Just some basic code cleanup in TextureCacheCommon.

GPU/Common/TextureCacheCommon.cpp

@@ -63,6 +63,9 @@
 #define TEXCACHE_MIN_PRESSURE 16 * 1024 * 1024  // Total in VRAM
 #define TEXCACHE_SECOND_MIN_PRESSURE 4 * 1024 * 1024
 
+static bool MatchFramebuffer(const TextureDefinition &entry, VirtualFramebuffer *framebuffer, u32 texaddrOffset, RasterChannel channel, bool useBufferedRendering, FramebufferMatchInfo *matchInfo);
+static bool GetBestFramebufferCandidate(FramebufferManagerCommon *fbManager, const TextureDefinition &entry, u32 texAddrOffset, AttachCandidate *bestCandidate, const char *context);
+
 // Just for reference
 
 // PSP Color formats:
@@ -339,6 +342,7 @@ SamplerCacheKey TextureCacheCommon::GetSamplingParams(int maxLevel, const TexCac
 }
 
 SamplerCacheKey TextureCacheCommon::GetFramebufferSamplingParams(u16 bufferWidth, u16 bufferHeight) {
+	// TODO: This call is pretty pointless, we overwrite most of it.
 	SamplerCacheKey key = GetSamplingParams(0, nullptr, true);
 
 	// In case auto max quality was on, restore min filt. Another fix for water in Outrun.
@@ -358,6 +362,7 @@ SamplerCacheKey TextureCacheCommon::GetFramebufferSamplingParams(u16 bufferWidth
 	int w = gstate.getTextureWidth(0);
 	int h = gstate.getTextureHeight(0);
 	if (w != bufferWidth || h != bufferHeight) {
+		// Actually, maybe we should always do this?
 		key.sClamp = true;
 		key.tClamp = true;
 	}
@@ -626,7 +631,7 @@ TexCacheEntry *TextureCacheCommon::SetTexture() {
 	def.bufw = bufw;
 
 	AttachCandidate bestCandidate;
-	if (GetBestFramebufferCandidate(def, 0, &bestCandidate, "texture")) {
+	if (GetBestFramebufferCandidate(framebufferManager_, def, 0, &bestCandidate, "texture")) {
 		// If we had a texture entry here, let's get rid of it.
 		if (entryIter != cache_.end()) {
 			DeleteTexture(entryIter);
@@ -709,20 +714,22 @@ TexCacheEntry *TextureCacheCommon::SetTexture() {
 	return entry;
 }
 
-bool TextureCacheCommon::GetBestFramebufferCandidate(const TextureDefinition &entry, u32 texAddrOffset, AttachCandidate *bestCandidate, const char *context) const {
+static bool GetBestFramebufferCandidate(FramebufferManagerCommon *fbManager, const TextureDefinition &entry, u32 texAddrOffset, AttachCandidate *bestCandidate, const char *context) {
 	gpuStats.perFrame.numFramebufferEvaluations++;
 
 	TinySet<AttachCandidate, 6> candidates;
 
-	const std::vector<VirtualFramebuffer *> &framebuffers = framebufferManager_->Framebuffers();
+	const std::vector<VirtualFramebuffer *> &framebuffers = fbManager->Framebuffers();
+
+	const bool useBufferedRendering = fbManager->UseBufferedRendering();
 
 	for (VirtualFramebuffer *framebuffer : framebuffers) {
 		FramebufferMatchInfo match{};
-		if (MatchFramebuffer(entry, framebuffer, texAddrOffset, RASTER_COLOR, &match)) {
+		if (MatchFramebuffer(entry, framebuffer, texAddrOffset, RASTER_COLOR, useBufferedRendering, &match)) {
 			candidates.push_back(AttachCandidate{ framebuffer, match, RASTER_COLOR });
 		}
 		match = {};
-		if (MatchFramebuffer(entry, framebuffer, texAddrOffset, RASTER_DEPTH, &match)) {
+		if (MatchFramebuffer(entry, framebuffer, texAddrOffset, RASTER_DEPTH, useBufferedRendering, &match)) {
 			candidates.push_back(AttachCandidate{ framebuffer, match, RASTER_DEPTH });
 		}
 	}
@@ -768,7 +775,7 @@ bool TextureCacheCommon::GetBestFramebufferCandidate(const TextureDefinition &en
 		// Avoid binding as texture the framebuffer we're rendering to.
 		// In Killzone, we split the framebuffer but the matching algorithm can still pick the wrong one,
 		// which this avoids completely.
-		if (kzCompat && candidate.fb == framebufferManager_->GetCurrentRenderVFB()) {
+		if (kzCompat && candidate.fb == fbManager->GetCurrentRenderVFB()) {
 			continue;
 		}
 
@@ -853,7 +860,7 @@ void TextureCacheCommon::Decimate(TexCacheEntry *exceptThisOne, bool forcePressu
 			// In low memory mode, we kill them all since secondary cache is disabled.
 			if (lowMemoryMode_ || iter->second->lastFrame + TEXTURE_SECOND_KILL_AGE < gpuStats.totals.numFlips) {
 				ReleaseTexture(iter->second.get(), true);
-				secondCacheSizeEstimate_ -= EstimateTexMemoryUsage(iter->second.get());
+				secondCacheSizeEstimate_ -= iter->second->EstimateTexMemoryUsage();
 				iter = secondCache_.erase(iter);
 			} else {
 				++iter;
@@ -863,11 +870,7 @@ void TextureCacheCommon::Decimate(TexCacheEntry *exceptThisOne, bool forcePressu
 		VERBOSE_LOG(Log::G3D, "Decimated second texture cache, saved %d estimated bytes - now %d bytes", had - secondCacheSizeEstimate_, secondCacheSizeEstimate_);
 	}
 
-	DecimateVideos();
-	replacer_.Decimate(forcePressure ? ReplacerDecimateMode::FORCE_PRESSURE : ReplacerDecimateMode::NEW_FRAME);
-}
-
-void TextureCacheCommon::DecimateVideos() {
+	// Decimate known videos.
 	for (auto iter = videos_.begin(); iter != videos_.end(); ) {
 		if (iter->flips + VIDEO_DECIMATE_AGE < gpuStats.totals.numFlips) {
 			iter = videos_.erase(iter);
@@ -875,6 +878,8 @@ void TextureCacheCommon::DecimateVideos() {
 			++iter;
 		}
 	}
+
+	replacer_.Decimate(forcePressure ? ReplacerDecimateMode::FORCE_PRESSURE : ReplacerDecimateMode::NEW_FRAME);
 }
 
 bool TextureCacheCommon::IsVideo(u32 texaddr) const {
@@ -891,7 +896,7 @@ bool TextureCacheCommon::IsVideo(u32 texaddr) const {
 }
 
 void TextureCacheCommon::HandleTextureChange(TexCacheEntry *const entry, const char *reason, bool initialMatch, bool doDelete) {
-	cacheSizeEstimate_ -= EstimateTexMemoryUsage(entry);
+	cacheSizeEstimate_ -= entry->EstimateTexMemoryUsage();
 	entry->numInvalidated++;
 	gpuStats.perFrame.numTextureInvalidations++;
 	DEBUG_LOG(Log::G3D, "Texture different or overwritten, reloading at %08x: %s", entry->addr, reason);
@@ -980,9 +985,8 @@ void TextureCacheCommon::NotifyFramebuffer(VirtualFramebuffer *framebuffer, Fram
 	}
 }
 
-bool TextureCacheCommon::MatchFramebuffer(
-	const TextureDefinition &entry,
-	VirtualFramebuffer *framebuffer, u32 texaddrOffset, RasterChannel channel, FramebufferMatchInfo *matchInfo) const {
+static bool MatchFramebuffer(const TextureDefinition &entry,
+	VirtualFramebuffer *framebuffer, u32 texaddrOffset, RasterChannel channel, bool useBufferedRendering, FramebufferMatchInfo *matchInfo) {
 	static const u32 MAX_SUBAREA_Y_OFFSET_SAFE = 32;
 
 	uint32_t fb_address = channel == RASTER_DEPTH ? framebuffer->z_address : framebuffer->fb_address;
@@ -1052,12 +1056,9 @@ bool TextureCacheCommon::MatchFramebuffer(
 			// Format incompatible, ignoring without comment. (maybe some really gnarly hacks will end up here...)
 			return false;
 		}
+	} else if (!useBufferedRendering) {
+		return false;
 	} else {
-		// Apply to buffered mode only.
-		if (!framebufferManager_->UseBufferedRendering()) {
-			return false;
-		}
-
 		// Check works for D16 too.
 		// These are combinations that we have special-cased handling for. There are more
 		// ones possible, but rare - we'll add them as we find them used.
@@ -1275,7 +1276,7 @@ bool TextureCacheCommon::SetOffsetTexture(u32 yOffset) {
 	def.dim = gstate.getTextureDimension(0);
 
 	AttachCandidate bestCandidate;
-	if (GetBestFramebufferCandidate(def, texaddrOffset, &bestCandidate, "offsetTexture")) {
+	if (GetBestFramebufferCandidate(framebufferManager_, def, texaddrOffset, &bestCandidate, "offsetTexture")) {
 		SetTextureFramebuffer(bestCandidate);
 		return true;
 	} else {
@@ -1547,14 +1548,13 @@ bool TextureCacheCommon::GetCurrentClutBuffer(GPUDebugBuffer &buffer) {
 }
 
 // Host memory usage, not PSP memory usage.
-u32 TextureCacheCommon::EstimateTexMemoryUsage(const TexCacheEntry *entry) {
-	const u16 dim = entry->dim;
+u32 TexCacheEntry::EstimateTexMemoryUsage() const {
 	// TODO: This does not take into account the HD remaster's larger textures.
 	const u8 dimW = ((dim >> 0) & 0xf);
 	const u8 dimH = ((dim >> 8) & 0xf);
 
 	u32 pixelSize = 2;
-	switch (entry->format) {
+	switch (format) {
 	case GE_TFMT_CLUT4:
 	case GE_TFMT_CLUT8:
 	case GE_TFMT_CLUT16:
@@ -2579,7 +2579,7 @@ void TextureCacheCommon::Clear(bool delete_them) {
 
 void TextureCacheCommon::DeleteTexture(TexCache::iterator it) {
 	ReleaseTexture(it->second.get(), true);
-	cacheSizeEstimate_ -= EstimateTexMemoryUsage(it->second.get());
+	cacheSizeEstimate_ -= it->second->EstimateTexMemoryUsage();
 	cache_.erase(it);
 }
 
@@ -2644,7 +2644,7 @@ bool TextureCacheCommon::CheckFullHash(TexCacheEntry *entry, bool &doDelete) {
 				// It wasn't found, so we're about to throw away the entry and rebuild a texture.
 				// Let's save this in the secondary cache in case it gets used again.
 				secondKey = entry->fullhash | ((u64)entry->cluthash << 32);
-				secondCacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
+				secondCacheSizeEstimate_ += entry->EstimateTexMemoryUsage();
 
 				// If the entry already exists in the secondary texture cache, drop it nicely.
 				auto oldIter = secondCache_.find(secondKey);
@@ -2773,7 +2773,7 @@ bool TextureCacheCommon::PrepareBuildTexture(BuildTexturePlan &plan, TexCacheEnt
 	gpuStats.perFrame.numTexturesDecoded++;
 
 	// For the estimate, we assume cluts always point to 8888 for simplicity.
-	cacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
+	cacheSizeEstimate_ += entry->EstimateTexMemoryUsage();
 
 	plan.badMipSizes = false;
 	// maxLevel here is the max level to upload. Not the count.
@@ -2912,9 +2912,8 @@ bool TextureCacheCommon::PrepareBuildTexture(BuildTexturePlan &plan, TexCacheEnt
 	if (plan.scaleFactor > 1) {
 		plan.levelsToLoad = 1;
 
-		bool enableVideoUpscaling = false;
-
-		if (!enableVideoUpscaling && plan.isVideo) {
+		if (plan.isVideo) {
+			// No upscaling for video textures.
 			plan.scaleFactor = 1;
 			plan.levelsToCreate = 1;
 		}

GPU/Common/TextureCacheCommon.h

@@ -235,6 +235,7 @@ struct TexCacheEntry {
 	bool Matches(u16 dim2, u8 format2, u8 maxLevel2) const;
 	u64 CacheKey() const;
 	static u64 CacheKey(u32 addr, u8 format, u16 dim, u32 cluthash);
+	u32 EstimateTexMemoryUsage() const;
 };
 
 std::string TexStatusToString(TexCacheEntry::TexStatus status);
@@ -443,22 +444,17 @@ protected:
 		return (const T *)clutBufRaw_;
 	}
 
-	static u32 EstimateTexMemoryUsage(const TexCacheEntry *entry);
-
+	// These need to be member functions just for IsVideo and Replacer.
 	SamplerCacheKey GetSamplingParams(int maxLevel, const TexCacheEntry *entry, bool flatZ);
 	SamplerCacheKey GetFramebufferSamplingParams(u16 bufferWidth, u16 bufferHeight);
+
 	void UpdateMaxSeenV(TexCacheEntry *entry, bool throughMode);
 
-	bool MatchFramebuffer(const TextureDefinition &entry, VirtualFramebuffer *framebuffer, u32 texaddrOffset, RasterChannel channel, FramebufferMatchInfo *matchInfo) const;
-
-	bool GetBestFramebufferCandidate(const TextureDefinition &entry, u32 texAddrOffset, AttachCandidate *bestCandidate, const char *context) const;
-
 	void SetTextureFramebuffer(const AttachCandidate &candidate);
 	bool GetCurrentFramebufferTextureDebug(GPUDebugBuffer &buffer, bool *isFramebuffer);
 
 	virtual void BoundFramebufferTexture() {}
 
-	void DecimateVideos();
 	bool IsVideo(u32 texaddr) const;
 
 	static CheckAlphaResult CheckCLUTAlpha(const uint8_t *pixelData, GEPaletteFormat clutFmt, int w);

GPU/Vulkan/TextureCacheVulkan.cpp

@@ -570,20 +570,6 @@ void TextureCacheVulkan::ReleaseTexture(TexCacheEntry *entry, bool delete_them)
 	entry->vkTex = nullptr;
 }
 
-VkFormat getClutDestFormatVulkan(GEPaletteFormat format) {
-	switch (format) {
-	case GE_CMODE_16BIT_ABGR4444:
-		return VULKAN_4444_FORMAT;
-	case GE_CMODE_16BIT_ABGR5551:
-		return VULKAN_1555_FORMAT;
-	case GE_CMODE_16BIT_BGR5650:
-		return VULKAN_565_FORMAT;
-	case GE_CMODE_32BIT_ABGR8888:
-		return VULKAN_8888_FORMAT;
-	}
-	return VK_FORMAT_UNDEFINED;
-}
-
 static const VkFilter MagFiltVK[2] = {
 	VK_FILTER_NEAREST,
 	VK_FILTER_LINEAR
@@ -833,18 +819,18 @@ void TextureCacheVulkan::BuildTexture(TexCacheEntry *const entry) {
 	copyBatch.reserve(levels);
 
 	for (int i = 0; i < levels; i++) {
-		int mipUnscaledWidth = gstate.getTextureWidth(i);
-		int mipUnscaledHeight = gstate.getTextureHeight(i);
+		const int mipUnscaledWidth = gstate.getTextureWidth(i);
+		const int mipUnscaledHeight = gstate.getTextureHeight(i);
 
 		int mipWidth;
 		int mipHeight;
 		plan.GetMipSize(i, &mipWidth, &mipHeight);
 
-		int bpp = VkFormatBytesPerPixel(actualFmt);
+		const int bpp = VkFormatBytesPerPixel(actualFmt);
 		// RoundToNextPowerOf2 is probably not necessary as the optimal alignment is gonna be a power of 2.
-		int optimalStrideAlignment = RoundToNextPowerOf2(std::max(4, (int)vulkan->GetPhysicalDeviceProperties().properties.limits.optimalBufferCopyRowPitchAlignment));
-		int byteStride = RoundUpToMultipleOf(mipWidth * bpp, optimalStrideAlignment);  // output stride
-		int pixelStride = byteStride / bpp;
+		const int optimalStrideAlignment = RoundToNextPowerOf2(std::max(4, (int)vulkan->GetPhysicalDeviceProperties().properties.limits.optimalBufferCopyRowPitchAlignment));
+		const int byteStride = RoundUpToMultipleOf(mipWidth * bpp, optimalStrideAlignment);  // output stride
+		const int pixelStride = byteStride / bpp;
 		int uploadSize = byteStride * mipHeight;
 
 		uint32_t bufferOffset;
@@ -968,7 +954,18 @@ VkFormat TextureCacheVulkan::GetDestFormat(GETextureFormat format, GEPaletteForm
 	case GE_TFMT_CLUT8:
 	case GE_TFMT_CLUT16:
 	case GE_TFMT_CLUT32:
-		return getClutDestFormatVulkan(clutFormat);
+		switch (clutFormat) {
+		case GE_CMODE_16BIT_ABGR4444:
+			return VULKAN_4444_FORMAT;
+		case GE_CMODE_16BIT_ABGR5551:
+			return VULKAN_1555_FORMAT;
+		case GE_CMODE_16BIT_BGR5650:
+			return VULKAN_565_FORMAT;
+		case GE_CMODE_32BIT_ABGR8888:
+			return VULKAN_8888_FORMAT;
+		default:
+			return VK_FORMAT_UNDEFINED;
+		}
 	case GE_TFMT_4444:
 		return VULKAN_4444_FORMAT;
 	case GE_TFMT_5551:

GPU/Vulkan/TextureCacheVulkan.h

@@ -154,5 +154,3 @@ private:
 	VulkanBuffer textureScaleCBuffer_;
 	bool cbufferInited_ = true;
 };
-
-VkFormat getClutDestFormatVulkan(GEPaletteFormat format);