Duplicate framebuffers textured onto themselves.

This fixes hrydgard#2917.

I verified that, at least for my card, blitting is much faster than
glReadPixels (by quite a margin.)

GPU/GLES/Framebuffer.cpp

@@ -305,7 +305,8 @@ FramebufferManager::FramebufferManager() :
 	shaderManager_(0),
 	usePostShader_(false),
 	postShaderAtOutputResolution_(false),
-	resized_(false)
+	resized_(false),
+	renderCopy_(0)
 #ifndef USING_GLES2
 	,
 	pixelBufObj_(0),
@@ -919,6 +920,48 @@ void FramebufferManager::SetLineWidth() {
 #endif
 }
 
+void FramebufferManager::BindFramebufferColor(VirtualFramebuffer *framebuffer) {
+	if (!framebuffer->fbo || !useBufferedRendering_) {
+		glBindTexture(GL_TEXTURE_2D, 0);
+		gstate_c.skipDrawReason |= SKIPDRAW_BAD_FB_TEXTURE;
+		return;
+	}
+
+	if (MaskedEqual(framebuffer->fb_address, gstate.getFrameBufRawAddress())) {
+#ifndef USING_GLES2
+		if (gl_extensions.FBO_ARB) {
+#else
+		if (gl_extensions.GLES3) {
+#endif
+#ifdef MAY_HAVE_GLES3
+
+			// TODO: Maybe merge with bvfbs_?  Not sure if those could be packing, and they're created at a different size.
+			if (!renderCopy_ || renderCopyWidth_ != framebuffer->renderWidth || renderCopyHeight_ != framebuffer->renderHeight) {
+				if (renderCopy_)
+					fbo_destroy(renderCopy_);
+				renderCopy_ = fbo_create(framebuffer->renderWidth, framebuffer->renderHeight, 1, true, framebuffer->colorDepth);
+				renderCopyWidth_ = framebuffer->renderWidth;
+				renderCopyHeight_ = framebuffer->renderHeight;
+			}
+
+			fbo_bind_as_render_target(renderCopy_);
+			glViewport(0, 0, framebuffer->renderWidth, framebuffer->renderHeight);
+			fbo_bind_for_read(framebuffer->fbo);
+			glBlitFramebuffer(0, 0, framebuffer->renderWidth, framebuffer->renderHeight, 0, 0, framebuffer->renderWidth, framebuffer->renderHeight, GL_COLOR_BUFFER_BIT, GL_NEAREST);
+
+			fbo_bind_as_render_target(currentRenderVfb_->fbo);
+			if (gl_extensions.gpuVendor != GPU_VENDOR_POWERVR)
+				glstate.viewport.restore();
+			fbo_bind_color_as_texture(renderCopy_, 0);
+#endif
+		} else {
+			fbo_bind_color_as_texture(framebuffer->fbo, 0);
+		}
+	} else {
+		fbo_bind_color_as_texture(framebuffer->fbo, 0);
+	}
+}
+
 void FramebufferManager::CopyDisplayToOutput() {
 	fbo_unbind();
 	glstate.viewport.set(0, 0, PSP_CoreParameter().pixelWidth, PSP_CoreParameter().pixelHeight);

GPU/GLES/Framebuffer.h

@@ -138,6 +138,9 @@ class FramebufferManager {
 	void UpdateFromMemory(u32 addr, int size, bool safe);
 	void SetLineWidth();
 
+	// For use when texturing from a framebuffer.  May create a duplicate if target.
+	void BindFramebufferColor(VirtualFramebuffer *framebuffer);
+
 #ifdef USING_GLES2
   void ReadFramebufferToMemory(VirtualFramebuffer *vfb, bool sync = true);
 #else
@@ -223,6 +226,9 @@ class FramebufferManager {
 	bool useBufferedRendering_;
 
 	std::vector<VirtualFramebuffer *> bvfbs_; // blitting FBOs
+	FBO *renderCopy_;
+	int renderCopyWidth_;
+	int renderCopyHeight_;
 
 	std::set<std::pair<u32, u32>> knownFramebufferCopies_;
 

GPU/GLES/GLES_GPU.cpp

@@ -397,6 +397,7 @@ GLES_GPU::GLES_GPU()
 	transformDraw_.SetFramebufferManager(&framebufferManager_);
 	framebufferManager_.SetTextureCache(&textureCache_);
 	framebufferManager_.SetShaderManager(shaderManager_);
+	textureCache_.SetFramebufferManager(&framebufferManager_);
 
 	// Sanity check gstate
 	if ((int *)&gstate.transferstart - (int *)&gstate != 0xEA) {

GPU/GLES/TextureCache.cpp

@@ -830,22 +830,15 @@ bool SetDebugTexture() {
 }
 #endif
 
-void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry)
-{
+void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry) {
 	entry->framebuffer->usageFlags |= FB_USAGE_TEXTURE;
 	bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
 	if (useBufferedRendering) {
-		// For now, let's not bind FBOs that we know are off (invalidHint will be -1.)
-		// But let's still not use random memory.
-		if (entry->framebuffer->fbo) {
-			fbo_bind_color_as_texture(entry->framebuffer->fbo, 0);
-			// Keep the framebuffer alive.
-			// TODO: Dangerous if it sets a new one?
-			entry->framebuffer->last_frame_used = gpuStats.numFlips;
-		} else {
-			glBindTexture(GL_TEXTURE_2D, 0);
-			gstate_c.skipDrawReason |= SKIPDRAW_BAD_FB_TEXTURE;
-		}
+		framebufferManager_->BindFramebufferColor(entry->framebuffer);
+
+		// Keep the framebuffer alive.
+		entry->framebuffer->last_frame_used = gpuStats.numFlips;
+
 		// We need to force it, since we may have set it on a texture before attaching.
 		UpdateSamplingParams(*entry, true);
 		gstate_c.curTextureWidth = entry->framebuffer->width;
@@ -883,11 +876,6 @@ void TextureCache::SetTexture(bool force) {
 	int w = gstate.getTextureWidth(0);
 	int h = gstate.getTextureHeight(0);
 
-	u32 fb_addr = gstate.getFrameBufRawAddress() | 0x04000000;
-	if (fb_addr == gstate.getTextureAddress(0)) {
-		WARN_LOG_REPORT(HLE, "Texturing from same buffer as target (%08x : %ix%i)", fb_addr, w, h);
-	}
-
 	GETextureFormat format = gstate.getTextureFormat();
 	if (format >= 11) {
 		ERROR_LOG_REPORT(G3D, "Unknown texture format %i", format);

GPU/GLES/TextureCache.h

@@ -24,6 +24,7 @@
 #include "TextureScaler.h"
 
 struct VirtualFramebuffer;
+class FramebufferManager;
 
 enum TextureFiltering {
 	AUTO = 1,
@@ -56,6 +57,10 @@ class TextureCache {
 	// are being rendered to. This is barebones so far.
 	void NotifyFramebuffer(u32 address, VirtualFramebuffer *framebuffer, FramebufferNotification msg);
 
+	void SetFramebufferManager(FramebufferManager *fbManager) {
+		framebufferManager_ = fbManager;
+	}
+
 	size_t NumLoadedTextures() const {
 		return cache.size();
 	}
@@ -160,5 +165,6 @@ class TextureCache {
 	float maxAnisotropyLevel;
 
 	int decimationCounter_;
+	FramebufferManager *framebufferManager_;
 };