Remove duplicate image scratch setup from push constant test

Co-authored-by: TarkusR <90155059+TarkusR@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

@@ -43,9 +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
@@ -155,7 +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
@@ -317,8 +314,8 @@ jobs:
             libxkbcommon-dev \
             libxrandr-dev \
             libxss-dev \
+            libxtst-dev \
             lld \
-            ninja-build \
             pkg-config \
             wayland-protocols
 
@@ -363,7 +360,7 @@ jobs:
             -DCMAKE_PREFIX_PATH="$Qt6_DIR" 2>&1 | tee _Build/configure.log
           exit "${PIPESTATUS[0]}"
 
-      - name: Verify SDL2 backends
+      - name: Verify SDL3 backends
         shell: bash
         run: |
           status=0

.gitmodules

@@ -1,7 +1,7 @@
-[submodule "3rdparty/SDL2"]
-	path = 3rdparty/SDL2
+[submodule "3rdparty/SDL3"]
+	path = 3rdparty/SDL3
 	url = https://github.com/libsdl-org/SDL.git
-	branch = SDL2
+	branch = release-3.4.x
 [submodule "3rdparty/magic_enum"]
 	path = 3rdparty/magic_enum
 	url = https://github.com/Neargye/magic_enum.git

.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

@@ -29,7 +29,7 @@ FetchContent_Declare(
 FetchContent_MakeAvailable(xbyak zydis)
 
 set(SDL_SHARED OFF CACHE BOOL "" FORCE)
-add_subdirectory(SDL2)
+add_subdirectory(SDL3)
 
 add_library(Vulkan-Headers INTERFACE)
 add_library(Vulkan::Headers ALIAS Vulkan-Headers)
@@ -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

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.12...4.3)
+cmake_minimum_required(VERSION 3.16...4.3)
 
 if(CMAKE_VERSION VERSION_GREATER_EQUAL 4.0)
 	set(CMAKE_POLICY_VERSION_MINIMUM 3.5 CACHE STRING "Minimum policy version for bundled dependencies")
@@ -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")
@@ -315,7 +331,7 @@ if (KYTY_CLANG_CL)
 	set_source_files_properties(${kyty_emulator_src} PROPERTIES COMPILE_FLAGS "-Wno-pragma-pack -Wno-deprecated-declarations -D_TIMESPEC_DEFINED")
 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)
+set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL3::SDL3-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
 
 # Linux system libraries.
 if(LINUX)
@@ -325,7 +341,6 @@ endif()
 
 set(inc_headers
 	${KYTY_SOURCE_DIR}
-	${KYTY_THIRD_PARTY_DIR}/SDL2/include
 	${KYTY_THIRD_PARTY_DIR}/VulkanMemoryAllocator/include
 	${KYTY_THIRD_PARTY_DIR}/SPIRV-Tools/include
 	${KYTY_THIRD_PARTY_DIR}/SPIRV-Headers/include
@@ -450,6 +465,12 @@ 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"
@@ -534,6 +555,7 @@ 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)
@@ -601,6 +623,21 @@ 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" ${kyty_emulator_src})
 configure_macos_guest_address_space(kyty_emulator)
@@ -643,6 +680,12 @@ elseif(WIN32 OR LINUX)
 	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${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`.
 
@@ -165,15 +166,15 @@ The finished application and its runtime dependencies will be placed in
 
 ### Building on Linux
 
-Install the toolchain and the libraries the bundled SDL2 needs. Without the audio, Wayland and
-udev development packages SDL2 quietly configures itself without those backends, and the resulting
+Install the toolchain and the libraries the bundled SDL3 needs. Without the audio, Wayland and
+udev development packages SDL3 quietly configures itself without those backends, and the resulting
 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 pkg-config \
   libgl1-mesa-dev libx11-dev libxcursor-dev libxext-dev libxfixes-dev \
-  libxi-dev libxrandr-dev libxss-dev libxkbcommon-dev \
+  libxi-dev libxrandr-dev libxss-dev libxtst-dev libxkbcommon-dev \
   libasound2-dev libpulse-dev libudev-dev libdbus-1-dev libwayland-dev wayland-protocols
 ```
 
@@ -203,7 +204,7 @@ 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
+A development shell provides Clang, CMake, Ninja, Qt 6, the Vulkan headers, and the SDL3 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:

nix/devshell.nix

@@ -1,8 +1,7 @@
 { 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.
+  # Build-time dependencies for the bundled SDL3 audio, video and input backends.
   buildDeps = with pkgs; [
     qt6.qtbase # Concurrent, Network, Widgets (launcher)
     vulkan-headers
@@ -18,6 +17,7 @@ let
     libXi
     libXrandr
     libXScrnSaver
+    libXtst
     libxkbcommon
     # Wayland
     wayland
@@ -46,6 +46,7 @@ let
     libXi
     libXrandr
     libXScrnSaver
+    libXtst
     libxkbcommon
     wayland
     libdecor

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

@@ -135,6 +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

@@ -66,6 +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;
@@ -110,6 +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));
 }
 

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

@@ -253,10 +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);

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();

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/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -73,6 +73,9 @@ public:
 	void SetIndexBufferSize(uint32_t index_buffer_size);
 	void SetDrawIndirectArgsBaseAddress(uint64_t draw_indirect_args_base_addr);
 	void SetDispatchIndirectArgsBaseAddress(uint64_t dispatch_indirect_args_base_addr);
+	[[nodiscard]] uint64_t GetDispatchIndirectArgsBaseAddress() const {
+		return m_dispatch_indirect_args_base_addr;
+	}
 	void SetNumInstances(uint32_t num_instances);
 	void DrawIndex(DrawIndexArgs args);
 	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count);
@@ -99,7 +102,7 @@ public:
 	void TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id);
 	void DispatchDirect(uint32_t thread_group_x, uint32_t thread_group_y, uint32_t thread_group_z,
 	                    uint32_t mode);
-	void DispatchIndirect(uint32_t data_offset, uint32_t mode);
+	void DispatchIndirect(uint64_t args_addr, uint32_t mode);
 	void WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index);
 	void TriggerEvent(uint32_t event_type, uint32_t event_index, uint64_t event_address = 0);
 
@@ -148,7 +151,6 @@ private:
 	void SuspendPm4();
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }
-	void                CheckBuffer() const { GetScheduler().CheckActive(); }
 
 	RenderContext&   m_renderer;
 	HW::Context      m_ctx;

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -1331,26 +1331,13 @@ KYTY_CP_OP_PARSER(CpOpDispatchIndirect) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0011600 && cmd_id != 0xc0021600);
 
 	if (cmd_id == 0xc0021600) {
-		struct DispatchIndirectArgs {
-			uint32_t thread_group_x;
-			uint32_t thread_group_y;
-			uint32_t thread_group_z;
-		};
-
-		auto* args = reinterpret_cast<const DispatchIndirectArgs*>(
-		    buffer[0] | (static_cast<uint64_t>(buffer[1]) << 32u));
-		uint32_t mode = buffer[2];
-
-		EXIT_NOT_IMPLEMENTED(args == nullptr);
-		cp.DispatchDirect(args->thread_group_x, args->thread_group_y, args->thread_group_z, mode);
-
+		cp.DispatchIndirect(buffer[0] | (static_cast<uint64_t>(buffer[1]) << 32u), buffer[2]);
 		return 3;
 	}
 
-	uint32_t data_offset = buffer[0];
-	uint32_t mode        = buffer[1];
-
-	cp.DispatchIndirect(data_offset, mode);
+	const auto base_addr = cp.GetDispatchIndirectArgsBaseAddress();
+	EXIT_NOT_IMPLEMENTED(base_addr == 0);
+	cp.DispatchIndirect(base_addr + buffer[0], buffer[1]);
 
 	return 2;
 }
@@ -2017,6 +2004,14 @@ 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);
 		}
 

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,
@@ -870,13 +874,11 @@ void CommandProcessor::SetPredication(uint32_t condition, uint32_t op, uint32_t
 }
 
 void CommandProcessor::DrawIndex(DrawIndexArgs args) {
-	CheckBuffer();
-
 	args.index_type_and_size = m_index_type_and_size;
 	if (args.instance_count == 0) {
 		args.instance_count = m_num_instances;
 	}
-	if (args.base_vertex != 0 || args.first_instance != 0) {
+	if (GraphicsRunDebugDumpEnabled() && (args.base_vertex != 0 || args.first_instance != 0)) {
 		LOGF("\t draw indexed offsets: base_vertex = %" PRId32 ", first_instance = %" PRIu32 "\n",
 		     args.base_vertex, args.first_instance);
 	}
@@ -908,17 +910,6 @@ 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,
@@ -930,16 +921,6 @@ 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) {
@@ -1001,17 +982,6 @@ 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,
@@ -1022,17 +992,6 @@ 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) {
@@ -1078,7 +1037,6 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 	// uint32_t local_z   = 1;
 
 	{
-		CheckBuffer();
 		frame_num = m_renderer.GetGpu().GetFrameNum();
 		if (GraphicsRunDebugDumpEnabled()) {
 			static std::atomic<uint32_t> log_count {0};
@@ -1129,24 +1087,18 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 	}*/
 }
 
-void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
-	struct DispatchIndirectArgs {
-		uint32_t thread_group_x;
-		uint32_t thread_group_y;
-		uint32_t thread_group_z;
-	};
-
-	EXIT_NOT_IMPLEMENTED(m_dispatch_indirect_args_base_addr == 0);
-
-	const auto args_addr = m_dispatch_indirect_args_base_addr + data_offset;
-	auto*      args      = reinterpret_cast<const DispatchIndirectArgs*>(args_addr);
-
-	DispatchDirect(args->thread_group_x, args->thread_group_y, args->thread_group_z, mode);
+void CommandProcessor::DispatchIndirect(uint64_t args_addr, uint32_t mode) {
+	EXIT_NOT_IMPLEMENTED(args_addr == 0 || (args_addr & 3u) != 0);
+	if ((mode & Pm4::COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS) != 0) {
+		const auto* args = reinterpret_cast<const vk::DispatchIndirectCommand*>(args_addr);
+		DispatchDirect(args->x, args->y, args->z, mode);
+		return;
+	}
+	m_sh_ctx.SetCsWaveSize(Pm4::ComputeWaveSize(mode));
+	m_renderer.GetRenderExecutor().DispatchIndirect(m_submit_id, CurrentBuffer(), args_addr, mode);
 }
 
 void CommandProcessor::DrawIndexAuto(DrawAutoArgs args) {
-	CheckBuffer();
-
 	if (args.instance_count == 0) {
 		args.instance_count = m_num_instances;
 	}
@@ -1168,7 +1120,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
                                         uint32_t interrupt_context_id) {
 	static_assert(sizeof(T) == sizeof(uint32_t) || sizeof(T) == sizeof(uint64_t));
 
-	CheckBuffer();
+	auto& command = CurrentBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		const auto bits      = static_cast<unsigned>(sizeof(T) * 8u);
@@ -1199,7 +1151,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 		case 0x03: with_interrupt = false; break;
 		case 0x01:
 			if (!IsAsyncComputeQueue()) {
-				Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id,
+				Sync::TriggerEopEventAtEndOfPipe(command, m_interrupt_event_id,
 				                                 interrupt_context_id);
 				return;
 			}
@@ -1216,17 +1168,17 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 
 		if (with_interrupt) {
 			if (with_writeback) {
-				Sync::WriteAtEndOfPipeWithInterruptWriteBack32(m_submit_id, CurrentBuffer(), dst,
+				Sync::WriteAtEndOfPipeWithInterruptWriteBack32(m_submit_id, command, dst,
 				                                               data, m_interrupt_event_id,
 				                                               interrupt_context_id);
 			} else {
-				Sync::WriteAtEndOfPipeWithInterrupt32(m_submit_id, CurrentBuffer(), dst, data,
+				Sync::WriteAtEndOfPipeWithInterrupt32(m_submit_id, command, dst, data,
 				                                      m_interrupt_event_id, interrupt_context_id);
 			}
 		} else if (with_writeback) {
-			Sync::WriteAtEndOfPipeWithWriteBack32(m_submit_id, CurrentBuffer(), dst, data);
+			Sync::WriteAtEndOfPipeWithWriteBack32(m_submit_id, command, dst, data);
 		} else {
-			Sync::WriteAtEndOfPipe32(m_submit_id, CurrentBuffer(), dst, data);
+			Sync::WriteAtEndOfPipe32(m_submit_id, command, dst, data);
 		}
 	};
 
@@ -1238,7 +1190,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					SynchronizeGpu();
 					Sync::ReadGds(*m_renderer.GetBufferCache().GetGdsBuffer(), dst, value & 0xffffu,
 					              value >> 16u);
-					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
+					Sync::WriteAtEndOfPipeGds32(m_submit_id, command, dst, value & 0xffffu,
 					                            value >> 16u);
 					if (with_interrupt) {
 						m_renderer.TriggerInterrupt(m_interrupt_event_id, interrupt_context_id);
@@ -1267,18 +1219,18 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					if (with_interrupt) {
 						if (with_writeback) {
 							Sync::WriteAtEndOfPipeWithInterruptWriteBack64(
-							    m_submit_id, CurrentBuffer(), dst, value, m_interrupt_event_id,
+							    m_submit_id, command, dst, value, m_interrupt_event_id,
 							    interrupt_context_id);
 						} else {
-							Sync::WriteAtEndOfPipeWithInterrupt64(m_submit_id, CurrentBuffer(), dst,
+							Sync::WriteAtEndOfPipeWithInterrupt64(m_submit_id, command, dst,
 							                                      value, m_interrupt_event_id,
 							                                      interrupt_context_id);
 						}
 					} else if (with_writeback) {
-						Sync::WriteAtEndOfPipeWithWriteBack64(m_submit_id, CurrentBuffer(), dst,
+						Sync::WriteAtEndOfPipeWithWriteBack64(m_submit_id, command, dst,
 						                                      value);
 					} else {
-						Sync::WriteAtEndOfPipe64(m_submit_id, CurrentBuffer(), dst, value);
+						Sync::WriteAtEndOfPipe64(m_submit_id, command, dst, value);
 					}
 				};
 
@@ -1359,10 +1311,10 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 						    (eop_event_type == 0x28 && event_index == 0x00)) {
 							if (with_interrupt) {
 								Sync::WriteAtEndOfPipeWithInterrupt64(
-								    m_submit_id, CurrentBuffer(), dst, clock, m_interrupt_event_id,
+								    m_submit_id, command, dst, clock, m_interrupt_event_id,
 								    interrupt_context_id);
 							} else {
-								Sync::WriteAtEndOfPipeClockCounter(m_submit_id, CurrentBuffer(),
+								Sync::WriteAtEndOfPipeClockCounter(m_submit_id, command,
 								                                   dst, clock);
 							}
 							return;
@@ -1374,11 +1326,11 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 						    (eop_event_type == 0x28 && event_index == 0x00)) {
 							if (with_interrupt) {
 								Sync::WriteAtEndOfPipeWithInterruptWriteBack64(
-								    m_submit_id, CurrentBuffer(), dst, clock, m_interrupt_event_id,
+								    m_submit_id, command, dst, clock, m_interrupt_event_id,
 								    interrupt_context_id);
 							} else {
 								Sync::WriteAtEndOfPipeClockCounterWithWriteBack(
-								    m_submit_id, CurrentBuffer(), dst, clock);
+								    m_submit_id, command, dst, clock);
 							}
 							return;
 						}
@@ -1416,8 +1368,6 @@ void CommandProcessor::WriteAtEndOfPipe64(uint32_t cache_policy, uint32_t event_
 }
 
 void CommandProcessor::EmitGlobalBarrier() {
-	CheckBuffer();
-
 	Common::LockGuard lock(m_renderer.GetMutex());
 
 	vk::MemoryBarrier2 barrier {};
@@ -1434,8 +1384,6 @@ void CommandProcessor::EmitGlobalBarrier() {
 }
 
 void CommandProcessor::TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id) {
-	CheckBuffer();
-
 	Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id, interrupt_context_id);
 }
 
@@ -1519,8 +1467,6 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index,
 }
 
 void CommandProcessor::Flip() {
-	CheckBuffer();
-
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n");
 	}
@@ -1534,7 +1480,7 @@ void CommandProcessor::Flip() {
 }
 
 void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
-	CheckBuffer();
+	auto& command = CurrentBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n"
@@ -1544,7 +1490,6 @@ void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
 	}
 
 	std::memcpy(dst_gpu_addr, &value, sizeof(value));
-	auto& command = CurrentBuffer();
 	auto request = Sync::PrepareVideoOutFlip(command, m_flip.handle, m_flip.index, m_flip.flip_mode,
 	                                         m_flip.flip_arg);
 	Sync::WriteAtEndOfPipeWithFlip32(m_submit_id, command, static_cast<uint32_t*>(dst_gpu_addr),
@@ -1555,7 +1500,7 @@ void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
 
 void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache_action,
                                          void* dst_gpu_addr, uint32_t value) {
-	CheckBuffer();
+	auto& command = CurrentBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::FlipWithInterrupt()\n"
@@ -1570,7 +1515,6 @@ void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache
 		EXIT("unknown event type\n");
 	}
 	std::memcpy(dst_gpu_addr, &value, sizeof(value));
-	auto& command = CurrentBuffer();
 	auto request = Sync::PrepareVideoOutFlip(command, m_flip.handle, m_flip.index, m_flip.flip_mode,
 	                                         m_flip.flip_arg);
 	Sync::WriteAtEndOfPipeWithInterruptWriteBackFlip32(
@@ -1580,7 +1524,7 @@ void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache
 }
 
 void CommandProcessor::PrepareCpuFlip(uint64_t request_id) {
-	CheckBuffer();
+	auto& command = CurrentBuffer();
 	if (g_current_processor != nullptr) {
 		EXIT("invalid graphics-thread CPU flip preparation\n");
 	}
@@ -1590,7 +1534,7 @@ void CommandProcessor::PrepareCpuFlip(uint64_t request_id) {
 	};
 	ProcessorScope processor_scope(*this);
 
-	m_renderer.GetVideoOut().PrepareFlip(request_id, CurrentBuffer());
+	m_renderer.GetVideoOut().PrepareFlip(request_id, command);
 	GetScheduler().Flush();
 	m_renderer.GetVideoOut().CompleteFlip(request_id);
 }

src/graphics/guest_gpu/pm4.h

@@ -910,6 +910,7 @@ constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_MASK    = 0x1;
 constexpr uint32_t COMPUTE_PGM_RSRC1_FWD_PROGRESS_SHIFT = 31;
 constexpr uint32_t COMPUTE_PGM_RSRC1_FWD_PROGRESS_MASK  = 0x1;
 
+constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS = 1u << 5u;
 constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_SHIFT = 15;
 constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_MASK  = 0x1;
 

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/bufferCache.cpp

@@ -129,7 +129,7 @@ bool BufferCache::DownloadBufferMemory(Buffer& buffer, uint64_t vaddr, uint64_t
 
 	const auto [mapped, offset] = m_download_buffer.Map(total_size, 64);
 	if (mapped == nullptr) {
-		EXIT("BufferCache: download exceeds 32 MiB staging buffer capacity\n");
+		EXIT("BufferCache: download exceeds 64 MiB staging buffer capacity\n");
 	}
 	m_download_buffer.Commit();
 	for (auto& copy: copies) {
@@ -183,7 +183,7 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
       m_memory_tracker(page_manager),
       m_staging_buffer(graphics, scheduler, MemoryUsage::Upload, 512 * MiB),
       m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
-      m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
+      m_download_buffer(graphics, scheduler, MemoryUsage::Download, 64 * MiB),
       m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
       m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));

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

@@ -138,14 +138,15 @@ void NameImageBinding(GraphicContext& graphics, Image& image, vk::ImageView view
 }
 
 [[nodiscard]] std::vector<vk::BufferImageCopy> BuildDepthCopies(const ImageInfo& info,
-                                                                uint64_t         slice_stride) {
+                                                              uint64_t slice_stride,
+                                                              vk::ImageAspectFlags aspect) {
 	std::vector<vk::BufferImageCopy> copies(info.resources.layers);
 	for (uint32_t layer = 0; layer < info.resources.layers; ++layer) {
 		auto& copy             = copies[layer];
 		copy.bufferOffset      = slice_stride * layer;
 		copy.bufferRowLength   = info.pitch;
 		copy.bufferImageHeight = info.extent.height;
-		copy.imageSubresource  = {vk::ImageAspectFlagBits::eDepth, 0, layer, 1};
+		copy.imageSubresource  = {aspect, 0, layer, 1};
 		copy.imageExtent       = {info.extent.width, info.extent.height, 1};
 	}
 	return copies;
@@ -208,6 +209,10 @@ 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;
 	}
@@ -843,6 +848,16 @@ 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;
@@ -852,9 +867,6 @@ 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
@@ -1020,16 +1032,17 @@ TextureCache::ImageDownload TextureCache::BuildDownload(const Image& image) cons
 }
 
 void TextureCache::UploadImage(Image& image, Buffer& source, uint64_t source_offset) {
-	const auto& info    = image.info;
-	const auto  binding = UploadBinding(image);
+	auto& destination = image.depth_id ? m_slot_images[image.depth_id] : image;
+	const auto binding = image.depth_id ? BindingType::DepthTarget : UploadBinding(image);
 	const auto  upload  = [&](std::vector<vk::BufferImageCopy>& copies, TileManager::Result linear) {
 		for (auto& copy: copies) {
 			copy.bufferOffset += linear.offset;
 		}
-		image.Upload(copies, linear.buffer, linear.offset, linear.size);
+		destination.Upload(copies, linear.buffer, linear.offset, linear.size);
 	};
 
 	if (binding != BindingType::DepthTarget) {
+		const auto& info = image.info;
 		auto transfer = BuildTextureTransfer(image, binding, TransferDirection::Upload);
 		if (!transfer.valid) {
 			EXIT("TextureCache: invalid texture upload: binding=%u addr=0x%016" PRIx64
@@ -1053,21 +1066,32 @@ void TextureCache::UploadImage(Image& image, Buffer& source, uint64_t source_off
 		return;
 	}
 
-	if (info.samples != 1 || image.backing.samples != 1 ||
+	// The stencil plane has its own row pitch and shares the native image
+	// with the depth plane.
+	auto info = destination.info;
+	if (image.depth_id) {
+		info.data            = image.info.data;
+		info.guest_format    = Prospero::BufferFormat::k8UInt;
+		info.bytes_per_block = 1;
+		if (info.IsTiled()) info.pitch = TileGetDepthPitch(info.extent.width, 1, 0);
+	}
+	if (info.samples != 1 || destination.backing.samples != 1 ||
 	    info.resources.layers == 0 || info.data.size % info.resources.layers != 0 ||
 	    Prospero::NumBytesPerElement(info.guest_format) != info.bytes_per_block) {
 		EXIT("TextureCache: invalid depth upload\n");
 	}
 	const auto          layers          = info.resources.layers;
 	const auto          full_slice_size = info.data.size / layers;
-	auto                copies          = BuildDepthCopies(info, full_slice_size);
+	auto copies = BuildDepthCopies(info, full_slice_size, image.depth_id
+	                                                        ? vk::ImageAspectFlagBits::eStencil
+	                                                        : vk::ImageAspectFlagBits::eDepth);
 	TileManager::Result linear {source.Handle(), source_offset, source.Size() - source_offset};
 	if (info.IsTiled()) {
 		const auto tiles = BuildDepthTiles(info);
 		linear =
 		    m_tiler.Detile(source.Handle(), source_offset, info.data.size, info.data.size, tiles);
 	}
-	const auto transfer_bytes = DepthAspectTransferBytes(info.pixel_format);
+	const auto transfer_bytes = image.depth_id ? 1u : DepthAspectTransferBytes(info.pixel_format);
 	if (transfer_bytes != info.bytes_per_block) {
 		const uint64_t texels_per_slice = static_cast<uint64_t>(info.pitch) * info.extent.height;
 		EXIT_NOT_IMPLEMENTED(info.bytes_per_block != sizeof(uint16_t) ||
@@ -1186,13 +1210,20 @@ void TextureCache::MaterializeDccClear(ImageId id, const ImageDesc& desc,
 		}
 		// 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);
+		if (desc.type != BindingType::VideoOut) {
+			m_buffer_cache.FillBuffer(address, slice_size, UINT32_MAX, false);
+		}
 	}
 }
 
 void TextureCache::RefreshImage(ImageId id) {
-	TrackImage(id);
 	auto& image = m_slot_images[id];
+	if (image.depth_id &&
+	    (m_slot_images[image.depth_id].info.metadata.stencil_compressed ||
+	     m_slot_images[image.depth_id].info.samples != 1)) {
+		return;
+	}
+	TrackImage(id);
 	if (image.IsMaybeCpuDirty()) {
 		const auto hash = image.HashGuestEdges();
 		if (image.NeedsMaybeCpuHash()) {
@@ -1214,7 +1245,7 @@ void TextureCache::RefreshImage(ImageId id) {
 	InitializeImage(id);
 }
 
-void TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
+ImageId TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
 	if (!stencil.Valid()) {
 		EXIT("TextureCache: invalid stencil association range\n");
 	}
@@ -1226,7 +1257,8 @@ 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;
 		}
 	}
@@ -1239,6 +1271,7 @@ void TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
 	auto& record = m_slot_images[association];
 	TouchImage(record);
 	record.depth_id = depth_id;
+	return association;
 }
 
 ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
@@ -1300,16 +1333,6 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 			}
 		}
 		auto& image = m_slot_images[result];
-		if (desc.type == BindingType::VideoOut &&
-		    desc.info.metadata.compression != VideoOutCompression::Uncompressed) {
-			const bool guest_dirty = image.IsBufferModified() || image.IsCpuDirty();
-			const bool native_current =
-			    (image.usage.render_target || image.IsGpuModified()) && !guest_dirty;
-			if (!native_current) {
-				EXIT("TextureCache: compressed video-out read requires clean native GPU "
-				     "contents\n");
-			}
-		}
 		if (view_mip >= 0) {
 			desc.view_info.base_level = static_cast<uint32_t>(view_mip);
 		}
@@ -1320,6 +1343,18 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		TouchImage(image);
 	}
 	MaterializeDccClear(result, desc, metadata_base_layer);
+	if (desc.type == BindingType::VideoOut &&
+	    desc.info.metadata.compression != VideoOutCompression::Uncompressed) {
+		std::scoped_lock lock {m_lock};
+		const auto& image = m_slot_images[result];
+		const bool guest_dirty = image.IsBufferModified() || image.IsCpuDirty();
+		const bool native_current =
+		    (image.usage.render_target || image.IsGpuModified()) && !guest_dirty;
+		if (!native_current) {
+			EXIT("TextureCache: compressed video-out read requires clean native GPU "
+			     "contents\n");
+		}
+	}
 	return result;
 }
 
@@ -1384,6 +1419,19 @@ vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
 	}
 	if (!image.info.data.Empty()) {
 		RefreshImage(id);
+		if (image.info.HasStencil() &&
+		    desc.info.data.address >= image.info.stencil.address &&
+		    desc.info.data.End() <= image.info.stencil.End()) {
+			for (const auto stencil_id:
+			     FindImagesInRegion(image.info.stencil.address, image.info.stencil.size, false)) {
+				const auto* stencil = m_slot_images.try_get(stencil_id);
+				if (stencil != nullptr && stencil->depth_id == id &&
+				    stencil->info.data == image.info.stencil) {
+					RefreshImage(stencil_id);
+					break;
+				}
+			}
+		}
 	}
 	switch (desc.type) {
 		case BindingType::Texture: break;
@@ -1432,16 +1480,17 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	TouchImage(image);
 	image.MarkGpuModified();
 	image.usage.depth_target = true;
-	RefreshImage(id);
+	image.info.stencil = desc.info.stencil;
+	image.info.metadata = desc.info.metadata;
 	if (desc.info.HasMetadata()) {
-		image.info.metadata = desc.info.metadata;
 		m_surface_metas.emplace(desc.info.metadata.range.address,
 		                        MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
 		                                      .clear_mask = image.info.htile_clear_mask});
 	}
+	RefreshImage(id);
 	CommitGpuWrite(image);
 	if (desc.info.HasStencil()) {
-		AssociateStencil(id, desc.info.stencil);
+		RefreshImage(AssociateStencil(id, desc.info.stencil));
 	}
 	const auto view = image.FindView(desc.view_info);
 	NameImageBinding(m_graphics, image, view, desc.type, desc.view_info);
@@ -1456,11 +1505,16 @@ void TextureCache::MarkGpuWritten(ImageId id) {
 	}
 	TrackImage(id);
 	CommitGpuWrite(image);
+	if (image.info.HasStencil()) {
+		const auto stencil_id = AssociateStencil(id, image.info.stencil);
+		TrackImage(stencil_id);
+		CommitGpuWrite(m_slot_images[stencil_id]);
+	}
 }
 
 void TextureCache::CommitGpuWrite(Image& image) {
-	if (image.depth_id || image.backing.image == nullptr) {
-		EXIT("TextureCache: stencil association cannot own image contents\n");
+	if (!image.depth_id && image.backing.image == nullptr) {
+		EXIT("TextureCache: GPU writes require a native image or stencil association\n");
 	}
 	image.ClearBufferModified();
 	if (image.IsCpuDirty()) {
@@ -1544,9 +1598,10 @@ void TextureCache::ClearImage(CommandBuffer& command, ImageId id, vk::Format for
 	        range.layerCount == 0 || range.baseArrayLayer >= layers ||
 	        range.layerCount > layers - range.baseArrayLayer ||
 	        (range.aspectMask & aspects) != range.aspectMask);
-	const bool full_image = range.aspectMask == aspects && range.baseMipLevel == 0 &&
-	                        range.levelCount == image.info.resources.levels &&
-	                        range.baseArrayLayer == 0 && range.layerCount == layers;
+	const bool full_subresources = range.baseMipLevel == 0 &&
+	                               range.levelCount == image.info.resources.levels &&
+	                               range.baseArrayLayer == 0 && range.layerCount == layers;
+	const bool full_image = range.aspectMask == aspects && full_subresources;
 	TrackImage(id);
 	if (!full_image && (image.IsBufferModified() || image.IsCpuDirty())) {
 		InitializeImage(id);
@@ -1554,6 +1609,15 @@ void TextureCache::ClearImage(CommandBuffer& command, ImageId id, vk::Format for
 			EXIT("TextureCache: image clear retained guest ownership\n");
 		}
 	}
+	if (image.info.HasStencil() && (range.aspectMask & vk::ImageAspectFlagBits::eStencil)) {
+		const auto stencil_id = AssociateStencil(id, image.info.stencil);
+		if (!full_subresources) {
+			RefreshImage(stencil_id);
+		} else {
+			TrackImage(stencil_id);
+			CommitGpuWrite(m_slot_images[stencil_id]);
+		}
+	}
 	command.EndRendering();
 	// Transfer clears use the backing format; aliased clears must encode through their view.
 	if (format != image.backing.format || (image.info.IsVolume() && !full_image)) {
@@ -1626,7 +1690,7 @@ void TextureCache::DownloadDepth(Image& image, Buffer& destination, uint64_t des
 	EXIT_NOT_IMPLEMENTED(transfer_slice > UINT64_MAX / layers);
 	const uint64_t transfer_size = transfer_slice * layers;
 	EXIT_NOT_IMPLEMENTED(guest_slice > full_slice_size);
-	auto copies = BuildDepthCopies(info, full_slice_size);
+	auto copies = BuildDepthCopies(info, full_slice_size, vk::ImageAspectFlagBits::eDepth);
 	if (transfer_bytes == info.bytes_per_block) {
 		if (!info.IsTiled()) {
 			for (auto& copy: copies) {
@@ -1819,7 +1883,7 @@ void TextureCache::InvalidateMemoryFromGPU(uint64_t address, uint64_t size) {
 	std::scoped_lock lock {m_lock};
 	for (const auto id: FindImagesInRegion(address, size, true)) {
 		auto& image = m_slot_images[id];
-		if (image.depth_id || !image.Overlaps(address, size)) {
+		if (!image.Overlaps(address, size)) {
 			continue;
 		}
 		if (image.IsGpuModified()) {

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

@@ -157,7 +157,7 @@ private:
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);
 	void               CopyImage(ImageId destination, ImageId source);
-	void               AssociateStencil(ImageId depth, GuestRange stencil);
+	[[nodiscard]] ImageId AssociateStencil(ImageId depth, GuestRange stencil);
 	void CopyImageMip(ImageId destination, ImageId source, uint32_t mip, uint32_t layer);
 	void ValidateImageDesc(const ImageDesc& desc) const;
 

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

@@ -159,8 +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) {

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -41,13 +41,42 @@ struct RenderDepthInfo {
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;
 };
 
+inline vk::ImageAspectFlags DepthFeedbackAspects(vk::ImageAspectFlags draw_writes,
+                                                 const ImageViewInfo& target,
+                                                 const ImageViewInfo& sampled) {
+	if (!ImageRangeOverlaps(target.base_level, target.level_count, sampled.base_level,
+	                        sampled.level_count) ||
+	    !ImageRangeOverlaps(target.base_layer, target.layer_count, sampled.base_layer,
+	                        sampled.layer_count)) {
+		return {};
+	}
+	return draw_writes & sampled.aspect;
+}
+
+inline vk::ImageAspectFlags DepthReadableAspects(vk::ImageLayout layout) {
+	switch (layout) {
+		case vk::ImageLayout::eDepthReadOnlyOptimal:
+		case vk::ImageLayout::eDepthReadOnlyStencilAttachmentOptimal:
+			return vk::ImageAspectFlagBits::eDepth;
+		case vk::ImageLayout::eStencilReadOnlyOptimal:
+		case vk::ImageLayout::eDepthAttachmentStencilReadOnlyOptimal:
+			return vk::ImageAspectFlagBits::eStencil;
+		case vk::ImageLayout::eDepthStencilReadOnlyOptimal:
+		case vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT:
+		case vk::ImageLayout::eGeneral:
+			return vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil;
+		default:
+			return {};
+	}
+}
+
 inline vk::ImageLayout depth_attachment_layout(const RenderDepthInfo& depth) {
 	const auto available     = ImageViewOps::DepthAspectMask(depth.desc.view_info.format);
 	const auto writes        = depth.AttachmentWriteAspects();
 	const bool has_depth     = static_cast<bool>(available & vk::ImageAspectFlagBits::eDepth);
 	const bool has_stencil   = static_cast<bool>(available & vk::ImageAspectFlagBits::eStencil);
-	// LoadOp clears are separate from guest depth writes.
-	const bool depth_write   = depth.depth_write_enable;
+	// The attachment layout must permit load clears as well as draw writes.
+	const bool depth_write   = static_cast<bool>(writes & vk::ImageAspectFlagBits::eDepth);
 	const bool stencil_write = static_cast<bool>(writes & vk::ImageAspectFlagBits::eStencil);
 	if (!has_stencil) {
 		return depth_write ? vk::ImageLayout::eDepthAttachmentOptimal

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

@@ -159,6 +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     = {};
@@ -167,9 +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/pipeline/descriptors.cpp

@@ -17,6 +17,7 @@
 #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/depthRenderTarget.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
@@ -184,17 +185,18 @@ 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;
 }
@@ -252,10 +254,10 @@ 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()) ||
@@ -293,13 +295,8 @@ 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();
 }
@@ -457,6 +454,14 @@ 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
@@ -507,6 +512,21 @@ 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);
@@ -523,24 +543,28 @@ 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;
+	// IMAGE_STORE addresses BASE_LEVEL; only IMAGE_STORE_MIP selects other view mips.
+	const bool single_storage_mip =
+	    storage && resource.mip_mode != ShaderRecompiler::IR::ImageMipMode::DynamicStorage;
+	const auto view_levels = multisampled || single_storage_mip
+	                             ? 1u
+	                             : static_cast<uint32_t>(last_level - base_level) + 1u;
+	const auto levels =
+	    multisampled ? 1u : std::max(physical_levels, base_level + view_levels);
 	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) ||
@@ -558,8 +582,6 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		     descriptor.fields[6], descriptor.fields[7]);
 	}
 	const auto samples = multisampled ? 1u << last_level : 1u;
-	const auto view_levels =
-	    multisampled ? 1u : static_cast<uint32_t>(view_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);
@@ -578,7 +600,19 @@ 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);
@@ -591,8 +625,8 @@ 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);
@@ -629,7 +663,7 @@ 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};
@@ -716,18 +750,27 @@ void RenderExecutor::ResetBindings() {
 	m_bound_images.clear();
 }
 
-PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime) {
+void RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime,
+                                     PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(!runtime);
 	const auto& program  = *runtime.program;
-	const auto& snapshot = runtime.resources;
-	PreparedBindings prepared;
+	const auto& snapshot = *runtime.resources;
 	prepared.runtime = &runtime;
-	prepared.images.reserve(program.info.images.size());
+	prepared.gds = {nullptr, 0, VK_WHOLE_SIZE};
+	prepared.flattened_srt = {};
+	prepared.shader_data_buffer = {};
+	prepared.buffer_sources.clear();
+	prepared.buffers.clear();
+	prepared.images.resize(program.info.images.size());
+	prepared.samplers.clear();
+	prepared.shader_data.clear();
 	for (uint32_t i = 0; i < program.info.images.size(); i++) {
 		auto binding = ResolveTexture(program.info.images[i], snapshot.images[i]);
 		BindImage(binding.image_id, binding.desc.type == TextureCache::BindingType::Storage);
-		prepared.images.push_back(std::move(binding));
+		binding.mip_views.swap(prepared.images[i].mip_views);
+		binding.mip_views.clear();
+		prepared.images[i] = std::move(binding);
 	}
 	prepared.samplers.reserve(program.info.samplers.size());
 	for (uint32_t i = 0; i < program.info.samplers.size(); i++) {
@@ -742,14 +785,13 @@ PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runti
 	        program.bindings, ShaderRecompiler::IR::DescriptorBindingKind::Gds) != nullptr) {
 		prepared.gds.buffer = m_context.GetBufferCache().GetGdsBuffer()->Handle();
 	}
-	return prepared;
 }
 
 void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program  = *prepared.runtime->program;
-	const auto& snapshot = prepared.runtime->resources;
+	const auto& snapshot = *prepared.runtime->resources;
 	auto&       cache    = m_context.GetBufferCache();
 
 	prepared.buffer_sources.clear();
@@ -771,7 +813,7 @@ void RenderExecutor::RebindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program   = *prepared.runtime->program;
-	const auto& snapshot  = prepared.runtime->resources;
+	const auto& snapshot  = *prepared.runtime->resources;
 	const auto& layout    = program.bindings;
 	EXIT_IF(prepared.buffer_sources.size() != program.info.buffers.size());
 
@@ -806,7 +848,7 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program  = *prepared.runtime->program;
-	const auto& snapshot = prepared.runtime->resources;
+	const auto& snapshot = *prepared.runtime->resources;
 	auto&       images   = prepared.images;
 	EXIT_IF(images.size() != program.info.images.size());
 	auto& texture_cache = m_context.GetTextureCache();
@@ -838,11 +880,7 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 			}
 			binding.image_view = binding.mip_views.front();
 		} else {
-			auto desc = binding.desc;
-			if (desc.type == TextureCache::BindingType::Storage) {
-				desc.view_info.level_count = 1;
-			}
-			binding.image_view = texture_cache.FindTexture(binding.image_id, desc);
+			binding.image_view = texture_cache.FindTexture(binding.image_id, binding.desc);
 		}
 		auto&      image   = texture_cache.GetImage(binding.image_id);
 		const bool storage = binding.desc.type == TextureCache::BindingType::Storage;
@@ -960,19 +998,7 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 					    std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
 					EXIT_IF(storage || host_view == image.views.end());
 					const auto aspect = host_view->info.aspect;
-					const bool depth_feedback =
-					    layout == vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT &&
-					    program.stage == ShaderType::Pixel;
-					const bool depth_read =
-					    depth_feedback || layout == vk::ImageLayout::eDepthReadOnlyOptimal ||
-					    layout == vk::ImageLayout::eDepthStencilReadOnlyOptimal ||
-					    layout == vk::ImageLayout::eDepthReadOnlyStencilAttachmentOptimal;
-					const bool stencil_read =
-					    layout == vk::ImageLayout::eStencilReadOnlyOptimal ||
-					    layout == vk::ImageLayout::eDepthStencilReadOnlyOptimal ||
-					    layout == vk::ImageLayout::eDepthAttachmentStencilReadOnlyOptimal;
-					if ((aspect & vk::ImageAspectFlagBits::eDepth && !depth_read) ||
-					    (aspect & vk::ImageAspectFlagBits::eStencil && !stencil_read)) {
+					if (aspect & ~DepthReadableAspects(layout)) {
 						EXIT("sampling a writable depth/stencil attachment aspect\n");
 					}
 				}

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

@@ -92,9 +92,9 @@ void PipelineCacheLog(fmt::format_string<Args...> format, Args&&... args) {
 	Log::WriteToConsoleAndLog(message);
 }
 
-bool ReadShaderGuestMemory(void*, uint64_t address, uint32_t* value) {
-	return value != nullptr &&
-	       Libs::LibKernel::Memory::TryReadGpuCleanBacking(address, value, sizeof(*value));
+bool ReadShaderGuestMemory(void*, uint64_t address, std::span<uint32_t> values) {
+	return !values.empty() &&
+	       Libs::LibKernel::Memory::TryReadGpuCleanBacking(address, values.data(), values.size_bytes());
 }
 
 void DumpShaderSpirv(const char* stage_name, uint64_t shader_hash,
@@ -200,8 +200,10 @@ struct PipelineCache::ProgramCache {
 			permutations.reserve(8);
 		}
 
-		ShaderRecompiler::IR::ResourcePlan resource_plan;
-		std::vector<Permutation>           permutations;
+		ShaderRecompiler::IR::ResourcePlan           resource_plan;
+		ShaderRecompiler::IR::ResourceSnapshot       resources;
+		ShaderRecompiler::IR::ResourceSpecialization specialization;
+		std::vector<Permutation>                    permutations;
 	};
 
 	struct ProgramKeyHash {
@@ -274,33 +276,33 @@ struct PipelineCache::ProgramCache {
 			stage = ShaderType::Compute;
 		}
 
+		const auto user_data = std::span(params.user_data).first(params.user_data_count);
 		lookup_key.stage           = stage;
 		lookup_key.hash            = params.hash;
-		lookup_key.user_data_count = static_cast<uint32_t>(params.user_data.size());
+		lookup_key.user_data_count = params.user_data_count;
 		lookup_key.code_size       = static_cast<uint32_t>(params.code.size());
 		BuildStageStaticKey(input_info, lookup_key.static_state);
 		auto                                         entry = programs.find(lookup_key);
-		ShaderRecompiler::IR::ResourceSnapshot       resources;
-		ShaderRecompiler::IR::ResourceSpecialization specialization;
 		const ShaderRecompiler::IR::SrtRuntime       runtime {
-		    .user_data                  = params.user_data,
+		    .user_data                  = user_data,
 		    .shader_base                = params.Base(),
 		    .read_specialization_memory = ReadShaderGuestMemory,
 		};
 		if (entry != programs.end()) {
 			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(
-			    entry->second.resource_plan, runtime, resources, specialization));
+			    entry->second.resource_plan, runtime, entry->second.resources,
+			    entry->second.specialization));
 			if (const auto permutation = std::ranges::find_if(
 			        entry->second.permutations, [&](const Permutation& candidate) {
 				        const auto& layout = candidate.program.bindings;
 				        return layout.push_data_start_dword ==
 				                   ShaderRecompiler::IR::PushData::StartFor(
 				                       push_data_cursor, layout.ShaderDataDwords()) &&
-				               candidate.specialization == specialization;
+				               candidate.specialization == entry->second.specialization;
 			        });
 			    permutation != entry->second.permutations.end()) {
 				input_info.stage = {.program   = &permutation->program,
-				                    .resources = std::move(resources)};
+				                    .resources = &entry->second.resources};
 				permutation->program.bindings.AdvancePushData(push_data_cursor);
 				return permutation->handle;
 			}
@@ -328,7 +330,7 @@ struct PipelineCache::ProgramCache {
 		ShaderRecompiler::CompileOptions options;
 		options.stage       = stage;
 		options.shader_hash = params.hash;
-		options.user_data   = params.user_data;
+		options.user_data   = user_data;
 		options.back_code      = params.back_code;
 		options.dump_ir     = Config::GetShaderLogDirection() != Config::LogDirection::Silent;
 		options.early_dump  = options.dump_ir;
@@ -337,6 +339,7 @@ struct PipelineCache::ProgramCache {
 
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
 			options.user_data_base = 8;
+			options.wave_size = input_info.wave_size;
 			if (stage == ShaderType::Mesh || stage == ShaderType::TessellationControl) {
 				options.user_data_base = 0;
 				options.wave_size = stage == ShaderType::Mesh ? input_info.mesh.wave_size : 64u;
@@ -346,15 +349,16 @@ struct PipelineCache::ProgramCache {
 		}
 		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
 		if (entry == programs.end()) {
-			auto resource_plan = ShaderRecompiler::IR::ExtractResourcePlan(translated.program);
-			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(resource_plan, runtime, resources,
-			                                                    specialization));
-			entry = programs.try_emplace(lookup_key, std::move(resource_plan)).first;
+			entry = programs.try_emplace(lookup_key,
+			    ShaderRecompiler::IR::ExtractResourcePlan(translated.program)).first;
+			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(
+			    entry->second.resource_plan, runtime, entry->second.resources,
+			    entry->second.specialization));
 		}
 		entry->second.permutations.push_back(CompilePermutation(
-		    params, options, std::move(translated), std::move(specialization), push_data_cursor));
+		    params, options, std::move(translated), entry->second.specialization, push_data_cursor));
 		const auto& permutation = entry->second.permutations.back();
-		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
+		input_info.stage = {.program = &permutation.program, .resources = &entry->second.resources};
 		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
 		std::array<size_t, static_cast<size_t>(ShaderType::TessellationEvaluation) + 1> counts {};
@@ -593,6 +597,21 @@ 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];
@@ -666,13 +685,16 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	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) {
@@ -681,14 +703,14 @@ 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);
@@ -737,8 +759,7 @@ 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;
-	const bool rect_list =
-	    command.GetUserConfig().GetPrimType() == Prospero::PrimitiveType::kRectList;
+	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;

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

@@ -46,8 +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 |
@@ -58,15 +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;
@@ -90,7 +90,7 @@ vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
 	EXIT_IF(!runtime);
 	const auto& program   = *runtime.program;
-	const auto& resources = runtime.resources;
+	const auto& resources = *runtime.resources;
 	EXIT_IF(resources.buffers.size() != program.info.buffers.size());
 	bool has_writes = false;
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {

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

@@ -10,9 +10,9 @@ 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/render.h

@@ -45,6 +45,7 @@ enum class CommandBufferDebugOp : uint32_t {
 	EopFlip,
 	EopWriteBackFlip,
 	EopOnlyFlip,
+	DispatchIndirect,
 	Unknown,
 };
 
@@ -156,8 +157,10 @@ public:
 
 	void DispatchDirect(uint64_t submit_id, CommandBuffer& buffer, uint32_t thread_group_x,
 	                    uint32_t thread_group_y, uint32_t thread_group_z, uint32_t mode);
+	void DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer, uint64_t args_addr,
+	                      uint32_t mode);
 
-	[[nodiscard]] PreparedBindings PrepareBindings(const ShaderStageRuntime& runtime);
+	void PrepareBindings(const ShaderStageRuntime& runtime, PreparedBindings& prepared);
 	void                           FindBuffers(PreparedBindings& bindings);
 	void                           RebindBuffers(PreparedBindings& bindings);
 	void                           RebindImages(PreparedBindings& bindings);
@@ -193,7 +196,8 @@ private:
 	                         bool primitive_restart_enable);
 	[[nodiscard]] RenderState AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
 	                                               uint32_t color_count, RenderDepthInfo& depth,
-	                                               const std::optional<PreparedBindings>& pixel = std::nullopt);
+	                                               vk::ImageAspectFlags& feedback_aspects,
+	                                               std::span<PreparedBindings* const> stages = {});
 	[[nodiscard]] bool        ResolveColorTargets(CommandBuffer& buffer,
 	                                              uint32_t render_target_slice_offset);
 	void                      BindImage(ImageId id, bool storage);
@@ -206,6 +210,8 @@ private:
 	                                              uint32_t group_y, uint32_t group_z, uint32_t mode);
 
 	RenderContext&                        m_context;
+	GraphicsBindings                     m_graphics_bindings;
+	PreparedBindings                     m_compute_bindings;
 	std::vector<ImageId>                  m_bound_images;
 	std::vector<vk::DescriptorBufferInfo> m_descriptor_buffers;
 	std::vector<vk::DescriptorImageInfo>  m_descriptor_images;

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -9,6 +9,7 @@
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/guest_gpu/hardwareContext.h"
+#include "graphics/guest_gpu/pm4.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/image/imageInfo.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
@@ -51,7 +52,7 @@ static bool FillSourcesDisjoint(std::span<const ShaderRecompiler::IR::Descriptor
 bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
                                                 const CommandBuffer&          buffer) {
 	const auto& program   = *input.stage.program;
-	const auto& resources = input.stage.resources;
+	const auto& resources = *input.stage.resources;
 	if (resources.buffers.size() != program.info.buffers.size()) {
 		EXIT("compute runtime buffer count does not match shader metadata\n");
 	}
@@ -85,7 +86,7 @@ bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t grou
                               uint32_t group_y, uint32_t group_z, uint32_t mode,
                               ShaderBufferResource& resolved_descriptor, uint32_t& resolved_clear,
                               uint64_t& resolved_size) {
-	const auto& resources = input.stage.resources;
+	const auto& resources = *input.stage.resources;
 	const auto& fill      = resources.uniform_fill;
 	if (fill.kind != ShaderRecompiler::IR::UniformFillKind::Buffer) {
 		return false;
@@ -128,7 +129,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
                                                 CommandBuffer& command, uint32_t group_x,
                                                 uint32_t group_y, uint32_t group_z, uint32_t mode) {
 	const auto& program   = *input.stage.program;
-	const auto& resources = input.stage.resources;
+	const auto& resources = *input.stage.resources;
 	const auto& fill      = resources.uniform_fill;
 	auto&       cache     = command.GetContext().GetTextureCache();
 	if (fill.kind == ShaderRecompiler::IR::UniformFillKind::Image) {
@@ -206,6 +207,17 @@ 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, "
@@ -214,7 +226,7 @@ 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;
 	}
 
@@ -251,7 +263,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	}
 
 	const auto& program   = *input_info.stage.program;
-	const auto& resources = input_info.stage.resources;
+	const auto& resources = *input_info.stage.resources;
 	if (TryConsumeComputeMetaClear(input_info, buffer)) {
 		ResetBindings();
 		return;
@@ -347,21 +359,11 @@ 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);
-	auto bindings = PrepareBindings(input_info.stage);
+	auto& bindings = m_compute_bindings;
+	PrepareBindings(input_info.stage, bindings);
 	FindBuffers(bindings);
 	if (program.info.uses_dma) {
 		m_context.PrepareBda();
@@ -395,4 +397,61 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	ResetBindings();
 }
 
+void RenderExecutor::DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer,
+                                      uint64_t args_addr, uint32_t mode) {
+	EXIT_IF(buffer.IsInvalid() || args_addr == 0 || (args_addr & 3u) != 0 ||
+	        (mode & Pm4::COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS) != 0);
+	m_context.GetCommandScheduler().PopPendingOperations();
+	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::DispatchIndirect), submit_id,
+	                    static_cast<uint32_t>(args_addr), static_cast<uint32_t>(args_addr >> 32u),
+	                    0, mode, buffer.GetShaders().GetCs().cs_regs.data_addr);
+	Common::LockGuard lock(m_context.GetMutex());
+	const auto& cs_regs = buffer.GetShaders().GetCs();
+	if (cs_regs.cs_regs.data_addr == 0) {
+		return;
+	}
+	ShaderComputeInputInfo input_info {};
+	const auto compute_program = m_context.GetPipelineCache().GetComputeProgram(
+	    cs_regs, buffer.GetRegisters().GetShaderRegisters(), input_info);
+	buffer.EndRendering();
+	auto& pipeline = m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
+	auto& bindings = m_compute_bindings;
+	PrepareBindings(input_info.stage, bindings);
+	FindBuffers(bindings);
+	const auto& program = *input_info.stage.program;
+	if (program.info.uses_dma) {
+		m_context.PrepareBda();
+	}
+	RebindImages(bindings);
+	// Acquiring arguments can merge cache buffers; finalize shader bindings afterward.
+	const auto [args_buffer, args_offset] = m_context.GetBufferCache().ObtainBuffer(
+	    args_addr, sizeof(vk::DispatchIndirectCommand), false);
+	EXIT_IF(args_buffer == nullptr || (args_offset & 3u) != 0);
+	RebindBuffers(bindings);
+	PreparedBindings* descriptor_stage = &bindings;
+	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline,
+	               std::span {&descriptor_stage, 1u});
+	const auto vk_buffer = buffer.Handle();
+	const bool has_storage_writes = HasShaderBufferWrites(input_info.stage) ||
+	    std::any_of(program.info.images.begin(), program.info.images.end(), [](const auto& image) {
+		    return image.written && image.resource_class ==
+		                                ShaderRecompiler::IR::ImageResourceClass::Storage;
+	    });
+	if (has_storage_writes) {
+		ShaderWriteHazardBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
+	}
+	vk::MemoryBarrier barrier {};
+	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite | vk::AccessFlagBits::eTransferWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits::eIndirectCommandRead;
+	vk_buffer.pipelineBarrier(vk::PipelineStageFlagBits::eAllGraphics |
+	                              vk::PipelineStageFlagBits::eComputeShader |
+	                              vk::PipelineStageFlagBits::eTransfer,
+	                          vk::PipelineStageFlagBits::eDrawIndirect, {},
+	                          1, &barrier, 0, nullptr, 0, nullptr);
+	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
+	vk_buffer.dispatchIndirect(args_buffer->Handle(), args_offset);
+	ShaderAccessBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
+	ResetBindings();
+}
+
 } // namespace Libs::Graphics

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"
@@ -56,7 +58,7 @@ std::pair<int32_t, uint32_t> ResolveDrawOffsets(uint32_t index_offset,
 
 	EXIT_IF(!vs_input_info.stage);
 	const auto& program   = *vs_input_info.stage.program;
-	const auto& resources = vs_input_info.stage.resources;
+	const auto& resources = *vs_input_info.stage.resources;
 	if (index_offset == 0 &&
 	    program.info.vertex_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
 		const auto index =
@@ -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) {
@@ -420,15 +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
 }
@@ -436,7 +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) {
@@ -467,15 +453,16 @@ struct DrawCallInfo {
 
 RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
                                                  uint32_t color_count, RenderDepthInfo& depth,
-                                                 const std::optional<PreparedBindings>& pixel) {
+                                                 vk::ImageAspectFlags& feedback_aspects,
+                                                 std::span<PreparedBindings* const> stages) {
 	EXIT_IF(colors == nullptr || color_count > RENDER_COLOR_ATTACHMENTS_MAX);
+	feedback_aspects = {};
 	auto&       cache = m_context.GetTextureCache();
 	RenderState state {};
 	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);
@@ -496,12 +483,6 @@ 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;
@@ -516,7 +497,8 @@ 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;
 	}
@@ -549,34 +531,29 @@ 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 auto draw_writes = depth.AttachmentWriteAspects();
+		vk::ImageAspectFlags sampled_aspects;
+		for (const auto* stage: stages) {
+			for (const auto& binding: stage->images) {
+				if (binding.image_id != depth.image_id ||
+				    binding.desc.type != TextureCache::BindingType::Texture) continue;
+				const auto native =
+				    std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
+				EXIT_IF(native == image.views.end());
+				sampled_aspects |= native->info.aspect;
+				feedback_aspects |= DepthFeedbackAspects(draw_writes, depth.desc.view_info,
+				                                         native->info);
+			}
 		}
-		const bool feedback = depth.depth_write_enable && pixel &&
-		    std::ranges::any_of(pixel->images, [&](const TextureBinding& binding) {
-			    if (binding.image_id != depth.image_id ||
-			        binding.desc.type != TextureCache::BindingType::Texture) {
-				    return false;
-			    }
-			    const auto native =
-			        std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
-			    EXIT_IF(native == image.views.end());
-			    const auto& sampled = native->info;
-			    const auto& target = depth.desc.view_info;
-			    return (sampled.aspect & vk::ImageAspectFlagBits::eDepth) &&
-			           ImageRangeOverlaps(sampled.base_level, sampled.level_count,
-			                              target.base_level, target.level_count) &&
-			           ImageRangeOverlaps(sampled.base_layer, sampled.layer_count,
-			                              target.base_layer, target.layer_count);
-		    });
-		if (feedback && !m_context.GetGraphics().attachment_feedback_loop_enabled) {
+		if (feedback_aspects && !m_context.GetGraphics().attachment_feedback_loop_enabled) {
 			EXIT("depth attachment feedback loop is not supported by the host\n");
 		}
-		const auto layout = feedback ? vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
-		                             : depth_attachment_layout(depth);
+		auto layout = depth_attachment_layout(depth);
+		if (sampled_aspects & ~DepthReadableAspects(layout)) {
+			layout = m_context.GetGraphics().attachment_feedback_loop_enabled
+			             ? vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
+			             : vk::ImageLayout::eGeneral;
+		}
 		// The attachment store writes even when guest depth/stencil tests do not.
 		const auto access = vk::AccessFlagBits2::eDepthStencilAttachmentRead |
 		                    vk::AccessFlagBits2::eDepthStencilAttachmentWrite;
@@ -605,9 +582,6 @@ 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;
@@ -622,7 +596,7 @@ 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));
 }
 
@@ -803,8 +777,7 @@ 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;
 
@@ -827,14 +800,11 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 			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;
@@ -891,42 +861,6 @@ 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();
@@ -940,8 +874,14 @@ 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()) {
@@ -952,6 +892,45 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
 	    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,
                                               const DrawIndexBufferSource& source) {
 	PreparedIndexBuffer prepared;
@@ -1004,8 +983,7 @@ 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;
 	}
 
@@ -1018,8 +996,7 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 }
 
 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:
@@ -1028,6 +1005,7 @@ 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,
@@ -1037,14 +1015,6 @@ 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) {
@@ -1073,7 +1043,13 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	uint32_t   mesh_groups = 0;
 	if (mesh_active) {
 		const auto& mesh = state.vertex_info[0].mesh;
-		if (primitive_restart_enable || mesh.primitives_per_group == 0) {
+		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);
 		}
@@ -1100,15 +1076,16 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		                              index_source.guest_element_size);
 	}
 	LogDrawPhase(draw.Name(), "PrepareBindings");
-	GraphicsBindings                 bindings;
+	auto&                            bindings = m_graphics_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);
+		PrepareBindings(state.vertex_info[i].stage, bindings.vertex[i]);
 		descriptor_stages[stage_count++] = &bindings.vertex[i];
 	}
 	if (state.ps_active) {
-		bindings.pixel.emplace(PrepareBindings(state.ps_input_info.stage));
+		if (!bindings.pixel) bindings.pixel.emplace();
+		PrepareBindings(state.ps_input_info.stage, *bindings.pixel);
 		descriptor_stages[stage_count++] = &*bindings.pixel;
 	}
 	const auto stages = std::span {descriptor_stages.data(), stage_count};
@@ -1120,10 +1097,6 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		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");
 	}
@@ -1131,6 +1104,10 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	    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);
+	vk::ImageAspectFlags feedback_aspects;
+	const auto rendering =
+	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info,
+	                         feedback_aspects, stages);
 
 	// 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
@@ -1140,7 +1117,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (!mesh_active) {
 		CommitVertexBuffers(vk_buffer, vertex_bindings);
 	}
-	if (bindings.pixel && !draw.IsIndexed()) {
+	if (state.ps_active && !draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
 	}
 	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline, stages);
@@ -1160,14 +1137,9 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		CommitIndexBuffer(vk_buffer, index_binding);
 	}
 
-	SetGraphicsDynamicParams(buffer, vk_buffer, vertex_stages.back(), 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 ==
-		            vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
-		        ? vk::ImageAspectFlags {vk::ImageAspectFlagBits::eDepth}
-		        : vk::ImageAspectFlags {});
+		vk_buffer.setAttachmentFeedbackLoopEnableEXT(feedback_aspects);
 	}
 
 	LogDrawPhase(draw.Name(), "BeginRendering");
@@ -1182,7 +1154,7 @@ 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.vertex_info[0], draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, draw, emit);
 	}
 
 	if (!draw.IsIndexed()) {
@@ -1226,7 +1198,8 @@ 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;
 	}
@@ -1257,7 +1230,7 @@ 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;
 	}
 
@@ -1300,8 +1273,6 @@ 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);
 
@@ -1338,7 +1309,8 @@ 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;
 	}
@@ -1367,20 +1339,18 @@ 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 = ucfg.GetPrimType() == Prospero::PrimitiveType::kRectList;
+	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) {

src/graphics/host_gpu/vulkanCommon.h

@@ -24,8 +24,6 @@
 
 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);

src/graphics/presentation/systemOverlay.cpp

@@ -1,6 +1,7 @@
 #include "graphics/presentation/systemOverlay.h"
 
-#include "SDL.h"
+#include <SDL3/SDL.h>
+
 #include "common/assert.h"
 #include "common/stringUtils.h"
 #include "graphics/host_gpu/graphicContext.h"
@@ -176,6 +177,7 @@ bool                         g_input_multiline            = false;
 bool                         g_input_lifecycle_active     = false;
 bool                         g_controller_captured        = false;
 OverlaySession               g_session;
+SDL_Window*                  g_input_window               = nullptr;
 
 void ClearInputEvents() {
 	std::scoped_lock lock(g_input_mutex);
@@ -212,7 +214,7 @@ void RetryVisibilityWakeup() {
 	}
 	SDL_Event event {};
 	event.type = type;
-	if (SDL_PushEvent(&event) > 0) {
+	if (SDL_PushEvent(&event)) {
 		g_missing_visibility_wakeups--;
 	}
 }
@@ -252,7 +254,7 @@ void RefreshVisibility() {
 		std::scoped_lock input_lock(g_input_mutex);
 		g_visibility_updates.push_back({snapshot.session, visible, capture_controller,
 		                                capture_keyboard, text_input, multiline});
-		if (g_missing_visibility_wakeups != 0 || SDL_PushEvent(&event) <= 0) {
+		if (g_missing_visibility_wakeups != 0 || !SDL_PushEvent(&event)) {
 			g_missing_visibility_wakeups++;
 		}
 	}
@@ -268,16 +270,16 @@ void OnDialogVisibilityChanged(bool, uint64_t) {
 
 uint32_t ExternalKeyStatus(SDL_Keymod modifiers, bool character_valid) {
 	uint32_t status = 0x00000001 | (character_valid ? 0x00000002 : 0);
-	if ((modifiers & KMOD_LCTRL) != 0) status |= 0x00000100;
-	if ((modifiers & KMOD_LSHIFT) != 0) status |= 0x00000200;
-	if ((modifiers & KMOD_LALT) != 0) status |= 0x00000400;
-	if ((modifiers & KMOD_LGUI) != 0) status |= 0x00000800;
-	if ((modifiers & KMOD_RCTRL) != 0) status |= 0x00001000;
-	if ((modifiers & KMOD_RSHIFT) != 0) status |= 0x00002000;
-	if ((modifiers & KMOD_RALT) != 0) status |= 0x00004000;
-	if ((modifiers & KMOD_RGUI) != 0) status |= 0x00008000;
-	if ((modifiers & KMOD_NUM) != 0) status |= 0x00010000;
-	if ((modifiers & KMOD_CAPS) != 0) status |= 0x00020000;
+	if ((modifiers & SDL_KMOD_LCTRL) != 0) status |= 0x00000100;
+	if ((modifiers & SDL_KMOD_LSHIFT) != 0) status |= 0x00000200;
+	if ((modifiers & SDL_KMOD_LALT) != 0) status |= 0x00000400;
+	if ((modifiers & SDL_KMOD_LGUI) != 0) status |= 0x00000800;
+	if ((modifiers & SDL_KMOD_RCTRL) != 0) status |= 0x00001000;
+	if ((modifiers & SDL_KMOD_RSHIFT) != 0) status |= 0x00002000;
+	if ((modifiers & SDL_KMOD_RALT) != 0) status |= 0x00004000;
+	if ((modifiers & SDL_KMOD_RGUI) != 0) status |= 0x00008000;
+	if ((modifiers & SDL_KMOD_NUM) != 0) status |= 0x00010000;
+	if ((modifiers & SDL_KMOD_CAPS) != 0) status |= 0x00020000;
 	return status;
 }
 
@@ -392,14 +394,14 @@ float AlignmentPivot(Ime::Alignment alignment) {
 
 ImGuiKey ControllerButtonToKey(int button) {
 	switch (button) {
-		case SDL_CONTROLLER_BUTTON_A: return ImGuiKey_GamepadFaceDown;
-		case SDL_CONTROLLER_BUTTON_B: return ImGuiKey_GamepadFaceRight;
-		case SDL_CONTROLLER_BUTTON_X: return ImGuiKey_GamepadFaceLeft;
-		case SDL_CONTROLLER_BUTTON_Y: return ImGuiKey_GamepadFaceUp;
-		case SDL_CONTROLLER_BUTTON_DPAD_LEFT: return ImGuiKey_GamepadDpadLeft;
-		case SDL_CONTROLLER_BUTTON_DPAD_RIGHT: return ImGuiKey_GamepadDpadRight;
-		case SDL_CONTROLLER_BUTTON_DPAD_UP: return ImGuiKey_GamepadDpadUp;
-		case SDL_CONTROLLER_BUTTON_DPAD_DOWN: return ImGuiKey_GamepadDpadDown;
+		case SDL_GAMEPAD_BUTTON_SOUTH: return ImGuiKey_GamepadFaceDown;
+		case SDL_GAMEPAD_BUTTON_EAST: return ImGuiKey_GamepadFaceRight;
+		case SDL_GAMEPAD_BUTTON_WEST: return ImGuiKey_GamepadFaceLeft;
+		case SDL_GAMEPAD_BUTTON_NORTH: return ImGuiKey_GamepadFaceUp;
+		case SDL_GAMEPAD_BUTTON_DPAD_LEFT: return ImGuiKey_GamepadDpadLeft;
+		case SDL_GAMEPAD_BUTTON_DPAD_RIGHT: return ImGuiKey_GamepadDpadRight;
+		case SDL_GAMEPAD_BUTTON_DPAD_UP: return ImGuiKey_GamepadDpadUp;
+		case SDL_GAMEPAD_BUTTON_DPAD_DOWN: return ImGuiKey_GamepadDpadDown;
 		default: return ImGuiKey_None;
 	}
 }
@@ -415,9 +417,11 @@ void CheckVulkanResult(VkResult result) {
 
 } // namespace
 
-void InitializeSystemOverlayInput() {
+void InitializeSystemOverlayInput(SDL_Window* window) {
+	EXIT_IF(window == nullptr);
 	const Uint32 type = SDL_RegisterEvents(1);
 	EXIT_IF(type == static_cast<Uint32>(-1));
+	g_input_window = window;
 	{
 		std::scoped_lock lock(g_visibility_mutex);
 		g_input_lifecycle_active = true;
@@ -453,9 +457,10 @@ void ShutdownSystemOverlayInput() {
 	g_input_controller = false;
 	g_input_keyboard   = false;
 	g_input_multiline  = false;
-	if (SDL_IsTextInputActive() == SDL_TRUE) {
-		SDL_StopTextInput();
+	if (g_input_window != nullptr && SDL_TextInputActive(g_input_window)) {
+		SDL_StopTextInput(g_input_window);
 	}
+	g_input_window = nullptr;
 }
 
 SystemOverlayVisualState GetSystemOverlayVisualState() noexcept {
@@ -487,13 +492,13 @@ bool ProcessSystemOverlayInput(const SDL_Event& event) {
 		g_input_keyboard        = update.capture_keyboard;
 		g_input_multiline       = update.multiline;
 		if (update.text_input) {
-			SDL_StartTextInput();
-		} else if (SDL_IsTextInputActive() == SDL_TRUE) {
-			SDL_StopTextInput();
+			SDL_StartTextInput(g_input_window);
+		} else if (SDL_TextInputActive(g_input_window)) {
+			SDL_StopTextInput(g_input_window);
 		}
 		return true;
 	}
-	if (event.type == SDL_CONTROLLERDEVICEREMOVED) {
+	if (event.type == SDL_EVENT_GAMEPAD_REMOVED) {
 		if (g_input_active && g_input_controller) {
 			QueueInput({InputKind::ResetController, g_input_session, 0, 0.0f, 0.0f});
 		}
@@ -505,26 +510,26 @@ bool ProcessSystemOverlayInput(const SDL_Event& event) {
 
 	const auto     session          = g_input_session;
 	const uint64_t generation       = session.generation;
-	const bool     controller_event = event.type == SDL_CONTROLLERBUTTONDOWN ||
-	                                  event.type == SDL_CONTROLLERBUTTONUP ||
-	                                  event.type == SDL_CONTROLLERAXISMOTION;
+	const bool     controller_event = event.type == SDL_EVENT_GAMEPAD_BUTTON_DOWN ||
+	                                  event.type == SDL_EVENT_GAMEPAD_BUTTON_UP ||
+	                                  event.type == SDL_EVENT_GAMEPAD_AXIS_MOTION;
 	if (controller_event && !g_input_controller) {
 		return false;
 	}
-	const bool keyboard_event = event.type == SDL_TEXTINPUT || event.type == SDL_TEXTEDITING ||
-	                            event.type == SDL_KEYDOWN || event.type == SDL_KEYUP;
+	const bool keyboard_event = event.type == SDL_EVENT_TEXT_INPUT || event.type == SDL_EVENT_TEXT_EDITING ||
+	                            event.type == SDL_EVENT_KEY_DOWN || event.type == SDL_EVENT_KEY_UP;
 	if (keyboard_event && !g_input_keyboard) {
 		return false;
 	}
 	if (keyboard_event && session.kind == OverlayKind::Error) {
-		if (event.type == SDL_KEYDOWN && event.key.repeat == 0 &&
-		    (event.key.keysym.sym == SDLK_RETURN || event.key.keysym.sym == SDLK_KP_ENTER)) {
+		if (event.type == SDL_EVENT_KEY_DOWN && !event.key.repeat &&
+		    (event.key.key == SDLK_RETURN || event.key.key == SDLK_KP_ENTER)) {
 			ErrorDialog::HostAccept(generation);
 		}
 		return true;
 	}
 	switch (event.type) {
-		case SDL_TEXTINPUT: {
+		case SDL_EVENT_TEXT_INPUT: {
 			const auto text = Utf8ToUtf16(event.text.text);
 			if (!text.empty()) {
 				auto input = MakeExternalInput(Ime::ExternalAction::Text, g_last_external_keycode,
@@ -535,26 +540,24 @@ bool ProcessSystemOverlayInput(const SDL_Event& event) {
 			}
 			return true;
 		}
-		case SDL_TEXTEDITING: return true;
-		case SDL_KEYDOWN: {
-			g_last_external_keycode = static_cast<uint16_t>(event.key.keysym.scancode);
-			g_last_external_status =
-			    ExternalKeyStatus(static_cast<SDL_Keymod>(event.key.keysym.mod), false);
+		case SDL_EVENT_TEXT_EDITING: return true;
+		case SDL_EVENT_KEY_DOWN: {
+			g_last_external_keycode = static_cast<uint16_t>(event.key.scancode);
+			g_last_external_status  = ExternalKeyStatus(event.key.mod, false);
 			auto action = Ime::ExternalAction::Text;
 			bool queue  = true;
-			if (event.key.keysym.sym == SDLK_BACKSPACE) {
+			if (event.key.key == SDLK_BACKSPACE) {
 				action = Ime::ExternalAction::Backspace;
-			} else if (event.key.keysym.sym == SDLK_LEFT) {
+			} else if (event.key.key == SDLK_LEFT) {
 				action = Ime::ExternalAction::MoveLeft;
-			} else if (event.key.keysym.sym == SDLK_RIGHT) {
+			} else if (event.key.key == SDLK_RIGHT) {
 				action = Ime::ExternalAction::MoveRight;
-			} else if (event.key.keysym.sym == SDLK_ESCAPE) {
+			} else if (event.key.key == SDLK_ESCAPE) {
 				action = Ime::ExternalAction::Cancel;
-			} else if (event.key.keysym.sym == SDLK_RETURN ||
-			           event.key.keysym.sym == SDLK_KP_ENTER) {
+			} else if (event.key.key == SDLK_RETURN || event.key.key == SDLK_KP_ENTER) {
 				action =
 				    g_input_multiline ? Ime::ExternalAction::Newline : Ime::ExternalAction::Accept;
-			} else if (event.key.keysym.sym == SDLK_TAB) {
+			} else if (event.key.key == SDLK_TAB) {
 				action = Ime::ExternalAction::None;
 			} else {
 				queue = false;
@@ -566,28 +569,28 @@ bool ProcessSystemOverlayInput(const SDL_Event& event) {
 			}
 			return true;
 		}
-		case SDL_KEYUP: return true;
-		case SDL_CONTROLLERBUTTONDOWN:
-		case SDL_CONTROLLERBUTTONUP:
-			QueueInput({InputKind::Button, session, event.cbutton.button,
-			            event.type == SDL_CONTROLLERBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
+		case SDL_EVENT_KEY_UP: return true;
+		case SDL_EVENT_GAMEPAD_BUTTON_DOWN:
+		case SDL_EVENT_GAMEPAD_BUTTON_UP:
+			QueueInput({InputKind::Button, session, event.gbutton.button,
+			            event.type == SDL_EVENT_GAMEPAD_BUTTON_DOWN ? 1.0f : 0.0f, 0.0f});
 			return true;
-		case SDL_CONTROLLERAXISMOTION:
-			QueueInput({InputKind::Axis, session, event.caxis.axis,
-			            static_cast<float>(event.caxis.value), 0.0f});
+		case SDL_EVENT_GAMEPAD_AXIS_MOTION:
+			QueueInput({InputKind::Axis, session, event.gaxis.axis,
+			            static_cast<float>(event.gaxis.value), 0.0f});
 			return true;
-		case SDL_MOUSEMOTION:
+		case SDL_EVENT_MOUSE_MOTION:
 			QueueInput({InputKind::MousePosition, session, 0, static_cast<float>(event.motion.x),
 			            static_cast<float>(event.motion.y)});
 			return true;
-		case SDL_MOUSEBUTTONDOWN:
-		case SDL_MOUSEBUTTONUP:
+		case SDL_EVENT_MOUSE_BUTTON_DOWN:
+		case SDL_EVENT_MOUSE_BUTTON_UP:
 			QueueInput({InputKind::MousePosition, session, 0, static_cast<float>(event.button.x),
 			            static_cast<float>(event.button.y)});
 			QueueInput({InputKind::MouseButton, session, event.button.button,
-			            event.type == SDL_MOUSEBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
+			            event.type == SDL_EVENT_MOUSE_BUTTON_DOWN ? 1.0f : 0.0f, 0.0f});
 			return true;
-		case SDL_MOUSEWHEEL:
+		case SDL_EVENT_MOUSE_WHEEL:
 			QueueInput({InputKind::MouseWheel, session, 0, static_cast<float>(event.wheel.x),
 			            static_cast<float>(event.wheel.y)});
 			return true;
@@ -681,9 +684,9 @@ struct SystemOverlay::Impl {
 				case InputKind::Button: {
 					const bool down = event.x != 0.0f;
 					if (session.kind == OverlayKind::Ime && down) {
-						if (event.id == SDL_CONTROLLER_BUTTON_B) {
+						if (event.id == SDL_GAMEPAD_BUTTON_EAST) {
 							Ime::HostCancel(session.generation);
-						} else if (event.id == SDL_CONTROLLER_BUTTON_Y) {
+						} else if (event.id == SDL_GAMEPAD_BUTTON_NORTH) {
 							Ime::HostBackspace(session.generation);
 						}
 					}
@@ -695,19 +698,19 @@ struct SystemOverlay::Impl {
 				}
 				case InputKind::Axis: {
 					const float value = event.x / (event.x < 0.0f ? 32768.0f : 32767.0f);
-					if (event.id == SDL_CONTROLLER_AXIS_LEFTX) {
+					if (event.id == SDL_GAMEPAD_AXIS_LEFTX) {
 						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickLeft, value < -0.25f,
 						                     std::max(-value, 0.0f));
 						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickRight, value > 0.25f,
 						                     std::max(value, 0.0f));
-					} else if (event.id == SDL_CONTROLLER_AXIS_LEFTY) {
+					} else if (event.id == SDL_GAMEPAD_AXIS_LEFTY) {
 						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickUp, value < -0.25f,
 						                     std::max(-value, 0.0f));
 						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickDown, value > 0.25f,
 						                     std::max(value, 0.0f));
-					} else if (event.id == SDL_CONTROLLER_AXIS_RIGHTX) {
+					} else if (event.id == SDL_GAMEPAD_AXIS_RIGHTX) {
 						right_stick.x = std::abs(value) > 0.2f ? value : 0.0f;
-					} else if (event.id == SDL_CONTROLLER_AXIS_RIGHTY) {
+					} else if (event.id == SDL_GAMEPAD_AXIS_RIGHTY) {
 						right_stick.y = std::abs(value) > 0.2f ? value : 0.0f;
 					}
 					break;

src/graphics/presentation/systemOverlay.h

@@ -7,6 +7,7 @@
 #include <memory>
 
 union SDL_Event;
+struct SDL_Window;
 
 namespace Libs::Graphics {
 
@@ -17,7 +18,7 @@ struct SystemOverlayVisualState {
 	uint64_t revision;
 };
 
-void                     InitializeSystemOverlayInput();
+void                     InitializeSystemOverlayInput(SDL_Window* window);
 void                     ShutdownSystemOverlayInput();
 bool                     ProcessSystemOverlayInput(const SDL_Event& event);
 SystemOverlayVisualState GetSystemOverlayVisualState() noexcept;

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>
@@ -564,7 +565,12 @@ Graphics::ImageInfo BufferAttributeGroup::ImageInfo(const VideoOutBuffer& buffer
 	info.bgra16          = pixel_format.bgra16;
 	info.mip_layout[0]   = {0, total.size, pitch, attribute.height};
 	if (compression != Graphics::VideoOutCompression::Uncompressed) {
-		info.metadata.range       = {buffer.metadata_address, 0};
+		Graphics::TileSizeAlign dcc_size {};
+		if (!Graphics::TileGetDccSize(attribute.width, attribute.height, 1,
+		                              pixel_format.bytes_per_element, 1, tile_mode, dcc_size)) {
+			EXIT("invalid video-out DCC footprint\n");
+		}
+		info.metadata.range       = {buffer.metadata_address, dcc_size.size};
 		info.metadata.kind        = Graphics::ImageMetadataKind::Dcc;
 		info.metadata.control     = attribute.dcc_control;
 		info.metadata.compression = compression;
@@ -1654,8 +1660,12 @@ 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
@@ -1670,6 +1680,21 @@ KYTY_SYSV_ABI int VideoOutGetOutputStatus(int handle, VideoOutOutputStatus* stat
 	return OK;
 }
 
+KYTY_SYSV_ABI int VideoOutGetVrrStatus(int handle, int32_t* status) {
+	PRINT_NAME();
+
+	if (status == nullptr) {
+		return VIDEO_OUT_ERROR_INVALID_ADDRESS;
+	}
+	if (DriverState().Get(handle) == nullptr) {
+		return VIDEO_OUT_ERROR_INVALID_HANDLE;
+	}
+
+	// Kyty currently presents at a fixed refresh rate and does not negotiate VRR.
+	*status = 0;
+	return OK;
+}
+
 static int ValidateOutputConfig(int handle, uint64_t mode, const VideoOutOutputOptions* options,
                                 void* reserved_ptr, uint64_t reserved) {
 	if (!DriverState().IsOpened(handle)) {

src/graphics/presentation/videoOut.h

@@ -87,6 +87,7 @@ KYTY_SYSV_ABI int VideoOutGetEventData(const LibKernel::EventQueue::KernelEvent*
 KYTY_SYSV_ABI int VideoOutGetEventCount(const LibKernel::EventQueue::KernelEvent* ev);
 KYTY_SYSV_ABI int VideoOutWaitVblank(int handle);
 KYTY_SYSV_ABI int VideoOutGetOutputStatus(int handle, VideoOutOutputStatus* status);
+KYTY_SYSV_ABI int VideoOutGetVrrStatus(int handle, int32_t* status);
 KYTY_SYSV_ABI int VideoOutInitializeOutputOptions(VideoOutOutputOptions* options);
 KYTY_SYSV_ABI int VideoOutIsOutputSupported(int handle, uint64_t mode,
                                             const VideoOutOutputOptions* options,

src/graphics/presentation/window/hostInput.cpp

@@ -1,12 +1,7 @@
 #include "graphics/presentation/window/hostInput.h"
 
-#include "SDL_error.h"
-#include "SDL_events.h"
-#include "SDL_keyboard.h"
-#include "SDL_keycode.h"
-#include "SDL_mouse.h"
-#include "SDL_timer.h"
-#include "SDL_video.h"
+#include <SDL3/SDL.h>
+
 #include "common/assert.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
@@ -80,6 +75,7 @@ struct MouseJoystickState {
 };
 
 MouseJoystickState g_mouse;
+SDL_Window*        g_mouse_window = nullptr;
 
 std::size_t ControlFromName(std::string_view name) {
 	const auto info = std::find_if(CONTROL_INFO.begin(), CONTROL_INFO.end(),
@@ -141,9 +137,8 @@ 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());
@@ -214,14 +209,14 @@ uint32_t DefaultKeyboardButton(int key_code) {
 		case SDLK_LEFT: return Controller::PAD_BUTTON_LEFT;
 		case SDLK_DOWN: return Controller::PAD_BUTTON_DOWN;
 		case SDLK_RIGHT: return Controller::PAD_BUTTON_RIGHT;
-		case SDLK_j: return Controller::PAD_BUTTON_CROSS;
-		case SDLK_i: return Controller::PAD_BUTTON_TRIANGLE;
-		case SDLK_k: return Controller::PAD_BUTTON_SQUARE;
-		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_J: return Controller::PAD_BUTTON_CROSS;
+		case SDLK_I: return Controller::PAD_BUTTON_TRIANGLE;
+		case SDLK_K: return Controller::PAD_BUTTON_SQUARE;
+		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:
@@ -285,35 +280,35 @@ void DefaultKeyboardInput(int key_code, bool down) {
 	switch (NormalizeKey(static_cast<SDL_Keycode>(key_code))) {
 		case SDLK_BACKSPACE: SetTouchPad(0.25f, down); return;
 		case SDLK_TAB: SetTouchPad(0.75f, down); return;
-		case SDLK_a:
+		case SDLK_A:
 			left.left = down;
 			SetStickAxis(Controller::Axis::LeftX, left.left, left.right);
 			return;
-		case SDLK_d:
+		case SDLK_D:
 			left.right = down;
 			SetStickAxis(Controller::Axis::LeftX, left.left, left.right);
 			return;
-		case SDLK_w:
+		case SDLK_W:
 			left.up = down;
 			SetStickAxis(Controller::Axis::LeftY, left.up, left.down);
 			return;
-		case SDLK_s:
+		case SDLK_S:
 			left.down = down;
 			SetStickAxis(Controller::Axis::LeftY, left.up, left.down);
 			return;
-		case SDLK_f:
+		case SDLK_F:
 			right.left = down;
 			SetStickAxis(Controller::Axis::RightX, right.left, right.right);
 			return;
-		case SDLK_h:
+		case SDLK_H:
 			right.right = down;
 			SetStickAxis(Controller::Axis::RightX, right.left, right.right);
 			return;
-		case SDLK_t:
+		case SDLK_T:
 			right.up = down;
 			SetStickAxis(Controller::Axis::RightY, right.up, right.down);
 			return;
-		case SDLK_g:
+		case SDLK_G:
 			right.down = down;
 			SetStickAxis(Controller::Axis::RightY, right.up, right.down);
 			return;
@@ -321,11 +316,11 @@ void DefaultKeyboardInput(int key_code, bool down) {
 	}
 }
 
-void MouseToJoystick(int delta_x, int delta_y) {
+void MouseToJoystick(float delta_x, float delta_y) {
 	const double distance = std::hypot(delta_x, delta_y);
 	const double scale =
 	    std::clamp(distance * GetInputMap().MouseSensitivity() + 16.0, 64.0, 128.0) / distance;
-	const auto map_axis = [scale](int delta) {
+	const auto map_axis = [scale](float delta) {
 		return std::clamp(128 + static_cast<int>(std::lround(delta * scale)), 0, 255);
 	};
 	Controller::SetRightStick(Controller::HOST_INPUT_CONTROLLER_ID, map_axis(delta_x),
@@ -341,16 +336,45 @@ void CenterMouseStick() {
 }
 
 bool SetRelativeMouseMode(bool enabled) {
-	if (SDL_SetRelativeMouseMode(enabled ? SDL_TRUE : SDL_FALSE) == 0) {
+	if (SDL_SetWindowRelativeMouseMode(g_mouse_window, enabled)) {
 		return true;
 	}
 	LOGF("Mouse-to-joystick relative mode failed: %s\n", SDL_GetError());
 	return false;
 }
+
+int PollMouse(uint64_t now_ms) {
+	if (now_ms < g_mouse.next_poll) {
+		return static_cast<int>(g_mouse.next_poll - now_ms);
+	}
+	g_mouse.next_poll = now_ms + MOUSE_POLL_INTERVAL_MS;
+
+	float delta_x = 0;
+	float delta_y = 0;
+	SDL_GetRelativeMouseState(&delta_x, &delta_y);
+	if (delta_x == 0 && delta_y == 0) {
+		CenterMouseStick();
+	} else {
+		MouseToJoystick(delta_x, delta_y);
+		g_mouse.output = true;
+	}
+	return MOUSE_POLL_INTERVAL_MS;
+}
+
 } // namespace
 
-void HostInputInit() {
+void HostInputInit(SDL_Window* window) {
 	GetInputMap();
+	g_mouse_window = window;
+}
+
+void HostInputShutdown() {
+	if (g_mouse.enabled) {
+		SetRelativeMouseMode(false);
+		CenterMouseStick();
+		g_mouse = {};
+	}
+	g_mouse_window = nullptr;
 }
 
 void HostInputKey(int key_code, bool down) {
@@ -381,51 +405,33 @@ void HostInputToggleMouseToJoystick() {
 	if (!SetRelativeMouseMode(true)) {
 		return;
 	}
-	int ignored_x = 0;
-	int ignored_y = 0;
-	SDL_GetRelativeMouseState(&ignored_x, &ignored_y);
-	g_mouse.enabled   = true;
-	g_mouse.next_poll = SDL_GetTicks64() + MOUSE_POLL_INTERVAL_MS;
+	g_mouse.enabled = true;
 	LOGF("Mouse to right stick: enabled (F7 to release)\n");
 }
 
-int PollMouse(uint64_t now_ms) {
-	if (now_ms < g_mouse.next_poll) {
-		return static_cast<int>(g_mouse.next_poll - now_ms);
-	}
-	g_mouse.next_poll = now_ms + MOUSE_POLL_INTERVAL_MS;
-
-	int delta_x = 0;
-	int delta_y = 0;
-	SDL_GetRelativeMouseState(&delta_x, &delta_y);
-	if (delta_x == 0 && delta_y == 0) {
-		CenterMouseStick();
-		return MOUSE_POLL_INTERVAL_MS;
-	}
-
-	MouseToJoystick(delta_x, delta_y);
-	g_mouse.output = true;
-	return MOUSE_POLL_INTERVAL_MS;
-}
-
 bool HostInputWaitEvent(SDL_Event* event) {
-	if (!g_mouse.enabled || SDL_GetKeyboardFocus() == nullptr) {
+	bool has_event;
+	if (!g_mouse.enabled || SDL_GetKeyboardFocus() != g_mouse_window) {
+		g_mouse.next_poll = 0;
 		CenterMouseStick();
-		if (SDL_WaitEvent(event) == 0) {
+		has_event = SDL_WaitEvent(event);
+		if (!has_event) {
 			EXIT("%s\n", SDL_GetError());
 		}
-		return true;
+	} else {
+		if (g_mouse.next_poll == 0) {
+			SDL_GetRelativeMouseState(nullptr, nullptr);
+			g_mouse.next_poll = SDL_GetTicks() + MOUSE_POLL_INTERVAL_MS;
+		}
+		has_event = SDL_WaitEventTimeout(event, PollMouse(SDL_GetTicks()));
 	}
 
-	const int timeout_ms = PollMouse(SDL_GetTicks64());
-	SDL_ClearError();
-	if (SDL_WaitEventTimeout(event, timeout_ms) != 0) {
-		return true;
+	if (has_event && event->type == SDL_EVENT_WINDOW_FOCUS_LOST &&
+	    event->window.windowID == SDL_GetWindowID(g_mouse_window)) {
+		g_mouse.next_poll = 0;
+		CenterMouseStick();
 	}
-	if (SDL_GetError()[0] != '\0') {
-		EXIT("%s\n", SDL_GetError());
-	}
-	return false;
+	return has_event;
 }
 
 } // namespace Libs::Graphics

src/graphics/presentation/window/hostInput.h

@@ -3,11 +3,13 @@
 
 #include <cstdint>
 
+struct SDL_Window;
 union SDL_Event;
 
 namespace Libs::Graphics {
 
-void               HostInputInit();
+void               HostInputInit(SDL_Window* window);
+void               HostInputShutdown();
 void               HostInputKey(int key_code, bool down);
 void               HostInputMouseButton(uint8_t mouse_button, bool down);
 void               HostInputToggleMouseToJoystick();

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 {
@@ -174,6 +171,13 @@ void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent,
 		EXIT("unsupported prepared frame, extent=%ux%u format=%d\n", extent.width, extent.height,
 		     static_cast<int>(format));
 	}
+	auto&      dst        = image;
+	const bool compatible = dst.image != nullptr && dst.extent.width == extent.width &&
+	                        dst.extent.height == extent.height && dst.format == format;
+	if (compatible) {
+		return;
+	}
+
 	const auto features = graphics.GetFormatProperties(format).optimalTilingFeatures;
 	const auto required =
 	    vk::FormatFeatureFlagBits::eBlitSrc | vk::FormatFeatureFlagBits::eSampledImageFilterLinear |
@@ -183,12 +187,6 @@ void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent,
 		     static_cast<int>(format), static_cast<vk::FormatFeatureFlags::MaskType>(features));
 	}
 
-	auto&      dst        = image;
-	const bool compatible = dst.image != nullptr && dst.extent.width == extent.width &&
-	                        dst.extent.height == extent.height && dst.format == format;
-	if (compatible) {
-		return;
-	}
 	if (dst.image != nullptr) {
 		graphics.DeleteImage(dst);
 	}
@@ -282,6 +280,7 @@ public:
 
 	void                 Create();
 	void                 Recreate(bool surface_lost = false);
+	[[nodiscard]] bool   NeedsResize() const;
 	[[nodiscard]] Status AcquireNextImage();
 	[[nodiscard]] bool   PrepareSystemOverlay();
 	void                 RecordPresentCommands(CommandBuffer& command, VulkanImage& source,
@@ -301,6 +300,8 @@ private:
 	vk::SwapchainKHR            m_handle = nullptr;
 	vk::Format                  m_format = vk::Format::eUndefined;
 	vk::Extent2D                m_extent {};
+	// Drawable pixel size observed when this swapchain was created.
+	vk::Extent2D                m_window_extent {};
 	std::vector<vk::Image>      m_images;
 	std::vector<vk::ImageView>  m_image_views;
 	std::vector<vk::Semaphore>  m_image_acquired;
@@ -360,19 +361,25 @@ void Swapchain::Create() {
 	EXIT_IF(graphics.device == nullptr);
 	EXIT_IF(m_window.surface == nullptr);
 
-	Common::LockGuard lock(m_window.mutex);
-	EXIT_IF(graphics.screen_width == 0);
-	EXIT_IF(graphics.screen_height == 0);
+	// Capture the size before querying surface capabilities. If the window changes
+	// during creation, the next presentation will detect the newer size.
+	{
+		Common::LockGuard lock(m_window.mutex);
+		m_window_extent = {graphics.screen_width, graphics.screen_height};
+	}
+	EXIT_IF(m_window_extent.width == 0);
+	EXIT_IF(m_window_extent.height == 0);
+	m_window.RefreshSurfaceCapabilities();
 	const auto&       surface = m_window.surface_capabilities;
 	EXIT_NOT_IMPLEMENTED(surface.formats.empty());
 
 	m_extent = surface.capabilities.currentExtent;
 	if (m_extent.width == std::numeric_limits<uint32_t>::max()) {
 		m_extent.width =
-		    std::clamp(graphics.screen_width, surface.capabilities.minImageExtent.width,
+		    std::clamp(m_window_extent.width, surface.capabilities.minImageExtent.width,
 		               surface.capabilities.maxImageExtent.width);
 		m_extent.height =
-		    std::clamp(graphics.screen_height, surface.capabilities.minImageExtent.height,
+		    std::clamp(m_window_extent.height, surface.capabilities.minImageExtent.height,
 		               surface.capabilities.maxImageExtent.height);
 	}
 	uint32_t image_count = surface.capabilities.minImageCount + 1;
@@ -476,8 +483,8 @@ void Swapchain::Create() {
 		    graphics.device.createSemaphore(&semaphore_info, nullptr, &m_render_complete[i]),
 		    "create swapchain render-complete semaphore");
 	}
-	m_image_index = static_cast<uint32_t>(-1);
-	m_frame_index = 0;
+	m_image_index   = static_cast<uint32_t>(-1);
+	m_frame_index   = 0;
 }
 
 Swapchain::~Swapchain() {
@@ -518,11 +525,12 @@ void Swapchain::Destroy() {
 		graphics.device.destroySwapchainKHR(m_handle, nullptr);
 	}
 
-	m_handle      = nullptr;
-	m_format      = vk::Format::eUndefined;
-	m_extent      = {};
-	m_image_index = static_cast<uint32_t>(-1);
-	m_frame_index = 0;
+	m_handle        = nullptr;
+	m_format        = vk::Format::eUndefined;
+	m_extent        = {};
+	m_window_extent = {};
+	m_image_index   = static_cast<uint32_t>(-1);
+	m_frame_index   = 0;
 	m_images.clear();
 	m_image_views.clear();
 	m_image_acquired.clear();
@@ -535,15 +543,22 @@ void Swapchain::Recreate(bool surface_lost) {
 #if defined(__APPLE__)
 		// Surface recreation goes through SDL_Vulkan_CreateSurface, which touches the
 		// window's view/layer and must run on the main thread on macOS.
-		m_window.RunOnMainThread([this] { m_window.RecreateSurface(); });
+		EXIT_IF(!SDL_RunOnMainThread(
+		    [](void* window) { static_cast<WindowContext*>(window)->RecreateSurface(); },
+		    &m_window, true));
 #else
 		m_window.RecreateSurface();
 #endif
 	}
-	m_window.RefreshSurfaceCapabilities();
 	Create();
 }
 
+bool Swapchain::NeedsResize() const {
+	Common::LockGuard lock(m_window.mutex);
+	return m_window_extent.width != m_window.graphic_ctx.screen_width ||
+	       m_window_extent.height != m_window.graphic_ctx.screen_height;
+}
+
 Swapchain::Status Swapchain::AcquireNextImage() {
 	EXIT_IF(m_handle == nullptr || m_frame_index >= m_image_acquired.size());
 	m_image_index     = static_cast<uint32_t>(-1);
@@ -769,6 +784,10 @@ void Presenter::Present(Frame& frame, bool reuse) {
 
 	const auto overlay_visual = GetSystemOverlayVisualState();
 	auto&      swapchain  = m_impl->swapchain;
+	// Some window systems keep presenting an old swapchain after a resize.
+	if (swapchain.NeedsResize()) {
+		m_impl->RecoverSwapchain(Swapchain::Status::Recreate);
+	}
 	for (uint32_t attempt = 0; attempt < 2; attempt++) {
 		auto status = swapchain.AcquireNextImage();
 		if (status != Swapchain::Status::Success) {

src/graphics/presentation/window/vulkanWindow.cpp

@@ -1,20 +1,6 @@
-#include "SDL.h"
-#include "SDL_error.h"
-#include "SDL_events.h"
-#include "SDL_gamecontroller.h"
-#include "SDL_hints.h"
-#include "SDL_joystick.h"
-#include "SDL_keyboard.h"
-#include "SDL_keycode.h"
-#include "SDL_mouse.h"
-#include "SDL_pixels.h"
-#include "SDL_rwops.h"
-#include "SDL_stdinc.h"
-#include "SDL_surface.h"
-#include "SDL_thread.h"
-#include "SDL_touch.h"
-#include "SDL_video.h"
-#include "SDL_vulkan.h"
+#include <SDL3/SDL.h>
+#include <SDL3/SDL_vulkan.h>
+
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/emulatorConfig.h"
@@ -49,7 +35,6 @@
 // IWYU pragma: no_include <intrin.h>
 
 #define KYTY_ENABLE_DEBUG_PRINTF
-#define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
 
@@ -201,6 +186,8 @@ 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 +222,10 @@ 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 +505,14 @@ 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 +596,11 @@ 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 +611,7 @@ 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;
@@ -690,25 +686,13 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 	return device;
 }
 
-static void VulkanGetExtensions(SDL_Window* window, VulkanExtensions& r) {
-	EXIT_IF(window == nullptr);
-
+static void VulkanGetExtensions(VulkanExtensions& r) {
 	uint32_t required_extensions_count = 0;
 
-	auto sdl_result = SDL_Vulkan_GetInstanceExtensions(window, &required_extensions_count, nullptr);
-
-	EXIT_NOT_IMPLEMENTED(sdl_result == SDL_FALSE);
+	const char* const* extensions = SDL_Vulkan_GetInstanceExtensions(&required_extensions_count);
+	EXIT_NOT_IMPLEMENTED(extensions == nullptr);
 	EXIT_NOT_IMPLEMENTED(required_extensions_count == 0);
-
-	r.required_extensions =
-	    std::vector<const char*>(required_extensions_count); // @suppress("Ambiguous problem")
-
-	sdl_result = SDL_Vulkan_GetInstanceExtensions(window, &required_extensions_count,
-	                                              r.required_extensions.data());
-
-	EXIT_NOT_IMPLEMENTED(sdl_result == SDL_FALSE);
-	EXIT_NOT_IMPLEMENTED(required_extensions_count == 0);
-	EXIT_NOT_IMPLEMENTED(required_extensions_count != r.required_extensions.size());
+	r.required_extensions.assign(extensions, extensions + required_extensions_count);
 
 	r.available_extensions =
 	    EnumerateVulkan<vk::ExtensionProperties>( // @suppress("Ambiguous problem")
@@ -889,7 +873,7 @@ void WindowContext::CreateVulkan() {
 	VULKAN_HPP_DEFAULT_DISPATCHER.init(get_instance_proc_addr);
 
 	VulkanExtensions r;
-	VulkanGetExtensions(window, r);
+	VulkanGetExtensions(r);
 	VulkanCheckInstanceVersion();
 
 	vk::ApplicationInfo app_info {};
@@ -976,15 +960,16 @@ void WindowContext::CreateVulkan() {
 	}
 
 	vk::SurfaceKHR::CType native_surface = VK_NULL_HANDLE;
-	if (SDL_Vulkan_CreateSurface(window, static_cast<vk::Instance::CType>(graphic_ctx.instance),
-	                             &native_surface) == SDL_FALSE) {
+	if (!SDL_Vulkan_CreateSurface(window, static_cast<vk::Instance::CType>(graphic_ctx.instance),
+	                              nullptr, &native_surface)) {
 		EXIT("Could not create a Vulkan surface");
 	}
 	surface = native_surface;
 
 	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
@@ -1089,8 +1074,8 @@ void WindowContext::RecreateSurface() {
 		surface = nullptr;
 	}
 	vk::SurfaceKHR::CType native_surface = VK_NULL_HANDLE;
-	if (SDL_Vulkan_CreateSurface(window, static_cast<vk::Instance::CType>(graphic_ctx.instance),
-	                             &native_surface) == SDL_FALSE) {
+	if (!SDL_Vulkan_CreateSurface(window, static_cast<vk::Instance::CType>(graphic_ctx.instance),
+	                              nullptr, &native_surface)) {
 		EXIT("Could not recreate the Vulkan surface: %s\n", SDL_GetError());
 	}
 	surface = native_surface;
@@ -1126,7 +1111,7 @@ WindowContext::~WindowContext() {
 		SDL_DestroyWindow(window);
 		window = nullptr;
 	}
-	SDL_QuitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMECONTROLLER);
+	SDL_QuitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMEPAD);
 }
 
 } // namespace Libs::Graphics

src/graphics/presentation/window/window.cpp

@@ -1,23 +1,7 @@
 #include "graphics/presentation/window.h"
 
-#include "SDL.h"
-#include "SDL_error.h"
-#include "SDL_events.h"
-#include "SDL_filesystem.h"
-#include "SDL_gamecontroller.h"
-#include "SDL_hints.h"
-#include "SDL_joystick.h"
-#include "SDL_keyboard.h"
-#include "SDL_keycode.h"
-#include "SDL_mouse.h"
-#include "SDL_pixels.h"
-#include "SDL_rwops.h"
-#include "SDL_stdinc.h"
-#include "SDL_surface.h"
-#include "SDL_thread.h"
-#include "SDL_touch.h"
-#include "SDL_video.h"
-#include "SDL_vulkan.h"
+#include <SDL3/SDL.h>
+
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/emulatorConfig.h"
@@ -27,6 +11,7 @@
 #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 +30,7 @@
 #include <string>
 #include <vector>
 #include <vulkan/vk_platform.h>
+#include <filesystem>
 
 #define STB_IMAGE_IMPLEMENTATION
 #define STBI_NO_SIMD
@@ -52,7 +38,6 @@
 
 // IWYU pragma: no_include <intrin.h>
 
-#define KYTY_ENABLE_DEBUG_PRINTF
 #define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
@@ -71,39 +56,39 @@ struct EventKeyboard {
 
 static uint32_t ControllerButtonToPadButton(int button) {
 	switch (button) {
-		case SDL_CONTROLLER_BUTTON_A: return Controller::PAD_BUTTON_CROSS;
-		case SDL_CONTROLLER_BUTTON_B: return Controller::PAD_BUTTON_CIRCLE;
-		case SDL_CONTROLLER_BUTTON_X: return Controller::PAD_BUTTON_SQUARE;
-		case SDL_CONTROLLER_BUTTON_Y: return Controller::PAD_BUTTON_TRIANGLE;
-		case SDL_CONTROLLER_BUTTON_START: return Controller::PAD_BUTTON_OPTIONS;
-		case SDL_CONTROLLER_BUTTON_LEFTSTICK: return Controller::PAD_BUTTON_L3;
-		case SDL_CONTROLLER_BUTTON_RIGHTSTICK: return Controller::PAD_BUTTON_R3;
-		case SDL_CONTROLLER_BUTTON_LEFTSHOULDER: return Controller::PAD_BUTTON_L1;
-		case SDL_CONTROLLER_BUTTON_RIGHTSHOULDER: return Controller::PAD_BUTTON_R1;
-		case SDL_CONTROLLER_BUTTON_DPAD_UP: return Controller::PAD_BUTTON_UP;
-		case SDL_CONTROLLER_BUTTON_DPAD_DOWN: return Controller::PAD_BUTTON_DOWN;
-		case SDL_CONTROLLER_BUTTON_DPAD_LEFT: return Controller::PAD_BUTTON_LEFT;
-		case SDL_CONTROLLER_BUTTON_DPAD_RIGHT: return Controller::PAD_BUTTON_RIGHT;
-		case SDL_CONTROLLER_BUTTON_TOUCHPAD: return Controller::PAD_BUTTON_TOUCH_PAD;
+		case SDL_GAMEPAD_BUTTON_SOUTH: return Controller::PAD_BUTTON_CROSS;
+		case SDL_GAMEPAD_BUTTON_EAST: return Controller::PAD_BUTTON_CIRCLE;
+		case SDL_GAMEPAD_BUTTON_WEST: return Controller::PAD_BUTTON_SQUARE;
+		case SDL_GAMEPAD_BUTTON_NORTH: return Controller::PAD_BUTTON_TRIANGLE;
+		case SDL_GAMEPAD_BUTTON_START: return Controller::PAD_BUTTON_OPTIONS;
+		case SDL_GAMEPAD_BUTTON_LEFT_STICK: return Controller::PAD_BUTTON_L3;
+		case SDL_GAMEPAD_BUTTON_RIGHT_STICK: return Controller::PAD_BUTTON_R3;
+		case SDL_GAMEPAD_BUTTON_LEFT_SHOULDER: return Controller::PAD_BUTTON_L1;
+		case SDL_GAMEPAD_BUTTON_RIGHT_SHOULDER: return Controller::PAD_BUTTON_R1;
+		case SDL_GAMEPAD_BUTTON_DPAD_UP: return Controller::PAD_BUTTON_UP;
+		case SDL_GAMEPAD_BUTTON_DPAD_DOWN: return Controller::PAD_BUTTON_DOWN;
+		case SDL_GAMEPAD_BUTTON_DPAD_LEFT: return Controller::PAD_BUTTON_LEFT;
+		case SDL_GAMEPAD_BUTTON_DPAD_RIGHT: return Controller::PAD_BUTTON_RIGHT;
+		case SDL_GAMEPAD_BUTTON_TOUCHPAD: return Controller::PAD_BUTTON_TOUCH_PAD;
 		default: return 0;
 	}
 }
 
 static Controller::Axis ControllerAxisFromSdl(int axis_id) {
 	switch (axis_id) {
-		case SDL_CONTROLLER_AXIS_LEFTX: return Controller::Axis::LeftX;
-		case SDL_CONTROLLER_AXIS_LEFTY: return Controller::Axis::LeftY;
-		case SDL_CONTROLLER_AXIS_RIGHTX: return Controller::Axis::RightX;
-		case SDL_CONTROLLER_AXIS_RIGHTY: return Controller::Axis::RightY;
-		case SDL_CONTROLLER_AXIS_TRIGGERLEFT: return Controller::Axis::TriggerLeft;
-		case SDL_CONTROLLER_AXIS_TRIGGERRIGHT: return Controller::Axis::TriggerRight;
+		case SDL_GAMEPAD_AXIS_LEFTX: return Controller::Axis::LeftX;
+		case SDL_GAMEPAD_AXIS_LEFTY: return Controller::Axis::LeftY;
+		case SDL_GAMEPAD_AXIS_RIGHTX: return Controller::Axis::RightX;
+		case SDL_GAMEPAD_AXIS_RIGHTY: return Controller::Axis::RightY;
+		case SDL_GAMEPAD_AXIS_LEFT_TRIGGER: return Controller::Axis::TriggerLeft;
+		case SDL_GAMEPAD_AXIS_RIGHT_TRIGGER: return Controller::Axis::TriggerRight;
 		default: return Controller::Axis::AxisMax;
 	}
 }
 
 static bool ControllerAxisIsTrigger(int axis_id) {
-	return axis_id == SDL_CONTROLLER_AXIS_TRIGGERLEFT ||
-	       axis_id == SDL_CONTROLLER_AXIS_TRIGGERRIGHT;
+	return axis_id == SDL_GAMEPAD_AXIS_LEFT_TRIGGER ||
+	       axis_id == SDL_GAMEPAD_AXIS_RIGHT_TRIGGER;
 }
 
 static int ControllerAxisValueFromSdl(int axis_id, int axis_value) {
@@ -164,28 +149,6 @@ struct EventController {
 	double timestamp_seconds;
 };
 
-enum class DisplayOrientation {
-	Unknown,   /* The display orientation can't be determined */
-	Landscape, /* The display is in landscape mode, with the right side up, relative to portrait
-	              mode */
-	LandscapeFlipped, /* The display is in landscape mode, with the left side up, relative to
-	                     portrait mode */
-	Portrait,         /* The display is in portrait mode */
-	PortraitFlipped,  /* The display is in portrait mode, upside down */
-
-	DisplayEventOrientation = 0xF0
-};
-
-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 +156,6 @@ 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");
@@ -219,16 +181,13 @@ static void ToggleDesktopFullscreen() {
 	}
 
 	const auto flags = static_cast<uint32_t>(SDL_GetWindowFlags(g_window->window));
-	const bool fullscreen =
-	    (flags & static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP)) != 0u;
-	const auto mode =
-	    fullscreen ? 0u : static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP);
-	if (SDL_SetWindowFullscreen(g_window->window, mode) != 0) {
+	const bool fullscreen = (flags & SDL_WINDOW_FULLSCREEN) != 0u;
+	if (!SDL_SetWindowFullscreen(g_window->window, !fullscreen)) {
 		LOGF("Toggle fullscreen failed: %s\n", SDL_GetError());
 	}
 }
 
-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 +200,6 @@ 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();
@@ -254,7 +212,7 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 				break;
 			case SDLK_RETURN:
 			case SDLK_KP_ENTER:
-				if (!key.repeat && (key.mod & KMOD_ALT) != 0) {
+				if (!key.repeat && (key.mod & SDL_KMOD_ALT) != 0) {
 					ToggleDesktopFullscreen();
 					fullscreen_key = key.key_code;
 				}
@@ -347,15 +305,15 @@ static void GameEventController([[maybe_unused]] const EventController& f) {
 #endif
 
 	if (f.added) {
-		auto* pad = SDL_GameControllerOpen(f.id);
+		auto* pad = SDL_OpenGamepad(f.id);
 		EXIT_NOT_IMPLEMENTED(pad == nullptr);
-		int id = SDL_JoystickInstanceID(SDL_GameControllerGetJoystick(pad));
+		int id = SDL_GetJoystickID(SDL_GetGamepadJoystick(pad));
 		Controller::Connect(id);
 	}
 
 	if (f.removed) {
 		Controller::Disconnect(f.id);
-		SDL_GameControllerClose(SDL_GameControllerFromInstanceID(f.id));
+		SDL_CloseGamepad(SDL_GetGamepadFromID(f.id));
 	}
 
 	if (f.down || f.up) {
@@ -407,81 +365,73 @@ void WindowContext::Resize(uint32_t new_width, uint32_t new_height) {
 
 void WindowContext::ProcessWindowEvent(const SDL_WindowEvent& event) {
 	const auto& window_event = event;
-	switch (window_event.event) {
-		case SDL_WINDOWEVENT_SHOWN:
+	switch (window_event.type) {
+		case SDL_EVENT_WINDOW_SHOWN:
 			LOGF("Window %" PRIu32 " shown\n", window_event.windowID);
 			break;
 
-		case SDL_WINDOWEVENT_HIDDEN:
+		case SDL_EVENT_WINDOW_HIDDEN:
 			LOGF("Window %" PRIu32 " hidden\n", window_event.windowID);
 			break;
 
-		case SDL_WINDOWEVENT_EXPOSED:
+		case SDL_EVENT_WINDOW_EXPOSED:
 			LOGF("Window %" PRIu32 " exposed\n", window_event.windowID);
 			break;
 
-		case SDL_WINDOWEVENT_MOVED:
+		case SDL_EVENT_WINDOW_MOVED:
 			LOGF("Window %" PRIu32 " moved to %" PRId32 ",%" PRId32 "\n", window_event.windowID,
 			     window_event.data1, window_event.data2);
 			break;
 
-		case SDL_WINDOWEVENT_RESIZED:
+		case SDL_EVENT_WINDOW_RESIZED:
 			LOGF("Window %" PRIu32 " resized to %" PRId32 "x%" PRId32 "\n", window_event.windowID,
 			     window_event.data1, window_event.data2);
-
-			LOGF("m: %d\n", static_cast<int>(SDL_ThreadID()));
-			Resize(window_event.data1, window_event.data2);
-
 			break;
 
-		case SDL_WINDOWEVENT_SIZE_CHANGED:
+		case SDL_EVENT_WINDOW_PIXEL_SIZE_CHANGED:
 			LOGF("Window %" PRIu32 " size changed to %" PRId32 "x%" PRId32 "\n",
 			     window_event.windowID, window_event.data1, window_event.data2);
 
-			LOGF("m: %d\n", static_cast<int>(SDL_ThreadID()));
 			Resize(window_event.data1, window_event.data2);
 
 			break;
 
-		case SDL_WINDOWEVENT_MINIMIZED:
+		case SDL_EVENT_WINDOW_MINIMIZED:
 			LOGF("Window %" PRIu32 " minimized\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_MAXIMIZED:
+		case SDL_EVENT_WINDOW_MAXIMIZED:
 			LOGF("Window %" PRIu32 " maximized\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_RESTORED:
+		case SDL_EVENT_WINDOW_RESTORED:
 			LOGF("Window %" PRIu32 " restored\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_ENTER:
+		case SDL_EVENT_WINDOW_MOUSE_ENTER:
 			LOGF("Mouse entered window %" PRIu32 "\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_LEAVE:
+		case SDL_EVENT_WINDOW_MOUSE_LEAVE:
 			LOGF("Mouse left window %" PRIu32 "\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_FOCUS_GAINED:
+		case SDL_EVENT_WINDOW_FOCUS_GAINED:
 			LOGF("Window %" PRIu32 " gained keyboard focus\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_FOCUS_LOST:
+		case SDL_EVENT_WINDOW_FOCUS_LOST:
 			LOGF("Window %" PRIu32 " lost keyboard focus\n", window_event.windowID);
 			break;
-		case SDL_WINDOWEVENT_CLOSE:
+		case SDL_EVENT_WINDOW_CLOSE_REQUESTED:
 			LOGF("Window %" PRIu32 " closed\n", window_event.windowID);
 			break;
 		default:
-			LOGF("Window %" PRIu32 " got unknown event %" PRIu8 "\n", window_event.windowID,
-			     window_event.event);
+			LOGF("Window %" PRIu32 " got unknown event %" PRIu32 "\n", window_event.windowID,
+			     static_cast<uint32_t>(window_event.type));
 			break;
 	}
 }
 
 void WindowContext::ProcessDisplayEvent(const SDL_DisplayEvent& display) {
-	bool sdl = false;
-
-	switch (display.event) {
-		case SDL_DISPLAYEVENT_ORIENTATION: sdl = true; [[fallthrough]];
-		case static_cast<Uint8>(DisplayOrientation::DisplayEventOrientation): {
-			LOGF("Display %" PRIu32 "[%s] changed orientation to %d - ", display.display,
-			     sdl ? "SDL" : "Kyty", static_cast<int>(display.data1));
+	switch (display.type) {
+		case SDL_EVENT_DISPLAY_ORIENTATION: {
+			LOGF("Display %" PRIu32 " changed orientation to %d - ", display.displayID,
+			     static_cast<int>(display.data1));
 
 			switch (display.data1) {
 				case SDL_ORIENTATION_UNKNOWN: LOGF("UNKNOWN\n"); break;
@@ -495,8 +445,8 @@ void WindowContext::ProcessDisplayEvent(const SDL_DisplayEvent& display) {
 			break;
 		}
 		default:
-			LOGF("Display %" PRIu32 " got unknown event 0x%" PRIx8 "\n", display.display,
-			     display.event);
+			LOGF("Display %" PRIu32 " got unknown event 0x%" PRIx32 "\n", display.displayID,
+			     static_cast<uint32_t>(display.type));
 			break;
 	}
 }
@@ -504,10 +454,9 @@ void WindowContext::ProcessDisplayEvent(const SDL_DisplayEvent& display) {
 void WindowContext::ProcessEvent(double time_s) {
 	auto& game  = loop;
 	auto* event = &game.event;
-	EXIT_IF(SDL_GetEventState(SDL_DISPLAYEVENT) != SDL_ENABLE);
-	if ((event->type == SDL_KEYDOWN || event->type == SDL_KEYUP) &&
-	    event->key.keysym.sym == SDLK_F7) {
-		if (event->type == SDL_KEYDOWN && event->key.repeat == 0) {
+	if ((event->type == SDL_EVENT_KEY_DOWN || event->type == SDL_EVENT_KEY_UP) &&
+	    event->key.key == SDLK_F7) {
+		if (event->type == SDL_EVENT_KEY_DOWN && event->key.repeat == 0) {
 			HostInputToggleMouseToJoystick();
 		}
 		return;
@@ -515,59 +464,63 @@ void WindowContext::ProcessEvent(double time_s) {
 	if (ProcessSystemOverlayInput(*event)) {
 		return;
 	}
+	if (event->type >= SDL_EVENT_WINDOW_FIRST && event->type <= SDL_EVENT_WINDOW_LAST) {
+		ProcessWindowEvent(event->window);
+		return;
+	}
+	if (event->type >= SDL_EVENT_DISPLAY_FIRST && event->type <= SDL_EVENT_DISPLAY_LAST) {
+		ProcessDisplayEvent(event->display);
+		return;
+	}
 
 	switch (event->type) {
-		case SDL_QUIT: GameEventQuit(game); break;
+		case SDL_EVENT_QUIT: GameEventQuit(game); break;
 
-		case SDL_APP_TERMINATING: GameEventTerminate(game); break;
+		case SDL_EVENT_TERMINATING: GameEventTerminate(game); break;
 
-		case SDL_APP_LOWMEMORY: GameEventLowMemory(); break;
+		case SDL_EVENT_LOW_MEMORY: GameEventLowMemory(); break;
 
-		case SDL_APP_WILLENTERBACKGROUND: GameEventWillEnterBackground(game); break;
+		case SDL_EVENT_WILL_ENTER_BACKGROUND: GameEventWillEnterBackground(game); break;
 
-		case SDL_APP_DIDENTERBACKGROUND: GameEventDidEnterBackground(); break;
+		case SDL_EVENT_DID_ENTER_BACKGROUND: GameEventDidEnterBackground(); break;
 
-		case SDL_APP_WILLENTERFOREGROUND: GameEventWillEnterForeground(); break;
+		case SDL_EVENT_WILL_ENTER_FOREGROUND: GameEventWillEnterForeground(); break;
 
-		case SDL_APP_DIDENTERFOREGROUND: GameEventDidEnterForeground(game); break;
+		case SDL_EVENT_DID_ENTER_FOREGROUND: GameEventDidEnterForeground(game); break;
 
-		case SDL_KEYDOWN:
-		case SDL_KEYUP: {
+		case SDL_EVENT_KEY_DOWN:
+		case SDL_EVENT_KEY_UP: {
 			EventKeyboard key {};
 
-			key.down              = (event->type == SDL_KEYDOWN);
-			key.up                = (event->type == SDL_KEYUP);
-			key.pressed           = (event->key.state == SDL_PRESSED);
-			key.released          = (event->key.state == SDL_RELEASED);
+			key.down              = (event->type == SDL_EVENT_KEY_DOWN);
+			key.up                = (event->type == SDL_EVENT_KEY_UP);
+			key.pressed           = event->key.down;
+			key.released          = !event->key.down;
 			key.repeat            = (event->key.repeat != 0u);
-			key.scan_code         = event->key.keysym.scancode;
-			key.key_code          = event->key.keysym.sym;
-			key.mod               = event->key.keysym.mod;
+			key.scan_code         = event->key.scancode;
+			key.key_code          = event->key.key;
+			key.mod               = event->key.mod;
 			key.timestamp_seconds = time_s;
 
-			GameEventKeyboard(game, key);
+			GameEventKeyboard(key);
 
 			break;
 		}
 
-		case SDL_WINDOWEVENT: ProcessWindowEvent(event->window); break;
-
-		case SDL_DISPLAYEVENT: ProcessDisplayEvent(event->display); break;
-
-		case SDL_MOUSEBUTTONDOWN:
-		case SDL_MOUSEBUTTONUP: {
+		case SDL_EVENT_MOUSE_BUTTON_DOWN:
+		case SDL_EVENT_MOUSE_BUTTON_UP: {
 			EventMouse mb {};
 
-			mb.down              = (event->button.type == SDL_MOUSEBUTTONDOWN);
-			mb.up                = (event->button.type == SDL_MOUSEBUTTONUP);
+			mb.down              = (event->button.type == SDL_EVENT_MOUSE_BUTTON_DOWN);
+			mb.up                = (event->button.type == SDL_EVENT_MOUSE_BUTTON_UP);
 			mb.left              = (event->button.button == SDL_BUTTON_LEFT);
 			mb.middle            = (event->button.button == SDL_BUTTON_MIDDLE);
 			mb.right             = (event->button.button == SDL_BUTTON_RIGHT);
 			mb.x1                = (event->button.button == SDL_BUTTON_X1);
 			mb.x2                = (event->button.button == SDL_BUTTON_X2);
 			mb.touch             = (event->button.which == SDL_TOUCH_MOUSEID);
-			mb.pressed           = (event->button.state == SDL_PRESSED);
-			mb.released          = (event->button.state == SDL_RELEASED);
+			mb.pressed           = event->button.down;
+			mb.released          = !event->button.down;
 			mb.num_of_clicks     = event->button.clicks;
 			mb.wheel             = false;
 			mb.x                 = event->button.x;
@@ -582,7 +535,7 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_MOUSEWHEEL: {
+		case SDL_EVENT_MOUSE_WHEEL: {
 			EventMouse mb {};
 
 			mb.down              = false;
@@ -609,16 +562,16 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_MOUSEMOTION: {
+		case SDL_EVENT_MOUSE_MOTION: {
 			EventMouse mb {};
 
 			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;
@@ -636,16 +589,16 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_FINGERMOTION:
-		case SDL_FINGERDOWN:
-		case SDL_FINGERUP: {
+		case SDL_EVENT_FINGER_MOTION:
+		case SDL_EVENT_FINGER_DOWN:
+		case SDL_EVENT_FINGER_UP: {
 			EventFinger f {};
 
-			f.down              = (event->tfinger.type == SDL_FINGERDOWN);
-			f.up                = (event->tfinger.type == SDL_FINGERUP);
-			f.motion            = (event->tfinger.type == SDL_FINGERMOTION);
-			f.finger_id         = static_cast<int>(event->tfinger.fingerId);
-			f.touch_id          = static_cast<int>(event->tfinger.touchId);
+			f.down              = (event->tfinger.type == SDL_EVENT_FINGER_DOWN);
+			f.up                = (event->tfinger.type == SDL_EVENT_FINGER_UP);
+			f.motion            = (event->tfinger.type == SDL_EVENT_FINGER_MOTION);
+			f.finger_id         = static_cast<int>(event->tfinger.fingerID);
+			f.touch_id          = static_cast<int>(event->tfinger.touchID);
 			f.x                 = event->tfinger.x;
 			f.y                 = event->tfinger.y;
 			f.dx                = event->tfinger.dx;
@@ -658,13 +611,13 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_CONTROLLERAXISMOTION: {
+		case SDL_EVENT_GAMEPAD_AXIS_MOTION: {
 			EventController c {};
 
-			c.id                = event->caxis.which;
-			c.button            = SDL_CONTROLLER_BUTTON_INVALID;
-			c.axis_id           = event->caxis.axis;
-			c.axis_value        = event->caxis.value;
+			c.id                = event->gaxis.which;
+			c.button            = SDL_GAMEPAD_BUTTON_INVALID;
+			c.axis_id           = event->gaxis.axis;
+			c.axis_value        = event->gaxis.value;
 			c.down              = false;
 			c.up                = false;
 			c.added             = false;
@@ -680,22 +633,22 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_CONTROLLERBUTTONDOWN:
-		case SDL_CONTROLLERBUTTONUP: {
+		case SDL_EVENT_GAMEPAD_BUTTON_DOWN:
+		case SDL_EVENT_GAMEPAD_BUTTON_UP: {
 			EventController c {};
 
-			c.id                = event->cbutton.which;
-			c.button            = event->cbutton.button;
-			c.axis_id           = SDL_CONTROLLER_AXIS_INVALID;
+			c.id                = event->gbutton.which;
+			c.button            = event->gbutton.button;
+			c.axis_id           = SDL_GAMEPAD_AXIS_INVALID;
 			c.axis_value        = 0;
-			c.down              = (event->cbutton.type == SDL_CONTROLLERBUTTONDOWN);
-			c.up                = (event->cbutton.type == SDL_CONTROLLERBUTTONUP);
+			c.down              = event->gbutton.down;
+			c.up                = !event->gbutton.down;
 			c.added             = false;
 			c.removed           = false;
 			c.remapped          = false;
 			c.axis              = false;
-			c.pressed           = (event->cbutton.state == SDL_PRESSED);
-			c.released          = (event->cbutton.state == SDL_RELEASED);
+			c.pressed           = event->gbutton.down;
+			c.released          = !event->gbutton.down;
 			c.timestamp_seconds = time_s;
 
 			GameEventController(c);
@@ -703,30 +656,44 @@ void WindowContext::ProcessEvent(double time_s) {
 			break;
 		}
 
-		case SDL_CONTROLLERTOUCHPADDOWN:
-		case SDL_CONTROLLERTOUCHPADMOTION:
-		case SDL_CONTROLLERTOUCHPADUP:
-			if (event->ctouchpad.touchpad == 0) {
-				Controller::SetTouchPad(event->ctouchpad.which, event->ctouchpad.finger,
-				                        event->ctouchpad.type != SDL_CONTROLLERTOUCHPADUP,
-				                        event->ctouchpad.x, event->ctouchpad.y);
+		case SDL_EVENT_GAMEPAD_TOUCHPAD_DOWN:
+		case SDL_EVENT_GAMEPAD_TOUCHPAD_MOTION:
+		case SDL_EVENT_GAMEPAD_TOUCHPAD_UP:
+			if (event->gtouchpad.touchpad == 0) {
+				Controller::SetTouchPad(event->gtouchpad.which, event->gtouchpad.finger,
+				                        event->gtouchpad.type != SDL_EVENT_GAMEPAD_TOUCHPAD_UP,
+				                        event->gtouchpad.x, event->gtouchpad.y);
 			}
 			break;
 
-		case SDL_CONTROLLERDEVICEADDED:
-		case SDL_CONTROLLERDEVICEREMOVED:
-		case SDL_CONTROLLERDEVICEREMAPPED: {
+		case SDL_EVENT_GAMEPAD_SENSOR_UPDATE: {
+			const auto& sensor = event->gsensor;
+			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.sensor_timestamp != 0
+				                          ? sensor.sensor_timestamp / 1000
+				                          : sensor.timestamp / 1000);
+			}
+			break;
+		}
+
+		case SDL_EVENT_GAMEPAD_ADDED:
+		case SDL_EVENT_GAMEPAD_REMOVED:
+		case SDL_EVENT_GAMEPAD_REMAPPED: {
 			EventController c {};
 
-			c.id                = event->cdevice.which;
-			c.button            = SDL_CONTROLLER_BUTTON_INVALID;
-			c.axis_id           = SDL_CONTROLLER_AXIS_INVALID;
+			c.id                = event->gdevice.which;
+			c.button            = SDL_GAMEPAD_BUTTON_INVALID;
+			c.axis_id           = SDL_GAMEPAD_AXIS_INVALID;
 			c.axis_value        = 0;
 			c.down              = false;
 			c.up                = false;
-			c.added             = (event->cdevice.type == SDL_CONTROLLERDEVICEADDED);
-			c.removed           = (event->cdevice.type == SDL_CONTROLLERDEVICEREMOVED);
-			c.remapped          = (event->cdevice.type == SDL_CONTROLLERDEVICEREMAPPED);
+			c.added             = (event->gdevice.type == SDL_EVENT_GAMEPAD_ADDED);
+			c.removed           = (event->gdevice.type == SDL_EVENT_GAMEPAD_REMOVED);
+			c.remapped          = (event->gdevice.type == SDL_EVENT_GAMEPAD_REMAPPED);
 			c.axis              = false;
 			c.pressed           = false;
 			c.released          = false;
@@ -739,47 +706,6 @@ void WindowContext::ProcessEvent(double time_s) {
 	}
 }
 
-void WindowContext::RunOnMainThread(std::function<void()> task) {
-	if (Common::Thread::IsMainThread()) {
-		task();
-		return;
-	}
-
-	uint64_t ticket = 0;
-	{
-		Common::LockGuard lock(main_task_mutex);
-		main_tasks.push_back(std::move(task));
-		ticket = ++main_tasks_queued;
-	}
-
-	// Wake the main loop in case it is blocked in SDL_WaitEvent.
-	SDL_Event event {};
-	event.type = SDL_USEREVENT;
-	SDL_PushEvent(&event);
-
-	Common::LockGuard lock(main_task_mutex);
-	while (main_tasks_run < ticket) {
-		main_task_done.Wait(&main_task_mutex);
-	}
-}
-
-void WindowContext::DrainMainThreadTasks() {
-	std::vector<std::function<void()>> tasks;
-	{
-		Common::LockGuard lock(main_task_mutex);
-		tasks.swap(main_tasks);
-	}
-	if (tasks.empty()) {
-		return;
-	}
-	for (auto& task: tasks) {
-		task();
-	}
-	Common::LockGuard lock(main_task_mutex);
-	main_tasks_run += tasks.size();
-	main_task_done.SignalAll();
-}
-
 void WindowContext::Run() {
 	Common::Timer timer;
 	timer.Start();
@@ -789,7 +715,6 @@ void WindowContext::Run() {
 	loop.paused.store(false, std::memory_order_release);
 
 	while (!loop.need_exit) {
-		DrainMainThreadTasks();
 		if (loop.paused.load(std::memory_order_acquire)) {
 			if (!timer.IsPaused()) {
 				timer.Pause();
@@ -813,17 +738,15 @@ static void WindowCreate(WindowContext& context) {
 	int width  = static_cast<int>(context.graphic_ctx.screen_width);
 	int height = static_cast<int>(context.graphic_ctx.screen_height);
 
-	SDL_SetHint(SDL_HINT_JOYSTICK_HIDAPI_PS4_RUMBLE, "1");
-	SDL_SetHint(SDL_HINT_JOYSTICK_HIDAPI_PS5_RUMBLE, "1");
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	SDL_SetHint(SDL_HINT_WINDOWS_DPI_SCALING, "0");
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX
+	// RenderDoc can hide Wayland Vulkan surface support; prefer X11 for captures.
+	if (Config::RenderDocEnabled()) {
+		SDL_SetHint(SDL_HINT_VIDEO_DRIVER, "x11,wayland");
+	}
 #endif
-
-	if (SDL_InitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMECONTROLLER) < 0) {
+	if (!SDL_InitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMEPAD)) {
 		EXIT("%s\n", SDL_GetError());
 	}
-	HostInputInit();
-	InitializeSystemOverlayInput();
 
 	LOGF("WindowCreate(): width = %d, height = %d\n", width, height);
 
@@ -833,15 +756,14 @@ static void WindowCreate(WindowContext& context) {
 	// MoltenVK loader before calling SDL_CreateWindow. Keep an explicit user
 	// override, and fall back to SDL's normal loader search when no bundle is present.
 	if (std::getenv("SDL_VULKAN_LIBRARY") == nullptr) {
-		if (char* base_path = SDL_GetBasePath(); base_path != nullptr) {
+		if (const char* base_path = SDL_GetBasePath(); base_path != nullptr) {
 			const std::string base_path_str = base_path;
-			SDL_free(base_path);
 			std::string moltenvk_path = base_path_str + "libMoltenVK.dylib";
 			if (!Common::File::IsFileExisting(moltenvk_path)) {
 				moltenvk_path = base_path_str + "../Frameworks/libMoltenVK.dylib";
 			}
 			if (Common::File::IsFileExisting(moltenvk_path) &&
-			    SDL_setenv("SDL_VULKAN_LIBRARY", moltenvk_path.c_str(), 0) == 0) {
+			    SDL_setenv_unsafe("SDL_VULKAN_LIBRARY", moltenvk_path.c_str(), 0) == 0) {
 				LOGF("Vulkan loader: %s\n", moltenvk_path.c_str());
 			}
 		}
@@ -854,21 +776,22 @@ 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, width, height, window_flags);
 
 	if (context.window == nullptr) {
 		EXIT("%s\n", SDL_GetError());
 	}
 
-	SDL_SetWindowResizable(context.window, SDL_FALSE);
+	SDL_SetWindowPosition(context.window, SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED);
+	HostInputInit(context.window);
+	InitializeSystemOverlayInput(context.window);
 	context.UpdateIcon();
 }
 
 uint32_t WindowContext::InitialWindowFlags(bool fullscreen) noexcept {
-	auto flags = static_cast<uint32_t>(SDL_WINDOW_VULKAN);
+	auto flags = static_cast<uint32_t>(SDL_WINDOW_VULKAN | SDL_WINDOW_RESIZABLE);
 	if (fullscreen) {
-		flags |= static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP);
+		flags |= static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN);
 	}
 	return flags;
 }
@@ -899,6 +822,7 @@ void WindowRun() {
 
 void WindowShutdown() {
 	if (g_window != nullptr) {
+		HostInputShutdown();
 		Controller::EmergencyShutdown();
 		g_window.reset();
 	}
@@ -934,15 +858,15 @@ struct WindowIcon {
 	void*        pixels  = nullptr;
 
 	~WindowIcon() {
-		SDL_FreeSurface(surface);
+		SDL_DestroySurface(surface);
 		stbi_image_free(pixels);
 	}
 };
 
-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;
@@ -958,8 +882,8 @@ static void WindowLoadPngIcon(const std::string& path, WindowIcon* icon) {
 
 	EXIT_IF(icon->pixels == nullptr);
 
-	icon->surface = SDL_CreateRGBSurfaceWithFormatFrom(icon->pixels, width, height, 32, width * 4,
-	                                                   SDL_PIXELFORMAT_RGBA32);
+	icon->surface = SDL_CreateSurfaceFrom(width, height, SDL_PIXELFORMAT_RGBA32, icon->pixels,
+	                                      width * 4);
 	EXIT_NOT_IMPLEMENTED(icon->surface == nullptr);
 }
 
@@ -968,7 +892,7 @@ 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);
 		}
@@ -1021,7 +945,16 @@ void WindowContext::UpdateTitle() {
 	    (has_title_id ? ", " : ""), (has_app_ver ? app_ver : ""), (has_app_ver ? " " : ""),
 	    device_name, processor_name, frame_num, current_fps);
 
-	RunOnMainThread([this, text = std::move(text)] { SDL_SetWindowTitle(window, text.c_str()); });
+	struct TitleUpdate {
+		SDL_Window*  window;
+		std::string* text;
+	} update {window, &text};
+	EXIT_IF(!SDL_RunOnMainThread(
+	    [](void* data) {
+		    auto& title = *static_cast<TitleUpdate*>(data);
+		    SDL_SetWindowTitle(title.window, title.text->c_str());
+	    },
+	    &update, true));
 }
 
 } // namespace Libs::Graphics

src/graphics/presentation/window/windowInternal.h

@@ -1,15 +1,14 @@
 #ifndef EMULATOR_SRC_GRAPHICS_PRESENTATION_WINDOW_WINDOWINTERNAL_H_
 #define EMULATOR_SRC_GRAPHICS_PRESENTATION_WINDOW_WINDOWINTERNAL_H_
 
-#include "SDL_events.h"
-#include "SDL_video.h"
+#include <SDL3/SDL.h>
+
 #include "common/threads.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <atomic>
 #include <cstdint>
-#include <functional>
 #include <memory>
 #include <vector>
 
@@ -48,9 +47,6 @@ struct WindowContext {
 	void ProcessDisplayEvent(const SDL_DisplayEvent& event);
 	void ProcessEvent(double time_seconds);
 	void Run();
-	// SDL window operations must complete on the main thread.
-	void RunOnMainThread(std::function<void()> task);
-	void DrainMainThreadTasks();
 
 	GraphicContext                 graphic_ctx;
 	SDL_Window*                    window        = nullptr;
@@ -61,12 +57,6 @@ struct WindowContext {
 	WindowLoopState                loop;
 
 	Common::Mutex mutex;
-
-	Common::Mutex                      main_task_mutex;
-	Common::CondVar                    main_task_done;
-	std::vector<std::function<void()>> main_tasks;            // guarded by main_task_mutex
-	uint64_t                           main_tasks_queued = 0; // guarded by main_task_mutex
-	uint64_t                           main_tasks_run    = 0; // guarded by main_task_mutex
 };
 
 } // namespace Libs::Graphics

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -105,8 +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;
@@ -210,13 +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) {
+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;
-	EmbeddedFetchData data;
+	Frontend::EmbeddedFetchPlan data;
 	data.loads.reserve(input_info->resources_num);
 	int32_t vertex_offset_candidate   = -1;
 	int32_t instance_offset_candidate = -1;
@@ -560,7 +557,7 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		     static_cast<uint64_t>(cfg.natural_loops.size()), phase_ms());
 	}
 
-	EmbeddedFetchData embedded_fetch;
+	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(

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");

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

@@ -138,7 +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);
@@ -559,6 +559,10 @@ void EmitVoid(ValueEmitContext&) {}
 
 void EmitBarrier(EmitterState& state) {
 	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 |
@@ -611,7 +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 {
@@ -61,10 +63,10 @@ uint32_t AddressF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::
 	           : Unary(ctx.state, spv::OpBitcast, TypeF32(ctx.state), value);
 }
 
-ImageSampleLayout Layout(const IR::MemoryInfo& mem, ImageDimension dimension) {
+ImageSampleLayout Layout(const IR::MemoryInfo& mem) {
 	ImageSampleLayout layout;
 	uint32_t          cursor = 0;
-	const auto&       info   = ImageDimensionInfoFor(dimension);
+	const auto&       info   = ImageDimensionInfoFor(mem.image_dimension);
 	if (HasFlag(mem, Decoder::ImageSampleFlagOffset)) layout.offset = cursor++;
 	if (HasFlag(mem, Decoder::ImageSampleFlagBias)) layout.bias = cursor++;
 	if (HasFlag(mem, Decoder::ImageSampleFlagCompare)) layout.dref = cursor++;
@@ -102,8 +104,7 @@ uint32_t CubeLayer(EmitterState& state, uint32_t value) {
 }
 
 uint32_t CoordF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
-                  uint32_t first, uint32_t components) {
-	const bool cube = ctx.state.program.info.images.at(mem.resource).cube;
+                  uint32_t first, uint32_t components, bool cube = false) {
 	auto x = AddressF32(ctx, mem, address, first);
 	if (components == 1u) return x;
 	auto y = mem.image_address_components > first + 1u ? AddressF32(ctx, mem, address, first + 1u)
@@ -574,7 +575,8 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto lod       = state.builder.AllocateId();
 		state.builder.AddFunction(
 		    spv::OpImageQueryLod, TypeF32Vector(state, 2), lod, sampled,
-		    CoordF32(ctx, mem, *address, 0, ImageDimensionInfoFor(dimension).spatial_components));
+		    CoordF32(ctx, mem, *address, 0, ImageDimensionInfoFor(dimension).spatial_components,
+		             image.cube));
 		uint32_t values[4] = {ConstantU32(state, 0), ConstantU32(state, 0), ConstantU32(state, 0),
 		                      ConstantU32(state, 0)};
 		for (uint32_t index = 0; index < 2u; index++) {
@@ -635,7 +637,7 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 	if (op == IR::ValueOpcode::ImageSampleRaw || op == IR::ValueOpcode::ImageGatherRaw) {
 		const auto  dimension      = image.dimension;
 		const auto& dimension_info = ImageDimensionInfoFor(dimension);
-		const auto  layout         = Layout(mem, dimension);
+		const auto  layout         = Layout(mem);
 		const auto  numeric_class  = image.numeric_class;
 		const bool  dref           = HasFlag(mem, Decoder::ImageSampleFlagCompare);
 		if (dref && state.program.info.images[mem.resource].conversion_format !=
@@ -643,9 +645,16 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			ctx.Fail(inst, "uses depth comparison with a packed integer image");
 			return;
 		}
-		const auto coord =
-		    CoordF32(ctx, mem, *address, layout.coord, dimension_info.coordinate_components);
 		if (op == IR::ValueOpcode::ImageGatherRaw) {
+			const auto coord = CoordF32(ctx, mem, *address, layout.coord,
+			                            dimension_info.coordinate_components, image.cube);
+			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) ||
@@ -743,6 +752,11 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			operands.push_back(AddressF32(ctx, mem, *address, layout.bias));
 		}
 		const auto EmitSample = [&](uint32_t resource) {
+			const auto& candidate = state.program.info.images[resource];
+			const auto coord =
+			    CoordF32(ctx, mem, *address, layout.coord,
+			             ImageDimensionInfoFor(candidate.dimension).coordinate_components,
+			             candidate.cube);
 			const auto            sampled = MakeSampledImage(state, resource, mem.sampler);
 			const auto            sample  = state.builder.AllocateId();
 			std::vector<uint32_t> sample_operands {result_type, sample, sampled, coord};
@@ -770,11 +784,11 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		                         ? &state.program.descriptor_sources[image.source]
 		                         : nullptr;
 		if (handle == nullptr || source == nullptr || !source->indirect_image.has_value() ||
-		    source->indirect_image->key_arg >= handle->NumArgs()) {
+		    handle->NumArgs() == 0u) {
 			ctx.Fail(inst, "has invalid indirect image key provenance");
 			return;
 		}
-		const auto key = ctx.Def(handle->Arg(source->indirect_image->key_arg));
+		const auto key = ctx.Def(handle->Arg(0));
 		if (state.flattened_srt_variable == 0 || image.indirect_search_iterations == 0u ||
 		    image.indirect_resources.size() < 2u) {
 			ctx.Fail(inst, "has no indirect image runtime mapping");

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

@@ -308,8 +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);
@@ -329,7 +327,8 @@ 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);
@@ -338,7 +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);
 
@@ -495,22 +494,18 @@ template <typename Fn>
 uint32_t EmitValueOrDefaultIfCondition(EmitterState& state, uint32_t condition, uint32_t type,
                                        uint32_t default_value, Fn&& fn) {
 	const auto then_label  = state.builder.AllocateId();
-	const auto then_exit   = state.builder.AllocateId();
-	const auto else_label  = state.builder.AllocateId();
+	const auto header     = state.current_label;
 	const auto merge_label = state.builder.AllocateId();
 	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
-	state.builder.AddFunction(spv::OpBranchConditional, condition, then_label, else_label);
+	state.builder.AddFunction(spv::OpBranchConditional, condition, then_label, merge_label);
 	EmitLabel(state, then_label);
 	const auto then_value = fn();
-	state.builder.AddFunction(spv::OpBranch, then_exit);
-	EmitLabel(state, then_exit);
-	state.builder.AddFunction(spv::OpBranch, merge_label);
-	EmitLabel(state, else_label);
+	const auto then_exit  = state.current_label;
 	state.builder.AddFunction(spv::OpBranch, merge_label);
 	EmitLabel(state, merge_label);
 	const auto value = state.builder.AllocateId();
 	state.builder.AddFunction(spv::OpPhi, type, value, then_value, then_exit, default_value,
-	                          else_label);
+	                          header);
 	return value;
 }
 

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

@@ -27,61 +27,71 @@ 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) {
+	const auto zero = ConstantU32(state, 0);
+	const auto one  = ConstantU32(state, 1);
+	const auto add = [&](uint32_t lhs, uint32_t rhs) {
+		return lhs == zero ? rhs : rhs == zero ? lhs
+		                                      : Binary(state, spv::OpIAdd, TypeU32(state), lhs, rhs);
+	};
+	const auto mul = [&](uint32_t lhs, uint32_t rhs) {
+		return lhs == zero || rhs == zero ? zero
+		       : lhs == one              ? rhs
+		       : rhs == one              ? lhs
+		                                 : Binary(state, spv::OpIMul, TypeU32(state), lhs, rhs);
+	};
+	if (immediate != 0u) {
+		offset = add(offset, ConstantU32(state, immediate));
+	}
+	auto address = add(mul(index, stride), 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 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);
+		const auto msb = mul(add(mul(index_msb, stride), offset_msb), indices);
+		const auto lsb = add(Binary(state, spv::OpShiftLeftLogical, TypeU32(state), index_lsb,
+		                            ConstantU32(state, 2u)), offset_lsb);
+		address = Select(state, TypeU32(state), swizzle, add(msb, lsb), address);
 	}
+	return {offset, add(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 +206,8 @@ 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 +243,7 @@ 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 +520,6 @@ 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 +710,15 @@ 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 (bits == 16u && (plan.info.type == Format::ComponentType::Snorm ||
+	                    plan.info.type == Format::ComponentType::Float)) {
+		const auto value = EmitBitCastF32U32(ctx.state, data);
+		const auto pair = EmitCompositeConstructF32x2(ctx.state, value,
+		                                               ConstantF32Value(ctx.state, 0.0f));
+		data = plan.info.type == Format::ComponentType::Float
+		           ? EmitPackHalf2x16(ctx.state, pair)
+		           : EmitPackSnorm2x16(ctx.state, pair);
+	}
 	if (bits == 8u || bits == 16u) {
 		StoreSubwordInBounds(ctx, mem, plan.resource, plan.addresses[component],
 		                     plan.indices[component], bits, data);
@@ -737,12 +727,103 @@ 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 +1172,8 @@ 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) {
@@ -558,8 +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;
 			}
 		}

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/MemoryOps.cpp

@@ -86,6 +86,7 @@ constexpr MemoryOpcodeInfo FLAT_OPCODE_LIST[] = {
 
 constexpr MemoryOpcodeInfo DS_OPCODE_LIST[] = {
     {0x00u, Opcode::DS_ADD_U32, 1, 32},          {0x01u, Opcode::DS_SUB_U32, 1, 32},
+    {0x03u, Opcode::DS_INC_U32, 1, 32},          {0x04u, Opcode::DS_DEC_U32, 1, 32},
     {0x05u, Opcode::DS_MIN_I32, 1, 32},          {0x06u, Opcode::DS_MAX_I32, 1, 32},
     {0x07u, Opcode::DS_MIN_U32, 1, 32},          {0x08u, Opcode::DS_MAX_U32, 1, 32},
     {0x09u, Opcode::DS_AND_B32, 1, 32},          {0x0au, Opcode::DS_OR_B32, 1, 32},
@@ -107,6 +108,7 @@ constexpr MemoryOpcodeInfo DS_OPCODE_LIST[] = {
     {0x4du, Opcode::DS_WRITE_B64, 2, 32},        {0x4eu, Opcode::DS_WRITE2_B64, 4, 32},
     {0x4fu, Opcode::DS_WRITE2ST64_B64, 4, 32},   {0x76u, Opcode::DS_READ_B64, 2, 32},
     {0x77u, Opcode::DS_READ2_B64, 4, 32},        {0x78u, Opcode::DS_READ2ST64_B64, 4, 32},
+    {0xa0u, Opcode::DS_WRITE_B8_D16_HI, 1, 8},
     {0xa1u, Opcode::DS_WRITE_B16_D16_HI, 1, 16},
     {0xa6u, Opcode::DS_READ_U16_D16, 1, 16},
     {0xa7u, Opcode::DS_READ_U16_D16_HI, 1, 16},
@@ -147,6 +149,7 @@ bool IsDsWriteOpcode(Opcode opcode) {
 	switch (opcode) {
 		case Opcode::DS_WRITE_B8:
 		case Opcode::DS_WRITE_B16:
+		case Opcode::DS_WRITE_B8_D16_HI:
 		case Opcode::DS_WRITE_B16_D16_HI:
 		case Opcode::DS_WRITE2_B32:
 		case Opcode::DS_WRITE2ST64_B32:
@@ -166,7 +169,9 @@ bool IsDsAtomicOpcode(Opcode opcode) {
 		case Opcode::DS_ADD_RTN_U32:
 		case Opcode::DS_SUB_U32:
 		case Opcode::DS_SUB_RTN_U32:
+		case Opcode::DS_INC_U32:
 		case Opcode::DS_INC_RTN_U32:
+		case Opcode::DS_DEC_U32:
 		case Opcode::DS_DEC_RTN_U32:
 		case Opcode::DS_MIN_I32:
 		case Opcode::DS_MIN_RTN_I32:
@@ -429,7 +434,9 @@ void DecodeDs(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
 	}
 	DecodeVectorGpr(addr, inst.src0);
 	DecodeVectorGpr(data0, inst.src1);
-	if (inst.opcode == Opcode::DS_WRITE_B16_D16_HI) {
+	if (inst.opcode == Opcode::DS_WRITE_B8_D16_HI) {
+		inst.src1.sdwa_sel = 2u;
+	} else if (inst.opcode == Opcode::DS_WRITE_B16_D16_HI) {
 		inst.src1.sdwa_sel = 5u;
 	}
 	DecodeVectorGpr(data1, inst.src2);

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},
 };
 
@@ -137,6 +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:
@@ -241,7 +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) {
@@ -259,7 +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;
@@ -475,8 +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;
 }
@@ -580,6 +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:
@@ -660,7 +659,9 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::DS_ADD_RTN_U32:
 		case Opcode::DS_SUB_U32:
 		case Opcode::DS_SUB_RTN_U32:
+		case Opcode::DS_INC_U32:
 		case Opcode::DS_INC_RTN_U32:
+		case Opcode::DS_DEC_U32:
 		case Opcode::DS_DEC_RTN_U32:
 		case Opcode::DS_MIN_I32:
 		case Opcode::DS_MIN_RTN_I32:
@@ -695,6 +696,7 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::DS_READ_B128:
 		case Opcode::DS_WRITE_B8:
 		case Opcode::DS_WRITE_B16:
+		case Opcode::DS_WRITE_B8_D16_HI:
 		case Opcode::DS_WRITE_B16_D16_HI:
 		case Opcode::DS_WRITE2_B32:
 		case Opcode::DS_WRITE2ST64_B32:

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,
@@ -495,7 +502,9 @@ enum class Opcode {
 	DS_ADD_RTN_U32,
 	DS_SUB_U32,
 	DS_SUB_RTN_U32,
+	DS_INC_U32,
 	DS_INC_RTN_U32,
+	DS_DEC_U32,
 	DS_DEC_RTN_U32,
 	DS_MIN_I32,
 	DS_MIN_RTN_I32,
@@ -534,6 +543,7 @@ enum class Opcode {
 	DS_READ_B128,
 	DS_WRITE_B8,
 	DS_WRITE_B16,
+	DS_WRITE_B8_D16_HI,
 	DS_WRITE_B16_D16_HI,
 	DS_WRITE2_B32,
 	DS_WRITE2ST64_B32,
@@ -559,6 +569,7 @@ enum class Opcode {
 	IMAGE_ATOMIC_OR,
 	IMAGE_ATOMIC_XOR,
 	IMAGE_SAMPLE,
+	IMAGE_GATHER4_L,
 	IMAGE_GATHER4_LZ,
 	IMAGE_GATHER4_C,
 	IMAGE_GATHER4_C_LZ,
@@ -664,6 +675,7 @@ struct Operand {
 	bool     absolute           = false;
 	bool     clamp              = false;
 	bool     dpp                = false;
+	bool     dpp8               = false;
 };
 
 struct Instruction {

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:
@@ -521,13 +521,14 @@ constexpr Vop1SdwaRule VOP1_SDWA_RULES[] = {
     {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;
@@ -234,6 +235,7 @@ Decoder::Operand MemorySourceAt(const Decoder::Instruction& decoded, uint32_t in
 				return index == 0u ? decoded.src1 : index == 1u ? decoded.src0 : decoded.src2;
 			case Decoder::Opcode::DS_WRITE_B8:
 			case Decoder::Opcode::DS_WRITE_B16:
+			case Decoder::Opcode::DS_WRITE_B8_D16_HI:
 			case Decoder::Opcode::DS_WRITE_B16_D16_HI:
 			case Decoder::Opcode::DS_WRITE_B32:
 			case Decoder::Opcode::DS_WRITE_B64:
@@ -961,8 +963,12 @@ bool Translator::EmitMemory(const Decoder::Instruction& inst) {
 			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicISub32, false);
 		case Decoder::Opcode::DS_SUB_RTN_U32:
 			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicISub32, true);
+		case Decoder::Opcode::DS_INC_U32:
+			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicInc32, false);
 		case Decoder::Opcode::DS_INC_RTN_U32:
 			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicInc32, true);
+		case Decoder::Opcode::DS_DEC_U32:
+			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicDec32, false);
 		case Decoder::Opcode::DS_DEC_RTN_U32:
 			return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicDec32, true);
 		case Decoder::Opcode::DS_MIN_I32:
@@ -1034,6 +1040,7 @@ 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:
@@ -1074,6 +1081,7 @@ bool Translator::EmitMemory(const Decoder::Instruction& inst) {
 		case Decoder::Opcode::DS_WRITE2ST64_B64: return DS_WRITE2(inst);
 		case Decoder::Opcode::DS_WRITE_B8:
 		case Decoder::Opcode::DS_WRITE_B16:
+		case Decoder::Opcode::DS_WRITE_B8_D16_HI:
 		case Decoder::Opcode::DS_WRITE_B16_D16_HI:
 		case Decoder::Opcode::DS_WRITE_B32:
 		case Decoder::Opcode::DS_WRITE_B64:

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);

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;
@@ -1065,8 +1063,8 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			}
 		} else if (options.stage == ShaderType::Mesh) {
 			const auto& mesh = options.input_info.vertex->mesh;
-			EXIT_NOT_IMPLEMENTED(options.wave_size != 64u || mesh.primitives_per_group == 0u ||
-			                     mesh.vertices_per_group > 64u || total_threads > 15u * 64u);
+			EXIT_NOT_IMPLEMENTED(mesh.primitives_per_group == 0u || mesh.vertices_per_group > 64u ||
+			                     total_threads > 15u * options.wave_size);
 			const auto u32  = [](uint32_t value) { return IR::U32(IR::Value(value)); };
 			const auto draw = [&](uint32_t index) {
 				return IR::U32(
@@ -1091,12 +1089,15 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			    entry_ir.IAdd(IR::U32(entry_ir.Emit(IR::ValueOpcode::UDiv32,
 			                                       {subtract_saturate(vertices, size), step})),
 			                  u32(1)));
-			const auto wave            = entry_ir.ShiftRightLogical(local, u32(6));
-			const auto wave_base       = entry_ir.BitwiseAnd(local, u32(~63u));
-			const auto vertex_count    = minimum(subtract_saturate(vertices, wave_base), u32(64));
-			const auto primitive_count = minimum(subtract_saturate(primitives, wave_base), u32(64));
+			const auto wave = entry_ir.ShiftRightLogical(local, u32(options.wave_size == 32u ? 5u : 6u));
+			const auto wave_base = entry_ir.BitwiseAnd(local, u32(~(options.wave_size - 1u)));
+			const auto vertex_count =
+			    minimum(subtract_saturate(vertices, wave_base), u32(options.wave_size));
+			const auto primitive_count =
+			    minimum(subtract_saturate(primitives, wave_base), u32(options.wave_size));
 			const auto wave_info = entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(wave, u32(24)),
-			                                          u32(((total_threads + 63u) / 64u) << 28u));
+			                                          u32(((total_threads + options.wave_size - 1u) /
+			                                               options.wave_size) << 28u));
 			entry_ir.SetScalarReg(
 			    static_cast<IR::ScalarReg>(3),
 			    entry_ir.BitwiseOr(wave_info, entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(
@@ -1227,6 +1228,12 @@ 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),

src/graphics/shader/recompiler/frontend/translate/Translator.h

@@ -44,7 +44,7 @@ private:
 	IR::U32 ReadU16LaneRaw(const Decoder::Operand& operand, bool high_lane);
 	IR::U32 ReadU16LaneAsU32(const Decoder::Operand& operand, bool high_lane, bool sign_extend);
 	IR::U32 ReadU16AsU32(const Decoder::Operand& operand, bool sign_extend);
-	IR::U32 ReadF16LaneBits(const Decoder::Operand& operand, bool high_lane);
+	IR::U32 Read16LaneBits(const Decoder::Operand& operand, bool high_lane);
 	std::array<IR::U32, 2> ExtractU64(IR::U64 value);
 	void    WriteU32Pair(const Decoder::Operand& operand, const std::array<IR::U32, 2>& value);
 	IR::U32 ConditionBit(const Decoder::Operand& operand);
@@ -164,6 +164,7 @@ private:
 	bool FloatCube(const Decoder::Instruction& inst, uint32_t result_kind);
 	bool Integer16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool arithmetic);
 	bool Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool sign);
+	bool V_MAD_I16(const Decoder::Instruction& inst);
 	bool V_MED3_I16(const Decoder::Instruction& inst);
 	bool PackedInteger16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
 	                          bool arithmetic);
@@ -176,6 +177,7 @@ private:
 	                     bool negate_result);
 	bool SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode opcode, IR::Type type,
 	                   bool reverse, bool mask_shift_count, bool update_scc);
+	bool S_ASHR_I64(const Decoder::Instruction& inst);
 	bool ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
 	                           bool negate_rhs, bool negate_result, bool update_scc);
 	bool V_AND_OR_B32(const Decoder::Instruction& inst);
@@ -203,6 +205,7 @@ private:
 	bool    V_BFE_U32(const Decoder::Instruction& inst, bool sign);
 	bool    V_BFI_B32(const Decoder::Instruction& inst);
 	bool    S_BITCMP_B32(const Decoder::Instruction& inst, bool expected);
+	bool    S_BITCMP_B64(const Decoder::Instruction& inst, bool expected);
 	bool    V_ALIGNBIT_B32(const Decoder::Instruction& inst);
 	bool    V_ALIGNBYTE_B32(const Decoder::Instruction& inst);
 	bool    V_LSHL_ADD_U32(const Decoder::Instruction& inst);

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

@@ -264,6 +264,9 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CMPX_LE_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPOrdLessThanEqual32, true, true);
 			return true;
+		case O::V_CMP_NGT_F16:
+			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordLessThanEqual32, true, false);
+			return true;
 		case O::V_CMPX_NGT_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordLessThanEqual32, true, true);
 			return true;
@@ -273,6 +276,9 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CMPX_NEQ_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordNotEqual32, true, true);
 			return true;
+		case O::V_CMP_NLT_F16:
+			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordGreaterThanEqual32, true, false);
+			return true;
 		case O::V_CMPX_NLT_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordGreaterThanEqual32, true, true);
 			return true;
@@ -321,6 +327,8 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_ADD_NC_I16: return Integer16Binary(inst, IR::ValueOpcode::IAdd32, false);
 		case O::V_SUB_NC_U16:
 		case O::V_SUB_NC_I16: return Integer16Binary(inst, IR::ValueOpcode::ISub32, false);
+		case O::V_MUL_LO_U16: return Integer16Binary(inst, IR::ValueOpcode::IMul32, false);
+		case O::V_MAD_I16: return V_MAD_I16(inst);
 		case O::V_MED3_I16: return V_MED3_I16(inst);
 		case O::V_MIN_I16: return Integer16Binary(inst, IR::ValueOpcode::SMin32, true);
 		case O::V_MAX_I16: return Integer16Binary(inst, IR::ValueOpcode::SMax32, true);

src/graphics/shader/recompiler/ir/Program.cpp

@@ -112,6 +112,10 @@ ResourcePlan& ResourcePlan::operator=(ResourcePlan&& other) noexcept {
 }
 
 Program::~Program() {
+	// Planning expressions can refer to block values but outlive block storage in the base class.
+	for (auto& inst: value_storage) {
+		inst.Invalidate();
+	}
 	// Values may cross block boundaries. Detach all arguments before any block starts destroying
 	// its instruction storage so reverse-use links always point to live definitions.
 	for (auto* block: blocks) {
@@ -184,6 +188,22 @@ Value ResolveInvariantPhi(const ResourcePlan& program, Value value) {
 	return invariant;
 }
 
+bool HasShaderMemoryWrites(const Program& program) {
+	for (const auto* block: program.blocks) {
+		for (const auto& inst: *block) {
+			const auto op     = inst.GetOpcode();
+			const auto buffer = BufferAccessOf(op);
+			const auto image  = ImageOpcodeInfoOf(op).access;
+			if (buffer == BufferAccess::Write || buffer == BufferAccess::Atomic ||
+			    image == ImageAccess::Write || image == ImageAccess::Atomic ||
+			    AddressOpcodeInfoOf(op).access == AddressAccess::Write) {
+				return true;
+			}
+		}
+	}
+	return false;
+}
+
 void ValidateProgram(const Program& program, bool require_ssa) {
 	if (program.blocks.size() != program.block_info.size() ||
 	    program.blocks.size() != program.block_storage.size()) {
@@ -427,13 +447,20 @@ void ValidateProgram(const Program& program, bool require_ssa) {
 					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
-				if (memory.kind != ResourceKind::Buffer &&
-				    memory.kind != ResourceKind::ScalarBuffer) {
+				const bool  vector_buffer = memory.kind == ResourceKind::Buffer ||
+				                            memory.kind == ResourceKind::IndirectBuffer;
+				if (!vector_buffer && memory.kind != ResourceKind::ScalarBuffer) {
 					return Fail(fmt::format("{} has a non-buffer resource kind",
 					                        ValueOpcodeName(inst.GetOpcode())));
 				}
+				if (memory.kind == ResourceKind::IndirectBuffer &&
+				    (memory.formatted || memory.typed ||
+				     (inst.GetOpcode() != ValueOpcode::LoadBufferU32x2 &&
+				      inst.GetOpcode() != ValueOpcode::LoadBufferU32x4))) {
+					return Fail("indirect buffer requires a raw DWORD x2/x4 load");
+				}
 				if (buffer_components > 1u &&
-				    (memory.kind != ResourceKind::Buffer || memory.data_bits != 32u ||
+				    (!vector_buffer || memory.data_bits != 32u ||
 				     memory.data_dwords != buffer_components || memory.component_index != 0u)) {
 					return Fail(fmt::format("{} has inconsistent native-wide metadata",
 					                        ValueOpcodeName(inst.GetOpcode())));

src/graphics/shader/recompiler/ir/ShaderIR.h

@@ -14,6 +14,7 @@
 
 #include <array>
 #include <bit>
+#include <deque>
 #include <list>
 #include <memory>
 #include <optional>
@@ -27,6 +28,7 @@ enum class ResourceKind {
 	ScalarBuffer,
 	ScalarAddress,
 	Buffer,
+	IndirectBuffer,
 	Flat,
 	Global,
 	Scratch,
@@ -457,11 +459,13 @@ struct BlockInfo {
 
 struct DescriptorSource {
 	struct IndirectImage {
-		uint32_t material_source = 0;
-		uint32_t heap_source     = 0;
+		uint32_t material_source = UINT32_MAX;
+		uint32_t table_source    = 0;
 		uint32_t selector_stride = 0;
 		uint32_t selector_offset = 0;
-		uint32_t key_arg         = 0;
+		uint32_t table_offset    = 0;
+		Value    key_count;
+		Value    selector_mask;
 
 		bool operator==(const IndirectImage& other) const = default;
 	};
@@ -505,9 +509,19 @@ struct UniformFillPlan {
 	std::array<Value, 4> values;
 };
 
-// Immutable runtime resource analysis retained by the shader cache. It owns descriptor/SRT,
-// uniform condition and fill values without retaining translated blocks.
+// Resource analysis retained by the shader cache. It owns immutable descriptor/SRT,
+// condition and fill values without translated blocks, plus reusable evaluation scratch.
 struct ResourcePlan {
+	struct EvaluationContext {
+		struct Entry {
+			uint64_t value      = 0;
+			uint64_t generation = 0;
+		};
+
+		std::vector<Entry> values;
+		uint64_t           generation = 0;
+	};
+
 	ResourcePlan() = default;
 	~ResourcePlan();
 
@@ -524,14 +538,23 @@ struct ResourcePlan {
 	std::vector<MemoryInfo>             memory_info;
 	std::vector<DescriptorSource>       descriptor_sources;
 	std::vector<ResourceBlock>          control_flow;
-	std::vector<uint32_t>               materialization_sources;
 	std::vector<SrtRead>                srt_reads;
 	std::vector<uint8_t>                clean_flat_slots;
 	bool                                requires_specialization_memory = false;
+	bool                                has_address_writes = false;
 	bool                                srt_plan_complete          = false;
 	bool                                resource_tracking_complete = false;
 	ShaderInfo                          info;
 	UniformFillPlan                     uniform_fill;
+	// GPU-thread scratch for nested clean/EXEC memos, activity and material keys.
+	mutable std::deque<EvaluationContext> evaluation_contexts;
+	mutable uint32_t                       evaluation_value_count = 0;
+	mutable uint32_t                       evaluation_depth       = 0;
+	mutable std::vector<uint8_t>            active_sources;
+	mutable std::vector<uint8_t>            visited_blocks;
+	mutable std::vector<uint32_t>           pending_blocks;
+	mutable std::vector<uint32_t>           material_keys;
+	mutable std::vector<std::pair<uint64_t, uint64_t>> specialization_reads;
 };
 
 struct Program: ResourcePlan {
@@ -563,6 +586,7 @@ struct Program: ResourcePlan {
 };
 
 std::string ProgramToString(const Program& program);
+bool        HasShaderMemoryWrites(const Program& program);
 
 void  ValidateProgram(const Program& program, bool require_ssa);
 void  ResolveControlFlowIdentities(Program& program);

src/graphics/shader/recompiler/ir/Value.h

@@ -121,6 +121,13 @@ public:
 	[[nodiscard]] Block*                  PhiBlock(size_t index) const;
 	[[nodiscard]] Block*                  Parent() const;
 	[[nodiscard]] const std::vector<Use>& Uses() const;
+	// Runtime indices belong to the resource plan that owns this instruction.
+	[[nodiscard]] uint32_t EvaluationIndex(uint32_t& count) const {
+		if (evaluation_index == UINT32_MAX) {
+			evaluation_index = count++;
+		}
+		return evaluation_index;
+	}
 
 	void SetParent(Block* block);
 	void SetArg(size_t index, Value value);
@@ -155,6 +162,7 @@ private:
 	std::vector<Value>  args;
 	std::vector<Block*> phi_blocks;
 	std::vector<Use>    uses;
+	mutable uint32_t    evaluation_index = UINT32_MAX;
 };
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR

src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.h

@@ -34,11 +34,12 @@ struct ImageOpcodeInfo {
 };
 
 struct DppMoveFlags {
-	uint16_t control        = 0;
-	uint8_t  row_mask       = 0xf;
-	uint8_t  bank_mask      = 0xf;
+	uint32_t control   : 24 = 0;
+	uint32_t row_mask  : 4  = 0xf;
+	uint32_t bank_mask : 4  = 0xf;
 	bool     fetch_inactive = false;
 	bool     bound_control  = false;
+	bool     dpp8           = false;
 };
 static_assert(sizeof(DppMoveFlags) <= sizeof(uint64_t));
 static_assert(std::is_trivially_copyable_v<DppMoveFlags>);

src/graphics/shader/recompiler/ir/passes/ConstantPropagation.cpp

@@ -200,6 +200,13 @@ void FoldInstruction(Block& block, Block::iterator instruction,
 	switch (inst.GetOpcode()) {
 		case ValueOpcode::Phi: FoldPhi(inst); return;
 		case ValueOpcode::SelectU1:
+			if (!FoldSelect(inst) && IsImmediate(Arg(inst, 2), Type::U1) &&
+			    !Arg(inst, 2).U1()) {
+				auto result = block.PrependNewInst(instruction, ValueOpcode::LogicalAnd,
+				                                   {Arg(inst, 0), Arg(inst, 1)});
+				Replace(inst, Value(&*result));
+			}
+			return;
 		case ValueOpcode::SelectU32:
 		case ValueOpcode::SelectF32: FoldSelect(inst); return;
 		case ValueOpcode::BitFieldInsert: {
@@ -431,6 +438,10 @@ void FoldInstruction(Block& block, Block::iterator instruction,
 			return;
 		case ValueOpcode::ShiftRightLogical32:
 			if (!FoldU32(inst, [](uint32_t a, uint32_t b) { return a >> (b & 31u); })) {
+				if (IsImmediate(Arg(inst, 0), Type::U32) && Arg(inst, 0).U32() == 0u) {
+					Replace(inst, Value(0u));
+					return;
+				}
 				const auto shift = Arg(inst, 1);
 				if (IsImmediate(shift, Type::U32) && (shift.U32() & 31u) == 0u) {
 					Replace(inst, Arg(inst, 0));
@@ -573,6 +584,23 @@ void FoldInstruction(Block& block, Block::iterator instruction,
 			if (!FoldLogical(inst, [](bool a, bool b) { return a && b; })) {
 				const auto lhs = Arg(inst, 0);
 				const auto rhs = Arg(inst, 1);
+				const auto simplify = [&](Value assumption, Value expression) {
+					const auto* disjunction = expression.TryInstruction();
+					if (disjunction == nullptr || disjunction->GetOpcode() != ValueOpcode::LogicalOr) {
+						return false;
+					}
+					for (uint32_t i = 0; i < 2u; ++i) {
+						const auto* inverse = disjunction->Arg(i).Resolve().TryInstruction();
+						if (inverse != nullptr && inverse->GetOpcode() == ValueOpcode::LogicalNot &&
+						    inverse->Arg(0).Resolve() == assumption) {
+							inst.SetArg(assumption == lhs ? 1u : 0u,
+							            disjunction->Arg(i ^ 1u));
+							return true;
+						}
+					}
+					return false;
+				};
+				if (simplify(lhs, rhs) || simplify(rhs, lhs)) return;
 				if (IsImmediate(lhs, Type::U1)) {
 					Replace(inst, lhs.U1() ? rhs : lhs);
 				} else if (IsImmediate(rhs, Type::U1)) {

src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp

@@ -21,18 +21,6 @@ namespace {
 constexpr uint64_t AddressMask            = 0x0000ffffffffffffull;
 constexpr uint64_t MaxIndirectImageProbes = 65536u;
 
-struct IndirectImage {
-	uint32_t                     resource = 0;
-	std::vector<uint32_t>        keys;
-	std::vector<uint32_t>        candidates;
-	std::vector<DescriptorValue> descriptors;
-};
-
-struct MaterializedSnapshot {
-	ResourceSnapshot           resources;
-	std::vector<IndirectImage> indirect_images;
-};
-
 bool SpecializationFail(std::string_view message) {
 	std::fprintf(stderr, "shader resource specialization failed: %.*s\n",
 	             static_cast<int>(message.size()), message.data());
@@ -74,6 +62,13 @@ bool NullImageDescriptor(const DescriptorValue& descriptor) {
 }
 
 bool ValidImageDescriptor(const DescriptorValue& descriptor, bool r128 = false) {
+	const auto& words = descriptor.dwords;
+	// Reject texture descriptors with nonzero reserved bits.
+	if ((words[1] & 0x20000000u) != 0u || (words[2] & 0x70003000u) != 0u ||
+	    (!r128 && ((words[4] & 0xe000e000u) != 0u || (words[5] & 0xf9000000u) != 0u ||
+	               (words[6] & 0x00007b00u) != 0u))) {
+		return false;
+	}
 	const auto type   = static_cast<Prospero::ImageType>((descriptor.dwords[3] >> 28u) & 0xfu);
 	const auto format = static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
 	if (type < Prospero::ImageType::kColor1D || format == Prospero::BufferFormat::kInvalid ||
@@ -84,6 +79,13 @@ bool ValidImageDescriptor(const DescriptorValue& descriptor, bool r128 = false)
 	    type != Prospero::ImageType::kColor2DMsaa) {
 		return false;
 	}
+	const bool array = type == Prospero::ImageType::kColor1DArray ||
+	                   type == Prospero::ImageType::kColor2DArray ||
+	                   type == Prospero::ImageType::kColor2DMsaaArray ||
+	                   type == Prospero::ImageType::kCube;
+	if (array && ((words[4] >> 16u) & 0x1fffu) > (words[4] & 0x1fffu)) {
+		return false;
+	}
 	if (type == Prospero::ImageType::kColor2DMsaa ||
 	    type == Prospero::ImageType::kColor2DMsaaArray) {
 		const auto base_level = (descriptor.dwords[3] >> 12u) & 0xfu;
@@ -136,6 +138,50 @@ bool DecodeBufferDescriptor(const DescriptorValue& descriptor, ShaderBufferResou
 	return true;
 }
 
+struct ReadCapture {
+	SrtRuntime                                  source;
+	std::vector<std::pair<uint64_t, uint64_t>>& ranges;
+};
+
+bool CaptureStrictRead(void* userdata, uint64_t address, std::span<uint32_t> values) {
+	auto& capture = *static_cast<ReadCapture*>(userdata);
+	if (!capture.source.read_specialization_memory(capture.source.userdata, address, values)) {
+		return false;
+	}
+	capture.ranges.emplace_back(address, values.size_bytes());
+	return true;
+}
+
+bool CaptureOrdinaryRead(void* userdata, uint64_t address, std::span<uint32_t> values) {
+	auto& capture = *static_cast<ReadCapture*>(userdata);
+	if (!capture.source.read_memory(capture.source.userdata, address, values)) {
+		return false;
+	}
+	capture.ranges.emplace_back(address, values.size_bytes());
+	return true;
+}
+
+bool WrittenBuffersDisjoint(const ResourcePlan& program, const ResourceSnapshot& snapshot,
+                            std::span<const std::pair<uint64_t, uint64_t>> reads) {
+	for (uint32_t i = 0; i < program.info.buffers.size(); ++i) {
+		if (!program.info.buffers[i].written) continue;
+		ShaderBufferResource buffer;
+		if (!DecodeBufferDescriptor(snapshot.buffers[i], buffer)) return false;
+		const auto base = buffer.Base48();
+		const auto size = buffer.GetSize();
+		if (size == 0u) continue;
+		if (size - 1u > AddressMask - base) return false;
+		for (const auto [address, bytes]: reads) {
+			if (bytes == 0u) continue;
+			if (address > AddressMask || bytes - 1u > AddressMask - address ||
+			    (address <= base ? base - address < bytes : address - base < size)) {
+				return false;
+			}
+		}
+	}
+	return true;
+}
+
 const DescriptorSource* Source(const ResourcePlan& program, uint32_t source) {
 	if (source >= program.descriptor_sources.size()) {
 		return nullptr;
@@ -144,12 +190,15 @@ const DescriptorSource* Source(const ResourcePlan& program, uint32_t source) {
 }
 
 void MarkCleanFlatSlots(const ResourcePlan& program, const DescriptorSource* source,
-                        std::span<uint8_t> slots) {
-	if (source == nullptr) {
+                        std::span<uint8_t> slots, Value extra = {}) {
+	if (source == nullptr && extra.IsEmpty()) {
 		return;
 	}
-	std::vector<Value>       pending(source->dwords.begin(),
-	                                 source->dwords.begin() + source->dword_count);
+	std::vector<Value>       pending;
+	if (source != nullptr) {
+		pending.assign(source->dwords.begin(), source->dwords.begin() + source->dword_count);
+	}
+	if (!extra.IsEmpty()) pending.push_back(extra);
 	std::vector<const Inst*> visited;
 	while (!pending.empty()) {
 		auto value = pending.back().Resolve();
@@ -173,200 +222,185 @@ void MarkCleanFlatSlots(const ResourcePlan& program, const DescriptorSource* sou
 	}
 }
 
-bool ReadSpecializationWord(const SrtRuntime& runtime, uint64_t address, uint32_t& word) {
-	return runtime.read_specialization_memory != nullptr &&
-	       runtime.read_specialization_memory(runtime.userdata, address, &word);
-}
-
-bool ReadScalarBufferWord(const ShaderBufferResource& descriptor, uint32_t dynamic_offset,
-                          uint32_t immediate_offset, const SrtRuntime& runtime, uint32_t& word) {
-	const auto byte_offset = static_cast<uint64_t>(dynamic_offset) + immediate_offset;
-	const auto aligned     = byte_offset & ~uint64_t {3};
-	const auto size        = descriptor.GetSize();
-	if (aligned > size || size - aligned < sizeof(uint32_t)) {
-		word = 0;
+bool ReadScalarTable(uint64_t base, uint64_t size, uint32_t dynamic_offset,
+                     const SrtRuntime& runtime, std::span<uint32_t> words) {
+	const auto offset = static_cast<uint64_t>(dynamic_offset) & ~uint64_t {3};
+	const auto count = std::min<uint64_t>(words.size(), offset < size ? (size - offset) / 4u : 0u);
+	std::ranges::fill(words.subspan(count), 0u);
+	if (count == 0u) {
 		return true;
 	}
-	const auto base = descriptor.Base48() & ~uint64_t {3};
-	if (aligned > AddressMask - base) {
+	base &= AddressMask & ~uint64_t {3};
+	if (offset > AddressMask - base) {
 		return false;
 	}
-	const auto address = base + aligned;
-	if (!ReadSpecializationWord(runtime, address, word)) {
-		return false;
-	}
-	return true;
+	const auto address = base + offset;
+	const auto prefix = words.first(count);
+	return prefix.size_bytes() - 1u <= AddressMask - address &&
+	       runtime.read_specialization_memory != nullptr &&
+	       runtime.read_specialization_memory(runtime.userdata, address, prefix);
 }
 
-bool MaterializeIndirectImage(const DescriptorSource::IndirectImage& indirect,
-                              const DescriptorValue&                 material_value,
-                              const DescriptorValue& heap_value, bool r128,
-                              const SrtRuntime& runtime, IndirectImage& result) {
-	ShaderBufferResource material;
-	ShaderBufferResource heap;
-	if (!DecodeBufferDescriptor(material_value, material) ||
-	    !DecodeBufferDescriptor(heap_value, heap)) {
+bool MaterializeIndirectImage(const ResourcePlan& program,
+                              const DescriptorSource::IndirectImage& indirect,
+                              const DescriptorValue& material_value,
+                              const DescriptorValue& table_value, uint32_t image_index,
+                              const SrtRuntime& runtime, SrtWalker& clean,
+                              ResourceSnapshot& snapshot,
+                              ResourceSpecialization& specialization) {
+	uint64_t table_base = 0;
+	uint64_t table_size = UINT64_MAX; // Scalar addresses have no buffer descriptor bounds.
+	ShaderBufferResource table;
+	if (table_value.dword_count == 2u) {
+		table_base = (static_cast<uint64_t>(table_value.dwords[1]) << 32u) | table_value.dwords[0];
+	} else if (DecodeBufferDescriptor(table_value, table)) {
+		table_base = table.Base48();
+		table_size = table.GetSize();
+	} else {
 		return false;
 	}
-	if (material.Stride() != indirect.selector_stride) {
-		return false;
-	}
-
-	// S_BUFFER_LOAD ignores vector-buffer swizzle/add-thread fields. The shader computes the
-	// record stride explicitly; enumerate every wrapped 32-bit offset that can pass bounds.
-	const auto period      = uint64_t {1} << 32u;
-	const auto step        = std::gcd<uint64_t>(indirect.selector_stride, period);
-	const auto residue     = static_cast<uint64_t>(indirect.selector_offset) % step;
-	const auto size        = material.GetSize();
-	const auto limit       = std::min<uint64_t>(UINT32_MAX, size + 3u);
-	const auto probe_count = residue <= limit ? (limit - residue) / step + 1u : 0u;
-	if (probe_count > MaxIndirectImageProbes) {
-		return false;
-	}
-
-	std::vector<uint32_t>        keys {0u};
-	std::unordered_set<uint32_t> seen {0u};
-	keys.reserve(static_cast<size_t>(probe_count) + 1u);
-	seen.reserve(static_cast<size_t>(probe_count) + 1u);
-	for (uint64_t offset = residue; offset <= limit && probe_count != 0u; offset += step) {
-		uint32_t key = 0;
-		if (!ReadScalarBufferWord(material, static_cast<uint32_t>(offset), 0u, runtime, key)) {
+	auto& keys = program.material_keys;
+	keys.clear();
+	if (indirect.material_source == UINT32_MAX) {
+		uint32_t key_count = 0;
+		const bool evaluated = clean.Evaluate(indirect.key_count, key_count);
+		if (std::bit_cast<int32_t>(key_count) <= 0) key_count = 0;
+		if (table_value.dword_count != 2u || !evaluated ||
+		    key_count > MaxIndirectImageProbes ||
+		    uint64_t {indirect.table_offset} + uint64_t {key_count} * 32u > UINT32_MAX + 1ull) {
 			return false;
 		}
-		if (seen.insert(key).second) {
+		keys.resize(key_count);
+		std::iota(keys.begin(), keys.end(), 0u);
+	} else if (!indirect.selector_mask.IsEmpty()) {
+		uint32_t mask = 0;
+		uint32_t count = 0;
+		if (material_value.dword_count != 2u || table_value.dword_count != 2u ||
+		    !clean.Evaluate(indirect.selector_mask, mask) ||
+		    !clean.Evaluate(indirect.key_count, count) || count == 0u || count > 32u) {
+			return false;
+		}
+		if (count < 32u) mask &= (1u << count) - 1u;
+		const auto material_base =
+		    (static_cast<uint64_t>(material_value.dwords[1]) << 32u) | material_value.dwords[0];
+		keys.reserve(std::popcount(mask));
+		while (mask != 0u) {
+			const auto index = std::countr_zero(mask);
+			const auto offset = static_cast<uint64_t>(indirect.selector_offset) +
+			                    static_cast<uint64_t>(index) * indirect.selector_stride;
+			if (offset > UINT32_MAX) return false;
+			uint32_t key = 0;
+			if (!ReadScalarTable(material_base, UINT64_MAX, static_cast<uint32_t>(offset),
+			                     runtime, {&key, 1})) return false;
 			keys.push_back(key);
+			mask &= mask - 1u;
 		}
-		if (limit - offset < step) {
-			break;
+		std::ranges::sort(keys);
+		keys.erase(std::unique(keys.begin(), keys.end()), keys.end());
+	} else {
+		ShaderBufferResource material;
+		if (!DecodeBufferDescriptor(material_value, material) || table_value.dword_count != 4u ||
+		    material.Stride() != indirect.selector_stride) {
+			return false;
 		}
+		// Enumerate every wrapped scalar-buffer offset that can pass the descriptor bounds.
+		const auto step = std::gcd<uint64_t>(indirect.selector_stride, uint64_t {1} << 32u);
+		const auto residue = static_cast<uint64_t>(indirect.selector_offset) % step;
+		const auto limit = std::min<uint64_t>(UINT32_MAX, material.GetSize() + 3u);
+		const auto probe_count = residue <= limit ? (limit - residue) / step + 1u : 0u;
+		if (probe_count > MaxIndirectImageProbes) {
+			return false;
+		}
+		keys.reserve(static_cast<size_t>(probe_count) + 1u);
+		keys.push_back(0u);
+		for (uint64_t offset = residue; offset <= limit && probe_count != 0u; offset += step) {
+			uint32_t key = 0;
+			if (!ReadScalarTable(material.Base48(), material.GetSize(),
+			                     static_cast<uint32_t>(offset), runtime, {&key, 1})) {
+				return false;
+			}
+			keys.push_back(key);
+			if (limit - offset < step) {
+				break;
+			}
+		}
+		std::ranges::sort(keys);
+		keys.erase(std::unique(keys.begin(), keys.end()), keys.end());
 	}
 
-	IndirectImage next;
-	next.keys = std::move(keys);
-	next.candidates.reserve(next.keys.size());
-	next.descriptors.reserve(
-	    std::min(next.keys.size(), static_cast<size_t>(ShaderInfo::MaxImages)));
-	for (const auto key: next.keys) {
+	const auto children_begin = snapshot.images.size();
+	const auto mapping_offset = snapshot.flattened_srt.size();
+	const auto root_image = specialization.images[image_index];
+	snapshot.flattened_srt.resize(mapping_offset + 1u + keys.size() * 2u);
+	snapshot.flattened_srt[mapping_offset] = static_cast<uint32_t>(keys.size());
+	for (uint32_t entry = 0; entry < keys.size(); ++entry) {
+		const auto key = keys[entry];
 		DescriptorValue candidate;
-		candidate.dword_count  = 8u;
-		const auto heap_offset = key << 5u;
-		for (uint32_t dword = 0; dword < candidate.dword_count; dword++) {
-			if (!ReadScalarBufferWord(heap, heap_offset, dword * sizeof(uint32_t), runtime,
-			                          candidate.dwords[dword])) {
-				return false;
-			}
+		candidate.dword_count = 8u;
+		const auto table_offset = (key << 5u) + indirect.table_offset;
+		if (!ReadScalarTable(table_base, table_size, table_offset, runtime, candidate.dwords)) {
+			return false;
 		}
-		if (NullImageDescriptor(candidate) || !ValidImageDescriptor(candidate, r128)) {
+		if (NullImageDescriptor(candidate) ||
+		    !ValidImageDescriptor(candidate, program.info.images[image_index].r128)) {
 			candidate.dwords.fill(0);
 		}
-		const auto found = std::ranges::find(next.descriptors, candidate);
-		if (found == next.descriptors.end()) {
-			if (next.descriptors.size() >= ShaderInfo::MaxImages) {
-				return false;
+		uint32_t ordinal = 0;
+		if (entry == 0) {
+			snapshot.images[image_index] = candidate;
+		} else if (snapshot.images[image_index] != candidate) {
+			const auto found = std::find(snapshot.images.begin() + children_begin,
+			                             snapshot.images.end(), candidate);
+			ordinal = static_cast<uint32_t>(found - snapshot.images.begin() - children_begin + 1u);
+			if (found == snapshot.images.end()) {
+				if (snapshot.images.size() >= ShaderInfo::MaxImages) {
+					return false;
+				}
+				snapshot.images.push_back(candidate);
+				auto child = root_image;
+				child.indirect_root = image_index;
+				specialization.images.push_back(child);
 			}
-			next.descriptors.push_back(candidate);
-			next.candidates.push_back(static_cast<uint32_t>(next.descriptors.size() - 1u));
-		} else {
-			next.candidates.push_back(static_cast<uint32_t>(found - next.descriptors.begin()));
 		}
+		snapshot.flattened_srt[mapping_offset + 1u + entry * 2u] = key;
+		snapshot.flattened_srt[mapping_offset + 2u + entry * 2u] = ordinal;
+	}
+	if (snapshot.images.size() == children_begin) {
+		snapshot.flattened_srt.resize(mapping_offset);
+	} else {
+		auto& root = specialization.images[image_index];
+		root.indirect_root = image_index;
+		root.indirect_mapping_offset = static_cast<uint32_t>(mapping_offset);
+		root.indirect_search_iterations = std::bit_width(keys.size());
 	}
-	result = std::move(next);
 	return true;
 }
 
 } // namespace
 
-static bool MaterializeSnapshot(const ResourcePlan& program, const SrtRuntime& runtime,
-                                MaterializedSnapshot& snapshot) {
-	if (!program.resource_tracking_complete) {
-		return false;
-	}
-
-	if (program.requires_specialization_memory && runtime.read_specialization_memory == nullptr) {
-		return false;
-	}
-	std::vector<DescriptorValue> values;
-	std::vector<uint32_t>        flattened_srt;
-	std::vector<uint8_t>         active_sources;
-	if (!EvaluateRuntimeSources(program, program.materialization_sources, runtime, values,
-	                            flattened_srt, program.clean_flat_slots, active_sources)) {
-		return false;
-	}
-
-	auto&                   next  = snapshot.resources;
-	const auto&             fill  = program.uniform_fill;
-	const auto              words = fill.fill.words;
-	std::array<uint32_t, 4> stored {};
-	if (words != 0 &&
-	    EvaluateUniformValues(program, std::span(fill.values).first(words), runtime,
-	                          std::span(stored).first(words)) &&
-	    std::all_of(stored.begin(), stored.begin() + words,
-	                [&](uint32_t value) { return value == stored[0]; })) {
-		next.uniform_fill       = fill.fill;
-		next.uniform_fill.value = stored[0];
-	}
-	auto cursor = values.begin();
-	next.buffers.assign(cursor, cursor + program.info.buffers.size());
-	cursor += program.info.buffers.size();
-	next.flattened_srt = std::move(flattened_srt);
-	next.images.resize(program.info.images.size());
-	for (uint32_t image_index = 0; image_index < program.info.images.size(); image_index++) {
-		const auto& image  = program.info.images[image_index];
-		const auto* source = Source(program, image.source);
-		if (source != nullptr && source->indirect_image.has_value()) {
-			if (!active_sources[image.source]) {
-				next.images[image_index].dword_count = 8u;
-				continue;
-			}
-			const std::array requests {source->indirect_image->material_source,
-			                           source->indirect_image->heap_source};
-			SrtRuntime       clean_runtime = runtime;
-			clean_runtime.read_memory      = runtime.read_specialization_memory;
-			std::vector<DescriptorValue> tables;
-			if (!EvaluateDescriptorSources(program, requests, clean_runtime, tables)) {
-				return false;
-			}
-			const auto&   material = tables[0];
-			const auto&   heap     = tables[1];
-			IndirectImage table;
-			if (!MaterializeIndirectImage(*source->indirect_image, material, heap, image.r128,
-			                              runtime, table)) {
-				return false;
-			}
-			next.images[image_index] = table.descriptors[table.candidates[0]];
-			if (table.descriptors.size() > 1u) {
-				table.resource = image_index;
-				snapshot.indirect_images.push_back(std::move(table));
-			}
-		} else {
-			auto descriptor = *cursor++;
-			if (!ValidImageDescriptor(descriptor, image.r128)) {
-				descriptor.dwords.fill(0);
-			}
-			next.images[image_index] = descriptor;
-		}
-	}
-	next.samplers.assign(cursor, cursor + program.info.samplers.size());
-	next.user_data.assign(runtime.user_data.begin(), runtime.user_data.end());
-	return true;
-}
-
 struct SamplerPlan {
 	std::array<uint32_t, ShaderInfo::MaxSamplers> point_sampler {};
 	uint32_t                                      sampler_count = 0;
 };
 
 struct ImageRemap {
-	explicit ImageRemap(const ResourceSpecialization& specialization) {
-		for (const auto& image: specialization.images) {
-			indices.push_back(image.fmask ? UINT32_MAX : count++);
+	explicit ImageRemap(const ResourceSpecialization& specialization)
+	    : source_count(static_cast<uint32_t>(specialization.images.size())) {
+		EXIT_IF(specialization.images.size() > indices.size());
+		for (uint32_t index = 0; index < source_count; index++) {
+			indices[index] = specialization.images[index].fmask ? UINT32_MAX : count++;
 		}
 	}
 
+	uint32_t operator[](uint32_t index) const {
+		EXIT_IF(index >= source_count);
+		return indices[index];
+	}
+
 	template <typename T>
 	void Apply(std::vector<T>& images) const {
-		EXIT_IF(images.size() != indices.size());
-		for (uint32_t index = 0; index < indices.size(); index++) {
+		EXIT_IF(images.size() != source_count);
+		for (uint32_t index = 0; index < source_count; index++) {
 			if (indices[index] != UINT32_MAX && indices[index] != index) {
 				images[indices[index]] = std::move(images[index]);
 			}
@@ -374,75 +408,21 @@ struct ImageRemap {
 		images.resize(count);
 	}
 
-	std::vector<uint32_t> indices;
-	uint32_t              count = 0;
+private:
+	std::array<uint32_t, ShaderInfo::MaxImages> indices;
+	uint32_t                                    source_count;
+	uint32_t                                    count = 0;
 };
 
 template <typename Images>
 bool BuildSamplerPlan(const ShaderInfo& base, const Images& images, SamplerPlan& plan);
 
-static bool BuildResourceSpecialization(const ResourcePlan& program, MaterializedSnapshot snapshot,
-                                        ResourceSnapshot&       specialized_snapshot,
+static bool BuildResourceSpecialization(const ResourcePlan& program, ResourceSnapshot& snapshot,
                                         ResourceSpecialization& specialization) {
-	auto                   next_snapshot = std::move(snapshot.resources);
-	ResourceSpecialization next_specialization;
-	next_specialization.buffers.reserve(program.info.buffers.size());
-	size_t image_count   = program.info.images.size();
-	size_t mapping_words = 0;
-	for (const auto& table: snapshot.indirect_images) {
-		if (table.resource >= program.info.images.size() || table.descriptors.size() < 2u ||
-		    image_count + table.descriptors.size() - 1u > ShaderInfo::MaxImages) {
-			return SpecializationFail(
-			    "indirect image candidates exceed the dense image resource limit");
-		}
-		image_count += table.descriptors.size() - 1u;
-		mapping_words += 1u + table.keys.size() * 2u;
-	}
-	next_snapshot.images.reserve(image_count);
-	next_snapshot.flattened_srt.reserve(next_snapshot.flattened_srt.size() + mapping_words);
-	next_specialization.images.reserve(image_count);
-	for (const auto& image: program.info.images) {
-		next_specialization.images.push_back({
-		    .numeric_class              = image.numeric_class,
-		    .dimension                  = image.dimension,
-		    .mip_count                  = image.mip_count,
-		    .conversion_format          = image.conversion_format,
-		    .shader_swizzle             = image.shader_swizzle,
-		    .indirect_root              = image.indirect_root,
-		    .indirect_mapping_offset    = image.indirect_mapping_offset,
-		    .indirect_search_iterations = image.indirect_search_iterations,
-		    .cube                       = image.cube,
-		});
-	}
-	for (const auto& table: snapshot.indirect_images) {
-		const auto root_image = next_specialization.images[table.resource];
-		for (uint32_t candidate = 1; candidate < table.descriptors.size(); candidate++) {
-			auto image          = root_image;
-			image.indirect_root = table.resource;
-			next_specialization.images.push_back(image);
-			next_snapshot.images.push_back(table.descriptors[candidate]);
-		}
-		auto& root                      = next_specialization.images[table.resource];
-		root.indirect_root              = table.resource;
-		root.indirect_mapping_offset    = static_cast<uint32_t>(next_snapshot.flattened_srt.size());
-		root.indirect_search_iterations = std::bit_width(table.keys.size());
-		next_snapshot.flattened_srt.resize(next_snapshot.flattened_srt.size() + 1u +
-		                                   table.keys.size() * 2u);
-		std::vector<uint32_t> order(table.keys.size());
-		std::iota(order.begin(), order.end(), 0u);
-		std::ranges::sort(order, {}, [&](uint32_t index) { return table.keys[index]; });
-		next_snapshot.flattened_srt[root.indirect_mapping_offset] =
-		    static_cast<uint32_t>(table.keys.size());
-		for (uint32_t entry = 0; entry < order.size(); entry++) {
-			const auto source                   = order[entry];
-			const auto offset                   = root.indirect_mapping_offset + 1u + entry * 2u;
-			next_snapshot.flattened_srt[offset] = table.keys[source];
-			next_snapshot.flattened_srt[offset + 1] = table.candidates[source];
-		}
-		next_snapshot.images[table.resource] = table.descriptors[0];
-	}
+	specialization.buffers.clear();
+	specialization.buffers.reserve(program.info.buffers.size());
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
-		auto&                descriptor_value = next_snapshot.buffers[i];
+		auto&                descriptor_value = snapshot.buffers[i];
 		ShaderBufferResource descriptor;
 		if (!DecodeBufferDescriptor(descriptor_value, descriptor)) {
 			return SpecializationFail(fmt::format("buffer descriptor {} has invalid width", i));
@@ -459,7 +439,7 @@ static bool BuildResourceSpecialization(const ResourcePlan& program, Materialize
 		} else if (!swizzle) {
 			packed_stride &= ~(3u << 16u);
 		}
-		next_specialization.buffers.push_back({
+		specialization.buffers.push_back({
 		    .packed_stride     = packed_stride,
 		    .descriptor_format = program.info.buffers[i].formatted
 		                             ? descriptor.Format()
@@ -468,9 +448,9 @@ static bool BuildResourceSpecialization(const ResourcePlan& program, Materialize
 		        program.info.buffers[i].formatted ? descriptor.DstSelXYZW() : DstSel(4, 5, 6, 7),
 		});
 	}
-	for (uint32_t i = 0; i < next_specialization.images.size(); i++) {
-		const auto& descriptor = next_snapshot.images[i];
-		auto&       image      = next_specialization.images[i];
+	for (uint32_t i = 0; i < specialization.images.size(); i++) {
+		const auto& descriptor = snapshot.images[i];
+		auto&       image      = specialization.images[i];
 		const auto  base_index = i < program.info.images.size() ? i : image.indirect_root;
 		if (base_index >= program.info.images.size()) {
 			return SpecializationFail(fmt::format("image resource {} has an invalid root", i));
@@ -545,42 +525,46 @@ static bool BuildResourceSpecialization(const ResourcePlan& program, Materialize
 			                static_cast<uint32_t>(format)));
 		}
 	}
-	for (uint32_t root_index = 0; root_index < next_specialization.images.size(); root_index++) {
-		auto& root = next_specialization.images[root_index];
+	for (uint32_t root_index = 0; root_index < specialization.images.size(); root_index++) {
+		auto& root = specialization.images[root_index];
 		if (root.indirect_root != root_index) {
 			continue;
 		}
-		const auto key_count = root.indirect_mapping_offset < next_snapshot.flattened_srt.size()
-		                           ? next_snapshot.flattened_srt[root.indirect_mapping_offset]
+		const auto key_count = root.indirect_mapping_offset < snapshot.flattened_srt.size()
+		                           ? snapshot.flattened_srt[root.indirect_mapping_offset]
 		                           : 0u;
 		if (root.indirect_search_iterations == 0u || key_count < 2u ||
 		    static_cast<size_t>(root.indirect_mapping_offset) + 1u +
 		            static_cast<size_t>(key_count) * 2u >
-		        next_snapshot.flattened_srt.size()) {
+		        snapshot.flattened_srt.size()) {
 			return SpecializationFail("indirect image specialization has an invalid key mapping");
 		}
 		uint32_t exemplar       = ImageResource::NoIndirectImage;
 		uint32_t resource_count = 0;
-		for (uint32_t resource = 0; resource < next_specialization.images.size(); resource++) {
-			if (next_specialization.images[resource].indirect_root != root_index) {
+		for (uint32_t resource = 0; resource < specialization.images.size(); resource++) {
+			if (specialization.images[resource].indirect_root != root_index) {
 				continue;
 			}
 			resource_count++;
 			if (exemplar == ImageResource::NoIndirectImage &&
-			    !NullImageDescriptor(next_snapshot.images[resource])) {
+			    !NullImageDescriptor(snapshot.images[resource])) {
 				exemplar = resource;
 			}
 		}
 		if (resource_count < 2u || exemplar == ImageResource::NoIndirectImage) {
 			return SpecializationFail("indirect image specialization has no typed candidate");
 		}
-		const auto& image_class = next_specialization.images[exemplar];
-		for (uint32_t candidate = 0; candidate < next_specialization.images.size(); candidate++) {
-			auto& image = next_specialization.images[candidate];
+		const auto& image_class = specialization.images[exemplar];
+		const auto is_2d = [](Decoder::ImageDimension dimension) {
+			return dimension == Decoder::ImageDimension::Dim2D ||
+			       dimension == Decoder::ImageDimension::Dim2DArray;
+		};
+		for (uint32_t candidate = 0; candidate < specialization.images.size(); candidate++) {
+			auto& image = specialization.images[candidate];
 			if (image.indirect_root != root_index) {
 				continue;
 			}
-			if (NullImageDescriptor(next_snapshot.images[candidate])) {
+			if (NullImageDescriptor(snapshot.images[candidate])) {
 				image.numeric_class     = image_class.numeric_class;
 				image.dimension         = image_class.dimension;
 				image.mip_count         = image_class.mip_count;
@@ -588,12 +572,13 @@ static bool BuildResourceSpecialization(const ResourcePlan& program, Materialize
 				image.shader_swizzle    = image_class.shader_swizzle;
 				image.cube              = image_class.cube;
 			}
+			const bool same_coordinates = image.dimension == image_class.dimension &&
+			                              image.cube == image_class.cube;
 			if (image.numeric_class != image_class.numeric_class ||
-			    image.dimension != image_class.dimension ||
+			    (!same_coordinates && !(is_2d(image.dimension) && is_2d(image_class.dimension))) ||
 			    image.mip_count != image_class.mip_count ||
 			    image.conversion_format != image_class.conversion_format ||
-			    image.shader_swizzle != image_class.shader_swizzle ||
-			    image.cube != image_class.cube) {
+			    image.shader_swizzle != image_class.shader_swizzle) {
 				return SpecializationFail(
 				    fmt::format("indirect image table at pc 0x{:08x} has incompatible candidates",
 				                program.info.images[root_index].first_use_pc));
@@ -601,18 +586,16 @@ static bool BuildResourceSpecialization(const ResourcePlan& program, Materialize
 		}
 	}
 	SamplerPlan sampler_plan;
-	if (!BuildSamplerPlan(program.info, next_specialization.images, sampler_plan)) {
+	if (!BuildSamplerPlan(program.info, specialization.images, sampler_plan)) {
 		return SpecializationFail("specialized sampler layout exceeds its resource limit");
 	}
 	for (uint32_t index = 0; index < program.info.samplers.size(); index++) {
 		const auto target = sampler_plan.point_sampler[index];
 		if (target != UINT32_MAX && target >= program.info.samplers.size()) {
-			next_snapshot.samplers.push_back(next_snapshot.samplers[index]);
+			snapshot.samplers.push_back(snapshot.samplers[index]);
 		}
 	}
-	ImageRemap(next_specialization).Apply(next_snapshot.images);
-	specialization       = std::move(next_specialization);
-	specialized_snapshot = std::move(next_snapshot);
+	ImageRemap(specialization).Apply(snapshot.images);
 	return true;
 }
 
@@ -647,7 +630,8 @@ bool BuildSamplerPlan(const ShaderInfo& base, const Images& images, SamplerPlan&
 }
 
 static std::vector<ResourceBlock> ResourceControlFlow(const Program& program) {
-	if (program.blocks.size() != program.block_info.size()) {
+	// Any shader write may alias a scalar predicate read, including on a later loop visit.
+	if (program.blocks.size() != program.block_info.size() || HasShaderMemoryWrites(program)) {
 		return {};
 	}
 	std::unordered_map<uint32_t, uint32_t> indices;
@@ -687,12 +671,6 @@ static std::vector<ResourceBlock> ResourceControlFlow(const Program& program) {
 			const auto op     = inst.GetOpcode();
 			const auto buffer = BufferAccessOf(op);
 			const auto image  = ImageOpcodeInfoOf(op);
-			// Any shader write may alias a scalar predicate read, including on a later loop visit.
-			if (buffer == BufferAccess::Write || buffer == BufferAccess::Atomic ||
-			    image.access == ImageAccess::Write || image.access == ImageAccess::Atomic ||
-			    AddressOpcodeInfoOf(op).access == AddressAccess::Write) {
-				return {};
-			}
 			if (buffer == BufferAccess::None && image.access == ImageAccess::None) {
 				continue;
 			}
@@ -878,6 +856,7 @@ ResourcePlan ExtractResourcePlan(const Program& program) {
 	plan.memory_info                = program.memory_info;
 	plan.srt_plan_complete          = program.srt_plan_complete;
 	plan.resource_tracking_complete = program.resource_tracking_complete;
+	plan.has_address_writes         = program.has_address_writes;
 
 	std::unordered_map<const Inst*, Inst*> cloned;
 	std::function<Value(Value)>            Clone = [&](Value value) -> Value {
@@ -915,6 +894,10 @@ ResourcePlan ExtractResourcePlan(const Program& program) {
 		auto& target          = plan.descriptor_sources.emplace_back();
 		target.dword_count    = source.dword_count;
 		target.indirect_image = source.indirect_image;
+		if (target.indirect_image.has_value()) {
+			target.indirect_image->key_count = Clone(target.indirect_image->key_count);
+			target.indirect_image->selector_mask = Clone(target.indirect_image->selector_mask);
+		}
 		for (uint32_t dword = 0; dword < source.dword_count; dword++) {
 			target.dwords[dword] = Clone(source.dwords[dword]);
 		}
@@ -931,31 +914,16 @@ ResourcePlan ExtractResourcePlan(const Program& program) {
 	for (uint32_t i = 0; i < plan.uniform_fill.fill.words; ++i) {
 		plan.uniform_fill.values[i] = Clone(plan.uniform_fill.values[i]);
 	}
-	plan.materialization_sources.reserve(plan.info.buffers.size() + plan.info.images.size() +
-	                                     plan.info.samplers.size());
-	for (const auto& buffer: plan.info.buffers) {
-		plan.materialization_sources.push_back(buffer.source);
-	}
-	for (const auto& image: plan.info.images) {
-		const auto* source = Source(plan, image.source);
-		if (source != nullptr && source->indirect_image.has_value()) {
-			plan.requires_specialization_memory = true;
-		} else {
-			plan.materialization_sources.push_back(image.source);
-		}
-	}
-	for (const auto& sampler: plan.info.samplers) {
-		plan.materialization_sources.push_back(sampler.source);
-	}
 	plan.clean_flat_slots.resize(plan.srt_reads.size());
 	for (const auto& image: plan.info.images) {
 		const auto* source = Source(plan, image.source);
 		if (source == nullptr || !source->indirect_image.has_value()) {
 			continue;
 		}
+		plan.requires_specialization_memory = true;
 		MarkCleanFlatSlots(plan, Source(plan, source->indirect_image->material_source),
-		                   plan.clean_flat_slots);
-		MarkCleanFlatSlots(plan, Source(plan, source->indirect_image->heap_source),
+		                   plan.clean_flat_slots, source->indirect_image->selector_mask);
+		MarkCleanFlatSlots(plan, Source(plan, source->indirect_image->table_source),
 		                   plan.clean_flat_slots);
 	}
 	return plan;
@@ -963,11 +931,130 @@ ResourcePlan ExtractResourcePlan(const Program& program) {
 
 bool MaterializeResources(const ResourcePlan& program, const SrtRuntime& runtime,
                           ResourceSnapshot& snapshot, ResourceSpecialization& specialization) {
-	MaterializedSnapshot materialized;
-	if (!MaterializeSnapshot(program, runtime, materialized)) {
+	if (!program.resource_tracking_complete ||
+	    (program.requires_specialization_memory && runtime.read_specialization_memory == nullptr)) {
 		return false;
 	}
-	return BuildResourceSpecialization(program, std::move(materialized), snapshot, specialization);
+	const bool masked_image = std::ranges::any_of(program.info.images, [&](const auto& image) {
+		const auto* source = Source(program, image.source);
+		return source != nullptr && source->indirect_image.has_value() &&
+		       !source->indirect_image->selector_mask.IsEmpty();
+	});
+	if (masked_image &&
+	    (program.has_address_writes ||
+	     std::ranges::any_of(program.info.images, &ImageResource::written))) {
+		return false;
+	}
+	const bool capture_reads = masked_image &&
+	    std::ranges::any_of(program.info.buffers, &BufferResource::written);
+	auto& reads = program.specialization_reads;
+	ReadCapture capture {runtime, reads};
+	SrtRuntime observed = runtime;
+	if (capture_reads) {
+		reads.clear();
+		observed.userdata = &capture;
+		observed.read_specialization_memory = CaptureStrictRead;
+		if (observed.read_memory != nullptr) observed.read_memory = CaptureOrdinaryRead;
+	}
+	SrtWalker clean(program, CleanRuntime(observed));
+	SrtWalker walker(program, observed, program.clean_flat_slots, &clean);
+	const auto active = clean.FindActiveSources();
+	if (!walker.RefreshFlatBuffer(snapshot.flattened_srt)) {
+		return false;
+	}
+	snapshot.uniform_fill = {};
+	const auto& fill = program.uniform_fill;
+	const auto words = fill.fill.words;
+	std::array<uint32_t, 4> stored {};
+	bool uniform_fill = words != 0;
+	for (uint32_t i = 0; i < words && uniform_fill; ++i) {
+		uniform_fill = clean.Evaluate(fill.values[i], stored[i]) && stored[i] == stored[0];
+	}
+	if (uniform_fill) {
+		snapshot.uniform_fill = fill.fill;
+		snapshot.uniform_fill.value = stored[0];
+	}
+	const auto evaluate = [&](uint32_t source, DescriptorValue& value) {
+		if (source >= program.descriptor_sources.size()) {
+			return false;
+		}
+		if (active.empty() || active[source]) {
+			return walker.EvaluateDescriptor(source, value);
+		}
+		value = {};
+		value.dword_count = program.descriptor_sources[source].dword_count;
+		return true;
+	};
+	snapshot.buffers.resize(program.info.buffers.size());
+	for (uint32_t i = 0; i < program.info.buffers.size(); ++i) {
+		if (!evaluate(program.info.buffers[i].source, snapshot.buffers[i])) {
+			return false;
+		}
+	}
+	if (capture_reads) {
+		for (uint32_t i = 0; i < program.info.buffers.size(); ++i) {
+			const auto& buffer = program.info.buffers[i];
+			if (!buffer.written || (!active.empty() && !active[buffer.source])) continue;
+			DescriptorValue strict;
+			if (!clean.EvaluateDescriptor(buffer.source, strict) ||
+			    strict != snapshot.buffers[i]) return false;
+		}
+	}
+	snapshot.images.resize(program.info.images.size());
+	specialization.images.clear();
+	specialization.images.reserve(program.info.images.size());
+	for (const auto& image: program.info.images) {
+		specialization.images.push_back({
+		    .numeric_class = image.numeric_class,
+		    .dimension = image.dimension,
+		    .mip_count = image.mip_count,
+		    .conversion_format = image.conversion_format,
+		    .shader_swizzle = image.shader_swizzle,
+		    .indirect_root = image.indirect_root,
+		    .indirect_mapping_offset = image.indirect_mapping_offset,
+		    .indirect_search_iterations = image.indirect_search_iterations,
+		    .cube = image.cube,
+		});
+	}
+	for (uint32_t i = 0; i < program.info.images.size(); ++i) {
+		const auto& image = program.info.images[i];
+		const auto* source = Source(program, image.source);
+		if (source == nullptr) {
+			return false;
+		}
+		if (source->indirect_image.has_value()) {
+			snapshot.images[i] = {.dword_count = 8u};
+			if (!active.empty() && !active[image.source]) {
+				continue;
+			}
+			const auto& indirect = *source->indirect_image;
+			DescriptorValue material;
+			DescriptorValue table;
+			if ((indirect.material_source != UINT32_MAX &&
+			     !clean.EvaluateDescriptor(indirect.material_source, material)) ||
+			    !clean.EvaluateDescriptor(indirect.table_source, table) ||
+			    !MaterializeIndirectImage(program, indirect, material, table, i, observed, clean, snapshot,
+			                              specialization)) {
+				return false;
+			}
+		} else {
+			if (!evaluate(image.source, snapshot.images[i])) {
+				return false;
+			}
+			if (!ValidImageDescriptor(snapshot.images[i], image.r128)) {
+				snapshot.images[i].dwords.fill(0);
+			}
+		}
+	}
+	snapshot.samplers.resize(program.info.samplers.size());
+	for (uint32_t i = 0; i < program.info.samplers.size(); ++i) {
+		if (!evaluate(program.info.samplers[i].source, snapshot.samplers[i])) {
+			return false;
+		}
+	}
+	if (capture_reads && !WrittenBuffersDisjoint(program, snapshot, reads)) return false;
+	snapshot.user_data.assign(runtime.user_data.begin(), runtime.user_data.end());
+	return BuildResourceSpecialization(program, snapshot, specialization);
 }
 
 void ApplyResourceSpecialization(Program& program, const ResourceSpecialization& specialization) {
@@ -1080,29 +1167,29 @@ void ApplyResourceSpecialization(Program& program, const ResourceSpecialization&
 	for (auto* block: program.blocks) {
 		for (auto& inst: *block) {
 			if (inst.GetOpcode() == ValueOpcode::GetImageResource) {
-				inst.SetFlags(image_remap.indices.at(inst.Flags<uint32_t>()));
+				inst.SetFlags(image_remap[inst.Flags<uint32_t>()]);
 			}
 		}
 	}
 	for (auto& memory: memory_info) {
 		if (memory.kind == ResourceKind::Image && !memory.planning_only) {
-			memory.resource = image_remap.indices.at(memory.resource);
+			memory.resource = image_remap[memory.resource];
 		}
 	}
 	for (auto& buffer: buffers) {
 		if (buffer.image_alias != BufferResource::NoImageAlias) {
-			buffer.image_alias = image_remap.indices.at(buffer.image_alias);
+			buffer.image_alias = image_remap[buffer.image_alias];
 		}
 	}
 	for (auto& pair: sampled_pairs) {
-		pair.image = image_remap.indices.at(pair.image);
+		pair.image = image_remap[pair.image];
 	}
 	for (auto& image: images) {
 		if (image.indirect_root != ImageResource::NoIndirectImage) {
-			image.indirect_root = image_remap.indices.at(image.indirect_root);
+			image.indirect_root = image_remap[image.indirect_root];
 		}
 		for (auto& resource: image.indirect_resources) {
-			resource = image_remap.indices.at(resource);
+			resource = image_remap[resource];
 		}
 	}
 	image_remap.Apply(images);

src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.h

@@ -39,8 +39,7 @@ struct ResourceSpecialization {
 // its values and is independent of the translated shader CFG.
 ResourcePlan ExtractResourcePlan(const Program& program);
 
-// Resolves and specializes the immutable resource plan in one transaction. On failure both
-// destinations are unchanged.
+// Refreshes cached resources and specialization in place. A failed refresh must not be used.
 bool MaterializeResources(const ResourcePlan& program, const SrtRuntime& runtime,
                           ResourceSnapshot& snapshot, ResourceSpecialization& specialization);
 

src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp

@@ -88,6 +88,7 @@ public:
 		m_info.samplers.clear();
 		m_info.sampled_pairs.clear();
 		m_info.uses_dma = false;
+		m_shader_writes = HasShaderMemoryWrites(program);
 	}
 
 	void Run() {
@@ -168,15 +169,121 @@ private:
 		std::abort();
 	}
 
+	Value LowerDescriptorPhi(Value value, const Block* use) {
+		value           = value.Resolve();
+		const auto* phi = value.TryInstruction();
+		if (m_shader_writes || phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
+		    phi->NumArgs() != 2u || phi->NumPhiBlocks() != 2u || phi->GetType() != Type::U32 ||
+		    m_program.blocks.size() != m_program.block_info.size()) {
+			return value;
+		}
+		const auto* merge = phi->Parent();
+		const auto* branch = phi->PhiBlock(0);
+		if (merge == nullptr || branch == nullptr || phi->PhiBlock(1) == nullptr ||
+		    branch == phi->PhiBlock(1)) {
+			return value;
+		}
+		// Structurization can merge the descriptor and its use predicate in parallel Phis.
+		// Match their incoming blocks to exclude only edges that cannot reach this use.
+		if (use != nullptr && use->ImmPredecessors().size() == 1u &&
+		    use->ImmPredecessors()[0] == merge) {
+			const auto merge_it = std::ranges::find(m_program.blocks, merge);
+			const auto use_it   = std::ranges::find(m_program.blocks, use);
+			if (merge_it != m_program.blocks.end() && use_it != m_program.blocks.end()) {
+				const auto& info = m_program.block_info[merge_it - m_program.blocks.begin()];
+				const auto& term = info.terminator;
+				const auto id    = m_program.block_info[use_it - m_program.blocks.begin()].id;
+				const auto* condition = info.condition.Resolve().TryInstruction();
+				if (term.kind == CFG::TerminatorKind::ConditionalBranch &&
+				    term.true_block != term.false_block &&
+				    (id == term.true_block || id == term.false_block) && condition != nullptr &&
+				    condition->GetOpcode() == ValueOpcode::Phi && condition->GetType() == Type::U1 &&
+				    condition->Parent() == merge && condition->NumArgs() == 2u &&
+				    condition->NumPhiBlocks() == 2u) {
+					const bool taken = id == term.true_block;
+					for (uint32_t skipped = 0; skipped < 2u; skipped++) {
+						const auto excluded = condition->Arg(skipped).Resolve();
+						const auto included = condition->Arg(skipped ^ 1u).Resolve();
+						if (!excluded.IsImmediate() || excluded.GetType() != Type::U1 ||
+						    excluded.U1() == taken ||
+						    (included.IsImmediate() && included.U1() != taken)) {
+							continue;
+						}
+						for (uint32_t selected = 0; selected < 2u; selected++) {
+							if (phi->PhiBlock(selected) == condition->PhiBlock(skipped ^ 1u) &&
+							    phi->PhiBlock(selected ^ 1u) == condition->PhiBlock(skipped)) {
+								return phi->Arg(selected);
+							}
+						}
+					}
+				}
+			}
+		}
+		for (const auto& [original, selected]: m_descriptor_selections) {
+			if (original == phi) {
+				return selected;
+			}
+		}
+		if (branch->ImmSuccessors().size() != 2u) {
+			if (branch->ImmPredecessors().size() != 1u) {
+				return value;
+			}
+			branch = branch->ImmPredecessors()[0];
+		}
+		if (branch == merge || branch->ImmSuccessors().size() != 2u) {
+			return value;
+		}
+		std::array<uint32_t, 2> target_ids;
+		for (uint32_t arm = 0; arm < 2; arm++) {
+			const auto* incoming = phi->PhiBlock(arm);
+			if (incoming == merge ||
+			    (incoming != branch &&
+			     (incoming->ImmPredecessors().size() != 1u ||
+			      incoming->ImmPredecessors()[0] != branch ||
+			      incoming->ImmSuccessors().size() != 1u ||
+			      incoming->ImmSuccessors()[0] != merge))) {
+				return value;
+			}
+			const auto* target = incoming == branch ? merge : incoming;
+			const auto  it     = std::ranges::find(m_program.blocks, target);
+			if (it == m_program.blocks.end()) {
+				return value;
+			}
+			target_ids[arm] = m_program.block_info[it - m_program.blocks.begin()].id;
+		}
+		const auto branch_it = std::ranges::find(m_program.blocks, branch);
+		if (branch_it == m_program.blocks.end()) {
+			return value;
+		}
+		const auto& info = m_program.block_info[branch_it - m_program.blocks.begin()];
+		const auto& term = info.terminator;
+		if (term.kind != CFG::TerminatorKind::ConditionalBranch ||
+		    !((term.true_block == target_ids[0] && term.false_block == target_ids[1]) ||
+		      (term.false_block == target_ids[0] && term.true_block == target_ids[1])) ||
+		    !ValidateRuntimeValue(m_program, info.condition, RuntimeValueType::Integer) ||
+		    !ValidateRuntimeValue(m_program, phi->Arg(0)) ||
+		    !ValidateRuntimeValue(m_program, phi->Arg(1))) {
+			return value;
+		}
+		// Retain a host expression; replacing the GPU Phi would break SSA dominance.
+		const auto true_arg = term.true_block == target_ids[0] ? 0u : 1u;
+		auto&      selected = m_program.value_storage.emplace_back(ValueOpcode::SelectU32);
+		selected.SetArg(0, info.condition);
+		selected.SetArg(1, phi->Arg(true_arg));
+		selected.SetArg(2, phi->Arg(true_arg ^ 1u));
+		m_descriptor_selections.emplace_back(phi, Value(&selected));
+		return Value(&selected);
+	}
+
 	void MakeSource(const Inst& handle, uint32_t width, bool sampler, bool sample_adjust,
-	                DescriptorSource& descriptor, uint32_t pc) const {
+	                DescriptorSource& descriptor, uint32_t pc) {
 		if (handle.NumArgs() != width) {
 			Fail(pc, fmt::format("{} has {} descriptor dwords, expected {}",
 			                     ValueOpcodeName(handle.GetOpcode()), handle.NumArgs(), width));
 		}
 		descriptor.dword_count = width;
 		for (uint32_t i = 0; i < width; i++) {
-			descriptor.dwords[i] = handle.Arg(i).Resolve();
+			descriptor.dwords[i] = LowerDescriptorPhi(handle.Arg(i), handle.Parent());
 		}
 		if (sample_adjust) {
 			descriptor.dwords[3] = CanonicalizeSampleAdjustDword3(descriptor.dwords[3]);
@@ -206,9 +313,20 @@ private:
 		for (uint32_t candidate = 0; candidate < m_sources.size(); candidate++) {
 			const auto& current = m_sources[candidate];
 			if (current.dword_count != descriptor.dword_count ||
-			    current.indirect_image != descriptor.indirect_image) {
+			    current.indirect_image.has_value() != descriptor.indirect_image.has_value()) {
 				continue;
 			}
+			if (current.indirect_image.has_value()) {
+				const auto& a = *current.indirect_image;
+				const auto& b = *descriptor.indirect_image;
+				if (a.material_source != b.material_source || a.table_source != b.table_source ||
+				    a.selector_stride != b.selector_stride || a.selector_offset != b.selector_offset ||
+				    a.table_offset != b.table_offset ||
+				    !EquivalentValue(m_program, a.key_count, b.key_count) ||
+				    a.selector_mask.IsEmpty() != b.selector_mask.IsEmpty() ||
+				    (!a.selector_mask.IsEmpty() &&
+				     !EquivalentValue(m_program, a.selector_mask, b.selector_mask))) continue;
+			}
 			bool same = true;
 			for (uint32_t i = 0; i < descriptor.dword_count; i++) {
 				same = same && EquivalentValue(m_program, current.dwords[i], descriptor.dwords[i]);
@@ -237,16 +355,26 @@ private:
 	}
 
 	const MemoryInfo* ScalarReadMemory(const Inst& read, uint32_t& index) const {
-		if (read.GetOpcode() != ValueOpcode::ReadConstBuffer || read.NumArgs() != 2u) {
+		const bool address = read.GetOpcode() == ValueOpcode::LoadAddressU32;
+		if (!(address ? read.NumArgs() == 4u
+		              : read.GetOpcode() == ValueOpcode::ReadConstBuffer && read.NumArgs() == 2u)) {
 			return nullptr;
 		}
+		if (address) {
+			const auto high = read.Arg(2).Resolve();
+			const auto enabled = read.Arg(3).Resolve();
+			if (!high.IsImmediate() || high.GetType() != Type::U32 || high.U32() != 0u ||
+			    !enabled.IsImmediate() || enabled.GetType() != Type::U1 || !enabled.U1()) {
+				return nullptr;
+			}
+		}
 		index = read.Flags<MemoryFlags>().index;
 		if (index >= m_program.memory_info.size()) {
 			return nullptr;
 		}
 		const auto& memory = m_program.memory_info[index];
-		return memory.kind == ResourceKind::ScalarBuffer && memory.data_bits == 32u &&
-		               memory.data_dwords == 1u
+		return memory.kind == (address ? ResourceKind::ScalarAddress : ResourceKind::ScalarBuffer) &&
+		               memory.data_bits == 32u && memory.data_dwords == 1u
 		           ? &memory
 		           : nullptr;
 	}
@@ -269,18 +397,16 @@ private:
 		return true;
 	}
 
-	bool MakeRuntimeBufferSource(const Inst& handle, uint32_t pc, uint32_t& source,
-	                             DescriptorSource& descriptor) {
-		if (handle.GetOpcode() != ValueOpcode::GetBufferResource) {
+	bool MakeRuntimeTableSource(const Inst& handle, uint32_t pc,
+	                            DescriptorSource& descriptor) {
+		const auto width = handle.GetOpcode() == ValueOpcode::GetBufferResource ? 4u
+		                 : handle.GetOpcode() == ValueOpcode::GetAddressResource ? 2u : 0u;
+		if (width == 0u) {
 			return false;
 		}
-		MakeSource(handle, 4u, false, false, descriptor, pc);
+		MakeSource(handle, width, false, false, descriptor, pc);
 		uint32_t bad_dword = 0;
-		if (!ValidateSource(descriptor, bad_dword)) {
-			return false;
-		}
-		source = InternSource(descriptor);
-		return true;
+		return ValidateSource(descriptor, bad_dword);
 	}
 
 	bool MatchMaterialOffset(Value value, Value& selector, uint32_t& stride,
@@ -317,106 +443,539 @@ private:
 		       selector_inst->GetOpcode() == ValueOpcode::ReadFirstLane;
 	}
 
+	bool NonzeroOnEntry(Value value, const Block* block) const {
+		if (m_program.blocks.size() != m_program.block_info.size()) {
+			return false;
+		}
+		// Each unique predecessor must execute before this use. Stop at joins: an
+		// unrelated comparison is not a bound on FindILsb's zero-input sentinel.
+		for (size_t depth = 0; block != nullptr && depth < m_program.blocks.size(); ++depth) {
+			if (block->ImmPredecessors().size() != 1u) {
+				return false;
+			}
+			const auto* previous = block->ImmPredecessors()[0];
+			const auto current_it = std::ranges::find(m_program.blocks, block);
+			const auto previous_it = std::ranges::find(m_program.blocks, previous);
+			if (current_it == m_program.blocks.end() || previous_it == m_program.blocks.end()) {
+				return false;
+			}
+			const auto& info = m_program.block_info[previous_it - m_program.blocks.begin()];
+			const auto id = m_program.block_info[current_it - m_program.blocks.begin()].id;
+			const auto& term = info.terminator;
+			if (term.kind == CFG::TerminatorKind::ConditionalBranch &&
+			    (term.true_block == id) != (term.false_block == id)) {
+				bool positive = term.true_block == id;
+				const auto* test = info.condition.Resolve().TryInstruction();
+				while (test != nullptr && test->GetOpcode() == ValueOpcode::LogicalNot) {
+					positive = !positive;
+					test = test->Arg(0).Resolve().TryInstruction();
+				}
+				if (test != nullptr && test->NumArgs() == 2u &&
+				    ((test->GetOpcode() == ValueOpcode::INotEqual32 && positive) ||
+				     (test->GetOpcode() == ValueOpcode::IEqual32 && !positive))) {
+					for (uint32_t arg = 0; arg < 2u; ++arg) {
+						uint32_t immediate;
+						if (ImmediateU32(test->Arg(arg), immediate) && immediate == 0u &&
+						    EquivalentValue(m_program, test->Arg(arg ^ 1u), value)) {
+							return true;
+						}
+					}
+				}
+			}
+			block = previous;
+		}
+		return false;
+	}
+
+	bool MatchTableOffset(Value value, Value& key, uint32_t& offset) const {
+		offset = 0;
+		for (;;) {
+			const auto* inst = value.Resolve().TryInstruction();
+			if (inst == nullptr || inst->NumArgs() != 2u) {
+				return false;
+			}
+			uint32_t immediate;
+			if (inst->GetOpcode() == ValueOpcode::ShiftLeftLogical32 &&
+			    ImmediateU32(inst->Arg(1), immediate) && immediate == 5u) {
+				key = inst->Arg(0).Resolve();
+				return key.GetType() == Type::U32;
+			}
+			if (inst->GetOpcode() != ValueOpcode::IAdd32) {
+				return false;
+			}
+			if (ImmediateU32(inst->Arg(0), immediate)) {
+				value = inst->Arg(1);
+			} else if (ImmediateU32(inst->Arg(1), immediate)) {
+				value = inst->Arg(0);
+			} else {
+				return false;
+			}
+			// These additions are shader U32 arithmetic, before the scalar memory offset.
+			offset += immediate;
+		}
+	}
+
+	const Block* FindBlock(uint32_t id) const {
+		const auto info = std::ranges::find(m_program.block_info, id, &BlockInfo::id);
+		return info == m_program.block_info.end()
+		           ? nullptr
+		           : m_program.blocks[info - m_program.block_info.begin()];
+	}
+
+	bool CanReach(const Block* start, const Block* target, const Block* avoid) const {
+		std::vector<const Block*> pending {start};
+		std::vector<const Block*> visited;
+		while (!pending.empty()) {
+			const auto* block = pending.back();
+			pending.pop_back();
+			if (block == nullptr || block == avoid ||
+			    std::ranges::find(visited, block) != visited.end()) continue;
+			if (block == target) return true;
+			visited.push_back(block);
+			for (const auto* next: block->ImmSuccessors()) pending.push_back(next);
+		}
+		return false;
+	}
+
+	bool ConditionalEdge(const Block* from, const Block* to, Value& condition,
+	                     bool& positive) const {
+		const auto position = std::ranges::find(m_program.blocks, from);
+		const auto target = std::ranges::find(m_program.blocks, to);
+		if (position == m_program.blocks.end() || target == m_program.blocks.end()) return false;
+		const auto& info = m_program.block_info[position - m_program.blocks.begin()];
+		const auto& term = info.terminator;
+		const auto id = m_program.block_info[target - m_program.blocks.begin()].id;
+		if (term.kind != CFG::TerminatorKind::ConditionalBranch ||
+		    (term.true_block == id) == (term.false_block == id)) return false;
+		condition = info.condition;
+		positive = term.true_block == id;
+		while (const auto* inst = condition.Resolve().TryInstruction()) {
+			if (inst->GetOpcode() != ValueOpcode::LogicalNot) break;
+			condition = inst->Arg(0);
+			positive = !positive;
+		}
+		condition = condition.Resolve();
+		return true;
+	}
+
+	Value PositiveUseGuard(const Block* use) const {
+		if (use == nullptr || use->ImmPredecessors().size() != 1u) return {};
+		Value condition;
+		bool positive = false;
+		return ConditionalEdge(use->ImmPredecessors()[0], use, condition, positive) && positive
+		           ? condition : Value {};
+	}
+
+	Value SimplifyGuard(Value guard) const {
+		const auto invariant = ResolveInvariantPhi(m_program, guard);
+		return (invariant.IsEmpty() ? guard : invariant).Resolve();
+	}
+
+	bool Implies(Value guard, Value required) const {
+		guard = SimplifyGuard(guard);
+		required = required.Resolve();
+		if (EquivalentValue(m_program, guard, required)) return true;
+		const auto* inst = guard.TryInstruction();
+		return inst != nullptr && inst->GetOpcode() == ValueOpcode::LogicalAnd &&
+		       (Implies(inst->Arg(0), required) || Implies(inst->Arg(1), required));
+	}
+
+	Value EqualLocalKey(Value guard, Value key) const {
+		guard = SimplifyGuard(guard);
+		const auto* inst = guard.TryInstruction();
+		if (inst == nullptr) return {};
+		if (inst->GetOpcode() == ValueOpcode::LogicalAnd) {
+			const auto left = EqualLocalKey(inst->Arg(0), key);
+			return left.IsEmpty() ? EqualLocalKey(inst->Arg(1), key) : left;
+		}
+		if (inst->GetOpcode() != ValueOpcode::IEqual32 || inst->NumArgs() != 2u) return {};
+		if (EquivalentValue(m_program, inst->Arg(0), key)) return inst->Arg(1).Resolve();
+		if (EquivalentValue(m_program, inst->Arg(1), key)) return inst->Arg(0).Resolve();
+		return {};
+	}
+
+	struct AffineOffset {
+		Value    index;
+		uint64_t stride = 0;
+		uint64_t offset = 0;
+	};
+
+	bool MatchAffineOffset(Value value, Value guard, AffineOffset& out,
+	                       uint32_t depth = 0) const {
+		if (depth > 16u) return false;
+		value = value.Resolve();
+		uint32_t immediate = 0;
+		if (ImmediateU32(value, immediate)) {
+			out.offset = immediate;
+			return true;
+		}
+		const auto* inst = value.TryInstruction();
+		if (inst == nullptr) return false;
+		if (inst->GetOpcode() == ValueOpcode::SelectU32 && inst->NumArgs() == 3u &&
+		    Implies(guard, inst->Arg(0))) {
+			return MatchAffineOffset(inst->Arg(1), guard, out, depth + 1u);
+		}
+		if (inst->GetOpcode() == ValueOpcode::IAdd32 && inst->NumArgs() == 2u) {
+			AffineOffset left, right;
+			if (!MatchAffineOffset(inst->Arg(0), guard, left, depth + 1u) ||
+			    !MatchAffineOffset(inst->Arg(1), guard, right, depth + 1u) ||
+			    (!left.index.IsEmpty() && !right.index.IsEmpty() &&
+			     !EquivalentValue(m_program, left.index, right.index))) return false;
+			out.index = left.index.IsEmpty() ? right.index : left.index;
+			out.stride = left.stride + right.stride;
+			out.offset = left.offset + right.offset;
+			return out.stride <= UINT32_MAX && out.offset <= UINT32_MAX;
+		}
+		if (inst->GetOpcode() == ValueOpcode::ShiftLeftLogical32 && inst->NumArgs() == 2u &&
+		    ImmediateU32(inst->Arg(1), immediate) && immediate < 32u) {
+			if (!MatchAffineOffset(inst->Arg(0), guard, out, depth + 1u)) return false;
+			out.stride <<= immediate;
+			out.offset <<= immediate;
+			return out.stride <= UINT32_MAX && out.offset <= UINT32_MAX;
+		}
+		out.index = value;
+		out.stride = 1u;
+		return value.GetType() == Type::U32;
+	}
+
+	bool ImpliesNonzero(Value guard, Value value) const {
+		guard = SimplifyGuard(guard);
+		const auto* inst = guard.TryInstruction();
+		if (inst == nullptr) return false;
+		if (inst->GetOpcode() == ValueOpcode::LogicalAnd) {
+			return ImpliesNonzero(inst->Arg(0), value) ||
+			       ImpliesNonzero(inst->Arg(1), value);
+		}
+		if (inst->GetOpcode() != ValueOpcode::INotEqual32 || inst->NumArgs() != 2u)
+			return false;
+		uint32_t zero = 1u;
+		return (ImmediateU32(inst->Arg(0), zero) && zero == 0u &&
+		        EquivalentValue(m_program, inst->Arg(1), value)) ||
+		       (ImmediateU32(inst->Arg(1), zero) && zero == 0u &&
+		        EquivalentValue(m_program, inst->Arg(0), value));
+	}
+
+	bool ImpliesIndexBelow32(Value guard, Value index) const {
+		guard = SimplifyGuard(guard);
+		const auto* inst = guard.TryInstruction();
+		if (inst == nullptr) return false;
+		if (inst->GetOpcode() == ValueOpcode::LogicalAnd) {
+			return ImpliesIndexBelow32(inst->Arg(0), index) ||
+			       ImpliesIndexBelow32(inst->Arg(1), index);
+		}
+		if (inst->NumArgs() != 2u) return false;
+		uint32_t limit = 0;
+		return (inst->GetOpcode() == ValueOpcode::SLessThan32 &&
+		        EquivalentValue(m_program, inst->Arg(0), index) &&
+		        ImmediateU32(inst->Arg(1), limit) && limit == 32u) ||
+		       (inst->GetOpcode() == ValueOpcode::SGreaterThan32 &&
+		        ImmediateU32(inst->Arg(0), limit) && limit == 32u &&
+		        EquivalentValue(m_program, inst->Arg(1), index));
+	}
+
+	bool ClearsFirstSetBit(Value value, Value mask) const {
+		const auto* update = value.Resolve().TryInstruction();
+		if (update == nullptr || update->GetOpcode() != ValueOpcode::BitwiseXor32 ||
+		    update->NumArgs() != 2u) return false;
+		Value bit;
+		if (EquivalentValue(m_program, update->Arg(0), mask)) bit = update->Arg(1);
+		else if (EquivalentValue(m_program, update->Arg(1), mask)) bit = update->Arg(0);
+		else return false;
+		const auto* shift = bit.Resolve().TryInstruction();
+		uint32_t one = 0;
+		if (shift == nullptr || shift->GetOpcode() != ValueOpcode::ShiftLeftLogical32 ||
+		    shift->NumArgs() != 2u || !ImmediateU32(shift->Arg(0), one) || one != 1u)
+			return false;
+		Value position = shift->Arg(1).Resolve();
+		const auto* masked = position.TryInstruction();
+		uint32_t lane_mask = 0;
+		if (masked != nullptr && masked->GetOpcode() == ValueOpcode::BitwiseAnd32 &&
+		    masked->NumArgs() == 2u) {
+			if (ImmediateU32(masked->Arg(0), lane_mask) && lane_mask == 31u)
+				position = masked->Arg(1);
+			else if (ImmediateU32(masked->Arg(1), lane_mask) && lane_mask == 31u)
+				position = masked->Arg(0);
+		}
+		const auto* first = position.Resolve().TryInstruction();
+		return first != nullptr && first->GetOpcode() == ValueOpcode::FindILsb32 &&
+		       first->NumArgs() == 1u && EquivalentValue(m_program, first->Arg(0), mask);
+	}
+
+	Value InitialClearingMask(Value value, const Block* update_block) const {
+		const auto* phi = value.Resolve().TryInstruction();
+		if (phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
+		    phi->NumArgs() != 2u || phi->GetType() != Type::U32) return {};
+		for (uint32_t back = 0; back < 2u; ++back) {
+			if (phi->PhiBlock(back) != update_block ||
+			    !ClearsFirstSetBit(phi->Arg(back), value)) continue;
+			const auto initial = phi->Arg(back ^ 1u).Resolve();
+			return ValidateRuntimeValue(m_program, initial, RuntimeValueType::Integer)
+			           ? initial : Value {};
+		}
+		return {};
+	}
+
+	bool EdgeOn(const Block* from, const Block* to, Value predicate, bool positive) const {
+		Value condition;
+		bool taken = false;
+		return ConditionalEdge(from, to, condition, taken) && taken == positive &&
+		       EquivalentValue(m_program, condition, predicate);
+	}
+
+	Value BoundedSetBitMask(Value index, Value guard) const {
+		const auto* phi = index.Resolve().TryInstruction();
+		if (phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
+		    phi->NumArgs() != 3u || phi->GetType() != Type::U32 ||
+		    !ImpliesIndexBelow32(guard, index)) return {};
+		for (uint32_t bit_arm = 0; bit_arm < 3u; ++bit_arm) {
+			const auto bit_guard = PositiveUseGuard(phi->PhiBlock(bit_arm));
+			const auto* selected = phi->Arg(bit_arm).Resolve().TryInstruction();
+			if (selected == nullptr || selected->GetOpcode() != ValueOpcode::SelectU32 ||
+			    selected->NumArgs() != 3u ||
+			    !Implies(bit_guard, selected->Arg(0))) continue;
+			const auto* first = selected->Arg(1).Resolve().TryInstruction();
+			if (first == nullptr || first->GetOpcode() != ValueOpcode::FindILsb32 ||
+			    first->NumArgs() != 1u || !ImpliesNonzero(bit_guard, first->Arg(0)))
+				continue;
+			const auto initial_mask = InitialClearingMask(first->Arg(0), phi->PhiBlock(bit_arm));
+			if (initial_mask.IsEmpty()) continue;
+			for (uint32_t sentinel_arm = 0; sentinel_arm < 3u; ++sentinel_arm) {
+				if (sentinel_arm == bit_arm) continue;
+				const auto sentinel_guard = PositiveUseGuard(phi->PhiBlock(sentinel_arm));
+				const auto* sentinel = phi->Arg(sentinel_arm).Resolve().TryInstruction();
+				uint32_t bound = 0;
+				if (sentinel == nullptr || sentinel->GetOpcode() != ValueOpcode::SelectU32 ||
+				    sentinel->NumArgs() != 3u ||
+				    !Implies(sentinel_guard, sentinel->Arg(0)) ||
+				    !ImmediateU32(sentinel->Arg(1), bound) || bound != 32u) continue;
+				const auto loop_active = sentinel->Arg(0).Resolve();
+				const auto* active_phi = loop_active.TryInstruction();
+				if (active_phi == nullptr || active_phi->GetOpcode() != ValueOpcode::Phi ||
+				    active_phi->NumArgs() != 2u) continue;
+				bool invariant = false;
+				for (uint32_t initial = 0; initial < 2u; ++initial) {
+					invariant = Implies(guard, active_phi->Arg(initial)) &&
+					            EdgeOn(active_phi->PhiBlock(initial ^ 1u), active_phi->Parent(),
+					                   active_phi->Arg(initial ^ 1u), true);
+					if (invariant) break;
+				}
+				if (!invariant) continue;
+				const auto other_arm = 3u - bit_arm - sentinel_arm;
+				if (EdgeOn(phi->PhiBlock(other_arm), phi->Parent(), loop_active, false))
+					return initial_mask;
+			}
+		}
+		return {};
+	}
+
+	bool MatchUniformizedMaterialKey(Value key, const Inst& image,
+	                                DescriptorSource::IndirectImage& indirect,
+	                                DescriptorSource& material_source, uint32_t pc) {
+		const auto guard = PositiveUseGuard(image.Parent());
+		if (guard.IsEmpty()) return false;
+		const auto local = EqualLocalKey(guard, key);
+		const auto* selected = local.Resolve().TryInstruction();
+		if (selected == nullptr || selected->GetOpcode() != ValueOpcode::SelectU32 ||
+		    selected->NumArgs() != 3u || !Implies(guard, selected->Arg(0))) return false;
+		const auto active = selected->Arg(0).Resolve();
+		const auto* read = selected->Arg(1).Resolve().TryInstruction();
+		if (read == nullptr || read->GetOpcode() != ValueOpcode::LoadAddressU32 ||
+		    read->NumArgs() != 4u || !EquivalentValue(m_program, read->Arg(3), active))
+			return false;
+		uint32_t high = 1u;
+		if (!ImmediateU32(read->Arg(2), high) || high != 0u) return false;
+		const auto memory_index = read->Flags<MemoryFlags>().index;
+		if (memory_index >= m_program.memory_info.size()) return false;
+		const auto& memory = m_program.memory_info[memory_index];
+		if (memory.kind != ResourceKind::Global || memory.data_bits != 32u ||
+		    memory.data_dwords != 1u || !MemoryIndexBelongsTo(memory_index, *read)) return false;
+		const auto* material_handle = read->Arg(0).Resolve().TryInstruction();
+		if (material_handle == nullptr ||
+		    material_handle->GetOpcode() != ValueOpcode::GetAddressResource ||
+		    !MakeRuntimeTableSource(*material_handle, pc, material_source)) return false;
+		AffineOffset offset;
+		if (!MatchAffineOffset(read->Arg(1), active, offset) || offset.index.IsEmpty() ||
+		    offset.stride == 0u || offset.offset + memory.offset > UINT32_MAX ||
+		    offset.offset + memory.offset + 31u * offset.stride + 4u >
+		        static_cast<uint64_t>(UINT32_MAX) + 1u) return false;
+		const auto mask = BoundedSetBitMask(offset.index, active);
+		if (mask.IsEmpty()) return false;
+		indirect.material_source = InternSource(material_source);
+		indirect.selector_stride = static_cast<uint32_t>(offset.stride);
+		indirect.selector_offset = static_cast<uint32_t>(offset.offset + memory.offset);
+		indirect.key_count = Value(32u);
+		indirect.selector_mask = mask;
+		return true;
+	}
+
+	Value BoundedLoopCount(Value key, const Block* use) const {
+		const auto* phi = key.Resolve().TryInstruction();
+		if (m_shader_writes || phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
+		    phi->GetType() != Type::U32 || phi->NumArgs() != 2u ||
+		    m_program.blocks.size() != m_program.block_info.size()) {
+			return {};
+		}
+		bool induction = false;
+		for (uint32_t initial = 0; initial < 2u; ++initial) {
+			const auto zero = phi->Arg(initial).Resolve();
+			const auto* step = phi->Arg(initial ^ 1u).Resolve().TryInstruction();
+			if (!zero.IsImmediate() || zero.GetType() != Type::U32 || zero.U32() != 0u ||
+			    step == nullptr || step->GetOpcode() != ValueOpcode::IAdd32 ||
+			    step->Parent() != phi->PhiBlock(initial ^ 1u)) {
+				continue;
+			}
+			uint32_t increment = 0;
+			induction = (step->Arg(0).Resolve() == key &&
+			             ImmediateU32(step->Arg(1), increment) && increment == 1u) ||
+			            (step->Arg(1).Resolve() == key &&
+			             ImmediateU32(step->Arg(0), increment) && increment == 1u);
+			if (induction) break;
+		}
+		if (!induction) return {};
+
+		const auto contains = [&](auto&& self, Value value, const Inst* comparison) -> bool {
+			value = value.Resolve();
+			if (value.TryInstruction() == comparison) return true;
+			const auto* inst = value.TryInstruction();
+			return inst != nullptr && inst->GetOpcode() == ValueOpcode::LogicalAnd &&
+			       (self(self, inst->Arg(0), comparison) || self(self, inst->Arg(1), comparison));
+		};
+		for (const auto& use_of_key: phi->Uses()) {
+			const auto* compare = use_of_key.user;
+			if (compare->GetOpcode() != ValueOpcode::SLessThan32 || use_of_key.operand != 0u ||
+			    !ValidateRuntimeValue(m_program, compare->Arg(1))) continue;
+			for (uint32_t i = 0; i < m_program.block_info.size(); ++i) {
+				const auto& info = m_program.block_info[i];
+				const auto* negated = info.condition.Resolve().TryInstruction();
+				if (info.terminator.kind != CFG::TerminatorKind::ConditionalBranch ||
+				    compare->Parent() != m_program.blocks[i] || negated == nullptr ||
+				    negated->GetOpcode() != ValueOpcode::LogicalNot ||
+				    !contains(contains, negated->Arg(0), compare) ||
+				    use == m_program.blocks[i] ||
+				    CanReach(m_program.blocks.front(), use, m_program.blocks[i]) ||
+				    CanReach(phi->Parent(), use, m_program.blocks[i]) ||
+				    CanReach(FindBlock(info.terminator.true_block), use, phi->Parent()) ||
+				    !CanReach(FindBlock(info.terminator.false_block), use, phi->Parent())) {
+					continue;
+				}
+				return compare->Arg(1);
+			}
+		}
+		return {};
+	}
+
 	bool TryMakeIndirectImage(Inst& handle, uint32_t pc, IndirectImagePlan& plan) {
 		if (handle.GetOpcode() != ValueOpcode::GetImageResource || handle.NumArgs() != 8u) {
 			return false;
 		}
-
-		std::array<Inst*, 8> heap_reads {};
-		Inst*                heap_handle = nullptr;
-		Value                heap_offset;
-		for (uint32_t dword = 0; dword < heap_reads.size(); dword++) {
-			heap_reads[dword] = handle.Arg(dword).Resolve().TryInstruction();
-			if (heap_reads[dword] == nullptr) {
+		Inst* table_handle = nullptr;
+		Value key;
+		uint32_t table_offset = 0;
+		for (uint32_t dword = 0; dword < plan.reads.size(); ++dword) {
+			auto* read = handle.Arg(dword).Resolve().TryInstruction();
+			if (read == nullptr) {
 				return false;
 			}
-			uint32_t    memory_index = 0;
-			const auto* memory       = ScalarReadMemory(*heap_reads[dword], memory_index);
-			if (memory == nullptr || memory->offset != dword * sizeof(uint32_t) ||
-			    !MemoryIndexBelongsTo(memory_index, *heap_reads[dword])) {
+			uint32_t memory_index = 0;
+			const auto* memory = ScalarReadMemory(*read, memory_index);
+			if (memory == nullptr || memory->offset > INT32_MAX || (memory->offset & 3u) != 0u ||
+			    !MemoryIndexBelongsTo(memory_index, *read)) {
 				return false;
 			}
-			auto* current_handle = heap_reads[dword]->Arg(0).Resolve().TryInstruction();
+			auto* current_handle = read->Arg(0).Resolve().TryInstruction();
+			Value current_key;
+			uint32_t offset = 0;
 			if (current_handle == nullptr ||
-			    (heap_handle != nullptr && current_handle != heap_handle)) {
+			    current_handle->GetOpcode() != (memory->kind == ResourceKind::ScalarAddress
+			                                      ? ValueOpcode::GetAddressResource
+			                                      : ValueOpcode::GetBufferResource) ||
+			    (memory->kind == ResourceKind::ScalarAddress && read->Parent() != handle.Parent()) ||
+			    (table_handle != nullptr &&
+			     !EquivalentValue(m_program, Value(table_handle), Value(current_handle))) ||
+			    !MatchTableOffset(read->Arg(1), current_key, offset) ||
+			    memory->offset > UINT32_MAX - offset) {
 				return false;
 			}
-			heap_handle = current_handle;
+			offset += memory->offset;
 			if (dword == 0u) {
-				heap_offset = heap_reads[dword]->Arg(1).Resolve();
-			} else if (!EquivalentValue(m_program, heap_offset, heap_reads[dword]->Arg(1))) {
+				key = current_key;
+				table_offset = offset;
+			} else if (!EquivalentValue(m_program, key, current_key) ||
+			           static_cast<uint64_t>(table_offset) + dword * sizeof(uint32_t) != offset) {
 				return false;
 			}
-			plan.memory[dword] = memory_index;
-			plan.reads[dword]  = heap_reads[dword];
-		}
-
-		const auto* shift        = heap_offset.TryInstruction();
-		uint32_t    shift_amount = 0;
-		if (shift == nullptr || shift->GetOpcode() != ValueOpcode::ShiftLeftLogical32 ||
-		    shift->NumArgs() != 2u || !ImmediateU32(shift->Arg(1), shift_amount) ||
-		    shift_amount != 5u) {
-			return false;
-		}
-		auto* material_read = shift->Arg(0).Resolve().TryInstruction();
-		if (material_read == nullptr) {
-			return false;
-		}
-		uint32_t    material_memory_index = 0;
-		const auto* material_memory       = ScalarReadMemory(*material_read, material_memory_index);
-		if (material_memory == nullptr || material_memory->offset != 0u ||
-		    !MemoryIndexBelongsTo(material_memory_index, *material_read)) {
-			return false;
-		}
-		auto* material_handle = material_read->Arg(0).Resolve().TryInstruction();
-		if (material_handle == nullptr) {
-			return false;
-		}
-
-		Value    selector;
-		uint32_t selector_stride = 0;
-		uint32_t selector_offset = 0;
-		if (!MatchMaterialOffset(material_read->Arg(1), selector, selector_stride,
-		                         selector_offset)) {
-			return false;
-		}
-
-		const std::array<const Inst*, 1> material_users {shift};
-		std::array<const Inst*, 8>       heap_users {};
-		std::copy(heap_reads.begin(), heap_reads.end(), heap_users.begin());
-		const std::array<const Inst*, 1> image_users {&handle};
-		if (!UsesOnly(*material_read, material_users) || !UsesOnly(*shift, heap_users)) {
-			return false;
-		}
-		for (const auto* read: heap_reads) {
+			table_handle = current_handle;
+			const std::array<const Inst*, 1> image_users {&handle};
 			if (!UsesOnly(*read, image_users)) {
 				return false;
 			}
+			plan.memory[dword] = memory_index;
+			plan.reads[dword] = read;
 		}
 
-		DescriptorSource material_source;
-		DescriptorSource heap_source;
-		uint32_t         material_source_index = 0;
-		uint32_t         heap_source_index     = 0;
-		if (!MakeRuntimeBufferSource(*material_handle, pc, material_source_index,
-		                             material_source) ||
-		    !MakeRuntimeBufferSource(*heap_handle, pc, heap_source_index, heap_source)) {
+		DescriptorSource table_source;
+		if (!MakeRuntimeTableSource(*table_handle, pc, table_source)) {
 			return false;
 		}
-
+		DescriptorSource material_source;
+		DescriptorSource::IndirectImage indirect;
+		indirect.table_offset = table_offset;
+		if (table_source.dword_count == 2u) {
+			const auto* selector = key.Resolve().TryInstruction();
+			const bool bitscan = selector != nullptr && selector->GetOpcode() == ValueOpcode::FindILsb32 &&
+			    selector->NumArgs() == 1u && !m_shader_writes &&
+			    NonzeroOnEntry(selector->Arg(0), handle.Parent());
+			if (bitscan) {
+				indirect.key_count = Value(32u);
+			} else {
+				indirect.key_count = BoundedLoopCount(key, handle.Parent());
+			}
+			if (indirect.key_count.IsEmpty() &&
+			    !MatchUniformizedMaterialKey(key, handle, indirect, material_source, pc)) {
+				return false;
+			}
+			if ((table_offset & 3u) != 0u ||
+			    (bitscan && table_offset > UINT32_MAX - (32u * 32u - 1u))) return false;
+		} else {
+			auto* material_read = key.Resolve().TryInstruction();
+			uint32_t material_memory_index = 0;
+			const auto* memory = material_read != nullptr
+			                         ? ScalarReadMemory(*material_read, material_memory_index) : nullptr;
+			if (table_offset != 0u || memory == nullptr || memory->kind != ResourceKind::ScalarBuffer ||
+			    memory->offset != 0u || !MemoryIndexBelongsTo(material_memory_index, *material_read)) {
+				return false;
+			}
+			Value selector;
+			if (!MatchMaterialOffset(material_read->Arg(1), selector, indirect.selector_stride,
+			                         indirect.selector_offset)) {
+				return false;
+			}
+			const auto* shift = plan.reads[0]->Arg(1).Resolve().TryInstruction();
+			const std::array<const Inst*, 1> material_users {shift};
+			if (!UsesOnly(*material_read, material_users) || !UsesOnly(*shift, plan.reads)) {
+				return false;
+			}
+			const auto* material_handle = material_read->Arg(0).Resolve().TryInstruction();
+			if (material_handle == nullptr ||
+			    material_handle->GetOpcode() != ValueOpcode::GetBufferResource ||
+			    !MakeRuntimeTableSource(*material_handle, pc, material_source)) {
+				return false;
+			}
+			indirect.material_source = InternSource(material_source);
+		}
+		indirect.table_source = InternSource(table_source);
 		DescriptorSource image_source;
 		image_source.dword_count = 8u;
-		std::copy(material_source.dwords.begin(), material_source.dwords.begin() + 4u,
-		          image_source.dwords.begin());
-		std::copy(heap_source.dwords.begin(), heap_source.dwords.begin() + 4u,
-		          image_source.dwords.begin() + 4u);
-		image_source.indirect_image = DescriptorSource::IndirectImage {
-		    material_source_index, heap_source_index, selector_stride, selector_offset, 0u};
-
+		image_source.dwords.fill(Value(0u));
+		std::copy_n(material_source.dwords.begin(), material_source.dword_count,
+		            image_source.dwords.begin());
+		std::copy_n(table_source.dwords.begin(), table_source.dword_count,
+		            image_source.dwords.begin() + 4u);
+		image_source.indirect_image = indirect;
 		plan.handle = &handle;
 		plan.source = InternSource(image_source);
-		plan.key    = Value(material_read);
-		plan.roots  = image_source.dwords;
+		plan.key = key;
+		plan.roots = image_source.dwords;
 		return true;
 	}
 
@@ -455,7 +1014,7 @@ private:
 		}
 	}
 
-	void GetHandle(Value value, ValueOpcode expected, uint32_t width, uint32_t pc, Inst*& handle,
+	bool GetHandle(Value value, ValueOpcode expected, uint32_t width, uint32_t pc, Inst*& handle,
 	               uint32_t& source, bool sampler = false, bool sample_adjust = false) {
 		handle = value.Resolve().TryInstruction();
 		if (handle == nullptr || handle->GetOpcode() != expected) {
@@ -464,21 +1023,17 @@ private:
 		DescriptorSource descriptor;
 		MakeSource(*handle, width, sampler, sample_adjust, descriptor, pc);
 		uint32_t bad_dword = 0;
-		if (expected == ValueOpcode::GetImageResource) {
-			for (; bad_dword < descriptor.dword_count; bad_dword++) {
-				const auto* value = descriptor.dwords[bad_dword].Resolve().TryInstruction();
-				if (value != nullptr && value->GetOpcode() == ValueOpcode::ReadConstBuffer) {
-					Fail(pc, fmt::format("{} dword {} is not a valid runtime value",
-					                     ValueOpcodeName(expected), bad_dword));
-				}
-			}
-			bad_dword = 0;
-		}
 		if (!ValidateSource(descriptor, bad_dword)) {
+			if (expected == ValueOpcode::GetBufferResource &&
+			    std::all_of(descriptor.dwords.begin(), descriptor.dwords.begin() + width,
+			                [](Value word) { return word.Resolve().GetType() == Type::U32; })) {
+				return false;
+			}
 			Fail(pc, fmt::format("{} dword {} is not a valid runtime value",
 			                     ValueOpcodeName(expected), bad_dword));
 		}
 		source = InternSource(descriptor);
+		return true;
 	}
 
 	void ValidateAddressHandle(Value value, uint32_t pc) const {
@@ -649,7 +1204,18 @@ private:
 		uint32_t resource = 0;
 
 		if (buffer != BufferAccess::None) {
-			GetHandle(inst.Arg(0), ValueOpcode::GetBufferResource, 4, flags.pc, handle, source);
+			if (!GetHandle(inst.Arg(0), ValueOpcode::GetBufferResource, 4, flags.pc, handle,
+			               source)) {
+				if (memory.kind != ResourceKind::Buffer || memory.formatted || memory.typed ||
+				    (op != ValueOpcode::LoadBufferU32x2 && op != ValueOpcode::LoadBufferU32x4)) {
+					Fail(flags.pc,
+					     "buffer descriptor is not a valid runtime value; GPU-selected access "
+					     "requires a raw DWORD x2/x4 load");
+				}
+				m_program.memory_info[flags.index].kind = ResourceKind::IndirectBuffer;
+				m_info.uses_dma                         = true;
+				return;
+			}
 			resource = AddBuffer(source, memory, op, flags.pc);
 			if (resource == UINT32_MAX) {
 				Fail(flags.pc, "buffer resource limit exceeded");
@@ -674,6 +1240,9 @@ private:
 				return;
 			}
 			ValidateAddressHandle(inst.Arg(0), flags.pc);
+			if (address_info.access == AddressAccess::Write) {
+				m_program.has_address_writes = true;
+			}
 			m_info.uses_dma = true;
 			return;
 		}
@@ -744,12 +1313,14 @@ private:
 		}
 	}
 
-	Program&                       m_program;
-	ShaderInfo                     m_info;
-	std::vector<DescriptorSource>  m_sources;
-	std::vector<HandlePatch>       m_handle_patches;
-	std::vector<MemoryPatch>       m_memory_patches;
-	std::vector<IndirectImagePlan> m_indirect_images;
+	Program&                                   m_program;
+	ShaderInfo                                 m_info;
+	std::vector<DescriptorSource>              m_sources;
+	std::vector<HandlePatch>                   m_handle_patches;
+	std::vector<MemoryPatch>                   m_memory_patches;
+	std::vector<IndirectImagePlan>             m_indirect_images;
+	std::vector<std::pair<const Inst*, Value>> m_descriptor_selections;
+	bool                                       m_shader_writes = false;
 };
 
 } // namespace