Some work toward better FBO handling. Fixes some things, breaks some things..

GPU/GLES/Framebuffer.cpp

@@ -235,7 +235,7 @@ void FramebufferManager::DrawActiveTexture(float x, float y, float w, float h, b
 	glsl_unbind();
 }
 
-FramebufferManager::VirtualFramebuffer *FramebufferManager::GetDisplayFBO() {
+VirtualFramebuffer *FramebufferManager::GetDisplayFBO() {
 	for (auto iter = vfbs_.begin(); iter != vfbs_.end(); ++iter) {
 		VirtualFramebuffer *v = *iter;
 		if (MaskedEqual(v->fb_address, displayFramebufPtr_) && v->format == displayFormat_) {
@@ -325,6 +325,7 @@ void FramebufferManager::SetRenderFrameBuffer() {
 		vfb->renderWidth = (u16)(drawing_width * renderWidthFactor);
 		vfb->renderHeight = (u16)(drawing_height * renderHeightFactor);
 		vfb->format = fmt;
+		vfb->usageFlags = FB_USAGE_RENDERTARGET;
 
 		switch (fmt) {
 		case GE_FORMAT_4444: vfb->colorDepth = FBO_4444; break;
@@ -342,7 +343,7 @@ void FramebufferManager::SetRenderFrameBuffer() {
 		//#endif
 
 		vfb->fbo = fbo_create(vfb->renderWidth, vfb->renderHeight, 1, true, vfb->colorDepth);
-		textureCache_->NotifyFramebuffer(vfb->fb_address, vfb->fbo);
+		textureCache_->NotifyFramebuffer(vfb->fb_address, vfb);
 
 		vfb->last_frame_used = gpuStats.numFrames;
 		vfbs_.push_back(vfb);
@@ -368,7 +369,7 @@ void FramebufferManager::SetRenderFrameBuffer() {
 		
 		fbo_bind_as_render_target(vfb->fbo);
 
-		textureCache_->NotifyFramebuffer(vfb->fb_address, vfb->fbo);
+		textureCache_->NotifyFramebuffer(vfb->fb_address, vfb);
 #ifdef USING_GLES2
 		// Some tiled mobile GPUs benefit IMMENSELY from clearing an FBO before rendering
 		// to it. This broke stuff before, so now it only clears on the first use of an
@@ -482,7 +483,7 @@ void FramebufferManager::DecimateFBOs() {
 		}
 		if ((*iter)->last_frame_used + FBO_OLD_AGE < gpuStats.numFrames) {
 			INFO_LOG(HLE, "Destroying FBO for %08x (%i x %i x %i)", vfb->fb_address, vfb->width, vfb->height, vfb->format)
-			textureCache_->NotifyFramebufferDestroyed(vfb->fb_address, vfb->fbo);
+			textureCache_->NotifyFramebufferDestroyed(vfb->fb_address, vfb);
 			fbo_destroy(vfb->fbo);
 			delete vfb;
 			vfbs_.erase(iter++);
@@ -495,7 +496,7 @@ void FramebufferManager::DecimateFBOs() {
 void FramebufferManager::DestroyAllFBOs() {
 	for (auto iter = vfbs_.begin(); iter != vfbs_.end(); ++iter) {
 		VirtualFramebuffer *vfb = *iter;
-		textureCache_->NotifyFramebufferDestroyed(vfb->fb_address, vfb->fbo);
+		textureCache_->NotifyFramebufferDestroyed(vfb->fb_address, vfb);
 		fbo_destroy(vfb->fbo);
 		delete vfb;
 	}

GPU/GLES/Framebuffer.h

@@ -38,6 +38,34 @@ enum PspDisplayPixelFormat {
 	PSP_DISPLAY_PIXEL_FORMAT_8888 = 3,
 };
 
+enum {
+	FB_USAGE_RENDERTARGET = 1,
+	FB_USAGE_TEXTURE = 2,
+};
+
+
+struct VirtualFramebuffer {
+	int last_frame_used;
+
+	u32 fb_address;
+	u32 z_address;
+	int fb_stride;
+	int z_stride;
+
+	// There's also a top left of the drawing region, but meh...
+	u16 width;
+	u16 height;
+	u16 renderWidth;
+	u16 renderHeight;
+
+	u16 usageFlags;
+
+	int format;  // virtual, right now they are all RGBA8888
+	FBOColorDepth colorDepth;
+	FBO *fbo;
+};
+
+
 class FramebufferManager {
 public:
 	FramebufferManager();
@@ -47,25 +75,6 @@ public:
 		textureCache_ = tc;
 	}
 
-	struct VirtualFramebuffer {
-		int last_frame_used;
-
-		u32 fb_address;
-		u32 z_address;
-		int fb_stride;
-		int z_stride;
-
-		// There's also a top left of the drawing region, but meh...
-		u16 width;
-		u16 height;
-		u16 renderWidth;
-		u16 renderHeight;
-
-		int format;  // virtual, right now they are all RGBA8888
-		FBOColorDepth colorDepth;
-		FBO *fbo;
-	};
-
 	void DrawPixels(const u8 *framebuf, int pixelFormat, int linesize);
 	void DrawActiveTexture(float x, float y, float w, float h, bool flip = false);
 

GPU/GLES/ShaderManager.cpp

@@ -243,7 +243,7 @@ void LinkedShader::updateUniforms() {
 	if (u_proj_through != -1 && (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX))
 	{
 		Matrix4x4 proj_through;
-		proj_through.setOrtho(0.0f, 480, 272, 0, 1, 0);
+		proj_through.setOrtho(0.0f, 480, 272, 0, 0, 1);
 		glUniformMatrix4fv(u_proj_through, 1, GL_FALSE, proj_through.getReadPtr());
 	}
 	if (u_texenv != -1 && (dirtyUniforms & DIRTY_TEXENV)) {

GPU/GLES/StateMapping.cpp

@@ -236,7 +236,7 @@ void TransformDrawEngine::UpdateViewportAndProjection() {
 	if (throughmode) {
 		// No viewport transform here. Let's experiment with using region.
 		glstate.viewport.set((0 + regionX1) * renderWidthFactor, (0 - regionY1) * renderHeightFactor, (regionX2 - regionX1) * renderWidthFactor, (regionY2 - regionY1) * renderHeightFactor);
-		glstate.depthRange.set(1.0, 0.0);
+		glstate.depthRange.set(0.0, 1.0);
 	} else {
 		// These we can turn into a glViewport call, offset by offsetX and offsetY. Math after.
 		float vpXa = getFloat24(gstate.viewportx1);

GPU/GLES/TextureCache.cpp

@@ -18,11 +18,12 @@
 #include <map>
 #include <algorithm>
 
-#include "../../Core/MemMap.h"
-#include "../ge_constants.h"
-#include "../GPUState.h"
-#include "TextureCache.h"
-#include "../Core/Config.h"
+#include "Core/MemMap.h"
+#include "GPU/ge_constants.h"
+#include "GPU/GPUState.h"
+#include "GPU/GLES/TextureCache.h"
+#include "GPU/GLES/Framebuffer.h"
+#include "Core/Config.h"
 
 // If a texture hasn't been seen for this many frames, get rid of it.
 #define TEXTURE_KILL_AGE 200
@@ -131,21 +132,21 @@ TextureCache::TexCacheEntry *TextureCache::GetEntryAt(u32 texaddr) {
 		return 0;
 }
 
-void TextureCache::NotifyFramebuffer(u32 address, FBO *fbo) {
+void TextureCache::NotifyFramebuffer(u32 address, VirtualFramebuffer *framebuffer) {
 	// Must be in VRAM so | 0x04000000 it is.
 	TexCacheEntry *entry = GetEntryAt(address | 0x04000000);
 	if (entry) {
 		DEBUG_LOG(HLE, "Render to texture detected at %08x!", address);
-		if (!entry->fbo)
-			entry->fbo = fbo;
+		if (!entry->framebuffer)
+			entry->framebuffer= framebuffer;
 		// TODO: Delete the original non-fbo texture too.
 	}
 }
 
-void TextureCache::NotifyFramebufferDestroyed(u32 address, FBO *fbo) {
+void TextureCache::NotifyFramebufferDestroyed(u32 address, VirtualFramebuffer *fbo) {
 	TexCacheEntry *entry = GetEntryAt(address | 0x04000000);
-	if (entry && entry->fbo) {
-		entry->fbo = 0;
+	if (entry && entry->framebuffer) {
+		entry->framebuffer = 0;
 	}
 }
 
@@ -699,19 +700,15 @@ void TextureCache::SetTexture() {
 	if (iter != cache.end()) {
 		entry = &iter->second;
 		// Check for FBO - slow!
-		if (entry->fbo) {
-			int w = 1 << (gstate.texsize[0] & 0xf);
-			int h = 1 << ((gstate.texsize[0] >> 8) & 0xf);
-			
-			fbo_bind_color_as_texture(entry->fbo, 0);
+		if (entry->framebuffer) {
+			fbo_bind_color_as_texture(entry->framebuffer->fbo, 0);
 			UpdateSamplingParams(*entry, false);
 
-			int fbow, fboh;
-			fbo_get_dimensions(entry->fbo, &fbow, &fboh);
-
 			// This isn't right.
-			gstate_c.curTextureWidth = w; //w;  //RoundUpToPowerOf2(fbow);
-			gstate_c.curTextureHeight = h;  // RoundUpToPowerOf2(fboh);
+			gstate_c.curTextureWidth = entry->framebuffer->width;
+			gstate_c.curTextureHeight = entry->framebuffer->height;
+			int h = 1 << ((gstate.texsize[0] >> 8) & 0xf);
+			gstate_c.actualTextureHeight = h;
 			gstate_c.flipTexture = true;
 			entry->lastFrame = gpuStats.numFrames;
 			return;
@@ -807,7 +804,7 @@ void TextureCache::SetTexture() {
 	entry->hash = texhash;
 	entry->format = format;
 	entry->lastFrame = gpuStats.numFrames;
-	entry->fbo = 0;
+	entry->framebuffer = 0;
 	entry->maxLevel = maxLevel;
 	entry->lodBias = 0.0f;
 

GPU/GLES/TextureCache.h

@@ -21,6 +21,8 @@
 #include "gfx_es2/fbo.h"
 #include "GPU/GPUState.h"
 
+struct VirtualFramebuffer;
+
 class TextureCache 
 {
 public:
@@ -36,16 +38,16 @@ public:
 
 	// FramebufferManager keeps TextureCache updated about what regions of memory
 	// are being rendered to. This is barebones so far.
-	void NotifyFramebuffer(u32 address, FBO *fbo);
-	void NotifyFramebufferDestroyed(u32 address, FBO *fbo);
+	void NotifyFramebuffer(u32 address, VirtualFramebuffer *framebuffer);
+	void NotifyFramebufferDestroyed(u32 address, VirtualFramebuffer *framebuffer);
 
 	size_t NumLoadedTextures() const {
 		return cache.size();
 	}
 
 	bool DecodeTexture(u8 *output, GPUgstate state);
+
 private:
-	
 	struct TexCacheEntry {
 		// After marking STATUS_UNRELIABLE, if it stays the same this many frames we'll trust it again.
 		const static int FRAMES_REGAIN_TRUST = 1000;
@@ -60,7 +62,7 @@ private:
 		int status;
 		u32 addr;
 		u32 hash;
-		FBO *fbo;  // if null, not sourced from an FBO.
+		VirtualFramebuffer *framebuffer;  // if null, not sourced from an FBO.
 		u32 sizeInRAM;
 		int lastFrame;
 		int numFrames;

GPU/GLES/TransformPipeline.cpp

@@ -640,7 +640,7 @@ void TransformDrawEngine::SoftwareTransformAndDraw(
 		transformed[index].fog = fogCoef;
 		memcpy(&transformed[index].u, uv, 2 * sizeof(float));
 		if (gstate_c.flipTexture)
-			transformed[index].v = 1.0f - transformed[index].v * 2.0f;
+			transformed[index].v = 1.0f - transformed[index].v; //(float)gstate_c.actualTextureHeight / gstate_c.curTextureHeight - transformed[index].v;
 		for (int i = 0; i < 4; i++) {
 			transformed[index].color0[i] = c0[i] * 255.0f;
 		}

GPU/GPUState.h

@@ -255,6 +255,7 @@ struct GPUStateCache
 
 	u32 curTextureWidth;
 	u32 curTextureHeight;
+	u32 actualTextureHeight;
 
 	float vpWidth;
 	float vpHeight;