GPU: Fix texture cache and framebuffer cache lifetime bugs

- Delete TexCacheEntry objects dropped on rehash instead of leaking them.
- Don't leave a released null entry in cache_ when the framebuffer match
  returns before the slot is refilled.
- Reset clutRenderAddress_ in Clear(), which releases the dynamic CLUT FBOs.
- Don't cache a null texture in drawPixelsCache_ when creation fails.
- Fix the reversed subtraction in the failed-FBO retry check.
- Remove the never-taken buffered-rendering early-out in UpdateRenderSize.
  Taking it would leave existing VFBs without an fbo.
- Include smoothedDepal in the depal shader cache key, and print/parse the
  debug IDs as 64-bit.
- Release depal pipelines through Draw2DPipeline::Release so the shader
  source isn't leaked, and make that null-safe.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

GPU/Common/Draw2D.h

@@ -49,7 +49,9 @@ struct Draw2DPipeline {
 	Draw2DPipelineInfo info;
 	char *code;
 	void Release() {
-		pipeline->Release();
+		if (pipeline) {
+			pipeline->Release();
+		}
 		delete[] code;
 		delete this;
 	}

GPU/Common/FramebufferManagerCommon.cpp

@@ -103,10 +103,8 @@ bool FramebufferManagerCommon::UpdateRenderSize(int msaaLevel) {
 
 	presentation_->UpdateRenderSize(renderWidth_, renderHeight_);
 
-	// If just switching TO buffered rendering, no need to pause the threads. In fact this causes problems due to the open backbuffer renderpass.
-	if (!useBufferedRendering_ && newBuffered) {
-		return false;
-	}
+	// Switching to buffered rendering must also recreate the framebuffers: the existing VFBs have no fbo,
+	// and nothing else gives them one.
 	return newRender || newSettings;
 }
 
@@ -1531,6 +1529,7 @@ Draw::Texture *FramebufferManagerCommon::MakePixelTexture(const u8 *srcPixels, G
 	Draw::Texture *tex = draw_->CreateTexture(desc);
 	if (!tex) {
 		ERROR_LOG(Log::G3D, "Failed to create DrawPixels texture");
+		return nullptr;
 	}
 	// We don't need to count here, already counted by numUploads by the caller.
 
@@ -1876,7 +1875,7 @@ void FramebufferManagerCommon::ResizeFramebufFBO(VirtualFramebuffer *vfb, int w,
 		}
 		return;
 	}
-	if (!old.fbo && vfb->last_frame_failed != 0 && vfb->last_frame_failed - gpuStats.totals.numFlips < 63) {
+	if (!old.fbo && vfb->last_frame_failed != 0 && gpuStats.totals.numFlips - vfb->last_frame_failed < 63) {
 		// Don't constantly retry FBOs which failed to create.
 		return;
 	}

GPU/Common/TextureCacheCommon.cpp

@@ -757,12 +757,14 @@ TextureApplyResult TextureCacheCommon::ApplyTexture(bool doBind) {
 							DEBUG_LOG(Log::TexCache, "%08x: Second cache already had one with hash %08x (tex has addr %08x)! (%dx%d)", texaddr, entry->fullhash, entry->addr, w, h);
 							// Just release the old entry, drop it on the ground.
 							ReleaseTexture(entry, true);
+							delete entry;
 							entry = nullptr;
 						}
 					} else {
 						// Just release the old video entry, drop it on the ground.
 						VERBOSE_LOG(Log::TexCache, "%08x: Dropping old invalidated video image (%dx%d)", texaddr, w, h);
 						ReleaseTexture(entry, true);
+						delete entry;
 						entry = nullptr;
 					}
 
@@ -802,9 +804,9 @@ TextureApplyResult TextureCacheCommon::ApplyTexture(bool doBind) {
 							// other stuff like Gran Turismo or Gods Eater font rendering unless there are hash collisions..
 							DEBUG_LOG(Log::TexCache, "%08x: No entry for hash %08x in secondary cache, creating new in main cache.", texaddr, newFullHash);
 						}
-						// Well, not found, so we need to create a new entry.
-						cache_.erase(entryIter);
 					}
+					// The slot was released above. Erase it, since the framebuffer match below can return before it's refilled.
+					cache_.erase(entryIter);
 					entry = nullptr;
 					entryIter = cache_.end();
 				} else {
@@ -2740,6 +2742,8 @@ void TextureCacheCommon::Clear(bool delete_them) {
 		dynamicClutTemp_->Release();
 		dynamicClutTemp_ = nullptr;
 	}
+	// The dynamic CLUT lived in the framebuffers released above. Fall back to the RAM copy until the next LoadClut.
+	clutRenderAddress_ = 0xFFFFFFFF;
 }
 
 // One type of texture update can happen without the involvement of the CPU: Block transfers.

GPU/Common/TextureShaderCommon.cpp

@@ -79,10 +79,7 @@ Draw::SamplerState *TextureShaderCache::GetSampler(bool linearFilter) {
 
 void TextureShaderCache::Clear() {
 	for (auto shader = pipelineCache_.begin(); shader != pipelineCache_.end(); ++shader) {
-		if (shader->second->pipeline) {
-			shader->second->pipeline->Release();
-		}
-		delete shader->second;
+		shader->second->Release();
 	}
 	pipelineCache_.clear();
 	if (nearestSampler_) {
@@ -99,7 +96,7 @@ Draw2DPipeline *TextureShaderCache::GetDepalettizeShader(uint32_t clutMode, GETe
 	using namespace Draw;
 
 	// Generate an ID for depal shaders.
-	u64 id = ((u64)depthUpperBits << 32) | (clutMode & 0xFFFFFF) | (textureFormat << 24) | (bufferFormat << 28);
+	u64 id = ((u64)smoothedDepal << 63) | ((u64)depthUpperBits << 32) | (clutMode & 0xFFFFFF) | (textureFormat << 24) | (bufferFormat << 28);
 
 	auto shader = pipelineCache_.find(id);
 	if (shader != pipelineCache_.end()) {
@@ -135,14 +132,14 @@ Draw2DPipeline *TextureShaderCache::GetDepalettizeShader(uint32_t clutMode, GETe
 std::vector<std::string> TextureShaderCache::DebugGetShaderIDs(DebugShaderType type) const {
 	std::vector<std::string> ids;
 	for (auto &entry : pipelineCache_) {
-		ids.push_back(StringFromFormat("%08x", entry.first));
+		ids.push_back(StringFromFormat("%016llx", (unsigned long long)entry.first));
 	}
 	return ids;
 }
 
 std::string TextureShaderCache::DebugGetShaderString(const std::string &idstr, DebugShaderType type, DebugShaderStringType stringType) const {
-	uint32_t id = 0;
-	if (sscanf(idstr.c_str(), "%08x", &id) == 0) {
+	unsigned long long id = 0;
+	if (sscanf(idstr.c_str(), "%llx", &id) != 1) {
 		return "";
 	}
 	auto iter = pipelineCache_.find(id);