GPU: Fix assorted leaks, null derefs and small rendering bugs

- Put the anisotropy level in the sampler key, so changing it applies on
  Vulkan and D3D11.
- Release CLUT textures at shutdown on GLES and D3D11.
- Fix the depth readback viewport, which squeezed the image whenever the
  read rectangle was smaller than the fbo.
- Test the computed depth, not the unset result, in the equal-depth clear
  check.
- Don't read back a CLUT from a framebuffer without an fbo.
- Tolerate null entries when releasing post-shader objects and CLUT
  textures after a failed creation.
- ImGe: Don't crash on a framebuffer without an fbo, or on GetVFB under the
  software renderer.

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

GPU/Common/DepthBufferCommon.cpp

@@ -199,7 +199,9 @@ bool FramebufferManagerCommon::ReadbackDepthbuffer(Draw::Framebuffer *fbo, int x
 
 		auto *blitFBO = GetTempFBO(TempFBO::Z_COPY, fbo->Width() * scaleX, fbo->Height() * scaleY);
 		draw_->BindFramebufferAsRenderTarget(blitFBO, { RPAction::DONT_CARE, RPAction::DONT_CARE, RPAction::DONT_CARE }, "ReadbackDepthbufferSync");
-		Draw::Viewport viewport = { 0.0f, 0.0f, (float)destW, (float)destH, 0.0f, 1.0f };
+		// The whole fbo is drawn, so the viewport has to cover all of it at the destination scale. Not just
+		// destW x destH, which would squeeze it whenever the read rectangle is smaller than the fbo.
+		Draw::Viewport viewport = { 0.0f, 0.0f, fbo->Width() * scaleX, fbo->Height() * scaleY, 0.0f, 1.0f };
 		draw_->SetViewport(viewport);
 		draw_->SetScissorRect(0, 0, fbo->Width() * scaleX, fbo->Height() * scaleY);
 

GPU/Common/FramebufferManagerCommon.cpp

@@ -3354,7 +3354,8 @@ void FramebufferManagerCommon::FlushBeforeCopy() {
 // TODO: Replace with with depal, reading the palette from the texture on the GPU directly.
 void FramebufferManagerCommon::DownloadFramebufferForClut(u32 fb_address, u32 loadBytes) {
 	VirtualFramebuffer *vfb = GetVFBAt(fb_address);
-	if (vfb && vfb->fb_stride != 0) {
+	// Without an fbo there's nothing to read back (ReadbackFramebuffer would read the backbuffer instead).
+	if (vfb && vfb->fb_stride != 0 && vfb->fbo) {
 		const u32 bpp = BufferFormatBytesPerPixel(vfb->fb_format);
 		int x = 0;
 		int y = 0;

GPU/Common/PresentationCommon.cpp

@@ -555,8 +555,12 @@ static void DoRelease(T *&obj) {
 
 template <typename T>
 static void DoReleaseVector(std::vector<T *> &list) {
-	for (auto &obj : list)
-		obj->Release();
+	for (auto &obj : list) {
+		// Can be null when a creation failed partway.
+		if (obj) {
+			obj->Release();
+		}
+	}
 	list.clear();
 }
 

GPU/Common/SoftwareTransformCommon.cpp

@@ -198,7 +198,7 @@ SoftwareTransformAction RunSoftwareTransform(SoftwareTransformParams &params, in
 				float depth = std::clamp(transformed[1].z, 0.0f, 65535.0f) / 65535.0f;
 				// Non-zero depth clears are unusual, but some drivers don't match drawn depth values to cleared values.
 				// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
-				if (!(params.everUsedEqualDepth && gstate.isClearModeDepthMask() && result->depth > 0.0f && result->depth < 1.0f)) {
+				if (!(params.everUsedEqualDepth && gstate.isClearModeDepthMask() && depth > 0.0f && depth < 1.0f)) {
 					result->color = transformed[1].color0_32;
 					result->depth = depth;
 					gpuStats.perFrame.numClears++;

GPU/Common/TextureCacheCommon.cpp

@@ -114,6 +114,8 @@ TextureCacheCommon::TextureCacheCommon(Draw::DrawContext *draw, Draw2D *draw2D)
 }
 
 TextureCacheCommon::~TextureCacheCommon() {
+	// Only DeviceLost cleared these, which the GLES and D3D11 backends don't go through at shutdown.
+	clutTextureCache_.Clear();
 	FreeAlignedMemory(clutBufConverted_);
 	FreeAlignedMemory(clutBufRaw_);
 	FreeAlignedMemory(expandClut_);
@@ -342,6 +344,9 @@ SamplerCacheKey TextureCacheCommon::GetSamplingParams(int maxLevel, const TexCac
 		break;
 	}
 
+	if (key.aniso) {
+		key.anisoLevel = (uint8_t)g_Config.iAnisotropyLevel;
+	}
 	return key;
 }
 

GPU/Common/TextureCacheCommon.h

@@ -86,6 +86,8 @@ struct SamplerCacheKey {
 			bool tClamp : 1;
 			bool aniso : 1;
 			bool texture3d : 1;
+			// Baked into the sampler objects, so it has to be part of the key, or a change wouldn't apply.
+			uint8_t anisoLevel;
 		};
 	};
 	bool operator < (const SamplerCacheKey &other) const {

GPU/Common/TextureShaderCommon.cpp

@@ -252,7 +252,9 @@ ClutTexture ClutTextureCache::GetClutTexture(GEPaletteFormat clutFormat, const u
 
 void ClutTextureCache::Clear() {
 	for (auto tex = texCache_.begin(); tex != texCache_.end(); ++tex) {
-		tex->second->texture->Release();
+		if (tex->second->texture) {
+			tex->second->texture->Release();
+		}
 		delete tex->second;
 	}
 	texCache_.clear();
@@ -261,7 +263,9 @@ void ClutTextureCache::Clear() {
 void ClutTextureCache::Decimate() {
 	for (auto tex = texCache_.begin(); tex != texCache_.end(); ) {
 		if (tex->second->lastFrame + DEPAL_TEXTURE_OLD_AGE < gpuStats.totals.numFlips) {
-			tex->second->texture->Release();
+			if (tex->second->texture) {
+				tex->second->texture->Release();
+			}
 			delete tex->second;
 			texCache_.erase(tex++);
 		} else {

GPU/D3D11/TextureCacheD3D11.cpp

@@ -90,7 +90,7 @@ HRESULT SamplerCacheD3D11::GetOrCreateSampler(ID3D11Device *device, const Sample
 	samp.AddressV = key.tClamp ? D3D11_TEXTURE_ADDRESS_CLAMP : D3D11_TEXTURE_ADDRESS_WRAP;
 	samp.AddressW = samp.AddressU;  // Mali benefits from all clamps being the same, and this one is irrelevant.
 	if (key.aniso) {
-		samp.MaxAnisotropy = (float)(1 << g_Config.iAnisotropyLevel);
+		samp.MaxAnisotropy = (float)(1 << key.anisoLevel);
 	} else {
 		samp.MaxAnisotropy = 1.0f;
 	}

GPU/Vulkan/TextureCacheVulkan.cpp

@@ -143,7 +143,7 @@ VkSampler SamplerCache::GetOrCreateSampler(const SamplerCacheKey &key) {
 
 	if (key.aniso) {
 		// Docs say the min of this value and the supported max are used.
-		samp.maxAnisotropy = 1 << g_Config.iAnisotropyLevel;
+		samp.maxAnisotropy = 1 << key.anisoLevel;
 		samp.anisotropyEnable = true;
 	} else {
 		samp.maxAnisotropy = 1.0f;

UI/ImDebugger/ImGe.cpp

@@ -96,9 +96,14 @@ void DrawFramebuffersWindow(ImConfig &cfg, FramebufferManagerCommon *framebuffer
 		ImGui::SliderFloat("Scale", &cfg.fbViewerZoom, 0.5f, 16.0f, "%.2f", ImGuiSliderFlags_Logarithmic);
 
 		// Now, draw the image of the selected framebuffer.
+		// Null without buffered rendering, or if creating it failed.
 		Draw::Framebuffer *fb = vfbs[cfg.selectedFramebuffer]->fbo;
-		ImTextureID texId = ImGui_ImplThin3d_AddFBAsTextureTemp(fb, Draw::Aspect::COLOR_BIT, ImGuiPipeline::TexturedOpaque);
-		ImGui::Image(texId, ImVec2(fb->Width() * cfg.fbViewerZoom, fb->Height() * cfg.fbViewerZoom));
+		if (fb) {
+			ImTextureID texId = ImGui_ImplThin3d_AddFBAsTextureTemp(fb, Draw::Aspect::COLOR_BIT, ImGuiPipeline::TexturedOpaque);
+			ImGui::Image(texId, ImVec2(fb->Width() * cfg.fbViewerZoom, fb->Height() * cfg.fbViewerZoom));
+		} else {
+			ImGui::TextUnformatted("(no framebuffer object)");
+		}
 	}
 
 	ImGui::End();
@@ -602,7 +607,8 @@ ImGeReadbackViewer::~ImGeReadbackViewer() {
 }
 
 VirtualFramebuffer *ImGeReadbackViewer::GetVFB(FramebufferManagerCommon *fbMan) const {
-	return fbMan->GetExactVFB(gstate.getFrameBufAddress(), gstate.FrameBufStride(), gstate.FrameBufFormat());
+	// fbMan is null with the software renderer.
+	return fbMan ? fbMan->GetExactVFB(gstate.getFrameBufAddress(), gstate.FrameBufStride(), gstate.FrameBufFormat()) : nullptr;
 }
 
 void ImGeReadbackViewer::DeviceLost() {