Merge branch 'master' into armjit

Core/HLE/__sceAudio.cpp

@@ -209,12 +209,12 @@ void __AudioUpdate()
 				outAudioQueue.push((s16)sampleL);
 				outAudioQueue.push((s16)sampleR);
 			}
+		} else {
+			// This happens quite a lot. There's still something slightly off
+			// about the amount of audio we produce.
+			DEBUG_LOG(HLE, "Audio outbuffer overrun! room = %i / %i", outAudioQueue.room(), (u32)outAudioQueue.capacity());
 		}
 		section.unlock();
-	} else {
-		// This happens quite a lot. There's still something slightly off
-		// about the amount of audio we produce.
-		DEBUG_LOG(HLE, "Audio outbuffer overrun! room = %i / %i", outAudioQueue.room(), (u32)outAudioQueue.capacity());
 	}
 	
 }

Core/HLE/sceDisplay.cpp

@@ -208,7 +208,7 @@ void hleEnterVblank(u64 userdata, int cyclesLate) {
 	// Here we will be drawing to the non buffered front surface.
 	if (g_Config.bShowDebugStats && gpuStats.numDrawCalls) {
 		gpu->UpdateStats();
-		char stats[512];
+		char stats[2048];
 		sprintf(stats,
 			"Frames: %i\n"
 			"DL processing time: %0.2f ms\n"
@@ -217,6 +217,7 @@ void hleEnterVblank(u64 userdata, int cyclesLate) {
 			"Draw calls: %i\n"
 			"Draw flushes: %i\n"
 			"Vertices Transformed: %i\n"
+			"FBOs active: %i\n"
 			"Textures active: %i\n"
 			"Textures decoded: %i\n"
 			"Texture invalidations: %i\n"
@@ -231,6 +232,7 @@ void hleEnterVblank(u64 userdata, int cyclesLate) {
 			gpuStats.numDrawCalls,
 			gpuStats.numFlushes,
 			gpuStats.numVertsTransformed,
+			gpuStats.numFBOs,
 			gpuStats.numTextures,
 			gpuStats.numTexturesDecoded,
 			gpuStats.numTextureInvalidations,

GPU/GLES/DisplayListInterpreter.cpp

@@ -37,6 +37,11 @@
 extern u32 curTextureWidth;
 extern u32 curTextureHeight;
 
+// Aggressively delete unused FBO:s to save gpu memory.
+enum {
+	FBO_OLD_AGE = 4
+};
+
 const int flushOnChangedBeforeCommandList[] = {
 	GE_CMD_VERTEXTYPE,
 	GE_CMD_BLENDMODE,
@@ -200,6 +205,7 @@ void GLES_GPU::DumpNextFrame() {
 
 void GLES_GPU::BeginFrame() {
 	TextureCache_StartFrame();
+	DecimateFBOs();
 
 	if (dumpNextFrame_) {
 		NOTICE_LOG(G3D, "DUMPING THIS FRAME");
@@ -277,6 +283,17 @@ void GLES_GPU::CopyDisplayToOutput() {
 	BeginDebugDraw();
 }
 
+void GLES_GPU::DecimateFBOs() {
+	for (auto iter = vfbs_.begin(); iter != vfbs_.end();) {
+		if ((*iter)->last_frame_used + FBO_OLD_AGE < gpuStats.numFrames) {
+			fbo_destroy((*iter)->fbo);
+			vfbs_.erase(iter++);
+		}
+		else
+			++iter;
+	}
+}
+
 GLES_GPU::VirtualFramebuffer *GLES_GPU::GetDisplayFBO() {
 	for (auto iter = vfbs_.begin(); iter != vfbs_.end(); ++iter) {
 		if (((*iter)->fb_address & 0x3FFFFFF) == (displayFramebufPtr_ & 0x3FFFFFF)) {
@@ -345,6 +362,7 @@ void GLES_GPU::SetRenderFrameBuffer() {
 		vfb->height = drawing_height;
 		vfb->format = fmt;
 		vfb->fbo = fbo_create(vfb->width * renderWidthFactor_, vfb->height * renderHeightFactor_, 1, true);
+		vfb->last_frame_used = gpuStats.numFrames;
 		vfbs_.push_back(vfb);
 		fbo_bind_as_render_target(vfb->fbo);
 		glstate.viewport.set(0, 0, renderWidth_, renderHeight_);
@@ -362,6 +380,7 @@ void GLES_GPU::SetRenderFrameBuffer() {
 		fbo_bind_as_render_target(vfb->fbo);
 		glstate.viewport.set(0, 0, renderWidth_, renderHeight_);
 		currentRenderVfb_ = vfb;
+		vfb->last_frame_used = gpuStats.numFrames;
 	}
 }
 
@@ -1064,6 +1083,7 @@ void GLES_GPU::UpdateStats() {
 	gpuStats.numFragmentShaders = shaderManager_->NumFragmentShaders();
 	gpuStats.numShaders = shaderManager_->NumPrograms();
 	gpuStats.numTextures = TextureCache_NumLoadedTextures();
+	gpuStats.numFBOs = vfbs_.size();
 }
 
 void GLES_GPU::DoBlockTransfer() {

GPU/GLES/DisplayListInterpreter.h

@@ -63,6 +63,9 @@ private:
 	void BeginDebugDraw();
 	void EndDebugDraw();
 
+	// Deletes old FBOs.
+	void DecimateFBOs();
+
 	FramebufferManager framebufferManager;
 	TransformDrawEngine transformDraw_;
 	ShaderManager *shaderManager_;
@@ -86,6 +89,8 @@ private:
 	};
 
 	struct VirtualFramebuffer {
+		int last_frame_used;
+
 		u32 fb_address;
 		u32 z_address;
 		int fb_stride;

GPU/GLES/TextureCache.cpp

@@ -61,11 +61,12 @@ u16 *clutBuf16;
 
 void TextureCache_Init() {
 	// TODO: Switch to aligned allocations for alignment. AllocateMemoryPages would do the trick.
-	tmpTexBuf32 = new u32[1024 * 512];
-	tmpTexBuf16 = new u16[1024 * 512];
-	tmpTexBufRearrange = new u32[1024 * 512];
-	clutBuf32 = new u32[4096];
-	clutBuf16 = new u16[4096];
+	// This is 5MB of temporary storage. Might be possible to shrink it.
+	tmpTexBuf32 = new u32[1024 * 512];  // 2MB
+	tmpTexBuf16 = new u16[1024 * 512];  // 1MB
+	tmpTexBufRearrange = new u32[1024 * 512];   // 2MB
+	clutBuf32 = new u32[4096];  // 4K
+	clutBuf16 = new u16[4096];  // 4K
 }
 
 void TextureCache_Shutdown() {

GPU/GPUState.h

@@ -283,6 +283,7 @@ struct GPUStatistics
 	int numVertexShaders;
 	int numFragmentShaders;
 	int numShaders;
+	int numFBOs;
 };
 
 void InitGfxState();