     <ClCompile Include="VertexLoaderBase.cpp">
       <Filter>Vertex Loading</Filter>
     </ClCompile>
+    <ClCompile Include="VertexLoaderX64.cpp">
+      <Filter>Vertex Loading</Filter>
+    </ClCompile>
     <ClCompile Include="VertexLoader_Color.cpp">
       <Filter>Vertex Loading</Filter>
     </ClCompile>