Add optional texture hashing

include/config.hpp

@@ -62,9 +62,13 @@ struct EmulatorConfig {
 	static constexpr RendererType rendererDefault = RendererType::OpenGL;
 #endif
 
+	static constexpr bool hashTexturesDefault = true;
+
 	bool shaderJitEnabled = shaderJitDefault;
 	bool useUbershaders = ubershaderDefault;
 	bool accelerateShaders = accelerateShadersDefault;
+	bool hashTextures = hashTexturesDefault;
+
 	bool accurateShaderMul = false;
 	bool discordRpcEnabled = false;
 

include/renderer.hpp

@@ -51,6 +51,9 @@ class Renderer {
 	u32 outputWindowWidth = 400;
 	u32 outputWindowHeight = 240 * 2;
 
+	// Should hw renderers hash textures? Stored separately from emulatorConfig because we'll be accessing it constantly, might be merged eventually
+	bool hashTextures = false;
+
 	EmulatorConfig* emulatorConfig = nullptr;
 
 	void doSoftwareTextureCopy(u32 inputAddr, u32 outputAddr, u32 copySize, u32 inputWidth, u32 inputGap, u32 outputWidth, u32 outputGap);
@@ -123,4 +126,5 @@ class Renderer {
 	}
 
 	void setConfig(EmulatorConfig* config) { emulatorConfig = config; }
+	void setHashTextures(bool setting) { hashTextures = setting; }
 };

include/renderer_gl/textures.hpp

@@ -1,6 +1,8 @@
 #pragma once
 #include <array>
 #include <string>
+
+#include "PICA/pica_hash.hpp"
 #include "PICA/regs.hpp"
 #include "boost/icl/interval.hpp"
 #include "helpers.hpp"
@@ -11,55 +13,55 @@ template <typename T>
 using Interval = boost::icl::right_open_interval<T>;
 
 struct Texture {
-    u32 location;
-    u32 config; // Magnification/minification filter, wrapping configs, etc
-    PICA::TextureFmt format;
-    OpenGL::uvec2 size;
-    bool valid;
+	using Hash = PICAHash::HashType;
 
-    // Range of VRAM taken up by buffer
-    Interval<u32> range;
-    // OpenGL resources allocated to buffer
-    OpenGL::Texture texture;
+	u32 location;
+	u32 config;  // Magnification/minification filter, wrapping configs, etc
+	Hash hash = Hash(0);
 
-    Texture() : valid(false) {}
+	PICA::TextureFmt format;
+	OpenGL::uvec2 size;
+	bool valid;
 
-    Texture(u32 loc, PICA::TextureFmt format, u32 x, u32 y, u32 config, bool valid = true)
-        : location(loc), format(format), size({x, y}), config(config), valid(valid) {
+	// Range of VRAM taken up by buffer
+	Interval<u32> range;
+	// OpenGL resources allocated to buffer
+	OpenGL::Texture texture;
 
-        u64 endLoc = (u64)loc + sizeInBytes();
-        // Check if start and end are valid here
-        range = Interval<u32>(loc, (u32)endLoc);
-    }
+	Texture() : valid(false) {}
 
-    // For 2 textures to "match" we only care about their locations, formats, and dimensions to match
-    // For other things, such as filtering mode, etc, we can just switch the attributes of the cached texture
-    bool matches(Texture& other) {
-        return location == other.location && format == other.format &&
-            size.x() == other.size.x() && size.y() == other.size.y();
-    }
+	Texture(u32 loc, PICA::TextureFmt format, u32 x, u32 y, u32 config, bool valid = true)
+		: location(loc), format(format), size({x, y}), config(config), valid(valid) {
+		u64 endLoc = (u64)loc + sizeInBytes();
+		// Check if start and end are valid here
+		range = Interval<u32>(loc, (u32)endLoc);
+	}
 
-    void allocate();
-    void setNewConfig(u32 newConfig);
-    void decodeTexture(std::span<const u8> data);
-    void free();
-    u64 sizeInBytes();
+	// For 2 textures to "match" we only care about their locations, formats, and dimensions to match
+	// For other things, such as filtering mode, etc, we can just switch the attributes of the cached texture
+	bool matches(Texture& other) {
+		return location == other.location && hash == other.hash && format == other.format && size.x() == other.size.x() && size.y() == other.size.y();
+	}
 
-    u32 decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data);
+	void allocate();
+	void setNewConfig(u32 newConfig);
+	void decodeTexture(std::span<const u8> data);
+	void free();
+	u64 sizeInBytes();
 
-    // Get the morton interleave offset of a texel based on its U and V values
-    static u32 mortonInterleave(u32 u, u32 v);
-    // Get the byte offset of texel (u, v) in the texture
-    static u32 getSwizzledOffset(u32 u, u32 v, u32 width, u32 bytesPerPixel);
-    static u32 getSwizzledOffset_4bpp(u32 u, u32 v, u32 width);
+	u32 decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data);
 
-    // Returns the format of this texture as a string
-    std::string_view formatToString() {
-        return PICA::textureFormatToString(format);
-    }
+	// Get the morton interleave offset of a texel based on its U and V values
+	static u32 mortonInterleave(u32 u, u32 v);
+	// Get the byte offset of texel (u, v) in the texture
+	static u32 getSwizzledOffset(u32 u, u32 v, u32 width, u32 bytesPerPixel);
+	static u32 getSwizzledOffset_4bpp(u32 u, u32 v, u32 width);
 
-    // Returns the texel at coordinates (u, v) of an ETC1(A4) texture
-    // TODO: Make hasAlpha a template parameter
-    u32 getTexelETC(bool hasAlpha, u32 u, u32 v, u32 width, std::span<const u8> data);
-    u32 decodeETC(u32 alpha, u32 u, u32 v, u64 colourData);
+	// Returns the format of this texture as a string
+	std::string_view formatToString() { return PICA::textureFormatToString(format); }
+
+	// Returns the texel at coordinates (u, v) of an ETC1(A4) texture
+	// TODO: Make hasAlpha a template parameter
+	u32 getTexelETC(bool hasAlpha, u32 u, u32 v, u32 width, std::span<const u8> data);
+	u32 decodeETC(u32 alpha, u32 u, u32 v, u64 colourData);
 };

include/renderer_mtl/mtl_texture.hpp

@@ -4,6 +4,7 @@
 #include <array>
 #include <string>
 
+#include "PICA/pica_hash.hpp"
 #include "PICA/regs.hpp"
 #include "boost/icl/interval.hpp"
 #include "helpers.hpp"
@@ -17,10 +18,14 @@ using Interval = boost::icl::right_open_interval<T>;
 
 namespace Metal {
 	struct Texture {
+		using Hash = PICAHash::HashType;
+
 		MTL::Device* device;
 
 		u32 location;
 		u32 config;  // Magnification/minification filter, wrapping configs, etc
+		Hash hash = Hash(0);
+
 		PICA::TextureFmt format;
 		OpenGL::uvec2 size;
 		bool valid;
@@ -45,7 +50,8 @@ namespace Metal {
 		// For 2 textures to "match" we only care about their locations, formats, and dimensions to match
 		// For other things, such as filtering mode, etc, we can just switch the attributes of the cached texture
 		bool matches(Texture& other) {
-			return location == other.location && format == other.format && size.x() == other.size.x() && size.y() == other.size.y();
+			return location == other.location && hash == other.hash && format == other.format && size.x() == other.size.x() &&
+				   size.y() == other.size.y();
 		}
 
 		void allocate();

src/config.cpp

@@ -90,6 +90,7 @@ void EmulatorConfig::load() {
 
 			forceShadergenForLights = toml::find_or<toml::boolean>(gpu, "ForceShadergenForLighting", true);
 			lightShadergenThreshold = toml::find_or<toml::integer>(gpu, "ShadergenLightThreshold", 1);
+			hashTextures = toml::find_or<toml::boolean>(gpu, "HashTextures", hashTexturesDefault);
 			enableRenderdoc = toml::find_or<toml::boolean>(gpu, "EnableRenderdoc", false);
 		}
 	}
@@ -101,7 +102,7 @@ void EmulatorConfig::load() {
 
 			auto dspCoreName = toml::find_or<std::string>(audio, "DSPEmulation", "HLE");
 			dspType = Audio::DSPCore::typeFromString(dspCoreName);
-			
+
 			audioEnabled = toml::find_or<toml::boolean>(audio, "EnableAudio", audioEnabledDefault);
 			aacEnabled = toml::find_or<toml::boolean>(audio, "EnableAACAudio", true);
 			printDSPFirmware = toml::find_or<toml::boolean>(audio, "PrintDSPFirmware", false);
@@ -180,7 +181,7 @@ void EmulatorConfig::save() {
 	data["Window"]["WindowPosY"] = windowSettings.y;
 	data["Window"]["WindowWidth"] = windowSettings.width;
 	data["Window"]["WindowHeight"] = windowSettings.height;
-	
+
 	data["GPU"]["EnableShaderJIT"] = shaderJitEnabled;
 	data["GPU"]["Renderer"] = std::string(Renderer::typeToString(rendererType));
 	data["GPU"]["EnableVSync"] = vsyncEnabled;
@@ -190,6 +191,7 @@ void EmulatorConfig::save() {
 	data["GPU"]["ShadergenLightThreshold"] = lightShadergenThreshold;
 	data["GPU"]["AccelerateShaders"] = accelerateShaders;
 	data["GPU"]["EnableRenderdoc"] = enableRenderdoc;
+	data["GPU"]["HashTextures"] = hashTextures;
 
 	data["Audio"]["DSPEmulation"] = std::string(Audio::DSPCore::typeToString(dspType));
 	data["Audio"]["EnableAudio"] = audioEnabled;

src/core/renderer_gl/renderer_gl.cpp

@@ -8,6 +8,7 @@
 #include "PICA/float_types.hpp"
 #include "PICA/gpu.hpp"
 #include "PICA/pica_frag_uniforms.hpp"
+#include "PICA/pica_hash.hpp"
 #include "PICA/pica_simd.hpp"
 #include "PICA/regs.hpp"
 #include "PICA/shader_decompiler.hpp"
@@ -359,17 +360,29 @@ void RendererGL::bindTexturesToSlots() {
 		const u32 width = getBits<16, 11>(dim);
 		const u32 addr = (regs[ioBase + 4] & 0x0FFFFFFF) << 3;
 		u32 format = regs[ioBase + (i == 0 ? 13 : 5)] & 0xF;
-
 		glActiveTexture(GL_TEXTURE0 + i);
 
 		if (addr != 0) [[likely]] {
 			Texture targetTex(addr, static_cast<PICA::TextureFmt>(format), width, height, config);
+
+			if (hashTextures) {
+				const u8* startPointer = gpu.getPointerPhys<u8>(targetTex.location);
+				const usize sizeInBytes = targetTex.sizeInBytes();
+
+				if (startPointer == nullptr || (sizeInBytes > 0 && gpu.getPointerPhys<u8>(targetTex.location + sizeInBytes - 1) == nullptr))
+					[[unlikely]] {
+					Helpers::warn("Out-of-bounds texture fetch");
+				} else {
+					targetTex.hash = PICAHash::computeHash((const char*)startPointer, sizeInBytes);
+				}
+			}
+
 			OpenGL::Texture tex = getTexture(targetTex);
 			tex.bind();
 		} else {
-			// Mapping a texture from NULL. PICA seems to read the last sampled colour, but for now we will display a black texture instead since it is far easier.
-			// Games that do this don't really care what it does, they just expect the PICA to not crash, since it doesn't have a PU/MMU and can do all sorts of
-			// Weird invalid memory accesses without crashing
+			// Mapping a texture from NULL. PICA seems to read the last sampled colour, but for now we will display a black texture instead since it
+			// is far easier. Games that do this don't really care what it does, they just expect the PICA to not crash, since it doesn't have a
+			// PU/MMU and can do all sorts of Weird invalid memory accesses without crashing
 			blankTexture.bind();
 		}
 	}

src/core/renderer_mtl/renderer_mtl.cpp

@@ -9,6 +9,7 @@
 #undef NO
 
 #include "PICA/gpu.hpp"
+#include "PICA/pica_hash.hpp"
 #include "SDL_metal.h"
 
 using namespace PICA;
@@ -729,6 +730,19 @@ void RendererMTL::bindTexturesToSlots() {
 
 		if (addr != 0) [[likely]] {
 			Metal::Texture targetTex(device, addr, static_cast<PICA::TextureFmt>(format), width, height, config);
+
+			if (hashTextures) {
+				const u8* startPointer = gpu.getPointerPhys<u8>(targetTex.location);
+				const usize sizeInBytes = targetTex.sizeInBytes();
+
+				if (startPointer == nullptr || (sizeInBytes > 0 && gpu.getPointerPhys<u8>(targetTex.location + sizeInBytes - 1) == nullptr))
+					[[unlikely]] {
+					Helpers::warn("Out-of-bounds texture fetch");
+				} else {
+					targetTex.hash = PICAHash::computeHash((const char*)startPointer, sizeInBytes);
+				}
+			}
+
 			auto tex = getTexture(targetTex);
 			commandEncoder.setFragmentTexture(tex.texture, i);
 			commandEncoder.setFragmentSamplerState(tex.sampler ? tex.sampler : nearestSampler, i);

src/emulator.cpp

@@ -32,18 +32,14 @@ Emulator::Emulator()
 	dspService.setDSPCore(dsp.get());
 
 	audioDevice.init(dsp->getSamples());
-	setAudioEnabled(config.audioEnabled);
-
-	if (Renderdoc::isSupported() && config.enableRenderdoc) {
-		loadRenderdoc();
-	}
-
 #ifdef PANDA3DS_ENABLE_DISCORD_RPC
 	if (config.discordRpcEnabled) {
 		discordRpc.init();
 		updateDiscord();
 	}
 #endif
+
+	reloadSettings();
 	reset(ReloadOption::NoReload);
 }
 
@@ -461,6 +457,8 @@ void Emulator::reloadSettings() {
 		loadRenderdoc();
 	}
 
+    gpu.getRenderer()->setHashTextures(config.hashTextures);
+
 #ifdef PANDA3DS_ENABLE_DISCORD_RPC
 	// Reload RPC setting if we're compiling with RPC support
 

src/libretro_core.cpp

@@ -176,6 +176,9 @@ static void configInit() {
 		{"panda3ds_use_ubershader", EmulatorConfig::ubershaderDefault ? "Use ubershaders (No stutter, maybe slower); enabled|disabled"
 																	  : "Use ubershaders (No stutter, maybe slower); disabled|enabled"},
 		{"panda3ds_use_vsync", "Enable VSync; enabled|disabled"},
+		{"panda3ds_hash_textures", EmulatorConfig::hashTexturesDefault ? "Hash textures (Better graphics, maybe slower); enabled|disabled"
+																	   : "Hash textures (Better graphics, maybe slower); disabled|enabled"},
+
 		{"panda3ds_system_language", "System language; En|Fr|Es|De|It|Pt|Nl|Ru|Ja|Zh|Ko|Tw"},
 		{"panda3ds_dsp_emulation", "DSP emulation; HLE|LLE|Null"},
 		{"panda3ds_use_audio", EmulatorConfig::audioEnabledDefault ? "Enable audio; enabled|disabled" : "Enable audio; disabled|enabled"},
@@ -216,6 +219,7 @@ static void configUpdate() {
 	config.accurateShaderMul = fetchVariableBool("panda3ds_accurate_shader_mul", false);
 	config.useUbershaders = fetchVariableBool("panda3ds_use_ubershader", EmulatorConfig::ubershaderDefault);
 	config.accelerateShaders = fetchVariableBool("panda3ds_accelerate_shaders", EmulatorConfig::accelerateShadersDefault);
+	config.hashTextures = fetchVariableBool("panda3ds_hash_textures", EmulatorConfig::hashTexturesDefault);
 
 	config.forceShadergenForLights = fetchVariableBool("panda3ds_ubershader_lighting_override", true);
 	config.lightShadergenThreshold = fetchVariableRange("panda3ds_ubershader_lighting_override_threshold", 1, 8);

src/panda_qt/config_window.cpp

@@ -231,6 +231,11 @@ ConfigWindow::ConfigWindow(ConfigCallback configCallback, MainWindowCallback win
 	connectCheckbox(accelerateShaders, config.accelerateShaders);
 	gpuLayout->addRow(accelerateShaders);
 
+	QCheckBox* hashTextures = new QCheckBox(tr("Hash textures"));
+	hashTextures->setToolTip(tr("Enable this to reduce texture mismatches at the cost of slightly lower performance"));
+	connectCheckbox(hashTextures, config.hashTextures);
+	gpuLayout->addRow(hashTextures);
+
 	QCheckBox* forceShadergenForLights = new QCheckBox(tr("Force shadergen when rendering lights"));
 	connectCheckbox(forceShadergenForLights, config.forceShadergenForLights);
 	gpuLayout->addRow(forceShadergenForLights);