Merge pull request #218 from unknownbrackets/texcache

Use sceKernelDcache*() to invalidate the texcache

Core/HLE/sceKernel.cpp

@@ -25,6 +25,8 @@
 #include "../PSPLoaders.h"
 #include "../../Core/CoreTiming.h"
 #include "../../Core/System.h"
+#include "../../GPU/GPUInterface.h"
+#include "../../GPU/GPUState.h"
 
 
 #include "__sceAudio.h"
@@ -187,6 +189,7 @@ void sceKernelGetGPI()
 // textures, and in the future display lists, in some cases though.
 void sceKernelDcacheInvalidateRange(u32 addr, int size)
 {
+	gpu->InvalidateCache(addr, size);
 }
 void sceKernelDcacheWritebackAll()
 {
@@ -196,9 +199,11 @@ void sceKernelDcacheWritebackRange(u32 addr, int size)
 }
 void sceKernelDcacheWritebackInvalidateRange(u32 addr, int size)
 {
+	gpu->InvalidateCache(addr, size);
 }
 void sceKernelDcacheWritebackInvalidateAll()
 {
+	gpu->InvalidateCache(0, -1);
 }
 
 KernelObjectPool kernelObjects;

GPU/GLES/DisplayListInterpreter.cpp

@@ -1259,7 +1259,7 @@ void GLES_GPU::DoBlockTransfer()
 {
 	// TODO: This is used a lot to copy data around between render targets and textures,
 	// and also to quickly load textures from RAM to VRAM. So we should do checks like the following:
-	//  * Does dstBasePtr point to an existing texture? If so invalidate it and reload it immediately.
+	//  * Does dstBasePtr point to an existing texture? If so maybe reload it immediately.
 	//
 	//  * Does srcBasePtr point to a render target, and dstBasePtr to a texture? If so
 	//    either copy between rt and texture or reassign the texture to point to the render target
@@ -1293,4 +1293,14 @@ void GLES_GPU::DoBlockTransfer()
 	}
 
 	// TODO: Notify all overlapping textures that it's time to die/reload.
+
+	TextureCache_Invalidate(srcBasePtr + srcY * srcStride + srcX, height * srcStride + width * bpp);
+}
+
+void GLES_GPU::InvalidateCache(u32 addr, int size)
+{
+	if (size > 0)
+		TextureCache_Invalidate(addr, size);
+	else
+		TextureCache_Clear(true);
 }

GPU/GLES/DisplayListInterpreter.h

@@ -49,6 +49,7 @@ public:
 	virtual void CopyDisplayToOutput();
 	virtual void BeginFrame();
 	virtual void UpdateStats();
+	virtual void InvalidateCache(u32 addr, int size);
 
 private:
 	// TransformPipeline.cpp

GPU/GLES/TextureCache.cpp

@@ -106,6 +106,26 @@ void TextureCache_Decimate()
 	}
 }
 
+void TextureCache_Invalidate(u32 addr, int size)
+{
+	u32 addr_end = addr + size;
+
+	for (TexCache::iterator iter = cache.begin(); iter != cache.end(); )
+	{
+		// Clear if either the addr or clutaddr is in the range.
+		bool invalidate = iter->second.addr >= addr && iter->second.addr < addr_end;
+		invalidate |= iter->second.clutaddr >= addr && iter->second.clutaddr < addr_end;
+
+		if (invalidate)
+		{
+			glDeleteTextures(1, &iter->second.texture);
+			cache.erase(iter++);
+		}
+		else
+			++iter;
+	}
+}
+
 int TextureCache_NumLoadedTextures() 
 {
 	return cache.size();

GPU/GLES/TextureCache.h

@@ -25,4 +25,5 @@ void TextureCache_Init();
 void TextureCache_Shutdown();
 void TextureCache_Clear(bool delete_them);
 void TextureCache_Decimate();  // Run this once per frame to get rid of old textures.
+void TextureCache_Invalidate(u32 addr, int size);
 int TextureCache_NumLoadedTextures();

GPU/GPUInterface.h

@@ -45,6 +45,10 @@ public:
 	// Tells the GPU to update the gpuStats structure.
 	virtual void UpdateStats() = 0;
 
+	// Invalidate any cached content sourced from the specified range.
+	// If size = -1, invalidate everything.
+	virtual void InvalidateCache(u32 addr, int size) = 0;
+
 	// Internal hack to avoid interrupts from "PPGe" drawing (utility UI, etc)
 	virtual void EnableInterrupts(bool enable) = 0;
 };

GPU/Null/NullGpu.cpp

@@ -838,3 +838,8 @@ void NullGPU::UpdateStats()
 	gpuStats.numShaders = 0;
 	gpuStats.numTextures = 0;
 }
+
+void NullGPU::InvalidateCache(u32 addr, int size)
+{
+	// Nothing to invalidate.
+}

GPU/Null/NullGpu.h

@@ -40,6 +40,7 @@ public:
 	virtual void SetDisplayFramebuffer(u32 framebuf, u32 stride, int format) {}
 	virtual void CopyDisplayToOutput() {}
 	virtual void UpdateStats();
+	virtual void InvalidateCache(u32 addr, int size);
 
 private:
 	bool ProcessDLQueue();