Flip texcoord vertically when texturing from FBO to compensate for OpenGL coordinate systems

GPU/GLES/DisplayListInterpreter.cpp

@@ -542,6 +542,7 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 		break;
 
 	case GE_CMD_TEXTUREMAPENABLE:
+		if (diff)
 			gstate_c.textureChanged = true;
 		break;
 
@@ -549,14 +550,17 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 		break;
 
 	case GE_CMD_FOGCOLOR:
+		if (diff)
 			shaderManager_->DirtyUniform(DIRTY_FOGCOLOR);
 		break;
 
 	case GE_CMD_FOG1:
+		if (diff)
 			shaderManager_->DirtyUniform(DIRTY_FOGCOEF);
 		break;
 
 	case GE_CMD_FOG2:
+		if (diff)
 			shaderManager_->DirtyUniform(DIRTY_FOGCOEF);
 		break;
 
@@ -618,6 +622,7 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 	case GE_CMD_TEXADDR6:
 	case GE_CMD_TEXADDR7:
 		gstate_c.textureChanged = true;
+		shaderManager_->DirtyUniform(DIRTY_UVSCALEOFFSET);
 		break;
 
 	case GE_CMD_TEXBUFWIDTH0:
@@ -667,7 +672,7 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 
 	case GE_CMD_TEXSIZE0:
 		gstate_c.curTextureWidth = 1 << (gstate.texsize[0] & 0xf);
-		gstate_c.curTextureHeight = 1 << ((gstate.texsize[0]>>8) & 0xf);
+		gstate_c.curTextureHeight = 1 << ((gstate.texsize[0] >> 8) & 0xf);
 		shaderManager_->DirtyUniform(DIRTY_UVSCALEOFFSET);
 		//fall thru - ignoring the mipmap sizes for now
 	case GE_CMD_TEXSIZE1:

GPU/GLES/ShaderManager.cpp

@@ -138,6 +138,7 @@ LinkedShader::LinkedShader(Shader *vs, Shader *fs)
 	glUniform1i(u_tex, 0);
 	// The rest, use the "dirty" mechanism.
 	dirtyUniforms = DIRTY_ALL;
+	use();
 }
 
 LinkedShader::~LinkedShader() {
@@ -264,8 +265,10 @@ void LinkedShader::updateUniforms() {
 
 	// Texturing
 	if (u_uvscaleoffset != -1 && (dirtyUniforms & DIRTY_UVSCALEOFFSET)) {
-		float uvscaleoff[4] = { gstate_c.uScale, gstate_c.vScale, gstate_c.uOff, gstate_c.vOff};
+		float uvscaleoff[4] = {gstate_c.uScale, gstate_c.vScale, gstate_c.uOff, gstate_c.vOff};
 		if (gstate.isModeThrough()) {
+			// We never get here because we don't use HW transform with through mode.
+			// Although - why don't we?
 			uvscaleoff[0] /= gstate_c.curTextureWidth;
 			uvscaleoff[1] /= gstate_c.curTextureHeight;
 			uvscaleoff[2] /= gstate_c.curTextureWidth;

GPU/GLES/TextureCache.cpp

@@ -24,9 +24,20 @@
 #include "TextureCache.h"
 #include "../Core/Config.h"
 
-// If a texture hasn't been seen for 200 frames, get rid of it.
+// If a texture hasn't been seen for this many frames, get rid of it.
 #define TEXTURE_KILL_AGE 200
-float maxAnisotropyLevel ;
+
+u32 RoundUpToPowerOf2(u32 v)
+{
+	v--;
+	v |= v >> 1;
+	v |= v >> 2;
+	v |= v >> 4;
+	v |= v >> 8;
+	v |= v >> 16;
+	v++;
+	return v;
+}
 
 TextureCache::TextureCache() {
 	lastBoundTexture = -1;
@@ -681,22 +692,28 @@ void TextureCache::SetTexture() {
 
 	TexCache::iterator iter = cache.find(cachekey);
 	TexCacheEntry *entry = NULL;
+	gstate_c.flipTexture = false;
+
 	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);
 			UpdateSamplingParams(*entry, false);
 
 			int fbow, fboh;
 			fbo_get_dimensions(entry->fbo, &fbow, &fboh);
 
-			// Almost certain this isn't right.
-			gstate_c.curTextureWidth = fbow;
-			gstate_c.curTextureHeight = fboh;
+			// This isn't right.
+			gstate_c.curTextureWidth = w; //w;  //RoundUpToPowerOf2(fbow);
+			gstate_c.curTextureHeight = h;  // RoundUpToPowerOf2(fboh);
+			gstate_c.flipTexture = true;
+			entry->lastFrame = gpuStats.numFrames;
 			return;
 		}
-
 		//Validate the texture here (width, height etc)
 
 		int dim = gstate.texsize[0] & 0xF0F;

GPU/GLES/TextureCache.h

@@ -101,5 +101,6 @@ private:
 	u16 *clutBuf16;
 
 	int lastBoundTexture;
+	float maxAnisotropyLevel;
 };
 

GPU/GLES/TransformPipeline.cpp

@@ -642,6 +642,8 @@ void TransformDrawEngine::SoftwareTransformAndDraw(
 		memcpy(&transformed[index].x, v, 3 * sizeof(float));
 		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;
 		for (int i = 0; i < 4; i++) {
 			transformed[index].color0[i] = c0[i] * 255.0f;
 		}

GPU/GLES/VertexShaderGenerator.cpp

@@ -60,6 +60,9 @@ void ComputeVertexShaderID(VertexShaderID *id, int prim)
 	id->d[0] |= ((int)enableFog) << 2;
 	id->d[0] |= doTexture << 3;
 	id->d[0] |= (hasColor & 1) << 4;
+	if (doTexture)
+		id->d[0] |= (gstate_c.flipTexture & 1) << 5;
+
 	if (CanUseHardwareTransform(prim)) {
 		id->d[0] |= 1 << 8;
 		id->d[0] |= (hasNormal & 1) << 9;
@@ -139,6 +142,7 @@ void GenerateVertexShader(int prim, char *buffer) {
 	bool hasColor = (gstate.vertType & GE_VTYPE_COL_MASK) != 0 || !hwXForm;
 	bool hasNormal = (gstate.vertType & GE_VTYPE_NRM_MASK) != 0 && hwXForm;
 	bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear();
+	bool flipV = gstate_c.flipTexture;
 
 	DoLightComputation doLight[4] = {LIGHT_OFF, LIGHT_OFF, LIGHT_OFF, LIGHT_OFF};
 	if (hwXForm) {
@@ -396,6 +400,8 @@ void GenerateVertexShader(int prim, char *buffer) {
 				// ILLEGAL
 				break;
 			}
+			if (flipV)
+				WRITE(p, "  v_texcoord.y = 1.0f - v_texcoord.y;\n");
 		}
 
 		// Compute fogdepth

GPU/GPUState.h

@@ -238,6 +238,8 @@ struct GPUStateCache
 
 	float uScale,vScale;
 	float uOff,vOff;
+	bool flipTexture;
+
 	float zMin, zMax;
 	float lightpos[4][3];
 	float lightdir[4][3];