Merge pull request #6202 from unknownbrackets/gpu-minor

Decimate in clut stuff, adjustments to IsReallyAClear()

GPU/GLES/DepalettizeShader.cpp

@@ -24,6 +24,8 @@
 #include "GPU/GPUState.h"
 #include "GPU/GLES/TextureCache.h"
 
+static const int DEPAL_TEXTURE_OLD_AGE = 120;
+
 #ifdef _WIN32
 #define SHADERLOG
 #endif
@@ -95,7 +97,7 @@ DepalShaderCache::DepalShaderCache() {
 	glShaderSource(vertexShader_, 1, useGL3_ ? &depalVShader300 : &depalVShader100, 0);
 	glCompileShader(vertexShader_);
 
-	if (CheckShaderCompileSuccess(vertexShader_, depalVShader100)) {
+	if (!CheckShaderCompileSuccess(vertexShader_, useGL3_ ? depalVShader300 : depalVShader100)) {
 		// ...
 	}
 }
@@ -322,6 +324,7 @@ GLuint DepalShaderCache::GetClutTexture(const u32 clutID, u32 *rawClut) {
 
 	auto oldtex = texCache_.find(realClutID);
 	if (oldtex != texCache_.end()) {
+		oldtex->second->lastFrame = gpuStats.numFlips;
 		return oldtex->second->texture;
 	}
 
@@ -348,6 +351,7 @@ GLuint DepalShaderCache::GetClutTexture(const u32 clutID, u32 *rawClut) {
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 
+	tex->lastFrame = gpuStats.numFlips;
 	texCache_[realClutID] = tex;
 	return tex->texture;
 }
@@ -367,7 +371,15 @@ void DepalShaderCache::Clear() {
 }
 
 void DepalShaderCache::Decimate() {
-	// TODO
+	for (auto tex = texCache_.begin(); tex != texCache_.end(); ) {
+		if (tex->second->lastFrame + DEPAL_TEXTURE_OLD_AGE < gpuStats.numFlips) {
+			glDeleteTextures(1, &tex->second->texture);
+			delete tex->second;
+			texCache_.erase(tex++);
+		} else {
+			++tex;
+		}
+	}
 }
 
 GLuint DepalShaderCache::GetDepalettizeShader(GEBufferFormat pixelFormat) {

GPU/GLES/DepalettizeShader.h

@@ -30,6 +30,7 @@ public:
 class DepalTexture {
 public:
 	GLuint texture;
+	int lastFrame;
 };
 
 // Caches both shaders and palette textures.

GPU/GLES/FragmentShaderGenerator.cpp

@@ -202,7 +202,6 @@ StencilValueType ReplaceAlphaWithStencilType() {
 				return STENCIL_VALUE_KEEP;
 			}
 
-		// Decrementing always zeros, since there's only one bit.
 		case GE_STENCILOP_DECR:
 		case GE_STENCILOP_INCR:
 		case GE_STENCILOP_INVERT:

GPU/GLES/Framebuffer.cpp

@@ -521,6 +521,8 @@ void FramebufferManager::DrawPlainColor(u32 color) {
 		((color & 0xFF000000) >> 24) / 255.0f,
 	};
 
+	shaderManager_->DirtyLastShader();
+
 	glsl_bind(program);
 	glUniform4fv(plainColorLoc_, 1, col);
 	glBindBuffer(GL_ARRAY_BUFFER, 0);
@@ -588,6 +590,8 @@ void FramebufferManager::DrawActiveTexture(GLuint texture, float x, float y, flo
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
 
+	shaderManager_->DirtyLastShader();  // dirty lastShader_
+
 	glsl_bind(program);
 	if (program == postShaderProgram_ && timeLoc_ != -1) {
 		int flipCount = __DisplayGetFlipCount();
@@ -606,8 +610,6 @@ void FramebufferManager::DrawActiveTexture(GLuint texture, float x, float y, flo
 	glDisableVertexAttribArray(program->a_texcoord0);
 
 	glsl_unbind();
-
-	shaderManager_->DirtyLastShader();  // dirty lastShader_
 }
 
 

GPU/GLES/GLES_GPU.cpp

@@ -411,6 +411,7 @@ GLES_GPU::GLES_GPU()
 	framebufferManager_.SetShaderManager(shaderManager_);
 	textureCache_.SetFramebufferManager(&framebufferManager_);
 	textureCache_.SetDepalShaderCache(&depalShaderCache_);
+	textureCache_.SetShaderManager(shaderManager_);
 
 	// Sanity check gstate
 	if ((int *)&gstate.transferstart - (int *)&gstate != 0xEA) {
@@ -545,6 +546,7 @@ void GLES_GPU::BeginFrameInternal() {
 
 	textureCache_.StartFrame();
 	transformDraw_.DecimateTrackedVertexArrays();
+	depalShaderCache_.Decimate();
 
 	if (dumpNextFrame_) {
 		NOTICE_LOG(G3D, "DUMPING THIS FRAME");

GPU/GLES/ShaderManager.cpp

@@ -684,11 +684,10 @@ Shader *ShaderManager::ApplyVertexShader(int prim, u32 vertType) {
 LinkedShader *ShaderManager::ApplyFragmentShader(Shader *vs, int prim, u32 vertType) {
 	FragmentShaderID FSID;
 	ComputeFragmentShaderID(&FSID);
-	if (lastVShaderSame_ && FSID == lastFSID_ && !gstate_c.shaderChanged) {
+	if (lastVShaderSame_ && FSID == lastFSID_) {
 		lastShader_->UpdateUniforms(vertType);
 		return lastShader_;
 	}
-	gstate_c.shaderChanged = false;
 
 	lastFSID_ = FSID;
 

GPU/GLES/SoftwareTransform.cpp

@@ -23,6 +23,7 @@
 #include "GPU/Math3D.h"
 #include "GPU/Common/VertexDecoderCommon.h"
 #include "GPU/Common/TransformCommon.h"
+#include "GPU/GLES/Framebuffer.h"
 #include "GPU/GLES/ShaderManager.h"
 #include "GPU/GLES/TransformPipeline.h"
 
@@ -88,8 +89,7 @@ bool TransformDrawEngine::IsReallyAClear(int numVerts) const {
 	if (transformed[0].x != 0.0f || transformed[0].y != 0.0f)
 		return false;
 
-	u32 matchcolor;
-	memcpy(&matchcolor, transformed[0].color0, 4);
+	u32 matchcolor = transformed[0].color0_32;
 	float matchz = transformed[0].z;
 
 	int bufW = gstate_c.curRTWidth;
@@ -97,9 +97,7 @@ bool TransformDrawEngine::IsReallyAClear(int numVerts) const {
 
 	float prevX = 0.0f;
 	for (int i = 1; i < numVerts; i++) {
-		u32 vcolor;
-		memcpy(&vcolor, transformed[i].color0, 4);
-		if (vcolor != matchcolor || transformed[i].z != matchz)
+		if (transformed[i].color0_32 != matchcolor || transformed[i].z != matchz)
 			return false;
 
 		if ((i & 1) == 0) {
@@ -385,9 +383,8 @@ void TransformDrawEngine::SoftwareTransformAndDraw(
 	//
 	// An alternative option is to simply ditch all the verts except the first and last to create a single
 	// rectangle out of many. Quite a small optimization though.
-	if (false && maxIndex > 1 && gstate.isModeClear() && prim == GE_PRIM_RECTANGLES && IsReallyAClear(maxIndex)) {
-		u32 clearColor;
-		memcpy(&clearColor, transformed[0].color0, 4);
+	if (maxIndex > 1 && gstate.isModeClear() && prim == GE_PRIM_RECTANGLES && IsReallyAClear(maxIndex)) {
+		u32 clearColor = transformed[0].color0_32;
 		float clearDepth = transformed[0].z;
 		const float col[4] = {
 			((clearColor & 0xFF)) / 255.0f,
@@ -398,22 +395,17 @@ void TransformDrawEngine::SoftwareTransformAndDraw(
 
 		bool colorMask = gstate.isClearModeColorMask();
 		bool alphaMask = gstate.isClearModeAlphaMask();
-		glstate.colorMask.set(colorMask, colorMask, colorMask, alphaMask);
-		if (alphaMask) {
-			glstate.stencilTest.set(true);
-			// Clear stencil
-			// TODO: extract the stencilValue properly, see below
-			int stencilValue = 0;
-			glstate.stencilFunc.set(GL_ALWAYS, stencilValue, 255);
-		} else {
-			// Don't touch stencil
-			glstate.stencilTest.set(false);
-		}
-		glstate.scissorTest.set(false);
 		bool depthMask = gstate.isClearModeDepthMask();
+		if (depthMask) {
+			framebufferManager_->SetDepthUpdated();
+		}
 
-		int target = 0;
-		if (colorMask || alphaMask) target |= GL_COLOR_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
+		glstate.stencilTest.set(false);
+		glstate.scissorTest.set(false);
+
+		GLbitfield target = 0;
+		if (colorMask || alphaMask) target |= GL_COLOR_BUFFER_BIT;
+		if (alphaMask) target |= GL_STENCIL_BUFFER_BIT;
 		if (depthMask) target |= GL_DEPTH_BUFFER_BIT;
 
 		glClearColor(col[0], col[1], col[2], col[3]);
@@ -422,7 +414,8 @@ void TransformDrawEngine::SoftwareTransformAndDraw(
 #else
 		glClearDepth(clearDepth);
 #endif
-		glClearStencil(0);  // TODO - take from alpha?
+		// Stencil takes alpha.
+		glClearStencil(clearColor >> 24);
 		glClear(target);
 		return;
 	}

GPU/GLES/TextureCache.cpp

@@ -28,6 +28,7 @@
 #include "GPU/GLES/Framebuffer.h"
 #include "GPU/GLES/FragmentShaderGenerator.h"
 #include "GPU/GLES/DepalettizeShader.h"
+#include "GPU/GLES/ShaderManager.h"
 #include "GPU/Common/TextureDecoder.h"
 #include "Core/Config.h"
 #include "Core/Host.h"
@@ -916,13 +917,17 @@ void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry) {
 			};
 			static const GLubyte indices[4] = { 0, 1, 3, 2 };
 
+			shaderManager_->DirtyLastShader();
+
 			glUseProgram(program);
-			gstate_c.shaderChanged = true;
+
+			GLint a_position = glGetAttribLocation(program, "a_position");
+			GLint a_texcoord0 = glGetAttribLocation(program, "a_texcoord0");
 
 			glBindBuffer(GL_ARRAY_BUFFER, 0);
 			glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
-			glEnableVertexAttribArray(0);
-			glEnableVertexAttribArray(1);
+			glEnableVertexAttribArray(a_position);
+			glEnableVertexAttribArray(a_texcoord0);
 
 			glActiveTexture(GL_TEXTURE1);
 			glBindTexture(GL_TEXTURE_2D, clutTexture);
@@ -944,14 +949,12 @@ void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry) {
 #endif
 			glViewport(0, 0, entry->framebuffer->renderWidth, entry->framebuffer->renderHeight);
 
-			glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 12, pos);
-			glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 8, uv);
+			glVertexAttribPointer(a_position, 3, GL_FLOAT, GL_FALSE, 12, pos);
+			glVertexAttribPointer(a_texcoord0, 2, GL_FLOAT, GL_FALSE, 8, uv);
 			glDrawElements(GL_TRIANGLE_STRIP, 4, GL_UNSIGNED_BYTE, indices);
+			glDisableVertexAttribArray(a_position);
+			glDisableVertexAttribArray(a_texcoord0);
 
-			/*
-			glDisableVertexAttribArray(0);
-			glDisableVertexAttribArray(1);
-			*/
 			fbo_bind_color_as_texture(entry->depalFBO, 0);
 			glstate.Restore();
 			framebufferManager_->RebindFramebuffer();

GPU/GLES/TextureCache.h

@@ -28,6 +28,7 @@
 struct VirtualFramebuffer;
 class FramebufferManager;
 class DepalShaderCache;
+class ShaderManager;
 
 enum TextureFiltering {
 	AUTO = 1,
@@ -74,6 +75,9 @@ public:
 	void SetDepalShaderCache(DepalShaderCache *dpCache) {
 		depalShaderCache_ = dpCache;
 	}
+	void SetShaderManager(ShaderManager *sm) {
+		shaderManager_ = sm;
+	}
 
 	size_t NumLoadedTextures() const {
 		return cache.size();
@@ -203,6 +207,7 @@ private:
 	int decimationCounter_;
 	FramebufferManager *framebufferManager_;
 	DepalShaderCache *depalShaderCache_;
+	ShaderManager *shaderManager_;
 };
 
 GLenum getClutDestFormat(GEPaletteFormat format);

GPU/GPUState.h

@@ -453,8 +453,6 @@ struct GPUStateCache
 	bool textureSimpleAlpha;
 	bool vertexFullAlpha;
 	bool framebufChanged;
-	// Doesn't need savestating.
-	bool shaderChanged;
 
 	int skipDrawReason;