Add Metal backend (cemu-project#1287)

Author: SamoZ256

.gitmodules

@@ -18,6 +18,10 @@
 	path = dependencies/imgui
 	url = https://github.com/ocornut/imgui
 	shallow = true
+[submodule "dependencies/metal-cpp"]
+	path = dependencies/metal-cpp
+	url = https://github.com/bkaradzic/metal-cpp.git
+	shallow = true
 [submodule "dependencies/xbyak_aarch64"]
 	path = dependencies/xbyak_aarch64
 	url = https://github.com/fujitsu/xbyak_aarch64

CMakeLists.txt

@@ -102,10 +102,20 @@ if (UNIX AND NOT APPLE)
 	option(ENABLE_BLUEZ "Build with Bluez support" ON)
 endif()
 
+if (APPLE)
+    set(ENABLE_METAL_DEFAULT ON)
+else()
+    set(ENABLE_METAL_DEFAULT OFF)
+endif()
+
 option(ENABLE_OPENGL "Enables the OpenGL backend" ON)
 option(ENABLE_VULKAN "Enables the Vulkan backend" ON)
+option(ENABLE_METAL "Enables the Metal backend" ${ENABLE_METAL_DEFAULT})
 option(ENABLE_DISCORD_RPC "Enables the Discord Rich Presence feature" ON)
 
+if (ENABLE_METAL AND NOT APPLE)
+    message(FATAL_ERROR "Metal backend is only supported on Apple platforms")
+endif()
 
 # input backends
 if (WIN32)
@@ -180,6 +190,12 @@ if (ENABLE_OPENGL)
 	find_package(OpenGL REQUIRED)
 endif()
 
+if (ENABLE_METAL)
+    include_directories(${CMAKE_SOURCE_DIR}/dependencies/metal-cpp)
+
+    add_definitions(-DENABLE_METAL=1)
+endif()
+
 if (ENABLE_DISCORD_RPC)
 	add_compile_definitions(ENABLE_DISCORD_RPC)
 endif()
@@ -204,7 +220,7 @@ endif()
 
 if (ENABLE_CUBEB)
 	if (NOT ENABLE_VCPKG)
-	find_package(cubeb)
+	    find_package(cubeb)
 	endif()
 	if (NOT cubeb_FOUND)
 		option(BUILD_TESTS "" OFF)

dependencies/metal-cpp

@@ -537,7 +537,69 @@ add_library(CemuCafe
 )
 
 if(APPLE)
-	target_sources(CemuCafe PRIVATE "HW/Latte/Renderer/Vulkan/CocoaSurface.mm")
+	target_sources(CemuCafe PRIVATE
+	    HW/Latte/Renderer/Vulkan/CocoaSurface.mm
+	    HW/Latte/Renderer/MetalView.mm
+	    HW/Latte/Renderer/MetalView.h
+	)
+endif()
+
+if(ENABLE_METAL)
+    target_sources(CemuCafe PRIVATE
+        HW/Latte/Renderer/Metal/CachedFBOMtl.cpp
+        HW/Latte/Renderer/Metal/CachedFBOMtl.h
+        HW/Latte/Renderer/Metal/LatteTextureMtl.cpp
+        HW/Latte/Renderer/Metal/LatteTextureMtl.h
+        HW/Latte/Renderer/Metal/LatteTextureReadbackMtl.cpp
+        HW/Latte/Renderer/Metal/LatteTextureReadbackMtl.h
+        HW/Latte/Renderer/Metal/LatteTextureViewMtl.cpp
+        HW/Latte/Renderer/Metal/LatteTextureViewMtl.h
+        HW/Latte/Renderer/Metal/LatteToMtl.cpp
+        HW/Latte/Renderer/Metal/LatteToMtl.h
+        HW/Latte/Renderer/Metal/MetalAttachmentsInfo.cpp
+        HW/Latte/Renderer/Metal/MetalAttachmentsInfo.h
+        HW/Latte/Renderer/Metal/MetalBufferAllocator.cpp
+        HW/Latte/Renderer/Metal/MetalBufferAllocator.h
+        HW/Latte/Renderer/Metal/MetalCommon.h
+        HW/Latte/Renderer/Metal/MetalCppImpl.cpp
+        HW/Latte/Renderer/Metal/MetalDepthStencilCache.cpp
+        HW/Latte/Renderer/Metal/MetalDepthStencilCache.h
+        HW/Latte/Renderer/Metal/MetalLayer.h
+        HW/Latte/Renderer/Metal/MetalLayer.mm
+        HW/Latte/Renderer/Metal/MetalLayerHandle.cpp
+        HW/Latte/Renderer/Metal/MetalLayerHandle.h
+        HW/Latte/Renderer/Metal/MetalMemoryManager.cpp
+        HW/Latte/Renderer/Metal/MetalMemoryManager.h
+        HW/Latte/Renderer/Metal/MetalOutputShaderCache.cpp
+        HW/Latte/Renderer/Metal/MetalOutputShaderCache.h
+        HW/Latte/Renderer/Metal/MetalPerformanceMonitor.h
+        HW/Latte/Renderer/Metal/MetalPipelineCache.cpp
+        HW/Latte/Renderer/Metal/MetalPipelineCache.h
+        HW/Latte/Renderer/Metal/MetalPipelineCompiler.cpp
+        HW/Latte/Renderer/Metal/MetalPipelineCompiler.h
+        HW/Latte/Renderer/Metal/MetalQuery.cpp
+        HW/Latte/Renderer/Metal/MetalQuery.h
+        HW/Latte/Renderer/Metal/MetalRenderer.cpp
+        HW/Latte/Renderer/Metal/MetalRenderer.h
+        HW/Latte/Renderer/Metal/MetalSamplerCache.cpp
+        HW/Latte/Renderer/Metal/MetalSamplerCache.h
+        HW/Latte/Renderer/Metal/MetalVoidVertexPipeline.cpp
+        HW/Latte/Renderer/Metal/MetalVoidVertexPipeline.h
+        HW/Latte/Renderer/Metal/RendererShaderMtl.cpp
+        HW/Latte/Renderer/Metal/RendererShaderMtl.h
+        HW/Latte/Renderer/Metal/UtilityShaderSource.h
+    )
+
+    target_sources(CemuCafe PRIVATE
+        HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitMSLAttrDecoder.cpp
+        HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitMSL.cpp
+        HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitMSLHeader.hpp
+    )
+
+    target_link_libraries(CemuCafe PRIVATE
+        "-framework Metal"
+        "-framework QuartzCore"
+    )
 endif()
 
 if(CEMU_ARCHITECTURE MATCHES "(aarch64)|(AARCH64)|(arm64)|(ARM64)")

src/Cafe/CMakeLists.txt

@@ -249,7 +249,17 @@ void InfoLog_PrintActiveSettings()
 	if (ActiveSettings::GetGraphicsAPI() == GraphicAPI::kVulkan)
 	{
 		cemuLog_log(LogType::Force, "Async compile: {}", GetConfig().async_compile.GetValue() ? "true" : "false");
-		if(!GetConfig().vk_accurate_barriers.GetValue())
+		if (!GetConfig().vk_accurate_barriers.GetValue())
+			cemuLog_log(LogType::Force, "Accurate barriers are disabled!");
+	}
+	else if (ActiveSettings::GetGraphicsAPI() == GraphicAPI::kMetal)
+	{
+	    cemuLog_log(LogType::Force, "Async compile: {}", GetConfig().async_compile.GetValue() ? "true" : "false");
+	    cemuLog_log(LogType::Force, "Force mesh shaders: {}", GetConfig().force_mesh_shaders.GetValue() ? "true" : "false");
+		cemuLog_log(LogType::Force, "Fast math: {}", g_current_game_profile->GetShaderFastMath() ? "true" : "false");
+		cemuLog_log(LogType::Force, "Buffer cache type: {}", g_current_game_profile->GetBufferCacheMode());
+		cemuLog_log(LogType::Force, "Position invariance: {}", g_current_game_profile->GetPositionInvariance());
+		if (!GetConfig().vk_accurate_barriers.GetValue())
 			cemuLog_log(LogType::Force, "Accurate barriers are disabled!");
 	}
 	cemuLog_log(LogType::Force, "Console language: {}", stdx::to_underlying(config.console_language.GetValue()));
@@ -1027,7 +1037,7 @@ namespace CafeSystem
 	{
 		// starting with Cemu 1.27.0 /vol/storage_mlc01/ is virtualized, meaning that it doesn't point to one singular host os folder anymore
 		// instead it now uses a more complex solution to source titles with various formats (folder, wud, wua) from the game paths and host mlc path
-		
+
 		// todo - mount /vol/storage_mlc01/ with base priority to the host mlc?
 
 		// since mounting titles is an expensive operation we have to avoid mounting all titles at once

src/Cafe/CafeSystem.cpp

@@ -127,7 +127,7 @@ bool gameProfile_loadIntegerOption(IniParser& iniParser, const char* optionName,
 	{
 		cemuLog_log(LogType::Force, "Value '{}' is out of range for option '{}' in game profile", *option_value, optionName);
 		return false;
-	}	
+	}
 }
 
 template<typename T>
@@ -224,8 +224,11 @@ bool GameProfile::Load(uint64_t title_id)
 			gameProfile_loadIntegerOption(&iniParser, "graphics_api", &graphicsApi, -1, 0, 1);
 			if (graphicsApi.value != -1)
 				m_graphics_api = (GraphicAPI)graphicsApi.value;
-			
+
 			gameProfile_loadEnumOption(iniParser, "accurateShaderMul", m_accurateShaderMul);
+			gameProfile_loadBooleanOption2(iniParser, "shaderFastMath", m_shaderFastMath);
+			gameProfile_loadEnumOption(iniParser, "metalBufferCacheMode2", m_metalBufferCacheMode);
+			gameProfile_loadEnumOption(iniParser, "positionInvariance2", m_positionInvariance);
 
 			// legacy support
 			auto option_precompiledShaders = iniParser.FindOption("precompiledShaders");
@@ -277,7 +280,7 @@ bool GameProfile::Load(uint64_t title_id)
 void GameProfile::Save(uint64_t title_id)
 {
 	auto gameProfileDir = ActiveSettings::GetConfigPath("gameProfiles");
-	if (std::error_code ex_ec; !fs::exists(gameProfileDir, ex_ec)) 
+	if (std::error_code ex_ec; !fs::exists(gameProfileDir, ex_ec))
 		fs::create_directories(gameProfileDir, ex_ec);
 	auto gameProfilePath = gameProfileDir / fmt::format("{:016x}.ini", title_id);
 	FileStream* fs = FileStream::createFile2(gameProfilePath);
@@ -292,22 +295,23 @@ void GameProfile::Save(uint64_t title_id)
 
 #define WRITE_OPTIONAL_ENTRY(__NAME) if (m_##__NAME) fs->writeLine(fmt::format("{} = {}", #__NAME, m_##__NAME.value()).c_str());
 #define WRITE_ENTRY(__NAME) fs->writeLine(fmt::format("{} = {}", #__NAME, m_##__NAME).c_str());
+#define WRITE_ENTRY_NUMBERED(__NAME, __NUM) fs->writeLine(fmt::format("{} = {}", #__NAME #__NUM, m_##__NAME).c_str());
 
 	fs->writeLine("[General]");
 	WRITE_OPTIONAL_ENTRY(loadSharedLibraries);
 	WRITE_ENTRY(startWithPadView);
-
 	fs->writeLine("");
 
-
 	fs->writeLine("[CPU]");
 	WRITE_OPTIONAL_ENTRY(cpuMode);
 	WRITE_ENTRY(threadQuantum);
-
 	fs->writeLine("");
 
 	fs->writeLine("[Graphics]");
 	WRITE_ENTRY(accurateShaderMul);
+	WRITE_ENTRY(shaderFastMath);
+	WRITE_ENTRY_NUMBERED(metalBufferCacheMode, 2);
+	WRITE_ENTRY_NUMBERED(positionInvariance, 2);
 	WRITE_OPTIONAL_ENTRY(precompiledShaders);
 	WRITE_OPTIONAL_ENTRY(graphics_api);
 	fs->writeLine("");
@@ -323,6 +327,7 @@ void GameProfile::Save(uint64_t title_id)
 
 #undef WRITE_OPTIONAL_ENTRY
 #undef WRITE_ENTRY
+#undef WRITE_ENTRY_NUMBERED
 
 	delete fs;
 }
@@ -337,6 +342,9 @@ void GameProfile::ResetOptional()
 
 	// graphic settings
 	m_accurateShaderMul = AccurateShaderMulOption::True;
+	m_shaderFastMath = true;
+	m_metalBufferCacheMode = MetalBufferCacheMode::Auto;
+	m_positionInvariance = PositionInvariance::Auto;
 	// cpu settings
 	m_threadQuantum = kThreadQuantumDefault;
 	m_cpuMode.reset(); // CPUModeOption::kSingleCoreRecompiler;
@@ -354,9 +362,12 @@ void GameProfile::Reset()
 	// general settings
 	m_loadSharedLibraries = true;
 	m_startWithPadView = false;
-	
+
 	// graphic settings
 	m_accurateShaderMul = AccurateShaderMulOption::True;
+	m_shaderFastMath = true;
+	m_metalBufferCacheMode = MetalBufferCacheMode::Auto;
+	m_positionInvariance = PositionInvariance::Auto;
 	m_precompiledShaders = PrecompiledShaderOption::Auto;
 	// cpu settings
 	m_threadQuantum = kThreadQuantumDefault;

src/Cafe/GameProfile/GameProfile.cpp

@@ -31,6 +31,9 @@ class GameProfile
 
 	[[nodiscard]] const std::optional<GraphicAPI>& GetGraphicsAPI() const { return m_graphics_api; }
 	[[nodiscard]] const AccurateShaderMulOption& GetAccurateShaderMul() const { return m_accurateShaderMul; }
+	[[nodiscard]] bool GetShaderFastMath() const { return m_shaderFastMath; }
+	[[nodiscard]] MetalBufferCacheMode GetBufferCacheMode() const { return m_metalBufferCacheMode; }
+	[[nodiscard]] PositionInvariance GetPositionInvariance() const { return m_positionInvariance; }
 	[[nodiscard]] const std::optional<PrecompiledShaderOption>& GetPrecompiledShadersState() const { return m_precompiledShaders; }
 
 	[[nodiscard]] uint32 GetThreadQuantum() const { return m_threadQuantum; }
@@ -54,6 +57,9 @@ class GameProfile
 	// graphic settings
 	std::optional<GraphicAPI> m_graphics_api{};
 	AccurateShaderMulOption m_accurateShaderMul = AccurateShaderMulOption::True;
+	bool m_shaderFastMath = true;
+	MetalBufferCacheMode m_metalBufferCacheMode = MetalBufferCacheMode::Auto;
+	PositionInvariance m_positionInvariance = PositionInvariance::Auto;
 	std::optional<PrecompiledShaderOption> m_precompiledShaders{};
 	// cpu settings
 	uint32 m_threadQuantum = kThreadQuantumDefault; // values: 20000 45000 60000 80000 100000