Merge pull request #8761 from unknownbrackets/gpu-minor

Assorted minor GPU bug fixes

GPU/Directx9/DrawEngineDX9.cpp

@@ -244,16 +244,6 @@ IDirect3DVertexDeclaration9 *DrawEngineDX9::SetupDecFmtForDraw(VSShader *vshader
 	}
 }
 
-VertexDecoder *DrawEngineDX9::GetVertexDecoder(u32 vtype) {
-	auto iter = decoderMap_.find(vtype);
-	if (iter != decoderMap_.end())
-		return iter->second;
-	VertexDecoder *dec = new VertexDecoder();
-	dec->SetVertexType(vtype, decOptions_, decJitCache_);
-	decoderMap_[vtype] = dec;
-	return dec;
-}
-
 void DrawEngineDX9::SetupVertexDecoder(u32 vertType) {
 	SetupVertexDecoderInternal(vertType);
 }

GPU/Directx9/DrawEngineDX9.h

@@ -201,8 +201,6 @@ private:
 	ReliableHashType ComputeHash();  // Reads deferred vertex data.
 	void MarkUnreliable(VertexArrayInfoDX9 *vai);
 
-	VertexDecoder *GetVertexDecoder(u32 vtype);
-
 	// Defer all vertex decoding to a Flush, so that we can hash and cache the
 	// generated buffers without having to redecode them every time.
 	struct DeferredDrawCall {

GPU/Directx9/TextureCacheDX9.cpp

@@ -649,28 +649,9 @@ void TextureCacheDX9::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFramebu
 	framebuffer->usageFlags |= FB_USAGE_TEXTURE;
 	bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
 	if (useBufferedRendering) {
-		const GEPaletteFormat clutFormat = gstate.getClutPaletteFormat();
 		const u64 cachekey = entry->CacheKey();
 		const auto &fbInfo = fbTexInfo_[cachekey];
 
-		LPDIRECT3DPIXELSHADER9 pshader = nullptr;
-		if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
-			pshader = depalShaderCache_->GetDepalettizePixelShader(clutFormat, framebuffer->drawnFormat);
-		}
-		if (pshader) {
-			const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
-			const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
-
-			TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormat(clutFormat), clutTotalColors, clutTotalColors, 1);
-			gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
-			gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
-		} else {
-			entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
-
-			gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
-			gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
-		}
-
 		// Keep the framebuffer alive.
 		framebuffer->last_frame_used = gpuStats.numFlips;
 
@@ -694,6 +675,10 @@ void TextureCacheDX9::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFramebu
 		pD3Ddevice->SetTexture(0, NULL);
 		gstate_c.needShaderTexClamp = false;
 	}
+
+	nextNeedsRehash_ = false;
+	nextNeedsChange_ = false;
+	nextNeedsRebuild_ = false;
 }
 
 void TextureCacheDX9::ApplyTexture() {
@@ -911,8 +896,20 @@ void TextureCacheDX9::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFrame
 		shaderApply.Shade();
 
 		fbo_bind_color_as_texture(depalFBO, 0);
+
+		const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
+		const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
+
+		TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormat(clutFormat), clutTotalColors, clutTotalColors, 1);
+		gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
+		gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
 	} else {
+		entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
+
 		framebufferManager_->BindFramebufferColor(0, framebuffer, BINDFBCOLOR_MAY_COPY_WITH_UV | BINDFBCOLOR_APPLY_TEX_OFFSET);
+
+		gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
+		gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
 	}
 
 	framebufferManager_->RebindFramebuffer();
@@ -1148,8 +1145,22 @@ void TextureCacheDX9::SetTexture(bool force) {
 	gstate_c.curTextureWidth = w;
 	gstate_c.curTextureHeight = h;
 
+	// Before we go reading the texture from memory, let's check for render-to-texture.
+	// We must do this early so we have the right w/h.
+	entry->framebuffer = 0;
+	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
+		auto framebuffer = fbCache_[i];
+		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
+	}
+
+	// If we ended up with a framebuffer, attach it - no texture decoding needed.
+	if (entry->framebuffer) {
+		SetTextureFramebuffer(entry, entry->framebuffer);
+	}
+
 	nextTexture_ = entry;
-	nextNeedsRehash_ = true;
+	nextNeedsRehash_ = entry->framebuffer == nullptr;
+	// We still need to rebuild, to allocate a texture.  But we'll bail early.
 	nextNeedsRebuild_= true;
 }
 
@@ -1263,16 +1274,8 @@ void TextureCacheDX9::BuildTexture(TexCacheEntry *const entry, bool replaceImage
 	// TODO: If a framebuffer is attached here, might end up with a bad entry.texture.
 	// Should just always create one here or something (like GLES.)
 
-	// Before we go reading the texture from memory, let's check for render-to-texture.
-	entry->framebuffer = 0;
-	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
-		auto framebuffer = fbCache_[i];
-		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
-	}
-
-	// If we ended up with a framebuffer, attach it - no texture decoding needed.
 	if (entry->framebuffer) {
-		SetTextureFramebuffer(entry, entry->framebuffer);
+		// Nothing else to do here.
 		return;
 	}
 

GPU/GLES/DrawEngineGLES.cpp

@@ -847,13 +847,6 @@ rotateVBO:
 			prim = indexGen.Prim();
 		}
 
-		if (gstate_c.Supports(GPU_SUPPORTS_VAO) && vbo == 0) {
-			vbo = BindBuffer(decoded, dec_->GetDecVtxFmt().stride * indexGen.MaxIndex());
-			if (useElements) {
-				ebo = BindElementBuffer(decIndex, sizeof(short) * indexGen.VertexCount());
-			}
-		}
-
 		VERBOSE_LOG(G3D, "Flush prim %i! %i verts in one go", prim, vertexCount);
 		bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
 		if (gstate.isModeThrough()) {
@@ -863,6 +856,14 @@ rotateVBO:
 		}
 
 		ApplyDrawStateLate();
+
+		if (gstate_c.Supports(GPU_SUPPORTS_VAO) && vbo == 0) {
+			vbo = BindBuffer(decoded, dec_->GetDecVtxFmt().stride * indexGen.MaxIndex());
+			if (useElements) {
+				ebo = BindElementBuffer(decIndex, sizeof(short) * indexGen.VertexCount());
+			}
+		}
+
 		LinkedShader *program = shaderManager_->ApplyFragmentShader(vsid, vshader, lastVType_, prim);
 		SetupDecFmtForDraw(program, dec_->GetDecVtxFmt(), vbo ? 0 : decoded);
 

GPU/GLES/Framebuffer.cpp

@@ -882,7 +882,7 @@ void FramebufferManager::BindFramebufferColor(int stage, u32 fbRawAddress, Virtu
 	}
 
 	if (stage != GL_TEXTURE0) {
-		glActiveTexture(stage);
+		glActiveTexture(GL_TEXTURE0);
 	}
 }
 

GPU/GLES/TextureCache.cpp

@@ -719,28 +719,9 @@ void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuffe
 	framebuffer->usageFlags |= FB_USAGE_TEXTURE;
 	bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
 	if (useBufferedRendering) {
-		const GEPaletteFormat clutFormat = gstate.getClutPaletteFormat();
 		const u64 cachekey = entry->CacheKey();
 		const auto &fbInfo = fbTexInfo_[cachekey];
 
-		DepalShader *depal = nullptr;
-		if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
-			depal = depalShaderCache_->GetDepalettizeShader(clutFormat, framebuffer->drawnFormat);
-		}
-		if (depal) {
-			const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
-			const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
-
-			TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormat(clutFormat), clutTotalColors, clutTotalColors, 1);
-			gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
-			gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
-		} else {
-			entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
-
-			gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
-			gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
-		}
-
 		// Keep the framebuffer alive.
 		framebuffer->last_frame_used = gpuStats.numFlips;
 
@@ -763,6 +744,10 @@ void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuffe
 		glBindTexture(GL_TEXTURE_2D, 0);
 		gstate_c.needShaderTexClamp = false;
 	}
+
+	nextNeedsRehash_ = false;
+	nextNeedsChange_ = false;
+	nextNeedsRebuild_ = false;
 }
 
 void TextureCache::ApplyTexture() {
@@ -985,8 +970,20 @@ void TextureCache::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuf
 		shaderApply.Shade();
 
 		fbo_bind_color_as_texture(depalFBO, 0);
+
+		const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
+		const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
+
+		TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormat(clutFormat), clutTotalColors, clutTotalColors, 1);
+		gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
+		gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
 	} else {
+		entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
+
 		framebufferManager_->BindFramebufferColor(GL_TEXTURE0, gstate.getFrameBufRawAddress(), framebuffer, BINDFBCOLOR_MAY_COPY_WITH_UV | BINDFBCOLOR_APPLY_TEX_OFFSET);
+
+		gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
+		gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
 	}
 
 	framebufferManager_->RebindFramebuffer();
@@ -1250,8 +1247,22 @@ void TextureCache::SetTexture(bool force) {
 	gstate_c.curTextureWidth = w;
 	gstate_c.curTextureHeight = h;
 
+	// Before we go reading the texture from memory, let's check for render-to-texture.
+	// We must do this early so we have the right w/h.
+	entry->framebuffer = 0;
+	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
+		auto framebuffer = fbCache_[i];
+		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
+	}
+
+	// If we ended up with a framebuffer, attach it - no texture decoding needed.
+	if (entry->framebuffer) {
+		SetTextureFramebuffer(entry, entry->framebuffer);
+	}
+
 	nextTexture_ = entry;
-	nextNeedsRehash_ = true;
+	nextNeedsRehash_ = entry->framebuffer == nullptr;
+	// We still need to rebuild, to allocate a texture.  But we'll bail early.
 	nextNeedsRebuild_= true;
 }
 
@@ -1367,16 +1378,8 @@ void TextureCache::BuildTexture(TexCacheEntry *const entry, bool replaceImages)
 		entry->textureName = AllocTextureName();
 	}
 
-	// Before we go reading the texture from memory, let's check for render-to-texture.
-	entry->framebuffer = 0;
-	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
-		auto framebuffer = fbCache_[i];
-		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
-	}
-
-	// If we ended up with a framebuffer, attach it - no texture decoding needed.
 	if (entry->framebuffer) {
-		SetTextureFramebuffer(entry, entry->framebuffer);
+		// Nothing else to do here.
 		return;
 	}
 

GPU/Vulkan/DrawEngineVulkan.cpp

@@ -268,16 +268,6 @@ void DrawEngineVulkan::EndFrame() {
 	curFrame_++;
 }
 
-VertexDecoder *DrawEngineVulkan::GetVertexDecoder(u32 vtype) {
-	auto iter = decoderMap_.find(vtype);
-	if (iter != decoderMap_.end())
-		return iter->second;
-	VertexDecoder *dec = new VertexDecoder();
-	dec->SetVertexType(vtype, decOptions_, decJitCache_);
-	decoderMap_[vtype] = dec;
-	return dec;
-}
-
 void DrawEngineVulkan::SetupVertexDecoder(u32 vertType) {
 	SetupVertexDecoderInternal(vertType);
 }

GPU/Vulkan/DrawEngineVulkan.h

@@ -168,8 +168,6 @@ private:
 
 	VkDescriptorSet GetDescriptorSet(VkImageView imageView, VkSampler sampler, VkBuffer base, VkBuffer light, VkBuffer bone);
 
-	VertexDecoder *GetVertexDecoder(u32 vtype);
-
 	VulkanContext *vulkan_;
 
 	// We use a single descriptor set layout for all PSP draws.

GPU/Vulkan/TextureCacheVulkan.cpp

@@ -659,28 +659,9 @@ void TextureCacheVulkan::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFram
 	framebuffer->usageFlags |= FB_USAGE_TEXTURE;
 	bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
 	if (useBufferedRendering) {
-		const GEPaletteFormat clutFormat = gstate.getClutPaletteFormat();
 		const u64 cachekey = entry->CacheKey();
 		const auto &fbInfo = fbTexInfo_[cachekey];
 
-		DepalShaderVulkan *depal = nullptr;
-		if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
-			// depal = depalShaderCache_->GetDepalettizeShader(clutFormat, framebuffer->drawnFormat);
-		}
-		if (depal) {
-			const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
-			const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
-
-			TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormatVulkan(clutFormat), clutTotalColors, clutTotalColors, 1);
-			gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
-			gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
-		} else {
-			entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
-
-			gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
-			gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
-		}
-
 		// Keep the framebuffer alive.
 		framebuffer->last_frame_used = gpuStats.numFlips;
 
@@ -702,6 +683,10 @@ void TextureCacheVulkan::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFram
 		}
 		gstate_c.needShaderTexClamp = false;
 	}
+
+	nextNeedsRehash_ = false;
+	nextNeedsChange_ = false;
+	nextNeedsRebuild_ = false;
 }
 
 void TextureCacheVulkan::ApplyTexture(VulkanPushBuffer *uploadBuffer, VkImageView &imageView, VkSampler &sampler) {
@@ -847,6 +832,18 @@ void TextureCacheVulkan::ApplyTextureFramebuffer(VkCommandBuffer cmd, TexCacheEn
 		//depalFBO->EndPass(cmd);
 		//depalFBO->TransitionToTexture(cmd);
 		//imageView = depalFBO->GetColorImageView();
+
+		const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
+		const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
+
+		TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormatVulkan(clutFormat), clutTotalColors, clutTotalColors, 1);
+		gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
+		gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
+	} else {
+		entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
+
+		gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
+		gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
 	}
 
 	/*
@@ -1120,8 +1117,22 @@ void TextureCacheVulkan::SetTexture() {
 	gstate_c.curTextureWidth = w;
 	gstate_c.curTextureHeight = h;
 
+	// Before we go reading the texture from memory, let's check for render-to-texture.
+	// We must do this early so we have the right w/h.
+	entry->framebuffer = 0;
+	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
+		auto framebuffer = fbCache_[i];
+		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
+	}
+
+	// If we ended up with a framebuffer, attach it - no texture decoding needed.
+	if (entry->framebuffer) {
+		SetTextureFramebuffer(entry, entry->framebuffer);
+	}
+
 	nextTexture_ = entry;
-	nextNeedsRehash_ = true;
+	nextNeedsRehash_ = entry->framebuffer == nullptr;
+	// We still need to rebuild, to allocate a texture.  But we'll bail early.
 	nextNeedsRebuild_= true;
 }
 
@@ -1225,16 +1236,8 @@ void TextureCacheVulkan::BuildTexture(TexCacheEntry *const entry,VulkanPushBuffe
 	// For the estimate, we assume cluts always point to 8888 for simplicity.
 	cacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
 
-	// Before we go reading the texture from memory, let's check for render-to-texture.
-	entry->framebuffer = 0;
-	for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
-		auto framebuffer = fbCache_[i];
-		AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
-	}
-
-	// If we ended up with a framebuffer, attach it - no texture decoding needed.
 	if (entry->framebuffer) {
-		SetTextureFramebuffer(entry, entry->framebuffer);
+		// Nothing else to do here.
 		return;
 	}