Merge branch 'StreamBuffers'

This branch drops our temporary buffer in VertexLoaderBase.
Instead, every backend now must provide a buffer to convert vertices and indices.
D3D just uses a temporary buffer like before.
OGL maps the gpu based buffer and stream to them directly.
So this will avoid an unneeded memcpy on OGL backend.

Source/Core/VideoBackends/D3D/VertexManager.cpp

@@ -77,6 +77,12 @@ void VertexManager::DestroyDeviceObjects()
 
 VertexManager::VertexManager()
 {
+	LocalVBuffer.resize(MAXVBUFFERSIZE);
+	s_pCurBufferPointer = s_pBaseBufferPointer = &LocalVBuffer[0];
+	s_pEndBufferPointer = s_pBaseBufferPointer + LocalVBuffer.size();
+
+	LocalIBuffer.resize(MAXIBUFFERSIZE);
+
 	CreateDeviceObjects();
 }
 
@@ -222,4 +228,10 @@ void VertexManager::vFlush()
 	g_renderer->RestoreState();
 }
 
+void VertexManager::ResetBuffer(u32 stride)
+{
+	s_pCurBufferPointer = s_pBaseBufferPointer;
+	IndexGenerator::Start(GetIndexBuffer());
+}
+
 }  // namespace

Source/Core/VideoBackends/D3D/VertexManager.h

@@ -22,6 +22,10 @@ class VertexManager : public ::VertexManager
 	void CreateDeviceObjects();
 	void DestroyDeviceObjects();
 
+protected:
+	virtual void ResetBuffer(u32 stride);
+	u16* GetIndexBuffer() { return &LocalIBuffer[0]; }
+
 private:
 
 	void PrepareDrawBuffers();
@@ -41,6 +45,9 @@ class VertexManager : public ::VertexManager
 
 	LineGeometryShader m_lineShader;
 	PointGeometryShader m_pointShader;
+
+	std::vector<u8> LocalVBuffer;
+	std::vector<u16> LocalIBuffer;
 };
 
 }  // namespace

Source/Core/VideoBackends/OGL/ProgramShaderCache.cpp

@@ -193,29 +193,19 @@ void ProgramShaderCache::UploadConstants()
 	{
 		if(PixelShaderManager::dirty || VertexShaderManager::dirty)
 		{
-			s_buffer->Alloc(s_ubo_buffer_size);
-			if (DriverDetails::HasBug(DriverDetails::BUG_BROKENBUFFERSTREAM))
-			{
-				// This is just a hack to support our BUFFERDATA upload method
-				// as it's broken to uploaded in a splited way
-				static u8 *tmpbuffer = new u8[s_ubo_buffer_size];
-				memcpy(tmpbuffer, &PixelShaderManager::constants, sizeof(PixelShaderConstants));
-				memcpy(tmpbuffer+ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align), &VertexShaderManager::constants, sizeof(VertexShaderConstants));
-				size_t offset = s_buffer->Upload(tmpbuffer, s_ubo_buffer_size);
-				glBindBufferRange(GL_UNIFORM_BUFFER, 1,
-						s_buffer->getBuffer(), offset, sizeof(PixelShaderConstants));
-				glBindBufferRange(GL_UNIFORM_BUFFER, 2,
-						s_buffer->getBuffer(), offset+ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align), sizeof(VertexShaderConstants));
-			}
-			else
-			{
-				size_t offset = s_buffer->Upload((u8*)&PixelShaderManager::constants, ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align));
-				glBindBufferRange(GL_UNIFORM_BUFFER, 1,
-						s_buffer->getBuffer(), offset, sizeof(PixelShaderConstants));
-				offset = s_buffer->Upload((u8*)&VertexShaderManager::constants, ROUND_UP(sizeof(VertexShaderConstants), s_ubo_align));
-				glBindBufferRange(GL_UNIFORM_BUFFER, 2,
-						s_buffer->getBuffer(), offset, sizeof(VertexShaderConstants));
-			}
+			auto buffer = s_buffer->Map(s_ubo_buffer_size, s_ubo_align);
+
+			memcpy(buffer.first,
+			       &PixelShaderManager::constants, sizeof(PixelShaderConstants));
+
+			memcpy(buffer.first + ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align),
+			       &VertexShaderManager::constants, sizeof(VertexShaderConstants));
+
+			s_buffer->Unmap(s_ubo_buffer_size);
+			glBindBufferRange(GL_UNIFORM_BUFFER, 1, s_buffer->m_buffer, buffer.second,
+			                  sizeof(PixelShaderConstants));
+			glBindBufferRange(GL_UNIFORM_BUFFER, 2, s_buffer->m_buffer, buffer.second + ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align),
+			                  sizeof(VertexShaderConstants));
 
 			PixelShaderManager::dirty = false;
 			VertexShaderManager::dirty = false;
@@ -481,7 +471,7 @@ void ProgramShaderCache::Init(void)
 		// We multiply by *4*4 because we need to get down to basic machine units.
 		// So multiply by four to get how many floats we have from vec4s
 		// Then once more to get bytes
-		s_buffer = new StreamBuffer(GL_UNIFORM_BUFFER, UBO_LENGTH);
+		s_buffer = StreamBuffer::Create(GL_UNIFORM_BUFFER, UBO_LENGTH);
 	}
 
 	// Read our shader cache, only if supported

Source/Core/VideoBackends/OGL/Render.cpp

@@ -1653,7 +1653,7 @@ void Renderer::RestoreAPIState()
 
 	VertexManager *vm = (OGL::VertexManager*)g_vertex_manager;
 	glBindBuffer(GL_ARRAY_BUFFER, vm->m_vertex_buffers);
-	vm->m_last_vao = 0;
+	glBindVertexArray(vm->m_last_vao);
 
 	TextureCache::SetStage();
 }