VideoCommon: Cache native vertex formats

We are used to have a 1:1 mapping of GX vertex formats and the native (OGL + D3D) ones, but there are by far more GX ones.
This new cache maps them directly so that we don't flush on GX vertex format changes as long as the native one doesn't change.

The idea is stolen from galop1n.

Source/Core/VideoCommon/NativeVertexFormat.h

@@ -76,6 +76,15 @@ struct PortableVertexDeclaration
 	AttributeFormat colors[2];
 	AttributeFormat texcoords[8];
 	AttributeFormat posmtx;
+
+	inline bool operator<(const PortableVertexDeclaration& b) const
+	{
+		return memcmp(this, &b, sizeof(PortableVertexDeclaration)) < 0;
+	}
+	inline bool operator==(const PortableVertexDeclaration& b) const
+	{
+		return memcmp(this, &b, sizeof(PortableVertexDeclaration)) == 0;
+	}
 };
 
 // The implementation of this class is specific for GL/DX, so NativeVertexFormat.cpp

Source/Core/VideoCommon/VertexLoader.cpp

@@ -463,7 +463,6 @@ VertexLoader::VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr)
 	m_compiledCode = nullptr;
 	m_numLoadedVertices = 0;
 	m_VertexSize = 0;
-	m_NativeFmt = nullptr;
 	loop_counter = 0;
 	VertexLoader_Normal::Init();
 	VertexLoader_Position::Init();
@@ -488,7 +487,6 @@ VertexLoader::~VertexLoader()
 	#ifdef USE_VERTEX_LOADER_JIT
 	FreeCodeSpace();
 	#endif
-	delete m_NativeFmt;
 }
 
 void VertexLoader::CompileVertexTranslator()
@@ -773,9 +771,7 @@ void VertexLoader::CompileVertexTranslator()
 	ABI_PopAllCalleeSavedRegsAndAdjustStack();
 	RET();
 #endif
-	m_NativeFmt = g_vertex_manager->CreateNativeVertexFormat();
-	m_NativeFmt->m_components = components;
-	m_NativeFmt->Initialize(vtx_decl);
+	m_NativeFmt = VertexLoaderManager::GetNativeVertexFormat(vtx_decl, components);
 }
 
 void VertexLoader::WriteCall(TPipelineFunction func)
@@ -825,9 +821,6 @@ void VertexLoader::SetupRunVertices(int vtx_attr_group, int primitive, int const
 	// Flush if our vertex format is different from the currently set.
 	if (g_nativeVertexFmt != nullptr && g_nativeVertexFmt != m_NativeFmt)
 	{
-		// We really must flush here. It's possible that the native representations
-		// of the two vtx formats are the same, but we have no way to easily check that
-		// now.
 		VertexManager::Flush();
 		// Also move the Set() here?
 	}

Source/Core/VideoCommon/VertexLoaderManager.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <memory>
 #include <unordered_map>
 #include <vector>
 
@@ -33,11 +34,13 @@ struct hash<VertexLoaderUID>
 }
 
 typedef std::unordered_map<VertexLoaderUID, VertexLoader*> VertexLoaderMap;
+typedef std::map<PortableVertexDeclaration, std::unique_ptr<NativeVertexFormat>> NativeVertexLoaderMap;
 
 namespace VertexLoaderManager
 {
 
 static VertexLoaderMap g_VertexLoaderMap;
+static NativeVertexLoaderMap s_native_vertex_map;
 // TODO - change into array of pointers. Keep a map of all seen so far.
 
 void Init()
@@ -55,6 +58,7 @@ void Shutdown()
 		delete p.second;
 	}
 	g_VertexLoaderMap.clear();
+	s_native_vertex_map.clear();
 }
 
 namespace
@@ -131,6 +135,19 @@ int GetVertexSize(int vtx_attr_group)
 	return RefreshLoader(vtx_attr_group)->GetVertexSize();
 }
 
+NativeVertexFormat* GetNativeVertexFormat(const PortableVertexDeclaration& format, u32 components)
+{
+	auto& native = s_native_vertex_map[format];
+	if (!native)
+	{
+		auto raw_pointer = g_vertex_manager->CreateNativeVertexFormat();
+		native = std::unique_ptr<NativeVertexFormat>(raw_pointer);
+		native->m_components = components;
+		native->Initialize(format);
+	}
+	return native.get();
+}
+
 }  // namespace
 
 void LoadCPReg(u32 sub_cmd, u32 value)

Source/Core/VideoCommon/VertexLoaderManager.h

@@ -7,6 +7,7 @@
 #include <string>
 
 #include "Common/Common.h"
+#include "VideoCommon/NativeVertexFormat.h"
 
 namespace VertexLoaderManager
 {
@@ -20,6 +21,8 @@ namespace VertexLoaderManager
 
 	// For debugging
 	void AppendListToString(std::string *dest);
+
+	NativeVertexFormat* GetNativeVertexFormat(const PortableVertexDeclaration& format, u32 components);
 };
 
 void RecomputeCachedArraybases();