perf(files): move directory-entry names instead of copying them (#758)

Co-authored-by: Techx3 <56268269+Techx3@users.noreply.github.com>
Co-authored-by: Ali Almohaya <17254073+Almo7aya@users.noreply.github.com>

.coderabbit.yaml

@@ -0,0 +1,37 @@
+# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
+
+# Keep inherited review profiles, automatic reviews, tools, and pre-merge checks.
+inheritance: true
+
+reviews:
+  # Trim supplemental output and keep the remaining walkthrough collapsed.
+  high_level_summary: false
+  review_status: false
+  review_details: false
+  collapse_walkthrough: true
+  changed_files_summary: false
+  sequence_diagrams: false
+  estimate_code_review_effort: false
+  related_issues: false
+  related_prs: false
+  in_progress_fortune: false
+  poem: false
+  enable_prompt_for_ai_agents: false
+
+  # Hide standard finishing-touch controls; this also disables their commands.
+  finishing_touches:
+    docstrings:
+      enabled: false
+    unit_tests:
+      enabled: false
+    simplify:
+      enabled: false
+    autofix:
+      enabled: false
+    fix_ci:
+      enabled: false
+    resolve_merge_conflict:
+      enabled: false
+
+chat:
+  art: false

.github/workflows/build.yml

@@ -6,7 +6,12 @@ on:
     branches: [ main, master ]
   pull_request:
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 env:
+  SCCACHE_GHA_ENABLED: "true"
   KYTY_BUILD_ORIGIN: ${{ github.repository != 'KytyPS5/KytyPS5' && 'Fork' || github.event_name == 'push' && 'Official' || 'Source' }}
   KYTY_BUILD_REPOSITORY: ${{ github.repository }}
 
@@ -38,9 +43,6 @@ jobs:
       - name: Setup Visual Studio environment
         uses: ilammy/msvc-dev-cmd@v1
 
-      - name: Setup Ninja
-        uses: seanmiddleditch/gha-setup-ninja@v5
-
       - name: Locate vcpkg
         id: vcpkg
         shell: pwsh
@@ -78,6 +80,9 @@ jobs:
           arch: win64_msvc2022_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: cmd
         run: |
@@ -100,6 +105,8 @@ jobs:
             -DKYTY_RELEASE_TAG=${{ needs.metadata.outputs.release_tag }} ^
             -DCMAKE_C_COMPILER=clang-cl ^
             -DCMAKE_CXX_COMPILER=clang-cl ^
+            -DCMAKE_C_COMPILER_LAUNCHER=sccache ^
+            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache ^
             -DCMAKE_PREFIX_PATH="%Qt6_DIR%"
 
       - name: Build
@@ -145,7 +152,7 @@ jobs:
 
       - name: Install build dependencies
         shell: bash
-        run: brew install glslang ninja
+        run: brew install glslang
 
       - name: Install Qt 6.10.3
         uses: jurplel/install-qt-action@v4
@@ -156,6 +163,9 @@ jobs:
           arch: clang_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: bash
         run: |
@@ -180,6 +190,8 @@ jobs:
             -DCMAKE_OSX_ARCHITECTURES=x86_64 \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
+            -DCMAKE_C_COMPILER_LAUNCHER=sccache \
+            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
             -DCMAKE_PREFIX_PATH="$QT_ROOT_DIR"
 
       - name: Build
@@ -303,7 +315,7 @@ jobs:
             libxrandr-dev \
             libxss-dev \
             lld \
-            ninja-build \
+            pkg-config \
             wayland-protocols
 
       - name: Install Qt 6.10.3
@@ -315,6 +327,9 @@ jobs:
           arch: linux_gcc_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: bash
         run: |
@@ -339,6 +354,8 @@ jobs:
             -DKYTY_RELEASE_TAG="${{ needs.metadata.outputs.release_tag }}" \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
+            -DCMAKE_C_COMPILER_LAUNCHER=sccache \
+            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
             -DCMAKE_PREFIX_PATH="$Qt6_DIR" 2>&1 | tee _Build/configure.log
           exit "${PIPESTATUS[0]}"
 

.gitmodules

@@ -22,7 +22,7 @@
 	url = https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator.git
 [submodule "3rdparty/ffmpeg-core"]
 	path = 3rdparty/ffmpeg-core
-	url = https://github.com/shadps4-emu/ext-ffmpeg-core.git
+	url = https://github.com/KytyPS5/ext-ffmpeg-core.git
 	shallow = true
 [submodule "3rdparty/fmt"]
 	path = 3rdparty/fmt

.vscode/settings.json

@@ -2,15 +2,17 @@
     "cmake.sourceDirectory": "${workspaceFolder}",
     "cmake.buildDirectory": "${workspaceFolder}/_Build/vscode-clang",
     "cmake.generator": "Ninja",
+    "cmake.useVsDeveloperEnvironment": "always",
     "cmake.environment": {
         "CC": "clang-cl",
-        "CXX": "clang-cl",
-        "PATH": "${env:ProgramFiles}/Microsoft Visual Studio/18/Community/VC/Tools/Llvm/x64/bin;${env:ProgramFiles}/Microsoft Visual Studio/18/Community/Common7/IDE/CommonExtensions/Microsoft/CMake/Ninja;${env:PATH}"
+        "CXX": "clang-cl"
     },
     "cmake.configureSettings": {
         "CMAKE_C_COMPILER": "clang-cl",
         "CMAKE_CXX_COMPILER": "clang-cl",
         "CMAKE_PREFIX_PATH": "C:/Qt/6.11.0/msvc2022_64"
-    }
+    },
+    "C_Cpp.default.intelliSenseMode": "windows-clang-x64",
+    "C_Cpp.default.compileCommands": "${workspaceFolder}/_Build/vscode-clang/compile_commands.json",
+    "C_Cpp.default.cppStandard": "c++20"
 }
-

3rdparty/CMakeLists.txt

@@ -86,6 +86,9 @@ set(SPDLOG_BUILD_WARNINGS OFF CACHE BOOL "" FORCE)
 set(SPDLOG_INSTALL OFF CACHE BOOL "" FORCE)
 set(SPDLOG_FMT_EXTERNAL ON CACHE BOOL "" FORCE)
 set(SPDLOG_NO_EXCEPTIONS ON CACHE BOOL "" FORCE)
+if(WIN32)
+	set(SPDLOG_WCHAR_FILENAMES ON CACHE BOOL "" FORCE)
+endif()
 add_subdirectory(spdlog EXCLUDE_FROM_ALL)
 
 if (NOT TARGET FFmpeg::ffmpeg)

CMakeLists.txt

@@ -12,6 +12,22 @@ endif()
 
 project(Kyty)
 
+find_program(SCCACHE_PROGRAM sccache)
+find_program(CCACHE_PROGRAM ccache)
+
+foreach(lang IN ITEMS C CXX)
+    # An explicitly empty launcher disables caching for this language.
+    if(NOT DEFINED CMAKE_${lang}_COMPILER_LAUNCHER)
+        if(SCCACHE_PROGRAM)
+            set(CMAKE_${lang}_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}")
+            message(STATUS "sccache enabled for ${lang}: ${SCCACHE_PROGRAM}")
+        elseif(CCACHE_PROGRAM)
+            set(CMAKE_${lang}_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
+            message(STATUS "ccache enabled for ${lang}: ${CCACHE_PROGRAM}")
+        endif()
+    endif()
+endforeach()
+
 set(KYTY_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/src")
 set(KYTY_TESTS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/tests")
 set(KYTY_THIRD_PARTY_DIR "${CMAKE_CURRENT_SOURCE_DIR}/3rdparty")
@@ -317,24 +333,10 @@ endif()
 
 set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL2-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
 
-# Linux system libraries required by the static FFmpeg archive.
+# Linux system libraries.
 if(LINUX)
 	find_package(Threads REQUIRED)
 	list(APPEND kyty_emulator_link_libraries m ${CMAKE_DL_LIBS} Threads::Threads)
-
-	# Optional FFmpeg dependencies.
-	find_package(ZLIB)
-	if(ZLIB_FOUND)
-		list(APPEND kyty_emulator_link_libraries ZLIB::ZLIB)
-	endif()
-	find_library(KYTY_BZ2_LIBRARY bz2)
-	if(KYTY_BZ2_LIBRARY)
-		list(APPEND kyty_emulator_link_libraries ${KYTY_BZ2_LIBRARY})
-	endif()
-	find_library(KYTY_LZMA_LIBRARY lzma)
-	if(KYTY_LZMA_LIBRARY)
-		list(APPEND kyty_emulator_link_libraries ${KYTY_LZMA_LIBRARY})
-	endif()
 endif()
 
 set(inc_headers
@@ -464,6 +466,12 @@ add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
 target_link_libraries(audio_out2_port_tests common fmt::fmt)
 target_include_directories(audio_out2_port_tests PRIVATE ${inc_headers})
 
+add_executable(save_data_memory_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/SaveDataMemoryTests.cpp"
+)
+target_link_libraries(save_data_memory_tests common fmt::fmt)
+target_include_directories(save_data_memory_tests PRIVATE ${inc_headers})
+
 add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/EventQueueLifetimeTests.cpp"
 	"${KYTY_SOURCE_DIR}/kernel/eventQueue.cpp"
@@ -548,6 +556,7 @@ if(BUILD_TESTING)
 	add_test(NAME kernel_file_system COMMAND $<TARGET_FILE:kernel_file_system_tests>)
 	add_test(NAME sync_on_address COMMAND $<TARGET_FILE:sync_on_address_tests>)
 	add_test(NAME audio_out2_port COMMAND $<TARGET_FILE:audio_out2_port_tests>)
+	add_test(NAME save_data_memory COMMAND $<TARGET_FILE:save_data_memory_tests>)
 	add_test(NAME shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
 	add_test(NAME shader_recompiler_alignbyte
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --alignbyte-only)
@@ -615,6 +624,21 @@ if(BUILD_TESTING)
 	)
 endif()
 
+# Test Darwin signal-context restoration for SSE4a instructions.
+if(APPLE AND (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR
+		(NOT CMAKE_OSX_ARCHITECTURES AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")))
+	add_executable(macos_sse4a_tests EXCLUDE_FROM_ALL
+		"${KYTY_TESTS_DIR}/macosSse4aTests.cpp"
+		"${KYTY_SOURCE_DIR}/loader/x64InstructionEmulator.cpp"
+	)
+	target_link_libraries(macos_sse4a_tests common fmt::fmt Zydis::Zydis)
+	target_include_directories(macos_sse4a_tests PRIVATE ${inc_headers})
+	if(BUILD_TESTING)
+		add_test(NAME macos_sse4a COMMAND $<TARGET_FILE:macos_sse4a_tests>)
+		set_tests_properties(macos_sse4a PROPERTIES TIMEOUT 15)
+		add_dependencies(kyty_tests macos_sse4a_tests)
+	endif()
+endif()
 
 add_executable(kyty_emulator "${KYTY_SOURCE_DIR}/main.cpp" ${kyty_emulator_src})
 configure_macos_guest_address_space(kyty_emulator)
@@ -657,6 +681,12 @@ elseif(WIN32 OR LINUX)
 	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${KYTY_LD_OPTIONS} -Wl,-Map=${KYTY_EMULATOR_MAP_LINK_PATH}")
 endif()
 
+if(LINUX)
+    target_link_options(kyty_emulator PRIVATE
+        $<$<CONFIG:Release>:-Wl,--strip-all>
+    )
+endif()
+
 if(APPLE)
 	# The emulator writes x86-64 trampolines/PLT stubs into memory and executes them, and
 	# runs guest code under Rosetta. That requires the JIT / unsigned-executable-memory

README.md

@@ -137,6 +137,7 @@ the Vulkan/SPIR-V validation rules.
 - Visual Studio 2022 or Build Tools 2022 with the **Desktop development with C++** workload and
   **C++ Clang tools for Windows** component
 - Qt 6 for MSVC 2022 64-bit, including Concurrent, Network, and Widgets
+- [glslang](https://github.com/KhronosGroup/glslang/releases) (`glslangValidator`) on `PATH`
 
 The Microsoft C++ compiler (`cl.exe`) is not supported; use `clang-cl`.
 
@@ -171,7 +172,7 @@ build has no working sound and no gamepad hotplug:
 
 ```bash
 sudo apt-get install --no-install-recommends \
-  clang lld ninja-build cmake git glslang-tools \
+  clang lld ninja-build cmake git glslang-tools pkg-config \
   libgl1-mesa-dev libx11-dev libxcursor-dev libxext-dev libxfixes-dev \
   libxi-dev libxrandr-dev libxss-dev libxkbcommon-dev \
   libasound2-dev libpulse-dev libudev-dev libdbus-1-dev libwayland-dev wayland-protocols
@@ -192,13 +193,37 @@ cmake --install _Build/linux --prefix _Build/linux/install
 ```
 
 The install step copies the Qt libraries and plugins next to the binaries, so
-`_Build/linux/install` runs without a matching system Qt.
+`_Build/linux/install` runs without a matching system Qt. FFmpeg is linked statically
+from the pinned [KytyPS5 FFmpeg core](https://github.com/KytyPS5/ext-ffmpeg-core)
+release, including VP9 and WebM support. System FFmpeg packages are not required.
 
 As on Windows, the MSVC compiler is not used; Clang is required. `cl.exe` is rejected at configure
 time.
 
 The CMake source root is the repository root.
 
+### Building on NixOS
+
+A development shell provides Clang, CMake, Ninja, Qt 6, the Vulkan headers, and the SDL2 backend
+libraries. Enter it and configure exactly as on other Linux distributions; the shell exports
+`CMAKE_PREFIX_PATH` and `QT_PLUGIN_PATH`, so the `-DCMAKE_PREFIX_PATH="$Qt6_DIR"` argument is not
+needed:
+
+```bash
+nix-shell # or: nix develop
+git submodule update --init --recursive
+
+cmake -S . -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \
+  -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++
+
+cmake --build _Build/linux --target launcher --parallel
+cmake --install _Build/linux --prefix _Build/linux/install
+```
+
+The configure step downloads the FFmpeg prebuilts and the `xbyak`/`zydis` sources, so it needs
+network access; a fully sandboxed `nix build` would require vendoring those inputs. A Vulkan 1.3
+driver must be available at runtime (on NixOS, `hardware.graphics.enable = true`).
+
 ### Building on macOS
 
 macOS builds target x86-64 and run under Rosetta 2 on Apple Silicon, so the PS5's x86-64 game

flake.lock

@@ -0,0 +1,27 @@
+{
+  "nodes": {
+    "nixpkgs": {
+      "locked": {
+        "lastModified": 1789426278,
+        "narHash": "sha256-GnotcvKtbTMnoVx6G4E7ZdUX6487Dky74dqq3pU7iRk=",
+        "owner": "NixOS",
+        "repo": "nixpkgs",
+        "rev": "efe6f071ede9d21c37462d2d6682d5e670099684",
+        "type": "github"
+      },
+      "original": {
+        "owner": "NixOS",
+        "ref": "nixos-unstable",
+        "repo": "nixpkgs",
+        "type": "github"
+      }
+    },
+    "root": {
+      "inputs": {
+        "nixpkgs": "nixpkgs"
+      }
+    }
+  },
+  "root": "root",
+  "version": 7
+}

flake.nix

@@ -0,0 +1,20 @@
+{
+  description = "KytyPS5 development shell";
+
+  inputs.nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
+
+  outputs = { self, nixpkgs }:
+    let
+      systems = [ "x86_64-linux" ];
+      forAllSystems = nixpkgs.lib.genAttrs systems;
+    in
+    {
+      devShells = forAllSystems (system:
+        let
+          pkgs = import nixpkgs { inherit system; };
+        in
+        {
+          default = import ./nix/devshell.nix { inherit pkgs; };
+        });
+    };
+}

nix/devshell.nix

@@ -0,0 +1,84 @@
+{ pkgs }:
+
+let
+  # Build-time dependencies. Every -dev package the bundled SDL2 probes for must
+  # be listed here, otherwise SDL2 quietly drops the audio/Wayland/udev backends.
+  buildDeps = with pkgs; [
+    qt6.qtbase # Concurrent, Network, Widgets (launcher)
+    vulkan-headers
+    vulkan-loader
+    vulkan-tools
+    vulkan-validation-layers
+    libglvnd
+    # X11 / input
+    libX11
+    libXcursor
+    libXext
+    libXfixes
+    libXi
+    libXrandr
+    libXScrnSaver
+    libxkbcommon
+    # Wayland
+    wayland
+    wayland-protocols
+    libdecor
+    # audio + device/dbus
+    alsa-lib
+    libpulseaudio
+    systemd # libudev
+    dbus
+  ];
+
+  # Libraries the built binaries need at runtime.
+  runtimeLibs = with pkgs; [
+    stdenv.cc.cc.lib # libstdc++ / libgcc_s
+    libgcc.lib
+    qt6.qtbase # libQt6Concurrent/Network/Widgets
+    vulkan-loader
+    libglvnd
+    alsa-lib
+    libpulseaudio
+    libX11
+    libXcursor
+    libXext
+    libXfixes
+    libXi
+    libXrandr
+    libXScrnSaver
+    libxkbcommon
+    wayland
+    libdecor
+    systemd
+    dbus
+  ];
+in
+pkgs.mkShell {
+  name = "kytyps5";
+
+  nativeBuildInputs = with pkgs; [
+    clang
+    lld
+    cmake
+    ninja
+    pkg-config
+    git
+    glslang # provides glslangValidator for the bundled shaders
+  ];
+
+  buildInputs = buildDeps;
+
+  shellHook = ''
+    export CMAKE_PREFIX_PATH="${pkgs.qt6.qtbase}:''${CMAKE_PREFIX_PATH:-}"
+    export QT_PLUGIN_PATH="${pkgs.qt6.qtbase}/lib/qt-6/plugins:${pkgs.qt6.qtbase}/lib/qt6/plugins:''${QT_PLUGIN_PATH:-}"
+    export LD_LIBRARY_PATH="${pkgs.lib.makeLibraryPath runtimeLibs}:''${LD_LIBRARY_PATH:-}"
+
+    if [ -z "''${KYTY_SHELL_QUIET:-}" ]; then
+      echo "KytyPS5 dev shell"
+      echo "  cmake -S . -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \\"
+      echo "    -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++"
+      echo "  cmake --build _Build/linux --target launcher --parallel"
+      echo "  cmake --install _Build/linux --prefix _Build/linux/install"
+    fi
+  '';
+}

shell.nix

@@ -0,0 +1,4 @@
+let
+  pkgs = import <nixpkgs> { };
+in
+import ./nix/devshell.nix { inherit pkgs; }

src/common/byteBuffer.h

@@ -1,166 +0,0 @@
-#ifndef KYTY_COMMON_BYTEBUFFER_H_
-#define KYTY_COMMON_BYTEBUFFER_H_
-
-#include "common/assert.h"
-#include "common/common.h"
-
-#include <algorithm>
-#include <cstddef> // IWYU pragma: export
-#include <cstring>
-#include <utility>
-#include <vector>
-
-namespace Common {
-
-using Byte = std::byte;
-
-class ByteBuffer {
-public:
-	using DataType = std::vector<Byte>;
-
-	static constexpr uint32_t INVALID_INDEX = static_cast<uint32_t>(-1);
-
-	using iterator       = DataType::iterator;
-	using const_iterator = DataType::const_iterator;
-
-	ByteBuffer()                  = default;
-	ByteBuffer(const ByteBuffer&) = default;
-	ByteBuffer(ByteBuffer&&)      = default;
-
-	ByteBuffer& operator=(const ByteBuffer&) = default;
-	ByteBuffer& operator=(ByteBuffer&&)      = default;
-
-	explicit ByteBuffer(uint32_t size): m_data(size) {}
-
-	ByteBuffer(std::initializer_list<uint8_t> list) {
-		for (const uint8_t& e: list) {
-			m_data.emplace_back(static_cast<Byte>(e));
-		}
-	}
-
-	explicit ByteBuffer(const void* buf, uint32_t size)
-	    : ByteBuffer(size) // @suppress("Ambiguous problem")
-	{
-		EXIT_IF(buf == nullptr && size != 0);
-
-		if (size != 0) {
-			std::memcpy(m_data.data(), buf, size);
-		}
-	}
-
-	[[nodiscard]] uint32_t Size() const { return static_cast<uint32_t>(m_data.size()); }
-	[[nodiscard]] bool     IsEmpty() const { return m_data.empty(); }
-
-	void Clear() { m_data.clear(); }
-
-	void Add(const Byte& val) { m_data.push_back(val); }
-	void Add(Byte&& val) { m_data.push_back(std::move(val)); }
-
-	void Add(const Byte* val, uint32_t num) {
-		EXIT_IF(val == nullptr && num != 0);
-		m_data.insert(m_data.end(), val, val + num);
-	}
-
-	void Add(const ByteBuffer& v) {
-		if (this == &v) {
-			const ByteBuffer copy(v);
-			Add(copy.GetDataConst(), copy.Size());
-			return;
-		}
-		Add(v.GetDataConst(), v.Size());
-	}
-
-	void InsertAt(uint32_t index, const Byte& val) {
-		EXIT_IF(index > Size());
-		m_data.insert(m_data.begin() + index, val);
-	}
-
-	void InsertAt(uint32_t index, Byte&& val) {
-		EXIT_IF(index > Size());
-		m_data.insert(m_data.begin() + index, std::move(val));
-	}
-
-	[[nodiscard]] uint32_t Find(const Byte& t) const {
-		const auto it = std::find(m_data.begin(), m_data.end(), t);
-		return (it == m_data.end() ? INVALID_INDEX
-		                           : static_cast<uint32_t>(std::distance(m_data.begin(), it)));
-	}
-
-	[[nodiscard]] bool Contains(const Byte& t) const { return Find(t) != INVALID_INDEX; }
-
-	bool Remove(const Byte& t) {
-		const auto it = std::find(m_data.begin(), m_data.end(), t);
-		if (it == m_data.end()) {
-			return false;
-		}
-		m_data.erase(it);
-		return true;
-	}
-
-	bool RemoveAt(uint32_t index, uint32_t count = 1) {
-		if (index >= Size() || count == 0) {
-			return false;
-		}
-
-		const uint32_t erase_count = std::min(count, Size() - index);
-		m_data.erase(m_data.begin() + index, m_data.begin() + index + erase_count);
-		return true;
-	}
-
-	[[nodiscard]] bool IndexValid(uint32_t index) const { return index < Size(); }
-
-	Byte& operator[](uint32_t index) {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	const Byte& operator[](uint32_t index) const {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	[[nodiscard]] const Byte& At(uint32_t index) const {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	Byte*                     GetData() { return m_data.data(); }
-	[[nodiscard]] const Byte* GetData() const { return m_data.data(); }
-	[[nodiscard]] const Byte* GetDataConst() const { return m_data.data(); }
-
-	void Sort() {
-		std::sort(m_data.begin(), m_data.end(), [](Byte a, Byte b) {
-			return std::to_integer<uint8_t>(a) < std::to_integer<uint8_t>(b);
-		});
-	}
-
-	template <typename OP>
-	void Sort(OP&& comp_func) {
-		std::sort(m_data.begin(), m_data.end(), std::forward<OP>(comp_func));
-	}
-
-	bool operator==(const ByteBuffer& s) const { return m_data == s.m_data; }
-	bool operator!=(const ByteBuffer& s) const { return !(*this == s); }
-
-	iterator begin() { return m_data.begin(); } // NOLINT(readability-identifier-naming)
-	iterator end() { return m_data.end(); }     // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator begin() const {
-		return m_data.begin();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator end() const {
-		return m_data.end();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator cbegin() const {
-		return m_data.cbegin();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator cend() const {
-		return m_data.cend();
-	} // NOLINT(readability-identifier-naming)
-
-private:
-	DataType m_data;
-};
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_BYTEBUFFER_H_ */

src/common/debug.cpp

@@ -24,8 +24,4 @@ std::string Debug::GetLinker() {
 #endif
 }
 
-std::string Debug::GetBitness() {
-	return "64";
-}
-
 } // namespace Common

src/common/debug.h

@@ -10,7 +10,6 @@ namespace Common {
 namespace Debug {
 std::string GetCompiler();
 std::string GetLinker();
-std::string GetBitness();
 } // namespace Debug
 
 } // namespace Common

src/common/emulatorConfig.cpp

@@ -43,6 +43,10 @@ int32_t GetUserId() {
 	return g_config->user_id;
 }
 
+const std::string& GetAudioInputDevice() {
+	return g_config->audio_input_device;
+}
+
 PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
@@ -55,6 +59,14 @@ bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }
 
+bool VrEnabled() {
+	return g_config->vr_enabled;
+}
+
+bool AmdCpuEnabled() {
+	return g_config->amd_cpu_enabled;
+}
+
 uint32_t GetVblankFrequency() {
 	return std::clamp(g_config->vblank_frequency, 30u, 360u);
 }
@@ -123,6 +135,10 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+bool TessellationEnabled() {
+	return g_config->tessellation_enabled;
+}
+
 bool PlayGoHackEnabled() {
 	return g_config->playgo_hack_enabled;
 }

src/common/emulatorConfig.h

@@ -43,9 +43,12 @@ struct ConfigOptions {
 	uint32_t               screen_height               = 720;
 	std::string            user_name                   = "Kyty";
 	int32_t                user_id                     = DEFAULT_USER_ID;
+	std::string            audio_input_device;
 	PresentMode            present_mode                = PresentMode::Mailbox;
 	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
+	bool                   vr_enabled                  = false;
+	bool                   amd_cpu_enabled             = false;
 	uint32_t               vblank_frequency            = 60;
 	uint32_t               console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = false;
@@ -63,6 +66,7 @@ struct ConfigOptions {
 	bool                   gpu_assisted_validation_enabled = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
+	bool                   tessellation_enabled        = false;
 	bool                   playgo_hack_enabled         = false;
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	bool red_zone_protection_enabled = false;
@@ -76,9 +80,12 @@ uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
+const std::string& GetAudioInputDevice();
 PresentMode GetPresentMode();
 int32_t GetGpuIndex();
 bool     FullscreenEnabled();
+bool     VrEnabled();
+bool     AmdCpuEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();
@@ -104,6 +111,7 @@ bool GpuAssistedValidationEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+bool TessellationEnabled();
 bool PlayGoHackEnabled();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();

src/common/file.cpp

@@ -136,10 +136,6 @@ bool File::OpenInMem(void* buf, uint32_t buf_size) {
 	return true;
 }
 
-bool File::OpenInMem(ByteBuffer& buf) {
-	return OpenInMem(buf.GetData(), buf.Size());
-}
-
 bool File::CreateInMem() {
 	EXIT_IF(m_p->f != nullptr);
 
@@ -249,88 +245,78 @@ void File::Printf(const char* format, ...) {
 	Write(s.data(), static_cast<uint32_t>(s.size()));
 }
 
+static std::filesystem::path WithoutTrailingSeparator(
+	const std::filesystem::path& path) {
+		if (!path.empty() && !path.has_filename() &&
+			path != path.root_path()) {
+			return path.parent_path();
+		}
+
+		return path;
+}
+
+
 bool File::IsDirectoryExisting(const std::filesystem::path& path) {
-	auto path_str = PathToGenericString(path);
-	return SysFileIsDirectoryExisting(
-	    Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1) : path_str);
+	return SysFileIsDirectoryExisting(WithoutTrailingSeparator(path));
 }
 
 bool File::IsFileExisting(const std::filesystem::path& name) {
 	return SysFileIsFileExisting(name);
 }
 
-bool File::CreateDirectory(
-    const std::filesystem::path& path) // @suppress("Member declaration not found")
-{
-	auto path_str = PathToGenericString(path);
-	return SysFileCreateDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
-	                                                              : path_str);
+bool File::CreateDirectory(const std::filesystem::path& path) {
+    return SysFileCreateDirectory(
+        WithoutTrailingSeparator(path));
 }
 
 bool File::DeleteDirectory(const std::filesystem::path& path) {
-	auto path_str = PathToGenericString(path);
-	return SysFileDeleteDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
-	                                                              : path_str);
+    return SysFileDeleteDirectory(
+        WithoutTrailingSeparator(path));
 }
 
 bool File::CreateDirectories(const std::filesystem::path& path) {
-	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+    const auto normalized_path =
+        WithoutTrailingSeparator(path);
 
-	std::vector<std::string> list = Common::Split(real_path, "/");
+    auto current_path = normalized_path.root_path();
 
-	std::string p;
+    for (const auto& component:
+         normalized_path.relative_path()) {
+        current_path /= component;
 
-	for (uint32_t si = 0; si < list.size(); si++) {
-		const std::string& s = list[si];
+        if (IsDirectoryExisting(current_path)) {
+            continue;
+        }
 
-		if (si != 0 || Common::StartsWith(real_path, "/")) {
-			p += "/";
-		}
+        if (!CreateDirectory(current_path)) {
+            return false;
+        }
+    }
 
-		p += s;
-
-		if (IsDirectoryExisting(p)) {
-			continue;
-		}
-
-		if (!CreateDirectory(p)) // @suppress("Invalid arguments")
-		{
-			return false;
-		}
-	}
-
-	return true;
+    return true;
 }
 
 bool File::DeleteDirectories(const std::filesystem::path& path) {
-	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+    const auto normalized_path =
+        WithoutTrailingSeparator(path);
 
-	std::vector<std::string> list = Common::Split(real_path, "/");
+    std::vector<std::filesystem::path> directories;
+    auto current_path = normalized_path.root_path();
 
-	std::string              p;
-	std::vector<std::string> list2;
+    for (const auto& component:
+         normalized_path.relative_path()) {
+        current_path /= component;
+        directories.push_back(current_path);
+    }
 
-	for (uint32_t si = 0; si < list.size(); si++) {
-		const std::string& s = list[si];
+    for (auto it = directories.rbegin();
+         it != directories.rend(); ++it) {
+        if (!DeleteDirectory(*it)) {
+            return false;
+        }
+    }
 
-		if (si != 0 || Common::StartsWith(real_path, "/")) {
-			p += "/";
-		}
-
-		p += s;
-
-		list2.push_back(p);
-	}
-
-	uint32_t num = list2.size();
-
-	for (uint32_t si = num - 1; si < num; si--) {
-		if (!DeleteDirectory(list2[si])) {
-			return false;
-		}
-	}
-
-	return true;
+    return true;
 }
 
 bool File::DeleteFile(
@@ -345,7 +331,7 @@ bool File::Flush() {
 	return SysFileFlush(*m_p->f);
 }
 
-ByteBuffer File::ReadWholeBuffer() {
+std::vector<std::byte> File::ReadWholeBuffer() {
 	EXIT_IF(IsInvalid());
 	EXIT_IF(Tell() != 0);
 
@@ -353,25 +339,14 @@ ByteBuffer File::ReadWholeBuffer() {
 
 	EXIT_IF((s >> 32u) != 0);
 
-	ByteBuffer buf(s);
+	const auto            read_size = static_cast<uint32_t>(s);
+	std::vector<std::byte> buf(read_size);
 
-	Read(buf.GetData(), s);
+	Read(buf.data(), read_size);
 
 	return buf;
 }
 
-ByteBuffer File::Read(uint32_t size) {
-	ByteBuffer buf(size);
-	uint32_t   b = 0;
-	Read(buf.GetData(), size, &b);
-	buf.RemoveAt(b, size - b);
-	return buf;
-}
-
-void File::Write(const ByteBuffer& buf, uint32_t* bytes_written) {
-	Write(buf.GetDataConst(), buf.Size(), bytes_written);
-}
-
 DateTime File::GetLastAccessTimeUTC(const std::filesystem::path& name) {
 	SysTimeStruct t {};
 	SysFileToSystemTimeUtc(SysFileGetLastAccessTimeUtc(name), t);
@@ -470,51 +445,6 @@ bool File::SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name, const
 	return SysFileSetLastAccessAndWriteTimeUtc(name, af, wf);
 }
 
-std::vector<File::FindInfo> File::FindFiles(const std::filesystem::path& path) {
-	std::vector<sys_file_find_t> files;
-
-	SysFileFindFiles(path, files);
-
-	auto     path_str = PathToGenericString(path);
-	uint32_t len      = static_cast<uint32_t>(path_str.size());
-
-	if (!Common::EndsWith(path_str, "/") && !Common::EndsWith(path_str, "\\")) {
-		len++;
-	}
-
-	std::vector<File::FindInfo> ret;
-	ret.reserve(files.size());
-
-	for (const auto& f: files) {
-		File::FindInfo r {};
-
-		r.path_with_name     = f.path_with_name;
-		r.rel_path_with_name = Common::Mid(PathToGenericString(f.path_with_name), len);
-		r.size               = f.size;
-
-		SysTimeStruct at {};
-		SysTimeStruct wt {};
-
-		SysFileToSystemTimeUtc(f.last_access_time, at);
-		SysFileToSystemTimeUtc(f.last_write_time, wt);
-
-		if (!at.is_invalid && !wt.is_invalid) {
-			r.last_access_time = DateTime(Date(at.Year, at.Month, at.Day),
-			                              Time(at.Hour, at.Minute, at.Second, at.Milliseconds));
-			r.last_write_time  = DateTime(Date(wt.Year, wt.Month, wt.Day),
-			                              Time(wt.Hour, wt.Minute, wt.Second, wt.Milliseconds));
-		}
-
-		// printf("%s, %s, %" PRIu64", %s, %s\n", r.path_with_name.c_str(),
-		// r.rel_path_with_name.c_str(), r.size, r.last_access_time.ToString().c_str(),
-		// r.last_write_time.ToString().c_str());
-
-		ret.push_back(std::move(r));
-	}
-
-	return ret;
-}
-
 std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& path) {
 	std::vector<sys_dir_entry_t> files;
 
@@ -523,13 +453,8 @@ std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& pat
 	std::vector<File::DirEntry> ret;
 	ret.reserve(files.size());
 
-	for (const auto& f: files) {
-		File::DirEntry r {};
-
-		r.name    = f.name;
-		r.is_file = f.is_file;
-
-		ret.push_back(std::move(r));
+	for (auto& f: files) {
+		ret.emplace_back(std::move(f.name), f.is_file);
 	}
 
 	return ret;

src/common/file.h

@@ -1,10 +1,10 @@
 #ifndef KYTY_COMMON_FILE_H_
 #define KYTY_COMMON_FILE_H_
 
-#include "common/byteBuffer.h"
 #include "common/common.h"
 #include "common/dateTime.h"
 
+#include <cstddef>
 #include <filesystem>
 #include <memory>
 #include <string>
@@ -30,14 +30,6 @@ class File {
 public:
 	enum class Mode { Read, Write, ReadWrite, WriteRead };
 
-	struct FindInfo {
-		std::filesystem::path path_with_name;
-		std::filesystem::path rel_path_with_name;
-		DateTime              last_access_time;
-		DateTime              last_write_time;
-		uint64_t              size;
-	};
-
 	struct DirEntry {
 		std::string name;
 		bool        is_file;
@@ -51,7 +43,6 @@ public:
 	bool Create(const std::filesystem::path& name);
 	bool Open(const std::filesystem::path& name, Mode mode);
 	bool OpenInMem(void* buf, uint32_t buf_size);
-	bool OpenInMem(ByteBuffer& buf); // NOLINT(google-runtime-references)
 	bool CreateInMem();
 
 	void Close();
@@ -73,14 +64,12 @@ public:
 
 	void GetLastAccessAndWriteTimeUTC(DateTime* access, DateTime* write);
 
-	void       Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
-	ByteBuffer Read(uint32_t size);
-	void       Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
-	void       Write(const ByteBuffer& buf, uint32_t* bytes_written = nullptr);
+	void Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
+	void Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
 
 	void Printf(const char* format, ...) KYTY_FORMAT_PRINTF(2, 3);
 
-	ByteBuffer ReadWholeBuffer();
+	std::vector<std::byte> ReadWholeBuffer();
 
 	static uint64_t Size(const std::filesystem::path& name);
 
@@ -101,7 +90,6 @@ public:
 	static bool SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name,
 	                                         const DateTime& access, const DateTime& write);
 
-	static std::vector<FindInfo> FindFiles(const std::filesystem::path& path);
 	static std::vector<DirEntry> GetDirEntries(const std::filesystem::path& path);
 	static bool CopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 	static bool RenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/hash.h

@@ -54,103 +54,6 @@ inline uint32_t hash(const void* key, uint32_t key_len) {
 	return hash;
 }
 
-inline uint32_t hash8(uint8_t key) {
-	uint32_t hash = 0;
-
-	uint8_t* ptr = &key;
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash16(uint16_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash32(uint32_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[2];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[3];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash64(uint64_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[2];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[3];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[4];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[5];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[6];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[7];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_HASH_H_ */

src/common/logging/log.cpp

@@ -3,7 +3,6 @@
 
 #include "common/assert.h"
 #include "common/emulatorConfig.h"
-#include "common/stringUtils.h"
 
 #include <cstdio>
 #include <filesystem>
@@ -47,7 +46,7 @@ std::shared_ptr<spdlog::logger> MakeFileLogger(std::string                  name
 	}
 
 	auto sink =
-	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(Common::PathToString(path), true);
+	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(path.native(), true);
 	return MakeLogger(std::move(name), std::move(sink));
 }
 
@@ -129,6 +128,14 @@ static void WriteImpl(std::string_view text, fmt::text_style style = {}) {
 	}
 }
 
+void WriteToConsoleAndLog(std::string_view text) {
+	WriteImpl(text);
+	if (g_initialized && g_direction != Direction::Console) {
+		WriteStdout(text);
+	}
+	Flush();
+}
+
 void WriteFatal(std::string_view text) {
 	if (g_direction == Direction::Silent || !g_initialized) {
 		WriteStdout(text);

src/common/logging/log.h

@@ -25,6 +25,7 @@ Direction GetDirection();
 bool      IsSilent();
 void      Write(std::string_view text);
 void      Write(fmt::text_style style, std::string_view text);
+void      WriteToConsoleAndLog(std::string_view text);
 void      WriteFatal(std::string_view text);
 void      WriteFatal(fmt::text_style style, std::string_view text);
 void      Flush();

src/common/magicEnum.h

@@ -1,42 +0,0 @@
-#ifndef KYTY_COMMON_MAGICENUM_H_
-#define KYTY_COMMON_MAGICENUM_H_
-
-#include "common/common.h"
-#include "common/stringUtils.h"
-#include "magic_enum.hpp" // IWYU pragma: export
-
-namespace Common {
-
-template <typename E>
-inline std::string EnumName(E v) {
-	auto str = magic_enum::enum_name(v);
-	return std::string(str.data(), str.length());
-}
-
-template <typename E>
-inline std::string EnumName8(E v) {
-	auto str = magic_enum::enum_name(v);
-	return std::string(str.data(), str.length());
-}
-
-template <typename E>
-inline E EnumValue(const std::string& str, E default_value) {
-	auto v = magic_enum::enum_cast<E>(str.c_str());
-	if (v.has_value()) {
-		return v.value();
-	}
-	return default_value;
-}
-
-#define KYTY_ENUM_RANGE(e, mx, mn)                                                                 \
-	namespace magic_enum::customize {                                                              \
-	template <>                                                                                    \
-	struct enum_range<e> {                                                                         \
-		static constexpr int min = (mx);                                                           \
-		static constexpr int max = (mn);                                                           \
-	};                                                                                             \
-	}
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_MAGICENUM_H_ */

src/common/platform/sysFileIO.h

@@ -18,14 +18,6 @@ enum sys_file_cache_type_t {
 	SYS_FILE_CACHE_SEQUENTIAL_SCAN = 2  // NOLINT(readability-identifier-naming)
 };
 
-// NOLINTNEXTLINE(readability-identifier-naming)
-struct sys_file_find_t {
-	std::filesystem::path path_with_name;
-	SysFileTimeStruct     last_access_time;
-	SysFileTimeStruct     last_write_time;
-	uint64_t              size;
-};
-
 // NOLINTNEXTLINE(readability-identifier-naming)
 struct sys_dir_entry_t {
 	std::string name;
@@ -52,7 +44,6 @@ uint64_t    SysFileTell(sys_file_t& f);                    // NOLINT(google-runt
 bool        SysFileTruncate(sys_file_t& f, uint64_t size); // NOLINT(google-runtime-references)
 bool        SysFileUnlink(sys_file_t&                  f,
                           const std::filesystem::path& name); // NOLINT(google-runtime-references)
-void        SysFileWrite(uint32_t n, sys_file_t& f);          // NOLINT(google-runtime-references)
 uint64_t    SysFileSize(const std::filesystem::path& file_name);
 bool        SysFileIsError(sys_file_t& f); // NOLINT(google-runtime-references)
 bool        SysFileIsDirectoryExisting(const std::filesystem::path& path);
@@ -76,8 +67,6 @@ bool SysFileSetLastWriteTimeUtc(const std::filesystem::path& name, SysFileTimeSt
 bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
                                          SysFileTimeStruct& access, SysFileTimeStruct& write);
 // NOLINTNEXTLINE(google-runtime-references)
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out);
-// NOLINTNEXTLINE(google-runtime-references)
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out);
 bool SysFileCopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 bool SysFileRenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/platform/sysLinuxFileIO.cpp

@@ -11,6 +11,7 @@
 
 #include <cerrno>
 #include <cstdlib>
+#include <cstring>
 #include <dirent.h>
 #include <fcntl.h>
 #include <filesystem>
@@ -151,10 +152,6 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
-void SysFileWrite(uint32_t n, sys_file_t& f) {
-	SysFileWrite(&n, 4, f);
-}
-
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 
@@ -613,51 +610,6 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	//	}
 }
 
-// Recursively collect regular files.
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
-	auto real_path = get_internal_name(path);
-
-	DIR* dir = opendir(real_path.string().c_str());
-	if (dir == nullptr) {
-		return;
-	}
-
-	for (const dirent* entry = readdir(dir); entry != nullptr; entry = readdir(dir)) {
-		const std::string file_name(entry->d_name);
-
-		if (file_name == "." || file_name == "..") {
-			continue;
-		}
-
-		auto        child = real_path / file_name;
-		struct stat s {};
-
-		// lstat, so a symlink is never followed into a cycle during the recursive walk.
-		if (0 != lstat(child.string().c_str(), &s)) {
-			continue;
-		}
-
-		if (S_ISDIR(s.st_mode)) {
-			SysFileFindFiles(child, out);
-		} else if (S_ISREG(s.st_mode)) {
-			sys_file_find_t r {};
-
-			r.path_with_name              = child;
-			r.size                        = static_cast<uint64_t>(s.st_size);
-			r.last_access_time.is_invalid = false;
-			r.last_access_time.time       = s.st_atime;
-			r.last_access_time.nanos      = KYTY_STAT_ATIME_NS(s);
-			r.last_write_time.is_invalid  = false;
-			r.last_write_time.time        = s.st_mtime;
-			r.last_write_time.nanos       = KYTY_STAT_MTIME_NS(s);
-
-			out.push_back(r);
-		}
-	}
-
-	closedir(dir);
-}
-
 // Keep "." and ".." to match FindFirstFileW.
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
 	auto real_path = get_internal_name(path);

src/common/platform/sysLinuxVirtual.cpp

@@ -5,7 +5,6 @@
 #else
 
 #include "common/assert.h"
-#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 #include <atomic>
@@ -28,12 +27,12 @@
 #define KYTY_FIXED_NOREPLACE
 #endif
 
-namespace Common {
+namespace Common::VirtualMemory {
 
 static pthread_mutex_t              g_virtual_mutex {};
 static std::map<uintptr_t, size_t>* g_allocs = nullptr;
 
-void SysVirtualInit() {
+void Init() {
 	pthread_mutexattr_t attr {};
 
 	pthread_mutexattr_init(&attr);
@@ -48,19 +47,19 @@ void SysVirtualInit() {
 	g_allocs = new std::map<uintptr_t, size_t>;
 }
 
-static int get_protection_flag(VirtualMemory::Mode mode) {
+static int get_protection_flag(Mode mode) {
 	int protect = PROT_NONE;
 	switch (mode) {
-		case VirtualMemory::Mode::Read: protect = PROT_READ; break;
-		case VirtualMemory::Mode::Write:
-		case VirtualMemory::Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
-		case VirtualMemory::Mode::Execute: protect = PROT_EXEC; break;
-		case VirtualMemory::Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
-		case VirtualMemory::Mode::ExecuteWrite:
-		case VirtualMemory::Mode::ExecuteReadWrite:
+		case Mode::Read: protect = PROT_READ; break;
+		case Mode::Write:
+		case Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
+		case Mode::Execute: protect = PROT_EXEC; break;
+		case Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
+		case Mode::ExecuteWrite:
+		case Mode::ExecuteReadWrite:
 			protect = PROT_EXEC | PROT_WRITE | PROT_READ;
 			break; // NOLINT
-		case VirtualMemory::Mode::NoAccess:
+		case Mode::NoAccess:
 		default: protect = PROT_NONE; break;
 	}
 	return protect;
@@ -130,7 +129,7 @@ static void* map_anonymous(uintptr_t addr, size_t size, int protect, int flags)
 	return mmap(nullptr, size, protect, flags, -1, 0); // NOLINT
 }
 
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -154,8 +153,7 @@ static uintptr_t align_up(uintptr_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment) {
+uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -224,7 +222,7 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 	}
 
 	if (ptr == MAP_FAILED) {
-		return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
+		return AllocAligned(address, size, mode, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
@@ -281,7 +279,7 @@ static bool is_mapped(void* ptr, size_t length) {
 }
 #endif
 
-bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr    = static_cast<uintptr_t>(address);
@@ -318,15 +316,15 @@ bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode m
 	return false;
 }
 
-bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
-	return SysVirtualProtect(address, size, mode);
+bool Commit(uint64_t address, uint64_t size, Mode mode) {
+	return Protect(address, size, mode);
 }
 
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
-	return SysVirtualReserveAligned(address, size, 1);
+uint64_t Reserve(uint64_t address, uint64_t size) {
+	return ReserveAligned(address, size, 1);
 }
 
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -395,7 +393,7 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 	}
 
 	if (ptr == MAP_FAILED) {
-		return SysVirtualReserveAligned(address, size, alignment << 1u);
+		return ReserveAligned(address, size, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
@@ -405,7 +403,7 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 	return ret_addr;
 }
 
-bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
+bool ReserveFixed(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -441,9 +439,9 @@ bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	return false;
 }
 
-bool SysVirtualDecommit(uint64_t address, uint64_t size) {
+bool Decommit(uint64_t address, uint64_t size) {
 	// Drop physical pages while preserving the reservation.
-	if (!SysVirtualProtect(address, size, VirtualMemory::Mode::NoAccess)) {
+	if (!Protect(address, size, Mode::NoAccess)) {
 		return false;
 	}
 
@@ -467,7 +465,7 @@ bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualFree(uint64_t address) {
+bool Free(uint64_t address) {
 	EXIT_IF(g_allocs == nullptr);
 	size_t size = 0;
 
@@ -487,7 +485,7 @@ bool SysVirtualFree(uint64_t address) {
 	return munmap(reinterpret_cast<void*>(addr), size) == 0;
 }
 
-bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
+bool FreeRange(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 	if (size == 0 || (address & 0xfffu) != 0 || (size & 0xfffu) != 0) {
 		return false;
@@ -543,7 +541,7 @@ bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Protect(uint64_t address, uint64_t size, Mode mode) {
 	const auto addr       = static_cast<uintptr_t>(address);
 	const auto page_start = addr >> 12u;
 	const auto page_end   = (addr + size - 1) >> 12u;
@@ -551,10 +549,10 @@ bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode
 	                get_protection_flag(mode)) == 0;
 }
 
-bool SysVirtualFlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
+bool FlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
 	return true;
 }
 
-} // namespace Common
+} // namespace Common::VirtualMemory
 
 #endif

src/common/platform/sysSwapByteOrder.h

@@ -1,53 +0,0 @@
-#ifndef KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
-#define KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
-
-#include "common/common.h"
-
-#include <type_traits>
-
-inline uint16_t SwapByteOrder16(uint16_t value) {
-	uint16_t hi = value << 8u;
-	uint16_t lo = value >> 8u;
-	return hi | lo;
-}
-
-inline uint32_t SwapByteOrder32(uint32_t value) {
-	return __builtin_bswap32(value);
-}
-
-inline uint64_t SwapByteOrder64(uint64_t value) {
-	return __builtin_bswap64(value);
-}
-
-template <typename T>
-inline void SwapByteOrder(T& x) {
-	if (sizeof(x) == 2) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder16(static_cast<std::make_unsigned_t<uint16_t>>(x)));
-		} else {
-			x = SwapByteOrder16(x);
-		}
-	}
-	if (sizeof(x) == 4) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder32(static_cast<std::make_unsigned_t<uint32_t>>(x)));
-		} else {
-			x = SwapByteOrder32(x);
-		}
-	}
-	if (sizeof(x) == 8) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder64(static_cast<std::make_unsigned_t<uint64_t>>(x)));
-		} else {
-			x = SwapByteOrder64(x);
-		}
-	}
-}
-
-template <typename T>
-inline void NoSwapByteOrder(T& x) {}
-
-#endif /* KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_ */

src/common/platform/sysTimer.h

@@ -151,10 +151,6 @@ inline void SysTimeTToSystem(time_t t, SysTimeStruct& s) {
 	SysFileToSystemTimeUtc(ft, s);
 }
 
-inline time_t SysTimegm(struct tm* tm) {
-	return timegm(tm);
-}
-
 inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t) {
 	struct tm i {};
 
@@ -165,7 +161,7 @@ inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t)
 	i.tm_min  = f.Minute;
 	i.tm_sec  = f.Second;
 
-	t.is_invalid = (f.is_invalid || (t.time = SysTimegm(&i)) == static_cast<time_t>(-1));
+	t.is_invalid = (f.is_invalid || (t.time = ::timegm(&i)) == static_cast<time_t>(-1));
 }
 
 // Retrieves the current local date and time.

src/common/platform/sysVirtual.h

@@ -1,27 +0,0 @@
-#ifndef KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
-#define KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
-
-#include "common/common.h"
-#include "common/virtualMemory.h"
-
-namespace Common {
-
-void SysVirtualInit();
-
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment);
-bool     SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-bool     SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size);
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment);
-bool     SysVirtualReserveFixed(uint64_t address, uint64_t size);
-bool     SysVirtualDecommit(uint64_t address, uint64_t size);
-bool     SysVirtualFree(uint64_t address);
-bool     SysVirtualFreeRange(uint64_t address, uint64_t size);
-bool     SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-bool     SysVirtualFlushInstructionCache(uint64_t address, uint64_t size);
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_ */

src/common/platform/sysWindowsFileIO.cpp

@@ -14,6 +14,7 @@
 #include "common/stringUtils.h"
 
 #include <cstdlib>
+#include <cstring>
 #include <vector>
 
 // NOLINTNEXTLINE(readability-identifier-naming)
@@ -56,10 +57,10 @@ static DWORD GetCacheAccessType(sys_file_cache_type_t t) {
 	}
 
 	if (t == SYS_FILE_CACHE_SEQUENTIAL_SCAN) {
-		return SYS_FILE_CACHE_SEQUENTIAL_SCAN;
+		return FILE_FLAG_SEQUENTIAL_SCAN;
 	}
 
-	return SYS_FILE_CACHE_AUTO;
+	return FILE_ATTRIBUTE_NORMAL;
 }
 
 void SysFileRead(void* data, uint32_t size, sys_file_t& f, uint32_t* bytes_read) {
@@ -135,10 +136,6 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
-void SysFileWrite(uint32_t n, sys_file_t& f) {
-	SysFileWrite(&n, 4, f);
-}
-
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 
@@ -479,66 +476,11 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	return ok;
 }
 
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
-	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
-	if (!Common::EndsWith(real_path, "/")) {
-		real_path += "/";
-	}
-
-	std::string pattern = real_path + "*";
-
-	HANDLE           h = nullptr;
-	WIN32_FIND_DATAW data;
-
-	auto wide_pattern = std::filesystem::path(pattern).wstring();
-	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
-
-	if (h == INVALID_HANDLE_VALUE) {
-		return;
-	}
-
-	do {
-		std::filesystem::path file_name(data.cFileName);
-		auto                  file_name_str = Common::PathToString(file_name);
-
-		if (file_name_str == "." || file_name_str == "..") {
-			continue;
-		}
-
-		if ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) != 0u) {
-			SysFileFindFiles(std::filesystem::path(real_path) / file_name, out);
-		} else {
-			sys_file_find_t r {};
-
-			r.path_with_name              = std::filesystem::path(real_path) / file_name;
-			r.size                        = (static_cast<uint64_t>(data.nFileSizeHigh) << 32u) +
-			                                static_cast<uint64_t>(data.nFileSizeLow);
-			r.last_access_time.is_invalid = false;
-			r.last_access_time.time       = data.ftLastAccessTime;
-			r.last_write_time.is_invalid  = false;
-			r.last_write_time.time        = data.ftLastWriteTime;
-
-			out.push_back(r);
-		}
-
-	} while (FindNextFileW(h, &data) != 0);
-
-	FindClose(h);
-}
-
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
-	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
-	if (!Common::EndsWith(real_path, "/")) {
-		real_path += "/";
-	}
+	const auto pattern = path / L"*";
 
-	std::string pattern = real_path + "*";
-
-	HANDLE           h = nullptr;
-	WIN32_FIND_DATAW data;
-
-	auto wide_pattern = std::filesystem::path(pattern).wstring();
-	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
+	WIN32_FIND_DATAW data {};
+	HANDLE h = FindFirstFileW(pattern.c_str(), &data);
 
 	if (h == INVALID_HANDLE_VALUE) {
 		return;
@@ -552,7 +494,7 @@ void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entr
 		r.is_file = ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) == 0u);
 		r.name    = Common::PathToString(file_name);
 
-		out.push_back(r);
+		out.push_back(std::move(r));
 
 	} while (FindNextFileW(h, &data) != 0);
 

src/common/platform/sysWindowsVirtual.cpp

@@ -10,7 +10,6 @@
 #include <windows.h> // IWYU pragma: keep
 
 #include "common/assert.h"
-#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 // IWYU pragma: no_include <basetsd.h>
@@ -22,33 +21,33 @@
 // IWYU pragma: no_include <winerror.h>
 // IWYU pragma: no_include <wtypes.h>
 
-namespace Common {
+namespace Common::VirtualMemory {
 
-static DWORD GetProtectionFlag(VirtualMemory::Mode mode) {
+static DWORD GetProtectionFlag(Mode mode) {
 	DWORD protect = PAGE_NOACCESS;
 	switch (mode) {
-		case VirtualMemory::Mode::Read: protect = PAGE_READONLY; break;
+		case Mode::Read: protect = PAGE_READONLY; break;
 
-		case VirtualMemory::Mode::Write:
-		case VirtualMemory::Mode::ReadWrite: protect = PAGE_READWRITE; break;
+		case Mode::Write:
+		case Mode::ReadWrite: protect = PAGE_READWRITE; break;
 
-		case VirtualMemory::Mode::Execute: protect = PAGE_EXECUTE; break;
+		case Mode::Execute: protect = PAGE_EXECUTE; break;
 
-		case VirtualMemory::Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
+		case Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
 
-		case VirtualMemory::Mode::ExecuteWrite:
-		case VirtualMemory::Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
+		case Mode::ExecuteWrite:
+		case Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
 
-		case VirtualMemory::Mode::NoAccess:
+		case Mode::NoAccess:
 		default: protect = PAGE_NOACCESS; break;
 	}
 	return protect;
 }
 
-void SysVirtualInit() {}
+void Init() {}
 
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
-	auto ptr = (address == 0 ? SysVirtualAllocAligned(address, size, mode, 1)
+uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
+	auto ptr = (address == 0 ? AllocAligned(address, size, mode, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE),
@@ -59,7 +58,7 @@ uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mo
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc() failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return SysVirtualAllocAligned(address, size, mode, 1);
+			return AllocAligned(address, size, mode, 1);
 		}
 	}
 	return ptr;
@@ -80,8 +79,7 @@ static uint64_t AlignUp(uint64_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment) {
+uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2 failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
 		return 0;
@@ -135,13 +133,13 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 			printf("VirtualAlloc2(alignment = 0x%016" PRIx64 ") failed: 0x%08" PRIx32 "\n",
 			       alignment, err);
 		} else {
-			return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
+			return AllocAligned(address, size, mode, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(VirtualAlloc(
 	    reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	    static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE), GetProtectionFlag(mode)));
@@ -161,7 +159,7 @@ bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode m
 	return true;
 }
 
-bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Commit(uint64_t address, uint64_t size, Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_COMMIT), GetProtectionFlag(mode)));
@@ -179,8 +177,8 @@ bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode)
 	return true;
 }
 
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
-	auto ptr = (address == 0 ? SysVirtualReserveAligned(address, size, 1)
+uint64_t Reserve(uint64_t address, uint64_t size) {
+	auto ptr = (address == 0 ? ReserveAligned(address, size, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS)));
@@ -190,13 +188,13 @@ uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc(MEM_RESERVE) failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return SysVirtualReserveAligned(address, size, 1);
+			return ReserveAligned(address, size, 1);
 		}
 	}
 	return ptr;
 }
 
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2(MEM_RESERVE) failed: 0x%08" PRIx32 "\n",
 		       static_cast<uint32_t>(GetLastError()));
@@ -250,13 +248,13 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 			       "\n",
 			       alignment, err);
 		} else {
-			return SysVirtualReserveAligned(address, size, alignment << 1u);
+			return ReserveAligned(address, size, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
+bool ReserveFixed(uint64_t address, uint64_t size) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS));
@@ -276,7 +274,7 @@ bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualDecommit(uint64_t address, uint64_t size) {
+bool Decommit(uint64_t address, uint64_t size) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                MEM_DECOMMIT) == 0) {
 		printf("VirtualFree(MEM_DECOMMIT) failed: 0x%08" PRIx32 "\n",
@@ -286,7 +284,7 @@ bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualFree(uint64_t address) {
+bool Free(uint64_t address) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), 0, MEM_RELEASE) ==
 	    0) {
 		printf("VirtualFree() failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
@@ -295,7 +293,7 @@ bool SysVirtualFree(uint64_t address) {
 	return true;
 }
 
-bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
+bool FreeRange(uint64_t address, uint64_t size) {
 	if (address == 0 || size == 0) {
 		return false;
 	}
@@ -313,10 +311,10 @@ bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 		}
 		current = next;
 	}
-	return current - address == size && SysVirtualFree(address);
+	return current - address == size && Free(address);
 }
 
-bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Protect(uint64_t address, uint64_t size, Mode mode) {
 	DWORD old_protect = 0;
 	if (VirtualProtect(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                   GetProtectionFlag(mode), &old_protect) == 0) {
@@ -326,7 +324,7 @@ bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode
 	return true;
 }
 
-bool SysVirtualFlushInstructionCache(uint64_t address, uint64_t size) {
+bool FlushInstructionCache(uint64_t address, uint64_t size) {
 	if (::FlushInstructionCache(GetCurrentProcess(),
 	                            reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)),
 	                            size) == 0) {
@@ -337,6 +335,6 @@ bool SysVirtualFlushInstructionCache(uint64_t address, uint64_t size) {
 	return true;
 }
 
-} // namespace Common
+} // namespace Common::VirtualMemory
 
 #endif

src/common/stringUtils.h

@@ -1,7 +1,6 @@
 #ifndef KYTY_COMMON_STRING_UTILS_H_
 #define KYTY_COMMON_STRING_UTILS_H_
 
-#include "common/byteBuffer.h"
 #include "common/common.h"
 
 #include <algorithm>
@@ -9,7 +8,6 @@
 #include <charconv>
 #include <codecvt>
 #include <filesystem>
-#include <fmt/format.h>
 #include <locale>
 #include <string>
 #include <string_view>
@@ -18,8 +16,6 @@
 
 namespace Common {
 
-constexpr uint32_t FIND_INVALID_INDEX = static_cast<uint32_t>(-1);
-
 inline std::string PathToString(const std::filesystem::path& path) {
 #if defined(__cpp_char8_t)
 	auto u8 = path.u8string();
@@ -38,8 +34,12 @@ inline std::string PathToGenericString(const std::filesystem::path& path) {
 #endif
 }
 
-inline bool IsSpace(char c) {
-	return std::isspace(static_cast<unsigned char>(c)) != 0;
+inline std::filesystem::path PathFromUtf8(std::string_view text) {
+#if defined(__cpp_char8_t)
+	return std::filesystem::path(std::u8string(text.begin(), text.end()));
+#else
+	return std::filesystem::u8path(text.begin(), text.end());
+#endif
 }
 
 inline bool EqualNoCase(std::string_view lhs, std::string_view rhs) {
@@ -63,65 +63,6 @@ inline std::string ToLower(std::string_view text) {
 	return ret;
 }
 
-inline bool ContainsStr(std::string_view text, std::string_view needle) {
-	return text.find(needle) != std::string_view::npos;
-}
-
-inline bool StartsWith(std::string_view text, std::string_view prefix) {
-	return text.starts_with(prefix);
-}
-
-inline bool StartsWith(std::string_view text, char prefix) {
-	return !text.empty() && text.front() == prefix;
-}
-
-inline bool EndsWith(std::string_view text, std::string_view suffix) {
-	return text.ends_with(suffix);
-}
-
-inline bool EndsWith(std::string_view text, char suffix) {
-	return !text.empty() && text.back() == suffix;
-}
-
-inline bool ContainsChar(std::string_view text, char needle) {
-	return text.find(needle) != std::string_view::npos;
-}
-
-inline bool IndexValid(std::string_view text, uint32_t index) {
-	return index < text.size();
-}
-
-inline uint32_t FindIndex(std::string_view text, std::string_view needle, size_t from = 0) {
-	const auto pos = text.find(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindIndex(std::string_view text, char needle, size_t from = 0) {
-	const auto pos = text.find(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindLastIndex(std::string_view text, std::string_view needle,
-                              size_t from = FIND_INVALID_INDEX) {
-	if (needle.empty()) {
-		return 0;
-	}
-	if (from == FIND_INVALID_INDEX || from > text.size()) {
-		from = text.size();
-	}
-	const auto pos = text.rfind(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindLastIndex(std::string_view text, char needle,
-                              size_t from = FIND_INVALID_INDEX) {
-	if (from == FIND_INVALID_INDEX || from > text.size()) {
-		from = text.size();
-	}
-	const auto pos = text.rfind(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
 inline std::string Mid(std::string_view text, size_t first, size_t count = std::string_view::npos) {
 	if (first >= text.size()) {
 		return {};
@@ -129,49 +70,18 @@ inline std::string Mid(std::string_view text, size_t first, size_t count = std::
 	return std::string(text.substr(first, count));
 }
 
-inline std::string Left(std::string_view text, size_t count) {
-	return Mid(text, 0, count);
-}
-
-inline std::string Right(std::string_view text, size_t count) {
-	if (count >= text.size()) {
-		return std::string(text);
-	}
-	return Mid(text, text.size() - count, count);
-}
-
 inline std::string RemoveLast(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return Left(text, text.size() - count);
+	return std::string(text.substr(0, text.size() - count));
 }
 
 inline std::string RemoveFirst(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return Right(text, text.size() - count);
-}
-
-inline std::string TrimRight(std::string_view text) {
-	size_t end = text.size();
-	while (end > 0 && IsSpace(text[end - 1])) {
-		end--;
-	}
-	return std::string(text.substr(0, end));
-}
-
-inline std::string TrimLeft(std::string_view text) {
-	size_t first = 0;
-	while (first < text.size() && IsSpace(text[first])) {
-		first++;
-	}
-	return std::string(text.substr(first));
-}
-
-inline std::string Trim(std::string_view text) {
-	return TrimRight(TrimLeft(text));
+	return std::string(text.substr(count));
 }
 
 inline std::string ReplaceChar(std::string_view text, char old_char, char new_char) {
@@ -204,16 +114,6 @@ inline std::string FilenameWithoutDirectory(std::string_view text) {
 	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(pos + 1));
 }
 
-inline std::string FilenameWithoutExtension(std::string_view text) {
-	const auto pos = text.find_last_of('.');
-	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(0, pos));
-}
-
-inline std::string ExtensionWithoutFilename(std::string_view text) {
-	const auto pos = text.find_last_of('.');
-	return pos == std::string_view::npos ? std::string() : std::string(text.substr(pos));
-}
-
 inline std::string FixFilenameSlash(std::string_view text) {
 	return ReplaceChar(text, '\\', '/');
 }
@@ -257,65 +157,12 @@ inline std::vector<std::string> Split(std::string_view text, char sep, bool keep
 	return Split(text, std::string_view(&sep, 1), keep_empty);
 }
 
-inline std::string Concat(const std::vector<std::string>& list, std::string_view sep) {
-	std::string ret;
-	for (size_t i = 0; i < list.size(); i++) {
-		if (i != 0) {
-			ret += sep;
-		}
-		ret += list[i];
-	}
-	return ret;
-}
-
-inline std::string Concat(const std::vector<std::string>& list, char sep) {
-	return Concat(list, std::string_view(&sep, 1));
-}
-
-inline uint32_t ToUint32(std::string_view text, int base = 10) {
-	uint32_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
-inline uint64_t ToUint64(std::string_view text, int base = 10) {
-	uint64_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
 inline int32_t ToInt32(std::string_view text, int base = 10) {
 	int32_t value = 0;
 	std::from_chars(text.data(), text.data() + text.size(), value, base);
 	return value;
 }
 
-inline int64_t ToInt64(std::string_view text, int base = 10) {
-	int64_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
-inline double ToDouble(std::string_view text) {
-	return std::stod(std::string(text));
-}
-
-inline float ToFloat(std::string_view text) {
-	return std::stof(std::string(text));
-}
-
-inline std::string SafeCsv(std::string_view text) {
-	const bool add_space =
-	    !text.empty() && (text.front() == '+' || text.front() == '=' || text.front() == '-');
-	if (text.find('"') != std::string_view::npos || text.find(';') != std::string_view::npos ||
-	    text.find('+') != std::string_view::npos || text.find('=') != std::string_view::npos ||
-	    text.find('-') != std::string_view::npos) {
-		return "\"" + std::string(add_space ? " " : "") +
-		       ReplaceStr(std::string(text), "\"", "\"\"") + "\"";
-	}
-	return std::string(text);
-}
-
 inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	if (utf16.empty()) {
 		return {};
@@ -324,23 +171,6 @@ inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
 }
 
-inline ByteBuffer HexToBin(std::string_view text) {
-	ByteBuffer ret;
-	for (size_t i = 0; i + 1 < text.size(); i += 2) {
-		ret.Add(static_cast<Byte>((ToUint32(text.substr(i, 2), 16) & 0xffu)));
-	}
-	return ret;
-}
-
-inline std::string HexFromBin(const ByteBuffer& bin) {
-	std::string ret;
-	ret.reserve(bin.Size() * 2);
-	for (uint32_t i = 0; i < bin.Size(); i++) {
-		ret += fmt::format("{:02X}", std::to_integer<uint8_t>(bin.At(i)));
-	}
-	return ret;
-}
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_STRING_UTILS_H_ */

src/common/threads.cpp

@@ -8,7 +8,6 @@
 #include <chrono>             // IWYU pragma: keep
 #include <condition_variable> // IWYU pragma: keep
 #include <mutex>
-#include <vector>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS && KYTY_COMPILER == KYTY_COMPILER_CLANG
 #define KYTY_WIN_CS
@@ -197,37 +196,7 @@ struct CondVarPrivate {
 #endif
 };
 
-static std::recursive_mutex                         g_cond_waiters_mutex;
-static std::vector<std::pair<int, CondVarPrivate*>> g_cond_waiters;
-static wait_poll_func_t                             g_cond_wait_poll_callback = nullptr;
-
-static void WakeCondVar(CondVarPrivate* cond_var) {
-#ifdef KYTY_WIN_CS
-	static auto func = ResolveWakeAllConditionVariable();
-	EXIT_NOT_IMPLEMENTED(func == nullptr);
-	func(&cond_var->m_cv);
-#else
-	cond_var->m_cv.notify_all();
-#endif
-}
-
-static void RegisterCondWaiter(CondVarPrivate* cond_var) {
-	std::lock_guard lock(g_cond_waiters_mutex);
-	g_cond_waiters.emplace_back(Thread::GetThreadIdUnique(), cond_var);
-}
-
-static void UnregisterCondWaiter(CondVarPrivate* cond_var) {
-	const auto      thread_id = Thread::GetThreadIdUnique();
-	std::lock_guard lock(g_cond_waiters_mutex);
-
-	const auto it = std::find_if(g_cond_waiters.begin(), g_cond_waiters.end(),
-	                             [thread_id, cond_var](const auto& waiter) {
-		                             return waiter.first == thread_id && waiter.second == cond_var;
-	                             });
-	if (it != g_cond_waiters.end()) {
-		g_cond_waiters.erase(it);
-	}
-}
+static wait_poll_func_t g_cond_wait_poll_callback = nullptr;
 
 struct ThreadPrivate {
 	ThreadPrivate(thread_func_t f, void* a): func(f), arg(a), m_thread(&Run, this) {}
@@ -284,10 +253,6 @@ void Thread::Detach() {
 	m_thread->m_thread.detach();
 }
 
-void Thread::Sleep(uint32_t millis) {
-	std::this_thread::sleep_for(std::chrono::milliseconds(millis));
-}
-
 void Thread::SleepMicro(uint32_t micros) {
 #ifdef KYTY_WIN_CS
 	SleepHighResolution100ns(static_cast<uint64_t>(micros) * 10);
@@ -365,7 +330,6 @@ CondVar::~CondVar() {
 }
 
 void CondVar::Wait(Mutex* mutex) {
-	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -406,7 +370,6 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 }
 
 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -415,7 +378,6 @@ void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
 
 bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	bool ok = false;
-	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -430,7 +392,6 @@ bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	      std::cv_status::no_timeout);
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 	return ok;
 }
 
@@ -445,16 +406,13 @@ void CondVar::Signal() {
 }
 
 void CondVar::SignalAll() {
-	WakeCondVar(m_cond_var.get());
-}
-
-void CondVar::SignalThread(int thread_id) {
-	std::lock_guard lock(g_cond_waiters_mutex);
-	for (const auto& waiter: g_cond_waiters) {
-		if (waiter.first == thread_id) {
-			WakeCondVar(waiter.second);
-		}
-	}
+#ifdef KYTY_WIN_CS
+	static auto func = ResolveWakeAllConditionVariable();
+	EXIT_NOT_IMPLEMENTED(func == nullptr);
+	func(&m_cond_var->m_cv);
+#else
+	m_cond_var->m_cv.notify_all();
+#endif
 }
 
 int Thread::GetThreadIdUnique() {

src/common/threads.h

@@ -31,7 +31,6 @@ public:
 	// The id is unique and can't be reused by another thread.
 	[[nodiscard]] int GetUniqueId() const;
 
-	static void Sleep(uint32_t millis);
 	static void SleepMicro(uint32_t micros);
 	static void SleepNano(uint64_t nanos);
 	static bool IsMainThread();
@@ -77,7 +76,6 @@ public:
 	void Signal();
 	void SignalAll();
 
-	static void SignalThread(int thread_id);
 	static void SetWaitPollCallback(wait_poll_func_t callback);
 
 	KYTY_CLASS_NO_COPY(CondVar);

src/common/timer.cpp

@@ -62,23 +62,6 @@ double Timer::GetTimeS() const {
 	return (static_cast<double>(current_time - m_StartTime)) / static_cast<double>(m_Frequency);
 }
 
-// return time in ticks
-uint64_t Timer::GetTicks() const {
-	if (m_is_paused) {
-		return (m_PauseTime - m_StartTime);
-	}
-
-	uint64_t current_time = 0;
-	SysQueryPerformanceCounter(&current_time);
-
-	return (current_time - m_StartTime);
-}
-
-// return ticks frequency
-uint64_t Timer::GetFrequency() const {
-	return m_Frequency;
-}
-
 uint64_t Timer::QueryPerformanceFrequency() {
 	uint64_t ret = 0;
 	SysQueryPerformanceFrequency(&ret);

src/common/timer.h

@@ -24,12 +24,6 @@ public:
 	// return time in seconds
 	[[nodiscard]] double GetTimeS() const;
 
-	// return time in ticks
-	[[nodiscard]] uint64_t GetTicks() const;
-
-	// return ticks frequency
-	[[nodiscard]] uint64_t GetFrequency() const;
-
 	KYTY_CLASS_NO_COPY(Timer);
 
 	[[nodiscard]] static uint64_t QueryPerformanceFrequency();

src/common/virtualMemory.cpp

@@ -1,63 +0,0 @@
-#include "common/virtualMemory.h"
-
-#include "common/platform/sysVirtual.h"
-
-namespace Common {
-
-namespace VirtualMemory {
-
-void Init() {
-	SysVirtualInit();
-}
-
-uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualAlloc(address, size, mode);
-}
-
-uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
-	return SysVirtualAllocAligned(address, size, mode, alignment);
-}
-
-bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualAllocFixed(address, size, mode);
-}
-
-bool Commit(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualCommit(address, size, mode);
-}
-
-uint64_t Reserve(uint64_t address, uint64_t size) {
-	return SysVirtualReserve(address, size);
-}
-
-uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
-	return SysVirtualReserveAligned(address, size, alignment);
-}
-
-bool ReserveFixed(uint64_t address, uint64_t size) {
-	return SysVirtualReserveFixed(address, size);
-}
-
-bool Decommit(uint64_t address, uint64_t size) {
-	return SysVirtualDecommit(address, size);
-}
-
-bool Free(uint64_t address) {
-	return SysVirtualFree(address);
-}
-
-bool FreeRange(uint64_t address, uint64_t size) {
-	return SysVirtualFreeRange(address, size);
-}
-
-bool Protect(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualProtect(address, size, mode);
-}
-
-bool FlushInstructionCache(uint64_t address, uint64_t size) {
-	return SysVirtualFlushInstructionCache(address, size);
-}
-
-} // namespace VirtualMemory
-
-} // namespace Common

src/emulator.cpp

@@ -196,6 +196,10 @@ void Run(const RunOptions& options) {
 	ClearDebugTextureFolder();
 
 	PrintSystemInfo();
+	std::string title_id;
+	if (Loader::SystemContentParamSfoGetString("TITLE_ID", &title_id) && !title_id.empty()) {
+		Log::WriteToConsoleAndLog(fmt::format("Title ID: {}\n", title_id));
+	}
 
 	int ok = atexit(KytyClose);
 	EXIT_NOT_IMPLEMENTED(ok != 0);

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -32,11 +32,12 @@ private:
 
 	struct BufferCursor {
 		std::span<const uint32_t> commands;
-		uint32_t                  offset_dw           = 0;
-		uint32_t                  deferred_advance_dw = 0;
+		uint32_t                  offset_dw = 0;
 	};
 
 	std::vector<BufferCursor> m_buffer_stack;
+	std::span<const uint32_t> m_next_buffer;
+	bool                      m_chain         = false;
 	bool                      m_suspended     = false;
 	bool                      m_made_progress = false;
 };
@@ -129,7 +130,7 @@ public:
 	[[nodiscard]] bool ShouldSkipPredicatedPackets() const { return m_predicate_skip; }
 
 	Pm4ProcessResult Process(Pm4Execution& execution, std::span<const uint32_t> commands);
-	void             ProcessIndirectBuffer(std::span<const uint32_t> commands);
+	void             ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain);
 
 	void SetFlip(const FlipInfo& flip) { m_flip = flip; }
 
@@ -143,7 +144,7 @@ private:
 	                      uint32_t cache_action, uint32_t event_index, uint32_t event_write_source,
 	                      void* dst_gpu_addr, T value, uint32_t interrupt_selector,
 	                      uint32_t interrupt_context_id);
-	void ProcessPm4(Pm4Execution& execution, size_t stop_depth);
+	void ProcessPm4(Pm4Execution& execution);
 	void SuspendPm4();
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -1495,10 +1495,10 @@ KYTY_CP_OP_PARSER(CpOpBranch) {
 	     reinterpret_cast<uint64_t>(else_buffer), else_num_dw);
 
 	if (take_then) {
-		cp.ProcessIndirectBuffer({then_buffer, then_num_dw});
+		cp.ProcessIndirectBuffer({then_buffer, then_num_dw}, true);
 	} else if (mode == 2 && else_num_dw != 0) {
 		EXIT_NOT_IMPLEMENTED(else_buffer == nullptr);
-		cp.ProcessIndirectBuffer({else_buffer, else_num_dw});
+		cp.ProcessIndirectBuffer({else_buffer, else_num_dw}, true);
 	}
 
 	return payload_dw;
@@ -1959,7 +1959,7 @@ KYTY_CP_OP_PARSER(CpOpIndirectBuffer) {
 
 	GraphicsDbgDumpDcb("ci", indirect_num_dw, indirect_buffer);
 
-	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw});
+	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw}, (control & (1u << 20u)) != 0);
 
 	return 3;
 }
@@ -2017,6 +2017,14 @@ KYTY_CP_OP_PARSER(CpOpIndirectCxRegs) {
 		auto pfunc = g_hw_ctx_indirect_func[cmd_offset & (Pm4::CX_NUM - 1)];
 
 		if (pfunc == nullptr) {
+			if (raw_cmd_offset == 0x24au && value == 0u) {
+				static std::atomic_flag logged = ATOMIC_FLAG_INIT;
+				if (!logged.test_and_set(std::memory_order_relaxed)) {
+					Log::WriteToConsoleAndLog(
+					    "\t diagnostic: ignoring indirect CX {0x24a, 0}; hardware effect unresolved\n");
+				}
+				continue;
+			}
 			EXIT("unknown cx reg at %05" PRIx32 ": 0x%" PRIx32 "\n", num_dw - dw, cmd_offset);
 		}
 
@@ -3635,6 +3643,11 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC4_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
+	// Toolkit state restoration emits the compiler's GS-front allocation
+	// metadata here. Native GS resource registers already carry that allocation.
+	g_hw_sh_indirect_func[0x0ca] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
@@ -3654,12 +3667,13 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
+	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS;
+	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS; offset++) {
+		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+			cp.GetShCtx().SetHsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS,
+			                                   value);
+		};
+	}
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		auto base = cp.GetShCtx().GetVs().hs_regs.data_addr;
 		base &= 0xFFFFFF00000000FFull;
@@ -3818,6 +3832,25 @@ void GraphicsInitJmpTablesUcIndirect() {
 	for (auto& func: g_hw_uc_indirect_func) {
 		func = nullptr;
 	}
+	for (uint32_t i = 0; i < 4; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(0, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(1, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_Y, value);
+		};
+	}
+	for (uint32_t i = 0; i < 2; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(0, cmd_offset - Pm4::FSR_ALPHA_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(1, cmd_offset - Pm4::FSR_ALPHA_LEFT_Y, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_WINDOW_LEFT + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrWindow(cmd_offset - Pm4::FSR_WINDOW_LEFT, value);
+		};
+	}
 
 	g_hw_uc_indirect_func[Pm4::GE_CNTL] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeControl r;

src/graphics/guest_gpu/gpu_defs.h

@@ -26,6 +26,10 @@ enum class PrimitiveType : uint32_t {
 	kPolygon            = 21,
 };
 
+constexpr bool IsRectList(PrimitiveType type) {
+	return type == PrimitiveType::kRectList || type == PrimitiveType::kRectListLegacy;
+}
+
 enum class IndexType : uint32_t {
 	kIndex16 = 0,
 	kIndex32 = 1,

src/graphics/guest_gpu/graphicsRun.cpp

@@ -386,8 +386,12 @@ void CommandProcessor::WriteReferenceClock(uint64_t dst_address, uint32_t num_by
 	}
 	const auto value = Sync::ReadReferenceClock();
 	std::memcpy(reinterpret_cast<void*>(dst_address), &value, num_bytes);
-	LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64 " size=%u\n",
-	     dst_address, value, num_bytes);
+	static std::atomic<uint32_t> clock_log_count {0};
+	if (clock_log_count.fetch_add(1) < 64) {
+		LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64
+		     " size=%u\n",
+		     dst_address, value, num_bytes);
+	}
 }
 
 void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cache_policy,
@@ -672,20 +676,16 @@ Pm4ProcessResult CommandProcessor::Process(Pm4Execution&             execution,
 		Pm4Execution*     previous_execution;
 	} execution_scope(*this, execution);
 
-	ProcessPm4(execution, 0);
+	ProcessPm4(execution);
 	return execution.m_buffer_stack.empty() ? Pm4ProcessResult::Complete
 	                                        : Pm4ProcessResult::Blocked;
 }
 
-void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands) {
+void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain) {
 	EXIT_IF(g_current_execution == nullptr);
-	if (commands.empty()) {
-		return;
-	}
-	auto&      execution  = *g_current_execution;
-	const auto stop_depth = execution.m_buffer_stack.size();
-	execution.m_buffer_stack.push_back({commands});
-	ProcessPm4(execution, stop_depth);
+	EXIT_IF(!g_current_execution->m_next_buffer.empty());
+	g_current_execution->m_next_buffer = commands;
+	g_current_execution->m_chain       = chain;
 }
 
 void CommandProcessor::SuspendPm4() {
@@ -693,21 +693,13 @@ void CommandProcessor::SuspendPm4() {
 	g_current_execution->m_suspended = true;
 }
 
-void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
-	while (execution.m_buffer_stack.size() > stop_depth) {
+void CommandProcessor::ProcessPm4(Pm4Execution& execution) {
+	while (!execution.m_buffer_stack.empty()) {
 		if (g_gpu_state != nullptr) {
 			g_gpu_state->ProcessCommands();
 		}
-		const auto buffer_index = execution.m_buffer_stack.size() - 1;
-		auto&      cursor       = execution.m_buffer_stack[buffer_index];
+		auto& cursor = execution.m_buffer_stack.back();
 		EXIT_IF(cursor.offset_dw > cursor.commands.size());
-		if (cursor.deferred_advance_dw != 0) {
-			EXIT_IF(cursor.deferred_advance_dw > cursor.commands.size() - cursor.offset_dw);
-			cursor.offset_dw += cursor.deferred_advance_dw;
-			cursor.deferred_advance_dw = 0;
-			execution.m_made_progress  = true;
-			continue;
-		}
 		if (cursor.offset_dw == cursor.commands.size()) {
 			execution.m_buffer_stack.pop_back();
 			continue;
@@ -783,14 +775,19 @@ void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
 		    handler(*this, packet_header & ~1u, packet + 1, remaining_dw, total_dw) + 1;
 		EXIT_IF(packet_dw > remaining_dw);
 		if (execution.m_suspended) {
-			if (execution.m_buffer_stack.size() > buffer_index + 1) {
-				execution.m_buffer_stack[buffer_index].deferred_advance_dw = packet_dw;
-			}
 			return;
 		}
-		EXIT_IF(execution.m_buffer_stack.size() != buffer_index + 1);
-		execution.m_buffer_stack[buffer_index].offset_dw += packet_dw;
+		cursor.offset_dw += packet_dw;
 		execution.m_made_progress = true;
+		if (!execution.m_next_buffer.empty()) {
+			// Chains and taken branches reuse the fetcher; only calls retain a return cursor.
+			if (execution.m_chain) {
+				cursor = {execution.m_next_buffer};
+			} else {
+				execution.m_buffer_stack.push_back({execution.m_next_buffer});
+			}
+			execution.m_next_buffer = {};
+		}
 	}
 }
 
@@ -825,36 +822,55 @@ void CommandProcessor::SetNumInstances(uint32_t num_instances) {
 
 void CommandProcessor::SetPredication(uint32_t condition, uint32_t op, uint32_t wait_op,
                                       const volatile void* address, uint32_t count_in_dwords) {
-	if (wait_op != 0) {
-		BufferFlushAndWait();
-	}
-
 	(void)count_in_dwords;
+	uint64_t value = 0;
 
 	switch (op) {
-		case 0x00: {
+		case 0x00:
 			m_predicate_skip = false;
-		} break;
-		case 0x03: {
+			return;
+		case 0x01: {
 			EXIT_NOT_IMPLEMENTED(address == nullptr);
-
-			auto value = *reinterpret_cast<const volatile uint64_t*>(address);
-
-			switch (condition) {
-				case 0x00: m_predicate_skip = (value != 0); break;
-				case 0x01: m_predicate_skip = (value == 0); break;
-				default: EXIT("unknown predication condition: 0x%08" PRIx32 "\n", condition);
-			}
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1) < 128) {
-				LOGF("\t bool predication: addr=0x%016" PRIx64 ", value=0x%016" PRIx64
-				     ", condition=%" PRIu32 ", skip=%u, wait_op=%" PRIu32 "\n",
-				     reinterpret_cast<uint64_t>(address), value, condition,
-				     m_predicate_skip ? 1u : 0u, wait_op);
+			// One begin/end pair per DB; bit 63 marks each counter ready.
+			constexpr uint64_t ready_bit = 1ull << 63u;
+			const auto* results = reinterpret_cast<const volatile uint64_t*>(address);
+			for (uint32_t db = 0; db < 16u; db++) {
+				const auto begin = results[db * 2u];
+				const auto end   = results[db * 2u + 1u];
+				if ((begin & end & ready_bit) == 0) {
+					if (wait_op == 0) {
+						SuspendPm4();
+					} else {
+						m_predicate_skip = false;
+					}
+					return;
+				}
+				value += end - begin;
 			}
 		} break;
+		case 0x03:
+			if (wait_op != 0) {
+				BufferFlushAndWait();
+			}
+			EXIT_NOT_IMPLEMENTED(address == nullptr);
+			value = *reinterpret_cast<const volatile uint64_t*>(address);
+			break;
 		default: EXIT("unknown predication op: 0x%08" PRIx32 "\n", op);
 	}
+	switch (condition) {
+		case 0x00: m_predicate_skip = (value != 0); break;
+		case 0x01: m_predicate_skip = (value == 0); break;
+		default: EXIT("unknown predication condition: 0x%08" PRIx32 "\n", condition);
+	}
+	if (op == 0x03) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1) < 128) {
+			LOGF("\t bool predication: addr=0x%016" PRIx64 ", value=0x%016" PRIx64
+			     ", condition=%" PRIu32 ", skip=%u, wait_op=%" PRIu32 "\n",
+			     reinterpret_cast<uint64_t>(address), value, condition,
+			     m_predicate_skip ? 1u : 0u, wait_op);
+		}
+	}
 }
 
 void CommandProcessor::DrawIndex(DrawIndexArgs args) {
@@ -896,17 +912,6 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 	if (!indexed) {
 		DrawIndirectArgs args {};
 		std::memcpy(&args, args_addr, sizeof(args));
-		if (args.instance_count != 1u || args.start_vertex_location != 0u ||
-		    args.start_instance_location != 0u) {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1) < 64) {
-				LOGF("\t warning: partial DrawIndirect args: vertex_count=%" PRIu32
-				     ", instance_count=%" PRIu32 ", start_vertex=%" PRIu32
-				     ", start_instance=%" PRIu32 "\n",
-				     args.vertex_count_per_instance, args.instance_count,
-				     args.start_vertex_location, args.start_instance_location);
-			}
-		}
 		m_num_instances = args.instance_count;
 		DrawIndexAuto({.vertex_count   = args.vertex_count_per_instance,
 		               .instance_count = args.instance_count,
@@ -918,16 +923,6 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 
 	DrawIndexedIndirectArgs args {};
 	std::memcpy(&args, args_addr, sizeof(args));
-	if (args.base_vertex_location != 0u || args.start_instance_location != 0u) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1) < 64) {
-			LOGF("\t warning: partial DrawIndexIndirect args: index_count=%" PRIu32
-			     ", instance_count=%" PRIu32 ", start_index=%" PRIu32 ", base_vertex=%" PRIu32
-			     ", start_instance=%" PRIu32 "\n",
-			     args.index_count_per_instance, args.instance_count, args.start_index_location,
-			     args.base_vertex_location, args.start_instance_location);
-		}
-	}
 
 	uint64_t index_size = 0;
 	switch (m_index_type_and_size) {
@@ -989,17 +984,6 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 
 		if (!indexed) {
 			auto* args = reinterpret_cast<const DrawIndirectArgs*>(args_addr);
-			if (args->instance_count != 1u || args->start_vertex_location != 0u ||
-			    args->start_instance_location != 0u) {
-				static std::atomic<uint32_t> log_count {0};
-				if (log_count.fetch_add(1) < 64) {
-					LOGF("\t warning: partial DrawIndirectMulti args[%u]: vertex_count=%" PRIu32
-					     ", instance_count=%" PRIu32 ", start_vertex=%" PRIu32
-					     ", start_instance=%" PRIu32 "\n",
-					     i, args->vertex_count_per_instance, args->instance_count,
-					     args->start_vertex_location, args->start_instance_location);
-				}
-			}
 			m_num_instances = args->instance_count;
 			DrawIndexAuto({.vertex_count   = args->vertex_count_per_instance,
 			               .instance_count = args->instance_count,
@@ -1010,17 +994,6 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 		}
 
 		auto* args = reinterpret_cast<const DrawIndexedIndirectArgs*>(args_addr);
-		if (args->base_vertex_location != 0u || args->start_instance_location != 0u) {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1) < 64) {
-				LOGF("\t warning: partial DrawIndexIndirectMulti args[%u]: index_count=%" PRIu32
-				     ", instance_count=%" PRIu32 ", start_index=%" PRIu32 ", base_vertex=%" PRIu32
-				     ", start_instance=%" PRIu32 "\n",
-				     i, args->index_count_per_instance, args->instance_count,
-				     args->start_index_location, args->base_vertex_location,
-				     args->start_instance_location);
-			}
-		}
 
 		uint64_t index_size = 0;
 		switch (m_index_type_and_size) {
@@ -1090,17 +1063,6 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		// local_x        = std::max(cs.num_thread_x, 1u);
 		// local_y        = std::max(cs.num_thread_y, 1u);
 		// local_z        = std::max(cs.num_thread_z, 1u);
-		if (cs.wave_size == 64u) {
-			static std::atomic_bool logged_wave64_shader {false};
-			if (!logged_wave64_shader.exchange(true, std::memory_order_relaxed)) {
-				LOGF("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
-				     cs.data_addr);
-				std::printf("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
-				            cs.data_addr);
-				std::fflush(stdout);
-			}
-		}
-
 		m_renderer.GetRenderExecutor().DispatchDirect(m_submit_id, CurrentBuffer(), thread_group_x,
 		                                              thread_group_y, thread_group_z, mode);
 	}
@@ -1292,6 +1254,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x14:
 							case 0x28:
+							case 0x2f:
 								if (event_index == 0x00) {
 									write64(false);
 									return;
@@ -1299,7 +1262,6 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x2b:
 							case 0x2d:
-							case 0x2f:
 							case 0x30:
 								if (event_index == 0x00 && !with_interrupt) {
 									write64(false);

src/graphics/guest_gpu/hardwareContext.h

@@ -498,7 +498,8 @@ struct LsStageRegisters {
 };
 
 struct HsStageRegisters {
-	uint64_t          data_addr = 0;
+	uint64_t          data_addr      = 0;
+	uint64_t          user_data_addr = 0;
 	HsShaderResource1 rsrc1;
 	HsShaderResource2 rsrc2;
 };
@@ -935,6 +936,12 @@ private:
 	ShaderRegisters m_sh_regs;
 };
 
+struct FsrView {
+	uint32_t control_points[2][4] {};
+	uint32_t alphas[2][2] {};
+	uint32_t window[2] {};
+};
+
 class UserConfig {
 public:
 	UserConfig()  = default;
@@ -964,8 +971,17 @@ public:
 	void SetGdsOaCntl(uint32_t value) { m_gds_oa.cntl = value; }
 	void SetGdsOaCounter(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].counter = value; }
 	void SetGdsOaAddress(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].address = value; }
+	void SetFsrControlPoint(uint32_t axis, uint32_t index, uint32_t value) {
+		m_fsr_view.control_points[axis][index] = value;
+	}
+	void SetFsrAlpha(uint32_t axis, uint32_t index, uint32_t value) {
+		m_fsr_view.alphas[axis][index] = value;
+	}
+	void SetFsrWindow(uint32_t index, uint32_t value) { m_fsr_view.window[index] = value; }
+	[[nodiscard]] const FsrView& GetFsrView() const { return m_fsr_view; }
 
 private:
+	FsrView m_fsr_view;
 	Prospero::PrimitiveType m_prim_type               = Prospero::PrimitiveType::kNone;
 	uint32_t                m_index_offset            = 0;
 	uint32_t                m_object_id               = 0;
@@ -988,14 +1004,14 @@ public:
 	void SetEsShaderBase(uint64_t addr) { m_vs.es_regs.data_addr = addr; }
 	void SetLsShaderBase(uint64_t addr) { m_vs.ls_regs.data_addr = addr; }
 	void SetHsShaderBase(uint64_t addr) { m_vs.hs_regs.data_addr = addr; }
+	void SetHsUserDataAddress(uint32_t word, uint32_t value) {
+		SetUserDataAddressWord(m_vs.hs_regs.user_data_addr, word, value);
+	}
 	void SetHsShaderResource1(const HsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
 	void SetHsShaderResource2(const HsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
 	void SetGsShaderBase(uint64_t addr) { m_vs.gs_regs.data_addr = addr; }
 	void SetGsUserDataAddress(uint32_t word, uint32_t value) {
-		const auto shift   = word * 32u;
-		auto&      address = m_vs.gs_regs.user_data_addr;
-		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
-		          (static_cast<uint64_t>(value) << shift);
+		SetUserDataAddressWord(m_vs.gs_regs.user_data_addr, word, value);
 	}
 	void SetGsShaderResource1(const GsShaderResource1& rsrc1) { m_vs.gs_regs.rsrc1 = rsrc1; }
 	void SetGsShaderResource2(const GsShaderResource2& rsrc2) { m_vs.gs_regs.rsrc2 = rsrc2; }
@@ -1037,6 +1053,12 @@ public:
 	[[nodiscard]] const ComputeShaderInfo& GetCs() const { return m_cs; }
 
 private:
+	static void SetUserDataAddressWord(uint64_t& address, uint32_t word, uint32_t value) {
+		const auto shift = word * 32u;
+		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
+		          (static_cast<uint64_t>(value) << shift);
+	}
+
 	VertexShaderInfo  m_vs;
 	PixelShaderInfo   m_ps;
 	ComputeShaderInfo m_cs;

src/graphics/guest_gpu/pm4.h

@@ -987,6 +987,12 @@ constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR2_MASK  = 0x1;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_SHIFT = 2;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_MASK  = 0x1;
 
+constexpr uint32_t FSR_WINDOW_LEFT           = 0x2BC;
+constexpr uint32_t FSR_CONTROL_POINTS_LEFT_X = 0x2C4;
+constexpr uint32_t FSR_CONTROL_POINTS_LEFT_Y = 0x2CC;
+constexpr uint32_t FSR_ALPHA_LEFT_X          = 0x2E4;
+constexpr uint32_t FSR_ALPHA_LEFT_Y          = 0x2E8;
+
 constexpr uint32_t TA_CS_BC_BASE_ADDR       = 0x380;
 constexpr uint32_t TA_CS_BC_BASE_ADDR_HI    = 0x381;
 constexpr uint32_t TEXTURE_GRADIENT_FACTORS = 0x383;

src/graphics/guest_gpu/tile.cpp

@@ -4,15 +4,12 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
-#include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/gpu_format.h"
 
 #include <algorithm>
 #include <array>
 #include <bit>
-#include <fmt/format.h>
-#include <vector>
 
 namespace Libs::Graphics {
 
@@ -1322,13 +1319,8 @@ void TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t
 		return;
 	}
 	if (total_size != nullptr && total_size->size == 0) {
-		std::vector<std::string> list;
-		list.push_back(fmt::format("format = {}", static_cast<uint32_t>(format)));
-		list.push_back(fmt::format("width  = {}", width));
-		list.push_back(fmt::format("height = {}", height));
-		list.push_back(fmt::format("levels = {}", levels));
-		list.push_back(fmt::format("tile   = {}", static_cast<uint32_t>(tile)));
-		EXIT("unknown format:\n%s\n", Common::Concat(list, '\n').c_str());
+		EXIT("unknown format:\nformat = %u\nwidth  = %u\nheight = %u\nlevels = %u\ntile   = %u\n",
+		     static_cast<uint32_t>(format), width, height, levels, static_cast<uint32_t>(tile));
 	}
 }
 

src/graphics/guest_gpu/tile.h

@@ -86,6 +86,7 @@ struct TileMipLayout {
 	uint32_t padded_height = 0;
 	uint32_t tail_x        = 0;
 	uint32_t tail_y        = 0;
+	bool     operator==(const TileMipLayout&) const = default;
 };
 
 struct TileSurfaceLayout {

src/graphics/host_gpu/regionManager.h

@@ -16,6 +16,7 @@
 #include <windows.h>
 #undef min
 #undef max
+#undef MemoryBarrier
 #elif defined(__APPLE__)
 #include <pthread.h>
 #elif defined(__linux__)

src/graphics/host_gpu/renderer/cache/faultManager.cpp

@@ -44,12 +44,7 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	                                                &m_fault_process_desc_layout),
 	    "create fault-buffer descriptor layout");
 
-	vk::ShaderModuleCreateInfo module_info {};
-	module_info.codeSize = std::size(FAULT_BUFFER_PROCESS_SPV) * sizeof(uint32_t);
-	module_info.pCode    = FAULT_BUFFER_PROCESS_SPV;
-	vk::ShaderModule module = nullptr;
-	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-	                     "create fault-buffer shader module");
+	const auto module = CompileSPV(FAULT_BUFFER_PROCESS_SPV, m_graphics.device);
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
 	pipeline_layout_info.setLayoutCount = 1;

src/graphics/host_gpu/renderer/cache/textureCache.cpp

@@ -33,22 +33,24 @@ namespace {
 
 constexpr uint64_t NumFramesBeforeRemoval = 32;
 
-[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, vk::Format format,
-                                  uint32_t fill, vk::ClearColorValue& clear) {
-	const auto code = static_cast<uint8_t>(fill);
-	if (fill != static_cast<uint32_t>(code) * 0x01010101u) {
-		return false;
+[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, uint8_t code,
+                                  vk::ClearColorValue& clear) {
+	switch (code) {
+		case 0x00:
+		case 0x20:
+		case 0x40:
+		case 0x80:
+		case 0xc0: break;
+		default: return false;
 	}
 	const auto& metadata = desc.info.metadata;
+	const auto  format   = desc.view_info.format;
 	if (code == 0x20) {
 		// Clear-to-register is a color-buffer operation; the texture pipe cannot decode it.
 		return desc.type == TextureCache::BindingType::RenderTarget &&
 		       metadata.dcc_clear_register_valid &&
 		       DecodePackedColorClear(format, metadata.dcc_clear_word, clear);
 	}
-	if (code != 0x00 && code != 0x40 && code != 0x80 && code != 0xc0) {
-		return false;
-	}
 	clear = {};
 	if (code == 0x00) {
 		return true;
@@ -206,6 +208,10 @@ bool TextureCache::SameBacking(const ImageInfo& cached, const ImageInfo& request
 	if (cached.extent != requested.extent) {
 		return false;
 	}
+	if (cached.resources.levels < requested.resources.levels ||
+	    cached.resources.layers < requested.resources.layers) {
+		return false;
+	}
 	if (cached.samples != requested.samples) {
 		return false;
 	}
@@ -323,10 +329,8 @@ void TextureCache::DeleteImage(ImageId id) {
 	if (image->info.HasMetadata()) {
 		const auto metadata = m_surface_metas.find(image->info.metadata.range.address);
 		if (metadata != m_surface_metas.end() &&
-		    ((image->info.metadata.kind == ImageMetadataKind::Dcc &&
-		      metadata->second.type == MetaDataInfo::Type::Dcc) ||
-		     (image->info.metadata.kind == ImageMetadataKind::Htile &&
-		      metadata->second.type == MetaDataInfo::Type::HTile))) {
+		    image->info.metadata.kind == ImageMetadataKind::Htile &&
+		    metadata->second.type == MetaDataInfo::Type::HTile) {
 			// A later binding may have reused this address for another metadata type.
 			m_surface_metas.erase(metadata);
 		}
@@ -810,17 +814,10 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 	const bool safe_to_delete =
 	    current_tick - std::min(current_tick, cached.tick_accessed_last) > NumFramesBeforeRemoval;
 
-	if (requested.data.address == cached.info.data.address) {
-		const uint32_t requested_block = requested.bytes_per_block * requested.samples;
-		const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
-		if (requested.BlockExtent() != cached.info.BlockExtent() ||
-		    requested_block != cached_block) {
-			if (safe_to_delete) {
-				FreeImage(cached_id);
-			}
-			return {merged_id};
-		}
-
+	const uint32_t requested_block = requested.bytes_per_block * requested.samples;
+	const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
+	if (requested.data.address == cached.info.data.address &&
+	    requested.BlockExtent() == cached.info.BlockExtent() && requested_block == cached_block) {
 		if (const auto depth_id = ResolveDepthOverlap(requested, binding, cached_id)) {
 			return {depth_id};
 		}
@@ -850,6 +847,16 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		    ImageViewOps::FormatsCompatible(cached.info.pixel_format, requested.pixel_format)) {
 			return {ExpandImage(requested, cached_id)};
 		}
+		// PS5 mip tails can expose more levels without increasing the guest allocation.
+		if (requested.pixel_format == cached.info.pixel_format &&
+		    requested.type == cached.info.type && requested.resources > cached.info.resources &&
+		    (requested.data.size > cached.info.data.size ||
+		     (requested.data.size == cached.info.data.size &&
+		      requested.extent == cached.info.extent &&
+		      cached.info.resources.levels > 1 &&
+		      requested.resources.layers == cached.info.resources.layers))) {
+			return {ExpandImage(requested, cached_id)};
+		}
 		if (requested.pixel_format != cached.info.pixel_format ||
 		    requested.data.size <= cached.info.data.size) {
 			const auto result_id = merged_id ? merged_id : cached_id;
@@ -859,9 +866,6 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 			            ? result_id
 			            : ImageId {}};
 		}
-		if (requested.type == cached.info.type && requested.resources > cached.info.resources) {
-			return {ExpandImage(requested, cached_id)};
-		}
 		EXIT("TextureCache: unresolvable equal-address image overlap, address=0x%016" PRIx64
 		     " requested=%ux%u "
 		     "cached=%ux%u requested_size=0x%016" PRIx64 " cached_size=0x%016" PRIx64
@@ -873,38 +877,27 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		     static_cast<uint32_t>(cached.info.tile_mode));
 	}
 
-	if (requested.data.address > cached.info.data.address) {
-		const int32_t mip = requested.MipOf(cached.info);
-		if (mip >= 0) {
-			const int32_t layer = requested.SliceOf(cached.info, mip);
-			if (layer >= 0) {
-				return {cached_id, mip, layer};
-			}
-		}
-		if (safe_to_delete) {
-			FreeImage(cached_id);
-		}
-		return {};
+	const int32_t requested_mip = requested.MipOf(cached.info);
+	if (requested_mip >= 0) {
+		const int32_t layer = requested.SliceOf(cached.info, requested_mip);
+		return {cached_id, requested_mip, layer};
 	}
 
 	const int32_t mip = cached.info.MipOf(requested);
 	if (mip >= 0) {
 		const int32_t layer = cached.info.SliceOf(requested, mip);
-		if (layer >= 0) {
-			if (cached.binding.is_target) {
-				cached.binding.needs_rebind = true;
-				if (merged_id) {
-					m_slot_images[merged_id].binding.is_target = true;
-				}
-				FreeImage(cached_id);
-				return {merged_id};
-			}
-			if (merged_id) {
-				CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
-				             static_cast<uint32_t>(layer));
-				FreeImage(cached_id);
-			}
+		if (!merged_id) {
+			return {ExpandImage(requested, cached_id)};
 		}
+		cached.binding.needs_rebind |= cached.binding.is_bound || cached.binding.is_target;
+		m_slot_images[merged_id].binding.is_target |= cached.binding.is_target;
+		CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+		FreeImage(cached_id);
+		return {merged_id};
+	}
+	if (requested.data.address >= cached.info.data.address && safe_to_delete) {
+		FreeImage(cached_id);
 	}
 	return {merged_id};
 }
@@ -919,7 +912,14 @@ ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId source_id) {
 		source.binding.needs_rebind = true;
 	}
 	InitializeImage(expanded_id);
-	CopyImage(expanded_id, source_id);
+	const int32_t mip = source.info.MipOf(info);
+	const int32_t layer = source.info.SliceOf(info, mip);
+	if (layer >= 0) {
+		CopyImageMip(expanded_id, source_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+	} else {
+		CopyImage(expanded_id, source_id);
+	}
 	FreeImage(source_id);
 	return expanded_id;
 }
@@ -1135,53 +1135,69 @@ void TextureCache::InitializeImage(ImageId id) {
 	}
 }
 
-void TextureCache::PrepareDccClear(ImageId id, const ImageDesc& desc) {
+void TextureCache::MaterializeDccClear(ImageId id, const ImageDesc& desc,
+                                       uint32_t metadata_base_layer) {
 	if (desc.info.metadata.kind != ImageMetadataKind::Dcc) {
 		return;
 	}
-	auto& image            = m_slot_images[id];
-	image.info.metadata    = desc.info.metadata;
-	auto [entry, inserted] = m_surface_metas.try_emplace(
-	    desc.info.metadata.range.address, MetaDataInfo {.type = MetaDataInfo::Type::Dcc});
-	auto& metadata = entry->second;
-	if (!inserted && metadata.type == MetaDataInfo::Type::PendingDcc) {
-		metadata.type = MetaDataInfo::Type::Dcc;
-	} else if (metadata.type != MetaDataInfo::Type::Dcc) {
-		EXIT("TextureCache: image reuses non-DCC metadata\n");
+	const auto range = desc.info.metadata.range;
+	{
+		std::scoped_lock lock {m_lock};
+		auto& image         = m_slot_images[id];
+		image.info.metadata = desc.info.metadata;
+		// A native DCC allocation must not retain a reused HTile/CMask/FMask interpretation.
+		m_surface_metas.erase(range.address);
+		if (range.size == 0 || desc.info.resources.levels != 1 || image.info.resources.levels != 1) {
+			return;
+		}
 	}
-	if (metadata.clear_mask == 0 || image.info.resources.levels != 1 ||
-	    desc.info.metadata.range.size == 0 || metadata.fill_size < desc.info.metadata.range.size) {
-		return;
-	}
-	vk::ClearValue clear {};
-	if (!DecodeDccClear(desc, image.backing.format, metadata.fill_value, clear.color)) {
-		return;
+	const auto layers = desc.info.TransferLayers();
+	// These one-mip surfaces use complete 4 KiB DCC metadata blocks.
+	constexpr uint64_t MetadataBlockSize = 0x1000;
+	if (!range.Valid() || range.address % MetadataBlockSize != 0 || layers == 0 ||
+	    range.size % layers != 0 || (range.size / layers) % MetadataBlockSize != 0) {
+		EXIT("TextureCache: DCC slices must contain aligned 4 KiB blocks\n");
 	}
 	const auto& view           = desc.view_info;
-	const bool  volume_texture = image.info.IsVolume() && view.type == vk::ImageViewType::e3D;
-	const auto  first          = volume_texture ? 0u : view.base_layer;
-	const auto  count = volume_texture ? std::max(image.info.extent.depth >> view.base_level, 1u)
-	                                   : view.layer_count;
-	if (first >= 32 || count > 32 - first) {
-		return;
+	const bool  volume_texture = desc.info.IsVolume() && view.type == vk::ImageViewType::e3D;
+	const auto  first          = volume_texture ? 0u : metadata_base_layer;
+	const auto  image_first    = volume_texture ? 0u : view.base_layer;
+	const auto  count          = volume_texture ? desc.info.extent.depth : view.layer_count;
+	if (first >= layers || count > layers - first) {
+		EXIT("TextureCache: DCC view exceeds its native metadata slices\n");
 	}
-	// The metadata fill covers the complete allocation. Consume each layer only after its
-	// native image contents exist; already materialized layers may have been rendered since.
-	for (uint32_t layer = first; layer < first + count;) {
-		if ((metadata.clear_mask & (1u << layer)) == 0) {
-			layer++;
+	// Finish native metadata writes before reading backing bytes. This can submit the scheduler,
+	// so discovery runs before final draw uploads and never holds the texture lock across it.
+	if (m_buffer_cache.IsRegionGpuModified(range.address, range.size)) {
+		m_buffer_cache.ReadMemory(range.address, range.size, false);
+	}
+	const auto slice_size = range.size / layers;
+	for (uint32_t slice = 0; slice < count; slice++) {
+		const auto address = range.address + slice_size * (first + slice);
+		uint8_t code = 0;
+		if (!LibKernel::Memory::TryReadBacking(address, &code, sizeof(code))) {
+			EXIT("TextureCache: failed to read DCC metadata backing\n");
+		}
+		vk::ClearValue clear {};
+		if (!DecodeDccClear(desc, code, clear.color)) {
 			continue;
 		}
-		const auto start = layer;
-		uint32_t   mask  = 0;
-		do {
-			mask |= 1u << layer++;
-		} while (layer < first + count && (metadata.clear_mask & (1u << layer)) != 0);
-		ClearImage(m_scheduler.Current(), id,
-		           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count, start,
-		            layer - start},
-		           clear);
-		metadata.clear_mask &= ~mask;
+		std::vector<uint8_t> bytes(slice_size);
+		if (!LibKernel::Memory::TryReadBacking(address, bytes.data(), bytes.size())) {
+			EXIT("TextureCache: failed to read DCC metadata slice\n");
+		}
+		if (!std::all_of(bytes.begin(), bytes.end(), [code](uint8_t byte) { return byte == code; })) {
+			continue;
+		}
+		{
+			std::scoped_lock lock {m_lock};
+			ClearImage(m_scheduler.Current(), id, view.format,
+			           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count,
+			            image_first + slice, 1}, clear);
+		}
+		// Native expanded keys own consumption. Existing buffer tracking publishes this CPU
+		// write to future GPU readers; FillBuffer can fault and must run outside the texture lock.
+		m_buffer_cache.FillBuffer(address, slice_size, UINT32_MAX, false);
 	}
 }
 
@@ -1221,7 +1237,8 @@ void TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
 	ImageId association {};
 	for (const auto id: FindImagesInRegion(stencil.address, stencil.size, false)) {
 		const auto owner = m_slot_images.try_get(id);
-		if (owner != nullptr && owner->info.data.address == stencil.address) {
+		if (owner != nullptr && owner->info.data == stencil &&
+		    owner->info.extent == depth.info.extent) {
 			association = id;
 		}
 	}
@@ -1246,6 +1263,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		std::scoped_lock lock {m_lock};
 		return GetNullImage(desc);
 	}
+	const auto metadata_base_layer = desc.view_info.base_layer;
 
 	ImageId result {};
 	{
@@ -1313,6 +1331,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		image.tick_accessed_last = m_scheduler.CurrentTick();
 		TouchImage(image);
 	}
+	MaterializeDccClear(result, desc, metadata_base_layer);
 	return result;
 }
 
@@ -1376,7 +1395,6 @@ vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
 		image.MarkGpuModified();
 	}
 	if (!image.info.data.Empty()) {
-		PrepareDccClear(id, desc);
 		RefreshImage(id);
 	}
 	switch (desc.type) {
@@ -1406,7 +1424,6 @@ vk::ImageView TextureCache::FindRenderTarget(ImageId id, const ImageDesc& desc)
 	TouchImage(image);
 	image.MarkGpuModified();
 	image.usage.render_target = true;
-	PrepareDccClear(id, desc);
 	RefreshImage(id);
 	CommitGpuWrite(image);
 	TrackImageDownload(id, image);
@@ -1430,16 +1447,9 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	RefreshImage(id);
 	if (desc.info.HasMetadata()) {
 		image.info.metadata = desc.info.metadata;
-		auto [metadata, inserted] =
-		    m_surface_metas.try_emplace(desc.info.metadata.range.address,
-		                                MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
-		                                              .clear_mask = image.info.htile_clear_mask});
-		if (!inserted && metadata->second.type != MetaDataInfo::Type::HTile) {
-			// PS5 allocations can reuse DCC storage as HTile while the old color image is cached.
-			// The depth binding defines the new type; incompatible fill state cannot carry over.
-			metadata->second = {.type       = MetaDataInfo::Type::HTile,
-			                    .clear_mask = image.info.htile_clear_mask};
-		}
+		m_surface_metas.emplace(desc.info.metadata.range.address,
+		                        MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
+		                                      .clear_mask = image.info.htile_clear_mask});
 	}
 	CommitGpuWrite(image);
 	if (desc.info.HasStencil()) {
@@ -1527,12 +1537,12 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 		}
 		clear.depthStencil.stencil = stencil_clear;
 	}
-	ClearImage(command, selected,
+	ClearImage(command, selected, image.backing.format,
 	           {aspect, 0, image.info.resources.levels, 0, image.info.TransferLayers()}, clear);
 	return true;
 }
 
-void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
+void TextureCache::ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
                               const vk::ImageSubresourceRange& range, const vk::ClearValue& clear) {
 	auto& image = m_slot_images[id];
 	const auto aspects = image.info.IsDepth() ? ImageViewOps::DepthAspectMask(image.backing.format)
@@ -1557,11 +1567,12 @@ void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
 		}
 	}
 	command.EndRendering();
-	if (image.info.IsVolume() && !full_image) {
+	// Transfer clears use the backing format; aliased clears must encode through their view.
+	if (format != image.backing.format || (image.info.IsVolume() && !full_image)) {
 		EXIT_NOT_IMPLEMENTED(range.aspectMask != vk::ImageAspectFlagBits::eColor ||
 		                     range.levelCount != 1);
 		ImageViewInfo view {};
-		view.format = image.backing.format;
+		view.format = format;
 		view.type   = range.layerCount == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
 		view.base_level  = range.baseMipLevel;
 		view.base_layer  = range.baseArrayLayer;
@@ -1872,72 +1883,32 @@ void TextureCache::InvalidateCpuAliases(uint64_t address, uint64_t size) {
 bool TextureCache::IsMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	return found != m_surface_metas.end() && found->second.type != MetaDataInfo::Type::PendingDcc;
+	return found != m_surface_metas.end();
 }
 
-bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice, uint32_t* fill_value) {
+bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
-	if (fill_value != nullptr) {
-		*fill_value = found->second.fill_value;
-	}
 	return (found->second.clear_mask & (1u << slice)) != 0;
 }
 
 bool TextureCache::ClearMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		// Preserve the broad metadata-clear operation for CMask/FMask/HTile. DCC requires a
-		// validated fill value, so an arbitrary compute write must not clear it.
+	if (found == m_surface_metas.end()) {
 		return false;
 	}
 	found->second.clear_mask = UINT32_MAX;
 	return true;
 }
 
-void TextureCache::TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value) {
-	if (!GuestRange {address, size}.Valid()) {
-		EXIT("TextureCache: invalid DCC fill range\n");
-	}
-	// DCC fills use a repeated byte code. Require all four bytes of the detected dword to agree,
-	// and mark only recognized deferred-clear encodings as logically clear.
-	const auto dcc_clear_mask = [fill_value] {
-		const auto code = static_cast<uint8_t>(fill_value);
-		if (fill_value != static_cast<uint32_t>(code) * 0x01010101u) {
-			return 0u;
-		}
-		switch (code) {
-			case 0x00:
-			case 0x20:
-			case 0x40:
-			case 0x80:
-			case 0xc0: return UINT32_MAX;
-			default: return 0u;
-		}
-	}();
-	std::scoped_lock lock {m_lock};
-	// The guest dispatch still writes metadata. An unknown address remains PendingDcc until an
-	// image descriptor confirms its role; never reinterpret CMask/FMask/HTile as DCC.
-	const auto found = m_surface_metas.try_emplace(address).first;
-	if (found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		found->second.clear_mask = dcc_clear_mask;
-		found->second.fill_value = fill_value;
-		found->second.fill_size  = size;
-	}
-}
-
 bool TextureCache::TouchMeta(uint64_t address, uint32_t slice, bool is_clear) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
 	if (is_clear) {
@@ -1953,10 +1924,13 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 		EXIT("TextureCache: invalid unmap range\n");
 	}
 	std::scoped_lock lock {m_lock};
-	const auto       end = address + size;
-	for (auto metadata = m_surface_metas.lower_bound(address);
-	     metadata != m_surface_metas.end() && metadata->first < end;) {
-		metadata = m_surface_metas.erase(metadata);
+	for (auto metadata = m_surface_metas.begin(); metadata != m_surface_metas.end();) {
+		const auto base = metadata->first;
+		if (base >= address && base < address + size) {
+			metadata = m_surface_metas.erase(metadata);
+		} else {
+			++metadata;
+		}
 	}
 	auto images = FindImagesInRegion(address, size, false);
 	for (const auto id: images) {

src/graphics/host_gpu/renderer/cache/textureCache.h

@@ -64,11 +64,8 @@ public:
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t address, uint64_t size);
 
 	[[nodiscard]] bool IsMeta(uint64_t address);
-	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice,
-	                                 uint32_t* fill_value = nullptr);
+	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice);
 	[[nodiscard]] bool ClearMeta(uint64_t address);
-	// Record deferred DCC state while the original guest dispatch writes the metadata.
-	void               TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
 	void UnmapMemory(uint64_t address, uint64_t size);
@@ -81,17 +78,10 @@ private:
 	struct ImageDownload;
 
 	struct MetaDataInfo {
-		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
-		// PendingDcc retains that exact fill until an image binding classifies the address,
-		// without exposing an unconfirmed buffer address to the normal metadata heuristics.
-		// Keep all surface metadata in one entry so CMask/FMask can be
-		// registered beside HTile and DCC without introducing parallel tracking paths.
-		enum class Type : uint8_t { PendingDcc, CMask, FMask, HTile, Dcc };
+		enum class Type : uint8_t { CMask, FMask, HTile };
 
-		Type     type       = Type::PendingDcc;
-		uint32_t clear_mask = 0;
-		uint32_t fill_value = 0xffffffffu;
-		uint64_t fill_size  = 0;
+		Type     type;
+		uint32_t clear_mask = UINT32_MAX;
 	};
 
 	struct OverlapResult {
@@ -149,7 +139,8 @@ private:
 	                                                ImageId cached);
 	[[nodiscard]] ImageId       ExpandImage(const ImageInfo& info, ImageId source);
 	void                        RefreshImage(ImageId id);
-	void                        PrepareDccClear(ImageId id, const ImageDesc& desc);
+	void                        MaterializeDccClear(ImageId id, const ImageDesc& desc,
+	                                                uint32_t metadata_base_layer);
 	void                        InitializeImage(ImageId id);
 	[[nodiscard]] TextureTransfer
 	BuildTextureTransfer(const Image& image, BindingType binding, TransferDirection direction) const;
@@ -160,8 +151,8 @@ private:
 	void DownloadDepth(Image& image, Buffer& destination, uint64_t destination_offset);
 	void CommitGpuWrite(Image& image);
 	// Caller holds m_lock. Volume layer ranges select depth slices.
-	void ClearImage(CommandBuffer& command, ImageId id, const vk::ImageSubresourceRange& range,
-	                const vk::ClearValue& clear);
+	void ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
+	                const vk::ImageSubresourceRange& range, const vk::ClearValue& clear);
 	void PrepareImageCopy(Image& image);
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);

src/graphics/host_gpu/renderer/debug.cpp

@@ -209,17 +209,16 @@ static void RtCheck(const HW::RenderTarget& rt) {
 		}
 		if (rt.info.fmask_compression_enable) {
 			EXIT_NOT_IMPLEMENTED(rt.attrib.num_samples == 0 && rt.attrib.num_fragments == 0);
+			// Native MSAA stores expanded samples, independent of FMASK metadata compression.
 			static bool logged = false;
 			if (!logged) {
-				LOGF("RenderTarget: using native Vulkan MSAA without guest FMASK metadata, "
+				LOGF("RenderTarget: using expanded native Vulkan MSAA samples, "
 				     "fmask=0x%016" PRIx64 "\n",
 				     rt.fmask.addr);
 				logged = true;
 			}
 		}
 
-		EXIT_NOT_IMPLEMENTED(rt.info.fmask_data_compression_disable != false);
-
 		if (rt.info.cmask_fast_clear_enable || rt.info.dcc_compression_enable) {
 			static bool logged = false;
 			if (!logged) {

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -19,6 +19,7 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cmath>
 #include <cstdarg>
@@ -45,8 +46,12 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
-		case Prospero::StencilOp::kReplaceTest:
-		case Prospero::StencilOp::kReplaceOp: return vk::StencilOp::eReplace;
+		case Prospero::StencilOp::kReplaceTest: return vk::StencilOp::eReplace;
+		case Prospero::StencilOp::kReplaceOp:
+			if ((op_value & write_mask) == 0) {
+				return vk::StencilOp::eZero;
+			}
+			return vk::StencilOp::eReplace;
 		case Prospero::StencilOp::kAddClamp: return vk::StencilOp::eIncrementAndClamp;
 		case Prospero::StencilOp::kSubClamp: return vk::StencilOp::eDecrementAndClamp;
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
@@ -66,9 +71,38 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	}
 }
 
-static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
-	constexpr auto replace_op = static_cast<uint8_t>(Prospero::StencilOp::kReplaceOp);
-	return fail == replace_op || pass == replace_op || depth_fail == replace_op;
+static vk::StencilOpState ConvertStencilState(
+    uint8_t compare, const std::array<uint8_t, 3>& operations, uint8_t op_value,
+    const vk::StencilOpState& state) {
+	const auto test_value = state.reference;
+	auto reference       = test_value;
+	auto required_bits   = state.compareMask;
+	if (compare == static_cast<uint8_t>(vk::CompareOp::eAlways) ||
+	    compare == static_cast<uint8_t>(vk::CompareOp::eNever)) {
+		required_bits = 0;
+	}
+	std::array<vk::StencilOp, 3> converted {};
+	for (size_t i = 0; i < operations.size(); i++) {
+		converted[i] = ConvertStencilOp(operations[i], state.writeMask, op_value);
+		if (converted[i] != vk::StencilOp::eReplace) {
+			continue;
+		}
+		auto replacement = test_value;
+		if (static_cast<Prospero::StencilOp>(operations[i]) == Prospero::StencilOp::kReplaceOp) {
+			replacement = op_value;
+		}
+		if (((reference ^ replacement) & required_bits & state.writeMask) != 0) {
+			DepthFatal("unsupported stencil replacement: compare=%u, compare mask=0x%02" PRIx32
+			           ", write mask=0x%02" PRIx32 ", operation value=0x%02" PRIx8
+			           ", test value=0x%02" PRIx32,
+			           compare, state.compareMask, state.writeMask, op_value, test_value);
+		}
+		// Vulkan shares one reference between comparison and every replacement on this face.
+		reference = (reference & ~state.writeMask) | (replacement & state.writeMask);
+		required_bits |= state.writeMask;
+	}
+	return {converted[0], converted[1], converted[2], static_cast<vk::CompareOp>(compare),
+	        state.compareMask, state.writeMask, reference};
 }
 
 [[nodiscard]] static vk::Format ResolveHostDepthAttachmentFormat(const CommandBuffer&     buffer,
@@ -125,8 +159,8 @@ static TextureCache::ImageDesc MakeDepthTargetDesc(const CommandBuffer& buffer,
 			DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
 			           z.depth_view.slice_max);
 	}
-	if (z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
-	    z.z_info.partially_resident ||
+	// EXPCLEAR permits an HTile acceleration state; the host attachment is already expanded.
+	if (z.z_info.partially_resident ||
 	    z.stencil_info.partially_resident || z.z_info.max_mip_level != 0 ||
 	    z.depth_view.current_mip_level != 0 || unsupported_shading_rate_encoding ||
 	    depth_address == 0 || (depth_address & 0xffffu) != 0) {
@@ -296,31 +330,20 @@ void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepth
 		const uint8_t back_write_mask  = stencil_ops_disabled ? 0 : sm.stencil_writemask_bf;
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
-		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
-		    (front_write_mask != 0 &&
-		     UsesStencilOpValue(sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail) &&
-		     sm.stencil_opval != sm.stencil_testval) ||
-		    (dc.backface_enable && back_write_mask != 0 &&
-		     UsesStencilOpValue(sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf) &&
-		     sm.stencil_opval_bf != sm.stencil_testval_bf)) {
-			DepthFatal("unsupported stencil compare or replacement state");
+		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways))) {
+			DepthFatal("unsupported stencil compare state");
 		}
-		r.stencil_static_front = {
-		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
-		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
-		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
-		    static_cast<vk::CompareOp>(dc.stencilfunc)};
-		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
+		r.stencil_front = ConvertStencilState(
+		    dc.stencilfunc, {sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail},
+		    sm.stencil_opval, {.compareMask = sm.stencil_mask, .writeMask = front_write_mask,
+		                       .reference = sm.stencil_testval});
 		if (dc.backface_enable) {
-			r.stencil_static_back = {
-			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
-			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
-			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
-			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
-			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
+			r.stencil_back = ConvertStencilState(
+			    dc.stencilfunc_bf, {sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf},
+			    sm.stencil_opval_bf, {.compareMask = sm.stencil_mask_bf, .writeMask = back_write_mask,
+			                          .reference = sm.stencil_testval_bf});
 		} else {
-			r.stencil_static_back  = r.stencil_static_front;
-			r.stencil_dynamic_back = r.stencil_dynamic_front;
+			r.stencil_back = r.stencil_front;
 		}
 	}
 	auto& cache = m_context.GetTextureCache();
@@ -406,14 +429,13 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		return writes;
 	}
 
-	const auto face_writes = [&](const PipelineStencilStaticState&  state,
-	                             const PipelineStencilDynamicState& dynamic) {
-		if (dynamic.writeMask == 0) {
+	const auto face_writes = [&](const vk::StencilOpState& state) {
+		if (state.writeMask == 0) {
 			return false;
 		}
 		bool can_pass = state.compareOp != vk::CompareOp::eNever;
 		bool can_fail = state.compareOp != vk::CompareOp::eAlways;
-		if (dynamic.compareMask == 0) {
+		if (state.compareMask == 0) {
 			switch (state.compareOp) {
 				case vk::CompareOp::eEqual:
 				case vk::CompareOp::eLessOrEqual:
@@ -439,8 +461,7 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		       (can_pass && depth_fail && state.depthFailOp != vk::StencilOp::eKeep);
 	};
 	if (stencil_clear_enable ||
-	    (stencil_test_enable && (face_writes(stencil_static_front, stencil_dynamic_front) ||
-	                             face_writes(stencil_static_back, stencil_dynamic_back)))) {
+	    (stencil_test_enable && (face_writes(stencil_front) || face_writes(stencil_back)))) {
 		writes |= vk::ImageAspectFlagBits::eStencil;
 	}
 	return writes;

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -34,10 +34,8 @@ struct RenderDepthInfo {
 	bool                        stencil_clear_enable     = false;
 	uint8_t                     stencil_clear_value      = 0;
 	bool                        stencil_test_enable      = false;
-	PipelineStencilStaticState  stencil_static_front;
-	PipelineStencilStaticState  stencil_static_back;
-	PipelineStencilDynamicState stencil_dynamic_front;
-	PipelineStencilDynamicState stencil_dynamic_back;
+	vk::StencilOpState          stencil_front;
+	vk::StencilOpState          stencil_back;
 	ImageId                     image_id;
 
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;

src/graphics/host_gpu/renderer/image/blitHelper.cpp

@@ -10,7 +10,6 @@
 
 #include <algorithm>
 #include <array>
-#include <iterator>
 
 namespace Libs::Graphics {
 
@@ -37,9 +36,8 @@ BlitHelper::BlitHelper(GraphicContext& graphics, CommandScheduler& scheduler)
 	    m_graphics.device.createPipelineLayout(&layout_info, nullptr, &m_pipeline_layout),
 	    "create BlitHelper pipeline layout");
 
-	m_vertex_shader = CreateShader(GPU_BLIT_FS_TRIANGLE_SPV, std::size(GPU_BLIT_FS_TRIANGLE_SPV));
-	m_fragment_shader =
-	    CreateShader(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV, std::size(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV));
+	m_vertex_shader   = CompileSPV(GPU_BLIT_FS_TRIANGLE_SPV, m_graphics.device);
+	m_fragment_shader = CompileSPV(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV, m_graphics.device);
 }
 
 BlitHelper::~BlitHelper() {
@@ -60,17 +58,6 @@ BlitHelper::~BlitHelper() {
 	}
 }
 
-vk::ShaderModule BlitHelper::CreateShader(const uint32_t* code, size_t words) const {
-	EXIT_IF(code == nullptr || words == 0);
-	vk::ShaderModuleCreateInfo create {};
-	create.codeSize         = words * sizeof(uint32_t);
-	create.pCode            = code;
-	vk::ShaderModule module = nullptr;
-	RequireVulkanSuccess(m_graphics.device.createShaderModule(&create, nullptr, &module),
-	                     "create BlitHelper shader module");
-	return module;
-}
-
 vk::Pipeline BlitHelper::GetPipeline(PipelineKey key) {
 	const auto cached = std::ranges::find(m_pipelines, key, &Pipeline::key);
 	if (cached != m_pipelines.end()) {

src/graphics/host_gpu/renderer/image/blitHelper.h

@@ -37,8 +37,7 @@ private:
 		vk::Pipeline handle = nullptr;
 	};
 
-	[[nodiscard]] vk::ShaderModule CreateShader(const uint32_t* code, size_t words) const;
-	[[nodiscard]] vk::Pipeline     GetPipeline(PipelineKey key);
+	[[nodiscard]] vk::Pipeline GetPipeline(PipelineKey key);
 
 	GraphicContext&         m_graphics;
 	CommandScheduler&       m_scheduler;

src/graphics/host_gpu/renderer/image/imageInfo.h

@@ -94,9 +94,24 @@ struct ImageInfo {
 		       bytes_per_block == other.bytes_per_block;
 	}
 	[[nodiscard]] int32_t MipOf(const ImageInfo& container) const noexcept {
-		if (!IsCompatible(container) || tile_mode != container.tile_mode || resources.levels != 1 ||
-		    container.resources.layers == 0 ||
-		    container.resources.levels > container.mip_layout.size()) {
+		if (container.resources.levels > container.mip_layout.size()) {
+			return -1;
+		}
+		for (uint32_t level = 0; level < container.resources.levels; level++) {
+			if (SliceOf(container, static_cast<int32_t>(level)) >= 0) {
+				return static_cast<int32_t>(level);
+			}
+		}
+		return -1;
+	}
+	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
+		if (!IsCompatible(container) || tile_mode != container.tile_mode || type != container.type ||
+		    IsVolume() || resources.levels != 1 || resources.layers == 0 ||
+		    container.resources.layers == 0 || mip < 0 ||
+		    static_cast<uint32_t>(mip) >= container.resources.levels ||
+		    container.resources.levels > container.mip_layout.size() ||
+		    !data.Valid() || !container.data.Valid() || data.address < container.data.address ||
+		    data.End() > container.data.End()) {
 			return -1;
 		}
 		if (HasStencil() != container.HasStencil() ||
@@ -105,75 +120,36 @@ struct ImageInfo {
 			return -1;
 		}
 
-		int32_t mip = -1;
-		for (uint32_t level = 0; level < container.resources.levels; level++) {
-			const auto& layout = container.mip_layout[level];
-			if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-			    container.data.address > UINT64_MAX - layout.offset) {
-				continue;
-			}
-			const auto mip_base   = container.data.address + layout.offset;
-			const auto slice_size = layout.size / container.resources.layers;
-			if (slice_size == 0 || mip_base > UINT64_MAX - layout.size) {
-				continue;
-			}
-			const auto mip_end = mip_base + layout.size;
-			if (data.address >= mip_base && data.address < mip_end &&
-			    (data.address - mip_base) % slice_size == 0) {
-				mip = static_cast<int32_t>(level);
-				break;
-			}
-		}
-		if (mip < 0) {
+		const auto  level  = static_cast<uint32_t>(mip);
+		const auto& layout = container.mip_layout[level];
+		if (extent.width != std::max(container.extent.width >> level, 1u) ||
+		    extent.height != std::max(container.extent.height >> level, 1u) ||
+		    extent.depth != container.extent.depth || mip_layout[0].pitch != layout.pitch ||
+		    mip_layout[0].height != layout.height || layout.size == 0 ||
+		    layout.size % container.resources.layers != 0 ||
+		    container.data.size % container.resources.layers != 0 ||
+		    data.size % resources.layers != 0) {
 			return -1;
 		}
 
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
+		// PS5 block slices contain the entire reversed mip chain.
+		const auto slice_stride = container.data.size / container.resources.layers;
+		const auto child_stride = data.size / resources.layers;
+		if (child_stride != layout.size / container.resources.layers ||
+		    layout.offset >= slice_stride || child_stride > slice_stride - layout.offset ||
+		    (resources.layers > 1 && child_stride != slice_stride)) {
 			return -1;
 		}
-		const auto mip_depth = std::max(container.extent.depth >> level, 1u);
-		if (container.type == Prospero::ImageType::kColor3D &&
-		    type == Prospero::ImageType::kColor2D) {
-			if (resources.layers != mip_depth) {
-				return -1;
-			}
-		} else if (type != container.type) {
+		const auto address_delta = data.address - container.data.address;
+		if (address_delta < layout.offset || (address_delta - layout.offset) % slice_stride != 0) {
 			return -1;
 		}
-		return mip;
-	}
-	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
-		if (!IsCompatible(container) || type != container.type || mip < 0 ||
-		    static_cast<uint32_t>(mip) >= container.resources.levels ||
-		    container.resources.levels > container.mip_layout.size() ||
-		    container.resources.layers == 0 || data.size == 0) {
+		const auto layer = (address_delta - layout.offset) / slice_stride;
+		if (layer > INT32_MAX || layer >= container.resources.layers ||
+		    resources.layers > container.resources.layers - layer) {
 			return -1;
 		}
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
-			return -1;
-		}
-		const auto& layout = container.mip_layout[level];
-		if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-		    container.data.address > UINT64_MAX - layout.offset) {
-			return -1;
-		}
-		const auto slice_size = layout.size / container.resources.layers;
-		if (slice_size == 0 || data.size % slice_size != 0) {
-			return -1;
-		}
-		const auto mip_base = container.data.address + layout.offset;
-		if (data.address < mip_base) {
-			return -1;
-		}
-		const auto address_delta = data.address - mip_base;
-		if (address_delta % data.size != 0 || address_delta / data.size > INT32_MAX) {
-			return -1;
-		}
-		return static_cast<int32_t>(address_delta / data.size);
+		return static_cast<int32_t>(layer);
 	}
 };
 
@@ -183,6 +159,7 @@ struct ImageViewInfo {
 	vk::ImageAspectFlags aspect      = vk::ImageAspectFlagBits::eColor;
 	uint32_t             base_level  = 0;
 	uint32_t             level_count = 1;
+	uint32_t             min_lod     = 0; // U4.8 clamp relative to base_level.
 	uint32_t             base_layer  = 0;
 	uint32_t             layer_count = 1;
 	vk::ComponentMapping mapping     = {};
@@ -191,9 +168,10 @@ struct ImageViewInfo {
 	[[nodiscard]] bool operator==(const ImageViewInfo& rhs) const noexcept {
 		return format == rhs.format && type == rhs.type && aspect == rhs.aspect &&
 		       base_level == rhs.base_level && level_count == rhs.level_count &&
-		       base_layer == rhs.base_layer && layer_count == rhs.layer_count &&
-		       mapping.r == rhs.mapping.r && mapping.g == rhs.mapping.g &&
-		       mapping.b == rhs.mapping.b && mapping.a == rhs.mapping.a && usage == rhs.usage;
+		       min_lod == rhs.min_lod && base_layer == rhs.base_layer &&
+		       layer_count == rhs.layer_count && mapping.r == rhs.mapping.r &&
+		       mapping.g == rhs.mapping.g && mapping.b == rhs.mapping.b &&
+		       mapping.a == rhs.mapping.a && usage == rhs.usage;
 	}
 };
 

src/graphics/host_gpu/renderer/image/imageView.cpp

@@ -362,6 +362,12 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	if (!is_storage) {
 		usage.usage &= ~vk::ImageUsageFlagBits::eStorage;
 	}
+	vk::ImageViewMinLodCreateInfoEXT min_lod {};
+	if (normalized.min_lod != 0) {
+		min_lod.minLod = static_cast<float>(normalized.base_level) +
+		                 static_cast<float>(normalized.min_lod) / 256.0f;
+		usage.pNext    = &min_lod;
+	}
 	vk::ImageViewCreateInfo create {};
 	create.pNext                           = &usage;
 	create.image                           = image.image;

src/graphics/host_gpu/renderer/image/tiler.cpp

@@ -260,12 +260,8 @@ vk::Pipeline TileManager::GetPipeline(uint32_t slot) {
 	const uint32_t                   values[] {1u << element_index, direction_index};
 	const vk::SpecializationMapEntry entries[] {{0, 0, 4}, {1, 4, 4}};
 	const vk::SpecializationInfo     specialization {2, entries, sizeof(values), values};
-	vk::ShaderModuleCreateInfo       module_info {};
-	module_info.codeSize    = shaders[family_index].words * sizeof(uint32_t);
-	module_info.pCode       = shaders[family_index].code;
-	vk::ShaderModule module = nullptr;
-	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-	                     "create TileManager shader module");
+	const auto module =
+	    CompileSPV({shaders[family_index].code, shaders[family_index].words}, m_graphics.device);
 	vk::PipelineShaderStageCreateInfo stage {};
 	stage.stage               = vk::ShaderStageFlagBits::eCompute;
 	stage.module              = module;
@@ -472,12 +468,7 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 			code  = GPU_TILER_DEMOTE_D16_SPV;
 			words = std::size(GPU_TILER_DEMOTE_D16_SPV);
 		}
-		vk::ShaderModuleCreateInfo module_info {};
-		module_info.codeSize    = words * sizeof(uint32_t);
-		module_info.pCode       = code;
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-		                     "create D16 conversion shader module");
+		const auto module = CompileSPV({code, words}, m_graphics.device);
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage               = vk::ShaderStageFlagBits::eCompute;
 		stage.module              = module;
@@ -611,12 +602,7 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 
 void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 	if (m_swap_bgra16 == nullptr) {
-		vk::ShaderModuleCreateInfo module_info {};
-		module_info.codeSize    = std::size(GPU_TILER_SWAP_BGRA16_SPV) * sizeof(uint32_t);
-		module_info.pCode       = GPU_TILER_SWAP_BGRA16_SPV;
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-		                     "create BGRA16 swap shader module");
+		const auto module = CompileSPV(GPU_TILER_SWAP_BGRA16_SPV, m_graphics.device);
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage  = vk::ShaderStageFlagBits::eCompute;
 		stage.module = module;

src/graphics/host_gpu/renderer/pipeline/descriptors.cpp

@@ -15,6 +15,7 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/hostMemory.h"
+#include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -93,22 +94,15 @@ static const char* ShaderStageResourceName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Vertex: return "Vertex";
 		case ShaderType::Mesh: return "Mesh";
+		case ShaderType::Local: return "Local";
+		case ShaderType::TessellationControl: return "Hull";
+		case ShaderType::TessellationEvaluation: return "Domain";
 		case ShaderType::Pixel: return "Pixel";
 		case ShaderType::Compute: return "Compute";
 		default: return "Unknown";
 	}
 }
 
-static vk::ShaderStageFlags NativeShaderStage(ShaderType stage) {
-	switch (stage) {
-		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
-		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
-		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
-		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
-		default: EXIT("unknown native shader stage\n");
-	}
-}
-
 static Prospero::ImageType TextureType(const ShaderTextureResource& descriptor) {
 	const auto type = descriptor.Type();
 	return type == Prospero::ImageType::kCube ? Prospero::ImageType::kColor2DArray : type;
@@ -154,7 +148,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	buffer_offset = static_cast<uint32_t>(adjustment);
 	const vk::DescriptorBufferInfo result {buffer->Handle(), aligned_offset, size + adjustment};
-	if (resource.formatted && resource.written) {
+	if (resource.written) {
 		context.GetTextureCache().InvalidateMemoryFromGPU(address, size);
 	}
 	const char* access = "Read";
@@ -190,17 +184,18 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, bo
 	}
 	const bool full = common && descriptor.fields[4] == field4_expected &&
 	                  descriptor.fields[5] == field5_expected;
-	if (!full || (descriptor.fields[6] == 0 && descriptor.fields[7] != 0) ||
+	if (!full ||
 	    (descriptor.MsaaDepth() && !IsMultisampledTexture(descriptor.Type()))) {
 		return false;
 	}
-	if (descriptor.fields[6] == 0) {
+	const auto metadata_control = descriptor.fields[6] & 0x00ffffffu;
+	if (metadata_control == 0) {
 		return true;
 	}
 	constexpr uint32_t htile_control = 0x00280000u;
 	const uint32_t expected_control  = htile_control | (descriptor.MsaaDepth() ? (1u << 10u) : 0u);
 	const auto     metadata_addr     = descriptor.MetaAddr() << 8u;
-	return (descriptor.fields[6] & 0x00ffffffu) == expected_control && metadata_addr != 0 &&
+	return metadata_control == expected_control && metadata_addr != 0 &&
 	       metadata_addr < TRACKER_ADDRESS_SIZE && (metadata_addr & 0x7fffu) == 0 &&
 	       descriptor.TileMode() == Prospero::TileMode::kDepth;
 }
@@ -258,10 +253,10 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	    (is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth());
 	const bool is_2d =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D && valid_2d_slice;
+	// Storage cube coordinates address individual faces, including partial cube views.
 	const bool is_cube = resource.cube && descriptor.Type() == Prospero::ImageType::kCube &&
 	                     descriptor.Width5() == descriptor.Height5() &&
-	                     descriptor.BaseArray5() <= descriptor.Depth() &&
-	                     (descriptor.Depth() - descriptor.BaseArray5() + 1u) % 6u == 0;
+	                     descriptor.BaseArray5() <= descriptor.Depth();
 	const bool is_2d_array =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray &&
 	    ((!resource.cube && is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth()) ||
@@ -299,13 +294,8 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
-	const auto max_mip = resource.r128 ? descriptor.LastLevel() : descriptor.MaxMip();
-	const auto view_last_level =
-	    resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::DynamicStorage
-	        ? descriptor.LastLevel()
-	        : std::min(descriptor.LastLevel(), max_mip);
 	return (is_1d || is_1d_array || is_2d || is_2d_array || is_3d) && supported_tile &&
-	       descriptor.BaseLevel() <= view_last_level && view_last_level <= max_mip &&
+	       descriptor.BaseLevel() <= descriptor.LastLevel() &&
 	       descriptor.MinLod() == 0 && supported_swizzle && descriptor.BCSwizzle() == 0 &&
 	       !descriptor.MsaaDepth();
 }
@@ -463,6 +453,14 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 	view.aspect      = vk::ImageAspectFlagBits::eColor;
 	view.base_level  = descriptor.BaseLevel();
 	view.level_count = view_levels;
+	if (descriptor.MinLod() > descriptor.LastLevel() * 256u) {
+		EXIT("texture minimum LOD exceeds last mip level: min_lod=%u last_level=%u\n",
+		     descriptor.MinLod(), descriptor.LastLevel());
+	}
+	const auto base_lod = view.base_level * 256u;
+	if (descriptor.MinLod() > base_lod) {
+		view.min_lod = descriptor.MinLod() - base_lod;
+	}
 	view.usage = storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlagBits::eSampled;
 	view.mapping =
 	    storage || surface_format.conversion_format != Prospero::BufferFormat::kInvalid
@@ -513,6 +511,21 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 	return view;
 }
 
+static bool TextureViewPreservesMipLayout(const TileSurfaceDescription& description,
+                                          uint32_t                      view_levels) {
+	TileSurfaceLayout physical {};
+	TileSurfaceLayout view {};
+	auto              view_description = description;
+	view_description.levels            = view_levels;
+	return TileGetTiledTextureLayout(description, physical) &&
+	       TileGetTiledTextureLayout(view_description, view) &&
+	       physical.first_tail_level == view.first_tail_level &&
+	       physical.block_slice_size == view.block_slice_size &&
+	       physical.total_size == view.total_size &&
+	       std::equal(std::begin(physical.mips), std::begin(physical.mips) + description.levels,
+	                  std::begin(view.mips));
+}
+
 TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   resource,
                                               const ShaderRecompiler::IR::DescriptorValue& value) {
 	auto descriptor = DecodeNativeDescriptor<ShaderTextureResource>(value);
@@ -529,24 +542,22 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		return {id, nullptr, std::move(desc)};
 	}
 
-	const auto address      = descriptor.Base40();
-	const auto width        = static_cast<uint32_t>(descriptor.Width5()) + 1u;
-	const auto height       = static_cast<uint32_t>(descriptor.Height5()) + 1u;
-	const auto base_level   = descriptor.BaseLevel();
-	const auto last_level   = descriptor.LastLevel();
-	const auto type         = TextureType(descriptor);
-	const bool multisampled = IsMultisampledTexture(type);
-	const auto max_mip      = resource.r128 ? last_level : descriptor.MaxMip();
-	const auto levels       = multisampled ? 1u : static_cast<uint32_t>(max_mip) + 1u;
-	const bool dynamic_storage =
-	    storage && resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::DynamicStorage;
-	const auto view_last_level =
-	    !multisampled && !dynamic_storage ? std::min(last_level, max_mip) : last_level;
+	const auto address         = descriptor.Base40();
+	const auto width           = static_cast<uint32_t>(descriptor.Width5()) + 1u;
+	const auto height          = static_cast<uint32_t>(descriptor.Height5()) + 1u;
+	const auto base_level      = descriptor.BaseLevel();
+	const auto last_level      = descriptor.LastLevel();
+	const auto type            = TextureType(descriptor);
+	const bool multisampled    = IsMultisampledTexture(type);
+	const auto max_mip         = resource.r128 ? last_level : descriptor.MaxMip();
+	const auto physical_levels = multisampled ? 1u : static_cast<uint32_t>(max_mip) + 1u;
+	const auto levels =
+	    multisampled ? 1u : std::max(physical_levels, static_cast<uint32_t>(last_level) + 1u);
 	const auto tile       = descriptor.TileMode();
 	const bool depth_tile = tile == Prospero::TileMode::kDepth;
 	const bool msaa_tile  = depth_tile || tile == Prospero::TileMode::kRenderTarget;
 	const bool msaa_array = type == Prospero::ImageType::kColor2DMsaaArray;
-	if ((!multisampled && (base_level > view_last_level || view_last_level >= levels)) ||
+	if ((!multisampled && base_level > last_level) ||
 	    (multisampled &&
 	     (base_level != 0 || last_level == 0 || last_level > 3 || max_mip != last_level ||
 	      !msaa_tile || (descriptor.MsaaDepth() && !depth_tile) ||
@@ -565,7 +576,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	}
 	const auto samples = multisampled ? 1u << last_level : 1u;
 	const auto view_levels =
-	    multisampled ? 1u : static_cast<uint32_t>(view_last_level - base_level) + 1u;
+	    multisampled ? 1u : static_cast<uint32_t>(last_level - base_level) + 1u;
 	const auto depth          = static_cast<uint32_t>(descriptor.Depth()) + 1u;
 	const auto format         = descriptor.Format();
 	const auto surface_format = TextureGetSurfaceFormatInfo(format);
@@ -584,7 +595,19 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	                             type == Prospero::ImageType::kColor2DArray ||
 	                             type == Prospero::ImageType::kColor2DMsaaArray;
 	const auto    image_layers = layered ? depth : 1u;
-	uint32_t      pitch        = 0;
+	if (levels > physical_levels) {
+		const TileSurfaceDescription physical {
+		    format, tile, volume ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
+		    width, height, volume ? depth : 1u, physical_levels, image_layers};
+		// Texture mip views take precedence over the resource count, but must keep its storage layout.
+		if (!TextureViewPreservesMipLayout(physical, levels)) {
+			EXIT("unsupported texture mip view changes physical layout: base=%u last=%u max=%u "
+			     "extent=%ux%ux%u tile=%u\n",
+			     base_level, last_level, max_mip, width, height, depth,
+			     static_cast<uint32_t>(tile));
+		}
+	}
+	uint32_t      pitch = 0;
 	TileSizeAlign size {};
 	if (multisampled) {
 		const auto bytes = Prospero::NumBytesPerElement(format);
@@ -597,8 +620,8 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		size.size *= image_layers;
 	} else {
 		pitch = TileGetTexturePitch(format, width, tile);
-		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, levels, tile,
-		                        volume, size);
+		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers,
+		                        physical_levels, tile, volume, size);
 	}
 	EXIT_NOT_IMPLEMENTED(size.size == 0 || size.align == 0 ||
 	                     (address & (static_cast<uint64_t>(size.align) - 1u)) != 0);
@@ -635,7 +658,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	    !desc.info.IsDepth()) {
 		TileSizeAlign metadata_size {};
 		(void)TileGetDccSize(width, height, volume ? depth : image_layers,
-		                     desc.info.bytes_per_block, levels, tile, metadata_size,
+		                     desc.info.bytes_per_block, physical_levels, tile, metadata_size,
 		                     std::countr_zero(samples));
 		desc.info.metadata.kind          = ImageMetadataKind::Dcc;
 		desc.info.metadata.range         = {descriptor.MetaAddr() << 8u, metadata_size.size};
@@ -857,32 +880,39 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	}
 }
 
-RenderExecutor::GraphicsBindings
-RenderExecutor::PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-                                        const ShaderStageRuntime& pixel, bool pixel_active) {
-	GraphicsBindings bindings {
-	    .vertex = PrepareBindings(vertex),
-	};
-	if (pixel_active) {
-		bindings.pixel.emplace(PrepareBindings(pixel));
+void RenderExecutor::PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+                                             std::span<RenderColorInfo> colors) {
+	bool uses_dma = false;
+	for (auto* stage: stages) {
+		FindBuffers(*stage);
+		uses_dma |= stage->runtime->program->info.uses_dma;
 	}
-	FindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		FindBuffers(*bindings.pixel);
-	}
-	if (bindings.vertex.runtime->program->info.uses_dma ||
-	    (bindings.pixel && bindings.pixel->runtime->program->info.uses_dma)) {
+	if (uses_dma) {
 		m_context.PrepareBda();
 	}
-	RebindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		RebindBuffers(*bindings.pixel);
+	for (auto* stage: stages) {
+		RebindImages(*stage);
 	}
-	RebindImages(bindings.vertex);
-	if (bindings.pixel) {
-		RebindImages(*bindings.pixel);
+	auto& cache = m_context.GetTextureCache();
+	for (auto& target: colors) {
+		EXIT_IF(!target.image_id);
+		const auto old_image = cache.m_slot_images.try_get(target.image_id);
+		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
+		    old_image->binding.needs_rebind) {
+			if (old_image != nullptr) {
+				old_image->binding = {};
+			}
+			target.desc.view_info.base_level = target.guest_mip_level;
+			target.desc.view_info.base_layer = target.guest_array_layer;
+			target.image_id = cache.FindImage(target.desc);
+			BindRenderTarget(target.image_id);
+		}
+	}
+	// Discovery can read back PS5 metadata and submit the scheduler. Reserve draw buffers only
+	// after image identities are final; attachment layout transitions follow buffer alias copies.
+	for (auto* stage: stages) {
+		RebindBuffers(*stage);
 	}
-	return bindings;
 }
 
 void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
@@ -895,9 +925,10 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 	size_t write_count      = 0;
 	ShaderRecompiler::IR::PushData push_data;
 	bool                           has_push_data = false;
-	constexpr auto                 GraphicsStages = vk::ShaderStageFlagBits::eVertex |
-	                                                vk::ShaderStageFlagBits::eMeshEXT |
-	                                                vk::ShaderStageFlagBits::eFragment;
+	constexpr auto                 GraphicsStages =
+	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eMeshEXT |
+	    vk::ShaderStageFlagBits::eTessellationControl |
+	    vk::ShaderStageFlagBits::eTessellationEvaluation | vk::ShaderStageFlagBits::eFragment;
 	vk::ShaderStageFlags push_stages = pipeline_bind_point == vk::PipelineBindPoint::eGraphics
 	                                       ? vk::ShaderStageFlagBits::eFragment
 	                                       : vk::ShaderStageFlags {};

src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp

@@ -89,12 +89,7 @@ template <typename... Args>
 void PipelineCacheLog(fmt::format_string<Args...> format, Args&&... args) {
 	auto message = fmt::format(format, std::forward<Args>(args)...);
 	message += '\n';
-	if (Log::GetDirection() != Log::Direction::Console) {
-		std::fwrite(message.data(), 1, message.size(), stdout);
-		std::fflush(stdout);
-	}
-	Log::Write(message);
-	Log::Flush();
+	Log::WriteToConsoleAndLog(message);
 }
 
 bool ReadShaderGuestMemory(void*, uint64_t address, uint32_t* value) {
@@ -236,6 +231,9 @@ struct PipelineCache::ProgramCache {
 		switch (options.stage) {
 			case ShaderType::Vertex: stage_name = "vs"; break;
 			case ShaderType::Mesh: stage_name = "ms"; break;
+			case ShaderType::Local: stage_name = "ls"; break;
+			case ShaderType::TessellationControl: stage_name = "hs"; break;
+			case ShaderType::TessellationEvaluation: stage_name = "ds"; break;
 			case ShaderType::Pixel: stage_name = "ps"; break;
 			case ShaderType::Compute: stage_name = "cs"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -250,12 +248,7 @@ struct PipelineCache::ProgramCache {
 		}
 		DumpShaderSpirv(stage_name, options.shader_hash, result.spirv);
 
-		vk::ShaderModuleCreateInfo create_info {};
-		create_info.codeSize    = result.spirv.size() * sizeof(uint32_t);
-		create_info.pCode       = result.spirv.data();
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
-		                     "create recompiled shader module");
+		const auto module = CompileSPV(result.spirv, device);
 		EXIT_IF(module == nullptr);
 		if (options.dump_ir) {
 			LOGF("%s SPIR-V words=%" PRIu64 " wave_size=%u\n", options.dump_label,
@@ -273,7 +266,7 @@ struct PipelineCache::ProgramCache {
 	                  uint32_t& push_data_cursor) {
 		ShaderType stage;
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
-			stage = input_info.mesh.threads_num[0] != 0 ? ShaderType::Mesh : ShaderType::Vertex;
+			stage = input_info.logical_stage;
 		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
 			stage = ShaderType::Pixel;
 		} else {
@@ -325,6 +318,9 @@ struct PipelineCache::ProgramCache {
 		switch (stage) {
 			case ShaderType::Vertex: label = "ShaderRecompiler VS"; break;
 			case ShaderType::Mesh: label = "ShaderRecompiler MS"; break;
+			case ShaderType::Local: label = "ShaderRecompiler LS"; break;
+			case ShaderType::TessellationControl: label = "ShaderRecompiler HS"; break;
+			case ShaderType::TessellationEvaluation: label = "ShaderRecompiler DS"; break;
 			case ShaderType::Pixel: label = "ShaderRecompiler PS"; break;
 			case ShaderType::Compute: label = "ShaderRecompiler CS"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -341,11 +337,12 @@ struct PipelineCache::ProgramCache {
 
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
 			options.user_data_base = 8;
-			if (stage == ShaderType::Mesh) {
+			options.wave_size = input_info.wave_size;
+			if (stage == ShaderType::Mesh || stage == ShaderType::TessellationControl) {
 				options.user_data_base = 0;
-				options.wave_size      = input_info.mesh.wave_size;
+				options.wave_size = stage == ShaderType::Mesh ? input_info.mesh.wave_size : 64u;
 			}
-		} else if constexpr (std::is_same_v<InputInfo, ShaderComputeInputInfo>) {
+		} else {
 			options.wave_size = input_info.wave_size;
 		}
 		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
@@ -361,16 +358,19 @@ struct PipelineCache::ProgramCache {
 		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
 		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
-		std::array<size_t, static_cast<size_t>(ShaderType::Mesh) + 1> counts {};
+		std::array<size_t, static_cast<size_t>(ShaderType::TessellationEvaluation) + 1> counts {};
 		for (const auto& [key, source]: programs) {
 			counts[static_cast<size_t>(key.stage)] += source.permutations.size();
 		}
 		// Guest geometry shaders are compiled through the host mesh stage.
-		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu\n",
+		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu | LS %zu | HS %zu | TES %zu\n",
 		            counts[static_cast<size_t>(ShaderType::Vertex)],
 		            counts[static_cast<size_t>(ShaderType::Pixel)],
 		            counts[static_cast<size_t>(ShaderType::Compute)],
-		            counts[static_cast<size_t>(ShaderType::Mesh)]);
+		            counts[static_cast<size_t>(ShaderType::Mesh)],
+		            counts[static_cast<size_t>(ShaderType::Local)],
+		            counts[static_cast<size_t>(ShaderType::TessellationControl)],
+		            counts[static_cast<size_t>(ShaderType::TessellationEvaluation)]);
 		return permutation.handle;
 	}
 
@@ -564,12 +564,18 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
     const HW::VertexShaderInfo& vertex_regs, const HW::PixelShaderInfo& pixel_regs,
     const HW::ShaderRegisters& sh, const HW::Context& context, const HW::UserConfig& user_config,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping, bool pixel_active,
-    ShaderVertexInputInfo& vertex_info, ShaderPixelInputInfo& pixel_info) {
-	const auto vertex_params = PrepareProgram(vertex_regs, context, user_config, vertex_info);
-	const bool mesh_active   = vertex_info.mesh.threads_num[0] != 0;
+    std::array<ShaderVertexInputInfo, 3>& vertex_info, ShaderPixelInputInfo& pixel_info) {
+	const bool tess_active = user_config.GetPrimType() == Prospero::PrimitiveType::kPatch;
+	std::array<ShaderParams, 3> vertex_params;
+	if (tess_active) {
+		vertex_params = PrepareTessellationPrograms(vertex_regs, context, vertex_info);
+	} else {
+		vertex_params[0] = PrepareProgram(vertex_regs, context, user_config, vertex_info[0]);
+	}
+	const bool mesh_active = vertex_info[0].logical_stage == ShaderType::Mesh;
 	if (mesh_active) {
 		EXIT_NOT_IMPLEMENTED(!m_graphics.mesh_shader_enabled);
-		auto& mesh              = vertex_info.mesh;
+		auto& mesh              = vertex_info[0].mesh;
 		mesh.host_subgroup_size = m_graphics.subgroup_size;
 		const auto& limits      = m_graphics.mesh_shader_properties;
 		const auto  logical_threads =
@@ -588,11 +594,26 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	ShaderParams pixel_params;
 	if (pixel_active) {
 		pixel_params = PrepareProgram(pixel_regs, sh, target_export_mapping, pixel_info);
+		const auto& blend          = context.GetBlendControl(0);
+		const auto  is_dual_source = [](uint8_t factor) {
+			return factor >= static_cast<uint8_t>(Prospero::BlendFactor::kSrc1Color) &&
+			       factor <= static_cast<uint8_t>(Prospero::BlendFactor::kOneMinusSrc1Alpha);
+		};
+		pixel_info.dual_source_blending =
+		    blend.enable && !context.GetRenderTarget(0).info.blend_bypass &&
+		    (is_dual_source(blend.color_srcblend) || is_dual_source(blend.color_destblend) ||
+		     (blend.separate_alpha_blend &&
+		      (is_dual_source(blend.alpha_srcblend) || is_dual_source(blend.alpha_destblend))));
+		if (pixel_info.dual_source_blending) {
+			// MRT1 supplies a second blend source for the same render target as MRT0.
+			pixel_info.target_output_mode[1]    = pixel_info.target_output_mode[0];
+			pixel_info.target_export_mapping[1] = pixel_info.target_export_mapping[0];
+		}
 	}
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];
 		const auto& limits   = m_graphics.GetPhysicalDeviceProperties().limits;
-		auto&       clip     = vertex_info.clip_space;
+		auto&       clip     = vertex_info[tess_active ? 2u : 0u].clip_space;
 		clip.scale[0]        = viewport.xscale;
 		clip.scale[1]        = viewport.yscale;
 		clip.offset[0]       = viewport.xoffset;
@@ -610,7 +631,9 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (pixel_active) {
 		result.pixel = m_program_cache->Get(pixel_params, pixel_info, push_data_cursor);
 	}
-	result.vertex = m_program_cache->Get(vertex_params, vertex_info, push_data_cursor);
+	for (uint32_t i = 0; i < (tess_active ? 3u : 1u); i++) {
+		result.vertex[i] = m_program_cache->Get(vertex_params[i], vertex_info[i], push_data_cursor);
+	}
 	return result;
 }
 
@@ -630,10 +653,12 @@ bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other)
 
 PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
     std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-    const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
+    std::span<const ShaderVertexInputInfo> vertex_info, CommandBuffer& command,
     const ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
-    bool primitive_restart_enable, const ShaderProgram& vertex_program,
-    const ShaderProgram& pixel_program) {
+    bool primitive_restart_enable, const GraphicsPrograms& programs) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Gfx)", profiler::colors::DeepOrangeA200);
 
 	EXIT_IF(colors.size() > RENDER_COLOR_ATTACHMENTS_MAX);
@@ -651,17 +676,22 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	const auto ps_id = ps_active ? pixel_program.id : 0;
 
 	GraphicsPipelineKey key {};
-	key.vs_shader_id            = vs_id;
+	for (uint32_t i = 0; i < programs.vertex.size(); i++) {
+		key.vertex_shader_ids[i] = programs.vertex[i].id;
+	}
 	key.ps_shader_id            = ps_id;
 	auto& static_params         = key.static_params;
 	auto& rendering             = key.rendering;
-	rendering.color_count       = color_count;
+	rendering.color_count       = 0;
 	uint32_t attachment_samples = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
+		const auto slot = colors[i].target_slot;
+		EXIT_IF(slot >= RENDER_COLOR_ATTACHMENTS_MAX);
+		rendering.color_count = std::max(rendering.color_count, slot + 1);
 		EXIT_IF(!colors[i].image_id || colors[i].desc.view_info.format == vk::Format::eUndefined);
-		static_params.color_mask[i] = colors[i].export_mapping.ApplyMask(
+		static_params.color_mask[slot] = colors[i].export_mapping.ApplyMask(
 		    render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot));
-		rendering.color_formats[i] = colors[i].desc.view_info.format;
+		rendering.color_formats[slot] = colors[i].desc.view_info.format;
 		if (attachment_samples == 0) {
 			attachment_samples = colors[i].desc.info.samples;
 		} else if (attachment_samples != colors[i].desc.info.samples) {
@@ -670,14 +700,14 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 		}
 		const auto& rt                        = ctx.GetRenderTarget(colors[i].target_slot);
 		const auto& bc                        = ctx.GetBlendControl(colors[i].target_slot);
-		static_params.color_srcblend[i]       = bc.color_srcblend;
-		static_params.color_comb_fcn[i]       = bc.color_comb_fcn;
-		static_params.color_destblend[i]      = bc.color_destblend;
-		static_params.alpha_srcblend[i]       = bc.alpha_srcblend;
-		static_params.alpha_comb_fcn[i]       = bc.alpha_comb_fcn;
-		static_params.alpha_destblend[i]      = bc.alpha_destblend;
-		static_params.separate_alpha_blend[i] = bc.separate_alpha_blend;
-		static_params.blend_enable[i]         = bc.enable && !rt.info.blend_bypass;
+		static_params.color_srcblend[slot]       = bc.color_srcblend;
+		static_params.color_comb_fcn[slot]       = bc.color_comb_fcn;
+		static_params.color_destblend[slot]      = bc.color_destblend;
+		static_params.alpha_srcblend[slot]       = bc.alpha_srcblend;
+		static_params.alpha_comb_fcn[slot]       = bc.alpha_comb_fcn;
+		static_params.alpha_destblend[slot]      = bc.alpha_destblend;
+		static_params.separate_alpha_blend[slot] = bc.separate_alpha_blend;
+		static_params.blend_enable[slot]         = bc.enable && !rt.info.blend_bypass;
 	}
 	const bool with_depth =
 	    depth.desc.view_info.format != vk::Format::eUndefined && static_cast<bool>(depth.image_id);
@@ -726,10 +756,7 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	static_params.depth_bounds_test_enable = depth.depth_bounds_test_enable;
 	static_params.depth_min_bounds         = depth.depth_min_bounds;
 	static_params.depth_max_bounds         = depth.depth_max_bounds;
-	static_params.stencil_test_enable      = depth.stencil_test_enable;
-	static_params.stencil_front            = depth.stencil_static_front;
-	static_params.stencil_back             = depth.stencil_static_back;
-	const bool rect_list     = topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list = Prospero::IsRectList(command.GetUserConfig().GetPrimType());
 	static_params.cull_back  = !rect_list && mc.cull_back;
 	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;
@@ -780,9 +807,8 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 
 	auto cached = std::make_unique<Pipeline>();
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id, ps_id);
-	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vs_input_info,
-	                       vertex_program, ps_input_info, pixel_program, static_params,
-	                       m_driver_cache);
+	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vertex_info,
+	                       ps_input_info, programs, static_params, m_driver_cache);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id, ps_id);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);

src/graphics/host_gpu/renderer/pipeline/pipelineCache.h

@@ -42,9 +42,6 @@ struct PipelineStaticParameters {
 	bool                       depth_bounds_test_enable = false;
 	float                      depth_min_bounds         = 0.0f;
 	float                      depth_max_bounds         = 0.0f;
-	bool                       stencil_test_enable      = false;
-	PipelineStencilStaticState stencil_front;
-	PipelineStencilStaticState stencil_back;
 	uint32_t                   color_mask[RENDER_COLOR_ATTACHMENTS_MAX]           = {};
 	bool                       cull_front                                         = false;
 	bool                       cull_back                                          = false;
@@ -68,7 +65,7 @@ struct PipelineStaticParameters {
 static_assert(std::is_trivially_copyable_v<PipelineStaticParameters>);
 static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
-static_assert(sizeof(PipelineStaticParameters) == 158);
+static_assert(sizeof(PipelineStaticParameters) == 125);
 
 struct PipelineRenderingState {
 	std::array<vk::Format, RENDER_COLOR_ATTACHMENTS_MAX> color_formats {};
@@ -121,8 +118,10 @@ public:
 	};
 
 	struct GraphicsPrograms {
-		ShaderProgram vertex;
+		std::array<ShaderProgram, 3> vertex;
 		ShaderProgram pixel;
+
+		[[nodiscard]] uint32_t VertexStageCount() const { return vertex[1] ? 3u : 1u; }
 	};
 
 	GraphicsPrograms
@@ -130,18 +129,18 @@ public:
 	                    const HW::PixelShaderInfo& pixel_regs, const HW::ShaderRegisters& sh,
 	                    const HW::Context& context, const HW::UserConfig& user_config,
 	                    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-	                    bool pixel_active, ShaderVertexInputInfo& vertex_info,
+	                    bool pixel_active, std::array<ShaderVertexInputInfo, 3>& vertex_info,
 	                    ShaderPixelInputInfo& pixel_info);
 	ShaderProgram GetComputeProgram(const HW::ComputeShaderInfo& regs,
 	                                const HW::ShaderRegisters&   sh,
 	                                ShaderComputeInputInfo&      input_info);
 
-	Pipeline&
-	GetGraphicsPipeline(std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-	                    const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
-	                    const ShaderPixelInputInfo* ps_input_info,
-	                    vk::PrimitiveTopology topology, bool primitive_restart_enable,
-	                    const ShaderProgram& vertex_program, const ShaderProgram& pixel_program);
+	Pipeline& GetGraphicsPipeline(std::span<const RenderColorInfo>       colors,
+	                              const RenderDepthInfo&                 depth,
+	                              std::span<const ShaderVertexInputInfo> vertex_info,
+	                              CommandBuffer& command, const ShaderPixelInputInfo* ps_input_info,
+	                              vk::PrimitiveTopology topology, bool primitive_restart_enable,
+	                              const GraphicsPrograms& programs);
 	Pipeline& GetComputePipeline(const ShaderComputeInputInfo& input_info,
 	                             const ShaderProgram&          compute_program);
 
@@ -150,13 +149,13 @@ private:
 
 	struct GraphicsPipelineKey {
 		PipelineRenderingState   rendering;
-		uint64_t                 vs_shader_id = 0;
+		std::array<uint64_t, 3>  vertex_shader_ids {};
 		uint64_t                 ps_shader_id = 0;
 		PipelineVertexInputState vertex_input;
 		PipelineStaticParameters static_params;
 
 		bool operator==(const GraphicsPipelineKey& other) const {
-			return rendering == other.rendering && vs_shader_id == other.vs_shader_id &&
+			return rendering == other.rendering && vertex_shader_ids == other.vertex_shader_ids &&
 			       ps_shader_id == other.ps_shader_id && vertex_input == other.vertex_input &&
 			       static_params == other.static_params;
 		}
@@ -189,7 +188,9 @@ private:
 		std::size_t operator()(const GraphicsPipelineKey& key) const {
 			std::size_t hash = 0;
 			PipelineKeyHash::MixRendering(hash, key.rendering);
-			PipelineKeyHash::Mix(hash, key.vs_shader_id);
+			for (const auto id: key.vertex_shader_ids) {
+				PipelineKeyHash::Mix(hash, id);
+			}
 			PipelineKeyHash::Mix(hash, key.ps_shader_id);
 			PipelineKeyHash::Mix(hash, key.vertex_input.binding_count);
 			for (uint32_t i = 0; i < key.vertex_input.binding_count; i++) {
@@ -219,12 +220,14 @@ private:
 };
 
 void LogPipelineTrace(const char* phase, uint64_t vertex_program_id, uint64_t pixel_program_id);
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache);
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache);
 void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
                             vk::ShaderModule compute_module, vk::PipelineCache driver_cache);

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.cpp

@@ -8,6 +8,20 @@
 
 namespace Libs::Graphics {
 
+vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage) {
+	switch (stage) {
+		case ShaderType::Local:
+		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
+		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
+		case ShaderType::TessellationControl: return vk::ShaderStageFlagBits::eTessellationControl;
+		case ShaderType::TessellationEvaluation:
+			return vk::ShaderStageFlagBits::eTessellationEvaluation;
+		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
+		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
+		default: EXIT("unknown native shader stage\n");
+	}
+}
+
 vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	vk::PipelineStageFlags result = {};
 	if (stages & vk::ShaderStageFlagBits::eVertex) {
@@ -16,6 +30,12 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	if (stages & vk::ShaderStageFlagBits::eMeshEXT) {
 		result |= vk::PipelineStageFlagBits::eMeshShaderEXT;
 	}
+	if (stages & vk::ShaderStageFlagBits::eTessellationControl) {
+		result |= vk::PipelineStageFlagBits::eTessellationControlShader;
+	}
+	if (stages & vk::ShaderStageFlagBits::eTessellationEvaluation) {
+		result |= vk::PipelineStageFlagBits::eTessellationEvaluationShader;
+	}
 	if (stages & vk::ShaderStageFlagBits::eFragment) {
 		result |= vk::PipelineStageFlagBits::eFragmentShader;
 	}
@@ -26,8 +46,8 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	return result;
 }
 
-VulkanMemoryBarrier MakeShaderWriteDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderWriteDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite |
 	                        vk::AccessFlagBits::eVertexAttributeRead |
@@ -38,15 +58,15 @@ VulkanMemoryBarrier MakeShaderWriteDependency() {
 	return barrier;
 }
 
-VulkanMemoryBarrier MakeShaderAccessDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderAccessDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	return barrier;
 }
 
-VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderWriteHazardDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	return barrier;

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h

@@ -8,10 +8,11 @@ namespace Libs::Graphics {
 
 struct ShaderStageRuntime;
 
+vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage);
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
-VulkanMemoryBarrier     MakeShaderAccessDependency();
-VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
-VulkanMemoryBarrier     MakeShaderWriteDependency();
+vk::MemoryBarrier       MakeShaderAccessDependency();
+vk::MemoryBarrier       MakeShaderWriteHazardDependency();
+vk::MemoryBarrier       MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
 void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);

src/graphics/host_gpu/renderer/pipeline/shaders.cpp

@@ -8,6 +8,7 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
+#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
@@ -205,12 +206,18 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache) {
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
+	const bool  tessellation   = vertex_info.size() == 3;
 	const bool ps_active = ps_input_info != nullptr;
 	EXIT_IF(!vertex_program || (ps_active && !pixel_program));
 	const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
@@ -218,46 +225,39 @@ void CreatePipelineInternal(
 	EXIT_IF(!vs_input_info.stage);
 	const bool mesh = vs_input_info.stage.program->stage == ShaderType::Mesh;
 	EXIT_NOT_IMPLEMENTED(mesh && !graphics.mesh_shader_enabled);
-	const auto vertex_stage =
-	    mesh ? vk::ShaderStageFlagBits::eMeshEXT : vk::ShaderStageFlagBits::eVertex;
-
-	const bool rect_list = !mesh && static_params.topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list =
+	    !mesh && !tessellation && static_params.topology == vk::PrimitiveTopology::ePatchList;
 
 	vk::ShaderModule tess_control_shader_module = nullptr;
 	vk::ShaderModule tess_eval_shader_module    = nullptr;
 
-	vk::ShaderModuleCreateInfo create_info {};
-	vk::Result result {};
 	if (rect_list) {
 		const auto shaders =
 		    BuildRectListShaders(vs_input_info, ps_active ? ps_input_info : nullptr);
-		create_info.codeSize = shaders.control.size() * 4;
-		create_info.pCode    = shaders.control.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_control_shader_module);
+		tess_control_shader_module = CompileSPV(shaders.control, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_control_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done module=%p\n",
+			     static_cast<void*>(tess_control_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
-		create_info.codeSize = shaders.evaluation.size() * 4;
-		create_info.pCode    = shaders.evaluation.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_eval_shader_module);
+		tess_eval_shader_module = CompileSPV(shaders.evaluation, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_eval_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done module=%p\n",
+			     static_cast<void*>(tess_eval_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
 
 	EXIT_NOT_IMPLEMENTED(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 
-	vk::PipelineShaderStageCreateInfo shader_stages[4] = {
-	    {.stage = vertex_stage, .module = vertex_program.module, .pName = "main"}};
-	uint32_t shader_stage_count = 1;
+	vk::PipelineShaderStageCreateInfo shader_stages[4] {};
+	uint32_t                          shader_stage_count = 0;
+	for (uint32_t i = 0; i < vertex_info.size(); i++) {
+		shader_stages[shader_stage_count++] = {.stage =
+		                                           NativeShaderStage(vertex_info[i].logical_stage),
+		                                       .module = programs.vertex[i].module,
+		                                       .pName  = "main"};
+	}
 	if (rect_list) {
 		shader_stages[shader_stage_count++] = {.stage =
 		                                           vk::ShaderStageFlagBits::eTessellationControl,
@@ -432,15 +432,19 @@ void CreatePipelineInternal(
 	color_blending.pAttachments    = color_blend_attachment;
 
 	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
-	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program, vertex_stage);
+	vk::ShaderStageFlags graphics_stages = vk::ShaderStageFlagBits::eFragment;
+	for (const auto& stage: vertex_info) {
+		const auto native_stage = NativeShaderStage(stage.logical_stage);
+		AddLayoutBindings(descriptor_bindings, *stage.stage.program, native_stage);
+		graphics_stages |= native_stage;
+	}
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
 		                  vk::ShaderStageFlagBits::eFragment);
 	}
 	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
-	const auto                  GraphicsStages = vertex_stage | vk::ShaderStageFlagBits::eFragment;
-	const vk::PushConstantRange push_constants {GraphicsStages, 0,
+	const vk::PushConstantRange push_constants {graphics_stages, 0,
 	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
@@ -456,8 +460,8 @@ void CreatePipelineInternal(
 		     " set_layouts=1 push_constants=%" PRIu32 "\n",
 		     vertex_program.id, ps_active ? pixel_program.id : 0, 1u);
 	}
-	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
-	                                              &pipeline.pipeline_layout);
+	auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
+	                                                   &pipeline.pipeline_layout);
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout done result=%s layout=%p\n",
 		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
@@ -473,15 +477,6 @@ void CreatePipelineInternal(
 #else
 	    (static_params.depth_bounds_test_enable ? VK_TRUE : VK_FALSE);
 #endif
-	depth_stencil_info.stencilTestEnable = (static_params.stencil_test_enable ? VK_TRUE : VK_FALSE);
-	depth_stencil_info.front.failOp      = static_params.stencil_front.failOp;
-	depth_stencil_info.front.passOp      = static_params.stencil_front.passOp;
-	depth_stencil_info.front.depthFailOp = static_params.stencil_front.depthFailOp;
-	depth_stencil_info.front.compareOp   = static_params.stencil_front.compareOp;
-	depth_stencil_info.back.failOp       = static_params.stencil_back.failOp;
-	depth_stencil_info.back.passOp       = static_params.stencil_back.passOp;
-	depth_stencil_info.back.depthFailOp  = static_params.stencil_back.depthFailOp;
-	depth_stencil_info.back.compareOp    = static_params.stencil_back.compareOp;
 	depth_stencil_info.minDepthBounds    = static_params.depth_min_bounds;
 	depth_stencil_info.maxDepthBounds    = static_params.depth_max_bounds;
 
@@ -494,6 +489,8 @@ void CreatePipelineInternal(
 	    vk::DynamicState::eDepthCompareOp,
 	    vk::DynamicState::eDepthBiasEnable,
 	    vk::DynamicState::eDepthBias,
+	    vk::DynamicState::eStencilTestEnable,
+	    vk::DynamicState::eStencilOp,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
@@ -524,8 +521,9 @@ void CreatePipelineInternal(
 	pipeline_info.pVertexInputState        = mesh ? nullptr : &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = mesh ? nullptr : &input_assembly;
 	vk::PipelineTessellationStateCreateInfo tessellation_state {};
-	tessellation_state.patchControlPoints = 3;
-	pipeline_info.pTessellationState      = (rect_list ? &tessellation_state : nullptr);
+	tessellation_state.patchControlPoints =
+	    tessellation ? vs_input_info.tess.input_control_points : 3u;
+	pipeline_info.pTessellationState = (rect_list || tessellation) ? &tessellation_state : nullptr;
 	pipeline_info.pViewportState          = &viewport_state;
 	pipeline_info.pRasterizationState     = &rasterizer;
 	pipeline_info.pMultisampleState       = &multisampling;

src/graphics/host_gpu/renderer/render.h

@@ -170,15 +170,14 @@ private:
 	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const DrawAutoArgs& args);
 
 	struct GraphicsBindings {
-		PreparedBindings                vertex;
+		std::array<PreparedBindings, 3> vertex;
 		std::optional<PreparedBindings> pixel;
 	};
 
 	[[nodiscard]] TextureBinding ResolveTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	                                            const ShaderRecompiler::IR::DescriptorValue& value);
-	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-	                                                       const ShaderStageRuntime& pixel,
-	                                                       bool                      pixel_active);
+	void PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+	                             std::span<RenderColorInfo> colors);
 	void ResolveRenderColorTarget(CommandBuffer& buffer, RenderColorInfo& target,
 	                              uint32_t render_target_slice_offset, uint32_t render_target_slot,
 	                              bool ignore_target_mask = false, bool exact_format = false);

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -172,7 +172,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		                                       1, view.base_layer, view.layer_count};
 		vk::ClearValue clear {};
 		clear.depthStencil = vk::ClearDepthStencilValue {0.0f, fill.value};
-		cache.ClearImage(command, binding.image_id, range, clear);
+		cache.ClearImage(command, binding.image_id, image.backing.format, range, clear);
 		return true;
 	}
 	ShaderBufferResource descriptor;
@@ -183,14 +183,6 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		return false;
 	}
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
-		// Track deferred DCC state while the original dispatch writes the metadata allocation.
-		cache.TrackDccFill(descriptor.Base48(), size, packed_clear);
-		static std::atomic<uint32_t> logged_metadata_clears {0};
-		if (logged_metadata_clears.fetch_add(1, std::memory_order_relaxed) < 32) {
-			LOGF("GraphicsRenderDispatchDirect: metadata fill shader=0x%016" PRIx64
-			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
-			     input.stage.program->shader_hash, descriptor.Base48(), size, packed_clear);
-		}
 		return false;
 	}
 	static std::atomic<uint32_t> logged_clears {0};
@@ -214,6 +206,17 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	                    thread_group_x, thread_group_y, thread_group_z, mode,
 	                    sh_ctx.GetCs().cs_regs.data_addr);
 
+	if (thread_group_x == 0 || thread_group_y == 0 || thread_group_z == 0) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1, std::memory_order_relaxed) < 32) {
+			LOGF("GraphicsRenderDispatchDirect: skipping zero-sized dispatch groups=%ux%ux%u "
+			     "mode=0x%08" PRIx32 " shader=0x%016" PRIx64 "\n",
+			     thread_group_x, thread_group_y, thread_group_z, mode,
+			     sh_ctx.GetCs().cs_regs.data_addr);
+		}
+		return;
+	}
+
 	Common::LockGuard lock(m_context.GetMutex());
 	if (sh_ctx.GetCs().cs_regs.data_addr == 0) {
 		LOGF("GraphicsRenderDispatchDirect: temporary: ignoring dispatch with null CS shader, "
@@ -222,7 +225,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (!ShaderAddressValid(sh_ctx.GetCs().cs_regs.data_addr)) {
+	if (sh_ctx.GetCs().cs_regs.data_addr == 0) {
 		return;
 	}
 
@@ -355,17 +358,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		}
 	}
 
-	if (thread_group_x == 0 || thread_group_y == 0 || thread_group_z == 0) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1, std::memory_order_relaxed) < 32) {
-			LOGF("GraphicsRenderDispatchDirect: skipping zero-sized dispatch groups=%ux%ux%u "
-			     "mode=0x%08" PRIx32 " shader=0x%016" PRIx64 "\n",
-			     thread_group_x, thread_group_y, thread_group_z, mode,
-			     sh_ctx.GetCs().cs_regs.data_addr);
-		}
-		return;
-	}
-
 	buffer.EndRendering();
 	auto& pipeline =
 	    m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
@@ -374,8 +366,8 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	if (program.info.uses_dma) {
 		m_context.PrepareBda();
 	}
-	RebindBuffers(bindings);
 	RebindImages(bindings);
+	RebindBuffers(bindings);
 
 	auto              vk_buffer        = buffer.Handle();
 	PreparedBindings* descriptor_stage = &bindings;

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -2,6 +2,7 @@
 
 #include "common/assert.h"
 #include "common/common.h"
+#include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -15,6 +16,7 @@
 #include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/depthRenderTarget.h"
+#include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
@@ -308,24 +310,12 @@ static void LogDrawInputState(const CommandBuffer& buffer, const RenderColorInfo
 
 static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuffer vk_buffer,
                                      const ShaderVertexInputInfo& vs_input_info,
-                                     const RenderColorInfo* colors, uint32_t color_count,
-                                     const RenderDepthInfo& depth) {
+                                     const RenderDepthInfo& depth, const RenderState& rendering) {
 	KYTY_PROFILER_FUNCTION();
 
-	EXIT_IF(colors == nullptr);
 	const auto& ctx = buffer.GetRegisters();
-
-	const auto&  vp = ctx.GetScreenViewport();
-	vk::Extent2D framebuffer_extent {};
-	if (color_count > 0 && colors[0].image_id) {
-		framebuffer_extent = colors[0].Extent();
-	} else if (depth.image_id) {
-		framebuffer_extent = {depth.desc.info.extent.width, depth.desc.info.extent.height};
-	} else {
-		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
-		framebuffer_extent = {limits.maxFramebufferWidth, limits.maxFramebufferHeight};
-	}
-
+	const auto&        vp  = ctx.GetScreenViewport();
+	const vk::Extent2D framebuffer_extent {rendering.width, rendering.height};
 	const auto& outputs = vs_input_info.stage.program->info.outputs;
 	const bool  indexed_viewports =
 	    std::any_of(outputs.begin(), outputs.end(), [](const auto& output) {
@@ -403,19 +393,16 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 		vk_buffer.setDepthBias(constant_factor, poly_offset.clamp, slope_factor);
 	}
 
+	vk_buffer.setStencilTestEnable(depth.stencil_test_enable ? VK_TRUE : VK_FALSE);
 	if (depth.stencil_test_enable) {
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
-		                                depth.stencil_dynamic_front.compareMask);
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eBack,
-		                                depth.stencil_dynamic_back.compareMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.writeMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.writeMask);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.reference);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.reference);
+		const auto set_stencil = [&](vk::StencilFaceFlagBits face, const vk::StencilOpState& state) {
+			vk_buffer.setStencilOp(face, state.failOp, state.passOp, state.depthFailOp, state.compareOp);
+			vk_buffer.setStencilCompareMask(face, state.compareMask);
+			vk_buffer.setStencilWriteMask(face, state.writeMask);
+			vk_buffer.setStencilReference(face, state.reference);
+		};
+		set_stencil(vk::StencilFaceFlagBits::eFront, depth.stencil_front);
+		set_stencil(vk::StencilFaceFlagBits::eBack, depth.stencil_back);
 	}
 
 #if defined(__APPLE__)
@@ -423,15 +410,11 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 	// eColorWriteEnableEXT dynamic state and relies on the static colorWriteMask instead.
 #else
 	vk::Bool32 enable[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	// Color-control operation selects special color-buffer paths, not the normal component write
-	// mask. Attachment availability therefore follows the target write mask.
-	for (uint32_t i = 0; i < color_count; i++) {
-		enable[i] = render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot) != 0
-		                ? VK_TRUE
-		                : VK_FALSE;
+	for (uint32_t slot = 0; slot < rendering.num_color_attachments; slot++) {
+		enable[slot] = rendering.color_attachments[slot].image_view != nullptr;
 	}
-	if (color_count != 0) {
-		vk_buffer.setColorWriteEnableEXT(color_count, enable);
+	if (rendering.num_color_attachments != 0) {
+		vk_buffer.setColorWriteEnableEXT(rendering.num_color_attachments, enable);
 	}
 #endif
 }
@@ -439,7 +422,7 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 static bool DrawHasValidVertexShader(const HW::Shader& sh_ctx) {
 
 	const auto& vs = sh_ctx.GetVs();
-	return ShaderAddressValid(vs.es_regs.data_addr);
+	return vs.es_regs.data_addr != 0;
 }
 
 static bool PixelShaderHasDepthOrCoverageSideEffects(const HW::ShaderRegisters& sh_regs) {
@@ -453,7 +436,7 @@ struct DrawRenderState {
 	RenderColorInfo       color_info[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	uint32_t              color_count                              = 0;
 	bool                  ps_active                                = true;
-	ShaderVertexInputInfo vs_input_info;
+	std::array<ShaderVertexInputInfo, 3> vertex_info;
 	ShaderPixelInputInfo  ps_input_info;
 	PipelineCache::GraphicsPrograms programs;
 };
@@ -477,19 +460,14 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 	state.width                 = std::numeric_limits<uint32_t>::max();
 	state.height                = std::numeric_limits<uint32_t>::max();
 	state.num_layers            = std::numeric_limits<uint32_t>::max();
-	state.num_color_attachments = color_count;
-	uint32_t attachment_samples = 0;
+	state.num_color_attachments = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
-		const auto old_image = cache.m_slot_images.try_get(target.image_id);
-		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
-		    old_image->binding.needs_rebind) {
-			if (old_image != nullptr) {
-				old_image->binding = {};
-			}
-			target.image_id = cache.FindImage(target.desc);
-			BindRenderTarget(target.image_id);
+		const auto owner = cache.m_slot_images.try_get(target.image_id);
+		if (owner == nullptr || (!owner->registered && !owner->info.data.Empty()) ||
+		    owner->binding.needs_rebind) {
+			EXIT("color target changed after render-state discovery\n");
 		}
 		const auto image_view = cache.FindRenderTarget(target.image_id, target.desc);
 		auto&      image      = cache.GetImage(target.image_id);
@@ -503,12 +481,6 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		                     target.desc.view_info.base_level, target.desc.view_info.base_layer,
 		                     target.desc.view_info.layer_count);
 		EXIT_IF(image.backing.samples != target.desc.info.samples || image_view == nullptr);
-		if (attachment_samples == 0) {
-			attachment_samples = target.desc.info.samples;
-		} else if (attachment_samples != target.desc.info.samples) {
-			EXIT("mixed color attachment sample counts are unsupported: %u and %u\n",
-			     attachment_samples, target.desc.info.samples);
-		}
 		const auto& view   = target.desc.view_info;
 		const auto  layout = image.binding.is_bound ? vk::ImageLayout::eGeneral
 		                                            : vk::ImageLayout::eColorAttachmentOptimal;
@@ -523,7 +495,8 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		state.width             = std::min(state.width, extent.width);
 		state.height            = std::min(state.height, extent.height);
 		state.num_layers        = std::min(state.num_layers, view.layer_count);
-		auto& attachment        = state.color_attachments[i];
+		state.num_color_attachments = std::max(state.num_color_attachments, target.target_slot + 1);
+		auto& attachment            = state.color_attachments[target.target_slot];
 		attachment.image_view   = image_view;
 		attachment.image_layout = layout;
 	}
@@ -556,12 +529,6 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		                     image.info.data.address, depth.desc.view_info.base_layer,
 		                     depth.desc.view_info.layer_count);
 		EXIT_IF(image_view == nullptr || image.backing.samples != depth.desc.info.samples);
-		if (attachment_samples == 0) {
-			attachment_samples = depth.desc.info.samples;
-		} else if (attachment_samples != depth.desc.info.samples) {
-			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n", attachment_samples,
-			     depth.desc.info.samples);
-		}
 		const bool feedback = depth.depth_write_enable && pixel &&
 		    std::ranges::any_of(pixel->images, [&](const TextureBinding& binding) {
 			    if (binding.image_id != depth.image_id ||
@@ -612,9 +579,6 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
 		state.width        = limits.maxFramebufferWidth;
 		state.height       = limits.maxFramebufferHeight;
-	} else if (attachment_samples == 0 ||
-	           vulkan_sample_count(attachment_samples) == vk::SampleCountFlagBits {}) {
-		EXIT("render state has no valid attachments\n");
 	}
 	if (state.num_layers == std::numeric_limits<uint32_t>::max()) {
 		state.num_layers = 1;
@@ -629,7 +593,7 @@ static bool DrawHasActivePixelShader(const CommandBuffer& buffer) {
 	const auto& ctx              = buffer.GetRegisters();
 	const auto& sh_regs          = ctx.GetShaderRegisters();
 	const bool  has_color_output = (ctx.GetRenderTargetMask() & sh_regs.m_cbShaderMask) != 0;
-	return ShaderAddressValid(buffer.GetShaders().GetPs().ps_regs.data_addr) &&
+	return buffer.GetShaders().GetPs().ps_regs.data_addr != 0 &&
 	       (has_color_output || PixelShaderHasDepthOrCoverageSideEffects(sh_regs));
 }
 
@@ -709,6 +673,7 @@ struct PreparedVertexBuffers {
 
 	std::array<vk::Buffer, MaxBuffers>     buffers {};
 	std::array<vk::DeviceSize, MaxBuffers> offsets {};
+	std::array<vk::DeviceSize, MaxBuffers> sizes {};
 	uint32_t                               count = 0;
 };
 
@@ -793,6 +758,7 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 
 		prepared.buffers[i] = range->binding.first->Handle();
 		prepared.offsets[i] = range->binding.second + vertex.addr - range->base_address;
+		prepared.sizes[i]   = std::min(size, range->acquired_end - vertex.addr);
 		SetVulkanObjectNameF(
 		    buffer.GetContext().GetGraphics().device, prepared.buffers[i],
 		    "Kyty.VertexBuffer[slot={} guest=0x{:016x} size=0x{:x} stride={} records={}]", i,
@@ -808,8 +774,7 @@ static void SetDrawDebugPhase(CommandBuffer& buffer, uint64_t submit_id, const D
 	                    draw.instance_count, draw.first_instance);
 }
 
-static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
-                            vk::PrimitiveTopology& topology) {
+static bool GetDrawTopology(const HW::UserConfig& ucfg, vk::PrimitiveTopology& topology) {
 
 	topology = vk::PrimitiveTopology::ePointList;
 
@@ -831,14 +796,12 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 		case Prospero::PrimitiveType::kTriStrip:
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
+		case Prospero::PrimitiveType::kPatch:
+			if (!Config::TessellationEnabled()) return false;
+			[[fallthrough]];
 		case Prospero::PrimitiveType::kRectList:
-			topology = vk::PrimitiveTopology::ePatchList;
-			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
-			if (!auto_draw) {
-				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
-			}
-			topology = vk::PrimitiveTopology::eTriangleStrip;
+			topology = vk::PrimitiveTopology::ePatchList;
 			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
@@ -895,42 +858,6 @@ static bool ResolvePrimitiveRestart(const CommandBuffer& buffer,
 	return false;
 }
 
-bool RenderExecutor::PrepareDrawRenderState(CommandBuffer& buffer, const DrawCallInfo& draw,
-                                            uint32_t            render_target_slice_offset,
-	                                        DrawRenderState& state) {
-	auto& ctx = buffer.GetRegisters();
-
-	if (ResolveColorTargets(buffer, render_target_slice_offset)) {
-		return false;
-	}
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "ResolveRenderColorTarget");
-	}
-	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
-		if (slot == 0 || (render_target_mask_slot(ctx.GetRenderTargetMask(), slot) != 0 &&
-		                  ctx.GetRenderTarget(slot).base.addr != 0)) {
-			ResolveRenderColorTarget(buffer, state.color_info[state.color_count],
-			                         render_target_slice_offset, slot);
-			if (state.color_info[state.color_count].image_id) {
-				state.color_count++;
-			}
-		}
-	}
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "ResolveRenderDepthTarget");
-	}
-	ResolveRenderDepthTarget(buffer, state.depth_info);
-
-	state.ps_active       = DrawHasActivePixelShader(buffer);
-	if (state.color_count == 0 && !state.depth_info.image_id && !state.ps_active) {
-		LogFramebufferSkip(draw.Name(), state.color_info[0], state.depth_info, buffer,
-		                   draw.index_count, 0);
-		return false;
-	}
-
-	return true;
-}
-
 static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
                            DrawRenderState& state) {
 	auto& ctx    = buffer.GetRegisters();
@@ -944,8 +871,14 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
 	state.ps_input_info = {};
 	std::array<Prospero::ColorComponentMapping, RENDER_COLOR_ATTACHMENTS_MAX>
 	    target_export_mapping {};
-	for (uint32_t i = 0; i < state.color_count; i++) {
-		target_export_mapping[state.color_info[i].target_slot] = state.color_info[i].export_mapping;
+	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
+		const auto& rt = ctx.GetRenderTarget(slot);
+		if (rt.base.addr != 0 && render_target_mask_slot(ctx.GetRenderTargetMask(), slot) != 0) {
+			target_export_mapping[slot] =
+			    TextureGetRenderTargetFormat(rt.info.format, rt.info.channel_type,
+			                                 rt.info.channel_order)
+			        .export_mapping;
+		}
 	}
 	auto& pipeline_cache = buffer.GetContext().GetPipelineCache();
 	if (draw.IsIndexed()) {
@@ -953,7 +886,46 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
 	}
 	state.programs = pipeline_cache.GetGraphicsPrograms(
 	    vertex_shader_info, pixel_shader_info, shader_regs, ctx, buffer.GetUserConfig(),
-	    target_export_mapping, state.ps_active, state.vs_input_info, state.ps_input_info);
+	    target_export_mapping, state.ps_active, state.vertex_info, state.ps_input_info);
+}
+
+bool RenderExecutor::PrepareDrawRenderState(CommandBuffer& buffer, const DrawCallInfo& draw,
+                                            uint32_t            render_target_slice_offset,
+	                                        DrawRenderState& state) {
+	state.ps_active = DrawHasActivePixelShader(buffer);
+	RefreshShaders(buffer, draw, state);
+	uint32_t mrt_mask = 0;
+	if (state.ps_active) {
+		for (const auto& output: state.ps_input_info.stage.program->info.outputs) {
+			if (output.kind == ShaderRecompiler::IR::StageOutputKind::Mrt) {
+				mrt_mask |= 1u << output.index;
+			}
+		}
+	}
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "ResolveRenderColorTarget");
+	}
+	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
+		if ((mrt_mask & (1u << slot)) != 0) {
+			ResolveRenderColorTarget(buffer, state.color_info[state.color_count],
+			                         render_target_slice_offset, slot);
+			if (state.color_info[state.color_count].image_id) {
+				state.color_count++;
+			}
+		}
+	}
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "ResolveRenderDepthTarget");
+	}
+	ResolveRenderDepthTarget(buffer, state.depth_info);
+
+	if (state.color_count == 0 && !state.depth_info.image_id && !state.ps_active) {
+		LogFramebufferSkip(draw.Name(), state.color_info[0], state.depth_info, buffer,
+		                   draw.index_count, 0);
+		return false;
+	}
+
+	return true;
 }
 
 static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffer,
@@ -988,8 +960,9 @@ static void CommitVertexBuffers(vk::CommandBuffer            vk_buffer,
 		EXIT_IF(prepared.buffers[i] == nullptr);
 	}
 	if (prepared.count != 0) {
-		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
-		                            prepared.offsets.data());
+		// Guest descriptor bounds must survive allocation merging in the cache.
+		vk_buffer.bindVertexBuffers2(0, prepared.count, prepared.buffers.data(),
+		                             prepared.offsets.data(), prepared.sizes.data(), nullptr);
 	}
 }
 
@@ -1007,8 +980,7 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 		return;
 	}
 
-	if (!draw.IsIndexed() &&
-	    buffer.GetUserConfig().GetPrimType() != Prospero::PrimitiveType::kRectListLegacy) {
+	if (!draw.IsIndexed() && !Prospero::IsRectList(buffer.GetUserConfig().GetPrimType())) {
 		return;
 	}
 
@@ -1016,13 +988,12 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 		LogDrawTargetState(draw.Name(), state.color_info[0], state.depth_info, buffer,
 		                   state.ps_input_info, draw.index_count, 0);
 	}
-	LogDrawInputState(buffer, state.color_info[0], state.vs_input_info, index_type_and_size,
+	LogDrawInputState(buffer, state.color_info[0], state.vertex_info[0], index_type_and_size,
 	                  draw.index_count, index_addr);
 }
 
 static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_buffer,
-                               const ShaderVertexInputInfo& vs_input_info, const DrawCallInfo& draw,
-                               const DrawEmitInfo& emit) {
+                               const DrawCallInfo& draw, const DrawEmitInfo& emit) {
 	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kPointList:
 		case Prospero::PrimitiveType::kLineList:
@@ -1031,6 +1002,8 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
 		case Prospero::PrimitiveType::kRectList:
+		case Prospero::PrimitiveType::kRectListLegacy:
+		case Prospero::PrimitiveType::kPatch:
 			if (draw.IsIndexed()) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
 				                      emit.first_instance);
@@ -1039,14 +1012,6 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 				               emit.first_instance);
 			}
 			break;
-		case Prospero::PrimitiveType::kRectListLegacy:
-			if (draw.IsIndexed()) {
-				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
-			}
-			// Sarah
-			EXIT_NOT_IMPLEMENTED(!(draw.index_count == 3 && vs_input_info.buffers_num == 0));
-			vk_buffer.draw(4, draw.instance_count, emit.first_vertex, emit.first_instance);
-			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			EXIT_NOT_IMPLEMENTED((draw.index_count & 0x3u) != 0);
 			for (uint32_t i = 0; i < draw.index_count; i += 4) {
@@ -1069,11 +1034,19 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
                                          const DrawIndexBufferSource& index_source,
 	                                     bool primitive_restart_enable) {
 	auto& ucfg = buffer.GetUserConfig();
-	const bool mesh_active = state.vs_input_info.stage.program->stage == ShaderType::Mesh;
+	const auto vertex_stages =
+	    std::span {state.vertex_info.data(), state.programs.VertexStageCount()};
+	const bool mesh_active = state.vertex_info[0].stage.program->stage == ShaderType::Mesh;
 	uint32_t   mesh_groups = 0;
 	if (mesh_active) {
-		const auto& mesh = state.vs_input_info.mesh;
-		if (primitive_restart_enable || mesh.primitives_per_group == 0) {
+		const auto& mesh = state.vertex_info[0].mesh;
+		static std::atomic_bool restart_warned = false;
+		if (primitive_restart_enable && !restart_warned.exchange(true, std::memory_order_relaxed)) {
+			std::printf("Warning: primitive restart is not implemented for mesh shaders; "
+			            "continuing draw (primitive=%u indexed=%u)\n",
+			            static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed());
+		}
+		if (mesh.primitives_per_group == 0) {
 			EXIT("unsupported mesh draw: primitive=%u indexed=%u restart=%u\n",
 			     static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed(), primitive_restart_enable);
 		}
@@ -1100,26 +1073,36 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		                              index_source.guest_element_size);
 	}
 	LogDrawPhase(draw.Name(), "PrepareBindings");
-	auto bindings = PrepareGraphicsBindings(state.vs_input_info.stage, state.ps_input_info.stage,
-	                                        state.ps_active);
+	GraphicsBindings                 bindings;
+	std::array<PreparedBindings*, 4> descriptor_stages {};
+	uint32_t                         stage_count = 0;
+	for (uint32_t i = 0; i < vertex_stages.size(); i++) {
+		bindings.vertex[i]               = PrepareBindings(state.vertex_info[i].stage);
+		descriptor_stages[stage_count++] = &bindings.vertex[i];
+	}
+	if (state.ps_active) {
+		bindings.pixel.emplace(PrepareBindings(state.ps_input_info.stage));
+		descriptor_stages[stage_count++] = &*bindings.pixel;
+	}
+	const auto stages = std::span {descriptor_stages.data(), stage_count};
+	PrepareGraphicsBindings(stages, std::span {state.color_info, state.color_count});
 	PreparedVertexBuffers vertex_bindings;
 	PreparedIndexBuffer   index_binding;
 	if (!mesh_active) {
 		LogDrawPhase(draw.Name(), "PrepareVertexBuffers");
-		vertex_bindings = AcquireVertexBuffers(buffer, state.vs_input_info);
+		vertex_bindings = AcquireVertexBuffers(buffer, state.vertex_info[0]);
 		index_binding   = PrepareIndexBuffer(buffer, index_source);
 	}
-	const auto rendering =
-	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info,
-	                         bindings.pixel);
-
 	if (draw.IsIndexed()) {
 		LogDrawPhase(draw.Name(), "CreatePipeline");
 	}
 	auto& pipeline = m_context.GetPipelineCache().GetGraphicsPipeline(
-	    std::span {state.color_info, state.color_count}, state.depth_info, state.vs_input_info, buffer,
+	    std::span {state.color_info, state.color_count}, state.depth_info, vertex_stages, buffer,
 	    state.ps_active ? &state.ps_input_info : nullptr, topology, primitive_restart_enable,
-	    state.programs.vertex, state.programs.pixel);
+	    state.programs);
+	const auto rendering =
+	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info,
+	                         bindings.pixel);
 
 	// Resource preparation above may synchronously finish and restart the scheduler. From this
 	// point onward, every operation targets the current command buffer and cannot touch guest
@@ -1132,13 +1115,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (bindings.pixel && !draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
 	}
-	std::array<PreparedBindings*, 2> descriptor_stages {&bindings.vertex, nullptr};
-	const size_t                     descriptor_stage_count = bindings.pixel.has_value() ? 2u : 1u;
-	if (bindings.pixel) {
-		descriptor_stages[1] = &*bindings.pixel;
-	}
-	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline,
-	               std::span {descriptor_stages.data(), descriptor_stage_count});
+	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline, stages);
 	if (mesh_active) {
 		const uint32_t draw_data[] {
 		    draw.index_count,
@@ -1155,8 +1132,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		CommitIndexBuffer(vk_buffer, index_binding);
 	}
 
-	SetGraphicsDynamicParams(buffer, vk_buffer, state.vs_input_info, state.color_info,
-	                         state.color_count, state.depth_info);
+	SetGraphicsDynamicParams(buffer, vk_buffer, vertex_stages.back(), state.depth_info, rendering);
 	if (m_context.GetGraphics().attachment_feedback_loop_enabled) {
 		vk_buffer.setAttachmentFeedbackLoopEnableEXT(
 		    rendering.depth_stencil_attachment.image_layout ==
@@ -1177,16 +1153,17 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (mesh_active) {
 		vk_buffer.drawMeshTasksEXT(mesh_groups, draw.instance_count, 1);
 	} else {
-		EmitDrawPrimitives(ucfg, vk_buffer, state.vs_input_info, draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, draw, emit);
 	}
 
 	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x600u);
 	}
 	vk::PipelineStageFlags shader_write_stages = {};
-	if (HasShaderBufferWrites(state.vs_input_info.stage)) {
-		shader_write_stages |= mesh_active ? vk::PipelineStageFlagBits::eMeshShaderEXT
-		                                   : vk::PipelineStageFlagBits::eVertexShader;
+	for (const auto& stage: vertex_stages) {
+		if (HasShaderBufferWrites(stage.stage)) {
+			shader_write_stages |= ShaderPipelineStages(NativeShaderStage(stage.logical_stage));
+		}
 	}
 	if (state.ps_active && HasShaderBufferWrites(state.ps_input_info.stage)) {
 		shader_write_stages |= vk::PipelineStageFlagBits::eFragmentShader;
@@ -1220,7 +1197,8 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer) ||
+	    ResolveColorTargets(buffer, args.render_target_slice_offset)) {
 		ResetBindings();
 		return;
 	}
@@ -1251,7 +1229,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	hw_check(buffer);
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, false, topology)) {
+	if (!GetDrawTopology(ucfg, topology)) {
 		return;
 	}
 
@@ -1294,15 +1272,13 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	RefreshShaders(buffer, draw, state);
-
 	LogDrawStateIfNeeded(buffer, draw, state, args.index_type_and_size,
 	                     args.index_addr);
 
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
 	const auto [vertex_offset, instance_offset] =
 	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vs_input_info);
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 
 	DrawEmitInfo emit {};
 	emit.vertex_offset  = vertex_offset + args.base_vertex;
@@ -1332,7 +1308,8 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer) ||
+	    ResolveColorTargets(buffer, args.render_target_slice_offset)) {
 		ResetBindings();
 		return;
 	}
@@ -1361,22 +1338,20 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	const DrawCallInfo draw {CommandBufferDebugOp::DrawIndexAuto,
 	                         args.vertex_count, args.instance_count, args.first_instance};
 
+	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
+	if (!GetDrawTopology(ucfg, topology)) {
+		ResetBindings();
+		return;
+	}
 	DrawRenderState state {};
 	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, state)) {
 		ResetBindings();
 		return;
 	}
 
-	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, true, topology)) {
-		ResetBindings();
-		return;
-	}
-	RefreshShaders(buffer, draw, state);
-
-	const bool rect_list = topology == vk::PrimitiveTopology::ePatchList;
-	if (rect_list && state.vs_input_info.buffers_num == 0 &&
-	    state.vs_input_info.stage.program->param_export_mask == 0 &&
+	const bool rect_list = Prospero::IsRectList(ucfg.GetPrimType());
+	if (rect_list && state.vertex_info[0].buffers_num == 0 &&
+	    state.vertex_info[0].stage.program->param_export_mask == 0 &&
 	    state.ps_input_info.input_num != 0) {
 		if (graphics_debug_dump_enabled()) {
 			LOGF("DrawIndexAuto: skipping rect-list draw with no VS param exports and PS inputs: "
@@ -1393,7 +1368,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
 	const auto [vertex_offset, instance_offset] =
 	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vs_input_info);
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 	DrawEmitInfo emit {};
 	emit.first_vertex = static_cast<uint32_t>(vertex_offset + static_cast<int32_t>(args.first_vertex));
 	emit.first_instance = instance_offset;

src/graphics/host_gpu/renderer/renderTarget.h

@@ -5,7 +5,6 @@
 
 #include <array>
 #include <cstdint>
-#include <type_traits>
 
 namespace Libs::Graphics {
 
@@ -67,33 +66,6 @@ inline constexpr TargetViewInfo ResolveTargetViewInfo(uint32_t base_layer, uint3
 	        base_layer, last_layer - base_layer + 1u, last_layer + 1u};
 }
 
-#pragma pack(push, 1)
-
-struct PipelineStencilStaticState {
-	vk::StencilOp failOp      = vk::StencilOp::eKeep;
-	vk::StencilOp passOp      = vk::StencilOp::eKeep;
-	vk::StencilOp depthFailOp = vk::StencilOp::eKeep;
-	vk::CompareOp compareOp   = vk::CompareOp::eNever;
-};
-
-struct PipelineStencilDynamicState {
-	uint32_t compareMask = 0;
-	uint32_t writeMask   = 0;
-	uint32_t reference   = 0;
-};
-
-#pragma pack(pop)
-
-static_assert(std::is_trivially_copyable_v<PipelineStencilStaticState>);
-static_assert(std::is_standard_layout_v<PipelineStencilStaticState>);
-static_assert(alignof(PipelineStencilStaticState) == 1);
-static_assert(sizeof(PipelineStencilStaticState) ==
-              sizeof(vk::StencilOp) * 3 + sizeof(vk::CompareOp));
-static_assert(std::is_trivially_copyable_v<PipelineStencilDynamicState>);
-static_assert(std::is_standard_layout_v<PipelineStencilDynamicState>);
-static_assert(alignof(PipelineStencilDynamicState) == 1);
-static_assert(sizeof(PipelineStencilDynamicState) == sizeof(uint32_t) * 3);
-
 } // namespace Libs::Graphics
 
 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERTARGET_H_

src/graphics/host_gpu/vulkanCommon.cpp

@@ -123,4 +123,14 @@ void RequireVulkanSuccess(vk::Result result, const char* operation) {
 	}
 }
 
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device) {
+	vk::ShaderModuleCreateInfo create_info {};
+	create_info.codeSize    = code.size_bytes();
+	create_info.pCode       = code.data();
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
+	                     "create SPIR-V shader module");
+	return module;
+}
+
 } // namespace Libs::Graphics

src/graphics/host_gpu/vulkanCommon.h

@@ -15,6 +15,7 @@
 
 #include <cstdint>
 #include <fmt/format.h>
+#include <span>
 #include <string>
 #include <utility>
 #include <vector>
@@ -23,10 +24,9 @@
 
 namespace Libs::Graphics {
 
-using VulkanMemoryBarrier = vk::MemoryBarrier;
-
 vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device);
 
 template <typename Handle, typename... Args>
 void SetVulkanObjectNameF(vk::Device device, Handle handle, fmt::format_string<Args...> format,

src/graphics/presentation/videoOut.cpp

@@ -20,6 +20,7 @@
 #include "kernel/pthread.h"
 #include "libs/errno.h"
 #include "libs/libs.h"
+#include "loader/systemContent.h"
 
 #include <algorithm>
 #include <array>
@@ -237,7 +238,9 @@ private:
 
 struct VideoOutDriver::Impl {
 public:
-	static constexpr int VIDEO_OUT_NUM_MAX = 4;
+	static constexpr std::array VIDEO_OUT_BUSES {
+	    VIDEO_OUT_BUS_TYPE_MAIN, VIDEO_OUT_BUS_TYPE_OVERLAY, VIDEO_OUT_BUS_TYPE_SUB};
+	static constexpr int VIDEO_OUT_NUM_MAX = VIDEO_OUT_BUSES.size() + 1;
 
 	Impl(uint32_t width, uint32_t height, Graphics::Presenter& presenter)
 	    : m_renderer(presenter.Renderer()), m_presenter(presenter), m_flip_queue(presenter) {
@@ -248,7 +251,7 @@ public:
 	~Impl();
 	KYTY_CLASS_NO_COPY(Impl);
 
-	int             Open(int bus_type);
+	int             Open(int bus_type, int index);
 	bool            Close(int handle);
 	VideoOutConfig* Get(int handle);
 	VideoOutConfig* Get(int handle, uint64_t& generation);
@@ -610,12 +613,17 @@ void VideoOutDriver::Impl::Init(uint32_t width, uint32_t height) {
 	}
 }
 
-int VideoOutDriver::Impl::Open(int bus_type) {
+int VideoOutDriver::Impl::Open(int bus_type, int index) {
+	const auto bus = std::find(VIDEO_OUT_BUSES.begin(), VIDEO_OUT_BUSES.end(), bus_type);
+	if (bus == VIDEO_OUT_BUSES.end()) {
+		return VIDEO_OUT_ERROR_INVALID_VALUE;
+	}
+	EXIT_NOT_IMPLEMENTED(index != 0);
 	Common::LockGuard lock(m_mutex);
 
-	const int handle = bus_type + 1;
+	const int handle = static_cast<int>(bus - VIDEO_OUT_BUSES.begin()) + 1;
 	if (m_video_out_ctx[handle].opened) {
-		return -1;
+		return VIDEO_OUT_ERROR_RESOURCE_BUSY;
 	}
 	auto&             config = m_video_out_ctx[handle];
 	Common::LockGuard config_lock(config.mutex);
@@ -1169,21 +1177,9 @@ KYTY_SYSV_ABI int VideoOutOpen(int user_id, int bus_type, int index, const void*
 	PRINT_NAME();
 
 	EXIT_NOT_IMPLEMENTED(user_id != 255 && user_id != 0);
-	if (bus_type != VIDEO_OUT_BUS_TYPE_MAIN && bus_type != VIDEO_OUT_BUS_TYPE_OVERLAY &&
-	    bus_type != VIDEO_OUT_BUS_TYPE_SUB) {
-		return VIDEO_OUT_ERROR_INVALID_VALUE;
-	}
-	EXIT_NOT_IMPLEMENTED(index != 0);
-
 	LOGF("\t param = 0x%016" PRIx64 "\n", reinterpret_cast<uint64_t>(param));
 
-	int handle = DriverState().Open(bus_type);
-
-	if (handle < 0) {
-		return VIDEO_OUT_ERROR_RESOURCE_BUSY;
-	}
-
-	return handle;
+	return DriverState().Open(bus_type, index);
 }
 
 KYTY_SYSV_ABI int VideoOutClose(int handle) {
@@ -1659,8 +1655,12 @@ KYTY_SYSV_ABI int VideoOutGetOutputStatus(int handle, VideoOutOutputStatus* stat
 		return VIDEO_OUT_ERROR_INVALID_HANDLE;
 	}
 
+	int32_t attribute3 = 0;
+	Loader::SystemContentParamSfoGetInt("ATTRIBUTE3", &attribute3);
 	ctx->mutex.Lock();
-	status->resolution   = (ctx->width >= 3840 || ctx->height >= 2160 ? 2u : 1u);
+	// Primary output reports 4K unless param.json Video-out Info enables resolution detection.
+	status->resolution =
+	    ((attribute3 & 4) != 0 && ctx->width < 3840 && ctx->height < 2160 ? 1u : 2u);
 	status->dynamicRange = 1;
 	status->refreshRate =
 	    (ctx->output_mode == VIDEO_OUT_OUTPUT_MODE_119_88HZ || Config::GetVblankFrequency() >= 119

src/graphics/presentation/window/hostInput.cpp

@@ -141,9 +141,8 @@ public:
 				}
 			}
 
-			const bool reserved = binding.key == SDLK_ESCAPE || binding.key == SDLK_SPACE ||
-			                      binding.key == SDLK_F1 || binding.key == SDLK_F7 ||
-			                      binding.key == SDLK_F11;
+			const bool reserved = binding.key == SDLK_ESCAPE || binding.key == SDLK_F1 ||
+			                      binding.key == SDLK_F7 || binding.key == SDLK_F11;
 			if (binding.control == INVALID_CONTROL || reserved ||
 			    (binding.key == SDLK_UNKNOWN && binding.mouse_button == 0)) {
 				EXIT("Invalid input mapping: %s\n", value.c_str());
@@ -220,6 +219,8 @@ uint32_t DefaultKeyboardButton(int key_code) {
 		case SDLK_l: return Controller::PAD_BUTTON_CIRCLE;
 		case SDLK_q: return Controller::PAD_BUTTON_L1;
 		case SDLK_e: return Controller::PAD_BUTTON_R1;
+		case SDLK_z: return Controller::PAD_BUTTON_L2;
+		case SDLK_c: return Controller::PAD_BUTTON_R2;
 		case SDLK_LSHIFT: return Controller::PAD_BUTTON_L3;
 		case SDLK_LCTRL: return Controller::PAD_BUTTON_R3;
 		case SDLK_RETURN:

src/graphics/presentation/window/swapchain.cpp

@@ -22,9 +22,6 @@
 
 // IWYU pragma: no_include <intrin.h>
 
-#define KYTY_ENABLE_DEBUG_PRINTF
-#define KYTY_DBG_INPUT
-
 namespace Libs::Graphics {
 
 struct Presenter::Frame {

src/graphics/presentation/window/vulkanWindow.cpp

@@ -49,7 +49,6 @@
 // IWYU pragma: no_include <intrin.h>
 
 #define KYTY_ENABLE_DEBUG_PRINTF
-#define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
 
@@ -201,6 +200,8 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 		vk::PhysicalDeviceVulkan13Features features13 {};
 
 		vk::PhysicalDeviceColorWriteEnableFeaturesEXT color_write_ext {};
+		vk::PhysicalDeviceImageViewMinLodFeaturesEXT  image_view_min_lod {};
+		color_write_ext.pNext = &image_view_min_lod;
 
 		vk::PhysicalDeviceDepthClipEnableFeaturesEXT depth_clip_enable {};
 		depth_clip_enable.pNext = &color_write_ext;
@@ -235,6 +236,10 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			skip_device = true;
 #endif
 		}
+		if (image_view_min_lod.minLod != VK_TRUE) {
+			LOGF("image view minLod is not supported\n");
+			skip_device = true;
+		}
 
 		if (depth_clip_control.depthClipControl != VK_TRUE) {
 			LOGF("depthClipControl is not supported\n");
@@ -514,11 +519,14 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 	depth_clip_enable.depthClipEnable = VK_TRUE;
 
 	vk::PhysicalDeviceDepthClipControlFeaturesEXT depth_clip_control {};
+	vk::PhysicalDeviceImageViewMinLodFeaturesEXT  image_view_min_lod {};
+	image_view_min_lod.minLod = VK_TRUE;
+	depth_clip_control.pNext  = &image_view_min_lod;
 	// MoltenVK lacks VK_EXT_depth_clip_enable and VK_EXT_color_write_enable, so drop those
 	// feature structs from the chain on macOS (the renderer falls back to default depth
 	// clipping and static color-write masks).
 #if !defined(__APPLE__)
-	depth_clip_control.pNext = &depth_clip_enable;
+	image_view_min_lod.pNext = &depth_clip_enable;
 #endif
 	depth_clip_control.depthClipControl = VK_TRUE;
 
@@ -602,10 +610,11 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 		     graphics.mesh_shader_properties.maxMeshOutputPrimitives,
 		     graphics.mesh_shader_properties.maxMeshSharedMemorySize);
 	}
-	// VulkanFindPhysicalDevice already checked the required creation features. These two
+	// VulkanFindPhysicalDevice already checked the required creation features. These
 	// requirements are specific to this creation path and are not part of device selection.
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderInt64 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.vertexPipelineStoresAndAtomics != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.dualSrcBlend != VK_TRUE);
 	vk::PhysicalDeviceFeatures device_features {};
 	device_features.fragmentStoresAndAtomics = VK_TRUE;
 	device_features.samplerAnisotropy        = VK_TRUE;
@@ -616,6 +625,7 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 	device_features.shaderStorageImageWriteWithoutFormat = VK_TRUE;
 	device_features.shaderImageGatherExtended            = VK_TRUE;
 	device_features.independentBlend                     = VK_TRUE;
+	device_features.dualSrcBlend                         = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
 	device_features.depthBiasClamp                       = VK_TRUE;
@@ -984,7 +994,8 @@ void WindowContext::CreateVulkan() {
 
 	std::vector<const char*> device_extensions = {
 	    VK_KHR_SWAPCHAIN_EXTENSION_NAME, VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME,
-	    VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME, "VK_KHR_maintenance1"};
+	    VK_EXT_IMAGE_VIEW_MIN_LOD_EXTENSION_NAME, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
+	    "VK_KHR_maintenance1"};
 
 #if defined(__APPLE__)
 	// MoltenVK lacks VK_EXT_depth_clip_enable and VK_EXT_color_write_enable; the renderer

src/graphics/presentation/window/window.cpp

@@ -27,6 +27,7 @@
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
+#include "common/stringUtils.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
@@ -45,6 +46,7 @@
 #include <string>
 #include <vector>
 #include <vulkan/vk_platform.h>
+#include <filesystem>
 
 #define STB_IMAGE_IMPLEMENTATION
 #define STBI_NO_SIMD
@@ -52,7 +54,6 @@
 
 // IWYU pragma: no_include <intrin.h>
 
-#define KYTY_ENABLE_DEBUG_PRINTF
 #define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
@@ -180,12 +181,6 @@ struct EventDisplay {
 	DisplayOrientation orientation;
 };
 
-constexpr uint32_t KYTY_SDL_BUTTON_LMASK  = SDL_BUTTON_LMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_MMASK  = SDL_BUTTON_MMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_RMASK  = SDL_BUTTON_RMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_X1MASK = SDL_BUTTON_X1MASK; // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_X2MASK = SDL_BUTTON_X2MASK; // NOLINT(hicpp-signed-bitwise)
-
 namespace {
 
 std::unique_ptr<WindowContext> g_window;
@@ -193,7 +188,6 @@ std::unique_ptr<WindowContext> g_window;
 } // namespace
 
 constexpr const char* KYTY_SDL_WINDOW_CAPTION = "Game";
-constexpr int KYTY_SDL_WINDOWPOS_CENTERED = SDL_WINDOWPOS_CENTERED; /*NOLINT(hicpp-signed-bitwise)*/
 
 static void SetPause(WindowLoopState& game, bool flag) {
 	LOGF("Pause: %s\n", flag ? "true" : "false");
@@ -228,7 +222,7 @@ static void ToggleDesktopFullscreen() {
 	}
 }
 
-static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
+static void GameEventKeyboard(const EventKeyboard& key) {
 	static SDL_Keycode fullscreen_key = SDLK_UNKNOWN;
 
 #ifdef KYTY_DBG_INPUT
@@ -241,7 +235,6 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS || KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 	if (key.down) {
 		switch (key.key_code) {
-			case SDLK_SPACE: SetPause(game, !game.paused.load(std::memory_order_acquire)); break;
 			case SDLK_F1:
 				if (!key.repeat) {
 					RenderDocRequestCapture();
@@ -545,7 +538,7 @@ void WindowContext::ProcessEvent(double time_s) {
 			key.mod               = event->key.keysym.mod;
 			key.timestamp_seconds = time_s;
 
-			GameEventKeyboard(game, key);
+			GameEventKeyboard(key);
 
 			break;
 		}
@@ -614,11 +607,11 @@ void WindowContext::ProcessEvent(double time_s) {
 
 			mb.down              = false;
 			mb.up                = false;
-			mb.left              = ((event->motion.state & KYTY_SDL_BUTTON_LMASK) != 0u);
-			mb.middle            = ((event->motion.state & KYTY_SDL_BUTTON_MMASK) != 0u);
-			mb.right             = ((event->motion.state & KYTY_SDL_BUTTON_RMASK) != 0u);
-			mb.x1                = ((event->motion.state & KYTY_SDL_BUTTON_X1MASK) != 0u);
-			mb.x2                = ((event->motion.state & KYTY_SDL_BUTTON_X2MASK) != 0u);
+			mb.left              = ((event->motion.state & SDL_BUTTON_LMASK) != 0u);
+			mb.middle            = ((event->motion.state & SDL_BUTTON_MMASK) != 0u);
+			mb.right             = ((event->motion.state & SDL_BUTTON_RMASK) != 0u);
+			mb.x1                = ((event->motion.state & SDL_BUTTON_X1MASK) != 0u);
+			mb.x2                = ((event->motion.state & SDL_BUTTON_X2MASK) != 0u);
 			mb.touch             = (event->motion.which == SDL_TOUCH_MOUSEID);
 			mb.pressed           = false;
 			mb.released          = false;
@@ -713,6 +706,20 @@ void WindowContext::ProcessEvent(double time_s) {
 			}
 			break;
 
+		case SDL_CONTROLLERSENSORUPDATE: {
+			const auto& sensor = event->csensor;
+			if (sensor.sensor == SDL_SENSOR_ACCEL || sensor.sensor == SDL_SENSOR_GYRO) {
+				Controller::SetSensor(sensor.which,
+				                      sensor.sensor == SDL_SENSOR_ACCEL ? Controller::Sensor::Accel
+				                                                        : Controller::Sensor::Gyro,
+				                      sensor.data,
+				                      sensor.timestamp_us != 0
+				                          ? sensor.timestamp_us
+				                          : static_cast<uint64_t>(sensor.timestamp) * 1000);
+			}
+			break;
+		}
+
 		case SDL_CONTROLLERDEVICEADDED:
 		case SDL_CONTROLLERDEVICEREMOVED:
 		case SDL_CONTROLLERDEVICEREMAPPED: {
@@ -854,8 +861,8 @@ static void WindowCreate(WindowContext& context) {
 		window_flags |= static_cast<uint32_t>(SDL_WINDOW_BORDERLESS);
 	}
 #endif
-	context.window = SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, KYTY_SDL_WINDOWPOS_CENTERED,
-	                                  KYTY_SDL_WINDOWPOS_CENTERED, width, height, window_flags);
+	context.window = SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, SDL_WINDOWPOS_CENTERED,
+	                                  SDL_WINDOWPOS_CENTERED, width, height, window_flags);
 
 	if (context.window == nullptr) {
 		EXIT("%s\n", SDL_GetError());
@@ -939,10 +946,10 @@ struct WindowIcon {
 	}
 };
 
-static void WindowLoadPngIcon(const std::string& path, WindowIcon* icon) {
+static void WindowLoadPngIcon(const std::filesystem::path& path, WindowIcon* icon) {
 	Common::File f;
 	if (!f.Open(path, Common::File::Mode::Read)) {
-		EXIT("Can't open icon file %s\n", path.c_str());
+		EXIT("Can't open icon file %s\n", Common::PathToString(path).c_str());
 	}
 
 	int width  = 0;
@@ -968,7 +975,7 @@ void WindowContext::UpdateIcon() {
 	static bool       icon_loaded = false;
 
 	if (!icon_loaded) {
-		std::string icon_path;
+		std::filesystem::path icon_path;
 		if (Loader::SystemContentGetIconPath(&icon_path)) {
 			WindowLoadPngIcon(icon_path, &icon);
 		}

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -1,4 +1,5 @@
 #include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/Tessellation.h"
 
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -47,6 +48,9 @@ const char* StageName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Compute: return "CS";
 		case ShaderType::Vertex: return "VS";
+		case ShaderType::Local: return "LS";
+		case ShaderType::TessellationControl: return "HS";
+		case ShaderType::TessellationEvaluation: return "TES";
 		case ShaderType::Mesh: return "MS";
 		case ShaderType::Pixel: return "PS";
 		default: return "unknown";
@@ -83,7 +87,6 @@ struct EmbeddedFetchSgprInfo {
 	EmbeddedFetchValueType type      = EmbeddedFetchValueType::Unknown;
 	int                    attrib_id = 0;
 	uint32_t               value     = 0;
-	std::vector<uint32_t>  prolog_loads;
 };
 
 using EmbeddedFetchVectorLanes = std::map<uint64_t, EmbeddedFetchSgprInfo>;
@@ -102,8 +105,6 @@ void ClearEmbeddedFetchVectorLanes(EmbeddedFetchVectorLanes* lanes, uint32_t reg
 	lanes->erase(first, last);
 }
 
-using EmbeddedFetchData = Frontend::EmbeddedFetchPlan;
-
 bool IsDecodedSgpr(const Decoder::Operand& op) {
 	return op.kind == Decoder::OperandKind::Sgpr || op.kind == Decoder::OperandKind::VccLo ||
 	       op.kind == Decoder::OperandKind::VccHi;
@@ -207,11 +208,12 @@ int BufferTableAttribFromOffset(uint32_t raw_offset, int dword) {
 	return static_cast<int>((raw_offset + static_cast<uint32_t>(dword) * 4u) / 16u);
 }
 
-EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded,
-                                            const ShaderVertexInputInfo* input_info,
-                                            uint32_t user_data_base, uint32_t user_data_count,
-                                            uint32_t wave_size) {
-	EmbeddedFetchData data;
+Frontend::EmbeddedFetchPlan DetectEmbeddedVertexFetch(
+    const Decoder::Program& decoded, const ShaderVertexInputInfo* input_info,
+    uint32_t user_data_base, uint32_t user_data_count, uint32_t wave_size) {
+	const uint32_t    vertex_index_reg   = input_info->logical_stage == ShaderType::Local ? 2u : 5u;
+	const uint32_t    instance_index_reg = input_info->logical_stage == ShaderType::Local ? 5u : 8u;
+	Frontend::EmbeddedFetchPlan data;
 	data.loads.reserve(input_info->resources_num);
 	int32_t vertex_offset_candidate   = -1;
 	int32_t instance_offset_candidate = -1;
@@ -251,16 +253,17 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		// corresponding direct-draw offsets before fetching.
 		const bool vertex_index_accumulator =
 		    IsDecodedVgpr(inst.dst) &&
-		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == 5));
+		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == vertex_index_reg));
 		const bool instance_index_accumulator =
-		    IsDecodedVgpr(inst.dst) && (inst.dst.reg == (user_data_base == 8 ? 8u : 3u));
+		    IsDecodedVgpr(inst.dst) &&
+		    (inst.dst.reg == (user_data_base == 8 ? instance_index_reg : 3u));
 		uint32_t   sad_zero = 0;
 		const bool index_offset_add =
-		    (vertex_index_accumulator || instance_index_accumulator) &&
-		    IsDecodedSgpr(inst.src0) &&
+		    (vertex_index_accumulator || instance_index_accumulator) && IsDecodedSgpr(inst.src0) &&
 		    ((inst.opcode == Decoder::Opcode::V_ADD_I32 && IsDecodedVgpr(inst.src1) &&
 		      inst.src1.reg == inst.dst.reg) ||
-		     (user_data_base == 8 && (inst.dst.reg == 5 || inst.dst.reg == 8) &&
+		     (user_data_base == 8 &&
+		      (inst.dst.reg == vertex_index_reg || inst.dst.reg == instance_index_reg) &&
 		      inst.opcode == Decoder::Opcode::V_SAD_U32 && IsDecodedVgpr(inst.src2) &&
 		      inst.src2.reg == inst.dst.reg &&
 		      TryDecodedOperandConstant(sgprs, inst.src1, sad_zero) && sad_zero == 0));
@@ -319,7 +322,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 						dst.type  = EmbeddedFetchValueType::Constant;
 						dst.value = value;
-						dst.prolog_loads.clear();
 					} else {
 						ClearEmbeddedFetchSgprs(sgprs, inst.dst, 1);
 					}
@@ -330,7 +332,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 					auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 					dst.type  = EmbeddedFetchValueType::Constant;
 					dst.value = inst.src0.value;
-					dst.prolog_loads.clear();
 				}
 				break;
 			default:
@@ -347,7 +348,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								auto& dst        = sgprs[register_id + i];
 								dst.type         = EmbeddedFetchValueType::Attrib;
 								dst.attrib_id    = index + static_cast<int>(i);
-								dst.prolog_loads = {inst.pc};
 							}
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
@@ -365,7 +365,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								dst.type  = EmbeddedFetchValueType::Buffer;
 								dst.attrib_id =
 								    BufferTableAttribFromOffset(raw_offset, static_cast<int>(i));
-								dst.prolog_loads = {inst.pc};
 							}
 						} else if (IsDecodedSgpr(inst.src1) &&
 						           DecodedSgprReg(inst.src1) < sgprs.size() &&
@@ -377,8 +376,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								auto& dst        = sgprs[register_id + i];
 								dst.type         = EmbeddedFetchValueType::Buffer;
 								dst.attrib_id    = sgprs[DecodedSgprReg(inst.src1)].attrib_id;
-								dst.prolog_loads = sgprs[DecodedSgprReg(inst.src1)].prolog_loads;
-								dst.prolog_loads.push_back(inst.pc);
 							}
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
@@ -391,8 +388,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg);
 					}
 					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgpr_is_index.size() &&
-					    IsDecodedVgpr(inst.src0) && inst.src0.reg == 8 &&
-					    IsDecodedVgpr(inst.src1) && inst.src1.reg == 5) {
+					    IsDecodedVgpr(inst.src0) && inst.src0.reg == instance_index_reg &&
+					    IsDecodedVgpr(inst.src1) && inst.src1.reg == vertex_index_reg) {
 						vgpr_is_index[inst.dst.reg] = true;
 					}
 				} else if (IsEmbeddedFetchAttribPropagationAlu(inst)) {
@@ -417,7 +414,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 									break;
 								default: dst.value = src0 + src1; break;
 							}
-							dst.prolog_loads.clear();
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, 1);
 						}
@@ -440,7 +436,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						load.pc           = inst.pc;
 						load.attrib_id    = buffer.attrib_id;
 						load.components   = DecodedDstSize(inst);
-						load.prolog_loads = buffer.prolog_loads;
 					}
 				}
 				break;
@@ -462,22 +457,9 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 Decoder::Program DecodeFusedProgram(std::span<const uint32_t> front, std::span<const uint32_t> back,
                                     std::vector<uint32_t>& joined_code) {
 	EXIT_IF(back.empty());
-	Decoder::Program result;
-	uint32_t         front_words = 0;
-	while (front_words < front.size()) {
-		auto& inst = result.instructions.emplace_back();
-		Decoder::DecodeInstruction(front, front_words, inst);
-		front_words += inst.word_count;
-		if (inst.opcode == Decoder::Opcode::S_SETPC_B64) {
-			EXIT_NOT_IMPLEMENTED(inst.src0.kind != Decoder::OperandKind::Sgpr ||
-			                     inst.src0.reg != 6u);
-			break;
-		}
-		EXIT_NOT_IMPLEMENTED(inst.opcode == Decoder::Opcode::S_ENDPGM);
-	}
-	EXIT_IF(result.instructions.empty() ||
-	        result.instructions.back().opcode != Decoder::Opcode::S_SETPC_B64);
-	joined_code.assign(front.begin(), front.begin() + front_words);
+	auto       result      = Decoder::DecodeFrontProgram(front);
+	const auto front_words = static_cast<uint32_t>(result.code.size());
+	joined_code.assign(result.code.begin(), result.code.end());
 	joined_code.insert(joined_code.end(), back.begin(), back.end());
 	// The merged-stage ABI passes the back shader in s[6:7]. Give that handoff an
 	// ordinary CFG edge, retaining both bodies in one register and LDS lifetime.
@@ -505,7 +487,9 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		EXIT("shader recompiler input is empty\n");
 	}
 	if (options.stage != ShaderType::Compute && options.stage != ShaderType::Vertex &&
-	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh) {
+	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh &&
+	    options.stage != ShaderType::Local && options.stage != ShaderType::TessellationControl &&
+	    options.stage != ShaderType::TessellationEvaluation) {
 		EXIT("shader recompiler received unsupported stage %u\n",
 		     static_cast<unsigned>(options.stage));
 	}
@@ -525,6 +509,12 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 	std::vector<uint32_t> joined_code;
 	if (!options.back_code.empty()) {
 		decoded = DecodeFusedProgram(code, options.back_code, joined_code);
+	} else if (options.stage == ShaderType::Local) {
+		decoded = Decoder::DecodeFrontProgram(code);
+		// The separately compiled hull half runs in the next Vulkan stage.
+		auto& handoff     = decoded.instructions.back();
+		handoff.opcode    = Decoder::Opcode::S_ENDPGM;
+		handoff.src_count = 0;
 	} else {
 		Decoder::DecodeProgram(code, decoded);
 	}
@@ -567,9 +557,9 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		     static_cast<uint64_t>(cfg.natural_loops.size()), phase_ms());
 	}
 
-	EmbeddedFetchData embedded_fetch;
-	if (options.stage == ShaderType::Vertex && options.input_info.vertex != nullptr &&
-	    options.input_info.vertex->fetch_embedded) {
+	Frontend::EmbeddedFetchPlan embedded_fetch;
+	if ((options.stage == ShaderType::Vertex || options.stage == ShaderType::Local) &&
+	    options.input_info.vertex != nullptr && options.input_info.vertex->fetch_embedded) {
 		embedded_fetch = DetectEmbeddedVertexFetch(
 		    decoded, options.input_info.vertex, options.user_data_base,
 		    static_cast<uint32_t>(options.user_data.size()), options.wave_size);
@@ -608,6 +598,7 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		IR::RemoveIdentities(ir.blocks);
 		IR::EliminateDeadCode(ir.blocks);
 	}
+	LowerTessellationMemory(ir, options);
 	IR::BuildSrtPlan(ir);
 	IR::EliminateDeadCode(ir.blocks);
 	IR::TrackResources(ir);

src/graphics/shader/recompiler/Tessellation.cpp

@@ -0,0 +1,330 @@
+#include "graphics/shader/recompiler/Tessellation.h"
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/frontend/decode/ShaderDecoder.h"
+#include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/recompiler/ir/passes/ConstantPropagation.h"
+#include "graphics/shader/recompiler/ir/passes/DeadCodeElimination.h"
+
+#include <algorithm>
+#include <array>
+#include <unordered_map>
+#include <utility>
+
+namespace Libs::Graphics::ShaderRecompiler {
+namespace {
+
+struct TessellationAddress {
+	enum class Kind { Unknown, Affine, PackedControlPoint };
+	Kind     kind        = Kind::Unknown;
+	uint32_t coefficient = 0;
+	uint32_t constant    = 0;
+};
+
+uint32_t ReflectTessellationStride(const Decoder::Program& program, bool local,
+                                   uint32_t control_points, uint32_t input_stride) {
+	using namespace Decoder;
+	using Address = TessellationAddress;
+	using Kind    = Address::Kind;
+	std::array<Address, IR::NumVectorRegs> registers {};
+	registers[local ? 3u : 1u] =
+	    local ? Address {Kind::Affine, 1u, 0u} : Address {Kind::PackedControlPoint};
+	const auto constant = [](uint32_t value) { return Address {Kind::Affine, 0u, value}; };
+	const auto read     = [&](const Operand& operand) -> Address {
+		if (operand.absolute || operand.negate || operand.sdwa_sext || operand.op_sel ||
+		    operand.op_sel_hi || operand.negate_hi || operand.dpp)
+			return {};
+		Address value;
+		if (operand.kind == OperandKind::IntegerInlineConstant ||
+		    operand.kind == OperandKind::LiteralConstant) {
+			value = constant(operand.value);
+		} else if (operand.kind == OperandKind::Vgpr) {
+			value = registers.at(operand.reg);
+		}
+		// The native packed HS ID's low byte is the patch ordinal. The logical
+		// interface addresses one patch, as in the frontend's v1 initialization.
+		if (value.kind == Kind::PackedControlPoint && operand.sdwa_sel == 0u) return constant(0u);
+		return operand.sdwa_sel == 6u ? value : Address {};
+	};
+	const auto affine = [](uint64_t coefficient, uint64_t offset) {
+		return coefficient <= UINT32_MAX && offset <= UINT32_MAX
+		           ? Address {Kind::Affine, static_cast<uint32_t>(coefficient),
+		                      static_cast<uint32_t>(offset)}
+		           : Address {};
+	};
+	const auto add = [&](Address lhs, Address rhs) {
+		return lhs.kind == Kind::Affine && rhs.kind == Kind::Affine
+		           ? affine(uint64_t {lhs.coefficient} + rhs.coefficient,
+		                    uint64_t {lhs.constant} + rhs.constant)
+		           : Address {};
+	};
+	const auto multiply = [&](Address lhs, Address rhs) {
+		if ((lhs.kind == Kind::Affine && lhs.coefficient == 0u && lhs.constant == 0u) ||
+		    (rhs.kind == Kind::Affine && rhs.coefficient == 0u && rhs.constant == 0u))
+			return constant(0u);
+		if (lhs.kind != Kind::Affine || rhs.kind != Kind::Affine) return Address {};
+		if (rhs.coefficient != 0u) std::swap(lhs, rhs);
+		return rhs.coefficient == 0u ? affine(uint64_t {lhs.coefficient} * rhs.constant,
+		                                      uint64_t {lhs.constant} * rhs.constant)
+		                             : Address {};
+	};
+	const auto low24 = [&](Address value) {
+		if (value.kind == Kind::Affine && value.coefficient == 0u)
+			return constant(value.constant & 0xffffffu);
+		return value.kind == Kind::Affine &&
+		               uint64_t {value.coefficient} * (control_points - 1u) + value.constant <=
+		                   0xffffffu
+		           ? value
+		           : Address {};
+	};
+	uint32_t stride = 0;
+	for (const auto& inst: program.instructions) {
+		// Stage exits preserve the active path's definitions. An internal join
+		// would require merging register definitions.
+		EXIT_NOT_IMPLEMENTED(IsDirectBranch(inst.opcode) &&
+		                     inst.branch_target != program.instructions.back().pc);
+		const bool local_store =
+		    inst.opcode == Opcode::DS_WRITE_B32 || inst.opcode == Opcode::DS_WRITE2_B32;
+		const bool buffer_store = inst.opcode == Opcode::BUFFER_STORE_DWORD ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX2 ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX3 ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX4;
+		const bool control_store =
+		    !local && buffer_store && inst.src2.kind == OperandKind::Sgpr && inst.src2.reg == 2u;
+		const bool control_read =
+		    !local && (inst.opcode == Opcode::DS_READ_B32 || inst.opcode == Opcode::DS_READ2_B32);
+		if ((local && local_store) || control_store || control_read) {
+			const auto address = read(inst.src0);
+			if (address.kind != Kind::Affine) {
+				EXIT("%s tessellation address is not affine at pc 0x%08x\n", local ? "LS" : "HS",
+				     inst.pc);
+			}
+			if (address.coefficient != 0u) {
+				const auto expected = control_read ? input_stride : stride;
+				EXIT_NOT_IMPLEMENTED((address.coefficient & 3u) != 0u ||
+				                     (expected != 0u && expected != address.coefficient));
+				if (!control_read) stride = address.coefficient;
+			}
+		}
+		// Store vdata is a source, despite occupying the decoder's dst field.
+		if (local_store || buffer_store || inst.dst.kind != OperandKind::Vgpr) continue;
+		const auto lhs   = read(inst.src0);
+		const auto rhs   = read(inst.src1);
+		const auto third = read(inst.src2);
+		Address    value;
+		switch (inst.opcode) {
+			case Opcode::V_BFE_U32:
+				if (lhs.kind == Kind::PackedControlPoint && rhs.kind == Kind::Affine &&
+				    rhs.coefficient == 0u && third.kind == Kind::Affine &&
+				    third.coefficient == 0u) {
+					if (rhs.constant == 8u && third.constant == 5u) value = {Kind::Affine, 1u, 0u};
+					if (rhs.constant == 0u && third.constant == 8u) value = constant(0u);
+				}
+				break;
+			case Opcode::V_MUL_U32_U24: value = multiply(low24(lhs), low24(rhs)); break;
+			case Opcode::V_MAD_U32_U24: value = add(multiply(low24(lhs), low24(rhs)), third); break;
+			case Opcode::V_LSHL_ADD_U32:
+				if (rhs.kind == Kind::Affine && rhs.coefficient == 0u && rhs.constant < 32u)
+					value = add(multiply(lhs, constant(1u << rhs.constant)), third);
+				break;
+			case Opcode::V_LSHLREV_B32:
+				if (lhs.kind == Kind::Affine && lhs.coefficient == 0u && lhs.constant < 32u)
+					value = multiply(rhs, constant(1u << lhs.constant));
+				break;
+			case Opcode::V_SUB_NC_U32:
+				if (lhs.kind == Kind::Affine && rhs.kind == Kind::Affine &&
+				    lhs.coefficient >= rhs.coefficient && lhs.constant >= rhs.constant)
+					value = affine(lhs.coefficient - rhs.coefficient, lhs.constant - rhs.constant);
+				break;
+			default: break;
+		}
+		if (inst.dst.sdwa_sel != 6u || inst.dst.op_sel || inst.dst.omod || inst.dst.clamp ||
+		    inst.dst.dpp)
+			value = {};
+		for (uint32_t index = 0;
+		     index < std::max(inst.data_dwords, 1u) && inst.dst.reg + index < registers.size(); index++) {
+			registers[inst.dst.reg + index] = {};
+		}
+		registers.at(inst.dst.reg) = value;
+	}
+	EXIT_NOT_IMPLEMENTED(stride == 0u);
+	return stride;
+}
+
+const IR::Inst* TessellationBufferBase(const IR::Inst& inst) {
+	if (IR::BufferAccessOf(inst.GetOpcode()) == IR::BufferAccess::None || inst.NumArgs() <= 3u)
+		return nullptr;
+	const auto* base = inst.Arg(3).Resolve().TryInstruction();
+	return base != nullptr && base->GetOpcode() == IR::ValueOpcode::TessellationBase ? base
+	                                                                                 : nullptr;
+}
+
+IR::Value ActiveAddress(IR::Block& block, IR::Block::iterator before, IR::Value value,
+                        IR::Value predicate, std::unordered_map<IR::Inst*, IR::Value>& resolved) {
+	using namespace IR;
+	value        = value.Resolve();
+	auto* source = value.TryInstruction();
+	if (source == nullptr) return value;
+	if (const auto found = resolved.find(source); found != resolved.end()) return found->second;
+	if (source->GetOpcode() == ValueOpcode::SelectU32) {
+		return source->Arg(0).Resolve() == predicate.Resolve()
+		           ? ActiveAddress(block, before, source->Arg(1), predicate, resolved)
+		           : value;
+	}
+	if (source->GetOpcode() != ValueOpcode::IAdd32 && source->GetOpcode() != ValueOpcode::ISub32)
+		return value;
+	const auto lhs = ActiveAddress(block, before, source->Arg(0), predicate, resolved);
+	const auto rhs = ActiveAddress(block, before, source->Arg(1), predicate, resolved);
+	if (lhs != source->Arg(0).Resolve() || rhs != source->Arg(1).Resolve()) {
+		auto copy = block.PrependNewInst(before, source->GetOpcode(), {lhs, rhs},
+		                                 source->Flags<uint64_t>());
+		value     = Value(&*copy);
+	}
+	resolved.emplace(source, value);
+	return value;
+}
+
+} // namespace
+
+void AnalyzeTessellationPrograms(std::span<const uint32_t> local, std::span<const uint32_t> control,
+                                 ShaderTessellationInputInfo& info) {
+	const auto       local_program = Decoder::DecodeFrontProgram(local);
+	Decoder::Program control_program;
+	Decoder::DecodeProgram(control, control_program);
+	info.ls_stride = ReflectTessellationStride(local_program, true, info.input_control_points, 0u);
+	info.hs_stride = ReflectTessellationStride(control_program, false, info.output_control_points,
+	                                           info.ls_stride);
+	LOGF("Tessellation interface: input_cp=%u output_cp=%u ls_stride=%u hs_stride=%u\n",
+	     info.input_control_points, info.output_control_points, info.ls_stride, info.hs_stride);
+}
+
+void LowerTessellationMemory(IR::Program& program, const CompileOptions& options) {
+	using namespace IR;
+	if (options.stage != ShaderType::Local && options.stage != ShaderType::TessellationControl &&
+	    options.stage != ShaderType::TessellationEvaluation) {
+		return;
+	}
+	const auto& tess = options.input_info.vertex->tess;
+	// Ring addresses can reuse data VGPRs. Their inactive values are irrelevant to
+	// a store guarded by the same EXEC predicate, but must remain intact elsewhere.
+	for (auto* block: program.blocks) {
+		for (auto it = block->begin(); it != block->end(); ++it) {
+			if (TessellationBufferBase(*it) == nullptr) continue;
+			std::unordered_map<Inst*, Value> resolved;
+			it->SetArg(
+			    2, ActiveAddress(*block, it, it->Arg(2), it->Arg(it->NumArgs() - 1u), resolved));
+		}
+	}
+	ConstantPropagationPass(program.blocks);
+	uint32_t reads = 0, writes = 0, factors = 0;
+	for (auto* block: program.blocks) {
+		for (auto it = block->begin(); it != block->end(); ++it) {
+			auto&      inst   = *it;
+			const auto shared = SharedAccessOf(inst.GetOpcode());
+			const auto buffer = BufferAccessOf(inst.GetOpcode());
+			if (shared == SharedAccess::None && buffer == BufferAccess::None) {
+				continue;
+			}
+			const auto&           memory = program.memory_info.at(inst.Flags<MemoryFlags>().index);
+			bool                  write  = false;
+			uint32_t              components = 0;
+			TessellationAttribute kind;
+			Value                 address, predicate;
+			if (shared != SharedAccess::None && memory.kind == ResourceKind::Lds) {
+				write = shared == SharedAccess::Write;
+				EXIT_NOT_IMPLEMENTED(memory.data_bits != 32u ||
+				                     (options.stage == ShaderType::Local
+				                          ? !write
+				                          : options.stage != ShaderType::TessellationControl ||
+				                                shared != SharedAccess::Read));
+				kind       = write ? TessellationAttribute::LocalOutput
+				                   : TessellationAttribute::ControlInput;
+				components = SharedComponentCount(inst.GetOpcode());
+				address    = inst.Arg(0);
+				predicate  = inst.Arg(inst.NumArgs() - 1u);
+			} else if (buffer != BufferAccess::None && memory.kind == ResourceKind::Buffer) {
+				const auto* base = TessellationBufferBase(inst);
+				if (base == nullptr) {
+					continue;
+				}
+				EXIT_NOT_IMPLEMENTED(memory.data_bits != 32u || memory.formatted || memory.idxen ||
+				                     !memory.offen || buffer == BufferAccess::Atomic);
+				write      = buffer == BufferAccess::Write;
+				components = BufferComponentCount(inst.GetOpcode());
+				address    = inst.Arg(2).Resolve();
+				predicate  = inst.Arg(inst.NumArgs() - 1u);
+				if (base->Arg(0).U32() == 1u) {
+					EXIT_NOT_IMPLEMENTED(!write ||
+					                     options.stage != ShaderType::TessellationControl);
+					kind = TessellationAttribute::Factor;
+					factors += components;
+				} else {
+					kind = write ? TessellationAttribute::ControlOutput
+					             : TessellationAttribute::EvaluationInput;
+					EXIT_NOT_IMPLEMENTED(write
+					                         ? options.stage != ShaderType::TessellationControl
+					                         : options.stage != ShaderType::TessellationEvaluation);
+					if (address.IsImmediate() &&
+					    address.U32() >= tess.hs_stride * tess.output_control_points) {
+						EXIT_NOT_IMPLEMENTED(!write);
+						kind = TessellationAttribute::PatchOutput;
+					}
+				}
+			} else {
+				continue;
+			}
+			const auto emit = [&](ValueOpcode opcode, std::initializer_list<Value> args) {
+				return Value(&*block->PrependNewInst(it, opcode, args));
+			};
+			std::array<Value, 4> values;
+			for (uint32_t component = 0; component < components; component++) {
+				const auto offset = memory.offset + 4u * component;
+				const auto byte_address =
+				    offset == 0u ? address : emit(ValueOpcode::IAdd32, {address, Value(offset)});
+				if (write) {
+					Value data = shared != SharedAccess::None ? inst.Arg(component + 1u)
+					             : components == 1u
+					                 ? inst.Arg(4)
+					                 : emit(components == 2u   ? ValueOpcode::CompositeExtractU32x2
+					                        : components == 3u ? ValueOpcode::CompositeExtractU32x3
+					                                           : ValueOpcode::CompositeExtractU32x4,
+					                        {inst.Arg(4), Value(component)});
+					emit(ValueOpcode::SetTessellationAttribute,
+					     {Value(static_cast<uint32_t>(kind)), byte_address, data, predicate});
+					writes++;
+				} else {
+					values[component] =
+					    emit(ValueOpcode::GetTessellationAttribute,
+					         {Value(static_cast<uint32_t>(kind)), byte_address, predicate});
+					reads++;
+				}
+			}
+			if (!write) {
+				Value replacement = values[0];
+				if (components == 2u)
+					replacement =
+					    emit(ValueOpcode::CompositeConstructU32x2, {values[0], values[1]});
+				if (components == 3u)
+					replacement = emit(ValueOpcode::CompositeConstructU32x3,
+					                   {values[0], values[1], values[2]});
+				if (components == 4u)
+					replacement = emit(ValueOpcode::CompositeConstructU32x4,
+					                   {values[0], values[1], values[2], values[3]});
+				inst.ReplaceUsesWith(replacement);
+			}
+			inst.Invalidate();
+		}
+	}
+	ConstantPropagationPass(program.blocks);
+	RemoveIdentities(program.blocks);
+	EliminateDeadCode(program.blocks);
+	LOGF("%s tessellation lowering: reads=%u writes=%u factors=%u\n",
+	     options.stage == ShaderType::Local                 ? "LS"
+	     : options.stage == ShaderType::TessellationControl ? "HS"
+	                                                       : "TES",
+	     reads, writes, factors);
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler

src/graphics/shader/recompiler/Tessellation.h

@@ -0,0 +1,21 @@
+#pragma once
+
+#include <cstdint>
+#include <span>
+
+namespace Libs::Graphics {
+struct ShaderTessellationInputInfo;
+}
+
+namespace Libs::Graphics::ShaderRecompiler {
+
+struct CompileOptions;
+namespace IR {
+struct Program;
+}
+
+void AnalyzeTessellationPrograms(std::span<const uint32_t> local, std::span<const uint32_t> control,
+                                 ShaderTessellationInputInfo& info);
+void LowerTessellationMemory(IR::Program& program, const CompileOptions& options);
+
+} // namespace Libs::Graphics::ShaderRecompiler

src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.cpp

@@ -132,12 +132,23 @@ void ValidateNativeProgram(const IR::Program& program) {
 			              program.memory_info[index].planning_only;
 		       });
 	};
+	const auto indirect_buffer_handle = [&](const IR::Inst& handle) {
+		return program.info.uses_dma && handle.NumArgs() == 4u && !handle.Uses().empty() &&
+		       std::ranges::all_of(handle.Uses(), [&](const IR::Use& use) {
+			       if (IR::BufferAccessOf(use.user->GetOpcode()) != IR::BufferAccess::Read) {
+				       return false;
+			       }
+			       const auto index = use.user->Flags<IR::MemoryFlags>().index;
+			       return index < program.memory_info.size() &&
+			              program.memory_info[index].kind == IR::ResourceKind::IndirectBuffer;
+		       });
+	};
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
 			const auto dense = inst.Flags<uint32_t>();
 			switch (inst.GetOpcode()) {
 				case IR::ValueOpcode::GetBufferResource:
-					if (planning_only_handle(inst)) {
+					if (planning_only_handle(inst) || indirect_buffer_handle(inst)) {
 						break;
 					}
 					if (dense >= program.info.buffers.size()) {
@@ -212,6 +223,9 @@ Emitter::SpirvRequirements Emitter::AnalyzeProgramRequirements(const IR::Program
 					Fail(program, "buffer operation has invalid memory metadata");
 				}
 				const auto& memory = program.memory_info[memory_index];
+				if (memory.kind == IR::ResourceKind::IndirectBuffer) {
+					requirements.subgroup_local_invocation_id = true;
+				}
 				if (memory.kind == IR::ResourceKind::Buffer) {
 					if (memory.resource >= program.info.buffers.size()) {
 						Fail(program, "buffer operation has invalid resource metadata");
@@ -277,7 +291,8 @@ Emitter::SpirvRequirements Emitter::AnalyzeProgramRequirements(const IR::Program
 					break;
 				}
 				case IR::ValueOpcode::LaneId:
-					requirements.subgroup_local_invocation_id = true;
+					requirements.subgroup_local_invocation_id |=
+					    program.stage != ShaderType::TessellationControl;
 					break;
 				case IR::ValueOpcode::ImageQueryLod: requirements.compute_derivatives = true; break;
 				case IR::ValueOpcode::ImageGatherRaw:
@@ -306,8 +321,10 @@ std::vector<uint32_t> EmitProgram(const IR::Program& program,
 	using namespace Emitter;
 
 	if (program.stage != ShaderType::Compute && program.stage != ShaderType::Vertex &&
-	    program.stage != ShaderType::Pixel && program.stage != ShaderType::Mesh) {
-		Fail(program, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
+	    program.stage != ShaderType::Pixel && program.stage != ShaderType::Mesh &&
+	    program.stage != ShaderType::Local && program.stage != ShaderType::TessellationControl &&
+	    program.stage != ShaderType::TessellationEvaluation) {
+		Fail(program, "binary SPIR-V emitter received an unsupported shader stage");
 	}
 	if (!program.srt_plan_complete || !program.resource_tracking_complete ||
 	    !program.shader_info_complete || !program.binding_layout_complete) {

src/graphics/shader/recompiler/backend/spirv/spirvEmitterAnalysis.cpp

@@ -250,7 +250,10 @@ void EmitStorageImageWrite(EmitterState& state, uint32_t resource, uint32_t mip_
 
 spv::ExecutionModel ExecutionModelForStage(ShaderType stage) {
 	switch (stage) {
+		case ShaderType::Local:
 		case ShaderType::Vertex: return spv::ExecutionModelVertex;
+		case ShaderType::TessellationControl: return spv::ExecutionModelTessellationControl;
+		case ShaderType::TessellationEvaluation: return spv::ExecutionModelTessellationEvaluation;
 		case ShaderType::Mesh: return spv::ExecutionModelMeshEXT; // MeshEXT
 		case ShaderType::Pixel: return spv::ExecutionModelFragment;
 		default: return spv::ExecutionModelGLCompute;

src/graphics/shader/recompiler/backend/spirv/spirvEmitterFlow.cpp

@@ -1,6 +1,9 @@
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h"
 
+#include "common/logging/log.h"
+
 #include <algorithm>
+#include <atomic>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 namespace {
@@ -38,7 +41,9 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 		return EmitAddU32(state, local,
 		                  EmitBinaryU32(state, spv::OpIMul, group, ConstantU32(state, size)));
 	}
-	const auto variable = InputVariableForKind(state, kind);
+	const bool centroid = kind == IR::StageInputKind::BaryCoordSmoothCentroid;
+	const auto variable = InputVariableForKind(
+	    state, centroid ? IR::StageInputKind::BaryCoordSmooth : kind);
 	if (variable == 0) {
 		return ConstantU32(state, 0);
 	}
@@ -46,11 +51,13 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 		const auto value = state.builder.AllocateId();
 		const auto bits  = state.builder.AllocateId();
 		state.builder.AddFunction(spv::OpLoad, TypeBool(state), value, variable);
-		state.builder.AddFunction(spv::OpSelect, TypeU32(state), bits, value, ConstantU32(state, 1),
-		                          ConstantU32(state, 0));
+		// PS5 initializes v_front_face with float +1.0/-1.0 bits.
+		state.builder.AddFunction(spv::OpSelect, TypeU32(state), bits, value,
+		                          ConstantU32(state, 0x3f800000u), ConstantU32(state, 0xbf800000u));
 		return bits;
 	}
 	if (kind == IR::StageInputKind::VertexIndex || kind == IR::StageInputKind::InstanceIndex ||
+	    kind == IR::StageInputKind::InvocationId || kind == IR::StageInputKind::PrimitiveId ||
 	    kind == IR::StageInputKind::Layer || kind == IR::StageInputKind::SampleId) {
 		const auto value = state.builder.AllocateId();
 		const auto bits  = state.builder.AllocateId();
@@ -58,7 +65,7 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
-	if (kind == IR::StageInputKind::FragCoord) {
+	if (kind == IR::StageInputKind::FragCoord || kind == IR::StageInputKind::TessCoord) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		const auto bits    = state.builder.AllocateId();
@@ -69,15 +76,24 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
-	if (kind == IR::StageInputKind::BaryCoordSmooth ||
+	if (centroid || kind == IR::StageInputKind::BaryCoordSmooth ||
 	    kind == IR::StageInputKind::BaryCoordNoPerspective) {
-		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		const auto bits    = state.builder.AllocateId();
-		state.builder.AddFunction(spv::OpAccessChain,
-		                          TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-		                          pointer, variable, ConstantU32(state, component + 1u));
-		state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
+		if (centroid) {
+			const auto coordinates = state.builder.AllocateId();
+			state.builder.RequireCapability(spv::CapabilityInterpolationFunction);
+			state.builder.AddFunction(spv::OpExtInst, TypeF32Vector(state, 3), coordinates,
+			                          GlslStd450(state), GLSLstd450InterpolateAtCentroid, variable);
+			state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), value,
+			                          coordinates, component + 1u);
+		} else {
+			const auto pointer = state.builder.AllocateId();
+			state.builder.AddFunction(spv::OpAccessChain,
+			                          TypePointer(state, spv::StorageClassInput, TypeF32(state)),
+			                          pointer, variable, ConstantU32(state, component + 1u));
+			state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
+		}
 		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
@@ -122,7 +138,7 @@ uint32_t EmitDppWriteCondition(ValueEmitContext& ctx, const IR::DppMoveFlags& fl
 	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), masks_ok, bank_ok, row_ok);
 	uint32_t writable = masks_ok;
 	if (!flags.bound_control) {
-		const auto target  = EmitDppTargetLane(state, flags.control);
+		const auto target  = EmitDppTargetLane(state, flags);
 		const auto bounded = state.builder.AllocateId();
 		state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), bounded, writable,
 		                          target.valid);
@@ -138,7 +154,7 @@ uint32_t EmitAttribute(EmitterState& state, uint32_t attr, uint32_t chan) {
 	if (input == nullptr || input->variable_id == 0) {
 		return ConstantU32(state, 0);
 	}
-	if (state.program.stage == ShaderType::Vertex) {
+	if (state.program.stage == ShaderType::Vertex || state.program.stage == ShaderType::Local) {
 		return EmitVertexParameterComponentU32(state, *input, chan & 3u);
 	}
 	const auto load_per_vertex = [&](uint32_t vertex) {
@@ -498,6 +514,31 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 			state.builder.AddFunction(spv::OpStore, MeshOutputPointer(state, kind, exp.index),
 			                          value);
 		} else if (exp.kind == IR::ExportTargetKind::Position) {
+			if (state.invalid_position_clip_distance != UINT32_MAX) {
+				const auto zero = state.builder.Constant(spv::OpConstantNull, TypeF32Vector(state, 4));
+				const auto equal = state.builder.AllocateId();
+				const auto invalid = state.builder.AllocateId();
+				const auto distance = state.builder.AllocateId();
+				const auto distance_pointer = state.builder.AllocateId();
+				state.builder.AddFunction(spv::OpFOrdEqual, TypeBoolVector(state, 4), equal,
+				                          value, zero);
+				state.builder.AddFunction(spv::OpAll, TypeBool(state), invalid, equal);
+				// Zero at valid vertices makes a primitive containing an invalid position
+				// collapse to its remaining edge, before the undefined 0/0 perspective divide.
+				state.builder.AddFunction(spv::OpSelect, TypeF32(state), distance, invalid,
+				                          ConstantF32Value(state, -1.0f),
+				                          ConstantF32Value(state, 0.0f));
+				state.builder.AddFunction(
+				    spv::OpAccessChain, TypePointer(state, spv::StorageClassOutput, TypeF32(state)),
+				    distance_pointer, state.clip_distance_variable,
+				    ConstantU32(state, state.invalid_position_clip_distance));
+				state.builder.AddFunction(spv::OpStore, distance_pointer, distance);
+				static std::atomic_bool logged = false;
+				if (!logged.exchange(true, std::memory_order_relaxed)) {
+					Log::WriteToConsoleAndLog(
+					    "Shader: emitted zero-position clip guard\n");
+				}
+			}
 			const auto pointer = state.builder.AllocateId();
 			state.builder.AddFunction(
 			    spv::OpAccessChain,
@@ -517,11 +558,23 @@ uint32_t EmitIdentity(ValueEmitContext&, uint32_t value) {
 void EmitVoid(ValueEmitContext&) {}
 
 void EmitBarrier(EmitterState& state) {
-	const auto semantics =
-	    spv::MemorySemanticsAcquireReleaseMask | spv::MemorySemanticsWorkgroupMemoryMask;
+	const auto tessellation = state.program.stage == ShaderType::TessellationControl;
+	if (!tessellation && ShaderWorkgroupInput(state.program.stage, state.input_info) == nullptr) {
+		// Independent graphics invocations have no native workgroup left to synchronize.
+		return;
+	}
+	const auto memory_scope = tessellation ? spv::ScopeInvocation : spv::ScopeWorkgroup;
+	const auto semantics    = tessellation ? spv::MemorySemanticsMaskNone
+	                                       : spv::MemorySemanticsAcquireReleaseMask |
+	                                             spv::MemorySemanticsWorkgroupMemoryMask;
 	state.builder.AddFunction(spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
-	                          ConstantU32(state, spv::ScopeWorkgroup),
-	                          ConstantU32(state, semantics));
+	                          ConstantU32(state, memory_scope), ConstantU32(state, semantics));
+}
+
+uint32_t EmitLaneId(EmitterState& state) {
+	return state.program.stage == ShaderType::TessellationControl
+	           ? EmitBuiltinU32(state, IR::StageInputKind::InvocationId, 0)
+	           : EmitSubgroupLocalInvocationId(state);
 }
 
 uint32_t EmitMeshDrawParameter(ValueEmitContext& ctx, const IR::Inst& inst) {
@@ -562,7 +615,7 @@ uint32_t EmitUndefU1(EmitterState& state, const IR::Inst& inst) {
 uint32_t EmitDppMoveU32(ValueEmitContext& ctx, const IR::Inst& inst) {
 	auto&      state    = ctx.state;
 	const auto flags    = inst.Flags<IR::DppMoveFlags>();
-	const auto target   = EmitDppTargetLane(state, flags.control);
+	const auto target   = EmitDppTargetLane(state, flags);
 	const auto shuffled = ctx.Shuffle(inst, 0, target.lane);
 	if (flags.fetch_inactive) {
 		return shuffled;

src/graphics/shader/recompiler/backend/spirv/spirvEmitterHelpers.cpp

@@ -2,24 +2,26 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
-DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control) {
-	const auto quad_base = state.builder.AllocateId();
+DppTargetLane EmitDppPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control,
+                                    uint32_t lane_bits) {
+	const auto lane_mask  = (1u << lane_bits) - 1u;
+	const auto group_base = state.builder.AllocateId();
 	const auto lane      = state.builder.AllocateId();
 	const auto shift     = state.builder.AllocateId();
 	const auto selected0 = state.builder.AllocateId();
 	const auto selected  = state.builder.AllocateId();
 	const auto target    = state.builder.AllocateId();
-	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), quad_base, subid,
-	                          ConstantU32(state, 0xfffffffcu));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), group_base, subid,
+	                          ConstantU32(state, ~lane_mask));
 	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
-	                          ConstantU32(state, 3));
-	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), shift, lane,
-	                          ConstantU32(state, 1));
+	                          ConstantU32(state, lane_mask));
+	state.builder.AddFunction(spv::OpIMul, TypeU32(state), shift, lane,
+	                          ConstantU32(state, lane_bits));
 	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), selected0,
 	                          ConstantU32(state, control), shift);
 	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), selected, selected0,
-	                          ConstantU32(state, 3));
-	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, quad_base, selected);
+	                          ConstantU32(state, lane_mask));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, group_base, selected);
 	return {target, ConstantBool(state, true)};
 }
 
@@ -89,10 +91,14 @@ DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool
 	return {target, ConstantBool(state, true)};
 }
 
-DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control) {
+DppTargetLane EmitDppTargetLane(EmitterState& state, const IR::DppMoveFlags& flags) {
 	const auto subid = EmitSubgroupLocalInvocationId(state);
+	const auto control = flags.control;
+	if (flags.dpp8) {
+		return EmitDppPermTargetLane(state, subid, control, 3u);
+	}
 	if (control <= 0xffu) {
-		return EmitDppQuadPermTargetLane(state, subid, control);
+		return EmitDppPermTargetLane(state, subid, control, 2u);
 	}
 	if (control >= 0x101u && control <= 0x10fu) {
 		return EmitDppRowShiftTargetLane(state, subid, control & 0xfu, true);

src/graphics/shader/recompiler/backend/spirv/spirvEmitterImage.cpp

@@ -3,7 +3,9 @@
 #include "graphics/shader/recompiler/frontend/decode/ImageOps.h"
 
 #include <algorithm>
+#include <atomic>
 #include <bit>
+#include <cstdio>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 namespace {
@@ -646,6 +648,13 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto coord =
 		    CoordF32(ctx, mem, *address, layout.coord, dimension_info.coordinate_components);
 		if (op == IR::ValueOpcode::ImageGatherRaw) {
+			if (HasFlag(mem, Decoder::ImageSampleFlagLod)) {
+				static std::atomic_flag warned = ATOMIC_FLAG_INIT;
+				if (!warned.test_and_set(std::memory_order_relaxed)) {
+					std::fputs("Warning: approximating IMAGE_GATHER4_L at mip level 0; explicit LOD is ignored.\n",
+					           stderr);
+				}
+			}
 			if (dimension == ImageDimension::Dim1D) {
 				if (dref || !HasFlag(mem, Decoder::ImageSampleFlagLevelZero) ||
 				    HasFlag(mem, Decoder::ImageSampleFlagOffset) ||

src/graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h

@@ -155,6 +155,8 @@ inline constexpr auto EmitInstPrefetch = EmitVoid;
 void                  EmitBarrier(EmitterState& state);
 void                  EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitMeshDrawParameter(ValueEmitContext& ctx, const IR::Inst& inst);
+uint32_t              EmitGetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst);
+void                  EmitSetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitGetUserData(EmitterState& state, IR::ScalarReg reg);
 uint32_t              EmitGetBuiltin(ValueEmitContext& ctx, IR::Value kind, IR::Value index);
 uint32_t              EmitUndefU1(EmitterState& state, const IR::Inst& inst);
@@ -165,7 +167,7 @@ inline constexpr auto EmitUndefU64 = EmitUndefU1;
 uint32_t              EmitDppMoveU32(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitDppUpdateU32(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitWqmU64(EmitterState& state, uint32_t value);
-inline constexpr auto EmitLaneId = EmitSubgroupLocalInvocationId;
+uint32_t              EmitLaneId(EmitterState& state);
 uint32_t              EmitBallot(ValueEmitContext& ctx, IR::Value predicate);
 uint32_t              EmitReadFirstLane(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitReadLane(ValueEmitContext& ctx, const IR::Inst& inst);
@@ -271,6 +273,7 @@ inline constexpr auto EmitImageAtomicOr32      = EmitImage;
 inline constexpr auto EmitImageAtomicXor32     = EmitImage;
 void                  EmitUnreachable(ValueEmitContext& ctx, const IR::Inst& inst);
 inline constexpr auto EmitPhi                        = EmitUnreachable;
+inline constexpr auto EmitTessellationBase           = EmitUnreachable;
 inline constexpr auto EmitGetThreadBitScalarRegister = EmitUnreachable;
 inline constexpr auto EmitSetThreadBitScalarRegister = EmitUnreachable;
 inline constexpr auto EmitGetScalarMaskTag           = EmitUnreachable;

src/graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h

@@ -80,6 +80,10 @@ struct EmitterState {
 	Builder                                          builder;
 	const IR::Program&                               program;
 	ShaderStageInputInfo                             input_info;
+	std::array<uint32_t, 6>                          tess_variables {};
+	uint32_t                                         tess_inner_variable = 0;
+	uint32_t                                         tess_patch_base     = 0;
+
 	const SpirvRequirements                          requirements;
 	uint32_t                                         lane_count              = 1;
 	uint32_t                                         lane_half               = 0;
@@ -112,6 +116,7 @@ struct EmitterState {
 	uint32_t                   per_vertex_variable                   = 0;
 	uint32_t                   point_size_variable                   = 0;
 	uint32_t                   clip_distance_variable                = 0;
+	uint32_t                   invalid_position_clip_distance        = UINT32_MAX;
 	uint32_t                   cull_distance_variable                = 0;
 	uint32_t                   layer_variable                        = 0;
 	uint32_t                   viewport_index_variable               = 0;
@@ -303,8 +308,6 @@ uint32_t ConstantI32(EmitterState& state, int32_t value);
 
 uint32_t ConstantF32(EmitterState& state, uint32_t bits);
 
-uint32_t FloatBits(float value);
-
 uint32_t ConstantF32Value(EmitterState& state, float value);
 
 uint32_t ConstantBool(EmitterState& state, bool value);
@@ -313,14 +316,19 @@ uint32_t ConstantU64(EmitterState& state, uint64_t value);
 
 uint32_t ConstantU32CompositeZero(EmitterState& state, uint32_t components);
 
+uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::StorageClass storage,
+                                 const char* name);
 void     DefineModule(EmitterState& state);
+void     DefineTessellationInterfaces(EmitterState& state);
+void     DefineTessellationExecutionModes(EmitterState& state);
 void     DefineMeshOutputs(EmitterState& state);
 void     EmitMeshEntryPoint(EmitterState& state);
 void     EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t MeshOutputPointer(EmitterState& state, IR::StageOutputKind kind, uint32_t index = 0);
 uint32_t MeshPrimitivePointer(EmitterState& state);
 
-DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control);
+DppTargetLane EmitDppPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control,
+                                    uint32_t lane_bits);
 
 DppTargetLane EmitDppRowShiftTargetLane(EmitterState& state, uint32_t subid, uint32_t amount,
                                         bool left);
@@ -329,7 +337,7 @@ DppTargetLane EmitDppRowRotateRightTargetLane(EmitterState& state, uint32_t subi
 
 DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool half_row);
 
-DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control);
+DppTargetLane EmitDppTargetLane(EmitterState& state, const IR::DppMoveFlags& flags);
 
 uint32_t EmitSubgroupLocalInvocationId(EmitterState& state);
 

src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp

@@ -27,61 +27,66 @@ uint32_t EmitDsMaskedLaneRead(EmitterState& state, uint32_t source, uint32_t tar
 	return Select(state, TypeU32(state), source_active, shuffled, ConstantU32(state, 0));
 }
 
-uint32_t BufferByteAddress(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-                           uint32_t index, uint32_t offset, uint32_t soffset) {
-	auto&          state   = ctx.state;
-	const uint32_t packed  = StorageBufferPackedStride(state, mem);
-	const uint32_t stride  = packed & 0x3fffu;
-	const bool     swizzle = stride != 0u && ((packed >> 14u) & 1u) != 0u;
-	if (((packed >> 20u) & 1u) != 0u) {
-		const auto lane = Binary(state, spv::OpBitwiseAnd, TypeU32(state),
-		                         EmitSubgroupLocalInvocationId(state), ConstantU32(state, 63));
-		index           = Binary(state, spv::OpIAdd, TypeU32(state), index, lane);
-	}
-	if (mem.offset != 0u) {
-		offset = Binary(state, spv::OpIAdd, TypeU32(state), offset, ConstantU32(state, mem.offset));
-	}
+struct BufferAddress {
+	uint32_t offset;
+	uint32_t byte;
+};
 
-	uint32_t address = 0;
-	if (!swizzle) {
-		if (stride == 0u) {
-			address = offset;
-		} else {
-			const auto indexed = stride == 1u ? index
-			                                  : Binary(state, spv::OpIMul, TypeU32(state), index,
-			                                           ConstantU32(state, stride));
-			address            = Binary(state, spv::OpIAdd, TypeU32(state), indexed, offset);
-		}
-	} else {
-		const uint32_t stride_enum  = (packed >> 16u) & 3u;
-		const uint32_t index_stride = 8u << stride_enum;
-		const auto     index_msb    = Binary(state, spv::OpShiftRightLogical, TypeU32(state), index,
-		                                     ConstantU32(state, stride_enum + 3u));
-		const auto     index_lsb    = Binary(state, spv::OpBitwiseAnd, TypeU32(state), index,
-		                                     ConstantU32(state, index_stride - 1u));
-		const auto     offset_msb =
+BufferAddress CalculateBufferAddress(EmitterState& state, uint32_t index, uint32_t offset,
+                                     uint32_t soffset, uint32_t immediate, uint32_t stride,
+                                     uint32_t swizzle, uint32_t index_stride) {
+	if (immediate != 0u) {
+		offset = Binary(state, spv::OpIAdd, TypeU32(state), offset, ConstantU32(state, immediate));
+	}
+	const auto indexed = Binary(state, spv::OpIMul, TypeU32(state), index, stride);
+	auto       address = Binary(state, spv::OpIAdd, TypeU32(state), indexed, offset);
+	if (swizzle != 0u) {
+		const auto index_shift =
+		    Binary(state, spv::OpIAdd, TypeU32(state), index_stride, ConstantU32(state, 3));
+		const auto indices = Binary(state, spv::OpShiftLeftLogical, TypeU32(state),
+		                            ConstantU32(state, 1), index_shift);
+		const auto index_msb =
+		    Binary(state, spv::OpShiftRightLogical, TypeU32(state), index, index_shift);
+		const auto index_lsb =
+		    Binary(state, spv::OpBitwiseAnd, TypeU32(state), index,
+		           Binary(state, spv::OpISub, TypeU32(state), indices, ConstantU32(state, 1)));
+		const auto offset_msb =
 		    Binary(state, spv::OpBitwiseAnd, TypeU32(state), offset, ConstantU32(state, ~3u));
 		const auto offset_lsb =
 		    Binary(state, spv::OpBitwiseAnd, TypeU32(state), offset, ConstantU32(state, 3u));
-		const auto indexed_msb = stride == 1u ? index_msb
-		                                      : Binary(state, spv::OpIMul, TypeU32(state),
-		                                               index_msb, ConstantU32(state, stride));
-		const auto msb = Binary(state, spv::OpIMul, TypeU32(state),
-		                        Binary(state, spv::OpIAdd, TypeU32(state), indexed_msb, offset_msb),
-		                        ConstantU32(state, index_stride));
+		const auto indexed_msb = Binary(state, spv::OpIMul, TypeU32(state), index_msb, stride);
+		const auto msb =
+		    Binary(state, spv::OpIMul, TypeU32(state),
+		           Binary(state, spv::OpIAdd, TypeU32(state), indexed_msb, offset_msb), indices);
 		const auto lsb = Binary(state, spv::OpIAdd, TypeU32(state),
 		                        Binary(state, spv::OpShiftLeftLogical, TypeU32(state), index_lsb,
 		                               ConstantU32(state, 2u)),
 		                        offset_lsb);
-		address        = Binary(state, spv::OpIAdd, TypeU32(state), msb, lsb);
+		address        = Select(state, TypeU32(state), swizzle,
+		                        Binary(state, spv::OpIAdd, TypeU32(state), msb, lsb), address);
 	}
+	return {offset, Binary(state, spv::OpIAdd, TypeU32(state), address, soffset)};
+}
 
-	const auto soffset_value = inst.Arg(3).Resolve();
-	if (soffset_value.IsImmediate() && soffset_value.GetType() == IR::Type::U32 &&
-	    soffset_value.U32() == 0u) {
-		return address;
+uint32_t BufferLane(EmitterState& state) {
+	return Binary(state, spv::OpBitwiseAnd, TypeU32(state), EmitSubgroupLocalInvocationId(state),
+	              ConstantU32(state, 63));
+}
+
+uint32_t BufferByteAddress(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
+	auto&      state  = ctx.state;
+	const auto packed = StorageBufferPackedStride(state, mem);
+	const auto stride = packed & 0x3fffu;
+	auto       index  = ctx.Arg(inst, 1);
+	if ((packed & (1u << 20u)) != 0u) {
+		index = Binary(state, spv::OpIAdd, TypeU32(state), index, BufferLane(state));
 	}
-	return Binary(state, spv::OpIAdd, TypeU32(state), address, soffset);
+	const bool swizzle = stride != 0u && (packed & (1u << 14u)) != 0u;
+	return CalculateBufferAddress(state, index, ctx.Arg(inst, 2), ctx.Arg(inst, 3), mem.offset,
+	                              ConstantU32(state, stride),
+	                              swizzle ? ConstantBool(state, true) : 0u,
+	                              ConstantU32(state, (packed >> 16u) & 3u))
+	    .byte;
 }
 
 uint32_t AddU64Low(EmitterState& state, uint32_t low, uint32_t high, uint32_t add_low,
@@ -196,11 +201,8 @@ uint32_t LoadBdaDword(ValueEmitContext& ctx, uint32_t address) {
 	});
 }
 
-uint32_t LoadBda(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-	             uint32_t bits) {
-	auto&      state   = ctx.state;
-	const auto address = GuestAddress(ctx, inst, mem);
-	const auto active  = ctx.Arg(inst, inst.NumArgs() - 1);
+uint32_t LoadBda(ValueEmitContext& ctx, uint32_t address, uint32_t active, uint32_t bits) {
+	auto& state = ctx.state;
 	return EmitValueOrZeroIfCondition(state, active, [&]() {
 		const auto aligned = Binary(state, spv::OpBitwiseAnd, TypeScalarU64(state), address,
 		                            ConstantDeviceAddress(state, ~uint64_t {3}));
@@ -236,8 +238,7 @@ uint32_t LoadBda(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryIn
 
 uint32_t ByteAddress(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
 	if (mem.kind == IR::ResourceKind::Buffer) {
-		return BufferByteAddress(ctx, inst, mem, ctx.Arg(inst, 1), ctx.Arg(inst, 2),
-		                         ctx.Arg(inst, 3));
+		return BufferByteAddress(ctx, inst, mem);
 	}
 	if (mem.kind == IR::ResourceKind::Lds || mem.kind == IR::ResourceKind::Gds) {
 		if (mem.offset == 0u) {
@@ -514,31 +515,6 @@ void StoreWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem)
 	});
 }
 
-void FormattedStorePrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-                            uint32_t component, const MemoryResourceAccess& resource,
-                            uint32_t data) {
-	const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
-	if (info.type == Format::ComponentType::Unknown) {
-		StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, component), resource, data);
-		return;
-	}
-	if (component >= info.component_count) return;
-	const auto bits          = info.component_bits[component];
-	const auto component_mem = RebaseFormattedComponent(mem, info, component);
-	if (bits == 8u || bits == 16u) {
-		StoreSubwordPrepared(ctx, inst, component_mem, resource, bits, data);
-	} else {
-		StoreWordPrepared(ctx, inst, component_mem, resource, data);
-	}
-}
-
-void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
-	EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
-		const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
-		FormattedStorePrepared(ctx, inst, mem, 0u, resource, ctx.Arg(inst, inst.NumArgs() - 2));
-	});
-}
-
 spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
 		case IR::ValueOpcode::BufferAtomicCmpSwap32: return spv::OpAtomicCompareExchange;
@@ -729,6 +705,11 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                             uint32_t data) {
 	if (component >= plan.info.component_count) return;
 	const auto bits = plan.info.component_bits[component];
+	if (plan.info.type == Format::ComponentType::Snorm && bits == 16u) {
+		const auto value = EmitBitCastF32U32(ctx.state, data);
+		data = EmitPackSnorm2x16(
+		    ctx.state, EmitCompositeConstructF32x2(ctx.state, value, ConstantF32Value(ctx.state, 0.0f)));
+	}
 	if (bits == 8u || bits == 16u) {
 		StoreSubwordInBounds(ctx, mem, plan.resource, plan.addresses[component],
 		                     plan.indices[component], bits, data);
@@ -737,12 +718,103 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
 	}
 }
 
+void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
+	EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
+		const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
+		const auto data     = ctx.Arg(inst, inst.NumArgs() - 2);
+		const auto info     = Format::GetFormatInfo(BufferFormat(ctx, mem));
+		if (info.type == Format::ComponentType::Unknown) {
+			StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, 0u), resource, data);
+			return;
+		}
+		const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info, 1u,
+		                                         FormattedAccess::Store);
+		EmitIfCondition(ctx.state, plan.in_bounds, [&]() {
+			StoreFormattedInBounds(ctx, mem, plan, 0u, data);
+		});
+	});
+}
+
+uint32_t LoadIndirectBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t components) {
+	auto&       state   = ctx.state;
+	const auto& handle  = *inst.Arg(0).ResolveInstruction();
+	const auto  word1   = ctx.Arg(handle, 1);
+	const auto  records = ctx.Arg(handle, 2);
+	const auto  word3   = ctx.Arg(handle, 3);
+	const auto  field   = [&](uint32_t word, uint32_t first, uint32_t count) {
+		return EmitBitFieldUExtract(state, word, ConstantU32(state, first),
+		                            ConstantU32(state, count));
+	};
+	const auto nonzero = [&](uint32_t value) {
+		return Binary(state, spv::OpINotEqual, TypeBool(state), value, ConstantU32(state, 0));
+	};
+	const auto has_dword = [&](uint32_t offset, uint32_t size) {
+		return AndCondition(
+		    state,
+		    Binary(state, spv::OpUGreaterThanEqual, TypeBool(state), size, ConstantU32(state, 4)),
+		    Binary(state, spv::OpULessThanEqual, TypeBool(state), offset,
+		           Binary(state, spv::OpISub, TypeU32(state), size, ConstantU32(state, 4))));
+	};
+	const auto stride       = field(word1, 16, 14);
+	const auto swizzle      = AndCondition(state, nonzero(stride), nonzero(field(word1, 31, 1)));
+	const auto index_stride = field(word3, 21, 2);
+	const auto add_tid      = nonzero(field(word3, 23, 1));
+	const auto index =
+	    Binary(state, spv::OpIAdd, TypeU32(state), ctx.Arg(inst, 1),
+	           Select(state, TypeU32(state), add_tid, BufferLane(state), ConstantU32(state, 0)));
+	const auto soffset = ctx.Arg(inst, 3);
+	const auto base    = DeviceAddressFromWords(state, ctx.Arg(handle, 0), field(word1, 0, 16));
+	const auto valid_format    = nonzero(field(word3, 12, 7));
+	const auto mode            = field(word3, 28, 2);
+	const auto index_in_bounds = Binary(state, spv::OpULessThan, TypeBool(state), index, records);
+	const auto scalar_in_bounds =
+	    Binary(state, spv::OpULessThanEqual, TypeBool(state), soffset, records);
+	const auto raw_records = Binary(state, spv::OpISub, TypeU32(state), records, soffset);
+	const auto raw_index_in_bounds =
+	    Binary(state, spv::OpULessThan, TypeBool(state), index, raw_records);
+	std::array<uint32_t, 4> values {};
+	for (uint32_t component = 0; component < components; component++) {
+		const auto address = CalculateBufferAddress(state, index, ctx.Arg(inst, 2), soffset,
+		                                            ctx.Memory(inst).offset + component * 4u,
+		                                            stride, swizzle, index_stride);
+		// RDNA2 raw DWORD vectors check each component, unlike formatted accesses.
+		const auto structured_bounds =
+		    AndCondition(state, index_in_bounds,
+		                 Binary(state, spv::OpULessThan, TypeBool(state), address.offset, stride));
+		// OOB_SELECT=3 reduces NUM_RECORDS by SOFFSET before the offset/index check.
+		const auto raw_bounds = AndCondition(
+		    state, scalar_in_bounds,
+		    Select(state, TypeBool(state), swizzle,
+		           AndCondition(state, raw_index_in_bounds, has_dword(address.offset, stride)),
+		           has_dword(address.offset, raw_records)));
+		auto in_bounds =
+		    Select(state, TypeBool(state),
+		           Binary(state, spv::OpIEqual, TypeBool(state), mode, ConstantU32(state, 0)),
+		           structured_bounds, index_in_bounds);
+		in_bounds = Select(
+		    state, TypeBool(state),
+		    Binary(state, spv::OpULessThan, TypeBool(state), mode, ConstantU32(state, 2)),
+		    in_bounds,
+		    Select(state, TypeBool(state),
+		           Binary(state, spv::OpIEqual, TypeBool(state), mode, ConstantU32(state, 2)),
+		           nonzero(records), raw_bounds));
+		const auto guest =
+		    Binary(state, spv::OpIAdd, TypeScalarU64(state), base,
+		           Unary(state, spv::OpUConvert, TypeScalarU64(state), address.byte));
+		values[component] = LoadBda(ctx, guest, AndCondition(state, valid_format, in_bounds), 32u);
+	}
+	return ConstructU32Composite(state, components, values);
+}
+
 uint32_t LoadWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t components) {
 	auto& state = ctx.state;
 	return EmitValueOrDefaultIfCondition(
 	    state, ctx.Arg(inst, inst.NumArgs() - 1), TypeU32Composite(state, components),
 	    ConstantU32CompositeZero(state, components), [&]() {
 		    const auto mem      = ctx.Memory(inst);
+		    if (mem.kind == IR::ResourceKind::IndirectBuffer) {
+			    return LoadIndirectBuffer(ctx, inst, components);
+		    }
 		    const auto resource = PrepareMemoryResourceAccess(state, mem);
 		    const auto info = Format::GetFormatInfo(
 		        mem.formatted ? BufferFormat(ctx, mem) : Prospero::BufferFormat::kInvalid);
@@ -1091,7 +1163,8 @@ void EmitLoadMemory(ValueEmitContext& ctx, const IR::Inst& inst) {
 		value = LoadWideShared(ctx, inst, shared_components);
 	else if (address_info.access == IR::AddressAccess::Read &&
 	         mem.kind != IR::ResourceKind::Scratch)
-		value = LoadBda(ctx, inst, mem, address_info.data_bits);
+		value = LoadBda(ctx, GuestAddress(ctx, inst, mem), ctx.Arg(inst, inst.NumArgs() - 1),
+		                address_info.data_bits);
 	else if (op == IR::ValueOpcode::LoadBufferU32 && mem.formatted)
 		value = FormattedLoad(ctx, inst, mem);
 	else if (inst.GetType() == IR::Type::U8)

src/graphics/shader/recompiler/backend/spirv/spirvEmitterMesh.cpp

@@ -1,4 +1,4 @@
-#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h"
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 namespace {
@@ -127,10 +127,7 @@ void EmitMeshEntryPoint(EmitterState& state) {
 	state.builder.AddFunction(spv::OpFunctionCall, TypeVoid(state), state.builder.AllocateId(),
 	                          state.mesh_guest_func);
 	// All guest waves finish before the uniform Vulkan allocation and output stores.
-	state.builder.AddFunction(spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
-	                          ConstantU32(state, spv::ScopeWorkgroup),
-	                          ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask |
-	                                                 spv::MemorySemanticsWorkgroupMemoryMask));
+	EmitBarrier(state);
 	const auto vertices   = MeshLoad(state, state.mesh_allocation, spv::StorageClassWorkgroup,
 	                                 TypeU32(state), ConstantU32(state, 0));
 	const auto primitives = MeshLoad(state, state.mesh_allocation, spv::StorageClassWorkgroup,

src/graphics/shader/recompiler/backend/spirv/spirvEmitterModule.cpp

@@ -1,6 +1,7 @@
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
 
 #include <algorithm>
+#include <bit>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
@@ -256,14 +257,8 @@ uint32_t ConstantF32(EmitterState& state, uint32_t bits) {
 	return state.builder.Constant(spv::OpConstant, TypeF32(state), bits);
 }
 
-uint32_t FloatBits(float value) {
-	uint32_t bits = 0;
-	std::memcpy(&bits, &value, sizeof(bits));
-	return bits;
-}
-
 uint32_t ConstantF32Value(EmitterState& state, float value) {
-	return ConstantF32(state, FloatBits(value));
+	return ConstantF32(state, std::bit_cast<uint32_t>(value));
 }
 
 uint32_t ConstantBool(EmitterState& state, bool value) {
@@ -290,7 +285,8 @@ uint32_t GlslStd450(EmitterState& state) {
 }
 
 VertexInputScalarKind VertexParameterScalarKind(const EmitterState& state, uint32_t location) {
-	if (state.program.stage != ShaderType::Vertex || location >= ShaderVertexInputInfo::RES_MAX ||
+	if ((state.program.stage != ShaderType::Vertex && state.program.stage != ShaderType::Local) ||
+	    location >= ShaderVertexInputInfo::RES_MAX ||
 	    location >= static_cast<uint32_t>(state.input_info.vertex->resources_num)) {
 		return VertexInputScalarKind::Float;
 	}
@@ -333,8 +329,6 @@ uint32_t VertexParameterScalarType(EmitterState& state, VertexInputScalarKind ki
 	}
 }
 
-namespace {
-
 uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::StorageClass storage,
                                  const char* name) {
 	const auto variable =
@@ -344,9 +338,14 @@ uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::Storag
 	return variable;
 }
 
+namespace {
+
 uint32_t BuiltInForInput(IR::StageInputKind kind) {
 	switch (kind) {
 		case IR::StageInputKind::VertexIndex: return spv::BuiltInVertexIndex;
+		case IR::StageInputKind::InvocationId: return spv::BuiltInInvocationId;
+		case IR::StageInputKind::PrimitiveId: return spv::BuiltInPrimitiveId;
+		case IR::StageInputKind::TessCoord: return spv::BuiltInTessCoord;
 		case IR::StageInputKind::InstanceIndex: return spv::BuiltInInstanceIndex;
 		case IR::StageInputKind::FragCoord: return spv::BuiltInFragCoord;
 		case IR::StageInputKind::FrontFacing: return spv::BuiltInFrontFacing;
@@ -384,9 +383,24 @@ void DefineInputs(EmitterState& state) {
 		}
 	}
 	for (auto& input: state.inputs) {
+		if (state.program.stage == ShaderType::Pixel &&
+		    input.kind == IR::StageInputKind::Parameter) {
+			const auto location = PixelParameterLocation(state, input.location);
+			const auto alias = std::ranges::find_if(state.inputs, [&](const InputBinding& other) {
+				return other.kind == IR::StageInputKind::Parameter && other.variable_id != 0 &&
+				       PixelParameterLocation(state, other.location) == location;
+			});
+			if (alias != state.inputs.end()) {
+				EXIT_IF(alias->per_vertex != input.per_vertex);
+				input.variable_id = alias->variable_id;
+				continue;
+			}
+		}
 		uint32_t type = TypeU32(state);
 		switch (input.kind) {
 			case IR::StageInputKind::VertexIndex:
+			case IR::StageInputKind::InvocationId:
+			case IR::StageInputKind::PrimitiveId:
 			case IR::StageInputKind::InstanceIndex:
 			case IR::StageInputKind::Layer:
 			case IR::StageInputKind::SampleId: type = TypeI32(state); break;
@@ -394,11 +408,13 @@ void DefineInputs(EmitterState& state) {
 			case IR::StageInputKind::LocalInvocationId:
 			case IR::StageInputKind::GlobalInvocationId: type = TypeU32Vector(state, 3); break;
 			case IR::StageInputKind::FragCoord: type = TypeF32Vector(state, 4); break;
+			case IR::StageInputKind::TessCoord:
 			case IR::StageInputKind::BaryCoordSmooth:
 			case IR::StageInputKind::BaryCoordNoPerspective: type = TypeF32Vector(state, 3); break;
 			case IR::StageInputKind::FrontFacing: type = TypeBool(state); break;
 			case IR::StageInputKind::Parameter:
-				if (state.program.stage == ShaderType::Vertex) {
+				if (state.program.stage == ShaderType::Vertex ||
+				    state.program.stage == ShaderType::Local) {
 					type = VertexParameterScalarType(
 					    state, VertexParameterScalarKind(state, input.location));
 					const auto components = VertexParameterComponentCount(input);
@@ -468,6 +484,15 @@ void DefineOutputs(EmitterState& state) {
 		DefineMeshOutputs(state);
 		return;
 	}
+	if (state.program.stage == ShaderType::Vertex && clip_distance_count + cull_distance_count < 8u &&
+	    std::ranges::any_of(state.outputs, [](const OutputBinding& output) {
+		    return output.kind == IR::StageOutputKind::Position;
+	    })) {
+		// Reserve one plane for the enabled PA_CL_CLIP_CNTL clipping-error cull.
+		state.invalid_position_clip_distance = clip_distance_count++;
+		state.outputs.push_back({{IR::StageOutputKind::ClipDistance,
+		                          state.invalid_position_clip_distance, 0, "gl_ClipDistance"}});
+	}
 	const auto BuiltIn = [&](uint32_t& variable, uint32_t type, const char* name,
 	                         spv::BuiltIn builtin) {
 		if (variable == 0) {
@@ -528,8 +553,17 @@ void DefineOutputs(EmitterState& state) {
 				const auto type = uint_output ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4);
 				binding.variable_id = DefineInterfaceVariable(state, type, spv::StorageClassOutput,
 				                                              binding.debug_name.c_str());
+				const bool dual_source = binding.kind == IR::StageOutputKind::Mrt &&
+				                         state.program.stage == ShaderType::Pixel &&
+				                         state.input_info.pixel->dual_source_blending;
+				EXIT_NOT_IMPLEMENTED(dual_source && binding.index > 1);
 				state.builder.AddAnnotation(spv::OpDecorate, binding.variable_id,
-				                            spv::DecorationLocation, binding.location);
+				                            spv::DecorationLocation,
+				                            dual_source ? 0u : binding.location);
+				if (dual_source) {
+					state.builder.AddAnnotation(spv::OpDecorate, binding.variable_id,
+					                            spv::DecorationIndex, binding.index);
+				}
 				break;
 			}
 		}
@@ -543,6 +577,7 @@ void DefineModule(EmitterState& state) {
 	                                  state.program.info.outputs.size());
 	DefineInputs(state);
 	DefineOutputs(state);
+	DefineTessellationInterfaces(state);
 	DefineDescriptors(state);
 	if (state.requirements.function_lds) {
 		state.lds_variable = state.builder.AllocateId();
@@ -564,6 +599,10 @@ void DefineModule(EmitterState& state) {
 		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOutputPrimitivesEXT,
 		                               state.input_info.vertex->mesh.max_primitives);
 	}
+	if (state.program.stage == ShaderType::TessellationControl ||
+	    state.program.stage == ShaderType::TessellationEvaluation) {
+		DefineTessellationExecutionModes(state);
+	}
 	state.entry_label = state.builder.AllocateId();
 
 	state.builder.RequireCapability(spv::CapabilityShader);

src/graphics/shader/recompiler/backend/spirv/spirvEmitterTessellation.cpp

@@ -0,0 +1,161 @@
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h"
+
+#include <algorithm>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+namespace {
+
+uint32_t TessellationPointer(ValueEmitContext& ctx, const IR::Inst& inst) {
+	auto& state          = ctx.state;
+	using Attribute      = IR::TessellationAttribute;
+	const auto  kind     = static_cast<Attribute>(inst.Arg(0).U32());
+	const auto& tess     = state.input_info.vertex->tess;
+	const auto  variable = state.tess_variables.at(static_cast<uint32_t>(kind));
+	EXIT_IF(variable == 0);
+	const auto input   = kind == Attribute::ControlInput || kind == Attribute::EvaluationInput;
+	const auto storage = input ? spv::StorageClassInput : spv::StorageClassOutput;
+	const auto pointer = state.builder.AllocateId();
+	if (kind == Attribute::Factor) {
+		EXIT_NOT_IMPLEMENTED(!inst.Arg(1).IsImmediate());
+		const auto index = inst.Arg(1).U32() / 4u;
+		const auto outer = index < 3u;
+		EXIT_NOT_IMPLEMENTED(index >= 4u);
+		state.builder.AddFunction(spv::OpAccessChain, TypePointer(state, storage, TypeF32(state)),
+		                          pointer, outer ? variable : state.tess_inner_variable,
+		                          ConstantU32(state, outer ? index : index - 3u));
+		return pointer;
+	}
+	auto address = ctx.Arg(inst, 1);
+	if (kind == Attribute::PatchOutput) {
+		address =
+		    EmitBinaryU32(state, spv::OpISub, address, ConstantU32(state, state.tess_patch_base));
+	}
+	const bool local  = kind == Attribute::LocalOutput || kind == Attribute::ControlInput;
+	const auto stride = local ? tess.ls_stride : tess.hs_stride;
+	const auto offset = kind == Attribute::PatchOutput ? address
+	                                                   : EmitBinaryU32(state, spv::OpUMod, address,
+	                                                                   ConstantU32(state, stride));
+	const auto attribute =
+	    EmitBinaryU32(state, spv::OpShiftRightLogical, offset, ConstantU32(state, 4));
+	const auto component =
+	    EmitBinaryU32(state, spv::OpBitwiseAnd,
+	                  EmitBinaryU32(state, spv::OpShiftRightLogical, offset, ConstantU32(state, 2)),
+	                  ConstantU32(state, 3));
+	const auto type = TypePointer(state, storage, TypeU32(state));
+	if (kind == Attribute::LocalOutput || kind == Attribute::PatchOutput) {
+		state.builder.AddFunction(spv::OpAccessChain, type, pointer, variable, attribute,
+		                          component);
+	} else {
+		const auto vertex =
+		    kind == Attribute::ControlOutput
+		        ? EmitLaneId(state)
+		        : EmitBinaryU32(state, spv::OpUDiv, address, ConstantU32(state, stride));
+		state.builder.AddFunction(spv::OpAccessChain, type, pointer, variable, vertex, attribute,
+		                          component);
+	}
+	return pointer;
+}
+
+} // namespace
+
+void DefineTessellationInterfaces(EmitterState& state) {
+	using Attribute = IR::TessellationAttribute;
+	std::array<bool, 6> used {};
+	uint32_t            patch_begin = UINT32_MAX, patch_end = 0;
+	for (const auto* block: state.program.blocks) {
+		for (const auto& inst: *block) {
+			if (inst.GetOpcode() != IR::ValueOpcode::GetTessellationAttribute &&
+			    inst.GetOpcode() != IR::ValueOpcode::SetTessellationAttribute)
+				continue;
+			const auto kind = inst.Arg(0).U32();
+			used.at(kind)   = true;
+			if (kind == static_cast<uint32_t>(Attribute::PatchOutput)) {
+				EXIT_NOT_IMPLEMENTED(!inst.Arg(1).IsImmediate());
+				patch_begin = std::min(patch_begin, inst.Arg(1).U32());
+				patch_end   = std::max(patch_end, inst.Arg(1).U32() + 4u);
+			}
+		}
+	}
+	if (std::ranges::none_of(used, [](bool value) { return value; })) return;
+	const auto& tess  = state.input_info.vertex->tess;
+	const auto  array = [&](uint32_t type, uint32_t count) {
+		return state.builder.Type(spv::OpTypeArray, type, ConstantU32(state, count));
+	};
+	for (uint32_t index = 0; index < used.size(); index++) {
+		if (!used[index]) continue;
+		const auto kind    = static_cast<Attribute>(index);
+		const bool input   = kind == Attribute::ControlInput || kind == Attribute::EvaluationInput;
+		const auto storage = input ? spv::StorageClassInput : spv::StorageClassOutput;
+		uint32_t   type;
+		if (kind == Attribute::Factor) {
+			type = array(TypeF32(state), 4u);
+		} else if (kind == Attribute::PatchOutput) {
+			state.tess_patch_base = patch_begin;
+			type = array(TypeU32Vector(state, 4), (patch_end - patch_begin + 15u) / 16u);
+		} else {
+			const bool local  = kind == Attribute::LocalOutput || kind == Attribute::ControlInput;
+			const auto stride = local ? tess.ls_stride : tess.hs_stride;
+			type              = array(TypeU32Vector(state, 4), (stride + 15u) / 16u);
+			if (kind != Attribute::LocalOutput) {
+				type = array(type, local ? tess.input_control_points : tess.output_control_points);
+			}
+		}
+		auto& variable = state.tess_variables[index];
+		variable       = DefineInterfaceVariable(state, type, storage, "tess_attributes");
+		if (kind == Attribute::Factor) {
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationBuiltIn,
+			                            spv::BuiltInTessLevelOuter);
+			state.tess_inner_variable =
+			    DefineInterfaceVariable(state, array(TypeF32(state), 2u), storage, "tess_inner");
+			state.builder.AddAnnotation(spv::OpDecorate, state.tess_inner_variable,
+			                            spv::DecorationBuiltIn, spv::BuiltInTessLevelInner);
+			state.builder.AddAnnotation(spv::OpDecorate, state.tess_inner_variable,
+			                            spv::DecorationPatch);
+		} else {
+			state.builder.AddAnnotation(
+			    spv::OpDecorate, variable, spv::DecorationLocation,
+			    kind == Attribute::PatchOutput ? (tess.hs_stride + 15u) / 16u : 0u);
+		}
+		if (kind == Attribute::Factor || kind == Attribute::PatchOutput) {
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationPatch);
+		}
+	}
+}
+
+void DefineTessellationExecutionModes(EmitterState& state) {
+	const auto& tess = state.input_info.vertex->tess;
+	EXIT_NOT_IMPLEMENTED(tess.domain != 1u || tess.partitioning != 2u ||
+	                     tess.output_topology != 2u);
+	state.builder.RequireCapability(spv::CapabilityTessellation);
+	if (state.program.stage == ShaderType::TessellationControl) {
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOutputVertices,
+		                               tess.output_control_points);
+	} else {
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeTriangles);
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeSpacingFractionalOdd);
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeVertexOrderCw);
+	}
+}
+
+uint32_t EmitGetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
+	return EmitValueOrZeroIfCondition(ctx.state, ctx.Arg(inst, 2), [&] {
+		const auto value = ctx.state.builder.AllocateId();
+		ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), value,
+		                              TessellationPointer(ctx, inst));
+		return value;
+	});
+}
+
+void EmitSetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
+	EmitIfCondition(ctx.state, ctx.Arg(inst, 3), [&] {
+		auto value = ctx.Arg(inst, 2);
+		if (inst.Arg(0).U32() == static_cast<uint32_t>(IR::TessellationAttribute::Factor)) {
+			const auto floating = ctx.state.builder.AllocateId();
+			ctx.state.builder.AddFunction(spv::OpBitcast, TypeF32(ctx.state), floating, value);
+			value = floating;
+		}
+		ctx.state.builder.AddFunction(spv::OpStore, TessellationPointer(ctx, inst), value);
+	});
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

src/graphics/shader/recompiler/frontend/cfg/ShaderCFG.cpp

@@ -1206,13 +1206,12 @@ uint32_t FindSelectionMerge(const Graph& graph, const BasicBlock& block) {
 					return false_target;
 				}
 				// A return can leave a selection without reaching its merge. Keep the
-				// continuing arm as the merge instead of joining live state with a return.
-				if (graph.Dominates(block.id, false_target) &&
-				    HasLinearPathToTerminal(graph, true_target)) {
+				// continuing arm as the merge. Shared continuations receive a dedicated
+				// gateway in SplitSharedMergeBlock before structured control is emitted.
+				if (HasLinearPathToTerminal(graph, true_target)) {
 					return false_target;
 				}
-				if (graph.Dominates(block.id, true_target) &&
-				    HasLinearPathToTerminal(graph, false_target)) {
+				if (HasLinearPathToTerminal(graph, false_target)) {
 					return true_target;
 				}
 			}

src/graphics/shader/recompiler/frontend/decode/ImageOps.cpp

@@ -3,6 +3,7 @@
 #include "graphics/shader/recompiler/frontend/decode/OpcodeTable.h"
 
 #include <algorithm>
+#include <bit>
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
@@ -175,6 +176,7 @@ constexpr MimgSampleInfo MIMG_SAMPLE_OPCODE_LIST[] = {
 };
 
 constexpr MimgGatherInfo MIMG_GATHER_OPCODE_LIST[] = {
+    {0x44u, Opcode::IMAGE_GATHER4_L, ImageSampleFlagLod},
     {0x47u, Opcode::IMAGE_GATHER4_LZ, ImageSampleFlagLevelZero},
     {0x48u, Opcode::IMAGE_GATHER4_C, ImageSampleFlagCompare},
     {0x4fu, Opcode::IMAGE_GATHER4_C_LZ,
@@ -202,18 +204,6 @@ constexpr auto MIMG_SAMPLE_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_SAMPLE_OPCO
 constexpr auto MIMG_GATHER_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_GATHER_OPCODE_LIST);
 constexpr auto MIMG_ATOMIC_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_ATOMIC_OPCODE_LIST);
 
-const MimgSampleInfo* LookupSample(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
-}
-
-const MimgGatherInfo* LookupGather(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_GATHER_OPS, opcode);
-}
-
-const Detail::OpcodeMap* LookupAtomic(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_ATOMIC_OPS, opcode);
-}
-
 Opcode DecodeMimgOpcode(uint32_t opcode, const MimgSampleInfo* sample, const MimgGatherInfo* gather,
                         const Detail::OpcodeMap* atomic) {
 	if (sample != nullptr) {
@@ -279,10 +269,6 @@ uint32_t CountDmaskComponents(uint32_t dmask) {
 	return count != 0 ? count : 1u;
 }
 
-bool IsSingleDmaskBit(uint32_t dmask) {
-	return dmask != 0u && (dmask & (dmask - 1u)) == 0u;
-}
-
 } // namespace
 
 ImageAddressComponent ImageAddressComponentLayout(uint32_t flags, uint32_t component) {
@@ -331,9 +317,9 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	const bool     r128   = ((word0 >> 15u) & 0x1u) != 0u;
 	const bool     a16    = ((word1 >> 30u) & 0x1u) != 0u;
 	const bool     d16    = ((word1 >> 31u) & 0x1u) != 0u;
-	const auto*    sample = LookupSample(opcode);
-	const auto*    gather = LookupGather(opcode);
-	const auto*    atomic = LookupAtomic(opcode);
+	const auto*    sample = Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
+	const auto*    gather = Detail::FindOpcode(MIMG_GATHER_OPS, opcode);
+	const auto*    atomic = Detail::FindOpcode(MIMG_ATOMIC_OPS, opcode);
 
 	inst.pc                 = pc;
 	inst.word_count         = word_count;
@@ -363,7 +349,7 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::MIMG, opcode, "MIMG opcode is not implemented");
 	}
-	if (gather != nullptr && !IsSingleDmaskBit(inst.dmask)) {
+	if (gather != nullptr && !std::has_single_bit(inst.dmask)) {
 		SetUnsupported(inst, Family::MIMG, opcode,
 		               "MIMG image gather requires exactly one dmask bit");
 	}
@@ -381,7 +367,7 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 }
 
 const char* MimgSampleOpcodeName(uint32_t opcode) {
-	const auto* sample = LookupSample(opcode);
+	const auto* sample = Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
 	return sample != nullptr ? sample->name : nullptr;
 }
 

src/graphics/shader/recompiler/frontend/decode/ScalarAluOps.cpp

@@ -24,7 +24,8 @@ constexpr OpcodeMap SOP2_OPCODE_LIST[] = {
     {0x1cu, Opcode::S_XNOR_B32},        {0x1du, Opcode::S_XNOR_B64},
     {0x1eu, Opcode::S_LSHL_B32},        {0x1fu, Opcode::S_LSHL_B64},
     {0x20u, Opcode::S_LSHR_B32},        {0x21u, Opcode::S_LSHR_B64},
-    {0x22u, Opcode::S_ASHR_I32},        {0x24u, Opcode::S_BFM_B32},
+    {0x22u, Opcode::S_ASHR_I32},        {0x23u, Opcode::S_ASHR_I64},
+    {0x24u, Opcode::S_BFM_B32},
     {0x25u, Opcode::S_BFM_B64},         {0x26u, Opcode::S_MUL_I32},
     {0x27u, Opcode::S_BFE_U32},         {0x28u, Opcode::S_BFE_I32},
     {0x29u, Opcode::S_BFE_U64},         {0x2cu, Opcode::S_ABSDIFF_I32},
@@ -73,7 +74,8 @@ constexpr OpcodeMap SOPC_OPCODE_LIST[] = {
     {0x03u, Opcode::S_CMP_GE_I32},  {0x04u, Opcode::S_CMP_LT_I32},  {0x05u, Opcode::S_CMP_LE_I32},
     {0x06u, Opcode::S_CMP_EQ_U32},  {0x07u, Opcode::S_CMP_LG_U32},  {0x08u, Opcode::S_CMP_GT_U32},
     {0x09u, Opcode::S_CMP_GE_U32},  {0x0au, Opcode::S_CMP_LT_U32},  {0x0bu, Opcode::S_CMP_LE_U32},
-    {0x0cu, Opcode::S_BITCMP0_B32}, {0x0du, Opcode::S_BITCMP1_B32}, {0x12u, Opcode::S_CMP_EQ_U64},
+    {0x0cu, Opcode::S_BITCMP0_B32}, {0x0du, Opcode::S_BITCMP1_B32},
+    {0x0eu, Opcode::S_BITCMP0_B64}, {0x0fu, Opcode::S_BITCMP1_B64}, {0x12u, Opcode::S_CMP_EQ_U64},
     {0x13u, Opcode::S_CMP_LG_U64},
 };
 
@@ -105,6 +107,7 @@ constexpr OpcodeMap SOPP_OPCODE_LIST[] = {
     {0x10u, Opcode::S_SENDMSG},
     {0x12u, Opcode::S_TRAP},
     {0x16u, Opcode::S_TTRACEDATA},
+    {0x17u, Opcode::S_CBRANCH_CDBGSYS},
     {0x20u, Opcode::S_INST_PREFETCH},
     {0x23u, Opcode::S_WAITCNT_DEPCTR},
 };
@@ -136,6 +139,10 @@ void DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.family    = Family::SOP1;
 	inst.opcode_id = opcode;
 	inst.opcode    = Detail::LookupOpcode(SOP1_OPS, opcode);
+	if (opcode == 0x21u && sdst == 125u) {
+		// S_SWAPPC_B64 with NULL discards the return PC, so it is a plain jump.
+		inst.opcode = Opcode::S_SETPC_B64;
+	}
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.cpp

@@ -14,10 +14,6 @@
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
 
-uint32_t FloatBits(float value) {
-	return std::bit_cast<uint32_t>(value);
-}
-
 bool HasLiteral(const Instruction& inst) {
 	return inst.src0.kind == OperandKind::LiteralConstant ||
 	       inst.src1.kind == OperandKind::LiteralConstant ||
@@ -144,6 +140,7 @@ std::string FormatMimg(const Instruction& inst) {
 	}
 	switch (inst.opcode) {
 		case Opcode::IMAGE_SAMPLE:
+		case Opcode::IMAGE_GATHER4_L:
 		case Opcode::IMAGE_GATHER4_LZ:
 		case Opcode::IMAGE_GATHER4_C:
 		case Opcode::IMAGE_GATHER4_C_LZ:
@@ -187,6 +184,8 @@ std::string FormatExp(const Instruction& inst) {
 
 bool IsConditionalBranch(Opcode opcode) {
 	switch (opcode) {
+		// DevKit NGG should be disabled.
+		case Opcode::S_CBRANCH_CDBGSYS: return false;
 		case Opcode::S_CBRANCH_SCC0:
 		case Opcode::S_CBRANCH_SCC1:
 		case Opcode::S_CBRANCH_VCCZ:
@@ -239,7 +238,7 @@ void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 	if (code >= 240u && code <= 247u) {
 		constexpr float values[] = {0.5f, -0.5f, 1.0f, -1.0f, 2.0f, -2.0f, 4.0f, -4.0f};
 		operand.kind             = OperandKind::FloatInlineConstant;
-		operand.value            = FloatBits(values[code - 240u]);
+		operand.value            = std::bit_cast<uint32_t>(values[code - 240u]);
 		return;
 	}
 	if (code >= 256u && code <= 511u) {
@@ -257,7 +256,7 @@ void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 		case 239u: operand.kind = OperandKind::PopsExitingWaveId; return;
 		case 248u:
 			operand.kind      = OperandKind::FloatInlineConstant;
-			operand.value = FloatBits(0.15915494309189535f);
+			operand.value = std::bit_cast<uint32_t>(0.15915494309189535f);
 			return;
 		case 251u: operand.kind = OperandKind::VccZ; return;
 		case 252u: operand.kind = OperandKind::ExecZ; return;
@@ -381,6 +380,25 @@ void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Inst
 	}
 }
 
+Program DecodeFrontProgram(std::span<const uint32_t> front) {
+	Program  result;
+	uint32_t front_words = 0;
+	while (front_words < front.size()) {
+		auto& inst = result.instructions.emplace_back();
+		DecodeInstruction(front, front_words, inst);
+		front_words += inst.word_count;
+		if (inst.opcode == Opcode::S_SETPC_B64) {
+			EXIT_NOT_IMPLEMENTED(inst.src0.kind != OperandKind::Sgpr || inst.src0.reg != 6u);
+			break;
+		}
+		EXIT_NOT_IMPLEMENTED(inst.opcode == Opcode::S_ENDPGM);
+	}
+	EXIT_IF(result.instructions.empty() ||
+	        result.instructions.back().opcode != Opcode::S_SETPC_B64);
+	result.code = front.first(front_words);
+	return result;
+}
+
 void DecodeProgram(std::span<const uint32_t> code, Program& program) {
 	program.instructions.clear();
 	program.instructions.reserve(code.size());
@@ -454,8 +472,9 @@ std::string OperandToString(const Operand& operand) {
 		text += ".clamp";
 	}
 	if (operand.dpp) {
-		text += fmt::format(".dpp(ctrl=0x{:x},fi={},bc={})", operand.dpp_ctrl,
-		                    operand.dpp_fetch_inactive ? 1u : 0u, operand.dpp_bound_ctrl ? 1u : 0u);
+		text += fmt::format(".{}(ctrl=0x{:x},fi={},bc={})", operand.dpp8 ? "dpp8" : "dpp",
+		                    operand.dpp_ctrl, operand.dpp_fetch_inactive ? 1u : 0u,
+		                    operand.dpp_bound_ctrl ? 1u : 0u);
 	}
 	return text;
 }
@@ -535,6 +554,7 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::S_CBRANCH_VCCNZ:
 		case Opcode::S_CBRANCH_EXECZ:
 		case Opcode::S_CBRANCH_EXECNZ:
+		case Opcode::S_CBRANCH_CDBGSYS:
 			return WithUnsupportedReason(inst, fmt::format("0x{:08x}: {} 0x{:08x}", inst.pc,
 			                                               magic_enum::enum_name(inst.opcode),
 			                                               inst.branch_target));
@@ -558,6 +578,7 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::IMAGE_LOAD_MIP:
 		case Opcode::IMAGE_GET_RESINFO:
 		case Opcode::IMAGE_GET_LOD:
+		case Opcode::IMAGE_GATHER4_L:
 		case Opcode::IMAGE_GATHER4_LZ:
 		case Opcode::IMAGE_GATHER4_C:
 		case Opcode::IMAGE_GATHER4_C_LZ:

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h

@@ -2,8 +2,8 @@
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SHADERDECODER_H_
 
 #include "common/common.h"
-#include "common/magicEnum.h"
 
+#include <magic_enum.hpp>
 #include <span>
 #include <string_view>
 #include <vector>
@@ -73,6 +73,8 @@ enum class Opcode {
 	S_SUB_I32,
 	S_BITCMP0_B32,
 	S_BITCMP1_B32,
+	S_BITCMP0_B64,
+	S_BITCMP1_B64,
 	S_BITSET0_B32,
 	S_BITSET1_B32,
 	S_BITSET0_B64,
@@ -106,6 +108,7 @@ enum class Opcode {
 	S_LSHR_B32,
 	S_LSHR_B64,
 	S_ASHR_I32,
+	S_ASHR_I64,
 	S_MUL_I32,
 	S_MUL_HI_U32,
 	S_MUL_HI_I32,
@@ -227,6 +230,7 @@ enum class Opcode {
 	V_CVT_PKRTZ_F16_F32,
 	V_CVT_PK_U8_F32,
 	V_MAD_F32,
+	V_MAD_I16,
 	V_MAD_I32_I24,
 	V_MAD_U32_U24,
 	V_MAD_U64_U32,
@@ -299,6 +303,7 @@ enum class Opcode {
 	V_SUBREV_NC_U32,
 	V_ADD_NC_U16,
 	V_SUB_NC_U16,
+	V_MUL_LO_U16,
 	V_MAX_U16,
 	V_MAX_I16,
 	V_MIN_U16,
@@ -377,7 +382,9 @@ enum class Opcode {
 	V_CMP_GT_F16,
 	V_CMP_LG_F16,
 	V_CMP_GE_F16,
+	V_CMP_NGT_F16,
 	V_CMP_NEQ_F16,
+	V_CMP_NLT_F16,
 	V_CMPX_LT_F16,
 	V_CMPX_EQ_F16,
 	V_CMPX_LE_F16,
@@ -559,6 +566,7 @@ enum class Opcode {
 	IMAGE_ATOMIC_OR,
 	IMAGE_ATOMIC_XOR,
 	IMAGE_SAMPLE,
+	IMAGE_GATHER4_L,
 	IMAGE_GATHER4_LZ,
 	IMAGE_GATHER4_C,
 	IMAGE_GATHER4_C_LZ,
@@ -581,6 +589,7 @@ enum class Opcode {
 	S_CBRANCH_VCCNZ,
 	S_CBRANCH_EXECZ,
 	S_CBRANCH_EXECNZ,
+	S_CBRANCH_CDBGSYS,
 	S_SENDMSG,
 	S_SETREG_B32,
 	S_SLEEP,
@@ -663,6 +672,7 @@ struct Operand {
 	bool     absolute           = false;
 	bool     clamp              = false;
 	bool     dpp                = false;
+	bool     dpp8               = false;
 };
 
 struct Instruction {
@@ -722,6 +732,7 @@ struct Program {
 Family GetInstructionFamily(uint32_t word);
 // The output object must be freshly initialized.
 void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst);
+Program DecodeFrontProgram(std::span<const uint32_t> front);
 void DecodeProgram(std::span<const uint32_t> code, Program& program);
 bool IsConditionalBranch(Opcode opcode);
 bool IsDirectBranch(Opcode opcode);

src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp

@@ -83,10 +83,6 @@ constexpr Vop2OpcodeInfo VOP2_OPCODE_LIST[] = {
 
 constexpr auto VOP2_OPS = Detail::MakeOpcodeTable<0x40>(VOP2_OPCODE_LIST);
 
-constexpr Opcode LookupVop2Opcode(uint32_t encoding) {
-	return Detail::LookupOpcode(VOP2_OPS, encoding);
-}
-
 constexpr OpcodeMap VOP1_OPCODE_LIST[] = {
     {0x00u, Opcode::V_NOP},
     {0x01u, Opcode::V_MOV_B32},
@@ -250,7 +246,8 @@ constexpr VopcOpcodeInfo VOPC_OPCODE_LIST[] = {
     {0xf5u, Opcode::V_CMPX_NE_U64, false}, {0xc9u, Opcode::V_CMP_LT_F16},
     {0xcau, Opcode::V_CMP_EQ_F16},         {0xcbu, Opcode::V_CMP_LE_F16},
     {0xccu, Opcode::V_CMP_GT_F16},         {0xcdu, Opcode::V_CMP_LG_F16},
-    {0xceu, Opcode::V_CMP_GE_F16},         {0xedu, Opcode::V_CMP_NEQ_F16},
+    {0xceu, Opcode::V_CMP_GE_F16},         {0xebu, Opcode::V_CMP_NGT_F16},
+    {0xedu, Opcode::V_CMP_NEQ_F16},        {0xeeu, Opcode::V_CMP_NLT_F16},
     {0xd9u, Opcode::V_CMPX_LT_F16},        {0xdau, Opcode::V_CMPX_EQ_F16},
     {0xdbu, Opcode::V_CMPX_LE_F16},        {0xdcu, Opcode::V_CMPX_GT_F16},
     {0xdeu, Opcode::V_CMPX_GE_F16},        {0xfbu, Opcode::V_CMPX_NGT_F16},
@@ -298,6 +295,7 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
     {0x300u, Opcode::V_LSHRREV_B64},
     {0x303u, Opcode::V_ADD_NC_U16},
     {0x304u, Opcode::V_SUB_NC_U16},
+    {0x305u, Opcode::V_MUL_LO_U16},
     {0x307u, Opcode::V_LSHRREV_B16},
     {0x308u, Opcode::V_ASHRREV_I16},
     {0x309u, Opcode::V_MAX_U16},
@@ -314,6 +312,7 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
     {0x345u, Opcode::V_XAD_U32},
     {0x346u, Opcode::V_LSHL_ADD_U32},
     {0x347u, Opcode::V_ADD_LSHL_U32},
+    {0x35eu, Opcode::V_MAD_I16},
     {0x360u, Opcode::V_READLANE_B32},
     {0x361u, Opcode::V_WRITELANE_B32},
     {0x362u, Opcode::V_LDEXP_F32},
@@ -374,7 +373,7 @@ Opcode LookupVop3Opcode(uint32_t opcode) {
 		if (IsUnsupportedVop3EncodedVop2Alias(opcode - 0x100u)) {
 			return Opcode::UNSUPPORTED;
 		}
-		return LookupVop2Opcode(opcode - 0x100u);
+		return Detail::LookupOpcode(VOP2_OPS, opcode - 0x100u);
 	}
 	if (opcode >= 0x180u && opcode <= 0x1ffu) {
 		return Detail::LookupOpcode(VOP3_ENCODED_VOP1_OPS, opcode - 0x180u);
@@ -426,13 +425,14 @@ bool IsNativeVop3F16TernaryOpcode(Opcode opcode) {
 }
 
 bool IsNativeVop3I16TernaryOpcode(Opcode opcode) {
-	return opcode == Opcode::V_MED3_I16;
+	return opcode == Opcode::V_MED3_I16 || opcode == Opcode::V_MAD_I16;
 }
 
 bool IsNativeVop3B16BinaryOpcode(Opcode opcode) {
 	switch (opcode) {
 		case Opcode::V_ADD_NC_U16:
 		case Opcode::V_SUB_NC_U16:
+		case Opcode::V_MUL_LO_U16:
 		case Opcode::V_MAX_U16:
 		case Opcode::V_MAX_I16:
 		case Opcode::V_MIN_U16:
@@ -516,18 +516,19 @@ struct Vop1SdwaRule {
 };
 
 constexpr Vop1SdwaRule VOP1_SDWA_RULES[] = {
-    {Opcode::V_MOV_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
+    {Opcode::V_MOV_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelBytes() | SdwaSelWords(),
      SdwaSelWords() | SdwaSelFull(), false},
     {Opcode::V_CVT_F32_U32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_I32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_UBYTE0, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
-    {Opcode::V_NOT_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
+    {Opcode::V_NOT_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelBytes() | SdwaSelWords(),
+     SdwaSelFull(), false},
     {Opcode::V_FFBL_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_F16, SdwaSelWords() | SdwaSelFull(), 0, 0, true},
     {Opcode::V_CVT_F16_F32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
      SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), true},
-    {Opcode::V_CVT_F16_U16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
-     SdwaSelWords() | SdwaSelFull(), false},
+    {Opcode::V_CVT_F16_U16, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
+     SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), false},
     {Opcode::V_CVT_U16_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
      SdwaSelWords() | SdwaSelFull(), true},
     {Opcode::V_CVT_F16_I16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
@@ -617,13 +618,14 @@ bool SupportsVop1Clamp(Opcode opcode) {
 }
 
 bool ValidateVop1Sdwa(Instruction& inst, uint32_t opcode, uint32_t modifier) {
-	const auto dst_sel  = (modifier >> 8u) & 0x7u;
-	const auto dst_u    = (modifier >> 11u) & 0x3u;
-	const auto clamp    = (modifier >> 13u) & 0x1u;
-	const auto omod     = (modifier >> 14u) & 0x3u;
-	const auto src0_sel = (modifier >> 16u) & 0x7u;
-	const auto src0_neg = (modifier >> 20u) & 0x1u;
-	const auto src0_abs = (modifier >> 21u) & 0x1u;
+	const auto dst_sel   = (modifier >> 8u) & 0x7u;
+	const auto dst_u     = (modifier >> 11u) & 0x3u;
+	const auto clamp     = (modifier >> 13u) & 0x1u;
+	const auto omod      = (modifier >> 14u) & 0x3u;
+	const auto src0_sel  = (modifier >> 16u) & 0x7u;
+	const auto src0_sext = (modifier >> 19u) & 0x1u;
+	const auto src0_neg  = (modifier >> 20u) & 0x1u;
+	const auto src0_abs  = (modifier >> 21u) & 0x1u;
 
 	if (src0_sel > 6u || dst_sel > 6u) {
 		SetUnsupported(inst, Family::VOP1, opcode, "VOP1 SDWA selector is invalid");
@@ -639,7 +641,8 @@ bool ValidateVop1Sdwa(Instruction& inst, uint32_t opcode, uint32_t modifier) {
 		               "VOP1 SDWA destination selector is not supported");
 		return false;
 	}
-	if (!IsVop1SdwaSourceSupported(inst.opcode, src0_sel, src0_neg != 0u, src0_abs != 0u)) {
+	if (!IsVop1SdwaSourceSupported(inst.opcode, src0_sel, src0_neg != 0u, src0_abs != 0u) ||
+	    (inst.opcode == Opcode::V_CVT_F16_U16 && src0_sel <= 3u && src0_sext != 0u)) {
 		SetUnsupported(inst, Family::VOP1, opcode, "VOP1 SDWA source selector is not supported");
 		return false;
 	}
@@ -685,10 +688,15 @@ void DecodeVop1Sdwa(uint32_t pc, std::span<const uint32_t> code, uint32_t word_i
 	ReadLiteralOperands(code, word_index, inst);
 }
 
-void ApplyDppModifier(Operand& operand, uint32_t modifier) {
+void ApplyDppModifier(Operand& operand, uint32_t modifier, uint32_t encoding) {
+	operand.dpp = true;
+	if (encoding == 233u) {
+		operand.dpp8     = true;
+		operand.dpp_ctrl = modifier >> 8u;
+		return;
+	}
 	operand.negate             = ((modifier >> 20u) & 0x1u) != 0u;
 	operand.absolute           = ((modifier >> 21u) & 0x1u) != 0u;
-	operand.dpp                = true;
 	operand.dpp_ctrl           = (modifier >> 8u) & 0x1ffu;
 	operand.dpp_fetch_inactive = ((modifier >> 18u) & 0x1u) != 0u;
 	operand.dpp_bound_ctrl     = ((modifier >> 19u) & 0x1u) != 0u;
@@ -709,7 +717,7 @@ void DecodeVop1Dpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 		DecodeVectorGpr(vdst, inst.dst);
 	}
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src_count = 1;
 
 	if (!IsVop1FloatSourceOpcode(inst.opcode) && (inst.src0.negate || inst.src0.absolute)) {
@@ -824,7 +832,9 @@ bool IsVopcFloatCompareOpcode(Opcode opcode) {
 		case Opcode::V_CMP_GT_F16:
 		case Opcode::V_CMP_LG_F16:
 		case Opcode::V_CMP_GE_F16:
+		case Opcode::V_CMP_NGT_F16:
 		case Opcode::V_CMP_NEQ_F16:
+		case Opcode::V_CMP_NLT_F16:
 		case Opcode::V_CMPX_LT_F16:
 		case Opcode::V_CMPX_EQ_F16:
 		case Opcode::V_CMPX_LE_F16:
@@ -1046,7 +1056,7 @@ void DecodeVop2Dpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 	DecodeVectorGpr(vdst, inst.dst);
 	DecodeVectorGpr(vsrc1, inst.src1);
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src1.negate       = ((modifier >> 22u) & 0x1u) != 0u;
 	inst.src1.absolute     = ((modifier >> 23u) & 0x1u) != 0u;
 	const bool packed_fmac = inst.opcode == Opcode::V_PK_FMAC_F16;
@@ -1155,7 +1165,7 @@ void DecodeVopcDpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 	DecodeVectorGpr(vsrc1, inst.src1);
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
 	inst.dst.kind = IsVopcCompareExec(inst.opcode) ? OperandKind::ExecLo : OperandKind::VccLo;
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src1.negate   = ((modifier >> 22u) & 0x1u) != 0u;
 	inst.src1.absolute = ((modifier >> 23u) & 0x1u) != 0u;
 	inst.src_count     = 2;
@@ -1171,6 +1181,7 @@ uint32_t NativeVop3SourceCount(Opcode opcode) {
 		case Opcode::V_MUL_I32_I24:
 		case Opcode::V_ADD_NC_U16:
 		case Opcode::V_SUB_NC_U16:
+		case Opcode::V_MUL_LO_U16:
 		case Opcode::V_MAX_U16:
 		case Opcode::V_MAX_I16:
 		case Opcode::V_MIN_U16:
@@ -1355,7 +1366,8 @@ bool SupportsNativeVop3ResultModifiers(Opcode opcode) {
 }
 
 bool SupportsNativeVop3Clamp(Opcode opcode) {
-	return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode);
+	return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode) ||
+	       opcode == Opcode::V_MAD_I16;
 }
 
 bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mix,
@@ -1376,7 +1388,7 @@ bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mi
 		return opcode != Opcode::V_FMA_F16 && (clamp != 0u || omod != 0u);
 	}
 	if (IsNativeVop3I16TernaryOpcode(opcode)) {
-		return abs != 0u || clamp != 0u || omod != 0u || neg != 0u;
+		return abs != 0u || (clamp != 0u && !clamp_modifier) || omod != 0u || neg != 0u;
 	}
 	if (IsNativeVop3B16BinaryOpcode(opcode)) {
 		return abs != 0u || clamp != 0u || omod != 0u || neg != 0u;
@@ -1468,7 +1480,7 @@ void DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.pc        = pc;
 	inst.family    = Family::VOP2;
 	inst.opcode_id = opcode;
-	inst.opcode    = LookupVop2Opcode(opcode);
+	inst.opcode    = Detail::LookupOpcode(VOP2_OPS, opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
@@ -1510,6 +1522,7 @@ void DecodeVop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 		return;
 	}
 	switch (src0) {
+		case 233u: DecodeVop1Dpp(pc, code, word_index, opcode, vdst, inst); return;
 		case 249u: DecodeVop1Sdwa(pc, code, word_index, opcode, vdst, inst); return;
 		case 250u: DecodeVop1Dpp(pc, code, word_index, opcode, vdst, inst); return;
 		default: break;

src/graphics/shader/recompiler/frontend/translate/Convert.cpp

@@ -166,9 +166,9 @@ void Translator::V_CVT_PK_U8_F32(const Decoder::Instruction& inst) {
 }
 
 void Translator::V_PACK_B32_F16(const Decoder::Instruction& inst) {
-	const auto low = ReadF16LaneBits(inst.src0, false);
+	const auto low = Read16LaneBits(inst.src0, false);
 	const auto high =
-	    ir.ShiftLeftLogical(ReadF16LaneBits(inst.src1, false), IR::U32(IR::Value(16u)));
+	    ir.ShiftLeftLogical(Read16LaneBits(inst.src1, false), IR::U32(IR::Value(16u)));
 	WriteOperand(DestinationOperand(inst), ir.BitwiseOr(low, high));
 }
 

src/graphics/shader/recompiler/frontend/translate/Integer.cpp

@@ -22,6 +22,18 @@ bool Translator::Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpco
 	return true;
 }
 
+bool Translator::V_MAD_I16(const Decoder::Instruction& inst) {
+	const auto lhs    = ReadU16AsU32(inst.src0, true);
+	const auto rhs    = ReadU16AsU32(inst.src1, true);
+	auto       result = ir.IAdd(ir.IMul(lhs, rhs), ReadU16AsU32(inst.src2, true));
+	if (inst.dst.clamp) {
+		result = IR::U32(ir.Emit(IR::ValueOpcode::SMax32, {result, IR::Value(0xffff8000u)}));
+		result = IR::U32(ir.Emit(IR::ValueOpcode::SMin32, {result, IR::Value(0x7fffu)}));
+	}
+	Write16Bits(DestinationOperand(inst), ir.BitwiseAnd(result, IR::U32(IR::Value(0xffffu))));
+	return true;
+}
+
 bool Translator::V_MED3_I16(const Decoder::Instruction& inst) {
 	const auto result = IR::U32(ir.Emit(
 	    IR::ValueOpcode::SMedTri32, {ReadU16AsU32(inst.src0, true), ReadU16AsU32(inst.src1, true),
@@ -168,6 +180,21 @@ bool Translator::SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode
 	return true;
 }
 
+bool Translator::S_ASHR_I64(const Decoder::Instruction& inst) {
+	// Signed 64-bit sources sign-extend literals; generic B64 operands zero-extend them.
+	const auto source =
+	    inst.src0.kind == Decoder::OperandKind::LiteralConstant
+	        ? ir.ConstructU64(
+	              ReadU32(inst.src0),
+	              IR::U32(IR::Value((inst.src0.value & 0x80000000u) ? 0xffffffffu : 0u)))
+	        : ReadU64(inst.src0);
+	const auto result =
+	    ir.Emit(IR::ValueOpcode::ShiftRightArithmetic64, {source, ReadU32(inst.src1)});
+	WriteOperand(inst.dst, result);
+	ir.SetScc(IR::U1(ir.Emit(IR::ValueOpcode::INotEqual64, {result, IR::Value(uint64_t {0})})));
+	return true;
+}
+
 bool Translator::ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
                                        bool negate_rhs, bool negate_result, bool update_scc) {
 	const auto lhs = ReadU32(inst.src0);
@@ -480,6 +507,18 @@ bool Translator::S_BITCMP_B32(const Decoder::Instruction& inst, bool expected) {
 	return true;
 }
 
+bool Translator::S_BITCMP_B64(const Decoder::Instruction& inst, bool expected) {
+	const auto value    = ReadU32Pair(inst.src0);
+	const auto offset   = ir.BitwiseAnd(ReadU32(inst.src1), IR::U32(IR::Value(63u)));
+	const auto word_bit = ir.BitwiseAnd(offset, IR::U32(IR::Value(31u)));
+	const auto word =
+	    ir.Select(ir.ULessThan(offset, IR::U32(IR::Value(32u))), value[0], value[1]);
+	const auto bit =
+	    IR::U32(ir.Emit(IR::ValueOpcode::BitFieldUExtract, {word, word_bit, IR::Value(1u)}));
+	WriteCompareResult(inst.dst, ir.IEqual(bit, IR::U32(IR::Value(expected ? 1u : 0u))));
+	return true;
+}
+
 bool Translator::V_ALIGNBIT_B32(const Decoder::Instruction& inst) {
 	const auto hi      = ReadU32(inst.src0);
 	const auto lo      = ReadU32(inst.src1);

src/graphics/shader/recompiler/frontend/translate/Memory.cpp

@@ -34,6 +34,7 @@ Decoder::Operand OffsetDecodedRegister(const Decoder::Operand& operand, uint32_t
 	result.negate_hi          = false;
 	result.absolute           = false;
 	result.dpp_ctrl           = 0;
+	result.dpp8               = false;
 	result.dpp_row_mask       = 0xf;
 	result.dpp_bank_mask      = 0xf;
 	result.explicit_sdwa_dst  = false;
@@ -1034,6 +1035,7 @@ bool Translator::EmitMemory(const Decoder::Instruction& inst) {
 		case Decoder::Opcode::IMAGE_STORE:
 		case Decoder::Opcode::IMAGE_STORE_MIP: return IMAGE_STORE(inst);
 		case Decoder::Opcode::IMAGE_SAMPLE: return IMAGE_SAMPLE(inst);
+		case Decoder::Opcode::IMAGE_GATHER4_L:
 		case Decoder::Opcode::IMAGE_GATHER4_LZ:
 		case Decoder::Opcode::IMAGE_GATHER4_C:
 		case Decoder::Opcode::IMAGE_GATHER4_C_LZ:

src/graphics/shader/recompiler/frontend/translate/Scalar.cpp

@@ -179,6 +179,7 @@ bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 		case O::S_LSHR_B64:
 			return SimpleInteger(inst, IR::ValueOpcode::ShiftRightLogical64, IR::Type::U64, false,
 			                     false, true);
+		case O::S_ASHR_I64: return S_ASHR_I64(inst);
 
 		case O::S_ANDN2_B32:
 			return ComposedIntegerBinary(inst, IR::ValueOpcode::BitwiseAnd32, true, false, true);
@@ -207,6 +208,8 @@ bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 		case O::S_BFE_U64: return S_BFE_U64(inst);
 		case O::S_BITCMP0_B32: return S_BITCMP_B32(inst, false);
 		case O::S_BITCMP1_B32: return S_BITCMP_B32(inst, true);
+		case O::S_BITCMP0_B64: return S_BITCMP_B64(inst, false);
+		case O::S_BITCMP1_B64: return S_BITCMP_B64(inst, true);
 		case O::S_PACK_LL_B32_B16: return PackB16(inst, false, false);
 		case O::S_PACK_LH_B32_B16: return PackB16(inst, false, true);
 		case O::S_PACK_HH_B32_B16: return PackB16(inst, true, true);
@@ -227,6 +230,7 @@ bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 		case O::S_CBRANCH_VCCNZ:
 		case O::S_CBRANCH_EXECZ:
 		case O::S_CBRANCH_EXECNZ:
+		case O::S_CBRANCH_CDBGSYS:
 		case O::S_ENDPGM: return true;
 		default: return false;
 	}

src/graphics/shader/recompiler/frontend/translate/Translate.cpp

@@ -9,6 +9,17 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
+static IR::DppMoveFlags DppFlags(const Decoder::Operand& operand) {
+	return {
+	    .control        = operand.dpp_ctrl,
+	    .row_mask       = operand.dpp_row_mask,
+	    .bank_mask      = operand.dpp_bank_mask,
+	    .fetch_inactive = operand.dpp_fetch_inactive,
+	    .bound_control  = operand.dpp_bound_ctrl,
+	    .dpp8           = operand.dpp8,
+	};
+}
+
 const Decoder::Operand& Translator::SourceAt(const Decoder::Instruction& inst, uint32_t index) {
 	switch (index) {
 		case 0: return inst.src0;
@@ -30,6 +41,7 @@ Decoder::Operand Translator::DestinationOperand(const Decoder::Instruction& inst
 			continue;
 		}
 		destination.dpp                = true;
+		destination.dpp8               = source.dpp8;
 		destination.dpp_ctrl           = source.dpp_ctrl;
 		destination.dpp_row_mask       = source.dpp_row_mask;
 		destination.dpp_bank_mask      = source.dpp_bank_mask;
@@ -106,6 +118,7 @@ Decoder::Operand Translator::PlainOperand(const Decoder::Operand& operand) {
 	result.dpp_fetch_inactive = false;
 	result.dpp_bound_ctrl     = false;
 	result.dpp                = false;
+	result.dpp8               = false;
 	return result;
 }
 
@@ -167,14 +180,8 @@ IR::U32 Translator::ReadScalarCode(uint32_t code) {
 
 IR::U32 Translator::ApplyBitSourceModifiers(const Decoder::Operand& operand, IR::U32 value) {
 	if (operand.dpp) {
-		const IR::DppMoveFlags flags {
-		    .control        = static_cast<uint16_t>(operand.dpp_ctrl),
-		    .row_mask       = static_cast<uint8_t>(operand.dpp_row_mask),
-		    .bank_mask      = static_cast<uint8_t>(operand.dpp_bank_mask),
-		    .fetch_inactive = operand.dpp_fetch_inactive,
-		    .bound_control  = operand.dpp_bound_ctrl,
-		};
-		value = IR::U32(ir.Emit(IR::ValueOpcode::DppMoveU32, {value, ir.GetExec()}, flags));
+		value =
+		    IR::U32(ir.Emit(IR::ValueOpcode::DppMoveU32, {value, ir.GetExec()}, DppFlags(operand)));
 	}
 	if (operand.sdwa_sel != 6u) {
 		uint32_t offset = 0;
@@ -301,15 +308,8 @@ void Translator::WriteRawU32(const Decoder::Operand& operand, IR::U32 value) {
 			const auto reg = static_cast<IR::VectorReg>(operand.reg);
 			const auto old = ir.GetVectorReg(reg);
 			if (operand.dpp) {
-				const IR::DppMoveFlags flags {
-				    .control        = static_cast<uint16_t>(operand.dpp_ctrl),
-				    .row_mask       = static_cast<uint8_t>(operand.dpp_row_mask),
-				    .bank_mask      = static_cast<uint8_t>(operand.dpp_bank_mask),
-				    .fetch_inactive = operand.dpp_fetch_inactive,
-				    .bound_control  = operand.dpp_bound_ctrl,
-				};
-				value = IR::U32(
-				    ir.Emit(IR::ValueOpcode::DppUpdateU32, {value, old, ir.GetExec()}, flags));
+				value = IR::U32(ir.Emit(IR::ValueOpcode::DppUpdateU32, {value, old, ir.GetExec()},
+				                        DppFlags(operand)));
 			} else {
 				value = ir.Select(ir.GetExec(), value, old);
 			}
@@ -541,10 +541,7 @@ IR::U32 Translator::ReadU16LaneRaw(const Decoder::Operand& operand, bool high_la
 
 IR::U32 Translator::ReadU16LaneAsU32(const Decoder::Operand& operand, bool high_lane,
                                      bool sign_extend) {
-	auto value = ReadU16LaneRaw(operand, high_lane);
-	if (high_lane ? operand.negate_hi : operand.negate) {
-		value = ir.BitwiseAnd(ir.ISub(IR::U32(IR::Value(0u)), value), IR::U32(IR::Value(0xffffu)));
-	}
+	auto value = Read16LaneBits(operand, high_lane);
 	if (sign_extend || operand.sdwa_sext) {
 		value = IR::U32(
 		    ir.Emit(IR::ValueOpcode::BitFieldSExtract, {value, IR::Value(0u), IR::Value(16u)}));
@@ -556,12 +553,13 @@ IR::U32 Translator::ReadU16AsU32(const Decoder::Operand& operand, bool sign_exte
 	return ReadU16LaneAsU32(operand, false, sign_extend);
 }
 
-IR::U32 Translator::ReadF16LaneBits(const Decoder::Operand& operand, bool high_lane) {
+IR::U32 Translator::Read16LaneBits(const Decoder::Operand& operand, bool high_lane) {
 	auto value = ReadU16LaneRaw(operand, high_lane);
 	if (operand.absolute) {
 		value = ir.BitwiseAnd(value, IR::U32(IR::Value(0x7fffu)));
 	}
 	if (high_lane ? operand.negate_hi : operand.negate) {
+		// RDNA2 source NEG flips the sign bit, including packed integer operands.
 		value = ir.BitwiseXor(value, IR::U32(IR::Value(0x8000u)));
 	}
 	return value;
@@ -789,15 +787,6 @@ const EmbeddedFetchLoad* FindEmbeddedFetchLoad(const EmbeddedFetchPlan* plan, ui
 	return found != plan->loads.end() ? &*found : nullptr;
 }
 
-bool IsEmbeddedFetchPrologLoad(const EmbeddedFetchPlan* plan, uint32_t pc) {
-	if (plan == nullptr) {
-		return false;
-	}
-	return std::ranges::any_of(plan->loads, [pc](const auto& load) {
-		return std::ranges::find(load.prolog_loads, pc) != load.prolog_loads.end();
-	});
-}
-
 int ResolveEmbeddedFetchResource(const ShaderVertexInputInfo& input,
                                  const EmbeddedFetchLoad&     load) {
 	if (load.attrib_id >= 0 && load.attrib_id < input.resources_num &&
@@ -819,22 +808,6 @@ int ResolveEmbeddedFetchResource(const ShaderVertexInputInfo& input,
 	return -1;
 }
 
-bool IsScalarMemoryLoad(Decoder::Opcode opcode) {
-	switch (opcode) {
-		case Decoder::Opcode::S_LOAD_DWORD:
-		case Decoder::Opcode::S_LOAD_DWORDX2:
-		case Decoder::Opcode::S_LOAD_DWORDX4:
-		case Decoder::Opcode::S_LOAD_DWORDX8:
-		case Decoder::Opcode::S_LOAD_DWORDX16:
-		case Decoder::Opcode::S_BUFFER_LOAD_DWORD:
-		case Decoder::Opcode::S_BUFFER_LOAD_DWORDX2:
-		case Decoder::Opcode::S_BUFFER_LOAD_DWORDX4:
-		case Decoder::Opcode::S_BUFFER_LOAD_DWORDX8:
-		case Decoder::Opcode::S_BUFFER_LOAD_DWORDX16: return true;
-		default: return false;
-	}
-}
-
 bool IsBufferDwordLoad(Decoder::Opcode opcode) {
 	switch (opcode) {
 		case Decoder::Opcode::BUFFER_LOAD_DWORD:
@@ -898,11 +871,15 @@ void ValidateTranslateOptions(const TranslateOptions& options) {
 	if (options.wave_size != 32u && options.wave_size != 64u) {
 		EXIT("shader translation requires wave32 or wave64, got %u", options.wave_size);
 	}
-	if (options.embedded_fetch != nullptr && options.stage != ShaderType::Vertex) {
-		EXIT("embedded fetch requires the vertex shader stage");
+	if (options.embedded_fetch != nullptr && options.stage != ShaderType::Vertex &&
+	    options.stage != ShaderType::Local) {
+		EXIT("embedded fetch requires a vertex or local shader");
 	}
 	switch (options.stage) {
 		case ShaderType::Vertex:
+		case ShaderType::Local:
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 		case ShaderType::Mesh:
 			if (options.input_info.vertex == nullptr) {
 				EXIT("vertex shader translation has no vertex input metadata");
@@ -941,6 +918,9 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 	result.user_data_count     = options.user_data_count;
 	switch (options.stage) {
 		case ShaderType::Vertex:
+		case ShaderType::Local:
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 			result.scratch_dwords = options.input_info.vertex->scratch_size_dwords;
 			break;
 		case ShaderType::Mesh:
@@ -1120,26 +1100,32 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			    entry_ir.BitwiseOr(wave_info, entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(
 			                                                         primitive_count, u32(8)),
 			                                                     vertex_count)));
-			// GS adjacency addresses local ES records in LDS. Strip winding alternates
-			// with the global primitive number, including across subgroup boundaries.
-			const auto parity = mesh.input_primitive ==
-			                            static_cast<uint32_t>(Prospero::PrimitiveType::kTriStrip)
-			                        ? entry_ir.BitwiseAnd(entry_ir.IAdd(primitive_chunk, local), u32(1))
-			                        : u32(0);
+			// GS adjacency addresses local ES records in LDS. Fans retain the draw's
+			// center in every subgroup; strip winding follows the global primitive.
 			const auto vertex = entry_ir.IMul(local, step);
-			const auto first  = entry_ir.IAdd(vertex, parity);
-			const auto second = mesh.InputPrimitiveSize() >= 2u
-			                        ? entry_ir.ISub(entry_ir.IAdd(vertex, u32(1)), parity)
-			                        : u32(0);
-			const auto third = mesh.InputPrimitiveSize() == 3u
-			                       ? entry_ir.IAdd(vertex, u32(2))
-			                       : u32(0);
+			auto       first  = vertex;
+			auto       second = u32(0);
+			auto       third  = u32(0);
+			if (mesh.InputPrimitiveSize() >= 2u) {
+				second = entry_ir.IAdd(vertex, u32(1));
+			}
+			if (mesh.InputPrimitiveSize() == 3u) {
+				third = entry_ir.IAdd(vertex, u32(2));
+			}
+			auto input_vertex = entry_ir.IAdd(chunk, local);
+			if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriFan)) {
+				first = u32(0);
+				input_vertex = entry_ir.Select(entry_ir.IEqual(local, u32(0)), u32(0), input_vertex);
+			} else if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriStrip)) {
+				const auto parity = entry_ir.BitwiseAnd(entry_ir.IAdd(primitive_chunk, local), u32(1));
+				first = entry_ir.IAdd(first, parity);
+				second = entry_ir.ISub(second, parity);
+			}
 			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(0),
 			                      entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(first, u32(2)),
 			                                         entry_ir.ShiftLeftLogical(second, u32(18))));
 			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(1),
 			                      entry_ir.ShiftLeftLogical(third, u32(2)));
-			const auto input_vertex = entry_ir.IAdd(chunk, local);
 			const auto index_bytes  = draw(3);
 			const auto indexed      = entry_ir.INotEqual(index_bytes, u32(0));
 			const auto index_low    = draw(4);
@@ -1164,15 +1150,53 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			entry_ir.SetVectorReg(
 			    static_cast<IR::VectorReg>(8),
 			    entry_ir.IAdd(draw(2), builtin(IR::StageInputKind::WorkgroupId, 1)));
+		} else if (options.stage == ShaderType::Local) {
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3), IR::U32(IR::Value(64u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(2),
+			                      builtin(IR::StageInputKind::VertexIndex));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(3), IR::U32(IR::Value(0u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
+			                      builtin(IR::StageInputKind::InstanceIndex));
+		} else if (options.stage == ShaderType::TessellationControl) {
+			const auto& tess = options.input_info.vertex->tess;
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(2),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(0u)})));
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(4),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(1u)})));
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3),
+			                      IR::U32(IR::Value(0x81010000u | tess.input_control_points |
+			                                        (tess.output_control_points << 8u))));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(0),
+			                      builtin(IR::StageInputKind::PrimitiveId));
+			entry_ir.SetVectorReg(
+			    static_cast<IR::VectorReg>(1),
+			    entry_ir.ShiftLeftLogical(builtin(IR::StageInputKind::InvocationId),
+			                              IR::U32(IR::Value(8u))));
+		} else if (options.stage == ShaderType::TessellationEvaluation) {
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3), IR::U32(IR::Value(64u)));
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(4),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(0u)})));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
+			                      builtin(IR::StageInputKind::TessCoord, 0));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(6),
+			                      builtin(IR::StageInputKind::TessCoord, 1));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(7), IR::U32(IR::Value(0u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(8),
+			                      builtin(IR::StageInputKind::PrimitiveId));
 		} else if (options.stage == ShaderType::Pixel) {
 			const auto* ps = options.input_info.pixel;
-			if (ps->ps_perspective_center_vgpr != UINT32_MAX) {
-				entry_ir.SetVectorReg(static_cast<IR::VectorReg>(ps->ps_perspective_center_vgpr),
-				                      builtin(IR::StageInputKind::BaryCoordSmooth, 0));
-				entry_ir.SetVectorReg(
-				    static_cast<IR::VectorReg>(ps->ps_perspective_center_vgpr + 1u),
-				    builtin(IR::StageInputKind::BaryCoordSmooth, 1));
-			}
+			const auto barycentric_pair = [&](uint32_t reg, IR::StageInputKind kind) {
+				if (reg != UINT32_MAX) {
+					entry_ir.SetVectorReg(static_cast<IR::VectorReg>(reg), builtin(kind, 0));
+					entry_ir.SetVectorReg(static_cast<IR::VectorReg>(reg + 1u), builtin(kind, 1));
+				}
+			};
+			barycentric_pair(ps->ps_perspective_center_vgpr, IR::StageInputKind::BaryCoordSmooth);
+			barycentric_pair(ps->ps_perspective_centroid_vgpr,
+			                 IR::StageInputKind::BaryCoordSmoothCentroid);
 			uint32_t reg = ps->ps_system_input_base;
 			if (ps->ps_pos_x) {
 				entry_ir.SetVectorReg(static_cast<IR::VectorReg>(reg++),
@@ -1201,6 +1225,12 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 				                      builtin(IR::StageInputKind::PackedAncillary));
 			}
 		} else if (options.stage == ShaderType::Vertex) {
+			// Vulkan owns primitive assembly; each vertex subgroup is one NGG wave.
+			// Keep its full lane extent: mbcnt(-1) uses lane ordinals, not active counts.
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(2),
+			                      IR::U32(IR::Value(options.wave_size << 12u)));
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3),
+			                      IR::U32(IR::Value((1u << 28u) | options.wave_size)));
 			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
 			                      builtin(IR::StageInputKind::VertexIndex));
 			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(8),
@@ -1215,10 +1245,6 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			if (IsCodeTableLoad(cfg, instruction.pc)) {
 				continue;
 			}
-			if (IsScalarMemoryLoad(instruction.opcode) &&
-			    IsEmbeddedFetchPrologLoad(options.embedded_fetch, instruction.pc)) {
-				continue;
-			}
 			const auto* embedded = FindEmbeddedFetchLoad(options.embedded_fetch, instruction.pc);
 			if (embedded != nullptr && IsBufferDwordLoad(instruction.opcode) &&
 			    instruction.data_dwords == embedded->components &&