From 06b83115a5acf48ac22a58123ceb01ad1f466592 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Fri, 10 Mar 2023 12:20:55 +0100
Subject: [PATCH] Replacer: Avoid tracking video textures

---
 Common/GPU/Vulkan/VulkanDebug.cpp |  6 ++++++
 GPU/Common/ReplacedTexture.h      |  2 ++
 GPU/Common/TextureCacheCommon.cpp | 11 ++++++++++-
 GPU/Common/TextureCacheCommon.h   |  6 ++++--
 GPU/Common/TextureReplacer.h      |  3 +++
 GPU/GPU.h                         |  4 ++++
 GPU/GPUCommonHW.cpp               |  3 +++
 7 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/Common/GPU/Vulkan/VulkanDebug.cpp b/Common/GPU/Vulkan/VulkanDebug.cpp
index 4daf41bdf7..29a71cf61e 100644
--- a/Common/GPU/Vulkan/VulkanDebug.cpp
+++ b/Common/GPU/Vulkan/VulkanDebug.cpp
@@ -62,6 +62,12 @@ VKAPI_ATTR VkBool32 VKAPI_CALL VulkanDebugUtilsCallback(
 		// See https://github.com/hrydgard/ppsspp/pull/16354
 		return false;
 	}
+	if (messageCode == -375211665) {
+		// VUID-vkAllocateMemory-pAllocateInfo-01713
+		// Can happen when VMA aggressively tries to allocate aperture memory for upload. It gracefully
+		// falls back to regular video memory, so we just ignore this. I'd argue this is a VMA bug, actually.
+		return false;
+	}
 
 	int count;
 	{
diff --git a/GPU/Common/ReplacedTexture.h b/GPU/Common/ReplacedTexture.h
index 4e00d8df28..735be70590 100644
--- a/GPU/Common/ReplacedTexture.h
+++ b/GPU/Common/ReplacedTexture.h
@@ -59,6 +59,8 @@ struct ReplacedTextureLevel {
 
 ReplacedImageType Identify(VFSBackend *vfs, VFSOpenFile *openFile, std::string *outMagic);
 
+// These aren't actually all replaced, they can also represent a placeholder for a not-found
+// replacement.
 struct ReplacedTexture {
 	~ReplacedTexture();
 
diff --git a/GPU/Common/TextureCacheCommon.cpp b/GPU/Common/TextureCacheCommon.cpp
index 8751efbb6e..5d914c061b 100644
--- a/GPU/Common/TextureCacheCommon.cpp
+++ b/GPU/Common/TextureCacheCommon.cpp
@@ -144,6 +144,11 @@ void TextureCacheCommon::StartFrame() {
 	timesInvalidatedAllThisFrame_ = 0;
 	replacementTimeThisFrame_ = 0.0;
 
+	if (g_Config.bShowDebugStats) {
+		gpuStats.numReplacerTrackedTex = replacer_.GetNumTrackedTextures();
+		gpuStats.numCachedReplacedTextures = replacer_.GetNumCachedReplacedTextures();
+	}
+
 	if (texelsScaledThisFrame_) {
 		VERBOSE_LOG(G3D, "Scaled %d texels", texelsScaledThisFrame_);
 	}
@@ -1497,6 +1502,10 @@ ReplacedTexture *TextureCacheCommon::FindReplacement(TexCacheEntry *entry, int &
 		return nullptr;
 	}
 
+	if (entry->status & TexCacheEntry::STATUS_VIDEO) {
+		return nullptr;
+	}
+
 	// Allow some delay to reduce pop-in.
 	constexpr double MAX_BUDGET_PER_TEX = 0.25 / 60.0;
 
@@ -2039,7 +2048,7 @@ void TextureCacheCommon::ApplyTexture() {
 		// Regardless of hash fails or otherwise, if this is a video, mark it frequently changing.
 		// This prevents temporary scaling perf hits on the first second of video.
 		if (IsVideo(entry->addr)) {
-			entry->status |= TexCacheEntry::STATUS_CHANGE_FREQUENT;
+			entry->status |= TexCacheEntry::STATUS_CHANGE_FREQUENT | TexCacheEntry::STATUS_VIDEO;
 		}
 
 		if (nextNeedsRehash_) {
diff --git a/GPU/Common/TextureCacheCommon.h b/GPU/Common/TextureCacheCommon.h
index 7db6507b13..a51660d364 100644
--- a/GPU/Common/TextureCacheCommon.h
+++ b/GPU/Common/TextureCacheCommon.h
@@ -149,10 +149,12 @@ struct TexCacheEntry {
 		STATUS_3D = 0x4000,
 
 		STATUS_CLUT_GPU = 0x8000,
+
+		STATUS_VIDEO = 0x10000,
 	};
 
-	// Status, but int so we can zero initialize.
-	int status;
+	// TexStatus enum flag combination.
+	u32 status;
 
 	u32 addr;
 	u32 minihash;
diff --git a/GPU/Common/TextureReplacer.h b/GPU/Common/TextureReplacer.h
index 1efc9d7a54..27c6b1b83f 100644
--- a/GPU/Common/TextureReplacer.h
+++ b/GPU/Common/TextureReplacer.h
@@ -127,6 +127,9 @@ public:
 	static bool GenerateIni(const std::string &gameID, Path &generatedFilename);
 	static bool IniExists(const std::string &gameID);
 
+	int GetNumTrackedTextures() const { return (int)cache_.size(); }
+	int GetNumCachedReplacedTextures() const { return (int)levelCache_.size(); }
+
 protected:
 	bool LoadIni();
 	bool LoadIniValues(IniFile &ini, bool isOverride = false);
diff --git a/GPU/GPU.h b/GPU/GPU.h
index 12274e51fb..c44749cb63 100644
--- a/GPU/GPU.h
+++ b/GPU/GPU.h
@@ -100,6 +100,8 @@ struct GPUStatistics {
 		numColorCopies = 0;
 		numCopiesForShaderBlend = 0;
 		numCopiesForSelfTex = 0;
+		numReplacerTrackedTex = 0;
+		numCachedReplacedTextures = 0;
 		msProcessingDisplayLists = 0;
 		vertexGPUCycles = 0;
 		otherGPUCycles = 0;
@@ -133,6 +135,8 @@ struct GPUStatistics {
 	int numColorCopies;
 	int numCopiesForShaderBlend;
 	int numCopiesForSelfTex;
+	int numReplacerTrackedTex;
+	int numCachedReplacedTextures;
 	double msProcessingDisplayLists;
 	int vertexGPUCycles;
 	int otherGPUCycles;
diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
index 01d6dd02c2..e5fd62a0ae 100644
--- a/GPU/GPUCommonHW.cpp
+++ b/GPU/GPUCommonHW.cpp
@@ -1636,6 +1636,7 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 		"FBOs active: %d (evaluations: %d)\n"
 		"Textures: %d, dec: %d, invalidated: %d, hashed: %d kB\n"
 		"readbacks %d (%d non-block), uploads %d, depal %d\n"
+		"replacer: tracks %d textures, %d unique loaded\n"
 		"Copies: depth %d, color %d, reint %d, blend %d, selftex %d\n"
 		"GPU cycles executed: %d (%f per vertex)\n",
 		gpuStats.msProcessingDisplayLists * 1000.0f,
@@ -1659,6 +1660,8 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 		gpuStats.numReadbacks,
 		gpuStats.numUploads,
 		gpuStats.numDepal,
+		gpuStats.numReplacerTrackedTex,
+		gpuStats.numCachedReplacedTextures,
 		gpuStats.numDepthCopies,
 		gpuStats.numColorCopies,
 		gpuStats.numReinterpretCopies,