agc: fix linux build of gs oversubscription clamps (#232)

* vulkan: remove validation layer configuration from device creation

* Remove accidental dependency gitlinks

---------

Co-authored-by: nmzik <Nmzik@mail.ru>

.github/ISSUE_TEMPLATE/kytyps5-game-emulation.yaml

@@ -1,6 +1,6 @@
-name: Game Emulation Bug Report
-description: Report KytyPS5 game boot, crash, freeze, rendering, audio, input, or gameplay behavior problems.
-title: "[GAME BUG]: "
+name: Game Emulation Status Report
+description: Submit a KytyPS5 game compatibility result, including boot, rendering, audio, input, or gameplay behavior.
+title: "[GAME STATUS]: "
 body:
   - type: markdown
     attributes:
@@ -10,6 +10,7 @@ body:
         Before submitting, please:
         - Search existing issues first.
         - Re-test with a clean game dump.
+        - Use the latest KytyPS5 build available to you.
         - Attach the complete log file instead of edited snippets.
 
   - type: markdown
@@ -44,6 +45,28 @@ body:
     validations:
       required: true
 
+  - type: dropdown
+    id: compatibility-status
+    attributes:
+      label: Compatibility status
+      description: Select the furthest point the game consistently reaches on this build.
+      options:
+        - "Doesn't boot"
+        - Logo
+        - Main menu
+        - In game
+    validations:
+      required: true
+
+  - type: markdown
+    attributes:
+      value: |
+        **Status guide**
+        - **Doesn't boot:** The game does not show its first logo or startup screen.
+        - **Logo:** The game shows a logo or startup screen but does not reach the main menu.
+        - **Main menu:** The game reaches its menus but does not enter gameplay.
+        - **In game:** The game reaches controllable gameplay, even if bugs prevent completion.
+
   - type: input
     id: regression-builds
     attributes:
@@ -54,22 +77,22 @@ body:
   - type: markdown
     attributes:
       value: |
-        ## Problem details
+        ## Test result
 
   - type: textarea
     id: what-happened
     attributes:
-      label: What happened?
-      description: Describe the failure or incorrect behavior as directly as possible.
-      placeholder: "Example: The game reaches the main menu, then freezes after selecting Continue."
+      label: Result details
+      description: Describe what worked, what failed, and any visible or audible problems.
+      placeholder: "Example: The game reaches the main menu, then freezes after selecting Continue. Audio plays correctly until the freeze."
     validations:
       required: true
 
   - type: textarea
     id: steps-to-reproduce
     attributes:
-      label: Steps to reproduce
-      description: List the shortest path that triggers the problem from a fresh KytyPS5 start.
+      label: Steps to reproduce the result
+      description: List the shortest path to the reported status or failure from a fresh KytyPS5 start.
       placeholder: |
         1. Open KytyPS5.
         2. Boot the game.
@@ -81,7 +104,7 @@ body:
     id: expected-behavior
     attributes:
       label: Expected behavior
-      description: Explain what should happen at the same point in the game.
+      description: Explain what should happen at the same point in the game. If there is no known problem, write "No known issue".
       placeholder: "Example: The save should load and gameplay should continue normally."
     validations:
       required: true

.github/workflows/build.yml

@@ -94,6 +94,7 @@ jobs:
           cmake -S . -B _Build/windows ^
             -G Ninja ^
             -DCMAKE_BUILD_TYPE=Release ^
+            -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON ^
             -DKYTY_BUILD_ORIGIN=%KYTY_BUILD_ORIGIN% ^
             -DKYTY_BUILD_REPOSITORY=%KYTY_BUILD_REPOSITORY% ^
             -DKYTY_RELEASE_TAG=${{ needs.metadata.outputs.release_tag }} ^
@@ -104,12 +105,12 @@ jobs:
       - name: Build
         shell: cmd
         run: |
-          cmake --build _Build/windows --target launcher audio_out2_port_tests virtual_memory_allocation_tests --parallel
+          cmake --build _Build/windows --target launcher audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests --parallel
 
       - name: Test
         shell: cmd
         run: |
-          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|virtual_memory_allocation)$"
+          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|ime_dialog|virtual_memory_allocation)$"
 
       - name: Install
         shell: cmd
@@ -172,6 +173,7 @@ jobs:
           cmake -S . -B _Build/macos \
             -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON \
             -DKYTY_BUILD_ORIGIN="$KYTY_BUILD_ORIGIN" \
             -DKYTY_BUILD_REPOSITORY="$KYTY_BUILD_REPOSITORY" \
             -DKYTY_RELEASE_TAG="${{ needs.metadata.outputs.release_tag }}" \
@@ -184,14 +186,14 @@ jobs:
         shell: bash
         run: |
           cmake --build _Build/macos \
-            --target launcher audio_out2_port_tests virtual_memory_allocation_tests \
+            --target launcher audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests \
             --parallel
 
       - name: Test
         shell: bash
         run: |
           ctest --test-dir _Build/macos --output-on-failure \
-            -R '^(audio_out2_port|virtual_memory_allocation)$'
+            -R '^(audio_out2_port|ime_dialog|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -299,6 +301,7 @@ jobs:
           cmake -S . -B _Build/linux \
             -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON \
             -DKYTY_BUILD_ORIGIN="$KYTY_BUILD_ORIGIN" \
             -DKYTY_BUILD_REPOSITORY="$KYTY_BUILD_REPOSITORY" \
             -DKYTY_RELEASE_TAG="${{ needs.metadata.outputs.release_tag }}" \
@@ -326,14 +329,14 @@ jobs:
         run: |
           cmake --build _Build/linux \
             --target launcher page_manager_tests memory_tracker_tests \
-              audio_out2_port_tests virtual_memory_allocation_tests \
+              audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests \
             --parallel
 
       - name: Test
         shell: bash
         run: |
           ctest --test-dir _Build/linux --output-on-failure \
-            -R '^(audio_out2_port|page_manager|memory_tracker|virtual_memory_allocation)$'
+            -R '^(audio_out2_port|ime_dialog|page_manager|memory_tracker|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -370,7 +373,13 @@ jobs:
 
   release:
     name: Release KytyPS5
-    if: github.event_name == 'push' && github.repository == 'KytyPS5/KytyPS5'
+    if: >-
+      always() &&
+      github.event_name == 'push' &&
+      github.repository == 'KytyPS5/KytyPS5' &&
+      (needs.windows.result == 'success' ||
+       needs.macos.result == 'success' ||
+       needs.linux.result == 'success')
     needs: [metadata, windows, macos, linux]
     runs-on: ubuntu-latest
     permissions:
@@ -379,34 +388,62 @@ jobs:
       RELEASE_NAME: ${{ needs.metadata.outputs.release_tag }}
 
     steps:
-      - name: Download builds
+      - name: Download Windows build
+        if: needs.windows.result == 'success'
         uses: actions/download-artifact@v4
         with:
-          path: artifacts
+          name: KytyPS5-Windows-x64
+          path: artifacts/KytyPS5-Windows-x64
 
-      - name: Package builds
+      - name: Download macOS build
+        if: needs.macos.result == 'success'
+        uses: actions/download-artifact@v4
+        with:
+          name: KytyPS5-macOS-x86_64
+          path: artifacts/KytyPS5-macOS-x86_64
+
+      - name: Download Linux build
+        if: needs.linux.result == 'success'
+        uses: actions/download-artifact@v4
+        with:
+          name: KytyPS5-Linux-x86_64
+          path: artifacts/KytyPS5-Linux-x86_64
+
+      - name: Package Windows build
+        if: needs.windows.result == 'success'
         shell: bash
         run: |
           windows_dir="artifacts/KytyPS5-Windows-x64"
-          macos_dir="artifacts/KytyPS5-macOS-x86_64"
-          linux_dir="artifacts/KytyPS5-Linux-x86_64"
           test -d "$windows_dir"
-          test -d "$macos_dir"
-          test -d "$linux_dir"
-          chmod a+x \
-            "$macos_dir/launcher" \
-            "$macos_dir/kyty_emulator" \
-            "$macos_dir/libMoltenVK.dylib" \
-            "$linux_dir/launcher" \
-            "$linux_dir/kyty_emulator"
           (
             cd "$windows_dir"
             zip -r "$GITHUB_WORKSPACE/$RELEASE_NAME-Windows-x64.zip" .
           )
+
+      - name: Package macOS build
+        if: needs.macos.result == 'success'
+        shell: bash
+        run: |
+          macos_dir="artifacts/KytyPS5-macOS-x86_64"
+          test -d "$macos_dir"
+          chmod a+x \
+            "$macos_dir/launcher" \
+            "$macos_dir/kyty_emulator" \
+            "$macos_dir/libMoltenVK.dylib"
           (
             cd "$macos_dir"
             zip -r "$GITHUB_WORKSPACE/$RELEASE_NAME-macOS-x86_64.zip" .
           )
+
+      - name: Package Linux build
+        if: needs.linux.result == 'success'
+        shell: bash
+        run: |
+          linux_dir="artifacts/KytyPS5-Linux-x86_64"
+          test -d "$linux_dir"
+          chmod a+x \
+            "$linux_dir/launcher" \
+            "$linux_dir/kyty_emulator"
           tar -C "$linux_dir" -czf \
             "$GITHUB_WORKSPACE/$RELEASE_NAME-Linux-x86_64.tar.gz" .
 
@@ -415,11 +452,16 @@ jobs:
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
-          assets=(
-            "$RELEASE_NAME-Windows-x64.zip"
-            "$RELEASE_NAME-macOS-x86_64.zip"
-            "$RELEASE_NAME-Linux-x86_64.tar.gz"
-          )
+          assets=()
+          for artifact in \
+            "$RELEASE_NAME-Windows-x64.zip" \
+            "$RELEASE_NAME-macOS-x86_64.zip" \
+            "$RELEASE_NAME-Linux-x86_64.tar.gz"; do
+            if [[ -f "$artifact" ]]; then
+              assets+=("$artifact")
+            fi
+          done
+          test "${#assets[@]}" -gt 0
           if gh release view "$RELEASE_NAME" --repo "$GITHUB_REPOSITORY" > /dev/null 2>&1; then
             gh release upload "$RELEASE_NAME" "${assets[@]}" \
               --clobber \

.gitmodules

@@ -39,3 +39,6 @@
 [submodule "3rdparty/nlohmann_json"]
 	path = 3rdparty/nlohmann_json
 	url = https://github.com/nlohmann/json.git
+[submodule "3rdparty/imgui"]
+	path = 3rdparty/imgui
+	url = https://github.com/ocornut/imgui.git

3rdparty/CMakeLists.txt

@@ -2,6 +2,32 @@ set(BUILD_SHARED_LIBS OFF CACHE BOOL "" FORCE)
 
 add_subdirectory(cpuinfo)
 
+# The Windows guest red-zone patcher needs a decoder/encoder and a small
+# trampoline assembler. Pin the same revisions used by shadPS4's implementation.
+include(FetchContent)
+
+set(XBYAK_BUILD_TESTS OFF CACHE BOOL "" FORCE)
+FetchContent_Declare(
+	xbyak
+	GIT_REPOSITORY https://github.com/herumi/xbyak.git
+	GIT_TAG 44a72f369268f7d552650891b296693e91db86bb
+	GIT_SHALLOW FALSE
+)
+
+set(ZYDIS_BUILD_EXAMPLES OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_TOOLS OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_TESTS OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_DOXYGEN OFF CACHE BOOL "" FORCE)
+FetchContent_Declare(
+	zydis
+	GIT_REPOSITORY https://github.com/zyantific/zydis.git
+	GIT_TAG 120e0e705f8e3b507dc49377ac2879979f0d545c
+	GIT_SUBMODULES_RECURSE TRUE
+	GIT_SHALLOW FALSE
+)
+
+FetchContent_MakeAvailable(xbyak zydis)
+
 set(SDL_SHARED OFF CACHE BOOL "" FORCE)
 add_subdirectory(SDL2)
 
@@ -11,6 +37,25 @@ target_include_directories(Vulkan-Headers INTERFACE
 	"${CMAKE_CURRENT_SOURCE_DIR}/Vulkan-Headers/include"
 )
 
+add_library(imgui STATIC
+	imgui/imgui.cpp
+	imgui/imgui_draw.cpp
+	imgui/imgui_tables.cpp
+	imgui/imgui_widgets.cpp
+	imgui/backends/imgui_impl_vulkan.cpp
+)
+target_include_directories(imgui PUBLIC
+	"${CMAKE_CURRENT_SOURCE_DIR}/imgui"
+	"${CMAKE_CURRENT_SOURCE_DIR}/imgui/backends"
+)
+target_compile_definitions(imgui PRIVATE IMGUI_IMPL_VULKAN_NO_PROTOTYPES)
+target_link_libraries(imgui PUBLIC Vulkan::Headers)
+if(MSVC)
+	target_compile_options(imgui PRIVATE /W0)
+else()
+	target_compile_options(imgui PRIVATE -w)
+endif()
+
 set(SPIRV_SKIP_TESTS ON CACHE BOOL "" FORCE)
 set(SPIRV_SKIP_EXECUTABLES ON CACHE BOOL "" FORCE)
 set(SKIP_SPIRV_TOOLS_INSTALL ON CACHE BOOL "" FORCE)

CMakeLists.txt

@@ -4,6 +4,12 @@ if(CMAKE_VERSION VERSION_GREATER_EQUAL 4.0)
 	set(CMAKE_POLICY_VERSION_MINIMUM 3.5 CACHE STRING "Minimum policy version for bundled dependencies")
 endif()
 
+get_property(KYTY_GENERATOR_IS_MULTI_CONFIG GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG)
+if(NOT KYTY_GENERATOR_IS_MULTI_CONFIG AND NOT CMAKE_BUILD_TYPE)
+	set(CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build" FORCE)
+	set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS Debug Release RelWithDebInfo MinSizeRel)
+endif()
+
 project(Kyty)
 
 set(KYTY_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/src")
@@ -286,7 +292,7 @@ if (KYTY_CLANG_CL)
 	set_source_files_properties(${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 xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version)
+set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL2-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
 
 # Linux system libraries required by the static FFmpeg archive.
 if(LINUX)
@@ -379,6 +385,12 @@ add_executable(bit_array_tests EXCLUDE_FROM_ALL
 )
 target_include_directories(bit_array_tests PRIVATE ${inc_headers})
 
+add_executable(ime_dialog_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/ImeDialogTests.cpp"
+	"${KYTY_SOURCE_DIR}/libs/imeDialog.cpp"
+)
+target_include_directories(ime_dialog_tests PRIVATE ${inc_headers})
+
 add_executable(memory_tracker_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/MemoryTrackerTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/pageManager.cpp"
@@ -421,13 +433,6 @@ add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 target_link_libraries(resource_tracking_tests fmt::fmt)
 target_include_directories(resource_tracking_tests PRIVATE ${inc_headers})
 
-add_executable(resource_mutex_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ResourceMutexTests.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/cache/resourceMutex.cpp"
-)
-target_link_libraries(resource_mutex_tests common)
-target_include_directories(resource_mutex_tests PRIVATE ${inc_headers})
-
 add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/AudioOut2PortTests.cpp"
 	"${KYTY_SOURCE_DIR}/libs/libAudio2.cpp"
@@ -444,6 +449,13 @@ add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 target_link_libraries(event_queue_lifetime_tests common fmt::fmt)
 target_include_directories(event_queue_lifetime_tests PRIVATE ${inc_headers})
 
+add_executable(sync_on_address_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/SyncOnAddressTests.cpp"
+	"${KYTY_SOURCE_DIR}/kernel/syncOnAddress.cpp"
+)
+target_link_libraries(sync_on_address_tests common)
+target_include_directories(sync_on_address_tests PRIVATE ${inc_headers})
+
 add_executable(image_page_table_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ImagePageTableTests.cpp"
 )
@@ -486,6 +498,7 @@ if(NOT KYTY_CLANG_CL)
 endif()
 
 if(BUILD_TESTING)
+	add_test(NAME ime_dialog COMMAND $<TARGET_FILE:ime_dialog_tests>)
 	add_test(NAME shader_cfg COMMAND $<TARGET_FILE:shader_cfg_tests>)
 	add_test(NAME scalar_provenance COMMAND $<TARGET_FILE:scalar_provenance_tests>)
 	add_test(NAME image_page_table COMMAND $<TARGET_FILE:image_page_table_tests>)
@@ -495,18 +508,22 @@ if(BUILD_TESTING)
 	add_test(NAME shader_vertex_metadata COMMAND $<TARGET_FILE:shader_vertex_metadata_tests>)
 	add_test(NAME shader_stage_runtime COMMAND $<TARGET_FILE:shader_stage_runtime_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
-	add_test(NAME resource_mutex COMMAND $<TARGET_FILE:resource_mutex_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_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 shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
 	add_test(NAME virtual_memory_allocation
 		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests>)
+	add_test(NAME guest_red_zone_patcher
+		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests> --red-zone-patcher-only)
 	add_test(NAME command_scheduler_timeline
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --scheduler-only)
 	add_test(NAME stream_buffer_ring
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --stream-buffer-only)
 	add_test(NAME gpu_command_lane
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --gpu-command-lane-only)
+	add_test(NAME pm4_context_state
+		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --context-state-only)
 	add_test(NAME gpu_tiler
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --gpu-tiler-only)
 
@@ -532,6 +549,7 @@ if(BUILD_TESTING)
 	endif()
 
 	add_custom_target(kyty_tests DEPENDS
+		ime_dialog_tests
 		shader_cfg_tests
 		scalar_provenance_tests
 		image_page_table_tests
@@ -541,8 +559,8 @@ if(BUILD_TESTING)
 		shader_vertex_metadata_tests
 		shader_stage_runtime_tests
 		resource_tracking_tests
-		resource_mutex_tests
 		event_queue_lifetime_tests
+		sync_on_address_tests
 		shader_recompiler_compute_tests
 		virtual_memory_allocation_tests
 	)

README.md

@@ -31,6 +31,9 @@ macOS support is experimental. The emulator is built for x86-64 and runs on Appl
 Rosetta 2, with Vulkan provided by MoltenVK. A small number of titles have been verified in-game
 on Apple Silicon hardware; see [Building on macOS](#building-on-macos).
 
+Community game test results are available in the
+[KytyPS5 Compatibility List](https://kytyps5.github.io/).
+
 ## Bugs and Issues
 
 The project is in an early stage, so please be mindful when opening new issues. Expect crashes,

src/common/CMakeLists.txt

@@ -24,6 +24,7 @@ target_link_libraries(common
 		Tracy::TracyClient
 	PRIVATE
 		cpuinfo
+		kyty_git_version
 		spdlog::spdlog
 )
 

src/common/assert.cpp

@@ -3,6 +3,7 @@
 #include "common/debug.h"
 #include "common/logging/log.h"
 #include "common/subsystems.h"
+#include "kytyGitVersion.h"
 
 #include <cstdio>
 #include <cstdlib>
@@ -23,7 +24,7 @@ static std::string BuildFatalReport(const char* title, std::string_view text, co
 	DebugStack stack;
 	DebugStack::Trace(&stack);
 
-	std::string report = "--- Stack Trace ---\n";
+	std::string report = "--- Build ---\n" KYTY_BUILD_LABEL "\n--- Stack Trace ---\n";
 	for (int i = PRINT_STACK_FROM; i < stack.depth; i++) {
 		report += fmt::format("[{}] {:016x}\n", i - PRINT_STACK_FROM,
 		                      static_cast<uint64_t>(stack.GetAddr(i)));

src/common/emulatorConfig.cpp

@@ -33,6 +33,10 @@ uint32_t GetScreenHeight() {
 	return g_config->screen_height;
 }
 
+bool FullscreenEnabled() {
+	return g_config->fullscreen_enabled;
+}
+
 uint32_t GetVblankFrequency() {
 	return std::clamp(g_config->vblank_frequency, 30u, 360u);
 }
@@ -97,4 +101,14 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled() {
+	return g_config->red_zone_protection_enabled;
+}
+#endif
+
+const Keymap& GetKeymap() {
+	return g_config->keymap;
+}
+
 } // namespace Config

src/common/emulatorConfig.h

@@ -5,6 +5,8 @@
 #include "common/subsystems.h"
 
 #include <filesystem>
+#include <string>
+#include <vector>
 
 namespace Config {
 
@@ -18,12 +20,15 @@ enum class ProfilerDirection { None, Network };
 
 enum class OutputDirection { Silent, Console, File };
 
+using Keymap = std::vector<std::string>;
+
 constexpr uint32_t DEFAULT_CONSOLE_LANGUAGE = 1;
 constexpr uint32_t MAX_CONSOLE_LANGUAGE     = 29;
 
 struct ConfigOptions {
 	uint32_t               screen_width                = 1280;
 	uint32_t               screen_height               = 720;
+	bool                   fullscreen_enabled          = false;
 	uint32_t               vblank_frequency            = 60;
 	uint32_t               console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = false;
@@ -40,12 +45,17 @@ struct ConfigOptions {
 	bool                   spirv_debug_printf_enabled  = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	bool red_zone_protection_enabled = false;
+#endif
+	Keymap keymap;
 };
 
 void Load(const ConfigOptions& cfg);
 
 uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
+bool     FullscreenEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();
@@ -69,6 +79,11 @@ bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled();
+#endif
+
+const Keymap& GetKeymap();
 
 } // namespace Config
 

src/common/threads.cpp

@@ -28,7 +28,8 @@
 #ifdef KYTY_WIN_CS
 #include <windows.h> // IWYU pragma: keep
 // IWYU pragma: no_include <winbase.h>
-constexpr DWORD KYTY_CS_SPIN_COUNT = 4000;
+constexpr DWORD    KYTY_CS_SPIN_COUNT          = 4000;
+constexpr uint64_t KYTY_SLEEP_SPIN_LIMIT_100NS = 500; // 50 us
 
 #ifndef CREATE_WAITABLE_TIMER_HIGH_RESOLUTION
 #define CREATE_WAITABLE_TIMER_HIGH_RESOLUTION 0x00000002
@@ -39,7 +40,10 @@ static void SleepHighResolution100ns(uint64_t units_100ns) {
 		return;
 	}
 
-	if (units_100ns <= 10000) {
+	// Keep spinning only where a kernel transition is
+	// likely to cost more than the requested delay; ordinary millisecond sleeps use the
+	// per-thread high-resolution waitable timer below.
+	if (units_100ns <= KYTY_SLEEP_SPIN_LIMIT_100NS) {
 		LARGE_INTEGER frequency {};
 		LARGE_INTEGER start {};
 		if (QueryPerformanceFrequency(&frequency) != 0 && QueryPerformanceCounter(&start) != 0 &&

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -15,6 +15,13 @@ bool TestWaitRegMemValue(uint64_t value, uint64_t ref, uint64_t mask, uint32_t f
 
 enum class Pm4ProcessResult { Complete, Blocked };
 
+enum class ContextStateOperation : uint32_t {
+	Clear     = 0,
+	Push      = 1,
+	Pop       = 2,
+	PushClear = 3,
+};
+
 class Pm4Execution {
 public:
 	[[nodiscard]] bool MadeProgress() const noexcept { return m_made_progress; }
@@ -49,24 +56,12 @@ public:
 	KYTY_CLASS_NO_COPY(CommandProcessor);
 
 	void Reset();
+	void ApplyContextStateOperation(ContextStateOperation operation);
 
-	void BufferInit();
-	void BufferFlush();
-	void BufferFlushAndWait();
-	void BufferWait();
-	void BeginReadbackTransaction() {
-		if (m_readback_active) {
-			EXIT("nested command-processor readback transaction\n");
-		}
-		m_readback_active = true;
-	}
-	void EndReadbackTransaction() {
-		if (!m_readback_active) {
-			EXIT("command-processor readback transaction is not active\n");
-		}
-		m_readback_active = false;
-	}
-
+	void            BufferInit();
+	void            BufferFlush();
+	void            BufferFlushAndWait();
+	void            BufferWait();
 	HW::Context&    GetCtx() { return m_ctx; }
 	HW::UserConfig& GetUcfg() { return m_ucfg; }
 	HW::Shader&     GetShCtx() { return m_sh_ctx; }
@@ -162,6 +157,8 @@ private:
 
 	RenderContext&   m_renderer;
 	HW::Context      m_ctx;
+	HW::Context      m_saved_ctx;
+	bool             m_context_state_pushed = false;
 	HW::UserConfig   m_ucfg;
 	HW::Shader       m_sh_ctx;
 	HW::UserSgprType m_user_data_marker                 = HW::UserSgprType::Unknown;
@@ -173,10 +170,9 @@ private:
 	// Persistent draw state: indirect draws update it for subsequent draws.
 	uint32_t m_num_instances = 1;
 
-	uint32_t m_de_count        = 0;
-	uint32_t m_ce_count        = 0;
-	bool     m_ce_complete     = false;
-	bool     m_readback_active = false;
+	uint32_t m_de_count    = 0;
+	uint32_t m_ce_count    = 0;
+	bool     m_ce_complete = false;
 
 	uint32_t m_const_ram[0x3000] = {0};
 

src/graphics/guest_gpu/command_processor/pm4Dispatch.cpp

@@ -254,6 +254,7 @@ constexpr auto MakeCustomOpcodeDispatchTable() {
 	g_cp_op_custom_func[Pm4::R_WAIT_MEM_64]    = CpOpWaitRegMem64;
 	g_cp_op_custom_func[Pm4::R_FLIP]           = CpOpFlip;
 	g_cp_op_custom_func[Pm4::R_RELEASE_MEM]    = CpOpReleaseMem;
+	g_cp_op_custom_func[Pm4::R_CONTEXT_STATE]  = CpOpContextState;
 
 	return g_cp_op_custom_func;
 }

src/graphics/guest_gpu/command_processor/pm4Dispatch.h

@@ -121,6 +121,7 @@ uint32_t HwUcSetGeIndexOffset(CommandProcessor&, uint32_t, uint32_t, const uint3
 uint32_t HwUcSetGdsOaRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 
 uint32_t CpOpNop(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
+uint32_t CpOpContextState(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpSetBase(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpClearState(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpIndexBase(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -18,7 +18,9 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <bit>
 #include <cstdio>
+#include <cstring>
 #include <vector>
 
 #define KYTY_HW_CTX_PARSER_ARGS                                                                    \
@@ -181,6 +183,10 @@ static void HwCtxIgnoreAaMaskRegister([[maybe_unused]] uint32_t cmd_offset,
 
 static void HwCtxIgnoreAlphaToMaskRegister([[maybe_unused]] uint32_t value) {}
 
+static void HwCtxIgnoreDisabledUserClipPlane(CommandProcessor& cp, uint32_t value) {
+	EXIT_NOT_IMPLEMENTED(value != 0 || cp.GetCtx().GetClipControl().user_clip_planes != 0);
+}
+
 static void HwCtxIgnoreDrawPayloadControl([[maybe_unused]] uint32_t value) {}
 
 static void HwCtxIgnoreObjprimIdControl([[maybe_unused]] uint32_t value) {}
@@ -895,23 +901,44 @@ KYTY_HW_CTX_PARSER(HwCtxSetModeControl) {
 	return 1;
 }
 
-static void HwCtxIgnorePolyOffsetRegister(uint32_t cmd_offset, uint32_t value) {
-	static std::atomic<uint32_t> log_count = 0;
+static void HwCtxSetPolyOffsetRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_t value) {
+	HW::PolyOffset offset = cp.GetCtx().GetPolyOffset();
 
-	auto count = log_count.fetch_add(1);
-	if (count < 8) {
-		LOGF_COLOR(Log::Color::Red,
-		           "\t temporary: ignoring polygon offset context register 0x%03" PRIx32
-		           " = 0x%08" PRIx32 " (depth bias not implemented)\n",
-		           cmd_offset, value);
+	switch (cmd_offset) {
+		case Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL: {
+			const auto neg_bits =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, NEG_NUM_DB_BITS);
+			offset.neg_num_db_bits = std::bit_cast<int8_t>(static_cast<uint8_t>(neg_bits));
+			offset.db_is_float_fmt =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, DB_IS_FLOAT_FMT) != 0;
+			break;
+		}
+		case Pm4::PA_SU_POLY_OFFSET_CLAMP: offset.clamp = std::bit_cast<float>(value); break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE:
+			offset.front_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET:
+			offset.front_offset = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_SCALE:
+			offset.back_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET:
+			offset.back_offset = std::bit_cast<float>(value);
+			break;
+		default:
+			EXIT("unknown polygon offset context register 0x%03" PRIx32 " = 0x%08" PRIx32 "\n",
+			     cmd_offset, value);
 	}
+
+	cp.GetCtx().SetPolyOffset(offset);
 }
 
 KYTY_HW_CTX_PARSER(HwCtxSetPolyOffsetRegisters) {
 	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
 
 	for (uint32_t i = 0; i < num_values; i++) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset + i, buffer[i]);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset + i, buffer[i]);
 	}
 
 	return num_values;
@@ -1312,26 +1339,36 @@ static void HwShSetCsRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_VGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_VGPRS_MASK;
 			cs_regs.sgprs =
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_SGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_SGPRS_MASK;
-			cs_regs.bulky =
-			    (value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK;
-			cs_regs.wave_size = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
-			                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
-			                        ? 32u
-			                        : 64u;
+			cs_regs.float_mode    = (value >> Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_SHIFT) &
+			                        Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_MASK;
+			cs_regs.dx10_clamp    = ((value >> Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_MASK) != 0u;
+			cs_regs.ieee_mode     = ((value >> Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_MASK) != 0u;
+			cs_regs.fp16_overflow = ((value >> Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_MASK) != 0u;
+			cs_regs.bulky         = ((value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK) != 0u;
+			cs_regs.threadgroup_configuration = ((value >> Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT) &
+			                                     Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_MASK) != 0u;
+			cs_regs.wave_size                 = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
+			                                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
+			                                        ? 32u
+			                                        : 64u;
 			break;
 		case Pm4::COMPUTE_PGM_RSRC2:
-			cs_regs.scratch_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK;
+			cs_regs.scratch_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK) != 0u;
 			cs_regs.user_sgpr      = (value >> Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_MASK;
-			cs_regs.tgid_x_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK;
-			cs_regs.tgid_y_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK;
-			cs_regs.tgid_z_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK;
-			cs_regs.tg_size_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK;
+			cs_regs.tgid_x_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK) != 0u;
+			cs_regs.tgid_y_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK) != 0u;
+			cs_regs.tgid_z_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK) != 0u;
+			cs_regs.tg_size_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK) != 0u;
 			cs_regs.tidig_comp_cnt = (value >> Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_MASK;
 			cs_regs.lds_size       = (value >> Pm4::COMPUTE_PGM_RSRC2_LDS_SIZE_SHIFT) &
@@ -2102,11 +2139,24 @@ KYTY_CP_OP_PARSER(CpOpClearState) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0001200);
 	EXIT_NOT_IMPLEMENTED((buffer[0] & ~0xfu) != 0);
 
-	cp.Reset();
+	cp.ApplyContextStateOperation(ContextStateOperation::Clear);
 
 	return 1;
 }
 
+KYTY_CP_OP_PARSER(CpOpContextState) {
+	KYTY_PROFILER_FUNCTION();
+
+	const auto packet_size_dw = KYTY_PM4_LEN(cmd_id);
+	EXIT_NOT_IMPLEMENTED(KYTY_PM4_R(cmd_id) != Pm4::R_CONTEXT_STATE);
+	EXIT_NOT_IMPLEMENTED(packet_size_dw != 3u && packet_size_dw != 5u);
+	EXIT_NOT_IMPLEMENTED(buffer[0] > static_cast<uint32_t>(ContextStateOperation::PushClear));
+
+	cp.ApplyContextStateOperation(static_cast<ContextStateOperation>(buffer[0]));
+
+	return packet_size_dw - 1u;
+}
+
 KYTY_CP_OP_PARSER(CpOpDumpConstRam) {
 	KYTY_PROFILER_FUNCTION();
 
@@ -2636,14 +2686,21 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 	const bool gl2_writeback = ((gcr_cntl & GcrGl2Writeback) != 0);
 
 	auto trigger_interrupt = [&]() {
+		bool queued = false;
 		switch (interrupt_selector) {
 			case 0x00:
 			case 0x03: break;
 			case 0x01:
 			case 0x02:
-			case 0x04: cp.TriggerEopEventAtEndOfPipe(interrupt_context_id); break;
+			case 0x04:
+				cp.TriggerEopEventAtEndOfPipe(interrupt_context_id);
+				queued = true;
+				break;
 			default: EXIT("unknown release_mem interrupt selector\n");
 		}
+		if (queued) {
+			cp.BufferFlush();
+		}
 	};
 
 	if (data_sel == 0 || interrupt_selector == 4) {
@@ -2651,10 +2708,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2665,10 +2718,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2682,17 +2731,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 	cache_policy      = 0;
 
 	if (data_sel == 1) {
-		static std::atomic<uint32_t> log_count {0};
-		const auto                   dst_addr = reinterpret_cast<uint64_t>(dst_gpu_addr);
-		if ((dst_addr & 0xffffffffff000000ull) == 0x0000001021000000ull || log_count.load() < 64) {
-			if (log_count.fetch_add(1) < 256) {
-				LOGF(
-				    "\t runtime release_mem32: dst=0x%016" PRIx64 ", value=0x%08" PRIx32
-				    ", action=0x%02" PRIx32 ", gcr=0x%04" PRIx32 ", cache_action=0x%08" PRIx32 "\n",
-				    dst_addr, static_cast<uint32_t>(value), eop_event_type, gcr_cntl, cache_action);
-			}
-		}
-
 		eop_event_type    = 0x2f;
 		event_index       = 6;
 		auto event_source = 2u;
@@ -2707,7 +2745,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 
 	if (data_sel == 5) {
 		if (gl2_writeback) {
-			cp.SynchronizeGpu();
 			cache_action = 0;
 		}
 
@@ -2718,7 +2755,9 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 		cp.WriteAtEndOfPipe32(cache_policy, event_write_dest, eop_event_type, cache_action,
 		                      event_index, event_source, dst_gpu_addr, static_cast<uint32_t>(value),
 		                      interrupt_selector, interrupt_context_id);
-		cp.BufferFlush();
+		if (interrupt_selector == 0x01) {
+			cp.BufferFlush();
+		}
 
 		return 7;
 	}
@@ -3308,6 +3347,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 		};
 	}
 
+	for (auto cmd_offset = Pm4::PA_CL_UCP_0_X; cmd_offset <= Pm4::PA_CL_UCP_5_W; cmd_offset++) {
+		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
+			HwCtxIgnoreDisabledUserClipPlane(cp, value);
+		};
+	}
+
 	for (auto cmd_offset = Pm4::PA_SC_VPORT_SCISSOR_0_TL;
 	     cmd_offset <= Pm4::PA_SC_VPORT_SCISSOR_15_TL; cmd_offset += 2) {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3682,22 +3727,22 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_CLAMP] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_Z_INFO] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -4220,29 +4265,28 @@ void GraphicsInitJmpTablesShIndirect() {
 	};
 
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC1_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HW::VsShaderResource1 r1;
-		r1.vgprs                = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, VGPRS);
-		r1.sgprs                = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, SGPRS);
-		r1.priority             = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, PRIORITY);
-		r1.float_mode           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FLOAT_MODE);
-		r1.dx10_clamp           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, DX10_CLAMP) != 0;
-		r1.ieee_mode            = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, IEEE_MODE) != 0;
-		r1.vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, VGPR_COMP_CNT);
-		r1.cu_group_enable = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, CU_GROUP_ENABLE) != 0;
+		HW::HsShaderResource1 r1;
+		r1.vgprs      = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, VGPRS);
+		r1.priority   = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, PRIORITY);
+		r1.float_mode = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FLOAT_MODE);
+		r1.dx10_clamp = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, DX10_CLAMP) != 0;
+		r1.debug_mode = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, DEBUG_MODE) != 0;
+		r1.ieee_mode  = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, IEEE_MODE) != 0;
 		r1.require_forward_progress =
-		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FWD_PROGRESS) != 0;
-		r1.fp16_overflow = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FP16_OVFL) != 0;
+		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FWD_PROGRESS) != 0;
+		r1.threadgroup_configuration = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, WGP_MODE) != 0;
+		r1.ls_vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, LS_VGPR_COMP_CNT);
+		r1.fp16_overflow           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FP16_OVFL) != 0;
 		cp.GetShCtx().SetHsShaderResource1(r1);
 	};
 
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC2_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HW::VsShaderResource2 r2;
-		r2.scratch_en        = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SCRATCH_EN) != 0;
-		r2.user_sgpr         = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, USER_SGPR) +
-		                       (KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, USER_SGPR_MSB) << 5u);
-		r2.offchip_lds       = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, OC_LDS_EN) != 0;
-		r2.streamout_enabled = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SO_EN) != 0;
-		r2.shared_vgprs      = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SHARED_VGPR_CNT);
+		HW::HsShaderResource2 r2;
+		r2.scratch_en   = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, SCRATCH_EN) != 0;
+		r2.user_sgpr    = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, USER_SGPR) +
+		                  (KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, USER_SGPR_MSB) << 5u);
+		r2.lds_size     = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, LDS_SIZE);
+		r2.shared_vgprs = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, SHARED_VGPR_CNT);
 		cp.GetShCtx().SetHsShaderResource2(r2);
 	};
 
@@ -4330,7 +4374,7 @@ void GraphicsInitJmpTablesShIndirect() {
 		r1.cu_group_enable = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, CU_GROUP_ENABLE) != 0;
 		r1.require_forward_progress =
 		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, FWD_PROGRESS) != 0;
-		r1.lds_configuration       = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, WGP_MODE) != 0;
+		r1.threadgroup_configuration = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, WGP_MODE) != 0;
 		r1.gs_vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, GS_VGPR_COMP_CNT);
 		r1.fp16_overflow           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, FP16_OVFL) != 0;
 		cp.GetShCtx().SetGsShaderResource1(r1);
@@ -4390,7 +4434,7 @@ void GraphicsInitJmpTablesShIndirect() {
 		r2.wave_cnt_en    = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, WAVE_CNT_EN);
 		r2.extra_lds_size = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, EXTRA_LDS_SIZE);
 		r2.raster_ordered_shading =
-		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, LOAD_INTRAWAVE_COLLISION);
+		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, RASTER_ORDERED_SHADING);
 		r2.shared_vgprs = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, SHARED_VGPR_CNT);
 		cp.GetShCtx().SetPsShaderResource2(r2);
 	};
@@ -4407,6 +4451,11 @@ void GraphicsInitJmpTablesUcIndirect() {
 		r.vertex_group_size    = KYTY_PM4_GET(value, GE_CNTL, VERT_GRP_SIZE);
 		cp.GetUcfg().SetGeControl(r);
 	};
+	g_hw_uc_indirect_func[Pm4::GE_PC_ALLOC] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+		(void)cp;
+		(void)cmd_offset;
+		(void)value;
+	};
 
 	g_hw_uc_indirect_func[Pm4::GE_USER_VGPR_EN] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeUserVgprEn r;

src/graphics/guest_gpu/gpu_defs.h

@@ -60,24 +60,34 @@ enum class GsOutputPrimitiveType : uint32_t {
 };
 
 enum class ChannelLayout : uint32_t {
-	kInvalid     = 0,
-	k8           = 1,
-	k16          = 2,
-	k8_8         = 3,
-	k32          = 4,
-	k16_16       = 5,
-	k11_11_10    = 6,
-	k10_10_10_2  = 9,
-	k8_8_8_8     = 10,
-	k32_32       = 11,
-	k16_16_16_16 = 12,
-	k32_32_32_32 = 14,
-	k5_6_5       = 16,
-	k5_5_5_1     = 17,
-	k4_4_4_4     = 19,
-	kBc1         = 35,
-	kBc2         = 36,
-	kBc3         = 37,
+	kInvalid         = 0,
+	k8               = 1,
+	k16              = 2,
+	k8_8             = 3,
+	k32              = 4,
+	k16_16           = 5,
+	k11_11_10        = 6,
+	k10_11_11        = 7,
+	k2_10_10_10      = 8,
+	k10_10_10_2      = 9,
+	k8_8_8_8         = 10,
+	k32_32           = 11,
+	k16_16_16_16     = 12,
+	k32_32_32        = 13,
+	k32_32_32_32     = 14,
+	k5_6_5           = 16,
+	k5_5_5_1         = 17,
+	k1_5_5_5         = 18,
+	k4_4_4_4         = 19,
+	k10_10_10_2Float = 31,
+	k9_9_9_5         = 34,
+	kBc1             = 35,
+	kBc2             = 36,
+	kBc3             = 37,
+	kBc4             = 38,
+	kBc5             = 39,
+	kBc6             = 40,
+	kBc7             = 41,
 };
 
 enum class ChannelType : uint32_t {

src/graphics/guest_gpu/graphicsRun.cpp

@@ -36,6 +36,7 @@ static thread_local CommandProcessor* g_current_processor = nullptr;
 static thread_local Pm4Execution*     g_current_execution = nullptr;
 static thread_local bool              g_gpu_mutex_owned   = false;
 static thread_local bool              g_gpu_thread        = false;
+static thread_local GpuState*         g_gpu_state         = nullptr;
 
 class GpuMutexLock final {
 public:
@@ -98,11 +99,10 @@ public:
 	void SubmitFlipPreparation(uint64_t request_id);
 	void Done();
 	void Shutdown();
-	[[nodiscard]] bool IsStopping();
-	void               SendCommand(Common::UniqueFunction<void>&& command);
-	void               SendCommandSync(Common::UniqueFunction<void>&& command);
-	void SendCommandSyncWithProcessor(Common::UniqueFunction<void, CommandProcessor&>&& command);
-	int  GetFrameNum();
+	[[nodiscard]] bool        IsStopping();
+	void                      SendCommand(Common::UniqueFunction<void>&& command);
+	void                      SendCommandSync(Common::UniqueFunction<void>&& command);
+	int                       GetFrameNum();
 	[[nodiscard]] static bool IsGpuThread() noexcept { return g_gpu_thread; }
 
 private:
@@ -124,9 +124,9 @@ private:
 		uint64_t       flip_request_id               = 0;
 	};
 
-	void              WaitLocked();
 	void              Enqueue(Submission submission);
 	void              WaitForIdle();
+	void              ProcessCommands();
 	bool              Process(Submission& submission);
 	static void       ThreadRun(void* data);
 	CommandProcessor& GetProcessor(uint32_t queue_id);
@@ -139,6 +139,7 @@ private:
 	Common::CondVar                                m_idle;
 	std::array<std::deque<Submission>, QueueCount> m_queues;
 	std::deque<Common::UniqueFunction<void>>       m_commands;
+	std::atomic_uint32_t                           m_pending_commands {0};
 	uint32_t                                       m_next_queue        = 0;
 	uint32_t                                       m_submission_count  = 0;
 	bool                                           m_processing        = false;
@@ -153,6 +154,8 @@ private:
 	uint64_t        m_submit_id = 0;
 	std::atomic_int m_done_num  = 0;
 	std::jthread    m_thread;
+
+	friend class CommandProcessor;
 };
 
 static bool GraphicsRunDebugDumpEnabled() {
@@ -195,9 +198,25 @@ void GpuState::SendCommand(Common::UniqueFunction<void>&& command) {
 	Common::LockGuard lock(m_queue_mutex);
 	EXIT_IF(!m_accepting);
 	m_commands.push_back(std::move(command));
+	m_pending_commands.fetch_add(1, std::memory_order_release);
 	m_work_available.Signal();
 }
 
+void GpuState::ProcessCommands() {
+	EXIT_IF(!IsGpuThread());
+	while (m_pending_commands.load(std::memory_order_acquire) != 0) {
+		Common::UniqueFunction<void> command;
+		{
+			Common::LockGuard lock(m_queue_mutex);
+			EXIT_IF(m_commands.empty());
+			command = std::move(m_commands.front());
+			m_commands.pop_front();
+			EXIT_IF(m_pending_commands.fetch_sub(1, std::memory_order_acq_rel) == 0);
+		}
+		command();
+	}
+}
+
 void GpuState::SendCommandSync(Common::UniqueFunction<void>&& command) {
 	EXIT_IF(!command);
 	if (IsGpuThread()) {
@@ -212,17 +231,6 @@ void GpuState::SendCommandSync(Common::UniqueFunction<void>&& command) {
 	done.acquire();
 }
 
-void GpuState::SendCommandSyncWithProcessor(
-    Common::UniqueFunction<void, CommandProcessor&>&& command) {
-	EXIT_IF(!command);
-	SendCommandSync([this, operation = std::move(command)]() mutable {
-		EXIT_IF(g_current_processor != nullptr);
-		g_current_processor = m_gfx_cp.get();
-		operation(*m_gfx_cp);
-		g_current_processor = nullptr;
-	});
-}
-
 void GpuState::Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t* cmd_const_buffer,
                       uint32_t num_const_dw, bool trigger_agc_interrupt_on_done) {
 	GpuMutexLock lock(m_submission_mutex);
@@ -267,10 +275,8 @@ void GpuState::SubmitFlipPreparation(uint64_t request_id) {
 
 void GpuState::Done() {
 	GpuMutexLock lock(m_submission_mutex);
-	if (IsGpuThread()) {
-		m_gfx_cp->BufferWait();
-	} else {
-		WaitLocked();
+	if (!IsGpuThread()) {
+		WaitForIdle();
 	}
 	m_graphics_done = true;
 	m_done_num++;
@@ -280,11 +286,6 @@ int GpuState::GetFrameNum() {
 	return m_done_num;
 }
 
-void GpuState::WaitLocked() {
-	WaitForIdle();
-	SendCommandSync([this] { m_gfx_cp->BufferWait(); });
-}
-
 CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
 	EXIT_IF(queue_id >= QueueCount);
 	if (queue_id == 0) {
@@ -298,11 +299,11 @@ CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
 }
 
 void CommandProcessor::Reset() {
-	BufferWait();
-
 	m_sh_ctx.Reset();
 	m_ucfg.Reset();
 	m_ctx.Reset();
+	m_saved_ctx.Reset();
+	m_context_state_pushed             = false;
 	m_index_type_and_size              = 0;
 	m_index_buffer_size                = 0;
 	m_user_data_marker                 = HW::UserSgprType::Unknown;
@@ -312,6 +313,30 @@ void CommandProcessor::Reset() {
 	std::memset(m_const_ram, 0, sizeof(m_const_ram));
 }
 
+void CommandProcessor::ApplyContextStateOperation(ContextStateOperation operation) {
+	switch (operation) {
+		case ContextStateOperation::Clear: m_ctx.Reset(); break;
+		case ContextStateOperation::Push:
+			EXIT_IF(m_context_state_pushed);
+			m_saved_ctx            = m_ctx;
+			m_context_state_pushed = true;
+			break;
+		case ContextStateOperation::Pop:
+			EXIT_IF(!m_context_state_pushed);
+			m_ctx                  = m_saved_ctx;
+			m_saved_ctx            = {};
+			m_context_state_pushed = false;
+			break;
+		case ContextStateOperation::PushClear:
+			EXIT_IF(m_context_state_pushed);
+			m_saved_ctx            = m_ctx;
+			m_context_state_pushed = true;
+			m_ctx.Reset();
+			break;
+		default: EXIT("unknown context state operation: %u\n", static_cast<uint32_t>(operation));
+	}
+}
+
 void CommandProcessor::BufferInit() {
 	GetScheduler().Begin(m_ctx, m_ucfg, m_sh_ctx);
 }
@@ -350,7 +375,6 @@ void CommandProcessor::WaitDeDiff(uint32_t diff) {
 }
 
 void CommandProcessor::IncrementDe() {
-	BufferWait();
 	m_de_count++;
 }
 
@@ -449,7 +473,6 @@ void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cach
 	if (num_bytes == 0) {
 		return;
 	}
-	EXIT_NOT_IMPLEMENTED((num_bytes & 3u) != 0);
 	EXIT_NOT_IMPLEMENTED(dst_cache_policy > 3);
 	EXIT_NOT_IMPLEMENTED(src_cache_policy > 3);
 	EXIT_NOT_IMPLEMENTED(wait_for_previous > 1);
@@ -509,6 +532,7 @@ void GpuState::ThreadRun(void* data) {
 	EXIT_IF(gpu == nullptr);
 	KYTY_PROFILER_THREAD("Thread_Gpu");
 	g_gpu_thread = true;
+	g_gpu_state  = gpu;
 
 	for (;;) {
 		Submission                   submission;
@@ -529,6 +553,7 @@ void GpuState::ThreadRun(void* data) {
 			} else if (!gpu->m_commands.empty()) {
 				command = std::move(gpu->m_commands.front());
 				gpu->m_commands.pop_front();
+				EXIT_IF(gpu->m_pending_commands.fetch_sub(1, std::memory_order_acq_rel) == 0);
 				gpu->m_processing = true;
 			} else {
 				int selected_queue = -1;
@@ -560,6 +585,7 @@ void GpuState::ThreadRun(void* data) {
 		}
 		if (should_stop) {
 			gpu->m_gfx_cp->BufferWait();
+			g_gpu_state  = nullptr;
 			g_gpu_thread = false;
 			return;
 		}
@@ -741,6 +767,9 @@ void CommandProcessor::SuspendPm4() {
 
 void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
 	while (execution.m_buffer_stack.size() > stop_depth) {
+		if (g_gpu_state != nullptr) {
+			g_gpu_state->ProcessCommands();
+		}
 		const auto buffer_index = execution.m_buffer_stack.size() - 1;
 		auto&      cursor       = execution.m_buffer_stack[buffer_index];
 		if (cursor.deferred_advance_dw != 0) {
@@ -1135,9 +1164,9 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_group_y,
                                       uint32_t thread_group_z, uint32_t mode) {
 	uint32_t frame_num = 0;
-	uint32_t local_x   = 1;
-	uint32_t local_y   = 1;
-	uint32_t local_z   = 1;
+	// uint32_t local_x   = 1;
+	// uint32_t local_y   = 1;
+	// uint32_t local_z   = 1;
 
 	{
 		CheckBuffer();
@@ -1161,9 +1190,9 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		}
 
 		const auto& cs = m_sh_ctx.GetCs().cs_regs;
-		local_x        = std::max(cs.num_thread_x, 1u);
-		local_y        = std::max(cs.num_thread_y, 1u);
-		local_z        = std::max(cs.num_thread_z, 1u);
+		// local_x        = std::max(cs.num_thread_x, 1u);
+		// local_y        = std::max(cs.num_thread_y, 1u);
+		// local_z        = std::max(cs.num_thread_z, 1u);
 		if (cs.wave_size == 64u) {
 			static std::atomic_bool logged_wave64_shader {false};
 			if (!logged_wave64_shader.exchange(true, std::memory_order_relaxed)) {
@@ -1179,27 +1208,27 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		                                              thread_group_y, thread_group_z, mode);
 	}
 
-	constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
+	/*constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
 	auto               group_count = [](uint32_t threads, uint32_t group_size) {
-		return (threads == 0
-		            ? 0u
-		            : (threads + std::max(group_size, 1u) - 1u) / std::max(group_size, 1u));
+	    return (threads == 0
+	                ? 0u
+	                : (threads + std::max(group_size, 1u) - 1u) / std::max(group_size, 1u));
 	};
 
 	auto groups_x = thread_group_x;
 	auto groups_y = thread_group_y;
 	auto groups_z = thread_group_z;
 	if ((mode & DispatchInitiatorUseThreadDimensions) != 0) {
-		groups_x = group_count(thread_group_x, local_x);
-		groups_y = group_count(thread_group_y, local_y);
-		groups_z = group_count(thread_group_z, local_z);
+	    groups_x = group_count(thread_group_x, local_x);
+	    groups_y = group_count(thread_group_y, local_y);
+	    groups_z = group_count(thread_group_z, local_z);
 	}
 
 	const uint64_t invocations =
 	    static_cast<uint64_t>(groups_x) * groups_y * groups_z * local_x * local_y * local_z;
 	if (invocations != 0) {
-		BufferFlushAndWait();
-	}
+	    BufferFlushAndWait();
+	}*/
 }
 
 void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
@@ -1311,6 +1340,9 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					              value >> 16u);
 					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
 					                            value >> 16u);
+					if (with_interrupt) {
+						m_renderer.TriggerEopEvent(interrupt_context_id);
+					}
 					return;
 				}
 			} else if (eop_event_type == 0x04 && cache_action == 0x00 && event_index == 0x05) {
@@ -1513,7 +1545,6 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
 				     event_index);
 			}
 			EmitGlobalBarrier();
-			SynchronizeGpu();
 			break;
 		// DbDataWritebackInvalidate, DbMetadataWritebackInvalidate, CbMetadataWritebackInvalidate.
 		case 0x0000002a:
@@ -1650,10 +1681,6 @@ void Gpu::SendCommandSync(Common::UniqueFunction<void>&& command) {
 	m_state->SendCommandSync(std::move(command));
 }
 
-void Gpu::SendCommandSyncWithProcessor(Common::UniqueFunction<void, CommandProcessor&>&& command) {
-	m_state->SendCommandSyncWithProcessor(std::move(command));
-}
-
 void Gpu::Submit(uint32_t* draw_commands, uint32_t draw_size_dw, uint32_t* constant_commands,
                  uint32_t constant_size_dw, bool trigger_agc_interrupt_on_done) {
 	EXIT_IF(draw_commands == nullptr || draw_size_dw == 0);
@@ -1679,12 +1706,8 @@ int Gpu::GetFrameNum() const {
 	return m_state->GetFrameNum();
 }
 
-bool Gpu::IsCommandProcessorThread() noexcept {
-	return g_current_processor != nullptr;
-}
-
-CommandProcessor* Gpu::CurrentCommandProcessor() noexcept {
-	return g_current_processor;
+bool Gpu::IsGpuThread() noexcept {
+	return GpuState::IsGpuThread();
 }
 
 } // namespace Libs::Graphics

src/graphics/guest_gpu/graphicsRun.h

@@ -9,7 +9,6 @@
 
 namespace Libs::Graphics {
 
-class CommandProcessor;
 class GpuState;
 class RenderContext;
 
@@ -23,7 +22,6 @@ public:
 	[[nodiscard]] bool IsStopping();
 	void               SendCommand(Common::UniqueFunction<void>&& command);
 	void               SendCommandSync(Common::UniqueFunction<void>&& command);
-	void SendCommandSyncWithProcessor(Common::UniqueFunction<void, CommandProcessor&>&& command);
 
 	void Submit(uint32_t* draw_commands, uint32_t draw_size_dw, uint32_t* constant_commands,
 	            uint32_t constant_size_dw, bool trigger_agc_interrupt_on_done = false);
@@ -33,8 +31,7 @@ public:
 	void Done();
 	[[nodiscard]] int GetFrameNum() const;
 
-	[[nodiscard]] static bool              IsCommandProcessorThread() noexcept;
-	[[nodiscard]] static CommandProcessor* CurrentCommandProcessor() noexcept;
+	[[nodiscard]] static bool IsGpuThread() noexcept;
 
 private:
 	std::unique_ptr<GpuState> m_state;

src/graphics/guest_gpu/hardwareContext.h

@@ -348,6 +348,16 @@ struct ModeControl {
 	bool    persp_corr_dis           = false;
 };
 
+struct PolyOffset {
+	int8_t neg_num_db_bits = -23;
+	bool   db_is_float_fmt = true;
+	float  clamp           = 0.0f;
+	float  front_scale     = 0.0f;
+	float  front_offset    = 0.0f;
+	float  back_scale      = 0.0f;
+	float  back_offset     = 0.0f;
+};
+
 struct BlendControl {
 	uint8_t color_srcblend       = 1;
 	uint8_t color_comb_fcn       = 0;
@@ -482,6 +492,26 @@ struct VsShaderResource2 {
 	uint8_t shared_vgprs      = 0;
 };
 
+struct HsShaderResource1 {
+	uint8_t vgprs                     = 0;
+	uint8_t priority                  = 0;
+	uint8_t float_mode                = 0;
+	bool    dx10_clamp                = false;
+	bool    debug_mode                = false;
+	bool    ieee_mode                 = false;
+	bool    require_forward_progress  = false;
+	bool    threadgroup_configuration = false;
+	uint8_t ls_vgpr_component_count   = 0;
+	bool    fp16_overflow             = false;
+};
+
+struct HsShaderResource2 {
+	bool     scratch_en   = false;
+	uint8_t  user_sgpr    = 0;
+	uint16_t lds_size     = 0;
+	uint8_t  shared_vgprs = 0;
+};
+
 struct PsShaderResource1 {
 	uint8_t vgprs                    = 0;
 	uint8_t sgprs                    = 0;
@@ -500,7 +530,7 @@ struct PsShaderResource2 {
 	uint8_t user_sgpr              = 0;
 	bool    wave_cnt_en            = false;
 	uint8_t extra_lds_size         = 0;
-	bool    raster_ordered_shading = false;
+	uint8_t raster_ordered_shading = 0;
 	uint8_t shared_vgprs           = 0;
 };
 
@@ -513,22 +543,27 @@ struct PsStageRegisters {
 
 struct CsStageRegisters {
 
-	uint64_t data_addr      = 0;
-	uint32_t num_thread_x   = 0;
-	uint32_t num_thread_y   = 0;
-	uint32_t num_thread_z   = 0;
-	uint8_t  vgprs          = 0;
-	uint8_t  sgprs          = 0;
-	uint8_t  bulky          = 0;
-	uint8_t  wave_size      = 64;
-	uint8_t  scratch_en     = 0;
-	uint8_t  user_sgpr      = 0;
-	uint8_t  tgid_x_en      = 0;
-	uint8_t  tgid_y_en      = 0;
-	uint8_t  tgid_z_en      = 0;
-	uint8_t  tg_size_en     = 0;
-	uint8_t  tidig_comp_cnt = 0;
-	uint8_t  lds_size       = 0;
+	uint64_t data_addr                 = 0;
+	uint32_t num_thread_x              = 0;
+	uint32_t num_thread_y              = 0;
+	uint32_t num_thread_z              = 0;
+	uint8_t  vgprs                     = 0;
+	uint8_t  sgprs                     = 0;
+	uint8_t  float_mode                = 0;
+	bool     dx10_clamp                = false;
+	bool     ieee_mode                 = false;
+	bool     fp16_overflow             = false;
+	bool     bulky                     = false;
+	bool     threadgroup_configuration = false;
+	uint8_t  wave_size                 = 64;
+	bool     scratch_en                = false;
+	uint8_t  user_sgpr                 = 0;
+	bool     tgid_x_en                 = false;
+	bool     tgid_y_en                 = false;
+	bool     tgid_z_en                 = false;
+	bool     tg_size_en                = false;
+	uint8_t  tidig_comp_cnt            = 0;
+	uint16_t lds_size                  = 0;
 };
 
 struct EsStageRegisters {
@@ -543,24 +578,24 @@ struct LsStageRegisters {
 
 struct HsStageRegisters {
 	uint64_t          data_addr = 0;
-	VsShaderResource1 rsrc1;
-	VsShaderResource2 rsrc2;
+	HsShaderResource1 rsrc1;
+	HsShaderResource2 rsrc2;
 	uint64_t          chksum = 0;
 };
 
 struct GsShaderResource1 {
-	uint8_t vgprs                    = 0;
-	uint8_t sgprs                    = 0;
-	uint8_t priority                 = 0;
-	uint8_t float_mode               = 0;
-	bool    dx10_clamp               = false;
-	bool    debug_mode               = false;
-	bool    ieee_mode                = false;
-	bool    cu_group_enable          = false;
-	bool    require_forward_progress = false;
-	bool    lds_configuration        = false;
-	uint8_t gs_vgpr_component_count  = 0;
-	bool    fp16_overflow            = false;
+	uint8_t vgprs                     = 0;
+	uint8_t sgprs                     = 0;
+	uint8_t priority                  = 0;
+	uint8_t float_mode                = 0;
+	bool    dx10_clamp                = false;
+	bool    debug_mode                = false;
+	bool    ieee_mode                 = false;
+	bool    cu_group_enable           = false;
+	bool    require_forward_progress  = false;
+	bool    threadgroup_configuration = false;
+	uint8_t gs_vgpr_component_count   = 0;
+	bool    fp16_overflow             = false;
 };
 
 struct GsShaderResource2 {
@@ -880,6 +915,8 @@ public:
 	void SetDepthControl(const DepthControl& control) { m_depth_control = control; }
 	[[nodiscard]] const ModeControl& GetModeControl() const { return m_mode_control; }
 	void SetModeControl(const ModeControl& control) { m_mode_control = control; }
+	[[nodiscard]] const PolyOffset& GetPolyOffset() const { return m_poly_offset; }
+	void SetPolyOffset(const PolyOffset& offset) { m_poly_offset = offset; }
 	[[nodiscard]] const EqaaControl& GetEqaaControl() const { return m_eqaa_control; }
 	void SetEqaaControl(const EqaaControl& control) { m_eqaa_control = control; }
 	[[nodiscard]] const StencilControl& GetStencilControl() const { return m_stencil_control; }
@@ -977,6 +1014,7 @@ private:
 	uint8_t           m_stencil_clear_value = 0;
 
 	ModeControl m_mode_control;
+	PolyOffset  m_poly_offset;
 	EqaaControl m_eqaa_control;
 
 	ShaderRegisters m_sh_regs;
@@ -1034,8 +1072,8 @@ public:
 	void SetLsShaderResource1(const VsShaderResource1& rsrc1) { m_vs.ls_regs.rsrc1 = rsrc1; }
 	void SetLsShaderResource2(const VsShaderResource2& rsrc2) { m_vs.ls_regs.rsrc2 = rsrc2; }
 	void SetHsShaderBase(uint64_t addr) { m_vs.hs_regs.data_addr = addr; }
-	void SetHsShaderResource1(const VsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
-	void SetHsShaderResource2(const VsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
+	void SetHsShaderResource1(const HsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
+	void SetHsShaderResource2(const HsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
 	void SetHsShaderChksum(uint32_t value) { m_vs.hs_regs.chksum = value; }
 	void SetGsShaderBase(uint64_t addr) { m_vs.gs_regs.data_addr = addr; }
 	void SetGsShaderResource1(const GsShaderResource1& rsrc1) { m_vs.gs_regs.rsrc1 = rsrc1; }

src/graphics/guest_gpu/pm4.cpp

@@ -30,6 +30,7 @@ constexpr auto MakeRegisterNames() {
 	names[R_FLIP]           = "R_FLIP";
 	names[R_RELEASE_MEM]    = "R_RELEASE_MEM";
 	names[R_DMA_DATA]       = "R_DMA_DATA";
+	names[R_CONTEXT_STATE]  = "R_CONTEXT_STATE";
 	return names;
 }
 

src/graphics/guest_gpu/pm4.h

@@ -87,6 +87,7 @@ constexpr uint32_t R_WAIT_MEM_64    = 0x16;
 constexpr uint32_t R_FLIP           = 0x17;
 constexpr uint32_t R_RELEASE_MEM    = 0x18;
 constexpr uint32_t R_DMA_DATA       = 0x19;
+constexpr uint32_t R_CONTEXT_STATE  = 0x1A;
 
 constexpr uint32_t R_NUM = 0x3F + 1;
 
@@ -371,6 +372,7 @@ constexpr uint32_t PA_CL_UCP_0_X          = 0x16F;
 constexpr uint32_t PA_CL_UCP_0_Y          = 0x170;
 constexpr uint32_t PA_CL_UCP_0_Z          = 0x171;
 constexpr uint32_t PA_CL_UCP_0_W          = 0x172;
+constexpr uint32_t PA_CL_UCP_5_W          = 0x186;
 constexpr uint32_t SPI_PS_INPUT_CNTL_0    = 0x191;
 constexpr uint32_t SPI_PS_INPUT_CNTL_31   = 0x1B0;
 constexpr uint32_t SPI_VS_OUT_CONFIG      = 0x1B1;
@@ -577,24 +579,28 @@ constexpr uint32_t DB_HTILE_SURFACE_PREFETCH_HEIGHT_MASK          = 0x3F;
 constexpr uint32_t DB_HTILE_SURFACE_DST_OUTSIDE_ZERO_TO_ONE_SHIFT = 16;
 constexpr uint32_t DB_HTILE_SURFACE_DST_OUTSIDE_ZERO_TO_ONE_MASK  = 0x1;
 
-constexpr uint32_t DB_SRESULTS_COMPARE_STATE0     = 0x2B0;
-constexpr uint32_t DB_SRESULTS_COMPARE_STATE1     = 0x2B1;
-constexpr uint32_t VGT_GS_MAX_VERT_OUT            = 0x2CE;
-constexpr uint32_t GE_NGG_SUBGRP_CNTL             = 0x2D3;
-constexpr uint32_t VGT_TESS_DISTRIBUTION          = 0x2D4;
-constexpr uint32_t VGT_SHADER_STAGES_EN           = 0x2D5;
-constexpr uint32_t VGT_LS_HS_CONFIG               = 0x2D6;
-constexpr uint32_t VGT_TF_PARAM                   = 0x2DB;
-constexpr uint32_t DB_ALPHA_TO_MASK               = 0x2DC;
-constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL  = 0x2DE;
-constexpr uint32_t PA_SU_POLY_OFFSET_CLAMP        = 0x2DF;
-constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_SCALE  = 0x2E0;
-constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_OFFSET = 0x2E1;
-constexpr uint32_t PA_SU_POLY_OFFSET_BACK_SCALE   = 0x2E2;
-constexpr uint32_t PA_SU_POLY_OFFSET_BACK_OFFSET  = 0x2E3;
-constexpr uint32_t VGT_GS_INSTANCE_CNT            = 0x2E4;
-constexpr uint32_t PA_SC_CENTROID_PRIORITY_0      = 0x2F5;
-constexpr uint32_t PA_SC_CENTROID_PRIORITY_1      = 0x2F6;
+constexpr uint32_t DB_SRESULTS_COMPARE_STATE0                          = 0x2B0;
+constexpr uint32_t DB_SRESULTS_COMPARE_STATE1                          = 0x2B1;
+constexpr uint32_t VGT_GS_MAX_VERT_OUT                                 = 0x2CE;
+constexpr uint32_t GE_NGG_SUBGRP_CNTL                                  = 0x2D3;
+constexpr uint32_t VGT_TESS_DISTRIBUTION                               = 0x2D4;
+constexpr uint32_t VGT_SHADER_STAGES_EN                                = 0x2D5;
+constexpr uint32_t VGT_LS_HS_CONFIG                                    = 0x2D6;
+constexpr uint32_t VGT_TF_PARAM                                        = 0x2DB;
+constexpr uint32_t DB_ALPHA_TO_MASK                                    = 0x2DC;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL                       = 0x2DE;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_NEG_NUM_DB_BITS_SHIFT = 0;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_NEG_NUM_DB_BITS_MASK  = 0xFF;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_DB_IS_FLOAT_FMT_SHIFT = 8;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_DB_IS_FLOAT_FMT_MASK  = 0x1;
+constexpr uint32_t PA_SU_POLY_OFFSET_CLAMP                             = 0x2DF;
+constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_SCALE                       = 0x2E0;
+constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_OFFSET                      = 0x2E1;
+constexpr uint32_t PA_SU_POLY_OFFSET_BACK_SCALE                        = 0x2E2;
+constexpr uint32_t PA_SU_POLY_OFFSET_BACK_OFFSET                       = 0x2E3;
+constexpr uint32_t VGT_GS_INSTANCE_CNT                                 = 0x2E4;
+constexpr uint32_t PA_SC_CENTROID_PRIORITY_0                           = 0x2F5;
+constexpr uint32_t PA_SC_CENTROID_PRIORITY_1                           = 0x2F6;
 
 constexpr uint32_t PA_SC_AA_CONFIG                             = 0x2F8;
 constexpr uint32_t PA_SC_AA_CONFIG_MSAA_NUM_SAMPLES_SHIFT      = 0;
@@ -778,21 +784,21 @@ constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FWD_PROGRESS_MASK      = 0x1;
 constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FP16_OVFL_SHIFT        = 29;
 constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FP16_OVFL_MASK         = 0x1;
 
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS                                = 0xB;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_SHIFT               = 0;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_MASK                = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_SHIFT                = 1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MASK                 = 0x1F;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_SHIFT              = 7;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_MASK               = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_SHIFT           = 8;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_MASK            = 0xFF;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_LOAD_INTRAWAVE_COLLISION_SHIFT = 26;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_LOAD_INTRAWAVE_COLLISION_MASK  = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_SHIFT            = 27;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_MASK             = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_SHIFT          = 28;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_MASK           = 0xF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS                              = 0xB;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_SHIFT             = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_MASK              = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_SHIFT              = 1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MASK               = 0x1F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_SHIFT            = 7;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_MASK             = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_SHIFT         = 8;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_MASK          = 0xFF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_RASTER_ORDERED_SHADING_SHIFT = 25;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_RASTER_ORDERED_SHADING_MASK  = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_SHIFT          = 27;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_MASK           = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_SHIFT        = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_MASK         = 0xF;
 
 constexpr uint32_t SPI_SHADER_USER_DATA_PS_0  = 0xC;
 constexpr uint32_t SPI_SHADER_USER_DATA_PS_15 = 0x1B;
@@ -885,61 +891,101 @@ constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_USER_SGPR_MSB_MASK     = 0x1;
 constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_SHARED_VGPR_CNT_SHIFT  = 28;
 constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_SHARED_VGPR_CNT_MASK   = 0xF;
 
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_0       = 0x8C;
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_15      = 0x9B;
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_31      = 0xAB;
-constexpr uint32_t SPI_SHADER_REQ_CTRL_ESGS        = 0xB0;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_0    = 0xB2;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_3    = 0xB5;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES_GS      = 0xBC;
-constexpr uint32_t SPI_SHADER_PGM_LO_ES            = 0xC8;
-constexpr uint32_t SPI_SHADER_PGM_HI_ES            = 0xC9;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_ES         = 0xCA;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES         = 0xCB;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_0       = 0xCC;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_15      = 0xDB;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_31      = 0xEB;
-constexpr uint32_t SPI_SHADER_PGM_CHKSUM_HS        = 0x100;
-constexpr uint32_t SPI_SHADER_PGM_RSRC4_HS         = 0x101;
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_HS = 0x102;
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_HI_HS = 0x103;
-constexpr uint32_t SPI_SHADER_PGM_RSRC3_HS         = 0x107;
-constexpr uint32_t SPI_SHADER_PGM_LO_HS            = 0x108;
-constexpr uint32_t SPI_SHADER_PGM_HI_HS            = 0x109;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS         = 0x10A;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS         = 0x10B;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_0       = 0x10C;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_15      = 0x11B;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_31      = 0x12B;
-constexpr uint32_t SPI_SHADER_REQ_CTRL_LSHS        = 0x130;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_0    = 0x132;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_3    = 0x135;
-constexpr uint32_t SPI_SHADER_PGM_LO_LS            = 0x148;
-constexpr uint32_t SPI_SHADER_PGM_HI_LS            = 0x149;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_LS         = 0x14A;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_LS         = 0x14B;
-constexpr uint32_t COMPUTE_START_X                 = 0x204;
-constexpr uint32_t COMPUTE_START_Y                 = 0x205;
-constexpr uint32_t COMPUTE_START_Z                 = 0x206;
-constexpr uint32_t COMPUTE_NUM_THREAD_X            = 0x207;
-constexpr uint32_t COMPUTE_NUM_THREAD_Y            = 0x208;
-constexpr uint32_t COMPUTE_NUM_THREAD_Z            = 0x209;
-constexpr uint32_t COMPUTE_PGM_LO                  = 0x20C;
-constexpr uint32_t COMPUTE_PGM_HI                  = 0x20D;
-constexpr uint32_t COMPUTE_TBA_LO                  = 0x20E;
-constexpr uint32_t COMPUTE_TBA_HI                  = 0x20F;
-constexpr uint32_t COMPUTE_TMA_LO                  = 0x210;
-constexpr uint32_t COMPUTE_TMA_HI                  = 0x211;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_0                      = 0x8C;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_15                     = 0x9B;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_31                     = 0xAB;
+constexpr uint32_t SPI_SHADER_REQ_CTRL_ESGS                       = 0xB0;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_0                   = 0xB2;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_3                   = 0xB5;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES_GS                     = 0xBC;
+constexpr uint32_t SPI_SHADER_PGM_LO_ES                           = 0xC8;
+constexpr uint32_t SPI_SHADER_PGM_HI_ES                           = 0xC9;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_ES                        = 0xCA;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES                        = 0xCB;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_0                      = 0xCC;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_15                     = 0xDB;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_31                     = 0xEB;
+constexpr uint32_t SPI_SHADER_PGM_CHKSUM_HS                       = 0x100;
+constexpr uint32_t SPI_SHADER_PGM_RSRC4_HS                        = 0x101;
+constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_HS                = 0x102;
+constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_HI_HS                = 0x103;
+constexpr uint32_t SPI_SHADER_PGM_RSRC3_HS                        = 0x107;
+constexpr uint32_t SPI_SHADER_PGM_LO_HS                           = 0x108;
+constexpr uint32_t SPI_SHADER_PGM_HI_HS                           = 0x109;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS                        = 0x10A;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS                        = 0x10B;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_VGPRS_SHIFT            = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_VGPRS_MASK             = 0x3F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_PRIORITY_SHIFT         = 10;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_PRIORITY_MASK          = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FLOAT_MODE_SHIFT       = 12;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FLOAT_MODE_MASK        = 0xFF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DX10_CLAMP_SHIFT       = 21;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DX10_CLAMP_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DEBUG_MODE_SHIFT       = 22;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DEBUG_MODE_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_IEEE_MODE_SHIFT        = 23;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_IEEE_MODE_MASK         = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FWD_PROGRESS_SHIFT     = 25;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FWD_PROGRESS_MASK      = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_WGP_MODE_SHIFT         = 26;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_WGP_MODE_MASK          = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_LS_VGPR_COMP_CNT_SHIFT = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_LS_VGPR_COMP_CNT_MASK  = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FP16_OVFL_SHIFT        = 30;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FP16_OVFL_MASK         = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SCRATCH_EN_SHIFT       = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SCRATCH_EN_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_SHIFT        = 1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MASK         = 0x1F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_LDS_SIZE_SHIFT         = 18;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_LDS_SIZE_MASK          = 0x1FF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MSB_SHIFT    = 27;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MSB_MASK     = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SHARED_VGPR_CNT_SHIFT  = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SHARED_VGPR_CNT_MASK   = 0xF;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_0                      = 0x10C;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_15                     = 0x11B;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_31                     = 0x12B;
+constexpr uint32_t SPI_SHADER_REQ_CTRL_LSHS                       = 0x130;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_0                   = 0x132;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_3                   = 0x135;
+constexpr uint32_t SPI_SHADER_PGM_LO_LS                           = 0x148;
+constexpr uint32_t SPI_SHADER_PGM_HI_LS                           = 0x149;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_LS                        = 0x14A;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_LS                        = 0x14B;
+constexpr uint32_t COMPUTE_START_X                                = 0x204;
+constexpr uint32_t COMPUTE_START_Y                                = 0x205;
+constexpr uint32_t COMPUTE_START_Z                                = 0x206;
+constexpr uint32_t COMPUTE_NUM_THREAD_X                           = 0x207;
+constexpr uint32_t COMPUTE_NUM_THREAD_Y                           = 0x208;
+constexpr uint32_t COMPUTE_NUM_THREAD_Z                           = 0x209;
+constexpr uint32_t COMPUTE_PGM_LO                                 = 0x20C;
+constexpr uint32_t COMPUTE_PGM_HI                                 = 0x20D;
+constexpr uint32_t COMPUTE_TBA_LO                                 = 0x20E;
+constexpr uint32_t COMPUTE_TBA_HI                                 = 0x20F;
+constexpr uint32_t COMPUTE_TMA_LO                                 = 0x210;
+constexpr uint32_t COMPUTE_TMA_HI                                 = 0x211;
 
-constexpr uint32_t COMPUTE_PGM_RSRC1              = 0x212;
-constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_SHIFT  = 0;
-constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_MASK   = 0x3F;
-constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_SHIFT  = 6;
-constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_MASK   = 0xF;
-constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_SHIFT  = 24;
-constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_MASK   = 0x1;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_SHIFT = 30;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_MASK  = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1                  = 0x212;
+constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_SHIFT      = 0;
+constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_MASK       = 0x3F;
+constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_SHIFT      = 6;
+constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_MASK       = 0xF;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FLOAT_MODE_SHIFT = 12;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FLOAT_MODE_MASK  = 0xFF;
+constexpr uint32_t COMPUTE_PGM_RSRC1_DX10_CLAMP_SHIFT = 21;
+constexpr uint32_t COMPUTE_PGM_RSRC1_DX10_CLAMP_MASK  = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_IEEE_MODE_SHIFT  = 23;
+constexpr uint32_t COMPUTE_PGM_RSRC1_IEEE_MODE_MASK   = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_SHIFT      = 24;
+constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_MASK       = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT  = 26;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_MASK   = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT   = 29;
+constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_MASK    = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_SHIFT     = 30;
+constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_MASK      = 0x1;
 
 constexpr uint32_t COMPUTE_PGM_RSRC2                      = 0x213;
 constexpr uint32_t COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT     = 0;
@@ -1002,6 +1048,7 @@ constexpr uint32_t GE_USER_VGPR1  = 0x25C;
 constexpr uint32_t GE_USER_VGPR2  = 0x25D;
 constexpr uint32_t GE_USER_VGPR3  = 0x25E;
 constexpr uint32_t GE_STEREO_CNTL = 0x25F;
+constexpr uint32_t GE_PC_ALLOC    = 0x260;
 
 constexpr uint32_t GE_USER_VGPR_EN                     = 0x262;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR1_SHIFT = 0;

src/graphics/guest_gpu/tile.h

@@ -39,7 +39,7 @@ enum class TileBlockFamily : uint32_t {
 };
 
 struct TileBlockLayout {
-	TileBlockFamily family            = TileBlockFamily::Standard256B;
+	TileBlockFamily family            = TileBlockFamily::Count;
 	uint32_t        bytes_per_element = 0;
 	uint32_t        block_size        = 0;
 	uint32_t        block_width       = 0;
@@ -47,38 +47,69 @@ struct TileBlockLayout {
 	uint32_t        block_depth       = 0;
 };
 
-struct TileVolumeLayout {
-	TileBlockFamily family            = TileBlockFamily::Count;
-	uint32_t        bytes_per_element = 0;
-	uint32_t        texel_width       = 1;
-	uint32_t        texel_height      = 1;
-	uint32_t        first_tail_level  = 0;
-	uint32_t        block_depth       = 1;
-	uint64_t        block_slice_size  = 0;
-	uint64_t        total_size        = 0;
-	uint64_t        level_offsets[16] = {};
-	uint64_t        level_sizes[16]   = {};
-	uint32_t        tail_x[16]        = {};
-	uint32_t        tail_y[16]        = {};
-	uint32_t        level_widths[16]  = {};
-	uint32_t        level_heights[16] = {};
+struct TileTextureElementLayout {
+	uint32_t bytes        = 0;
+	uint32_t texel_width  = 1;
+	uint32_t texel_height = 1;
+};
+
+struct TileTextureBlockLayout {
+	TileBlockLayout block;
+	uint32_t        texel_width  = 1;
+	uint32_t        texel_height = 1;
+};
+
+enum class TileSurfaceDimension : uint32_t {
+	Dim2D,
+	Dim3D,
+};
+
+struct TileSurfaceDescription {
+	uint32_t             format    = 0;
+	uint32_t             tile_mode = 0;
+	TileSurfaceDimension dimension = TileSurfaceDimension::Dim2D;
+	uint32_t             width     = 0;
+	uint32_t             height    = 0;
+	uint32_t             depth     = 1;
+	uint32_t             levels    = 1;
+	uint32_t             layers    = 1;
+};
+
+// Width/height fields are measured in addressable elements, not texture texels.
+struct TileMipLayout {
+	uint64_t offset        = 0;
+	uint64_t size          = 0;
+	uint32_t width         = 0;
+	uint32_t height        = 0;
+	uint32_t padded_width  = 0;
+	uint32_t padded_height = 0;
+	uint32_t tail_x        = 0;
+	uint32_t tail_y        = 0;
+};
+
+struct TileSurfaceLayout {
+	TileSurfaceDescription description;
+	TileTextureBlockLayout texture;
+	uint32_t               first_tail_level = 0;
+	uint64_t               block_slice_size = 0;
+	uint64_t               total_size       = 0;
+	TileMipLayout          mips[16]         = {};
 };
 
 bool TileGetBlockLayout(TileBlockFamily family, uint32_t bytes_per_element,
                         TileBlockLayout& layout);
+bool TileGetTextureElementLayout(uint32_t format, TileTextureElementLayout& layout);
+bool TileGetTextureBlockLayout(uint32_t format, uint32_t tile, bool volume,
+                               TileTextureBlockLayout& layout);
+bool TileGetTiledTextureLayout(const TileSurfaceDescription& description,
+                               TileSurfaceLayout&            layout);
 bool TileGetBlockOffset(const TileBlockLayout& layout, uint32_t x, uint32_t y, uint32_t z,
                         uint32_t& byte_offset);
 bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t block_y,
                      uint32_t& byte_offset);
 bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t block_y,
                      uint32_t block_z, uint32_t& byte_offset);
-bool TileIsStandard256BTextureSupported(uint32_t format);
-bool TileIsStandard4KBTextureSupported(uint32_t format);
-bool TileIsStandard64KBTextureSupported(uint32_t format);
-bool TileGetTextureVolumeLayout(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                                uint32_t levels, uint32_t tile, TileVolumeLayout& layout);
 
-bool     TileGetHtileSize(uint32_t width, uint32_t height, TileSizeAlign& htile_size);
 bool     TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t z_format,
                           uint32_t stencil_format, bool htile, TileSizeAlign& stencil_size,
                           TileSizeAlign& htile_size, TileSizeAlign& depth_size,
@@ -94,13 +125,13 @@ bool     TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t
                                       uint32_t bytes_per_element, uint32_t levels,
                                       TileSizeAlign& total_size, TileSizeOffset* level_sizes,
                                       TilePaddedSize* padded_size);
-void     TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t pitch,
-                            uint32_t levels, uint32_t tile, TileSizeAlign* total_size,
-                            TileSizeOffset* level_sizes, TilePaddedSize* padded_size);
-void TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                             uint32_t pitch, uint32_t levels, uint32_t tile, bool volume_texture,
-                             TileSizeAlign& total_size);
-uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t levels, uint32_t tile);
+void     TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t levels,
+                            uint32_t tile, TileSizeAlign* total_size, TileSizeOffset* level_sizes,
+                            TilePaddedSize* padded_size);
+void     TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
+                                 uint32_t levels, uint32_t tile, bool volume_texture,
+                                 TileSizeAlign& total_size);
+uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t tile);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/memoryTracker.cpp

@@ -79,7 +79,6 @@ RegionManager* MemoryTracker::GetOrCreateRegion(uint64_t index) {
 
 bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	return Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Cpu>(offset, bytes);
@@ -88,7 +87,6 @@ bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 
 bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	return Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Gpu>(offset, bytes);
@@ -97,7 +95,6 @@ bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
@@ -106,7 +103,6 @@ void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Gpu, true>(manager->GetCpuAddr() + offset, bytes);
@@ -115,14 +111,13 @@ void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Gpu, false>(manager->GetCpuAddr() + offset, bytes);
 	});
 }
 
-void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
+void MemoryTracker::UntrackMemoryImpl(uint64_t vaddr, uint64_t size) {
 	std::vector<RegionManager*> managers;
 	managers.reserve((vaddr % TRACKER_REGION_SIZE + size + TRACKER_REGION_SIZE - 1) /
 	                 TRACKER_REGION_SIZE);
@@ -148,8 +143,7 @@ void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
-	UntrackMemoryLocked(vaddr, size);
+	UntrackMemoryImpl(vaddr, size);
 }
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/memoryTracker.h

@@ -29,45 +29,29 @@ public:
 	void               MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size);
 	void               UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size);
 	void               UntrackMemory(uint64_t vaddr, uint64_t size);
+	// Removes protection from a range and flushes GPU-owned data when required.
 	template <typename Flush>
-	void InvalidateRegion(uint64_t vaddr, uint64_t size, Flush&& on_flush) {
+	void InvalidateRegion(uint64_t vaddr, uint64_t size, Flush&& on_flush) noexcept {
 		static_assert(std::is_invocable_v<Flush&>);
 		CheckNotInUploadCallback();
 		ValidateRange(vaddr, size);
 
-		const auto update_cpu_state = [this, vaddr, size] {
-			std::lock_guard             access(m_access_mutex);
-			std::vector<RegionManager*> managers;
-			Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t, uint64_t) {
-				managers.push_back(manager);
-			});
-			std::vector<std::unique_lock<TrackingSpinLock>> locks;
-			locks.reserve(managers.size());
-			for (auto* manager: managers) {
-				locks.emplace_back(manager->lock);
+		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+			const bool should_flush = [&] {
+				// Perform both the GPU modification check and CPU state change with the lock in
+				// case the GPU thread is racing to mark the page modified. If a flush is needed,
+				// on_flush performs the CPU state change.
+				std::scoped_lock lock(manager->lock);
+				if (manager->IsModified<DirtySource::Gpu>(offset, bytes)) {
+					return true;
+				}
+				manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
+				return false;
+			}();
+			if (should_flush) {
+				on_flush();
 			}
-			const bool gpu_modified = Iterate<false>(
-			    vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-				    return manager->IsModified<DirtySource::Gpu>(offset, bytes);
-			    });
-			if (gpu_modified) {
-				return true;
-			}
-			Iterate<false>(vaddr, size,
-			               [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-				               manager->ChangeState<DirtySource::Cpu, true>(
-				                   manager->GetCpuAddr() + offset, bytes);
-			               });
-			return false;
-		};
-
-		if (!update_cpu_state()) {
-			return;
-		}
-		std::forward<Flush>(on_flush)();
-		if (update_cpu_state()) {
-			EXIT("memory invalidation retained GPU-owned pages\n");
-		}
+		});
 	}
 #if KYTY_BUILD == KYTY_BUILD_DEBUG
 	void ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr, uint64_t size,
@@ -84,7 +68,6 @@ public:
 		static_assert(std::is_nothrow_invocable_v<Preflight&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<Func&, uint64_t, uint64_t>);
 		CheckNotInUploadCallback();
-		std::lock_guard             access(m_access_mutex);
 		std::vector<RegionManager*> managers;
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t, uint64_t) {
 			managers.push_back(manager);
@@ -125,7 +108,6 @@ public:
 		static_assert(std::is_nothrow_invocable_v<RangeFunc&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<UploadFunc&>);
 		CheckNotInUploadCallback();
-		std::unique_lock access(m_access_mutex);
 		Iterate<true>(vaddr, size, [](RegionManager*, uint64_t, uint64_t) {});
 		const auto* previous_upload_owner = std::exchange(s_upload_owner, this);
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
@@ -189,13 +171,12 @@ private:
 	}
 
 	static void    ValidateRange(uint64_t vaddr, uint64_t size);
-	void           UntrackMemoryLocked(uint64_t vaddr, uint64_t size);
+	void           UntrackMemoryImpl(uint64_t vaddr, uint64_t size);
 	RegionManager* GetOrCreateRegion(uint64_t index);
 
 	std::unique_ptr<std::atomic<RegionManager*>[]> m_regions;
 	std::vector<std::unique_ptr<RegionManager>>    m_region_storage;
 	std::mutex                                     m_region_mutex;
-	std::mutex                                     m_access_mutex;
 	PageManager&                                   m_page_manager;
 };
 

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

@@ -3,15 +3,17 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
+#include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/render.h"
+#include "graphics/host_gpu/renderer/renderContext.h"
 #include "kernel/memory.h"
 
 #include <algorithm>
 #include <array>
+#include <cinttypes>
 #include <cstring>
 #include <utility>
 #include <vector>
@@ -20,30 +22,9 @@ namespace Libs::Graphics {
 
 namespace {
 
-thread_local const void* g_cache_lock_owner = nullptr;
-
 constexpr uint64_t MiB           = 1024 * 1024;
 constexpr uint64_t GdsBufferSize = 64 * 1024;
 
-class FaultSafeCacheLock final {
-public:
-	FaultSafeCacheLock(const void* owner, Common::Mutex& mutex): m_mutex(mutex) {
-		if (g_cache_lock_owner != nullptr) {
-			EXIT("BufferCache: recursive cache lock acquisition\n");
-		}
-		g_cache_lock_owner = owner;
-		m_mutex.Lock();
-	}
-
-	~FaultSafeCacheLock() {
-		m_mutex.Unlock();
-		g_cache_lock_owner = nullptr;
-	}
-
-private:
-	Common::Mutex& m_mutex;
-};
-
 } // namespace
 
 uint64_t BufferCache::AlignDown(uint64_t value) noexcept {
@@ -57,15 +38,6 @@ uint64_t BufferCache::AlignUp(uint64_t value) {
 	return (value + CACHING_PAGE_SIZE - 1) & ~(CACHING_PAGE_SIZE - 1);
 }
 
-bool BufferCache::PageOverlaps(uint64_t left, uint64_t left_size, uint64_t right,
-                               uint64_t right_size) noexcept {
-	const auto left_begin  = left & ~(TRACKER_PAGE_SIZE - 1);
-	const auto left_end    = (left + left_size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	const auto right_begin = right & ~(TRACKER_PAGE_SIZE - 1);
-	const auto right_end = (right + right_size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	return left_begin < right_end && right_begin < left_end;
-}
-
 void BufferCache::Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
                          const void* source, uint64_t size) {
 	auto* bytes = static_cast<const uint8_t*>(source);
@@ -198,27 +170,23 @@ void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, Ret
 	m_scheduler.DeferOperation(
 	    [this, downloads = std::move(downloads), retire = std::move(retire)]() mutable {
 		    PublishDownloads(downloads);
-		    {
-			    FaultSafeCacheLock lock(this, m_mutex);
-			    for (const auto& range: downloads) {
-				    m_gpu_modified_ranges.Subtract(range.address, range.size);
-			    }
-			    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
-			    // interval on those pages was downloaded and removed. Clearing the original
-			    // query therefore cannot orphan a dirty sibling on an edge page.
-			    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
-			    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
-			        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
-				    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
-			    }
-			    m_memory_tracker.UntrackMemory(retire.address, retire.size);
+		    for (const auto& range: downloads) {
+			    m_gpu_modified_ranges.Subtract(range.address, range.size);
 		    }
+		    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
+		    // interval on those pages was downloaded and removed. Clearing the original
+		    // query therefore cannot orphan a dirty sibling on an edge page.
+		    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
+		    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
+		        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
+			    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
+		    }
+		    m_memory_tracker.UntrackMemory(retire.address, retire.size);
 	    });
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
-                         PageManager& page_manager, TextureCache& texture_cache,
-                         ResourceMutex& resource_mutex)
+                         PageManager& page_manager, TextureCache& texture_cache)
     : m_graphics(graphics), m_scheduler(scheduler),
       m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
       m_memory_tracker(page_manager),
@@ -226,7 +194,7 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
       m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
       m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
       m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
-      m_texture_cache(texture_cache), m_resource_mutex(resource_mutex) {
+      m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
 	m_gds_buffer.Flush(0, m_gds_buffer.Size());
 	if (!m_graphics.CanReportMemoryUsage()) {
@@ -284,58 +252,77 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
 		EXIT("BufferCache: invalid memory-invalidation range\n");
 	}
-	if (!HasPageOverlap(vaddr, size)) {
-		return;
-	}
 	m_memory_tracker.InvalidateRegion(vaddr, size,
-	                                  [this, vaddr, size] { ReadMemory(vaddr, size); });
+	                                  [this, vaddr, size] { ReadMemory(vaddr, size, true); });
 }
 
-void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
-	std::vector<DownloadCopy> copies;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		m_memory_tracker.ForEachDownloadRange<false>(
-		    vaddr, size,
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-			                                           "memory invalidation");
-		    },
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-				    for (uint64_t copied = 0; copied < range.size;) {
-					    const auto copy_address = range.address + copied;
-					    auto       owner        = m_buffers.upper_bound(copy_address);
-					    if (owner == m_buffers.begin()) {
-						    EXIT("BufferCache: invalidation readback has no buffer owner\n");
-					    }
-					    auto& cached = *std::prev(owner)->second;
-					    if (!cached.buffer->IsInBounds(copy_address, 1)) {
-						    EXIT(
-						        "BufferCache: invalidation readback is outside its buffer owner\n");
-					    }
-					    const auto copy_size = std::min(range.size - copied,
-					                                    cached.vaddr + cached.size - copy_address);
-					    copies.push_back({cached.buffer, cached.buffer->Offset(copy_address),
-					                      copy_address, copy_size});
-					    copied += copy_size;
-				    }
-			    }
-		    });
+void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size, bool is_write) {
+	if (Gpu::IsGpuThread()) {
+		ReadMemoryOnGpu(vaddr, size, is_write);
+		return;
 	}
+	if (CommandScheduler::InDeferredOperation()) {
+		EXIT("unsupported buffer readback from an asynchronous GPU completion, "
+		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
+		     vaddr, size);
+	}
+	m_scheduler.Context().GetGpu().SendCommandSync(
+	    [this, vaddr, size, is_write] { ReadMemoryOnGpu(vaddr, size, is_write); });
+}
+
+void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
+	// CPU invalidation reaches this point only for a GPU-owned tracker page. Resolve the exact
+	// Buffer owner on the GPU thread so the cache index remains single-thread-owned.
+	if (is_write && !IsRegionRegistered(vaddr, size)) {
+		return;
+	}
+	std::vector<DownloadCopy> copies;
+	m_memory_tracker.ForEachDownloadRange<false>(
+	    vaddr, size,
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+		                                           "memory invalidation");
+	    },
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
+			    for (uint64_t copied = 0; copied < range.size;) {
+				    const auto copy_address = range.address + copied;
+				    auto       owner        = m_buffers.upper_bound(copy_address);
+				    if (owner == m_buffers.begin()) {
+					    EXIT("BufferCache: invalidation readback has no buffer owner\n");
+				    }
+				    auto& cached = *std::prev(owner)->second;
+				    if (!cached.buffer->IsInBounds(copy_address, 1)) {
+					    EXIT("BufferCache: invalidation readback is outside its buffer owner\n");
+				    }
+				    const auto copy_size =
+				        std::min(range.size - copied, cached.vaddr + cached.size - copy_address);
+				    copies.push_back({cached.buffer, cached.buffer->Offset(copy_address),
+				                      copy_address, copy_size});
+				    copied += copy_size;
+			    }
+		    }
+	    });
 	if (copies.empty()) {
+		if (!is_write) {
+			return;
+		}
+		// A preceding read fault can consume the last GPU-owned copy after this write invalidation
+		// has already chosen to flush. Complete the CPU ownership handoff even though this callback
+		// no longer has bytes to download.
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
 		return;
 	}
 	auto downloads = RecordDownloads(copies);
 	m_scheduler.FinishCurrent();
 	PublishDownloads(downloads);
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& range: downloads) {
-			m_gpu_modified_ranges.Subtract(range.address, range.size);
-		}
-		// The enumeration above covered whole dirty pages and every exact interval on them.
-		m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
+	for (const auto& range: downloads) {
+		m_gpu_modified_ranges.Subtract(range.address, range.size);
+	}
+	// The enumeration above covered whole dirty pages and every exact interval on them.
+	m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
+	if (is_write) {
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
 	}
 }
 
@@ -346,43 +333,40 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 	std::vector<DownloadCopy>                  copies;
 	std::vector<std::pair<uint64_t, uint64_t>> modified_buffers;
 	std::vector<std::pair<uint64_t, uint64_t>> retired_buffers;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& [begin, cached]: m_buffers) {
-			if (vaddr < begin + cached->size && begin < vaddr + size) {
-				retired_buffers.emplace_back(begin, cached->size);
-			}
+	for (const auto& [begin, cached]: m_buffers) {
+		if (vaddr < begin + cached->size && begin < vaddr + size) {
+			retired_buffers.emplace_back(begin, cached->size);
 		}
-		for (const auto& [begin, cached]: m_buffers) {
-			if (vaddr >= begin + cached->size || begin >= vaddr + size ||
-			    !m_memory_tracker.IsRegionGpuModified(begin, cached->size)) {
-				continue;
-			}
-			const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
-			if (dirty.empty()) {
-				EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
-			}
-			modified_buffers.emplace_back(begin, cached->size);
+	}
+	for (const auto& [begin, cached]: m_buffers) {
+		if (vaddr >= begin + cached->size || begin >= vaddr + size ||
+		    !m_memory_tracker.IsRegionGpuModified(begin, cached->size)) {
+			continue;
 		}
-		for (const auto& [begin, bytes]: modified_buffers) {
-			auto owner = m_buffers.find(begin);
-			if (owner == m_buffers.end() || owner->second->size != bytes) {
-				EXIT("BufferCache: unmap owner changed during collection\n");
-			}
-			auto& cached = *owner->second;
-			m_memory_tracker.ForEachDownloadRange<false>(
-			    begin, cached.size,
-			    [&](uint64_t address, uint64_t bytes) noexcept {
-				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-				                                           "unmap");
-			    },
-			    [&](uint64_t address, uint64_t bytes) noexcept {
-				    for (const auto& range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-					    copies.push_back(
-					        {cached.buffer, range.address - begin, range.address, range.size});
-				    }
-			    });
+		const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
+		if (dirty.empty()) {
+			EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
 		}
+		modified_buffers.emplace_back(begin, cached->size);
+	}
+	for (const auto& [begin, bytes]: modified_buffers) {
+		auto owner = m_buffers.find(begin);
+		if (owner == m_buffers.end() || owner->second->size != bytes) {
+			EXIT("BufferCache: unmap owner changed during collection\n");
+		}
+		auto& cached = *owner->second;
+		m_memory_tracker.ForEachDownloadRange<false>(
+		    begin, cached.size,
+		    [&](uint64_t address, uint64_t bytes) noexcept {
+			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+			                                           "unmap");
+		    },
+		    [&](uint64_t address, uint64_t bytes) noexcept {
+			    for (const auto& range: m_gpu_modified_ranges.Intersections(address, bytes)) {
+				    copies.push_back(
+				        {cached.buffer, range.address - begin, range.address, range.size});
+			    }
+		    });
 	}
 	if (!copies.empty()) {
 		auto downloads = RecordDownloads(copies);
@@ -393,29 +377,26 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		// command stream before removing such backing.
 		m_scheduler.FinishCurrent();
 	}
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& [begin, bytes]: modified_buffers) {
-			m_gpu_modified_ranges.Subtract(begin, bytes);
-			m_memory_tracker.UnmarkRegionAsGpuModified(begin, bytes);
-		}
-		for (const auto& [begin, bytes]: retired_buffers) {
-			m_memory_tracker.MarkRegionAsCpuModified(begin, bytes);
-		}
-		if (!m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
-			EXIT("BufferCache: unmap retained dirty byte ranges\n");
-		}
-		m_memory_tracker.UntrackMemory(vaddr, size);
-		for (auto it = m_buffers.begin(); it != m_buffers.end();) {
-			if (vaddr < it->first + it->second->size && it->first < vaddr + size) {
-				if (it->second->size > m_total_used_memory) {
-					EXIT("BufferCache: allocation accounting underflow\n");
-				}
-				m_total_used_memory -= it->second->size;
-				it = m_buffers.erase(it);
-			} else {
-				++it;
+	for (const auto& [begin, bytes]: modified_buffers) {
+		m_gpu_modified_ranges.Subtract(begin, bytes);
+		m_memory_tracker.UnmarkRegionAsGpuModified(begin, bytes);
+	}
+	for (const auto& [begin, bytes]: retired_buffers) {
+		m_memory_tracker.MarkRegionAsCpuModified(begin, bytes);
+	}
+	if (!m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
+		EXIT("BufferCache: unmap retained dirty byte ranges\n");
+	}
+	m_memory_tracker.UntrackMemory(vaddr, size);
+	for (auto it = m_buffers.begin(); it != m_buffers.end();) {
+		if (vaddr < it->first + it->second->size && it->first < vaddr + size) {
+			if (it->second->size > m_total_used_memory) {
+				EXIT("BufferCache: allocation accounting underflow\n");
 			}
+			m_total_used_memory -= it->second->size;
+			it = m_buffers.erase(it);
+		} else {
+			++it;
 		}
 	}
 }
@@ -497,7 +478,6 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 	if (command.IsInvalid() || command.IsExecute()) {
 		EXIT("BufferCache: buffer request requires a recording command buffer\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
 
 	if (is_read && !is_written && size <= CACHING_PAGE_SIZE &&
 	    !m_memory_tracker.IsRegionGpuModified(vaddr, size) &&
@@ -524,7 +504,6 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 		(void)m_texture_cache.InvalidateMemoryFromGPU(vaddr, size, true);
 	}
 
-	FaultSafeCacheLock                         lock(this, m_mutex);
 	auto&                                      cached = GetOrCreateBuffer(command, vaddr, size);
 	std::vector<std::pair<uint64_t, uint64_t>> uploads;
 	m_memory_tracker.ForEachUploadRange(
@@ -539,26 +518,23 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 	if (is_written) {
 		m_gpu_modified_ranges.Add(vaddr, size);
 	}
+	auto     owner  = cached.buffer;
+	uint64_t offset = cached.buffer->Offset(vaddr);
 	if (is_formatted && is_read && !is_written) {
-		(void)SynchronizeBufferFromImage(*cached.buffer, vaddr, size);
+		(void)SynchronizeBufferFromImage(*owner, vaddr, size);
 	}
-	return {cached.buffer, cached.buffer->Handle(), cached.buffer->Offset(vaddr)};
+	return {owner, owner->Handle(), offset};
 }
 
 std::shared_ptr<Buffer> BufferCache::ObtainNullBuffer() {
-	std::shared_ptr<Buffer> buffer;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		if (m_null_buffer != nullptr) {
-			return m_null_buffer;
-		}
-		m_null_buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal,
-		                                         0, AllFlags, 16);
-		buffer        = m_null_buffer;
+	if (m_null_buffer != nullptr) {
+		return m_null_buffer;
 	}
+	m_null_buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, 0,
+	                                         AllFlags, 16);
 	const std::array<uint8_t, 16> zeros {};
-	Upload(m_scheduler.Current(), *buffer, 0, zeros.data(), zeros.size());
-	return buffer;
+	Upload(m_scheduler.Current(), *m_null_buffer, 0, zeros.data(), zeros.size());
+	return m_null_buffer;
 }
 
 ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t size) {
@@ -576,11 +552,10 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	};
 
 	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		const bool         cpu_modified = m_memory_tracker.IsRegionCpuModified(vaddr, size);
-		const bool         gpu_modified = m_memory_tracker.IsRegionGpuModified(vaddr, size);
-		const auto         dirty        = m_gpu_modified_ranges.Intersections(vaddr, size);
-		const bool         has_dirty_buffer_source = !dirty.empty();
+		const bool cpu_modified            = m_memory_tracker.IsRegionCpuModified(vaddr, size);
+		const bool gpu_modified            = m_memory_tracker.IsRegionGpuModified(vaddr, size);
+		const auto dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
+		const bool has_dirty_buffer_source = !dirty.empty();
 		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
 		                                           "image source");
 
@@ -651,10 +626,9 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	}
 	m_staging_buffer.Commit();
 
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
-	const bool         has_dirty_buffer_source = !dirty.empty();
-	auto               owner                   = find_owner();
+	const auto dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
+	const bool has_dirty_buffer_source = !dirty.empty();
+	auto       owner                   = find_owner();
 	if (has_dirty_buffer_source && owner == m_buffers.end()) {
 		EXIT("BufferCache: GPU-dirty image source lost its native owner\n");
 	}
@@ -687,8 +661,7 @@ void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data)
 	}
 	Libs::LibKernel::Memory::WriteBacking(vaddr, data.data(), data.size());
 
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         end = vaddr + data.size();
+	const auto end = vaddr + data.size();
 	for (auto& [address, cached]: m_buffers) {
 		const auto cached_end = address + cached->size;
 		const auto begin      = std::max(vaddr, address);
@@ -718,8 +691,7 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 	}
 	(void)m_texture_cache.ClearMeta(vaddr);
 	{
-		std::lock_guard transaction(m_resource_mutex);
-		const auto      region = m_texture_cache.QueryRegion(vaddr, size);
+		const auto region = m_texture_cache.QueryRegion(vaddr, size);
 		if (!HasGpuDirtyBytes(vaddr, size) && !region.gpu_image_bytes) {
 			if (region.image_bytes) {
 				m_texture_cache.InvalidateMemory(vaddr, size);
@@ -750,16 +722,15 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	const bool dst_memory = !dst_gds;
 	const bool src_memory = !src_gds;
 	if ((dst_memory && dst_vaddr == 0) || (src_memory && src_vaddr == 0) || size == 0 ||
-	    ((dst_vaddr | src_vaddr | size) & 3u) != 0 || size > UINT64_MAX - dst_vaddr ||
-	    size > UINT64_MAX - src_vaddr || (dst_gds && src_gds) ||
+	    ((dst_gds || src_gds) && ((dst_vaddr | src_vaddr | size) & 3u) != 0) ||
+	    size > UINT64_MAX - dst_vaddr || size > UINT64_MAX - src_vaddr || (dst_gds && src_gds) ||
 	    (dst_gds == src_gds && src_vaddr < dst_vaddr + size && dst_vaddr < src_vaddr + size) ||
 	    (dst_gds && (dst_vaddr > m_gds_buffer.Size() || size > m_gds_buffer.Size() - dst_vaddr)) ||
 	    (src_gds && (src_vaddr > m_gds_buffer.Size() || size > m_gds_buffer.Size() - src_vaddr))) {
 		EXIT("BufferCache: invalid or overlapping copy range\n");
 	}
 	if (src_memory || dst_memory) {
-		std::lock_guard transaction(m_resource_mutex);
-		const auto      src_region =
+		const auto src_region =
 		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
 		const auto dst_region =
 		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
@@ -810,18 +781,19 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	destination.CopyFrom(command, source, src.offset, dst.offset, size);
 }
 
-bool BufferCache::HasPageOverlap(uint64_t vaddr, uint64_t size) {
+bool BufferCache::IsRegionRegistered(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid page-overlap query\n");
+		EXIT("BufferCache: invalid registered-region query\n");
 	}
-	FaultSafeCacheLock lock(this, m_mutex);
-	for (const auto& [address, cached]: m_buffers) {
-		if (PageOverlaps(vaddr, size, address, cached->size)) {
-			return true;
-		}
+	// Cached buffers are ordered and non-overlapping. The last buffer beginning before the query
+	// end is therefore the only possible intersection.
+	const auto candidate = m_buffers.lower_bound(vaddr + size);
+	if (candidate == m_buffers.begin()) {
+		return false;
 	}
-	return false;
+	const auto& [address, cached] = *std::prev(candidate);
+	return address + cached->size > vaddr;
 }
 
 bool BufferCache::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
@@ -829,7 +801,6 @@ bool BufferCache::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 }
 
 bool BufferCache::HasGpuDirtyBytes(uint64_t vaddr, uint64_t size) {
-	FaultSafeCacheLock lock(this, m_mutex);
 	return !m_gpu_modified_ranges.Intersections(vaddr, size).empty();
 }
 
@@ -838,8 +809,7 @@ bool BufferCache::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 }
 
 void BufferCache::RunGarbageCollector() {
-	std::lock_guard transaction(m_resource_mutex);
-	const auto      tick = m_gc_tick++;
+	const auto tick = m_gc_tick++;
 	if (m_graphics.CanReportMemoryUsage()) {
 		m_total_used_memory = m_graphics.GetDeviceMemoryUsage();
 	}
@@ -854,7 +824,6 @@ void BufferCache::RunGarbageCollector() {
 	std::vector<RetiredBuffer>                                       retires;
 	std::vector<std::pair<RetiredBuffer, std::vector<DownloadCopy>>> dirty_retires;
 	{
-		FaultSafeCacheLock    lock(this, m_mutex);
 		std::vector<uint64_t> candidates;
 		for (const auto& [address, owner]: m_buffers) {
 			const auto& cached = *owner;
@@ -902,7 +871,6 @@ void BufferCache::RunGarbageCollector() {
 		QueueGarbageDownload(copies, std::move(retire));
 	}
 
-	FaultSafeCacheLock lock(this, m_mutex);
 	for (const auto& retire: retires) {
 		auto found = m_buffers.find(retire.address);
 		if (found == m_buffers.end() || found->second->size != retire.size ||

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

@@ -3,7 +3,6 @@
 
 #include "common/abi.h"
 #include "common/common.h"
-#include "common/threads.h"
 #include "graphics/host_gpu/memoryTracker.h"
 #include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
@@ -20,7 +19,6 @@ struct GraphicContext;
 class CommandBuffer;
 class CommandScheduler;
 class TextureCache;
-class ResourceMutex;
 
 struct BufferBinding {
 	std::shared_ptr<void> owner;
@@ -41,12 +39,12 @@ public:
 	}
 
 	BufferCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
-	            TextureCache& texture_cache, ResourceMutex& resource_mutex);
+	            TextureCache& texture_cache);
 	~BufferCache();
 	KYTY_CLASS_NO_COPY(BufferCache);
 
 	void                        InvalidateMemory(uint64_t vaddr, uint64_t size);
-	void                        ReadMemory(uint64_t vaddr, uint64_t size);
+	void                        ReadMemory(uint64_t vaddr, uint64_t size, bool is_write = false);
 	void                        UnmapMemory(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] BufferBinding ObtainBuffer(CommandBuffer& command, uint64_t vaddr, uint64_t size,
 	                                         bool is_written = false, bool is_read = true,
@@ -61,7 +59,8 @@ public:
 	void FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds = false);
 	void CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t size, bool dst_gds = false,
 	                bool src_gds = false);
-	[[nodiscard]] bool HasPageOverlap(uint64_t vaddr, uint64_t size);
+	// Cache-index and exact dirty-range queries require GPU-thread serialization.
+	[[nodiscard]] bool IsRegionRegistered(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool HasGpuDirtyBytes(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool IsRegionCpuModified(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t vaddr, uint64_t size);
@@ -84,8 +83,6 @@ private:
 	[[nodiscard]] static constexpr uint64_t AlignDownload(uint64_t size) noexcept {
 		return (size + DOWNLOAD_ALIGNMENT - 1) & ~(DOWNLOAD_ALIGNMENT - 1);
 	}
-	[[nodiscard]] static bool PageOverlaps(uint64_t left, uint64_t left_size, uint64_t right,
-	                                       uint64_t right_size) noexcept;
 	[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
 	[[nodiscard]] static bool ResolveOverlap(CacheRange& merged, CacheRange candidate) noexcept;
 	void Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
@@ -97,11 +94,11 @@ private:
 	void PublishDownloads(std::span<const DownloadRange> downloads);
 	void QueueGarbageDownload(std::span<const DownloadCopy> copies, RetiredBuffer retire);
 	void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);
+	void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);
 
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;
 	Buffer                                            m_gds_buffer;
-	Common::Mutex                                     m_mutex;
 	std::shared_ptr<Buffer>                           m_null_buffer;
 	std::map<uint64_t, std::unique_ptr<CachedBuffer>> m_buffers;
 	RangeSet                                          m_gpu_modified_ranges;
@@ -111,7 +108,6 @@ private:
 	StreamBuffer                                      m_download_buffer;
 	StreamBuffer                                      m_device_buffer;
 	TextureCache&                                     m_texture_cache;
-	ResourceMutex&                                    m_resource_mutex;
 	uint64_t                                          m_total_used_memory = 0;
 	uint64_t m_trigger_gc_memory  = 1ull * 1024 * 1024 * 1024;
 	uint64_t m_critical_gc_memory = 2ull * 1024 * 1024 * 1024;

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

@@ -1,15 +1,13 @@
 #include "graphics/host_gpu/renderer/cache/gpuResourceManager.h"
 
 #include "common/assert.h"
-#include "graphics/guest_gpu/command_processor/commandProcessor.h"
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 namespace Libs::Graphics {
 
 GpuResourceManager::GpuResourceManager(GraphicContext& graphics, CommandScheduler& scheduler)
-    : m_scheduler(scheduler),
-      m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache, m_resource_mutex),
-      m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache, m_resource_mutex) {}
+    : m_scheduler(scheduler), m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache),
+      m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache) {}
 
 GpuResourceManager::~GpuResourceManager() = default;
 
@@ -18,69 +16,21 @@ bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vadd
 	if (!IsMapped(fault_vaddr, fault_size)) {
 		return false;
 	}
-	if (CommandScheduler::InDeferredOperation()) {
-		EXIT("unsupported guest-memory fault from an asynchronous GPU completion, "
-		     "addr=0x%016" PRIx64 " access=%u\n",
-		     fault_vaddr, static_cast<uint32_t>(access));
+	if (access == PageFaultAccess::Write) {
+		m_buffer_cache.InvalidateMemory(fault_vaddr, fault_size);
+		m_texture_cache.InvalidateMemory(fault_vaddr, fault_size);
+	} else {
+		m_buffer_cache.ReadMemory(fault_vaddr, fault_size);
 	}
-	bool       handled = false;
-	const auto resolve = [this, access, fault_vaddr, &handled](CommandProcessor& cp) {
-		cp.BeginReadbackTransaction();
-		{
-			ResourceMutex::FaultScope fault(m_resource_mutex);
-			if (access == PageFaultAccess::Write) {
-				m_buffer_cache.InvalidateMemory(fault_vaddr, fault_size);
-				m_texture_cache.InvalidateMemory(fault_vaddr, fault_size);
-			} else {
-				m_buffer_cache.ReadMemory(fault_vaddr, fault_size);
-			}
-			handled = true;
-		}
-		cp.EndReadbackTransaction();
-	};
-	if (auto* cp = Gpu::CurrentCommandProcessor(); cp != nullptr) {
-		resolve(*cp);
-		return handled;
-	}
-	if (m_resource_mutex.IsOwnedByCurrentThread()) {
-		EXIT("unsupported page fault from a pre-owned resource transaction, addr=0x%016" PRIx64
-		     " access=%u\n",
-		     fault_vaddr, static_cast<uint32_t>(access));
-	}
-	EXIT_IF(m_gpu == nullptr);
-	m_gpu->SendCommandSyncWithProcessor(resolve);
-	return handled;
+	return true;
 }
 
 bool GpuResourceManager::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 	if (!IsMapped(vaddr, size)) {
 		return false;
 	}
-	if (CommandScheduler::InDeferredOperation()) {
-		EXIT("unsupported memory invalidation from an asynchronous GPU completion, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
-	const auto resolve = [this, vaddr, size](CommandProcessor& cp) {
-		cp.BeginReadbackTransaction();
-		{
-			ResourceMutex::FaultScope fault(m_resource_mutex);
-			m_buffer_cache.InvalidateMemory(vaddr, size);
-			m_texture_cache.InvalidateMemory(vaddr, size);
-		}
-		cp.EndReadbackTransaction();
-	};
-	if (auto* cp = Gpu::CurrentCommandProcessor(); cp != nullptr) {
-		resolve(*cp);
-		return true;
-	}
-	if (m_resource_mutex.IsOwnedByCurrentThread()) {
-		EXIT("unsupported memory invalidation from a pre-owned resource transaction, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
-	EXIT_IF(m_gpu == nullptr);
-	m_gpu->SendCommandSyncWithProcessor(resolve);
+	m_buffer_cache.InvalidateMemory(vaddr, size);
+	m_texture_cache.InvalidateMemory(vaddr, size);
 	return true;
 }
 
@@ -107,11 +57,6 @@ void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 		     vaddr, size);
 	}
-	if (m_resource_mutex.IsOwnedByCurrentThread()) {
-		EXIT("unsupported memory unmap from a pre-owned resource transaction, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
 	const auto unmap = [this, vaddr, size] {
 		if (m_scheduler.Active()) {
 			const auto tick = m_scheduler.CurrentTick();

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

@@ -5,7 +5,6 @@
 #include "common/common.h"
 #include "graphics/host_gpu/pageManager.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 
 #include <cstdint>
@@ -35,7 +34,6 @@ public:
 
 private:
 	PageManager               m_page_manager;
-	ResourceMutex             m_resource_mutex;
 	CommandScheduler&         m_scheduler;
 	BufferCache               m_buffer_cache;
 	TextureCache              m_texture_cache;

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

@@ -1,93 +0,0 @@
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
-
-#include "common/assert.h"
-
-namespace Libs::Graphics {
-
-ResourceMutex::FaultScope::FaultScope(ResourceMutex& mutex): m_mutex(mutex) {
-	m_resource_preowned = m_mutex.BeginFault();
-}
-
-ResourceMutex::FaultScope::~FaultScope() {
-	m_mutex.EndFault(m_resource_preowned);
-}
-
-ResourceMutex::~ResourceMutex() {
-	std::lock_guard state(m_state);
-	if (m_resource_owner != std::thread::id {} || m_fault_owner != std::thread::id {}) {
-		EXIT("ResourceMutex destroyed with an active owner or fault transaction\n");
-	}
-}
-
-void ResourceMutex::lock() {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_resource_owner == current) {
-			EXIT("recursive resource transaction\n");
-		}
-		if (m_fault_owner == current) {
-			EXIT("resource transaction re-entered from a page-fault callback\n");
-		}
-	}
-	m_resource.lock();
-	std::lock_guard state(m_state);
-	if (m_resource_owner != std::thread::id {}) {
-		EXIT("resource mutex acquired with a stale owner\n");
-	}
-	m_resource_owner = current;
-}
-
-void ResourceMutex::unlock() {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_resource_owner != current) {
-			EXIT("resource transaction released without ownership\n");
-		}
-		m_resource_owner = {};
-	}
-	m_resource.unlock();
-}
-
-bool ResourceMutex::IsOwnedByCurrentThread() {
-	std::lock_guard state(m_state);
-	return m_resource_owner == std::this_thread::get_id();
-}
-
-bool ResourceMutex::BeginFault() {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_fault_owner == current) {
-			EXIT("nested resource page-fault transaction\n");
-		}
-		if (m_resource_owner == current) {
-			m_fault_owner = current;
-			return true;
-		}
-	}
-	lock();
-	std::lock_guard state(m_state);
-	if (m_resource_owner != current || m_fault_owner != std::thread::id {}) {
-		EXIT("resource fault transaction acquired inconsistent ownership\n");
-	}
-	m_fault_owner = current;
-	return false;
-}
-
-void ResourceMutex::EndFault(bool resource_preowned) {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_fault_owner != current || m_resource_owner != current) {
-			EXIT("resource page-fault transaction ended with inconsistent ownership\n");
-		}
-		m_fault_owner = {};
-	}
-	if (!resource_preowned) {
-		unlock();
-	}
-}
-
-} // namespace Libs::Graphics

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

@@ -1,48 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_
-
-#include "common/common.h"
-
-#include <mutex>
-#include <thread>
-
-namespace Libs::Graphics {
-
-// Owner-tracked shared buffer/image transaction. External faults pause GPU submissions first;
-// command-processor faults drain pending guest processors before entering this transaction.
-class ResourceMutex final {
-public:
-	class FaultScope final {
-	public:
-		explicit FaultScope(ResourceMutex& mutex);
-		~FaultScope();
-		KYTY_CLASS_NO_COPY(FaultScope);
-
-	private:
-		ResourceMutex& m_mutex;
-		bool           m_resource_preowned = false;
-	};
-
-	ResourceMutex() = default;
-	~ResourceMutex();
-	KYTY_CLASS_NO_COPY(ResourceMutex);
-
-	void               lock();
-	void               unlock();
-	[[nodiscard]] bool IsOwnedByCurrentThread();
-
-private:
-	friend class FaultScope;
-
-	[[nodiscard]] bool BeginFault();
-	void               EndFault(bool resource_preowned);
-
-	std::mutex      m_resource;
-	std::mutex      m_state;
-	std::thread::id m_resource_owner;
-	std::thread::id m_fault_owner;
-};
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_

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

@@ -8,7 +8,6 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -56,13 +55,12 @@ private:
 } // namespace
 
 TextureCache::TextureCache(GraphicContext& graphics, CommandScheduler& scheduler,
-                           PageManager& page_manager, BufferCache& buffer_cache,
-                           ResourceMutex& resource_mutex)
+                           PageManager& page_manager, BufferCache& buffer_cache)
     : m_graphics(graphics), m_scheduler(scheduler), m_page_manager(page_manager),
       m_blit_helper(graphics, scheduler),
       m_tiler(std::make_unique<TileManager>(graphics, scheduler,
                                             buffer_cache.GetUtilityBuffer(MemoryUsage::Stream))),
-      m_buffer_cache(buffer_cache), m_resource_mutex(resource_mutex),
+      m_buffer_cache(buffer_cache),
       m_readback_linear_images(Config::ReadbackLinearImagesEnabled()) {
 	if (m_graphics.CanReportMemoryUsage()) {
 		constexpr int64_t GiB = 1024ll * 1024 * 1024;
@@ -434,9 +432,8 @@ void TextureCache::UntrackImageTail(ImageId id) {
 }
 
 void TextureCache::TrackImageDownload(ImageId id) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	TrackImageDownloadLocked(id, image);
 }
 
@@ -911,7 +908,6 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 	uint32_t    format           = info.guest_format;
 	uint32_t    layers           = info.TransferLayers();
 	bool        volume           = info.IsVolume();
-	bool        layered          = info.IsLayered();
 	bool        allow_depth_tile = direction == TransferDirection::Upload;
 	const char* owner =
 	    direction == TransferDirection::Upload ? "TextureCache" : "TextureCache readback";
@@ -940,7 +936,6 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 				format           = info.guest_format;
 				layers           = info.resources.layers;
 				volume           = false;
-				layered          = layers > 1;
 				allow_depth_tile = false;
 				plan.swap_bgra16 = info.bgra16;
 				owner            = "VideoOut";
@@ -958,19 +953,14 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 		plan.swap_bgra16 = info.bgra16;
 	}
 
-	plan.layout = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
-	                                      info.resources.levels, layers, info.pitch, info.tile_mode,
-	                                      info.data.size, allow_depth_tile, volume, owner);
-	plan.regions = TextureBuildImageCopies(plan.layout, info.extent.width, info.extent.height,
-	                                       layers, info.resources.levels, layered, volume);
-	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.tile) != Prospero::TileMode::kLinear;
+	plan.layout  = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
+	                                       info.resources.levels, layers, info.tile_mode,
+	                                       info.data.size, allow_depth_tile, volume, owner);
+	plan.regions = TextureBuildImageCopies(plan.layout);
+	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.surface.description.tile_mode) !=
+	               Prospero::TileMode::kLinear;
 	if (plan.tiled) {
-		// CMASK/FMASK decoding is intentionally deferred.
-		if (plan.layout.tile_family == TileBlockFamily::Depth64KB &&
-		    Prospero::IsFmaskTextureFormat(format)) {
-			return plan;
-		}
-		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout, format, layers,
+		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout,
 		                              info.resources.levels, plan.tiles)) {
 			return plan;
 		}
@@ -1017,9 +1007,10 @@ void TextureCache::UploadImage(Image& image, const ImageDesc& desc, Buffer& sour
 			     " size=0x%016" PRIx64 " format=%u tile=%u family=%u extent=%ux%ux%u "
 			     "pitch=%u levels=%u layers=%u samples=%u\n",
 			     static_cast<uint32_t>(desc.type), info.data.address, info.data.size,
-			     info.guest_format, info.tile_mode, static_cast<uint32_t>(plan.layout.tile_family),
-			     info.extent.width, info.extent.height, info.extent.depth, info.pitch,
-			     info.resources.levels, info.resources.layers, info.samples);
+			     info.guest_format, info.tile_mode,
+			     static_cast<uint32_t>(plan.layout.surface.texture.block.family), info.extent.width,
+			     info.extent.height, info.extent.depth, info.pitch, info.resources.levels,
+			     info.resources.layers, info.samples);
 		}
 		TileManager::Result linear {source.Handle(), source_offset, info.data.size};
 		if (plan.tiled) {
@@ -1152,8 +1143,7 @@ void TextureCache::RefreshImage(ImageId id, const ImageDesc& desc) {
 }
 
 void TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
+	CacheLock lock(*this, m_lock);
 	AssociateStencilLocked(depth_id, stencil);
 }
 
@@ -1196,11 +1186,9 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	}
 
 	ImageId result {};
-	bool    inserted_new = false;
 	{
-		std::lock_guard transaction(m_resource_mutex);
-		CacheLock       lock(*this, m_lock);
-		const auto      candidates =
+		CacheLock  lock(*this, m_lock);
+		const auto candidates =
 		    FindImagesInRegion(desc.info.data.address, desc.info.data.size, false);
 
 		for (const auto id: candidates) {
@@ -1243,19 +1231,12 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		}
 		if (!result) {
 			result         = InsertImage(desc.info);
-			inserted_new   = true;
 			auto& inserted = ResolveImage(result);
 			if (m_buffer_cache.HasGpuDirtyBytes(inserted.info.data.address,
 			                                    inserted.info.data.size)) {
 				inserted.MarkBufferModified();
 			}
 		}
-		if (inserted_new) {
-			InitializeImage(result, desc);
-		} else {
-			RefreshImage(result, desc);
-		}
-
 		auto& image = ResolveImage(result);
 		if (desc.type == BindingType::VideoOut &&
 		    desc.info.metadata.compression != VideoOutCompression::Uncompressed) {
@@ -1280,6 +1261,13 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	return result;
 }
 
+void TextureCache::UpdateImage(ImageId id) {
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
+	TouchImage(image);
+	RefreshImage(id, ImageDesc {.info = image.info, .type = UploadBinding(image)});
+}
+
 ImageId TextureCache::FindImageFromRange(uint64_t address, uint64_t size, bool ensure_valid) {
 	if (!GuestRange {address, size}.Valid()) {
 		return {};
@@ -1324,22 +1312,24 @@ ImageId TextureCache::FindImageFromRange(uint64_t address, uint64_t size, bool e
 }
 
 vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	TouchImage(image);
 	if (!image.info.data.Empty()) {
 		if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 			EXIT("TextureCache: texture requires rediscovery before final acquisition\n");
 		}
+	}
+	if (desc.type == BindingType::Storage) {
+		image.MarkGpuModified();
+	}
+	if (!image.info.data.Empty()) {
 		RefreshImage(id, desc);
 	}
 	switch (desc.type) {
 		case BindingType::Texture: break;
 		case BindingType::Storage:
-			if (image.info.data.Empty()) {
-				image.MarkGpuModified();
-			} else {
+			if (!image.info.data.Empty()) {
 				if (!image.registered || image.depth_id) {
 					EXIT("TextureCache: cannot acquire an unavailable storage image\n");
 				}
@@ -1358,16 +1348,16 @@ vk::ImageView TextureCache::FindRenderTarget(ImageId id, const ImageDesc& desc)
 	if (desc.type != BindingType::RenderTarget) {
 		EXIT("TextureCache: invalid color-target binding\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 		EXIT("TextureCache: color target requires rediscovery before final acquisition\n");
 	}
 	TouchImage(image);
+	image.MarkGpuModified();
+	image.usage.render_target = true;
 	RefreshImage(id, desc);
 	CommitGpuWrite(image);
-	image.usage.render_target = true;
 	TrackImageDownloadLocked(id, image);
 	const auto view = image.FindView(desc.view_info);
 	RetainImage(m_scheduler.Current(), id);
@@ -1378,13 +1368,14 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	if (desc.type != BindingType::DepthTarget) {
 		EXIT("TextureCache: invalid depth-target binding\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 		EXIT("TextureCache: depth target requires rediscovery before final acquisition\n");
 	}
 	TouchImage(image);
+	image.MarkGpuModified();
+	image.usage.depth_target = true;
 	RefreshImage(id, desc);
 	if (desc.info.HasMetadata()) {
 		image.info.metadata = desc.info.metadata;
@@ -1392,7 +1383,6 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 		                            MetaDataInfo {.clear_mask = image.info.htile_clear_mask});
 	}
 	CommitGpuWrite(image);
-	image.usage.depth_target = true;
 	if (desc.info.HasStencil()) {
 		AssociateStencilLocked(id, desc.info.stencil);
 	}
@@ -1402,9 +1392,8 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 }
 
 void TextureCache::MarkGpuWritten(ImageId id) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id) {
 		EXIT("TextureCache: cannot mark an unavailable image GPU-written\n");
 	}
@@ -1428,7 +1417,6 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 	if (command.IsInvalid() || !GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid image clear\n");
 	}
-	std::lock_guard      transaction(m_resource_mutex);
 	CacheLock            lock(*this, m_lock);
 	ImageId              selected {};
 	vk::ImageAspectFlags aspect {};
@@ -1713,14 +1701,9 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uin
 			return false;
 		}
 		if (color.tiled) {
-			const auto binding = m_texture_cache.UploadBinding(image);
-			const auto format =
-			    binding == TextureCache::BindingType::RenderTarget
-			        ? ImageOps::RenderTargetTransferFormat(image.info.bytes_per_block)
-			        : image.info.guest_format;
 			color.tiles.clear();
-			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, format,
-			                              image.info.TransferLayers(), levels, color.tiles)) {
+			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, levels,
+			                              color.tiles)) {
 				return false;
 			}
 			color.linear_size = GetLinearSize(color.tiles);
@@ -1891,9 +1874,8 @@ bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice) {
 }
 
 bool TextureCache::ClearMeta(uint64_t address) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	const auto      found = m_surface_metas.find(address);
+	CacheLock  lock(*this, m_lock);
+	const auto found = m_surface_metas.find(address);
 	if (found == m_surface_metas.end()) {
 		return false;
 	}
@@ -1919,9 +1901,8 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 	if (!GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid unmap range\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto            images = FindImagesInRegion(address, size, false);
+	CacheLock lock(*this, m_lock);
+	auto      images = FindImagesInRegion(address, size, false);
 	if (!images.empty()) {
 		m_scheduler.Finish();
 	}
@@ -1938,9 +1919,8 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 }
 
 void TextureCache::RunGarbageCollector() {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	const uint64_t  tick = m_gc_tick++;
+	CacheLock      lock(*this, m_lock);
+	const uint64_t tick = m_gc_tick++;
 	if (m_graphics.CanReportMemoryUsage()) {
 		m_total_used_memory = m_graphics.GetDeviceMemoryUsage();
 	}
@@ -2000,8 +1980,7 @@ void TextureCache::RunGarbageCollector() {
 }
 
 void TextureCache::ProcessDownloadImages() {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
+	CacheLock lock(*this, m_lock);
 	for (const auto id: m_download_images) {
 		const auto owner = ResolveOwner(id);
 		if (owner != nullptr && owner->registered && owner->IsGpuModified()) {

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

@@ -26,7 +26,6 @@ class Buffer;
 class BufferCache;
 class CommandBuffer;
 class CommandScheduler;
-class ResourceMutex;
 class RenderExecutor;
 class StreamBuffer;
 class TileManager;
@@ -49,11 +48,12 @@ public:
 	};
 
 	TextureCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
-	             BufferCache& buffer_cache, ResourceMutex& resource_mutex);
+	             BufferCache& buffer_cache);
 	~TextureCache();
 	KYTY_CLASS_NO_COPY(TextureCache);
 
 	[[nodiscard]] ImageId       FindImage(ImageDesc& desc, bool exact_format = false);
+	void                        UpdateImage(ImageId id);
 	[[nodiscard]] ImageId       FindImageFromRange(uint64_t address, uint64_t size,
 	                                               bool ensure_valid = true);
 	[[nodiscard]] vk::ImageView FindTexture(ImageId id, const ImageDesc& desc);
@@ -168,7 +168,6 @@ private:
 	BlitHelper                                        m_blit_helper;
 	std::unique_ptr<TileManager>                      m_tiler;
 	BufferCache&                                      m_buffer_cache;
-	ResourceMutex&                                    m_resource_mutex;
 	std::vector<Slot>                                 m_slots;
 	std::vector<uint32_t>                             m_free_slots;
 	ImagePageTable                                    m_image_page_table;

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -178,10 +178,10 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		pitch = (rt.pitch.pitch_div8_minus1 + 1u) << 3u;
 	} else if (tile) {
 		if (volume) {
-			pitch = TileGetTexturePitch(transfer_format, width, levels, rt.attrib3.tile_mode);
+			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
 		} else if (standard64) {
 			pitch = TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
-			                            width, levels, rt.attrib3.tile_mode);
+			                            width, rt.attrib3.tile_mode);
 		} else {
 			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
 		}
@@ -192,13 +192,20 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		pitch = width;
 	}
 
-	TileSizeOffset   mip_sizes[16] {};
-	TilePaddedSize   mip_padded[16] {};
-	TileVolumeLayout volume_layout {};
-	uint64_t         backing_size = 0;
+	TileSizeOffset    mip_sizes[16] {};
+	TilePaddedSize    mip_padded[16] {};
+	TileSurfaceLayout volume_layout {};
+	uint64_t          backing_size = 0;
 	if (volume) {
-		if (!tile || !TileGetTextureVolumeLayout(transfer_format, width, height, depth, levels,
-		                                         rt.attrib3.tile_mode, volume_layout)) {
+		const TileSurfaceDescription description {transfer_format,
+		                                          rt.attrib3.tile_mode,
+		                                          TileSurfaceDimension::Dim3D,
+		                                          width,
+		                                          height,
+		                                          depth,
+		                                          levels,
+		                                          1};
+		if (!tile || !TileGetTiledTextureLayout(description, volume_layout)) {
 			EXIT("unsupported 3D render-target layout: %ux%ux%u levels=%u tile=%u\n", width, height,
 			     depth, levels, rt.attrib3.tile_mode);
 		}
@@ -209,7 +216,7 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		bool          valid_layout = false;
 		if (standard64) {
 			TileGetTextureSize(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float), width,
-			                   height, pitch, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
+			                   height, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
 			                   mip_padded);
 			valid_layout = layout.size != 0 && layout.align == 65536;
 		} else {
@@ -289,11 +296,12 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	desc.info.tile_mode       = rt.attrib3.tile_mode;
 	for (uint32_t level = 0; level < levels; level++) {
 		if (volume) {
+			const auto& mip             = volume_layout.mips[level];
 			desc.info.mip_layout[level] = {
-			    volume_layout.level_offsets[level],
-			    volume_layout.level_sizes[level],
-			    volume_layout.level_widths[level],
-			    volume_layout.level_heights[level],
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
 			};
 			continue;
 		}

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -99,7 +99,10 @@ bool CommandScheduler::InDeferredOperation() noexcept {
 
 CommandScheduler::CommandScheduler(RenderContext& context, GraphicContext& graphics)
     : m_master(graphics), m_context(context), m_graphics(graphics),
-      m_priority_thread([this](std::stop_token stop) { PriorityOperationsThread(stop); }) {}
+      m_priority_thread([this](std::stop_token stop) { PriorityOperationsThread(stop); }) {
+	m_buffers.reserve(CommandBufferGrowStep);
+	m_buffer_ticks.reserve(CommandBufferGrowStep);
+}
 
 CommandScheduler::~CommandScheduler() {
 	Shutdown();
@@ -154,10 +157,7 @@ void CommandScheduler::Begin(HW::Context& registers, HW::UserConfig& user_config
 	m_shaders     = &shaders;
 
 	if (!Active()) {
-		for (auto& buffer: m_buffers) {
-			buffer = std::make_unique<RenderCommandBuffer>(*this);
-		}
-		m_current = 0;
+		m_current = static_cast<int>(GrowCommandBuffers());
 	}
 
 	BindCurrent();
@@ -235,7 +235,13 @@ void CommandScheduler::Wait(uint64_t tick) {
 }
 
 void CommandScheduler::PopPendingOperations() {
-	m_master.Refresh();
+	PopPendingOperations(true);
+}
+
+void CommandScheduler::PopPendingOperations(bool refresh_gpu_tick) {
+	if (refresh_gpu_tick) {
+		m_master.Refresh();
+	}
 	for (;;) {
 		Common::UniqueFunction<void> callback;
 		{
@@ -362,7 +368,7 @@ uint64_t CommandScheduler::NextSubmitSequence() noexcept {
 }
 
 void CommandScheduler::CheckActive() const {
-	EXIT_IF(!Active() || m_current >= BufferCount);
+	EXIT_IF(!Active() || static_cast<size_t>(m_current) >= m_buffers.size());
 }
 
 RenderCommandBuffer& CommandScheduler::Current() const {
@@ -384,15 +390,52 @@ CommandBuffer& CommandScheduler::SubmitCurrent(SubmitInfo& submit) {
 	const auto signal_tick = m_master.NextTick();
 	submit.AddSignal(m_master.Handle(), signal_tick);
 	submitted.Execute(submit);
-	m_recording = false;
+	m_buffer_ticks[static_cast<size_t>(m_current)] = signal_tick;
+	m_recording                                    = false;
 	return submitted;
 }
 
+int CommandScheduler::FindReusableBuffer(uint64_t gpu_tick) const {
+	EXIT_IF(m_buffer_ticks.size() != m_buffers.size());
+	const auto buffer_count = m_buffers.size();
+	for (size_t offset = 1; offset <= buffer_count; ++offset) {
+		const auto candidate = (static_cast<size_t>(m_current) + offset) % buffer_count;
+		if (gpu_tick >= m_buffer_ticks[candidate]) {
+			return static_cast<int>(candidate);
+		}
+	}
+	return -1;
+}
+
+size_t CommandScheduler::GrowCommandBuffers() {
+	const auto first = m_buffers.size();
+	const auto end   = first + CommandBufferGrowStep;
+	m_buffers.reserve(end);
+	m_buffer_ticks.reserve(end);
+	for (size_t i = first; i < end; ++i) {
+		m_buffers.emplace_back(std::make_unique<RenderCommandBuffer>(*this));
+		m_buffer_ticks.push_back(0);
+	}
+	return first;
+}
+
 void CommandScheduler::BeginNext() {
 	EXIT_IF(m_recording);
-	m_current = (m_current + 1) % BufferCount;
+
+	auto candidate = FindReusableBuffer(m_master.KnownGpuTick());
+	bool refreshed = false;
+	if (candidate < 0) {
+		m_master.Refresh();
+		refreshed = true;
+		candidate = FindReusableBuffer(m_master.KnownGpuTick());
+	}
+	if (candidate < 0) {
+		candidate = static_cast<int>(GrowCommandBuffers());
+	}
+
+	m_current = candidate;
 	Current().WaitForFenceAndReset();
-	PopPendingOperations();
+	PopPendingOperations(!refreshed);
 	BindCurrent();
 	Current().Begin();
 	m_recording = true;

src/graphics/host_gpu/renderer/commandScheduler.h

@@ -16,6 +16,7 @@
 #include <queue>
 
 #include <thread>
+#include <vector>
 
 namespace Libs::Graphics {
 
@@ -31,8 +32,6 @@ struct CommandSlot {
 
 class CommandScheduler {
 public:
-	static constexpr int BufferCount = 8;
-
 	CommandScheduler(RenderContext& context, GraphicContext& graphics);
 	~CommandScheduler();
 	KYTY_CLASS_NO_COPY(CommandScheduler);
@@ -65,6 +64,8 @@ public:
 	[[nodiscard]] GraphicContext& Graphics() const noexcept { return m_graphics; }
 
 private:
+	static constexpr size_t CommandBufferGrowStep = 4;
+
 	class CommandPool {
 	public:
 		CommandPool() = default;
@@ -94,30 +95,34 @@ private:
 	void                       BindCurrent() const;
 	CommandBuffer&             SubmitCurrent(SubmitInfo& submit);
 	void                       BeginNext();
+	void                       PopPendingOperations(bool refresh_gpu_tick);
+	[[nodiscard]] int          FindReusableBuffer(uint64_t gpu_tick) const;
+	[[nodiscard]] size_t       GrowCommandBuffers();
 	void                       PriorityOperationsThread(std::stop_token stop);
 	void                       RunOperation(Common::UniqueFunction<void>&& operation);
 	[[nodiscard]] CommandSlot* AllocateCommandBuffer();
 	[[nodiscard]] uint64_t     NextSubmitSequence() noexcept;
 
-	MasterSemaphore                                               m_master;
-	RenderContext&                                                m_context;
-	GraphicContext&                                               m_graphics;
-	CommandPool                                                   m_command_pool;
-	std::array<std::unique_ptr<RenderCommandBuffer>, BufferCount> m_buffers;
-	std::queue<PendingOperation>                                  m_pending_operations;
-	std::queue<PendingOperation>                                  m_priority_operations;
-	std::mutex                                                    m_operation_mutex;
-	std::condition_variable                                       m_operation_available;
-	std::jthread                                                  m_priority_thread;
-	bool                                                          m_priority_active      = false;
-	uint64_t                                                      m_priority_active_tick = 0;
-	OperationState        m_operation_state = OperationState::Open;
-	int                   m_current         = -1;
-	bool                  m_recording       = false;
-	HW::Context*          m_registers       = nullptr;
-	HW::UserConfig*       m_user_config     = nullptr;
-	HW::Shader*           m_shaders         = nullptr;
-	std::atomic<uint64_t> m_submit_sequence = 0;
+	MasterSemaphore                                   m_master;
+	RenderContext&                                    m_context;
+	GraphicContext&                                   m_graphics;
+	CommandPool                                       m_command_pool;
+	std::vector<std::unique_ptr<RenderCommandBuffer>> m_buffers;
+	std::vector<uint64_t>                             m_buffer_ticks;
+	std::queue<PendingOperation>                      m_pending_operations;
+	std::queue<PendingOperation>                      m_priority_operations;
+	std::mutex                                        m_operation_mutex;
+	std::condition_variable                           m_operation_available;
+	std::jthread                                      m_priority_thread;
+	bool                                              m_priority_active      = false;
+	uint64_t                                          m_priority_active_tick = 0;
+	OperationState                                    m_operation_state      = OperationState::Open;
+	int                                               m_current              = -1;
+	bool                                              m_recording            = false;
+	HW::Context*                                      m_registers            = nullptr;
+	HW::UserConfig*                                   m_user_config          = nullptr;
+	HW::Shader*                                       m_shaders              = nullptr;
+	std::atomic<uint64_t>                             m_submit_sequence      = 0;
 
 	friend class CommandBuffer;
 };

src/graphics/host_gpu/renderer/debug.cpp

@@ -10,6 +10,7 @@
 #include "graphics/host_gpu/renderer/render.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cmath>
 #include <fmt/format.h>
@@ -686,14 +687,6 @@ static void McCheck(const HW::ModeControl& c) {
 	}
 	EXIT_NOT_IMPLEMENTED(c.polymode_front_ptype != 0 && c.polymode_front_ptype != 2);
 	EXIT_NOT_IMPLEMENTED(c.polymode_back_ptype != 0 && c.polymode_back_ptype != 2);
-	if (c.poly_offset_front_enable || c.poly_offset_back_enable) {
-		static bool logged = false;
-		if (!logged) {
-			LOGF("\t temporary: PA_SU_SC_MODE_CNTL.POLY_OFFSET_*_ENABLE is not implemented; "
-			     "continuing without depth bias\n");
-			logged = true;
-		}
-	}
 	if (c.vtx_window_offset_enable) {
 		static bool logged = false;
 		if (!logged) {
@@ -1067,18 +1060,24 @@ static ScissorRect ScissorRectClamp(ScissorRect r, uint32_t width, uint32_t heig
 	return r;
 }
 
+static constexpr std::array<uint16_t, 16> MakeScissorIntersectionRules() {
+	std::array<uint16_t, 16> rules {};
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		for (uint32_t combination = 0; combination < 16; combination++) {
+			if ((combination & candidate) == candidate) {
+				rules[candidate] |= static_cast<uint16_t>(1u << combination);
+			}
+		}
+	}
+	return rules;
+}
+
 static bool ScissorClipRuleToIntersectionMask(uint16_t rule, uint8_t* mask) {
 	EXIT_IF(mask == nullptr);
 
-	for (uint32_t candidate = 0; candidate < 16; candidate++) {
-		uint16_t expected = 0;
-		for (uint32_t combination = 0; combination < 16; combination++) {
-			if ((combination & candidate) == candidate) {
-				expected |= static_cast<uint16_t>(1u << combination);
-			}
-		}
-
-		if (expected == rule) {
+	static constexpr auto rules = MakeScissorIntersectionRules();
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		if (rules[candidate] == rule) {
 			*mask = static_cast<uint8_t>(candidate);
 			return true;
 		}

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -38,7 +38,10 @@ namespace Libs::Graphics {
 	EXIT("unsupported render state; details were printed above\n");
 }
 
-static vk::StencilOp ConvertStencilOp(uint8_t value) {
+static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t op_value) {
+	if (write_mask == 0) {
+		return vk::StencilOp::eKeep;
+	}
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
@@ -49,7 +52,17 @@ static vk::StencilOp ConvertStencilOp(uint8_t value) {
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
 		case Prospero::StencilOp::kAddWrap: return vk::StencilOp::eIncrementAndWrap;
 		case Prospero::StencilOp::kSubWrap: return vk::StencilOp::eDecrementAndWrap;
-		default: DepthFatal("unsupported stencil operation");
+		case Prospero::StencilOp::kXor:
+			if ((write_mask & op_value) == 0) {
+				return vk::StencilOp::eKeep;
+			}
+			if ((write_mask & ~op_value) != 0) {
+				DepthFatal("unsupported stencil XOR operands: write mask=0x%02" PRIx8
+				           ", operation value=0x%02" PRIx8,
+				           write_mask, op_value);
+			}
+			return vk::StencilOp::eInvert;
+		default: DepthFatal("unsupported stencil operation: 0x%02" PRIx8, value);
 	}
 }
 
@@ -192,8 +205,6 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		if (z.depth_view.slice_max >= 32) {
 			DepthFatal("HTile clear tracking supports at most 32 slices");
 		}
-	} else if (z.htile_data_base_addr != 0) {
-		DepthFatal("HTile address without an enabled tile surface");
 	}
 	const bool size_xy_valid = z.size.valid;
 	const bool wh_valid      = z.width_height_valid && z.width != 0 && z.height != 0;
@@ -283,6 +294,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	r.stencil_clear_value  = hw.GetStencilClearValue();
 	r.stencil_test_enable  = has_stencil && dc.stencil_enable;
 	if (r.stencil_test_enable) {
+		const uint8_t front_write_mask = rc.stencil_clear_enable ? 0 : sm.stencil_writemask;
+		const uint8_t back_write_mask  = rc.stencil_clear_enable ? 0 : sm.stencil_writemask_bf;
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
 		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
@@ -294,19 +307,18 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 			DepthFatal("unsupported stencil compare or replacement state");
 		}
 		r.stencil_static_front = {
-		    ConvertStencilOp(sc.stencil_fail), ConvertStencilOp(sc.stencil_zpass),
-		    ConvertStencilOp(sc.stencil_zfail), static_cast<vk::CompareOp>(dc.stencilfunc)};
-		r.stencil_dynamic_front = {sm.stencil_mask,
-		                           rc.stencil_clear_enable ? 0u : sm.stencil_writemask,
-		                           sm.stencil_testval};
+		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
+		    static_cast<vk::CompareOp>(dc.stencilfunc)};
+		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
 		if (dc.backface_enable) {
-			r.stencil_static_back  = {ConvertStencilOp(sc.stencil_fail_bf),
-			                          ConvertStencilOp(sc.stencil_zpass_bf),
-			                          ConvertStencilOp(sc.stencil_zfail_bf),
-			                          static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
-			r.stencil_dynamic_back = {sm.stencil_mask_bf,
-			                          rc.stencil_clear_enable ? 0u : sm.stencil_writemask_bf,
-			                          sm.stencil_testval_bf};
+			r.stencil_static_back = {
+			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
+			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
+			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
 		} else {
 			r.stencil_static_back  = r.stencil_static_front;
 			r.stencil_dynamic_back = r.stencil_dynamic_front;

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

@@ -663,7 +663,8 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
     : info(image_info), m_graphics(&graphics), m_scheduler(&scheduler) {
 	KYTY_PROFILER_FUNCTION();
 	ImageOps::Validate(info);
-	m_cpu_dirty = !info.data.Empty();
+	m_cpu_dirty =
+	    !info.data.Empty() && info.metadata.compression == VideoOutCompression::Uncompressed;
 	if (info.pixel_format == vk::Format::eUndefined) {
 		return;
 	}

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

@@ -22,6 +22,10 @@ struct RenderTargetFormatMapping {
 };
 
 constexpr RenderTargetFormatMapping kRenderTargetFormats[] = {
+    {Prospero::ChannelLayout::k8,
+     Prospero::ChannelType::kUInt,
+     Prospero::ChannelOrder::kStandard,
+     {vk::Format::eR8Uint, 1}},
     {Prospero::ChannelLayout::k8_8,
      Prospero::ChannelType::kUNorm,
      Prospero::ChannelOrder::kStandard,
@@ -164,14 +168,6 @@ RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_
 
 namespace {
 
-static uint64_t GetLevelSrcOffset(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_offset : level_size.offset);
-}
-
-static uint64_t GetLevelSrcSize(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_size : level_size.size);
-}
-
 static uint32_t GetTextureLevelDepth(uint32_t depth, uint32_t level, bool volume_texture) {
 	return volume_texture ? std::max(depth >> level, 1u) : depth;
 }
@@ -184,20 +180,6 @@ static size_t GetTextureRegionCount(uint32_t depth, uint64_t levels, bool volume
 	return count;
 }
 
-uint64_t TextureUploadSliceSourceOffset(const TextureUploadLayout& layout, uint32_t level,
-                                        uint32_t slice) {
-	if (level >= 16 || layout.level_sizes[level].size == 0) {
-		EXIT("invalid texture upload slice source, level=%u slice=%u\n", level, slice);
-	}
-	const auto level_offset = GetLevelSrcOffset(layout.level_sizes[level]);
-	const auto slice_stride =
-	    layout.source_slice_stride != 0 ? layout.source_slice_stride : layout.slice_stride;
-	if (slice_stride != 0 && slice > (UINT64_MAX - level_offset) / slice_stride) {
-		EXIT("texture upload slice source offset overflow, level=%u slice=%u\n", level, slice);
-	}
-	return level_offset + static_cast<uint64_t>(slice) * slice_stride;
-}
-
 vk::ComponentSwizzle TextureGetComponentSwizzle(uint8_t s) {
 	switch (static_cast<Prospero::CompSwizzle>(s)) {
 		case Prospero::CompSwizzle::kZero: return vk::ComponentSwizzle::eZero;
@@ -237,12 +219,11 @@ vk::Format TextureGetFormat(uint32_t fmt) {
 
 namespace {
 
-uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t levels,
+uint64_t CalcTextureSliceStride(const TextureUploadMipLayout* mips, uint64_t levels,
                                 uint64_t total_size, uint32_t depth) {
 	uint64_t stride = 0;
 	for (uint32_t i = 0; i < levels; i++) {
-		stride =
-		    std::max(stride, static_cast<uint64_t>(level_sizes[i].offset) + level_sizes[i].size);
+		stride = std::max(stride, mips[i].offset + mips[i].size);
 	}
 
 	if (depth > 1 && total_size != 0 && total_size % depth == 0) {
@@ -255,39 +236,26 @@ uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t leve
 	return stride;
 }
 
-uint64_t CalcLinearUploadLevelSize(uint32_t fmt, uint32_t pitch, uint32_t height) {
-	if (const uint32_t bytes_per_element = Prospero::NumBytesPerElement(fmt);
-	    bytes_per_element != 0) {
-		return static_cast<uint64_t>(pitch) * height * bytes_per_element;
-	}
-
-	if (const uint32_t bytes_per_block = Prospero::BlockCompressedBytesPerBlock(fmt);
-	    bytes_per_block != 0) {
-		const uint32_t blocks_w = std::max((pitch + 3u) / 4u, 1u);
-		const uint32_t blocks_h = std::max((height + 3u) / 4u, 1u);
-		return static_cast<uint64_t>(blocks_w) * blocks_h * bytes_per_block;
-	}
-
-	return 0;
+uint64_t CalcLinearUploadLevelSize(const TileTextureElementLayout& element, uint32_t pitch,
+                                   uint32_t height) {
+	const uint32_t elements_w =
+	    std::max((pitch + element.texel_width - 1u) / element.texel_width, 1u);
+	const uint32_t elements_h =
+	    std::max((height + element.texel_height - 1u) / element.texel_height, 1u);
+	return static_cast<uint64_t>(elements_w) * elements_h * element.bytes;
 }
 
-uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64_t height,
+uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
+                               const TileTextureElementLayout& element, uint64_t height,
                                uint64_t levels, uint32_t base_pitch) {
 	uint64_t offset = 0;
 	auto     pitch  = base_pitch;
 	auto     h      = static_cast<uint32_t>(height);
 
 	for (uint32_t i = 0; i < levels; i++) {
-		const auto size = CalcLinearUploadLevelSize(fmt, pitch, h);
-		EXIT_NOT_IMPLEMENTED(size > 0xffffffffull);
-		EXIT_NOT_IMPLEMENTED(offset > 0xffffffffull);
-
-		level_sizes[i].size       = static_cast<uint32_t>(size);
-		level_sizes[i].offset     = static_cast<uint32_t>(offset);
-		level_sizes[i].src_size   = 0;
-		level_sizes[i].src_offset = 0;
-		level_sizes[i].x          = 0;
-		level_sizes[i].y          = 0;
+		const auto size = CalcLinearUploadLevelSize(element, pitch, h);
+		mips[i].size    = size;
+		mips[i].offset  = offset;
 
 		offset += size;
 		if (pitch > 1) {
@@ -303,141 +271,123 @@ uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64
 
 } // namespace
 
-TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64_t height,
-                                            uint64_t levels, uint32_t depth, uint64_t pitch,
-                                            uint64_t tile, uint64_t upload_size,
-                                            bool allow_depth_tile, bool volume_texture,
-                                            const char* owner) {
+TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
+                                            uint32_t levels, uint32_t depth, uint32_t tile,
+                                            uint64_t upload_size, bool allow_depth_tile,
+                                            bool volume_texture, const char* owner) {
 	TextureUploadLayout layout {};
-	layout.tile           = static_cast<uint32_t>(tile);
-	layout.pitch          = static_cast<uint32_t>(pitch);
-	layout.texel_block    = Prospero::BlockCompressedBytesPerBlock(fmt) != 0 ? 4u : 1u;
-	layout.volume_texture = volume_texture;
+	layout.surface.description = {
+	    fmt,
+	    tile,
+	    volume_texture ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
+	    width,
+	    height,
+	    volume_texture ? depth : 1u,
+	    levels,
+	    volume_texture ? 1u : depth,
+	};
+	const auto&              description = layout.surface.description;
+	const auto               tile_mode   = static_cast<Prospero::TileMode>(description.tile_mode);
+	TileTextureElementLayout element {};
 
-	if (fmt != 0) {
-		if (layout.tile != 0) {
-			const auto tile_mode = static_cast<Prospero::TileMode>(layout.tile);
-			switch (tile_mode) {
-				case Prospero::TileMode::kStandard256B:
-					if (TileIsStandard256BTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard256B;
-					}
-					break;
-				case Prospero::TileMode::kStandard4KB:
-					if (TileIsStandard4KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard4KB;
-					}
-					break;
-				case Prospero::TileMode::kStandard64KB:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard64KB;
-					}
-					break;
-				case Prospero::TileMode::kPrt:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Prt64KB;
-					}
-					break;
-				case Prospero::TileMode::kRenderTarget:
-					if (Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::RenderTarget64KB;
-					}
-					break;
-				case Prospero::TileMode::kDepth:
-					if (allow_depth_tile && Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::Depth64KB;
-					}
-					break;
-				default: break;
-			}
-			if (layout.tile_family == TileBlockFamily::Count) {
-				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
-				     "size=%" PRIu64 " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64
-				     " levels=%" PRIu64 "\n",
-				     owner, static_cast<uint32_t>(fmt), layout.tile, upload_size, width, height,
-				     pitch, levels);
-			}
-		}
-
-		layout.pitch = TileGetTexturePitch(fmt, width, levels, layout.tile);
-
-		TileGetTextureSize(fmt, width, height, layout.pitch, levels, layout.tile, nullptr,
-		                   layout.level_sizes, layout.padded_sizes);
-
-		if (static_cast<Prospero::TileMode>(layout.tile) != Prospero::TileMode::kLinear) {
-			if (layout.volume_texture) {
-				layout.slice_stride = SetLinearUploadLevels(layout.level_sizes, fmt, height, levels,
-				                                            static_cast<uint32_t>(width));
-			} else {
-				TileSizeOffset tiled_levels[16] {};
-				std::copy_n(layout.level_sizes, levels, tiled_levels);
-				layout.source_slice_stride =
-				    CalcTextureSliceStride(tiled_levels, levels, upload_size, depth);
-				SetLinearUploadLevels(layout.level_sizes, fmt, height, levels, layout.pitch);
-				for (uint32_t i = 0; i < levels; ++i) {
-					if (tiled_levels[i].src_size > tiled_levels[i].size) {
-						layout.first_tail_level = std::min(layout.first_tail_level, i);
-					}
-					layout.level_sizes[i].src_offset = GetLevelSrcOffset(tiled_levels[i]);
-					layout.level_sizes[i].src_size   = GetLevelSrcSize(tiled_levels[i]);
-					layout.level_sizes[i].x          = tiled_levels[i].x;
-					layout.level_sizes[i].y          = tiled_levels[i].y;
-				}
-			}
-		} else if (layout.volume_texture) {
-			layout.slice_stride =
-			    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
-		}
-	} else {
+	if (fmt == 0) {
 		EXIT("%s: legacy texture upload format unsupported: fmt=0 tile=%u size=%" PRIu64
-		     " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64 " levels=%" PRIu64 "\n",
-		     owner, layout.tile, upload_size, width, height, pitch, levels);
+		     " extent=%ux%u levels=%u\n",
+		     owner, description.tile_mode, upload_size, width, height, levels);
 	}
 
-	if (!layout.volume_texture) {
-		layout.slice_stride =
-		    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
+	switch (tile_mode) {
+		case Prospero::TileMode::kLinear:
+			if (!TileGetTextureElementLayout(fmt, element)) {
+				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner, fmt);
+			}
+			break;
+		case Prospero::TileMode::kDepth:
+			if (!allow_depth_tile || !TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+		default:
+			if (!TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+	}
+
+	if (tile_mode != Prospero::TileMode::kLinear) {
+		element = {layout.surface.texture.block.bytes_per_element,
+		           layout.surface.texture.texel_width, layout.surface.texture.texel_height};
+	}
+	layout.pitch = TileGetTexturePitch(fmt, width, description.tile_mode);
+	if (tile_mode == Prospero::TileMode::kLinear) {
+		TileSizeOffset level_sizes[16] {};
+		TilePaddedSize padded_sizes[16] {};
+		TileGetTextureSize(fmt, width, height, levels, description.tile_mode, nullptr, level_sizes,
+		                   padded_sizes);
+		for (uint32_t level = 0; level < levels; ++level) {
+			layout.mips[level] = {level_sizes[level].offset, level_sizes[level].size,
+			                      padded_sizes[level].width, padded_sizes[level].height};
+		}
+	} else if (volume_texture) {
+		layout.slice_stride = SetLinearUploadLevels(layout.mips, element, height, levels, width);
+	} else {
+		layout.source_slice_stride = layout.surface.block_slice_size;
+		if (depth > 1 && upload_size != 0 && upload_size % depth == 0) {
+			layout.source_slice_stride = std::max(layout.source_slice_stride, upload_size / depth);
+		}
+		SetLinearUploadLevels(layout.mips, element, height, levels, layout.pitch);
+	}
+
+	if (!volume_texture || tile_mode == Prospero::TileMode::kLinear) {
+		layout.slice_stride = CalcTextureSliceStride(layout.mips, levels, upload_size, depth);
 	}
 	return layout;
 }
 
-std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
-                                                         uint32_t width, uint32_t height,
-                                                         uint32_t depth, uint64_t levels,
-                                                         bool array_texture, bool volume_texture) {
-	uint32_t mip_width  = width;
-	uint32_t mip_height = height;
-	uint32_t mip_pitch  = volume_texture && static_cast<Prospero::TileMode>(layout.tile) !=
-	                                            Prospero::TileMode::kLinear
-	                          ? width
-	                          : layout.pitch;
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout) {
+	const auto& description    = layout.surface.description;
+	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const auto  depth          = volume_texture ? description.depth : description.layers;
+	uint32_t    mip_width      = description.width;
+	uint32_t    mip_height     = description.height;
+	uint32_t mip_pitch = volume_texture && static_cast<Prospero::TileMode>(description.tile_mode) !=
+	                                           Prospero::TileMode::kLinear
+	                         ? description.width
+	                         : layout.pitch;
 
 	std::vector<vk::BufferImageCopy> regions;
-	regions.reserve(GetTextureRegionCount(depth, levels, volume_texture));
-	for (uint32_t i = 0; i < levels; i++) {
-		EXIT_NOT_IMPLEMENTED(layout.level_sizes[i].size == 0);
+	regions.reserve(GetTextureRegionCount(depth, description.levels, volume_texture));
+	for (uint32_t i = 0; i < description.levels; i++) {
+		EXIT_NOT_IMPLEMENTED(layout.mips[i].size == 0);
 
 		const auto mip_depth = GetTextureLevelDepth(depth, i, volume_texture);
 
 		for (uint32_t z = 0; z < mip_depth; z++) {
 			const auto          slice_offset = z * layout.slice_stride;
 			vk::BufferImageCopy region {};
-			region.bufferOffset     = layout.level_sizes[i].offset + slice_offset;
-			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, array_texture ? z : 0,
+			region.bufferOffset     = layout.mips[i].offset + slice_offset;
+			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, volume_texture ? 0u : z,
 			                           1};
 			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
 			region.imageExtent      = {mip_width, mip_height, 1};
-			const bool linear =
-			    static_cast<Prospero::TileMode>(layout.tile) == Prospero::TileMode::kLinear;
+			const bool linear       = static_cast<Prospero::TileMode>(description.tile_mode) ==
+			                          Prospero::TileMode::kLinear;
 			if (linear) {
-				region.bufferRowLength   = layout.padded_sizes[i].width;
-				region.bufferImageHeight = layout.padded_sizes[i].height;
+				region.bufferRowLength   = layout.mips[i].row_length;
+				region.bufferImageHeight = layout.mips[i].image_height;
 			} else {
-				const auto align = [](uint32_t value, uint32_t block) {
+				const auto texel_width = layout.surface.texture.texel_width;
+				const auto align       = [](uint32_t value, uint32_t block) {
 					return ((value + block - 1u) / block) * block;
 				};
-				const auto pitch       = align(mip_pitch, layout.texel_block);
-				region.bufferRowLength = pitch > align(mip_width, layout.texel_block) ? pitch : 0;
+				const auto pitch       = align(mip_pitch, texel_width);
+				region.bufferRowLength = pitch > align(mip_width, texel_width) ? pitch : 0;
 			}
 			regions.push_back(region);
 		}
@@ -456,24 +406,6 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 	return regions;
 }
 
-struct GpuTileElementLayout {
-	uint32_t bytes = 0;
-	uint32_t wide  = 1;
-	uint32_t tall  = 1;
-};
-
-static bool GetGpuTileElementLayout(uint32_t fmt, GpuTileElementLayout& out) {
-	if (const auto bytes = Prospero::NumBytesPerElement(fmt); bytes != 0) {
-		out = {bytes, 1, 1};
-		return true;
-	}
-	if (const auto bytes = Prospero::BlockCompressedBytesPerBlock(fmt); bytes != 0) {
-		out = {bytes, 4, 4};
-		return true;
-	}
-	return false;
-}
-
 static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity, uint64_t& size) {
 	if (offset > capacity || length > capacity - offset) {
 		return false;
@@ -483,42 +415,32 @@ static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity,
 }
 
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
-                              const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
-                              uint64_t levels, std::vector<GpuTileInfo>& out_infos) {
+                              const TextureUploadLayout& layout, uint32_t levels,
+                              std::vector<GpuTileInfo>& out_infos) {
+	const auto& description    = layout.surface.description;
+	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const auto  depth          = volume_texture ? description.depth : description.layers;
 	if (tiled_size == 0 || levels == 0 || levels > 16 || depth == 0 ||
-	    regions.size() != GetTextureRegionCount(depth, levels, layout.volume_texture) ||
-	    Prospero::IsFmaskTextureFormat(fmt)) {
+	    regions.size() != GetTextureRegionCount(depth, levels, volume_texture) ||
+	    Prospero::IsFmaskTextureFormat(description.format)) {
 		return false;
 	}
 
-	GpuTileElementLayout element {};
-	if (layout.tile_family == TileBlockFamily::RenderTarget64KB ||
-	    layout.tile_family == TileBlockFamily::Depth64KB) {
-		element.bytes = Prospero::RenderTargetBytesPerElement(fmt);
-	} else if (!GetGpuTileElementLayout(fmt, element)) {
-		return false;
-	}
-	if (element.bytes == 0) {
+	const auto& surface = layout.surface;
+	const auto& texture = surface.texture;
+	const auto& block   = texture.block;
+	if (surface.description.levels < levels || block.bytes_per_element == 0 ||
+	    block.family == TileBlockFamily::Count) {
 		return false;
 	}
 
 	std::vector<GpuTileInfo> infos;
 	infos.reserve(regions.size());
-	if (layout.volume_texture) {
-		TileVolumeLayout volume {};
-		if (!TileGetTextureVolumeLayout(fmt, regions[0].imageExtent.width,
-		                                regions[0].imageExtent.height, depth,
-		                                static_cast<uint32_t>(levels), layout.tile, volume)) {
-			return false;
-		}
-		element = {volume.bytes_per_element, volume.texel_width, volume.texel_height};
-		TileBlockLayout block {};
-		if (!TileGetBlockLayout(volume.family, element.bytes, block)) return false;
-
+	if (volume_texture) {
 		size_t region_base = 0;
 		for (uint32_t level = 0; level < levels; ++level) {
 			const uint32_t mip_depth     = GetTextureLevelDepth(depth, level, true);
-			const bool     tail          = level >= volume.first_tail_level;
+			const bool     tail          = level >= surface.first_tail_level;
 			const uint64_t linear_stride = layout.slice_stride;
 			for (uint32_t z = 0; z < mip_depth; z += block.block_depth) {
 				const uint32_t copy_depth = std::min(block.block_depth, mip_depth - z);
@@ -533,50 +455,43 @@ bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferI
 				info.bytes_per_element = block.bytes_per_element;
 				info.linear_offset     = region.bufferOffset;
 				info.tiled_offset =
-				    static_cast<uint64_t>(z / block.block_depth) * volume.block_slice_size +
-				    volume.level_offsets[level];
+				    static_cast<uint64_t>(z / block.block_depth) * surface.block_slice_size +
+				    surface.mips[level].offset;
 				const uint64_t linear_span =
 				    static_cast<uint64_t>(copy_depth - 1u) * linear_stride +
-				    layout.level_sizes[level].size;
+				    layout.mips[level].size;
 				if (!SetGpuTileSize(info.linear_offset, linear_span, UINT64_MAX,
 				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, volume.level_sizes[level], tiled_size,
+				    !SetGpuTileSize(info.tiled_offset, surface.mips[level].size, tiled_size,
 				                    info.tiled_size)) {
 					return false;
 				}
 				info.linear_slice_stride = linear_stride;
-				info.width =
-				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
-				info.depth  = copy_depth;
+				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
+				                           texture.texel_width,
+				                       1u);
+				info.height = std::max(
+				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+				info.depth = copy_depth;
 				info.surface_z =
 				    block.block_depth == 1 ? static_cast<uint32_t>(region.imageOffset.z) : 0;
-				info.pitch        = std::max((pitch + element.wide - 1u) / element.wide, 1u);
-				info.tail_x       = tail ? volume.tail_x[level] : 0;
-				info.tail_y       = tail ? volume.tail_y[level] : 0;
+				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
+				info.tail_x       = tail ? surface.mips[level].tail_x : 0;
+				info.tail_y       = tail ? surface.mips[level].tail_y : 0;
 				info.tail         = tail;
-				info.tiled_width  = volume.level_widths[level];
-				info.tiled_height = volume.level_heights[level];
+				info.tiled_width  = surface.mips[level].padded_width;
+				info.tiled_height = surface.mips[level].padded_height;
 				infos.push_back(info);
 			}
 			region_base += mip_depth;
 		}
 	} else {
-		const auto base_family = layout.tile_family;
-		if (base_family == TileBlockFamily::Count) {
-			return false;
-		}
-
 		size_t region_index = 0;
 		for (uint32_t level = 0; level < levels; level++) {
-			const auto&     level_size = layout.level_sizes[level];
-			const bool      tail       = level >= layout.first_tail_level;
-			const auto      family     = base_family;
-			TileBlockLayout block {};
-			if (!TileGetBlockLayout(family, element.bytes, block)) {
-				return false;
-			}
-			const auto level_depth = GetTextureLevelDepth(depth, level, layout.volume_texture);
+			const auto& level_size  = layout.mips[level];
+			const auto& mip         = surface.mips[level];
+			const bool  tail        = level >= surface.first_tail_level;
+			const auto  level_depth = GetTextureLevelDepth(depth, level, false);
 			for (uint32_t z = 0; z < level_depth; z++) {
 				const auto& region = regions[region_index++];
 				const auto  pitch =
@@ -585,39 +500,36 @@ bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferI
 				                                 ? region.bufferImageHeight
 				                                 : region.imageExtent.height;
 				GpuTileInfo info {};
-				info.family            = block.family;
-				info.bytes_per_element = block.bytes_per_element;
-				info.linear_offset     = region.bufferOffset;
-				info.tiled_offset      = TextureUploadSliceSourceOffset(layout, level, z);
-				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
-				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, GetLevelSrcSize(level_size), tiled_size,
-				                    info.tiled_size)) {
+				info.family              = block.family;
+				info.bytes_per_element   = block.bytes_per_element;
+				info.linear_offset       = region.bufferOffset;
+				const auto source_stride = layout.source_slice_stride != 0
+				                               ? layout.source_slice_stride
+				                               : surface.block_slice_size;
+				if (z > (UINT64_MAX - mip.offset) / source_stride) {
 					return false;
 				}
-				info.width =
-				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height    = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
-				info.surface_z = base_family == TileBlockFamily::RenderTarget64KB ||
-				                         base_family == TileBlockFamily::Depth64KB
+				info.tiled_offset = mip.offset + static_cast<uint64_t>(z) * source_stride;
+				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
+				                    info.linear_size) ||
+				    !SetGpuTileSize(info.tiled_offset, mip.size, tiled_size, info.tiled_size)) {
+					return false;
+				}
+				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
+				                           texture.texel_width,
+				                       1u);
+				info.height = std::max(
+				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+				info.surface_z = block.family == TileBlockFamily::RenderTarget64KB ||
+				                         block.family == TileBlockFamily::Depth64KB
 				                     ? region.imageSubresource.baseArrayLayer
 				                     : 0;
-				info.pitch     = std::max((pitch + element.wide - 1u) / element.wide, 1u);
-				info.tail      = tail;
-				info.tail_x    = tail ? level_size.x : 0;
-				info.tail_y    = tail ? level_size.y : 0;
-				info.tiled_width =
-				    layout.padded_sizes[level].width != 0
-				        ? std::max((layout.padded_sizes[level].width + element.wide - 1u) /
-				                       element.wide,
-				                   1u)
-				        : info.pitch;
-				info.tiled_height =
-				    layout.padded_sizes[level].height != 0
-				        ? std::max((layout.padded_sizes[level].height + element.tall - 1u) /
-				                       element.tall,
-				                   1u)
-				        : info.height;
+				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
+				info.tail  = tail;
+				info.tail_x       = tail ? mip.tail_x : 0;
+				info.tail_y       = tail ? mip.tail_y : 0;
+				info.tiled_width  = mip.padded_width;
+				info.tiled_height = mip.padded_height;
 				infos.push_back(info);
 			}
 		}

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

@@ -19,34 +19,32 @@ struct RenderTargetFormatInfo {
 	Prospero::ColorComponentMapping export_mapping;
 };
 
+struct TextureUploadMipLayout {
+	uint64_t offset       = 0;
+	uint64_t size         = 0;
+	uint32_t row_length   = 0;
+	uint32_t image_height = 0;
+};
+
 struct TextureUploadLayout {
-	uint32_t        tile                = 0;
-	uint32_t        pitch               = 0;
-	uint32_t        texel_block         = 1;
-	uint64_t        slice_stride        = 0;
-	uint64_t        source_slice_stride = 0;
-	uint32_t        first_tail_level    = 16;
-	TileBlockFamily tile_family         = TileBlockFamily::Count;
-	bool            volume_texture      = false;
-	TileSizeOffset  level_sizes[16]     = {};
-	TilePaddedSize  padded_sizes[16]    = {};
+	uint32_t               pitch               = 0;
+	uint64_t               slice_stride        = 0;
+	uint64_t               source_slice_stride = 0;
+	TileSurfaceLayout      surface;
+	TextureUploadMipLayout mips[16] = {};
 };
 
 vk::ComponentMapping   TextureGetComponentMapping(uint32_t swizzle);
 vk::Format             TextureGetFormat(uint32_t fmt);
 RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t layout, uint32_t type, uint32_t order);
-TextureUploadLayout    TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64_t height,
-                                               uint64_t levels, uint32_t depth, uint64_t pitch,
-                                               uint64_t tile, uint64_t upload_size,
-                                               bool allow_depth_tile, bool volume_texture,
-                                               const char* owner);
-std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
-                                                         uint32_t width, uint32_t height,
-                                                         uint32_t depth, uint64_t levels,
-                                                         bool array_texture, bool volume_texture);
+TextureUploadLayout    TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
+                                               uint32_t levels, uint32_t depth, uint32_t tile,
+                                               uint64_t upload_size, bool allow_depth_tile,
+                                               bool volume_texture, const char* owner);
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout);
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
-                              const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
-                              uint64_t levels, std::vector<GpuTileInfo>& infos);
+                              const TextureUploadLayout& layout, uint32_t levels,
+                              std::vector<GpuTileInfo>& infos);
 
 } // namespace Libs::Graphics
 

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

@@ -239,7 +239,7 @@ bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, c
 	const auto samples      = multisampled ? 1u << descriptor.LastLevel() : 1u;
 	const auto pitch =
 	    multisampled ? TileGetDepthPitch(width, image.info.bytes_per_block, descriptor.LastLevel())
-	                 : TileGetTexturePitch(descriptor.Format(), width, 1, descriptor.TileMode());
+	                 : TileGetTexturePitch(descriptor.Format(), width, descriptor.TileMode());
 	const bool supported_2d    = type == Prospero::ImageType::kColor2D &&
 	                             image.info.resources.layers == 1 && descriptor.Depth() == 0 &&
 	                             descriptor.BaseArray5() == 0;
@@ -321,7 +321,7 @@ static void ValidateDepthTargetBinding(const ShaderRecompiler::IR::ImageResource
 		return;
 	}
 	const auto descriptor_pitch =
-	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u, 1,
+	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u,
 	                        descriptor.TileMode());
 	EXIT("unsupported sampled depth target: resource=%d descriptor=%d encoding=%d format=%d "
 	     "kind=%u dimension=%u mip_mode=%u read=%d written=%d atomic=%d compare=%d "
@@ -373,21 +373,33 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool is_3d = resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D &&
 	                   descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor3D) &&
 	                   descriptor.BaseArray5() == 0;
-	TileBlockLayout depth_block {};
-	const auto      depth_bpe = Prospero::RenderTargetBytesPerElement(descriptor.Format());
-	const bool      supported_depth_tile =
-	    tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) && !resource.read &&
-	    !Prospero::IsFmaskTextureFormat(descriptor.Format()) && (is_2d || is_2d_array) &&
-	    TileGetBlockLayout(TileBlockFamily::Depth64KB, depth_bpe, depth_block);
-	const bool supported_standard_tile =
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB) &&
-	     TileIsStandard4KBTextureSupported(descriptor.Format())) ||
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB) &&
-	     TileIsStandard64KBTextureSupported(descriptor.Format()));
-	const bool supported_tile = tile == Prospero::GpuEnumValue(Prospero::TileMode::kLinear) ||
-	                            tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
-	                            supported_depth_tile || supported_standard_tile;
-	const auto swizzle        = descriptor.DstSelXYZW();
+	TileTextureBlockLayout tile_layout {};
+	bool                   supported_tile = false;
+	switch (static_cast<Prospero::TileMode>(tile)) {
+		case Prospero::TileMode::kLinear: supported_tile = true; break;
+		case Prospero::TileMode::kDepth:
+			supported_tile =
+			    !resource.read && !Prospero::IsFmaskTextureFormat(descriptor.Format()) &&
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard256B:
+			supported_tile =
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, is_3d, tile_layout);
+			break;
+		case Prospero::TileMode::kRenderTarget:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		default: break;
+	}
+	const auto swizzle = descriptor.DstSelXYZW();
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
@@ -493,18 +505,20 @@ static TextureCache::ImageDesc NullTextureDesc(const ShaderRecompiler::IR::Image
 
 static void PopulateTextureMipLayout(ImageInfo& info) {
 	if (info.IsVolume() && info.tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kLinear)) {
-		TileVolumeLayout volume {};
-		if (!TileGetTextureVolumeLayout(info.guest_format, info.extent.width, info.extent.height,
-		                                info.extent.depth, info.resources.levels, info.tile_mode,
-		                                volume)) {
+		TileSurfaceLayout            surface {};
+		const TileSurfaceDescription description {
+		    info.guest_format,  info.tile_mode,    TileSurfaceDimension::Dim3D, info.extent.width,
+		    info.extent.height, info.extent.depth, info.resources.levels,       1};
+		if (!TileGetTiledTextureLayout(description, surface)) {
 			EXIT("unsupported normalized volume texture layout\n");
 		}
 		for (uint32_t level = 0; level < info.resources.levels; level++) {
+			const auto& mip        = surface.mips[level];
 			info.mip_layout[level] = {
-			    volume.level_offsets[level],
-			    volume.level_sizes[level],
-			    volume.level_widths[level],
-			    volume.level_heights[level],
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
 			};
 		}
 		return;
@@ -512,7 +526,7 @@ static void PopulateTextureMipLayout(ImageInfo& info) {
 
 	TileSizeOffset levels[16] {};
 	TilePaddedSize padded[16] {};
-	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height, info.pitch,
+	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height,
 	                   info.resources.levels, info.tile_mode, nullptr, levels, padded);
 	for (uint32_t level = 0; level < info.resources.levels; level++) {
 		const auto offset =
@@ -670,9 +684,9 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		}
 		size.size *= image_layers;
 	} else {
-		pitch = TileGetTexturePitch(format, width, levels, tile);
-		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, pitch, levels,
-		                        tile, volume, size);
+		pitch = TileGetTexturePitch(format, width, tile);
+		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, levels, tile,
+		                        volume, size);
 	}
 	EXIT_NOT_IMPLEMENTED(size.size == 0 || size.align == 0 ||
 	                     (address & (static_cast<uint64_t>(size.align) - 1u)) != 0);
@@ -919,22 +933,21 @@ RenderExecutor::PrepareGraphicsBindings(CommandBuffer& buffer, const ShaderStage
 	return bindings;
 }
 
-static void RetainBindings(CommandBuffer&                            buffer,
-                           const DescriptorCache::NativeDescriptors& resources) {
+static void RetainBindings(CommandBuffer& buffer, DescriptorCache::NativeDescriptors& resources) {
 	if (resources.flattened_srt.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.flattened_srt.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.flattened_srt.owner));
 	}
 	if (resources.user_data.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.user_data.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.user_data.owner));
 	}
-	for (const auto& view: resources.buffers) {
+	for (auto& view: resources.buffers) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
-	for (const auto& view: resources.addresses) {
+	for (auto& view: resources.addresses) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
 }

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

@@ -36,6 +36,22 @@ VulkanMemoryBarrier MakeShaderWriteDependency() {
 	return barrier;
 }
 
+VulkanMemoryBarrier MakeShaderAccessDependency() {
+	VulkanMemoryBarrier barrier {};
+	barrier.sType         = vk::StructureType::eMemoryBarrier;
+	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
+	return barrier;
+}
+
+VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
+	VulkanMemoryBarrier barrier {};
+	barrier.sType         = vk::StructureType::eMemoryBarrier;
+	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
+	return barrier;
+}
+
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 	EXIT_IF(buffer == nullptr);
 
@@ -84,6 +100,21 @@ bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
 	return !CollectShaderBufferWrites(*runtime.program, *runtime.resources).empty();
 }
 
+void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
+	EXIT_IF(vk_buffer == nullptr || !source_stages);
+	const auto barrier = MakeShaderAccessDependency();
+	vk_buffer.pipelineBarrier(source_stages, vk::PipelineStageFlagBits::eAllCommands,
+	                          vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
+}
+
+void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,
+                              vk::PipelineStageFlags destination_stages) {
+	EXIT_IF(vk_buffer == nullptr || !destination_stages);
+	const auto barrier = MakeShaderWriteHazardDependency();
+	vk_buffer.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands, destination_stages,
+	                          vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
+}
+
 void ShaderWriteBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
 	EXIT_IF(vk_buffer == nullptr || !source_stages);
 	const auto barrier = MakeShaderWriteDependency();

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

@@ -18,12 +18,17 @@ struct ShaderBufferWriteRange {
 };
 
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
+VulkanMemoryBarrier     MakeShaderAccessDependency();
+VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
 VulkanMemoryBarrier     MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
 std::vector<ShaderBufferWriteRange>
 CollectShaderBufferWrites(const ShaderRecompiler::IR::Program&          program,
                           const ShaderRecompiler::IR::ResourceSnapshot& resources);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
+void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);
+void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,
+                              vk::PipelineStageFlags destination_stages);
 void ShaderWriteBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);
 
 } // namespace Libs::Graphics

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

@@ -947,15 +947,23 @@ void CreatePipelineInternal(
 	    vk::DynamicState::eViewport,
 	    vk::DynamicState::eScissor,
 	    vk::DynamicState::eLineWidth,
+	    vk::DynamicState::eDepthBiasEnable,
+	    vk::DynamicState::eDepthBias,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
 #if !defined(__APPLE__)
+	    // Keep last so depth-only pipelines can omit this dynamic state.
 	    vk::DynamicState::eColorWriteEnableEXT, // unsupported by MoltenVK; static mask instead
 #endif
 	};
-	const auto dynamic_states_count =
+	auto dynamic_states_count =
 	    static_cast<uint32_t>(sizeof(dynamic_states) / sizeof(dynamic_states[0]));
+#if !defined(__APPLE__)
+	if (static_params.color_count == 0) {
+		dynamic_states_count--;
+	}
+#endif
 
 	vk::PipelineDynamicStateCreateInfo dynamic_state {};
 	dynamic_state.sType             = vk::StructureType::ePipelineDynamicStateCreateInfo;

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -325,9 +325,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 
 	auto vk_buffer = buffer.Handle();
 	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline.pipeline_layout, bindings);
-	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
-	vk_buffer.dispatch(thread_group_x, thread_group_y, thread_group_z);
-
 	bool has_storage_writes = HasShaderBufferWrites(input_info.stage);
 	has_storage_writes =
 	    std::any_of(program.info.images.begin(), program.info.images.end(),
@@ -338,8 +335,15 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 	                }) ||
 	    has_storage_writes;
 	if (has_storage_writes) {
-		ShaderWriteBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
+		// A host fence used to serialize every dispatch. Preserve its read-before-write ordering
+		// while allowing the queue to execute asynchronously.
+		ShaderWriteHazardBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
 	}
+	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
+	vk_buffer.dispatch(thread_group_x, thread_group_y, thread_group_z);
+
+	// The removed host fence also ordered read-only dispatches before later writers.
+	ShaderAccessBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
 	ResetBindings();
 }
 

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -64,12 +64,32 @@ int32_t ResolveVertexOffset(uint32_t index_offset, const ShaderVertexInputInfo&
 	return 0;
 }
 
-static std::atomic<uint32_t> g_draw_state_log_count       = 0;
-static std::atomic<uint32_t> g_draw_input_log_count       = 0;
-static std::atomic<uint32_t> g_mrt_state_log_count        = 0;
-static std::atomic<uint32_t> g_shader_stage_log_count     = 0;
+static std::atomic<uint32_t> g_draw_state_log_count   = 0;
+static std::atomic<uint32_t> g_draw_input_log_count   = 0;
+static std::atomic<uint32_t> g_mrt_state_log_count    = 0;
+static std::atomic<uint32_t> g_shader_stage_log_count = 0;
+
 static std::atomic<uint32_t> g_framebuffer_skip_log_count = 0;
 
+static float ConvertPolygonOffsetConstantFactor(float guest_factor, const HW::PolyOffset& offset,
+                                                vk::Format host_depth_format) {
+	if (offset.db_is_float_fmt) {
+		return guest_factor;
+	}
+
+	int host_depth_bits = 0;
+	switch (host_depth_format) {
+		case vk::Format::eD16Unorm:
+		case vk::Format::eD16UnormS8Uint: host_depth_bits = 16; break;
+		case vk::Format::eD24UnormS8Uint: host_depth_bits = 24; break;
+		default:
+			// A fixed-point guest bias cannot be represented exactly by a floating-point host
+			// attachment without VK_EXT_depth_bias_control.
+			return guest_factor;
+	}
+	return std::ldexp(guest_factor, host_depth_bits + offset.neg_num_db_bits);
+}
+
 static const char* RenderColorTypeName(RenderColorType type) {
 	switch (type) {
 		case RenderColorType::NoColorOutput: return "NoColorOutput";
@@ -335,6 +355,22 @@ static PipelineDynamicParameters BuildGraphicsDynamicParams(const RenderCommandB
 	ret.stencil_front      = depth.stencil_dynamic_front;
 	ret.stencil_back       = depth.stencil_dynamic_back;
 
+	const auto& mode        = ctx.GetModeControl();
+	const auto& poly_offset = ctx.GetPolyOffset();
+	const bool  use_front   = mode.poly_offset_front_enable && !mode.cull_front;
+	const bool  use_back    = mode.poly_offset_back_enable && !mode.cull_back;
+	ret.depth_bias_enable   = use_front || use_back;
+	if (ret.depth_bias_enable) {
+		// Vulkan has one bias for both faces. Prefer a visible front face when both are enabled.
+		const float guest_constant_factor =
+		    use_front ? poly_offset.front_offset : poly_offset.back_offset;
+		ret.depth_bias_constant_factor =
+		    ConvertPolygonOffsetConstantFactor(guest_constant_factor, poly_offset, depth.format);
+		ret.depth_bias_clamp = poly_offset.clamp;
+		ret.depth_bias_slope_factor =
+		    (use_front ? poly_offset.front_scale : poly_offset.back_scale) / 16.0f;
+	}
+
 	// CB_COLOR_CONTROL.operation controls special CB operations, not the normal color component
 	// write mask. Use CB_TARGET_MASK here so scanout passes with mode=Disable do not go black.
 	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
@@ -379,6 +415,13 @@ static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffe
 	}
 	vk_buffer.setLineWidth(line_width);
 
+	vk_buffer.setDepthBiasEnable(dynamic_params.depth_bias_enable ? VK_TRUE : VK_FALSE);
+	if (dynamic_params.depth_bias_enable) {
+		vk_buffer.setDepthBias(dynamic_params.depth_bias_constant_factor,
+		                       dynamic_params.depth_bias_clamp,
+		                       dynamic_params.depth_bias_slope_factor);
+	}
+
 	if (dynamic_params.stencil_test_enable) {
 		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
 		                                dynamic_params.stencil_front.compareMask);
@@ -664,12 +707,10 @@ struct DrawIndexBufferSource {
 
 struct PreparedIndexBuffer {
 	std::shared_ptr<void> owner;
-	vk::Buffer            buffer   = nullptr;
-	uint64_t              address  = 0;
-	uint64_t              size     = 0;
-	vk::DeviceSize        offset   = 0;
-	vk::IndexType         type     = vk::IndexType::eUint16;
-	bool                  streamed = false;
+	vk::Buffer            buffer = nullptr;
+	uint64_t              size   = 0;
+	vk::DeviceSize        offset = 0;
+	vk::IndexType         type   = vk::IndexType::eUint16;
 };
 
 static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer) {
@@ -686,11 +727,24 @@ struct VertexBufferRange {
 	[[nodiscard]] uint64_t RequestedSize() const { return requested_end - base_address; }
 };
 
-static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&         buffer,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+struct PreparedVertexBuffers {
+	static constexpr uint32_t MaxBuffers = ShaderVertexInputInfo::RES_MAX;
+
+	std::array<vk::Buffer, MaxBuffers>            buffers {};
+	std::array<vk::DeviceSize, MaxBuffers>        offsets {};
+	std::array<std::shared_ptr<void>, MaxBuffers> owners {};
+	uint32_t                                      count       = 0;
+	uint32_t                                      owner_count = 0;
+};
+
+static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         buffer,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
+	EXIT_IF(vs_input_info.buffers_num < 0 ||
+	        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX);
+
 	// Collect the non-empty guest vertex ranges.
-	std::vector<VertexBufferRange> ranges;
-	ranges.reserve(vs_input_info.buffers_num);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> ranges {};
+	uint32_t                                                      range_count = 0;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
@@ -701,27 +755,31 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 			EXIT("invalid vertex buffer range: addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 			     vertex.addr, size);
 		}
-		ranges.push_back({vertex.addr, vertex.addr + size});
+		ranges[range_count++] = {vertex.addr, vertex.addr + size};
 	}
 
-	std::ranges::sort(ranges, [](const VertexBufferRange& left, const VertexBufferRange& right) {
-		return left.base_address < right.base_address;
-	});
+	std::sort(ranges.begin(), ranges.begin() + range_count,
+	          [](const VertexBufferRange& left, const VertexBufferRange& right) {
+		          return left.base_address < right.base_address;
+	          });
 
 	// Merge overlapping or touching ranges before acquiring host buffers.
-	std::vector<VertexBufferRange> merged_ranges;
-	merged_ranges.reserve(ranges.size());
-	for (const auto& range: ranges) {
-		if (!merged_ranges.empty() && merged_ranges.back().requested_end >= range.base_address) {
-			merged_ranges.back().requested_end =
-			    std::max(merged_ranges.back().requested_end, range.requested_end);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> merged_ranges {};
+	uint32_t                                                      merged_count = 0;
+	for (uint32_t i = 0; i < range_count; i++) {
+		const auto& range = ranges[i];
+		if (merged_count != 0 &&
+		    merged_ranges[merged_count - 1].requested_end >= range.base_address) {
+			merged_ranges[merged_count - 1].requested_end =
+			    std::max(merged_ranges[merged_count - 1].requested_end, range.requested_end);
 			continue;
 		}
-		merged_ranges.push_back(range);
+		merged_ranges[merged_count++] = {range.base_address, range.requested_end};
 	}
 
 	auto& cache = buffer.GetContext().GetBufferCache();
-	for (auto& range: merged_ranges) {
+	for (uint32_t i = 0; i < merged_count; i++) {
+		auto& range = merged_ranges[i];
 		// PPSA20298
 		const auto size =
 		    Libs::LibKernel::Memory::ClampRangeSize(range.base_address, range.RequestedSize());
@@ -730,30 +788,47 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 	}
 
 	// Rebuild slot bindings, offsetting non-empty slots into their acquired merged range.
-	std::vector<BufferBinding> bindings;
-	bindings.reserve(vs_input_info.buffers_num);
+	PreparedVertexBuffers prepared;
+	prepared.count = static_cast<uint32_t>(vs_input_info.buffers_num);
+	std::shared_ptr<Buffer> null_owner;
+	vk::Buffer              null_buffer = nullptr;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
 		if (size == 0) {
-			auto owner = cache.ObtainNullBuffer();
-			bindings.push_back({owner, owner->Handle(), 0});
+			if (null_owner == nullptr) {
+				null_owner  = cache.ObtainNullBuffer();
+				null_buffer = null_owner->Handle();
+			}
+			prepared.buffers[i] = null_buffer;
+			prepared.offsets[i] = 0;
 			continue;
 		}
 
-		const auto range = std::ranges::find_if(merged_ranges, [&](const VertexBufferRange& value) {
-			return vertex.addr >= value.base_address && vertex.addr < value.acquired_end;
-		});
-		if (range == merged_ranges.end()) {
+		const auto range = std::find_if(merged_ranges.begin(), merged_ranges.begin() + merged_count,
+		                                [&](const VertexBufferRange& value) {
+			                                return vertex.addr >= value.base_address &&
+			                                       vertex.addr < value.acquired_end;
+		                                });
+		if (range == merged_ranges.begin() + merged_count) {
 			EXIT("vertex buffer address is outside the acquired range: addr=0x%016" PRIx64 "\n",
 			     vertex.addr);
 		}
 
-		auto binding = range->binding;
-		binding.offset += vertex.addr - range->base_address;
-		bindings.push_back(std::move(binding));
+		prepared.buffers[i] = range->binding.buffer;
+		prepared.offsets[i] = range->binding.offset + vertex.addr - range->base_address;
 	}
-	return bindings;
+
+	if (null_owner != nullptr) {
+		prepared.owners[prepared.owner_count++] = std::move(null_owner);
+	}
+	for (uint32_t i = 0; i < merged_count; i++) {
+		if (merged_ranges[i].binding.owner != nullptr) {
+			EXIT_IF(prepared.owner_count >= PreparedVertexBuffers::MaxBuffers);
+			prepared.owners[prepared.owner_count++] = std::move(merged_ranges[i].binding.owner);
+		}
+	}
+	return prepared;
 }
 
 static void SetDrawDebugPhase(RenderCommandBuffer& buffer, uint64_t submit_id,
@@ -885,10 +960,9 @@ static void RefreshShaders(RenderCommandBuffer& buffer, const DrawCallInfo& draw
 	}
 }
 
-static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t                     submit_id,
-                                                       RenderCommandBuffer&         buffer,
-                                                       const DrawCallInfo&          draw,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+static PreparedVertexBuffers PrepareVertexBuffers(uint64_t submit_id, RenderCommandBuffer& buffer,
+                                                  const DrawCallInfo&          draw,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
 	EXIT_IF(draw.name == nullptr);
 	(void)submit_id;
 
@@ -896,12 +970,6 @@ static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t
 	return AcquireVertexBuffers(buffer, vs_input_info);
 }
 
-static void RebindVertexBuffers(RenderCommandBuffer&         buffer,
-                                const ShaderVertexInputInfo& vs_input_info,
-                                std::vector<BufferBinding>&  bindings) {
-	bindings = AcquireVertexBuffers(buffer, vs_input_info);
-}
-
 static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffer,
                                               const DrawIndexBufferSource& source) {
 	PreparedIndexBuffer prepared;
@@ -909,16 +977,14 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 		return prepared;
 	}
 	EXIT_IF(source.size == 0);
-	prepared.address = source.address;
-	prepared.size    = source.size;
-	prepared.type    = source.type;
+	prepared.size = source.size;
+	prepared.type = source.type;
 	if (source.host_data != nullptr) {
 		auto binding =
 		    buffer.GetContext().GetBufferCache().UploadTransient(source.host_data, source.size, 16);
-		prepared.owner    = std::move(binding.owner);
-		prepared.buffer   = binding.buffer;
-		prepared.offset   = binding.offset;
-		prepared.streamed = true;
+		prepared.owner  = std::move(binding.owner);
+		prepared.buffer = binding.buffer;
+		prepared.offset = binding.offset;
 	} else {
 		auto binding =
 		    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, source.address, source.size);
@@ -929,36 +995,29 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 	return prepared;
 }
 
-static void RebindIndexBuffer(RenderCommandBuffer& buffer, PreparedIndexBuffer& prepared) {
-	if (prepared.size == 0 || prepared.streamed) {
-		return;
-	}
-	auto binding =
-	    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, prepared.address, prepared.size);
-	prepared.owner  = std::move(binding.owner);
-	prepared.buffer = binding.buffer;
-	prepared.offset = binding.offset;
-}
-
 static void CommitVertexBuffers(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                                std::vector<BufferBinding>& bindings) {
-	for (uint32_t slot = 0; slot < bindings.size(); slot++) {
-		auto& binding = bindings[slot];
-		if (binding.owner != nullptr) {
-			buffer.RetainResourceUntilFence(binding.owner);
+                                PreparedVertexBuffers& prepared) {
+	for (uint32_t i = 0; i < prepared.owner_count; i++) {
+		if (prepared.owners[i] != nullptr) {
+			buffer.RetainResourceUntilFence(std::move(prepared.owners[i]));
 		}
-		EXIT_IF(binding.buffer == nullptr);
-		vk_buffer.bindVertexBuffers(slot, 1, &binding.buffer, &binding.offset);
+	}
+	for (uint32_t i = 0; i < prepared.count; i++) {
+		EXIT_IF(prepared.buffers[i] == nullptr);
+	}
+	if (prepared.count != 0) {
+		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
+		                            prepared.offsets.data());
 	}
 }
 
 static void CommitIndexBuffer(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                              const PreparedIndexBuffer& prepared) {
+                              PreparedIndexBuffer& prepared) {
 	if (prepared.size == 0) {
 		return;
 	}
 	if (prepared.owner != nullptr) {
-		buffer.RetainResourceUntilFence(prepared.owner);
+		buffer.RetainResourceUntilFence(std::move(prepared.owner));
 	}
 	EXIT_IF(prepared.buffer == nullptr);
 	vk_buffer.bindIndexBuffer(prepared.buffer, prepared.offset, prepared.type);
@@ -1046,8 +1105,6 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer
 	                                               state.ps_input_info.stage, state.ps_active);
 	auto vertex_bindings = PrepareVertexBuffers(submit_id, buffer, draw, state.vs_input_info);
 	auto index_binding   = PrepareIndexBuffer(buffer, index_source);
-	RebindVertexBuffers(buffer, state.vs_input_info, vertex_bindings);
-	RebindIndexBuffer(buffer, index_binding);
 	state.rendering =
 	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info);
 

src/graphics/host_gpu/renderer/renderDraw.h

@@ -15,12 +15,16 @@ struct ShaderVertexInputInfo;
 
 struct PipelineDynamicParameters {
 	bool stencil_test_enable = false;
+	bool depth_bias_enable   = false;
 
 	float viewport_scale[3]  = {};
 	float viewport_offset[3] = {};
 	int   scissor_ltrb[4]    = {0};
 
 	float    line_width                                       = 1.0f;
+	float    depth_bias_constant_factor                       = 0.0f;
+	float    depth_bias_clamp                                 = 0.0f;
+	float    depth_bias_slope_factor                          = 0.0f;
 	uint32_t color_write_count                                = 1;
 	bool     color_write_enable[RENDER_COLOR_ATTACHMENTS_MAX] = {true, true};
 
@@ -33,10 +37,6 @@ struct PipelineDynamicParameters {
 static_assert(std::is_trivially_copyable_v<PipelineDynamicParameters>);
 static_assert(std::is_standard_layout_v<PipelineDynamicParameters>);
 static_assert(alignof(PipelineDynamicParameters) == 1);
-static_assert(sizeof(PipelineDynamicParameters) ==
-              sizeof(bool) + sizeof(float[3]) + sizeof(float[3]) + sizeof(int[4]) + sizeof(float) +
-                  sizeof(uint32_t) + sizeof(bool[RENDER_COLOR_ATTACHMENTS_MAX]) +
-                  sizeof(PipelineStencilDynamicState) * 2);
 
 [[nodiscard]] int32_t ResolveVertexOffset(uint32_t                     index_offset,
                                           const ShaderVertexInputInfo& vs_input_info);

src/graphics/presentation/imeOverlay.h

@@ -0,0 +1,42 @@
+#ifndef EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_
+#define EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_
+
+#include "common/common.h"
+#include "graphics/host_gpu/vulkanCommon.h"
+
+#include <memory>
+
+union SDL_Event;
+
+namespace Libs::Graphics {
+
+struct GraphicContext;
+
+struct ImeVisualState {
+	bool     active;
+	uint64_t revision;
+};
+
+void           InitializeImeInput();
+void           ShutdownImeInput();
+bool           ProcessImeInput(const SDL_Event& event);
+ImeVisualState GetImeVisualState() noexcept;
+
+class ImeOverlay final {
+public:
+	explicit ImeOverlay(GraphicContext& graphics);
+	~ImeOverlay();
+	KYTY_CLASS_NO_COPY(ImeOverlay);
+
+	[[nodiscard]] bool PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count);
+	void               Record(vk::CommandBuffer command, vk::ImageView target);
+	void               ReleaseVulkan();
+
+private:
+	struct Impl;
+	std::unique_ptr<Impl> m_impl;
+};
+
+} // namespace Libs::Graphics
+
+#endif // EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_

src/graphics/presentation/presenter.h

@@ -25,6 +25,7 @@ public:
 	                                               CommandBuffer* producer = nullptr);
 	[[nodiscard]] Frame*         PrepareLastFrame();
 	[[nodiscard]] bool           IsGuestPaused() const noexcept;
+	[[nodiscard]] bool           NeedsImeRefresh() const noexcept;
 	[[nodiscard]] RenderContext& Renderer() const noexcept;
 	void                         Present(Frame& frame, bool reuse = false);
 	void                         Discard(Frame& frame);

src/graphics/presentation/renderDoc.cpp

@@ -128,8 +128,7 @@ static bool BindRenderDocApi(HMODULE module) {
 	g_api    = static_cast<RenderDocApi*>(api);
 
 	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
-	static RenderDocInputButton capture_keys[] = {eRENDERDOC_Key_F1};
-	g_api->SetCaptureKeys(capture_keys, 1);
+	g_api->SetCaptureKeys(nullptr, 0);
 	g_api->UnloadCrashHandler();
 
 	char module_path[MAX_PATH] = {};
@@ -170,8 +169,7 @@ static bool BindRenderDocApi(HMODULE module) {
 	g_api    = static_cast<RenderDocApi*>(api);
 
 	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
-	static RenderDocInputButton capture_keys[] = {eRENDERDOC_Key_F1};
-	g_api->SetCaptureKeys(capture_keys, 1);
+	g_api->SetCaptureKeys(nullptr, 0);
 	g_api->UnloadCrashHandler();
 
 	Dl_info info {};

src/graphics/presentation/videoOut.cpp

@@ -537,10 +537,10 @@ Graphics::ImageInfo BufferAttributeGroup::ImageInfo(const VideoOutBuffer& buffer
 	const auto tile_mode =
 	    Graphics::Prospero::GpuEnumValue(Graphics::Prospero::TileMode::kRenderTarget);
 	const auto pitch =
-	    Graphics::TileGetTexturePitch(pixel_format.guest_format, attribute.width, 1, tile_mode);
+	    Graphics::TileGetTexturePitch(pixel_format.guest_format, attribute.width, tile_mode);
 	Graphics::TileSizeAlign total {};
 	Graphics::TileGetTextureTotalSize(pixel_format.guest_format, attribute.width, attribute.height,
-	                                  1, pitch, 1, tile_mode, false, total);
+	                                  1, 1, tile_mode, false, total);
 	if (total.size == 0 || total.align != 65536 ||
 	    (buffer.data_address & (total.align - 1u)) != 0) {
 		EXIT("invalid video-out surface footprint or alignment\n");
@@ -788,14 +788,10 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 	while (!token.stop_requested()) {
 		const auto sleep_begin = Common::Timer::QueryPerformanceCounter();
 		if (total_wait > 0) {
-			const auto sleep_end = sleep_begin + static_cast<uint64_t>(total_wait);
-			auto       now       = sleep_begin;
-			while (!token.stop_requested() && now < sleep_end) {
-				const auto remaining_us = (sleep_end - now) * 1000000u / frequency;
-				Common::Thread::SleepMicro(
-				    static_cast<uint32_t>(std::clamp<uint64_t>(remaining_us, 1, 1000)));
-				now = Common::Timer::QueryPerformanceCounter();
-			}
+			const auto remaining_us =
+			    (static_cast<uint64_t>(total_wait) * 1000000u + frequency - 1) / frequency;
+			Common::Thread::SleepMicro(static_cast<uint32_t>(
+			    std::clamp<uint64_t>(remaining_us, 1, std::numeric_limits<uint32_t>::max())));
 		}
 		if (token.stop_requested()) {
 			break;
@@ -817,7 +813,24 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 		}
 
 		VblankBegin();
-		const bool presented = m_flip_queue.Flip(0);
+		bool presented = m_flip_queue.Flip(0);
+		if (!presented && m_presenter.NeedsImeRefresh()) {
+			if (auto* frame = m_presenter.PrepareLastFrame(); frame != nullptr) {
+				m_presenter.Present(*frame, true);
+				presented = true;
+			} else {
+				uint32_t width  = 0;
+				uint32_t height = 0;
+				{
+					Common::LockGuard lock(m_mutex);
+					width  = m_video_out_ctx[0].width;
+					height = m_video_out_ctx[0].height;
+				}
+				auto& blank = m_presenter.PrepareBlankFrame(width, height, true);
+				m_presenter.Present(blank);
+				presented = true;
+			}
+		}
 		if (!presented && total_wait < 0) {
 			bool     any_open = false;
 			uint32_t width    = 0;

src/graphics/presentation/window/hostInput.cpp

@@ -0,0 +1,305 @@
+#include "graphics/presentation/window/hostInput.h"
+
+#include "SDL_keyboard.h"
+#include "SDL_keycode.h"
+#include "SDL_mouse.h"
+#include "common/assert.h"
+#include "common/emulatorConfig.h"
+#include "libs/controller.h"
+
+#include <algorithm>
+#include <array>
+#include <cstddef>
+#include <string>
+#include <string_view>
+#include <utility>
+
+namespace Libs::Graphics {
+
+namespace {
+
+struct ControlInfo {
+	std::string_view name;
+	uint32_t         button   = 0;
+	Controller::Axis axis     = Controller::Axis::AxisMax;
+	bool             positive = false;
+};
+
+static constexpr std::array CONTROL_INFO = {
+    ControlInfo {"L3", Controller::PAD_BUTTON_L3},
+    ControlInfo {"R3", Controller::PAD_BUTTON_R3},
+    ControlInfo {"Options", Controller::PAD_BUTTON_OPTIONS},
+    ControlInfo {"Up", Controller::PAD_BUTTON_UP},
+    ControlInfo {"Right", Controller::PAD_BUTTON_RIGHT},
+    ControlInfo {"Down", Controller::PAD_BUTTON_DOWN},
+    ControlInfo {"Left", Controller::PAD_BUTTON_LEFT},
+    ControlInfo {"L2", Controller::PAD_BUTTON_L2},
+    ControlInfo {"R2", Controller::PAD_BUTTON_R2},
+    ControlInfo {"L1", Controller::PAD_BUTTON_L1},
+    ControlInfo {"R1", Controller::PAD_BUTTON_R1},
+    ControlInfo {"Triangle", Controller::PAD_BUTTON_TRIANGLE},
+    ControlInfo {"Circle", Controller::PAD_BUTTON_CIRCLE},
+    ControlInfo {"Cross", Controller::PAD_BUTTON_CROSS},
+    ControlInfo {"Square", Controller::PAD_BUTTON_SQUARE},
+    ControlInfo {"TouchPad", Controller::PAD_BUTTON_TOUCH_PAD},
+    ControlInfo {"LeftStickLeft", 0, Controller::Axis::LeftX, false},
+    ControlInfo {"LeftStickRight", 0, Controller::Axis::LeftX, true},
+    ControlInfo {"LeftStickUp", 0, Controller::Axis::LeftY, false},
+    ControlInfo {"LeftStickDown", 0, Controller::Axis::LeftY, true},
+    ControlInfo {"RightStickLeft", 0, Controller::Axis::RightX, false},
+    ControlInfo {"RightStickRight", 0, Controller::Axis::RightX, true},
+    ControlInfo {"RightStickUp", 0, Controller::Axis::RightY, false},
+    ControlInfo {"RightStickDown", 0, Controller::Axis::RightY, true},
+};
+
+constexpr std::size_t INVALID_CONTROL = CONTROL_INFO.size();
+
+struct Binding {
+	SDL_Keycode key          = SDLK_UNKNOWN;
+	uint8_t     mouse_button = 0;
+	std::size_t control      = INVALID_CONTROL;
+};
+
+std::size_t ControlFromName(std::string_view name) {
+	const auto info = std::find_if(CONTROL_INFO.begin(), CONTROL_INFO.end(),
+	                               [name](const auto& item) { return item.name == name; });
+	return static_cast<std::size_t>(std::distance(CONTROL_INFO.begin(), info));
+}
+
+const ControlInfo& Info(std::size_t control) {
+	EXIT_IF(control >= CONTROL_INFO.size());
+	return CONTROL_INFO[control];
+}
+
+SDL_Keycode NormalizeKey(SDL_Keycode key) {
+	switch (key) {
+		case SDLK_RSHIFT: return SDLK_LSHIFT;
+		case SDLK_RCTRL: return SDLK_LCTRL;
+		case SDLK_RALT: return SDLK_LALT;
+		case SDLK_RGUI: return SDLK_LGUI;
+		default: return key;
+	}
+}
+
+uint8_t MouseButtonFromName(std::string_view name) {
+	static constexpr std::array buttons = {
+	    std::pair {std::string_view("Mouse:Left"), uint8_t {SDL_BUTTON_LEFT}},
+	    std::pair {std::string_view("Mouse:Middle"), uint8_t {SDL_BUTTON_MIDDLE}},
+	    std::pair {std::string_view("Mouse:Right"), uint8_t {SDL_BUTTON_RIGHT}},
+	    std::pair {std::string_view("Mouse:X1"), uint8_t {SDL_BUTTON_X1}},
+	    std::pair {std::string_view("Mouse:X2"), uint8_t {SDL_BUTTON_X2}},
+	};
+
+	const auto button = std::find_if(buttons.begin(), buttons.end(),
+	                                 [name](const auto& item) { return item.first == name; });
+	return button != buttons.end() ? button->second : 0;
+}
+
+bool Conflicts(const Binding& first, const Binding& second) {
+	return first.control == second.control ||
+	       (first.key != SDLK_UNKNOWN && first.key == second.key) ||
+	       (first.mouse_button != 0 && first.mouse_button == second.mouse_button);
+}
+
+class InputMap {
+public:
+	InputMap(): m_custom(!Config::GetKeymap().empty()) {
+		for (const auto& value: Config::GetKeymap()) {
+			const std::string_view entry = value;
+			const auto             split = entry.find('=');
+
+			Binding binding;
+			if (split != std::string_view::npos) {
+				binding.control       = ControlFromName(entry.substr(0, split));
+				const auto host_input = entry.substr(split + 1);
+				binding.mouse_button  = MouseButtonFromName(host_input);
+				if (binding.mouse_button == 0) {
+					binding.key = NormalizeKey(SDL_GetKeyFromName(std::string(host_input).c_str()));
+				}
+			}
+
+			const bool reserved =
+			    binding.key == SDLK_ESCAPE || binding.key == SDLK_SPACE || binding.key == SDLK_F1;
+			if (binding.control == INVALID_CONTROL || reserved ||
+			    (binding.key == SDLK_UNKNOWN && binding.mouse_button == 0)) {
+				EXIT("Invalid input mapping: %s\n", value.c_str());
+			}
+			Add(binding);
+		}
+	}
+	[[nodiscard]] bool Custom() const { return m_custom; }
+
+	[[nodiscard]] std::size_t FindKey(int key_code) const {
+		key_code = NormalizeKey(static_cast<SDL_Keycode>(key_code));
+		const auto binding =
+		    std::find_if(m_bindings.begin(), m_bindings.begin() + m_size,
+		                 [key_code](const auto& item) { return item.key == key_code; });
+		return binding != m_bindings.begin() + m_size ? binding->control : INVALID_CONTROL;
+	}
+
+	[[nodiscard]] std::size_t FindMouseButton(uint8_t mouse_button) const {
+		const auto binding = std::find_if(
+		    m_bindings.begin(), m_bindings.begin() + m_size,
+		    [mouse_button](const auto& item) { return item.mouse_button == mouse_button; });
+		return binding != m_bindings.begin() + m_size ? binding->control : INVALID_CONTROL;
+	}
+
+private:
+	void Add(const Binding& binding) {
+		for (std::size_t index = 0; index < m_size;) {
+			if (Conflicts(m_bindings[index], binding)) {
+				m_bindings[index] = m_bindings[--m_size];
+			} else {
+				index++;
+			}
+		}
+		EXIT_IF(m_size >= m_bindings.size());
+		m_bindings[m_size++] = binding;
+	}
+
+	std::array<Binding, CONTROL_INFO.size()> m_bindings {};
+	std::size_t                              m_size = 0;
+	bool                                     m_custom;
+};
+
+const InputMap& GetInputMap() {
+	static const InputMap map;
+	return map;
+}
+
+void SetButton(uint32_t button, bool down) {
+	if (button == Controller::PAD_BUTTON_L2) {
+		Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID,
+		                           Controller::Axis::TriggerLeft, down ? 255 : 0);
+	} else if (button == Controller::PAD_BUTTON_R2) {
+		Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID,
+		                           Controller::Axis::TriggerRight, down ? 255 : 0);
+	} else if (button != 0) {
+		Controller::ControllerButton(Controller::HOST_INPUT_CONTROLLER_ID, button, down);
+	}
+}
+
+uint32_t DefaultKeyboardButton(int key_code) {
+	switch (NormalizeKey(static_cast<SDL_Keycode>(key_code))) {
+		case SDLK_UP: return Controller::PAD_BUTTON_UP;
+		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_LSHIFT: return Controller::PAD_BUTTON_L3;
+		case SDLK_LCTRL: return Controller::PAD_BUTTON_R3;
+		case SDLK_RETURN:
+		case SDLK_RETURN2: return Controller::PAD_BUTTON_OPTIONS;
+		case SDLK_BACKSPACE:
+		case SDLK_TAB: return Controller::PAD_BUTTON_TOUCH_PAD;
+		default: return 0;
+	}
+}
+
+struct StickKeys {
+	bool left  = false;
+	bool right = false;
+	bool up    = false;
+	bool down  = false;
+};
+
+void SetStickAxis(Controller::Axis axis, bool negative, bool positive) {
+	const int value = negative == positive ? 128 : negative ? 0 : 255;
+	Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID, axis, value);
+}
+
+void SetControl(std::size_t control, bool down) {
+	if (control == INVALID_CONTROL) {
+		return;
+	}
+
+	const auto& info = Info(control);
+	if (info.button != 0) {
+		SetButton(info.button, down);
+		return;
+	}
+
+	struct AxisKeys {
+		bool negative = false;
+		bool positive = false;
+	};
+	static std::array<AxisKeys, 4> axes;
+
+	const auto axis = static_cast<std::size_t>(info.axis);
+	EXIT_IF(axis >= axes.size());
+	if (info.positive) {
+		axes[axis].positive = down;
+	} else {
+		axes[axis].negative = down;
+	}
+	SetStickAxis(info.axis, axes[axis].negative, axes[axis].positive);
+}
+
+void DefaultKeyboardInput(int key_code, bool down) {
+	static StickKeys left;
+	static StickKeys right;
+
+	switch (NormalizeKey(static_cast<SDL_Keycode>(key_code))) {
+		case SDLK_a:
+			left.left = down;
+			SetStickAxis(Controller::Axis::LeftX, left.left, left.right);
+			return;
+		case SDLK_d:
+			left.right = down;
+			SetStickAxis(Controller::Axis::LeftX, left.left, left.right);
+			return;
+		case SDLK_w:
+			left.up = down;
+			SetStickAxis(Controller::Axis::LeftY, left.up, left.down);
+			return;
+		case SDLK_s:
+			left.down = down;
+			SetStickAxis(Controller::Axis::LeftY, left.up, left.down);
+			return;
+		case SDLK_f:
+			right.left = down;
+			SetStickAxis(Controller::Axis::RightX, right.left, right.right);
+			return;
+		case SDLK_h:
+			right.right = down;
+			SetStickAxis(Controller::Axis::RightX, right.left, right.right);
+			return;
+		case SDLK_t:
+			right.up = down;
+			SetStickAxis(Controller::Axis::RightY, right.up, right.down);
+			return;
+		case SDLK_g:
+			right.down = down;
+			SetStickAxis(Controller::Axis::RightY, right.up, right.down);
+			return;
+		default: SetButton(DefaultKeyboardButton(key_code), down); return;
+	}
+}
+} // namespace
+
+void HostInputInit() {
+	GetInputMap();
+}
+
+void HostInputKey(int key_code, bool down) {
+	const auto& map = GetInputMap();
+	if (map.Custom()) {
+		SetControl(map.FindKey(key_code), down);
+	} else {
+		DefaultKeyboardInput(key_code, down);
+	}
+}
+
+void HostInputMouseButton(uint8_t mouse_button, bool down) {
+	const auto& map = GetInputMap();
+	if (map.Custom() && mouse_button != 0) {
+		SetControl(map.FindMouseButton(mouse_button), down);
+	}
+}
+
+} // namespace Libs::Graphics

src/graphics/presentation/window/hostInput.h

@@ -0,0 +1,14 @@
+#ifndef KYTY_GRAPHICS_PRESENTATION_WINDOW_HOST_INPUT_H_
+#define KYTY_GRAPHICS_PRESENTATION_WINDOW_HOST_INPUT_H_
+
+#include <cstdint>
+
+namespace Libs::Graphics {
+
+void HostInputInit();
+void HostInputKey(int key_code, bool down);
+void HostInputMouseButton(uint8_t mouse_button, bool down);
+
+} // namespace Libs::Graphics
+
+#endif /* KYTY_GRAPHICS_PRESENTATION_WINDOW_HOST_INPUT_H_ */

src/graphics/presentation/window/swapchain.cpp

@@ -31,6 +31,7 @@
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/presentation/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
@@ -329,8 +330,9 @@ public:
 	void                 Create();
 	void                 Recreate(bool surface_lost = false);
 	[[nodiscard]] Status AcquireNextImage();
-	void                 RecordPresentCommands(CommandBuffer& command, VulkanImage& source);
-	void                 Submit(CommandBuffer& command);
+	[[nodiscard]] bool   PrepareImeOverlay();
+	void RecordPresentCommands(CommandBuffer& command, VulkanImage& source, bool draw_ime_overlay);
+	void Submit(CommandBuffer& command);
 	[[nodiscard]] Status Present();
 
 	[[nodiscard]] uint32_t ImageCount() const noexcept {
@@ -342,16 +344,17 @@ private:
 	void Destroy();
 	void RefreshSurfaceSize();
 
-	WindowContext&             m_window;
-	vk::SwapchainKHR           m_handle = nullptr;
-	vk::Format                 m_format = vk::Format::eUndefined;
-	vk::Extent2D               m_extent {};
-	std::vector<vk::Image>     m_images;
-	std::vector<vk::ImageView> m_image_views;
-	std::vector<vk::Semaphore> m_image_acquired;
-	std::vector<vk::Semaphore> m_render_complete;
-	uint32_t                   m_image_index = static_cast<uint32_t>(-1);
-	uint32_t                   m_frame_index = 0;
+	WindowContext&              m_window;
+	vk::SwapchainKHR            m_handle = nullptr;
+	vk::Format                  m_format = vk::Format::eUndefined;
+	vk::Extent2D                m_extent {};
+	std::vector<vk::Image>      m_images;
+	std::vector<vk::ImageView>  m_image_views;
+	std::vector<vk::Semaphore>  m_image_acquired;
+	std::vector<vk::Semaphore>  m_render_complete;
+	std::unique_ptr<ImeOverlay> m_ime_overlay;
+	uint32_t                    m_image_index = static_cast<uint32_t>(-1);
+	uint32_t                    m_frame_index = 0;
 };
 
 struct Presenter::Impl {
@@ -383,17 +386,20 @@ struct Presenter::Impl {
 		desc.view_info.usage       = vk::ImageUsageFlagBits::eTransferSrc;
 		desc.type                  = TextureCache::BindingType::VideoOut;
 
-		auto& cache           = renderer.GetTextureCache();
-		auto& image           = cache.GetImage(cache.FindImage(desc));
+		auto&      cache      = renderer.GetTextureCache();
+		const auto image_id   = cache.FindImage(desc);
+		auto&      image      = cache.GetImage(image_id);
 		image.usage.video_out = true;
+		cache.UpdateImage(image_id);
 		return image;
 	}
 
-	RenderContext&   renderer;
-	WindowContext&   window;
-	Swapchain        swapchain;
-	CommandScheduler present_scheduler;
-	FramePool        frames;
+	RenderContext&        renderer;
+	WindowContext&        window;
+	Swapchain             swapchain;
+	CommandScheduler      present_scheduler;
+	FramePool             frames;
+	std::atomic<uint64_t> presented_ime_revision {0};
 };
 
 void Swapchain::Create() {
@@ -522,6 +528,9 @@ void Swapchain::Destroy() {
 		Common::LockGuard queue_lock(graphics.queue_mutex);
 		RequireVulkanSuccess(graphics.queue.waitIdle(), "wait for swapchain queue");
 	}
+	if (m_ime_overlay != nullptr) {
+		m_ime_overlay->ReleaseVulkan();
+	}
 
 	for (const auto semaphore: m_image_acquired) {
 		if (semaphore != nullptr) {
@@ -606,7 +615,15 @@ Swapchain::Status Swapchain::AcquireNextImage() {
 	return Status::Success;
 }
 
-void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& source) {
+bool Swapchain::PrepareImeOverlay() {
+	if (m_ime_overlay == nullptr) {
+		m_ime_overlay = std::make_unique<ImeOverlay>(m_window.graphic_ctx);
+	}
+	return m_ime_overlay->PrepareFrame(m_extent, m_format, ImageCount());
+}
+
+void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& source,
+                                      bool draw_ime_overlay) {
 	if (source.state.layout != vk::ImageLayout::eTransferSrcOptimal) {
 		EXIT("invalid prepared presentation image, vk_image=%p layout=%d\n",
 		     static_cast<void*>(source.image), static_cast<int>(source.state.layout));
@@ -652,11 +669,14 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	                     vk::Filter::eLinear);
 
 	vk::ImageMemoryBarrier to_present {};
-	to_present.sType                           = vk::StructureType::eImageMemoryBarrier;
-	to_present.srcAccessMask                   = vk::AccessFlagBits::eTransferWrite;
-	to_present.dstAccessMask                   = vk::AccessFlagBits::eMemoryRead;
-	to_present.oldLayout                       = vk::ImageLayout::eTransferDstOptimal;
-	to_present.newLayout                       = vk::ImageLayout::ePresentSrcKHR;
+	to_present.sType         = vk::StructureType::eImageMemoryBarrier;
+	to_present.srcAccessMask = vk::AccessFlagBits::eTransferWrite;
+	to_present.dstAccessMask = draw_ime_overlay ? vk::AccessFlagBits::eColorAttachmentRead |
+	                                                  vk::AccessFlagBits::eColorAttachmentWrite
+	                                            : vk::AccessFlagBits::eMemoryRead;
+	to_present.oldLayout     = vk::ImageLayout::eTransferDstOptimal;
+	to_present.newLayout     = draw_ime_overlay ? vk::ImageLayout::eColorAttachmentOptimal
+	                                            : vk::ImageLayout::ePresentSrcKHR;
 	to_present.srcQueueFamilyIndex             = VK_QUEUE_FAMILY_IGNORED;
 	to_present.dstQueueFamilyIndex             = VK_QUEUE_FAMILY_IGNORED;
 	to_present.image                           = m_images[m_image_index];
@@ -665,9 +685,22 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	to_present.subresourceRange.levelCount     = 1;
 	to_present.subresourceRange.baseArrayLayer = 0;
 	to_present.subresourceRange.layerCount     = 1;
-	vk_command.pipelineBarrier(
-	    vk::PipelineStageFlagBits::eAllCommands, vk::PipelineStageFlagBits::eAllCommands,
-	    vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1, &to_present);
+	vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
+	                           draw_ime_overlay ? vk::PipelineStageFlagBits::eColorAttachmentOutput
+	                                            : vk::PipelineStageFlagBits::eAllCommands,
+	                           vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1,
+	                           &to_present);
+	if (draw_ime_overlay) {
+		m_ime_overlay->Record(vk_command, m_image_views[m_image_index]);
+		to_present.srcAccessMask = vk::AccessFlagBits::eColorAttachmentWrite;
+		to_present.dstAccessMask = vk::AccessFlagBits::eMemoryRead;
+		to_present.oldLayout     = vk::ImageLayout::eColorAttachmentOptimal;
+		to_present.newLayout     = vk::ImageLayout::ePresentSrcKHR;
+		vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eColorAttachmentOutput,
+		                           vk::PipelineStageFlagBits::eAllCommands,
+		                           vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1,
+		                           &to_present);
+	}
 	command.End();
 }
 
@@ -772,6 +805,12 @@ bool Presenter::IsGuestPaused() const noexcept {
 	return m_impl->window.loop.paused.load(std::memory_order_acquire);
 }
 
+bool Presenter::NeedsImeRefresh() const noexcept {
+	const auto visual = GetImeVisualState();
+	return visual.active ||
+	       visual.revision != m_impl->presented_ime_revision.load(std::memory_order_acquire);
+}
+
 RenderContext& Presenter::Renderer() const noexcept {
 	return m_impl->renderer;
 }
@@ -789,19 +828,22 @@ void Presenter::Present(Frame& frame, bool reuse) {
 		    [&window] {
 			    window.UpdateIcon();
 			    SDL_ShowWindow(window.window);
+			    SDL_RaiseWindow(window.window);
 		    },
 		    true);
 #else
 		window.UpdateIcon();
 
 		SDL_ShowWindow(window.window);
+		SDL_RaiseWindow(window.window);
 #endif
 
 		window.window_hidden = false;
 		m_impl->RecoverSwapchain(Swapchain::Status::Recreate);
 	}
 
-	auto& swapchain = m_impl->swapchain;
+	const auto ime_visual = GetImeVisualState();
+	auto&      swapchain  = m_impl->swapchain;
 	for (uint32_t attempt = 0; attempt < 2; attempt++) {
 		auto status = swapchain.AcquireNextImage();
 		if (status != Swapchain::Status::Success) {
@@ -814,8 +856,9 @@ void Presenter::Present(Frame& frame, bool reuse) {
 		{
 			Common::LockGuard render_lock(m_impl->renderer.GetMutex());
 			frame.present_commands->WaitForFenceAndReset();
-			auto& command = *frame.present_commands;
-			swapchain.RecordPresentCommands(command, frame.image);
+			auto&      command          = *frame.present_commands;
+			const bool draw_ime_overlay = ime_visual.active && swapchain.PrepareImeOverlay();
+			swapchain.RecordPresentCommands(command, frame.image, draw_ime_overlay);
 			swapchain.Submit(command);
 		}
 		status = swapchain.Present();
@@ -825,6 +868,7 @@ void Presenter::Present(Frame& frame, bool reuse) {
 		}
 
 		RenderDocOnPresent();
+		m_impl->presented_ime_revision.store(ime_visual.revision, std::memory_order_release);
 		window.UpdateTitle();
 		m_impl->frames.Release(&frame, true);
 		return;

src/graphics/presentation/window/vulkanWindow.cpp

@@ -31,6 +31,7 @@
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/presentation/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
@@ -259,6 +260,10 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("sampleRateShading is not supported\n");
 			skip_device = true;
 		}
+		if (device_features2.features.depthBiasClamp != VK_TRUE) {
+			LOGF("depthBiasClamp is not supported\n");
+			skip_device = true;
+		}
 
 		if (device_features2.features.fragmentStoresAndAtomics != VK_TRUE) {
 			LOGF("fragmentStoresAndAtomics is not supported\n");
@@ -548,6 +553,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	EXIT_NOT_IMPLEMENTED(required_features13.synchronization2 == VK_TRUE &&
 	                     supported_features13.synchronization2 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.sampleRateShading != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.depthBiasClamp != VK_TRUE);
 	features12.timelineSemaphore = VK_TRUE;
 
 	vk::PhysicalDeviceFeatures device_features {};
@@ -563,6 +569,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.independentBlend                     = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
+	device_features.depthBiasClamp                       = VK_TRUE;
 	graphics.sample_rate_shading_enabled                 = true;
 	device_features.vertexPipelineStoresAndAtomics =
 	    supported_features2.features.vertexPipelineStoresAndAtomics;
@@ -600,10 +607,6 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	create_info.flags                = {};
 	create_info.pQueueCreateInfos    = &queue_create_info;
 	create_info.queueCreateInfoCount = 1;
-	create_info.enabledLayerCount =
-	    (r.enable_validation_layers ? static_cast<uint32_t>(r.required_layers.size()) : 0);
-	create_info.ppEnabledLayerNames =
-	    (r.enable_validation_layers ? r.required_layers.data() : nullptr);
 	create_info.enabledExtensionCount   = static_cast<uint32_t>(device_extensions.size());
 	create_info.ppEnabledExtensionNames = device_extensions.data();
 	create_info.pEnabledFeatures        = &device_features;
@@ -1008,6 +1011,7 @@ void WindowContext::RecreateSurface() {
 }
 
 WindowContext::~WindowContext() {
+	ShutdownImeInput();
 	presenter.reset();
 	LibKernel::Memory::InstallGpuResources(nullptr);
 	render_context.reset();

src/graphics/presentation/window/window.cpp

@@ -31,7 +31,9 @@
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/presentation/imeOverlay.h"
 #include "graphics/presentation/renderDoc.h"
+#include "graphics/presentation/window/hostInput.h"
 #include "graphics/presentation/window/windowInternal.h"
 #include "kytyGitVersion.h"
 #include "libs/controller.h"
@@ -59,8 +61,6 @@
 
 namespace Libs::Graphics {
 
-constexpr int KEYBOARD_CONTROLLER_ID = -1000;
-
 struct EventKeyboard {
 	bool     down;
 	bool     up;
@@ -73,26 +73,6 @@ struct EventKeyboard {
 	double   timestamp_seconds;
 };
 
-static uint32_t KeyboardKeyToPadButton(int key_code) {
-	switch (key_code) {
-		case SDLK_w: return Controller::PAD_BUTTON_UP;
-		case SDLK_a: return Controller::PAD_BUTTON_LEFT;
-		case SDLK_s: return Controller::PAD_BUTTON_DOWN;
-		case SDLK_d: 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_RETURN:
-		case SDLK_RETURN2: return Controller::PAD_BUTTON_OPTIONS;
-		case SDLK_BACKSPACE:
-		case SDLK_TAB: return Controller::PAD_BUTTON_TOUCH_PAD;
-		default: return 0;
-	}
-}
-
 static uint32_t ControllerButtonToPadButton(int button) {
 	switch (button) {
 		case SDL_CONTROLLER_BUTTON_A: return Controller::PAD_BUTTON_CROSS;
@@ -261,16 +241,11 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 		}
 	}
 
-	const auto button = KeyboardKeyToPadButton(key.key_code);
-	if (button != 0 && (key.down || key.up) && !key.repeat) {
-		static bool keyboard_connected = false;
-		if (!keyboard_connected) {
-			Controller::ControllerConnect(KEYBOARD_CONTROLLER_ID);
-			keyboard_connected = true;
-		}
-		Controller::ControllerButton(KEYBOARD_CONTROLLER_ID, button, key.down);
-	}
 #endif
+
+	if ((key.down || key.up) && !key.repeat) {
+		HostInputKey(key.key_code, key.down);
+	}
 }
 
 static void GameEventMouse([[maybe_unused]] const EventMouse& mb) {
@@ -292,6 +267,23 @@ static void GameEventMouse([[maybe_unused]] const EventMouse& mb) {
 		     mb.x, mb.y);
 	}
 #endif
+
+	if (mb.down || mb.up) {
+		uint8_t mouse_button = 0;
+		if (mb.left) {
+			mouse_button = SDL_BUTTON_LEFT;
+		} else if (mb.middle) {
+			mouse_button = SDL_BUTTON_MIDDLE;
+		} else if (mb.right) {
+			mouse_button = SDL_BUTTON_RIGHT;
+		} else if (mb.x1) {
+			mouse_button = SDL_BUTTON_X1;
+		} else if (mb.x2) {
+			mouse_button = SDL_BUTTON_X2;
+		}
+
+		HostInputMouseButton(mouse_button, mb.down);
+	}
 }
 
 static void GameEventFinger([[maybe_unused]] const EventFinger& f) {
@@ -485,6 +477,9 @@ void WindowContext::ProcessEvent(double time_s) {
 	auto& game  = loop;
 	auto* event = &game.event;
 	EXIT_IF(SDL_GetEventState(SDL_DISPLAYEVENT) != SDL_ENABLE);
+	if (ProcessImeInput(*event)) {
+		return;
+	}
 
 	switch (event->type) {
 		case SDL_QUIT: GameEventQuit(game); break;
@@ -757,26 +752,18 @@ void WindowContext::Run() {
 #if defined(__APPLE__)
 		DrainMainThreadTasks();
 #endif
-		if (SDL_PollEvent(&loop.event) != 0) {
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
 		if (loop.paused.load(std::memory_order_acquire)) {
 			if (!timer.IsPaused()) {
 				timer.Pause();
 			}
-			if (SDL_WaitEvent(&loop.event) == 0) {
-				EXIT("%s\n", SDL_GetError());
-			}
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
-		if (timer.IsPaused()) {
+		} else if (timer.IsPaused()) {
 			timer.Resume();
 		}
-		Common::Thread::SleepMicro(1000);
+
+		if (SDL_WaitEvent(&loop.event) == 0) {
+			EXIT("%s\n", SDL_GetError());
+		}
+		ProcessEvent(timer.GetTimeS());
 	}
 }
 
@@ -795,10 +782,15 @@ static void WindowCreate(WindowContext& context) {
 	if (SDL_InitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMECONTROLLER) < 0) {
 		EXIT("%s\n", SDL_GetError());
 	}
+	HostInputInit();
+	InitializeImeInput();
 
 	LOGF("WindowCreate(): width = %d, height = %d\n", width, height);
 
 	uint32_t window_flags = KYTY_SDL_WINDOW_FLAGS;
+	if (Config::FullscreenEnabled()) {
+		window_flags |= static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP);
+	}
 #if defined(__APPLE__)
 	// macOS 26 window chrome (CoreUI asset decode, SwiftUI titlebar) has been observed
 	// throwing NSExceptions under Rosetta during the first CATransaction commit. A
@@ -946,14 +938,6 @@ void WindowContext::UpdateTitle() {
 	static constexpr auto build_type = "Unknown";
 #endif
 
-#if defined(KYTY_OFFICIAL_BUILD)
-	static constexpr auto build_label = "Official build " KYTY_RELEASE_TAG;
-#elif defined(KYTY_FORK_BUILD)
-	static constexpr auto build_label = "Fork build " KYTY_BUILD_REPOSITORY " " KYTY_GIT_HASH;
-#else
-	static constexpr auto build_label = "Source build " KYTY_GIT_HASH;
-#endif
-
 	const auto now       = Common::Timer::QueryPerformanceCounter();
 	const auto frequency = Common::Timer::QueryPerformanceFrequency();
 	frame_num++;
@@ -965,11 +949,11 @@ void WindowContext::UpdateTitle() {
 		fps_frames  = 0;
 	}
 
-	auto fps = fmt::format("[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", build_label,
-	                       build_type, (has_title ? title : ""), (has_title ? ", " : ""),
-	                       (has_title_id ? title_id : ""), (has_title_id ? ", " : ""),
-	                       (has_app_ver ? app_ver : ""), (has_app_ver ? " " : ""), device_name,
-	                       processor_name, frame_num, current_fps);
+	auto fps = fmt::format(
+	    "[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", KYTY_BUILD_LABEL, build_type,
+	    (has_title ? title : ""), (has_title ? ", " : ""), (has_title_id ? title_id : ""),
+	    (has_title_id ? ", " : ""), (has_app_ver ? app_ver : ""), (has_app_ver ? " " : ""),
+	    device_name, processor_name, frame_num, current_fps);
 
 #if defined(__APPLE__)
 	// AppKit traps on title changes off the main thread; fire-and-forget keeps present pacing.

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -84,6 +84,8 @@ bool InstructionMaySplitSpirvBlock(const IR::Instruction& inst) {
 		case IR::Opcode::AtomicAndU32:
 		case IR::Opcode::AtomicOrU32:
 		case IR::Opcode::AtomicXorU32:
+		case IR::Opcode::AtomicFMinF32:
+		case IR::Opcode::AtomicFMaxF32:
 		case IR::Opcode::FlatLoadUbyte:
 		case IR::Opcode::FlatLoadSbyte:
 		case IR::Opcode::FlatLoadUshort:

src/graphics/shader/recompiler/decompiler/MemoryOps.cpp

@@ -57,6 +57,8 @@ constexpr MemoryOpcodeInfo MUBUF_OPS[] = {
     {0x39u, Opcode::BufferAtomicAnd, 1, 32},
     {0x3au, Opcode::BufferAtomicOr, 1, 32},
     {0x3bu, Opcode::BufferAtomicXor, 1, 32},
+    {0x3fu, Opcode::BufferAtomicFMin, 1, 32},
+    {0x40u, Opcode::BufferAtomicFMax, 1, 32},
 };
 
 constexpr MemoryOpcodeInfo MTBUF_OPS[] = {

src/graphics/shader/recompiler/decompiler/OpcodeTable.h

@@ -0,0 +1,45 @@
+#ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_DECOMPILER_OPCODETABLE_H_
+#define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_DECOMPILER_OPCODETABLE_H_
+
+#include "graphics/shader/recompiler/decompiler/ShaderDecoder.h"
+
+#include <cstddef>
+
+namespace Libs::Graphics::ShaderRecompiler::Decoder::Detail {
+
+struct OpcodeMap {
+	uint32_t encoding = 0;
+	Opcode   decoded  = Opcode::Unknown;
+};
+
+template <typename Entry, size_t N>
+constexpr const Entry* FindOpcode(const Entry (&table)[N], uint32_t encoding) {
+	for (const auto& entry: table) {
+		if (entry.encoding == encoding) {
+			return &entry;
+		}
+	}
+	return nullptr;
+}
+
+template <typename Entry, size_t N>
+constexpr Opcode LookupOpcode(const Entry (&table)[N], uint32_t encoding) {
+	const auto* entry = FindOpcode(table, encoding);
+	return entry != nullptr ? entry->decoded : Opcode::Unsupported;
+}
+
+template <typename Entry, size_t N>
+constexpr bool HasUniqueEncodings(const Entry (&table)[N]) {
+	for (size_t i = 0; i < N; i++) {
+		for (size_t j = i + 1; j < N; j++) {
+			if (table[i].encoding == table[j].encoding) {
+				return false;
+			}
+		}
+	}
+	return true;
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Decoder::Detail
+
+#endif // EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_DECOMPILER_OPCODETABLE_H_

src/graphics/shader/recompiler/decompiler/ScalarAluOps.cpp

@@ -1,14 +1,11 @@
 #include "graphics/shader/recompiler/decompiler/ScalarAluOps.h"
 
-#include <iterator>
+#include "graphics/shader/recompiler/decompiler/OpcodeTable.h"
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
 
-struct OpcodeMap {
-	uint32_t opcode = 0;
-	Opcode   ir     = Opcode::Unknown;
-};
+using Detail::OpcodeMap;
 
 constexpr OpcodeMap SOP2_OPS[] = {
     {0x00u, Opcode::SAddU32},       {0x01u, Opcode::SSubU32},       {0x02u, Opcode::SAddI32},
@@ -58,7 +55,8 @@ constexpr OpcodeMap SOPC_OPS[] = {
     {0x03u, Opcode::SCmpGeI32},   {0x04u, Opcode::SCmpLtI32},   {0x05u, Opcode::SCmpLeI32},
     {0x06u, Opcode::SCmpEqU32},   {0x07u, Opcode::SCmpLgU32},   {0x08u, Opcode::SCmpGtU32},
     {0x09u, Opcode::SCmpGeU32},   {0x0au, Opcode::SCmpLtU32},   {0x0bu, Opcode::SCmpLeU32},
-    {0x0cu, Opcode::SBitcmp0B32}, {0x0du, Opcode::SBitcmp1B32}, {0x13u, Opcode::SCmpLgU64},
+    {0x0cu, Opcode::SBitcmp0B32}, {0x0du, Opcode::SBitcmp1B32}, {0x12u, Opcode::SCmpEqU64},
+    {0x13u, Opcode::SCmpLgU64},
 };
 
 constexpr OpcodeMap SOPK_OPS[] = {
@@ -78,15 +76,11 @@ constexpr OpcodeMap SOPP_OPS[] = {
     {0x0au, Opcode::SBarrier},      {0x0cu, Opcode::SWaitcnt},      {0x0eu, Opcode::SSleep},
     {0x10u, Opcode::SSendmsg},      {0x16u, Opcode::STtraceData},   {0x20u, Opcode::SInstPrefetch},
 };
-
-Opcode Lookup(const OpcodeMap* ops, uint32_t count, uint32_t opcode) {
-	for (uint32_t i = 0; i < count; i++) {
-		if (ops[i].opcode == opcode) {
-			return ops[i].ir;
-		}
-	}
-	return Opcode::Unsupported;
-}
+static_assert(Detail::HasUniqueEncodings(SOP1_OPS));
+static_assert(Detail::HasUniqueEncodings(SOP2_OPS));
+static_assert(Detail::HasUniqueEncodings(SOPK_OPS));
+static_assert(Detail::HasUniqueEncodings(SOPC_OPS));
+static_assert(Detail::HasUniqueEncodings(SOPP_OPS));
 
 bool DecodeBinarySources(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
                          Instruction& inst, uint32_t ssrc0, uint32_t ssrc1, std::string* error) {
@@ -111,7 +105,7 @@ bool DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::SOP1;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(SOP1_OPS, static_cast<uint32_t>(std::size(SOP1_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(SOP1_OPS, opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::Unsupported) {
@@ -153,7 +147,7 @@ bool DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::SOP2;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(SOP2_OPS, static_cast<uint32_t>(std::size(SOP2_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(SOP2_OPS, opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::Unsupported) {
@@ -180,7 +174,7 @@ bool DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word            = word;
 	inst.family          = Family::SOPK;
 	inst.opcode_id       = opcode;
-	inst.opcode          = Lookup(SOPK_OPS, static_cast<uint32_t>(std::size(SOPK_OPS)), opcode);
+	inst.opcode          = Detail::LookupOpcode(SOPK_OPS, opcode);
 	inst.src0.kind       = OperandKind::IntegerInlineConstant;
 	inst.src0.signed_val = imm;
 	inst.src0.value      = static_cast<uint32_t>(imm);
@@ -237,7 +231,7 @@ bool DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::SOPC;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(SOPC_OPS, static_cast<uint32_t>(std::size(SOPC_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(SOPC_OPS, opcode);
 	inst.dst.kind  = OperandKind::Scc;
 	SetRawWords(inst, code, word_index, 1);
 
@@ -261,7 +255,7 @@ bool DecodeSopp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word            = word;
 	inst.family          = Family::SOPP;
 	inst.opcode_id       = opcode;
-	inst.opcode          = Lookup(SOPP_OPS, static_cast<uint32_t>(std::size(SOPP_OPS)), opcode);
+	inst.opcode          = Detail::LookupOpcode(SOPP_OPS, opcode);
 	inst.src0.kind       = OperandKind::LiteralConstant;
 	inst.src0.value      = simm;
 	inst.src0.signed_val = static_cast<int16_t>(simm);

src/graphics/shader/recompiler/decompiler/ShaderDecoder.cpp

@@ -77,6 +77,16 @@ std::string FormatBinary(const Instruction& inst) {
 	return text;
 }
 
+std::string FormatSources(const Instruction& inst) {
+	std::string    text      = fmt::format("0x{:08x}: {}", inst.pc, OpcodeToString(inst.opcode));
+	const Operand* sources[] = {&inst.src0, &inst.src1, &inst.src2, &inst.src3};
+	for (uint32_t i = 0; i < inst.src_count && i < 4u; i++) {
+		text += i == 0 ? " " : ", ";
+		text += OperandToString(*sources[i]);
+	}
+	return text;
+}
+
 std::string FormatMemory(const Instruction& inst) {
 	std::string text = fmt::format("0x{:08x}: {} {}", inst.pc, OpcodeToString(inst.opcode).c_str(),
 	                               OperandToString(inst.dst).c_str());
@@ -511,6 +521,7 @@ std::string OpcodeToString(Opcode opcode) {
 		case Opcode::SCmpGeU32: return "s_cmp_ge_u32";
 		case Opcode::SCmpLtU32: return "s_cmp_lt_u32";
 		case Opcode::SCmpLeU32: return "s_cmp_le_u32";
+		case Opcode::SCmpEqU64: return "s_cmp_eq_u64";
 		case Opcode::SCmpLgU64: return "s_cmp_lg_u64";
 		case Opcode::VNop: return "v_nop";
 		case Opcode::VMovB32: return "v_mov_b32";
@@ -766,6 +777,7 @@ std::string OpcodeToString(Opcode opcode) {
 		case Opcode::VCmpNeU32: return "v_cmp_ne_u32";
 		case Opcode::VCmpGeU32: return "v_cmp_ge_u32";
 		case Opcode::VCmpTU32: return "v_cmp_t_u32";
+		case Opcode::VCmpEqI64: return "v_cmp_eq_i64";
 		case Opcode::VCmpNeU64: return "v_cmp_ne_u64";
 		case Opcode::VCmpxLtU32: return "v_cmpx_lt_u32";
 		case Opcode::VCmpxEqU32: return "v_cmpx_eq_u32";
@@ -823,6 +835,8 @@ std::string OpcodeToString(Opcode opcode) {
 		case Opcode::BufferAtomicAnd: return "buffer_atomic_and";
 		case Opcode::BufferAtomicOr: return "buffer_atomic_or";
 		case Opcode::BufferAtomicXor: return "buffer_atomic_xor";
+		case Opcode::BufferAtomicFMin: return "buffer_atomic_fmin";
+		case Opcode::BufferAtomicFMax: return "buffer_atomic_fmax";
 		case Opcode::FlatLoadUbyte: return "flat_load_ubyte";
 		case Opcode::FlatLoadSbyte: return "flat_load_sbyte";
 		case Opcode::FlatLoadUshort: return "flat_load_ushort";
@@ -983,6 +997,9 @@ std::string InstructionToString(const Instruction& inst) {
 		                   inst.pc, FamilyToString(inst.family).c_str(), inst.opcode_id,
 		                   RawWordsToString(inst).c_str(), inst.unsupported_reason.c_str());
 	}
+	if (inst.family == Family::SOPC) {
+		return WithUnsupportedReason(inst, FormatSources(inst));
+	}
 
 	switch (inst.opcode) {
 		case Opcode::SMovB32:
@@ -1110,6 +1127,8 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::BufferAtomicAnd:
 		case Opcode::BufferAtomicOr:
 		case Opcode::BufferAtomicXor:
+		case Opcode::BufferAtomicFMin:
+		case Opcode::BufferAtomicFMax:
 		case Opcode::BufferLoadSbyte:
 		case Opcode::BufferLoadSshort:
 		case Opcode::FlatLoadUbyte:
@@ -1171,25 +1190,6 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::DsWriteB128:
 		case Opcode::DsWriteAddtidB32:
 		case Opcode::DsReadAddtidB32: return WithUnsupportedReason(inst, FormatMemory(inst));
-		case Opcode::SCmpEqU32:
-		case Opcode::SCmpEqI32:
-		case Opcode::SCmpLgU32:
-		case Opcode::SCmpLgI32:
-		case Opcode::SCmpGtU32:
-		case Opcode::SCmpGtI32:
-		case Opcode::SCmpGeU32:
-		case Opcode::SCmpGeI32:
-		case Opcode::SCmpLtU32:
-		case Opcode::SCmpLtI32:
-		case Opcode::SCmpLeU32:
-		case Opcode::SCmpLeI32:
-		case Opcode::SBitcmp0B32:
-		case Opcode::SBitcmp1B32:
-		case Opcode::SCmpLgU64:
-			return WithUnsupportedReason(inst, fmt::format("0x{:08x}: {} {}, {}", inst.pc,
-			                                               OpcodeToString(inst.opcode).c_str(),
-			                                               OperandToString(inst.src0).c_str(),
-			                                               OperandToString(inst.src1).c_str()));
 		default: return WithUnsupportedReason(inst, FormatBinary(inst));
 	}
 }

src/graphics/shader/recompiler/decompiler/ShaderDecoder.h

@@ -118,6 +118,7 @@ enum class Opcode {
 	SCmpGeU32,
 	SCmpLtU32,
 	SCmpLeU32,
+	SCmpEqU64,
 	SCmpLgU64,
 
 	VNop,
@@ -374,6 +375,7 @@ enum class Opcode {
 	VCmpNeU32,
 	VCmpGeU32,
 	VCmpTU32,
+	VCmpEqI64,
 	VCmpNeU64,
 	VCmpxLtU32,
 	VCmpxEqU32,
@@ -432,6 +434,8 @@ enum class Opcode {
 	BufferAtomicAnd,
 	BufferAtomicOr,
 	BufferAtomicXor,
+	BufferAtomicFMin,
+	BufferAtomicFMax,
 	FlatLoadUbyte,
 	FlatLoadSbyte,
 	FlatLoadUshort,

src/graphics/shader/recompiler/decompiler/VectorAluOps.cpp

@@ -1,43 +1,91 @@
 #include "graphics/shader/recompiler/decompiler/VectorAluOps.h"
 
+#include "graphics/shader/recompiler/decompiler/OpcodeTable.h"
+
 #include <fmt/format.h>
-#include <iterator>
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
 
-struct OpcodeMap {
-	uint32_t opcode = 0;
-	Opcode   ir     = Opcode::Unknown;
+using Detail::OpcodeMap;
+
+// These profiles describe the selectors that the current lowering implements. They are not
+// architectural SDWA legality classes.
+enum class Vop2SdwaProfile {
+	None,
+	Cndmask,
+	Float32,
+	Float16,
+	IntegerFullDestination,
+	IntegerPartialDestination,
+	ReverseLogicalRight,
+	ReverseLogicalLeft,
+	Bitwise,
+	Count,
 };
 
-constexpr OpcodeMap VOP2_OPS[] = {
-    {0x01u, Opcode::VCndmaskB32},    {0x02u, Opcode::VDot2cF32F16},
-    {0x03u, Opcode::VAddF32},        {0x04u, Opcode::VSubF32},
-    {0x05u, Opcode::VSubrevF32},     {0x08u, Opcode::VMulF32},
-    {0x09u, Opcode::VMulI32I24},     {0x0bu, Opcode::VMulU32U24},
-    {0x0fu, Opcode::VMinF32},        {0x10u, Opcode::VMaxF32},
-    {0x11u, Opcode::VMinI32},        {0x12u, Opcode::VMaxI32},
-    {0x13u, Opcode::VMinU32},        {0x14u, Opcode::VMaxU32},
-    {0x15u, Opcode::VLshrB32},       {0x16u, Opcode::VLshrrevB32},
-    {0x17u, Opcode::VAshrI32},       {0x18u, Opcode::VAshrrevI32},
-    {0x19u, Opcode::VLshlB32},       {0x1au, Opcode::VLshlrevB32},
-    {0x1bu, Opcode::VAndB32},        {0x1cu, Opcode::VOrB32},
-    {0x1du, Opcode::VXorB32},        {0x1eu, Opcode::VXnorB32},
-    {0x1fu, Opcode::VMacF32},        {0x20u, Opcode::VMadmkF32},
-    {0x21u, Opcode::VMadakF32},      {0x22u, Opcode::VBcntU32B32},
-    {0x23u, Opcode::VMbcntLoU32B32}, {0x24u, Opcode::VMbcntHiU32B32},
-    {0x25u, Opcode::VAddNcU32},      {0x28u, Opcode::VAddcU32},
-    {0x26u, Opcode::VSubNcU32},      {0x27u, Opcode::VSubrevNcU32},
-    {0x2bu, Opcode::VMacF32},        {0x2cu, Opcode::VMadmkF32},
-    {0x2du, Opcode::VMadakF32},      {0x2fu, Opcode::VCvtPkrtzF16F32},
-    {0x32u, Opcode::VAddF16},        {0x33u, Opcode::VSubF16},
-    {0x34u, Opcode::VSubrevF16},     {0x35u, Opcode::VMulF16},
-    {0x36u, Opcode::VFmacF16},       {0x37u, Opcode::VFmamkF16},
-    {0x38u, Opcode::VFmaakF16},      {0x39u, Opcode::VMaxF16},
-    {0x3au, Opcode::VMinF16},        {0x3cu, Opcode::VPkFmacF16},
+struct Vop2OpcodeInfo {
+	uint32_t        encoding     = 0;
+	Opcode          decoded      = Opcode::Unknown;
+	Vop2SdwaProfile sdwa_profile = Vop2SdwaProfile::None;
 };
 
+constexpr Vop2OpcodeInfo VOP2_OPS[] = {
+    {0x01u, Opcode::VCndmaskB32, Vop2SdwaProfile::Cndmask},
+    {0x02u, Opcode::VDot2cF32F16},
+    {0x03u, Opcode::VAddF32, Vop2SdwaProfile::Float32},
+    {0x04u, Opcode::VSubF32, Vop2SdwaProfile::Float32},
+    {0x05u, Opcode::VSubrevF32},
+    {0x08u, Opcode::VMulF32, Vop2SdwaProfile::Float32},
+    {0x09u, Opcode::VMulI32I24, Vop2SdwaProfile::IntegerFullDestination},
+    {0x0bu, Opcode::VMulU32U24, Vop2SdwaProfile::IntegerFullDestination},
+    {0x0fu, Opcode::VMinF32},
+    {0x10u, Opcode::VMaxF32},
+    {0x11u, Opcode::VMinI32},
+    {0x12u, Opcode::VMaxI32},
+    {0x13u, Opcode::VMinU32, Vop2SdwaProfile::IntegerPartialDestination},
+    {0x14u, Opcode::VMaxU32, Vop2SdwaProfile::IntegerFullDestination},
+    {0x15u, Opcode::VLshrB32},
+    {0x16u, Opcode::VLshrrevB32, Vop2SdwaProfile::ReverseLogicalRight},
+    {0x17u, Opcode::VAshrI32},
+    {0x18u, Opcode::VAshrrevI32},
+    {0x19u, Opcode::VLshlB32},
+    {0x1au, Opcode::VLshlrevB32, Vop2SdwaProfile::ReverseLogicalLeft},
+    {0x1bu, Opcode::VAndB32, Vop2SdwaProfile::Bitwise},
+    {0x1cu, Opcode::VOrB32, Vop2SdwaProfile::Bitwise},
+    {0x1du, Opcode::VXorB32, Vop2SdwaProfile::Bitwise},
+    {0x1eu, Opcode::VXnorB32, Vop2SdwaProfile::Bitwise},
+    {0x1fu, Opcode::VMacF32},
+    {0x20u, Opcode::VMadmkF32},
+    {0x21u, Opcode::VMadakF32},
+    {0x22u, Opcode::VBcntU32B32},
+    {0x23u, Opcode::VMbcntLoU32B32},
+    {0x24u, Opcode::VMbcntHiU32B32},
+    {0x25u, Opcode::VAddNcU32, Vop2SdwaProfile::IntegerFullDestination},
+    {0x28u, Opcode::VAddcU32},
+    {0x26u, Opcode::VSubNcU32, Vop2SdwaProfile::IntegerPartialDestination},
+    {0x27u, Opcode::VSubrevNcU32, Vop2SdwaProfile::IntegerFullDestination},
+    {0x2bu, Opcode::VMacF32},
+    {0x2cu, Opcode::VMadmkF32},
+    {0x2du, Opcode::VMadakF32},
+    {0x2fu, Opcode::VCvtPkrtzF16F32},
+    {0x32u, Opcode::VAddF16, Vop2SdwaProfile::Float16},
+    {0x33u, Opcode::VSubF16, Vop2SdwaProfile::Float16},
+    {0x34u, Opcode::VSubrevF16, Vop2SdwaProfile::Float16},
+    {0x35u, Opcode::VMulF16, Vop2SdwaProfile::Float16},
+    {0x36u, Opcode::VFmacF16},
+    {0x37u, Opcode::VFmamkF16},
+    {0x38u, Opcode::VFmaakF16},
+    {0x39u, Opcode::VMaxF16, Vop2SdwaProfile::Float16},
+    {0x3au, Opcode::VMinF16, Vop2SdwaProfile::Float16},
+    {0x3cu, Opcode::VPkFmacF16},
+};
+static_assert(Detail::HasUniqueEncodings(VOP2_OPS));
+
+constexpr Opcode LookupVop2Opcode(uint32_t encoding) {
+	return Detail::LookupOpcode(VOP2_OPS, encoding);
+}
+
 constexpr OpcodeMap VOP1_OPS[] = {
     {0x00u, Opcode::VNop},
     {0x01u, Opcode::VMovB32},
@@ -135,37 +183,59 @@ constexpr OpcodeMap VOP3_ENCODED_VOP1_OPS[] = {
     {0x5eu, Opcode::VRndneF16},
 };
 
-constexpr OpcodeMap VOPC_OPS[] = {
-    {0x00u, Opcode::VCmpFF32},     {0x01u, Opcode::VCmpLtF32},   {0x02u, Opcode::VCmpEqF32},
-    {0x03u, Opcode::VCmpLeF32},    {0x04u, Opcode::VCmpGtF32},   {0x05u, Opcode::VCmpLgF32},
-    {0x06u, Opcode::VCmpGeF32},    {0x07u, Opcode::VCmpOF32},    {0x08u, Opcode::VCmpUF32},
-    {0x09u, Opcode::VCmpNgeF32},   {0x0au, Opcode::VCmpNlgF32},  {0x0bu, Opcode::VCmpNgtF32},
-    {0x0cu, Opcode::VCmpNleF32},   {0x0du, Opcode::VCmpNeqF32},  {0x0eu, Opcode::VCmpNltF32},
-    {0x0fu, Opcode::VCmpTruF32},   {0x11u, Opcode::VCmpxLtF32},  {0x12u, Opcode::VCmpxEqF32},
-    {0x13u, Opcode::VCmpxLeF32},   {0x14u, Opcode::VCmpxGtF32},  {0x15u, Opcode::VCmpxLgF32},
-    {0x16u, Opcode::VCmpxGeF32},   {0x19u, Opcode::VCmpxNgeF32}, {0x1au, Opcode::VCmpxNlgF32},
-    {0x1bu, Opcode::VCmpxNgtF32},  {0x1cu, Opcode::VCmpxNleF32}, {0x1du, Opcode::VCmpxNeqF32},
-    {0x1eu, Opcode::VCmpxNltF32},  {0x80u, Opcode::VCmpFI32},    {0x81u, Opcode::VCmpLtI32},
-    {0x82u, Opcode::VCmpEqI32},    {0x83u, Opcode::VCmpLeI32},   {0x84u, Opcode::VCmpGtI32},
-    {0x85u, Opcode::VCmpNeI32},    {0x86u, Opcode::VCmpGeI32},   {0x87u, Opcode::VCmpTI32},
-    {0x88u, Opcode::VCmpClassF32}, {0x89u, Opcode::VCmpLtI16},   {0x8au, Opcode::VCmpEqI16},
-    {0x8bu, Opcode::VCmpLeI16},    {0x8cu, Opcode::VCmpGtI16},   {0x8du, Opcode::VCmpNeI16},
-    {0x8eu, Opcode::VCmpGeI16},    {0x91u, Opcode::VCmpxLtI32},  {0x92u, Opcode::VCmpxEqI32},
-    {0x93u, Opcode::VCmpxLeI32},   {0x94u, Opcode::VCmpxGtI32},  {0x95u, Opcode::VCmpxNeI32},
-    {0x96u, Opcode::VCmpxGeI32},   {0xa9u, Opcode::VCmpLtU16},   {0xaau, Opcode::VCmpEqU16},
-    {0xabu, Opcode::VCmpLeU16},    {0xacu, Opcode::VCmpGtU16},   {0xadu, Opcode::VCmpNeU16},
-    {0xaeu, Opcode::VCmpGeU16},    {0xc0u, Opcode::VCmpFU32},    {0xc1u, Opcode::VCmpLtU32},
-    {0xc2u, Opcode::VCmpEqU32},    {0xc3u, Opcode::VCmpLeU32},   {0xc4u, Opcode::VCmpGtU32},
-    {0xc5u, Opcode::VCmpNeU32},    {0xc6u, Opcode::VCmpGeU32},   {0xc7u, Opcode::VCmpTU32},
-    {0xd1u, Opcode::VCmpxLtU32},   {0xd2u, Opcode::VCmpxEqU32},  {0xd3u, Opcode::VCmpxLeU32},
-    {0xd4u, Opcode::VCmpxGtU32},   {0xd5u, Opcode::VCmpxNeU32},  {0xd6u, Opcode::VCmpxGeU32},
-    {0xe5u, Opcode::VCmpNeU64},    {0xc9u, Opcode::VCmpLtF16},   {0xcau, Opcode::VCmpEqF16},
-    {0xcbu, Opcode::VCmpLeF16},    {0xccu, Opcode::VCmpGtF16},   {0xcdu, Opcode::VCmpLgF16},
-    {0xceu, Opcode::VCmpGeF16},    {0xedu, Opcode::VCmpNeqF16},  {0xd9u, Opcode::VCmpxLtF16},
-    {0xdau, Opcode::VCmpxEqF16},   {0xdbu, Opcode::VCmpxLeF16},  {0xdcu, Opcode::VCmpxGtF16},
-    {0xdeu, Opcode::VCmpxGeF16},   {0xfdu, Opcode::VCmpxNeqF16}, {0xfeu, Opcode::VCmpxNltF16},
+struct VopcOpcodeInfo {
+	uint32_t encoding     = 0;
+	Opcode   decoded      = Opcode::Unknown;
+	bool     supports_dpp = true;
 };
 
+constexpr VopcOpcodeInfo VOPC_OPS[] = {
+    {0x00u, Opcode::VCmpFF32},         {0x01u, Opcode::VCmpLtF32},
+    {0x02u, Opcode::VCmpEqF32},        {0x03u, Opcode::VCmpLeF32},
+    {0x04u, Opcode::VCmpGtF32},        {0x05u, Opcode::VCmpLgF32},
+    {0x06u, Opcode::VCmpGeF32},        {0x07u, Opcode::VCmpOF32},
+    {0x08u, Opcode::VCmpUF32},         {0x09u, Opcode::VCmpNgeF32},
+    {0x0au, Opcode::VCmpNlgF32},       {0x0bu, Opcode::VCmpNgtF32},
+    {0x0cu, Opcode::VCmpNleF32},       {0x0du, Opcode::VCmpNeqF32},
+    {0x0eu, Opcode::VCmpNltF32},       {0x0fu, Opcode::VCmpTruF32},
+    {0x11u, Opcode::VCmpxLtF32},       {0x12u, Opcode::VCmpxEqF32},
+    {0x13u, Opcode::VCmpxLeF32},       {0x14u, Opcode::VCmpxGtF32},
+    {0x15u, Opcode::VCmpxLgF32},       {0x16u, Opcode::VCmpxGeF32},
+    {0x19u, Opcode::VCmpxNgeF32},      {0x1au, Opcode::VCmpxNlgF32},
+    {0x1bu, Opcode::VCmpxNgtF32},      {0x1cu, Opcode::VCmpxNleF32},
+    {0x1du, Opcode::VCmpxNeqF32},      {0x1eu, Opcode::VCmpxNltF32},
+    {0x80u, Opcode::VCmpFI32},         {0x81u, Opcode::VCmpLtI32},
+    {0x82u, Opcode::VCmpEqI32},        {0x83u, Opcode::VCmpLeI32},
+    {0x84u, Opcode::VCmpGtI32},        {0x85u, Opcode::VCmpNeI32},
+    {0x86u, Opcode::VCmpGeI32},        {0x87u, Opcode::VCmpTI32},
+    {0x88u, Opcode::VCmpClassF32},     {0x89u, Opcode::VCmpLtI16},
+    {0x8au, Opcode::VCmpEqI16},        {0x8bu, Opcode::VCmpLeI16},
+    {0x8cu, Opcode::VCmpGtI16},        {0x8du, Opcode::VCmpNeI16},
+    {0x8eu, Opcode::VCmpGeI16},        {0x91u, Opcode::VCmpxLtI32},
+    {0x92u, Opcode::VCmpxEqI32},       {0x93u, Opcode::VCmpxLeI32},
+    {0x94u, Opcode::VCmpxGtI32},       {0x95u, Opcode::VCmpxNeI32},
+    {0x96u, Opcode::VCmpxGeI32},       {0xa9u, Opcode::VCmpLtU16},
+    {0xaau, Opcode::VCmpEqU16},        {0xabu, Opcode::VCmpLeU16},
+    {0xacu, Opcode::VCmpGtU16},        {0xadu, Opcode::VCmpNeU16},
+    {0xaeu, Opcode::VCmpGeU16},        {0xc0u, Opcode::VCmpFU32},
+    {0xc1u, Opcode::VCmpLtU32},        {0xc2u, Opcode::VCmpEqU32},
+    {0xc3u, Opcode::VCmpLeU32},        {0xc4u, Opcode::VCmpGtU32},
+    {0xc5u, Opcode::VCmpNeU32},        {0xc6u, Opcode::VCmpGeU32},
+    {0xc7u, Opcode::VCmpTU32},         {0xa2u, Opcode::VCmpEqI64, false},
+    {0xd1u, Opcode::VCmpxLtU32},       {0xd2u, Opcode::VCmpxEqU32},
+    {0xd3u, Opcode::VCmpxLeU32},       {0xd4u, Opcode::VCmpxGtU32},
+    {0xd5u, Opcode::VCmpxNeU32},       {0xd6u, Opcode::VCmpxGeU32},
+    {0xe5u, Opcode::VCmpNeU64, false}, {0xc9u, Opcode::VCmpLtF16},
+    {0xcau, Opcode::VCmpEqF16},        {0xcbu, Opcode::VCmpLeF16},
+    {0xccu, Opcode::VCmpGtF16},        {0xcdu, Opcode::VCmpLgF16},
+    {0xceu, Opcode::VCmpGeF16},        {0xedu, Opcode::VCmpNeqF16},
+    {0xd9u, Opcode::VCmpxLtF16},       {0xdau, Opcode::VCmpxEqF16},
+    {0xdbu, Opcode::VCmpxLeF16},       {0xdcu, Opcode::VCmpxGtF16},
+    {0xdeu, Opcode::VCmpxGeF16},       {0xfdu, Opcode::VCmpxNeqF16},
+    {0xfeu, Opcode::VCmpxNltF16},
+};
+static_assert(Detail::HasUniqueEncodings(VOPC_OPS));
+
 constexpr OpcodeMap VOP3_OPS[] = {
     {0x141u, Opcode::VMadF32},          {0x142u, Opcode::VMadI32I24},
     {0x143u, Opcode::VMadU32U24},       {0x176u, Opcode::VMadU64U32},
@@ -214,16 +284,6 @@ constexpr OpcodeMap VOP3P_OPS[] = {
     {0x22u, Opcode::VMadMixhiF16},
 };
 
-Opcode Lookup(const OpcodeMap* ops, uint32_t count, uint32_t opcode) {
-	for (uint32_t i = 0; i < count; i++) {
-		const auto& op = ops[i];
-		if (op.opcode == opcode) {
-			return op.ir;
-		}
-	}
-	return Opcode::Unsupported;
-}
-
 bool IsVop2LiteralMadOpcode(uint32_t opcode) {
 	return opcode == 0x20u || opcode == 0x21u || opcode == 0x2cu || opcode == 0x2du;
 }
@@ -234,24 +294,23 @@ bool IsUnsupportedVop3EncodedVop2Alias(uint32_t opcode) {
 }
 
 Opcode LookupVop3Opcode(uint32_t opcode) {
-	const auto direct = Lookup(VOP3_OPS, static_cast<uint32_t>(std::size(VOP3_OPS)), opcode);
+	const auto direct = Detail::LookupOpcode(VOP3_OPS, opcode);
 	if (direct != Opcode::Unsupported) {
 		return direct;
 	}
 	if (opcode <= 0xffu) {
-		return Lookup(VOPC_OPS, static_cast<uint32_t>(std::size(VOPC_OPS)), opcode);
+		return Detail::LookupOpcode(VOPC_OPS, opcode);
 	}
 	if (opcode >= 0x100u && opcode <= 0x13fu) {
 		if (IsUnsupportedVop3EncodedVop2Alias(opcode - 0x100u)) {
 			return Opcode::Unsupported;
 		}
-		return Lookup(VOP2_OPS, static_cast<uint32_t>(std::size(VOP2_OPS)), opcode - 0x100u);
+		return LookupVop2Opcode(opcode - 0x100u);
 	}
 	if (opcode >= 0x180u && opcode <= 0x1ffu) {
-		return Lookup(VOP3_ENCODED_VOP1_OPS,
-		              static_cast<uint32_t>(std::size(VOP3_ENCODED_VOP1_OPS)), opcode - 0x180u);
+		return Detail::LookupOpcode(VOP3_ENCODED_VOP1_OPS, opcode - 0x180u);
 	}
-	return Lookup(VOP3_OPS, static_cast<uint32_t>(std::size(VOP3_OPS)), opcode);
+	return Detail::LookupOpcode(VOP3_OPS, opcode);
 }
 
 bool IsVop3EncodedVopc(uint32_t opcode) {
@@ -791,7 +850,6 @@ Vop2SdwaFields DecodeVop2SdwaFields(uint32_t modifier) {
 }
 
 struct Vop2SdwaRule {
-	Opcode   opcode           = Opcode::Unknown;
 	uint32_t dst_selectors    = SdwaSelFull();
 	uint32_t src0_selectors   = SdwaSelFull();
 	uint32_t src1_selectors   = SdwaSelFull();
@@ -804,46 +862,27 @@ constexpr uint32_t SdwaSelAll() {
 }
 
 constexpr Vop2SdwaRule VOP2_SDWA_RULES[] = {
-    {Opcode::VCndmaskB32, SdwaSelWords(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, false},
-    {Opcode::VAddF32, SdwaSelFull(), SdwaSelFull(), SdwaSelFull(), false, true},
-    {Opcode::VSubF32, SdwaSelFull(), SdwaSelFull(), SdwaSelFull(), false, true},
-    {Opcode::VMulF32, SdwaSelFull(), SdwaSelFull(), SdwaSelFull(), false, true},
-    {Opcode::VAddF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VSubF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VSubrevF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VMulF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VMaxF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VMinF16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
-     SdwaSelWords() | SdwaSelFull(), true, true},
-    {Opcode::VMulI32I24, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VMulU32U24, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VMinU32, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VMaxU32, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VLshrrevB32, SdwaSelFull(), SdwaSelFull(), SdwaSelWords() | SdwaSelFull(), false,
-     false},
-    {Opcode::VLshlrevB32, SdwaSelFull(), SdwaSelFull(), SdwaSelAll(), false, false},
-    {Opcode::VAndB32, SdwaSelWords() | SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), true, false},
-    {Opcode::VOrB32, SdwaSelWords() | SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), true, false},
-    {Opcode::VXorB32, SdwaSelWords() | SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), true, false},
-    {Opcode::VXnorB32, SdwaSelWords() | SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), true, false},
-    {Opcode::VAddNcU32, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VSubNcU32, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
-    {Opcode::VSubrevNcU32, SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
+    {},
+    {SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
+     true, false},
+    {SdwaSelFull(), SdwaSelFull(), SdwaSelFull(), false, true},
+    {SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(), SdwaSelWords() | SdwaSelFull(),
+     true, true},
+    {SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), false, false},
+    {SdwaSelAll(), SdwaSelAll(), SdwaSelAll(), true, false},
+    {SdwaSelFull(), SdwaSelFull(), SdwaSelWords() | SdwaSelFull(), false, false},
+    {SdwaSelFull(), SdwaSelFull(), SdwaSelAll(), false, false},
+    {SdwaSelWords() | SdwaSelFull(), SdwaSelAll(), SdwaSelAll(), true, false},
 };
+static_assert(sizeof(VOP2_SDWA_RULES) / sizeof(VOP2_SDWA_RULES[0]) ==
+              static_cast<size_t>(Vop2SdwaProfile::Count));
 
-const Vop2SdwaRule* FindVop2SdwaRule(Opcode opcode) {
-	for (const auto& rule: VOP2_SDWA_RULES) {
-		if (rule.opcode == opcode) {
-			return &rule;
-		}
+const Vop2SdwaRule* FindVop2SdwaRule(uint32_t encoding) {
+	const auto* info = Detail::FindOpcode(VOP2_OPS, encoding);
+	if (info == nullptr || info->sdwa_profile == Vop2SdwaProfile::None) {
+		return nullptr;
 	}
-	return nullptr;
+	return &VOP2_SDWA_RULES[static_cast<size_t>(info->sdwa_profile)];
 }
 
 bool IsVop2SdwaDestinationSupported(const Vop2SdwaRule& rule, const Vop2SdwaFields& fields) {
@@ -881,7 +920,7 @@ bool ValidateVop2Sdwa(Instruction& inst, uint32_t opcode, const Vop2SdwaFields&
 		return true;
 	}
 
-	const auto* rule = FindVop2SdwaRule(inst.opcode);
+	const auto* rule = FindVop2SdwaRule(opcode);
 	if (rule == nullptr) {
 		SetUnsupported(inst, Family::VOP2, opcode,
 		               "VOP2 SDWA modifier is not supported for opcode");
@@ -1135,6 +1174,11 @@ bool DecodeVopcDpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 		SetUnsupported(inst, Family::VOPC, opcode, "VOPC opcode is not implemented");
 		return true;
 	}
+	const auto* info = Detail::FindOpcode(VOPC_OPS, opcode);
+	if (info == nullptr || !info->supports_dpp) {
+		SetUnsupported(inst, Family::VOPC, opcode, "VOPC DPP modifier is not supported for opcode");
+		return true;
+	}
 	if (!DecodeScalarSource(src0 + 256u, pc, inst.src0, error) ||
 	    !DecodeVectorGpr(vsrc1, inst.src1, error)) {
 		return false;
@@ -1427,7 +1471,7 @@ bool DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::VOP2;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(VOP2_OPS, static_cast<uint32_t>(std::size(VOP2_OPS)), opcode);
+	inst.opcode    = LookupVop2Opcode(opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	switch (opcode) {
@@ -1468,7 +1512,7 @@ bool DecodeVop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::VOP1;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(VOP1_OPS, static_cast<uint32_t>(std::size(VOP1_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(VOP1_OPS, opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::Unsupported) {
@@ -1509,7 +1553,7 @@ bool DecodeVopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.word      = word;
 	inst.family    = Family::VOPC;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(VOPC_OPS, static_cast<uint32_t>(std::size(VOPC_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(VOPC_OPS, opcode);
 	inst.dst.kind  = IsVopcCompareExec(inst.opcode) ? OperandKind::ExecLo : OperandKind::VccLo;
 	SetRawWords(inst, code, word_index, 1);
 
@@ -1723,7 +1767,7 @@ bool DecodeVop3p(uint32_t pc, std::span<const uint32_t> code, uint32_t word_inde
 	inst.word      = word0;
 	inst.family    = Family::VOP3P;
 	inst.opcode_id = opcode;
-	inst.opcode    = Lookup(VOP3P_OPS, static_cast<uint32_t>(std::size(VOP3P_OPS)), opcode);
+	inst.opcode    = Detail::LookupOpcode(VOP3P_OPS, opcode);
 	SetRawWords(inst, code, word_index, 2);
 
 	if (inst.opcode == Opcode::Unsupported) {

src/graphics/shader/recompiler/emitter/SpirvEmitter.cpp

@@ -131,7 +131,9 @@ bool IsAtomic(IR::Opcode op) {
 		case IR::Opcode::AtomicUMaxU32:
 		case IR::Opcode::AtomicAndU32:
 		case IR::Opcode::AtomicOrU32:
-		case IR::Opcode::AtomicXorU32: return true;
+		case IR::Opcode::AtomicXorU32:
+		case IR::Opcode::AtomicFMinF32:
+		case IR::Opcode::AtomicFMaxF32: return true;
 		default: return false;
 	}
 }
@@ -141,6 +143,9 @@ uint32_t BufferAddressOperandCount(const IR::Instruction& inst) {
 }
 
 bool ValidateInstructionContract(const IR::Instruction& inst, std::string* error) {
+	if (static_cast<size_t>(inst.op) >= static_cast<size_t>(IR::Opcode::Count)) {
+		return Fail(error, "IR instruction opcode is out of range");
+	}
 	if (inst.src_count > std::size(inst.src)) {
 		return Fail(error, "instruction source count exceeds fixed IR storage");
 	}
@@ -200,6 +205,12 @@ bool ValidateInstructionContract(const IR::Instruction& inst, std::string* error
 	     inst.src_count != 2)) {
 		return Fail(error, "storage image instruction has an invalid resource class");
 	}
+	const auto float_atomic =
+	    inst.op == IR::Opcode::AtomicFMinF32 || inst.op == IR::Opcode::AtomicFMaxF32;
+	if (float_atomic &&
+	    (kind != IR::ResourceKind::Buffer || inst.src_count != buffer_address_count + 1u)) {
+		return Fail(error, "floating-point atomic instruction must target a buffer");
+	}
 	if (IsAtomic(inst.op) &&
 	    !((kind == IR::ResourceKind::Buffer && inst.src_count == buffer_address_count + 1u) ||
 	      ((kind == IR::ResourceKind::Lds || kind == IR::ResourceKind::Gds) &&
@@ -389,7 +400,7 @@ bool ProgramRequiresExactSubgroupSize(const IR::Program& program) {
 			if (Emitter::InstructionHasDppSource(inst)) {
 				return true;
 			}
-			if (Emitter::IsCompareOpcode(inst.op) && !Emitter::IsSccOperand(inst.dst)) {
+			if (IR::IsCompareOpcode(inst.op) && !Emitter::IsSccOperand(inst.dst)) {
 				return true;
 			}
 			if ((IsExecDestination(inst.dst) || IsExecDestination(inst.dst2)) &&
@@ -488,6 +499,13 @@ bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resourc
 	AllocateBlockLabels(state, program);
 	AllocateDispatcherState(state, program);
 	EmitFunction(state, program);
+	if (state.unsupported_ir_instruction) {
+		char pc[9] {};
+		std::snprintf(pc, sizeof(pc), "%08x", state.unsupported_ir_pc);
+		return Fail(error, "SPIR-V emitter has no handler for IR opcode " +
+		                       std::string(IR::OpcodeName(state.unsupported_ir_opcode)) +
+		                       " at pc 0x" + pc);
+	}
 
 	ReportReserveExceeded(program, "registers", state.registers.size());
 	ReportReserveExceeded(program, "inputs", state.inputs.size());

src/graphics/shader/recompiler/emitter/spirvEmitterAluOps.cpp

@@ -1486,17 +1486,20 @@ void EmitBitCompareB32(EmitterState& state, const IR::Instruction& inst, bool bi
 	EmitStoreU32(state, inst.dst, ret);
 }
 
-void EmitCompareNeU64(EmitterState& state, const IR::Instruction& inst) {
-	const auto lhs_low  = EmitSequentialValueLoad(state, inst.src[0], 0);
-	const auto lhs_high = EmitSequentialValueLoad(state, inst.src[0], 1);
-	const auto rhs_low  = EmitSequentialValueLoad(state, inst.src[1], 0);
-	const auto rhs_high = EmitSequentialValueLoad(state, inst.src[1], 1);
-	const auto ne_low   = state.builder.AllocateId();
-	const auto ne_high  = state.builder.AllocateId();
-	const auto cond     = state.builder.AllocateId();
-	state.builder.AddFunction({OpINotEqual, state.bool_type, ne_low, lhs_low, rhs_low});
-	state.builder.AddFunction({OpINotEqual, state.bool_type, ne_high, lhs_high, rhs_high});
-	state.builder.AddFunction({OpLogicalOr, state.bool_type, cond, ne_low, ne_high});
+void EmitCompareU64(EmitterState& state, const IR::Instruction& inst) {
+	const bool equal      = inst.op == IR::Opcode::CompareEqU64;
+	const auto lhs_low    = EmitSequentialValueLoad(state, inst.src[0], 0);
+	const auto lhs_high   = EmitSequentialValueLoad(state, inst.src[0], 1);
+	const auto rhs_low    = EmitSequentialValueLoad(state, inst.src[1], 0);
+	const auto rhs_high   = EmitSequentialValueLoad(state, inst.src[1], 1);
+	const auto low        = state.builder.AllocateId();
+	const auto high       = state.builder.AllocateId();
+	const auto cond       = state.builder.AllocateId();
+	const auto compare_op = equal ? OpIEqual : OpINotEqual;
+	state.builder.AddFunction({compare_op, state.bool_type, low, lhs_low, rhs_low});
+	state.builder.AddFunction({compare_op, state.bool_type, high, lhs_high, rhs_high});
+	state.builder.AddFunction(
+	    {equal ? OpLogicalAnd : OpLogicalOr, state.bool_type, cond, low, high});
 	EmitCompareResult(state, inst.dst, cond);
 }
 

src/graphics/shader/recompiler/emitter/spirvEmitterAnalysis.cpp

@@ -75,8 +75,6 @@ bool IsSccOperand(const IR::Operand& operand) {
 	return operand.kind == IR::OperandKind::Register && operand.reg.file == IR::RegisterFile::Scc;
 }
 
-bool IsCompareOpcode(IR::Opcode op);
-
 void CollectMaskStateRegisters(std::vector<RegisterBinding>& registers) {
 	CollectRegister(registers, {IR::RegisterFile::Exec, 0});
 	CollectRegister(registers, {IR::RegisterFile::Exec, 1});
@@ -189,11 +187,11 @@ void CollectRegisters(const IR::Program& program, std::vector<RegisterBinding>&
 			if (inst.op == IR::Opcode::BitCountU64 || inst.op == IR::Opcode::FindMsbFromHighU64) {
 				CollectSequentialRegisters(registers, inst.src[0], 2);
 			}
-			if (inst.op == IR::Opcode::CompareNeU64) {
+			if (IR::IsCompare64Opcode(inst.op)) {
 				CollectSequentialRegisters(registers, inst.src[0], 2);
 				CollectSequentialRegisters(registers, inst.src[1], 2);
 			}
-			if (IsCompareOpcode(inst.op) && inst.dst.kind == IR::OperandKind::Register &&
+			if (IR::IsCompareOpcode(inst.op) && inst.dst.kind == IR::OperandKind::Register &&
 			    inst.dst.reg.file != IR::RegisterFile::Scc) {
 				CollectSequentialRegisters(registers, inst.dst, 2);
 			}
@@ -384,93 +382,6 @@ bool ProgramNeedsSubgroupShuffle(const IR::Program& program) {
 	return false;
 }
 
-bool IsCompareOpcode(IR::Opcode op) {
-	switch (op) {
-		case IR::Opcode::CompareFalse:
-		case IR::Opcode::CompareTrue:
-		case IR::Opcode::CompareEqU32:
-		case IR::Opcode::CompareNeU32:
-		case IR::Opcode::CompareGtU32:
-		case IR::Opcode::CompareGeU32:
-		case IR::Opcode::CompareLtU32:
-		case IR::Opcode::CompareLeU32:
-		case IR::Opcode::CompareNeU64:
-		case IR::Opcode::CompareMaskEqU32:
-		case IR::Opcode::CompareMaskNeU32:
-		case IR::Opcode::CompareMaskGtU32:
-		case IR::Opcode::CompareMaskGeU32:
-		case IR::Opcode::CompareMaskLtU32:
-		case IR::Opcode::CompareMaskLeU32:
-		case IR::Opcode::CompareEqI32:
-		case IR::Opcode::CompareNeI32:
-		case IR::Opcode::CompareGtI32:
-		case IR::Opcode::CompareGeI32:
-		case IR::Opcode::CompareLtI32:
-		case IR::Opcode::CompareLeI32:
-		case IR::Opcode::CompareEqI16:
-		case IR::Opcode::CompareNeI16:
-		case IR::Opcode::CompareGtI16:
-		case IR::Opcode::CompareGeI16:
-		case IR::Opcode::CompareLtI16:
-		case IR::Opcode::CompareLeI16:
-		case IR::Opcode::CompareMaskEqI32:
-		case IR::Opcode::CompareMaskNeI32:
-		case IR::Opcode::CompareMaskGtI32:
-		case IR::Opcode::CompareMaskGeI32:
-		case IR::Opcode::CompareMaskLtI32:
-		case IR::Opcode::CompareMaskLeI32:
-		case IR::Opcode::CompareEqU16:
-		case IR::Opcode::CompareNeU16:
-		case IR::Opcode::CompareGtU16:
-		case IR::Opcode::CompareGeU16:
-		case IR::Opcode::CompareLtU16:
-		case IR::Opcode::CompareLeU16:
-		case IR::Opcode::CompareEqF32:
-		case IR::Opcode::CompareNeF32:
-		case IR::Opcode::CompareGtF32:
-		case IR::Opcode::CompareGeF32:
-		case IR::Opcode::CompareLtF32:
-		case IR::Opcode::CompareLeF32:
-		case IR::Opcode::CompareOrderedF32:
-		case IR::Opcode::CompareUnorderedF32:
-		case IR::Opcode::CompareUnordEqF32:
-		case IR::Opcode::CompareUnordNeF32:
-		case IR::Opcode::CompareUnordGtF32:
-		case IR::Opcode::CompareUnordGeF32:
-		case IR::Opcode::CompareUnordLtF32:
-		case IR::Opcode::CompareUnordLeF32:
-		case IR::Opcode::CompareClassF32:
-		case IR::Opcode::CompareEqF16:
-		case IR::Opcode::CompareNeF16:
-		case IR::Opcode::CompareGtF16:
-		case IR::Opcode::CompareGeF16:
-		case IR::Opcode::CompareLtF16:
-		case IR::Opcode::CompareLeF16:
-		case IR::Opcode::CompareUnordNeF16:
-		case IR::Opcode::CompareMaskEqF16:
-		case IR::Opcode::CompareMaskNeF16:
-		case IR::Opcode::CompareMaskGtF16:
-		case IR::Opcode::CompareMaskGeF16:
-		case IR::Opcode::CompareMaskLtF16:
-		case IR::Opcode::CompareMaskLeF16:
-		case IR::Opcode::CompareMaskUnordNeF16:
-		case IR::Opcode::CompareMaskUnordGeF16:
-		case IR::Opcode::CompareMaskEqF32:
-		case IR::Opcode::CompareMaskNeF32:
-		case IR::Opcode::CompareMaskGtF32:
-		case IR::Opcode::CompareMaskGeF32:
-		case IR::Opcode::CompareMaskLtF32:
-		case IR::Opcode::CompareMaskLeF32:
-		case IR::Opcode::CompareMaskUnordEqF32:
-		case IR::Opcode::CompareMaskUnordNeF32:
-		case IR::Opcode::CompareMaskUnordGtF32:
-		case IR::Opcode::CompareMaskUnordGeF32:
-		case IR::Opcode::CompareMaskUnordLtF32:
-		case IR::Opcode::CompareMaskUnordLeF32: return true;
-		default: return false;
-	}
-}
-
 bool ProgramNeedsSubgroupLocalInvocationId(const IR::Program& program) {
 	for (const auto& block: program.blocks) {
 		for (const auto& inst: block.instructions) {
@@ -481,7 +392,7 @@ bool ProgramNeedsSubgroupLocalInvocationId(const IR::Program& program) {
 			    InstructionHasDppSource(inst)) {
 				return true;
 			}
-			if (IsCompareOpcode(inst.op)) {
+			if (IR::IsCompareOpcode(inst.op)) {
 				return true;
 			}
 			if (inst.dst.kind == IR::OperandKind::Register &&

src/graphics/shader/recompiler/emitter/spirvEmitterControlFlow.cpp

@@ -516,7 +516,8 @@ void EmitInstruction(EmitterState& state, const IR::Instruction& inst) {
 		case IR::Opcode::CompareGeU32: EmitCompareU32(state, inst, OpUGreaterThanEqual); break;
 		case IR::Opcode::CompareLtU32: EmitCompareU32(state, inst, OpULessThan); break;
 		case IR::Opcode::CompareLeU32: EmitCompareU32(state, inst, OpULessThanEqual); break;
-		case IR::Opcode::CompareNeU64: EmitCompareNeU64(state, inst); break;
+		case IR::Opcode::CompareEqU64:
+		case IR::Opcode::CompareNeU64: EmitCompareU64(state, inst); break;
 		case IR::Opcode::CompareMaskEqU32: EmitCompareMaskU32(state, inst, OpIEqual); break;
 		case IR::Opcode::CompareMaskNeU32: EmitCompareMaskU32(state, inst, OpINotEqual); break;
 		case IR::Opcode::CompareMaskGtU32: EmitCompareMaskU32(state, inst, OpUGreaterThan); break;
@@ -774,6 +775,12 @@ void EmitInstruction(EmitterState& state, const IR::Instruction& inst) {
 		case IR::Opcode::AtomicXorU32:
 			EmitGuardedByExec(state, [&]() { EmitAtomicU32(state, inst, OpAtomicXor); });
 			break;
+		case IR::Opcode::AtomicFMinF32:
+			EmitGuardedByExec(state, [&]() { EmitAtomicFMinF32(state, inst); });
+			break;
+		case IR::Opcode::AtomicFMaxF32:
+			EmitGuardedByExec(state, [&]() { EmitAtomicFMaxF32(state, inst); });
+			break;
 		case IR::Opcode::FlatLoadUbyte: EmitFlatLoadUbyte(state, inst); break;
 		case IR::Opcode::FlatLoadSbyte: EmitFlatLoadSbyte(state, inst); break;
 		case IR::Opcode::FlatLoadUshort: EmitFlatLoadUshort(state, inst); break;
@@ -850,7 +857,13 @@ void EmitInstruction(EmitterState& state, const IR::Instruction& inst) {
 			}
 			EmitGuardedByExec(state, [&]() { EmitExport(state, inst); });
 			break;
-		default: break;
+		default:
+			if (!state.unsupported_ir_instruction) {
+				state.unsupported_ir_instruction = true;
+				state.unsupported_ir_opcode      = inst.op;
+				state.unsupported_ir_pc          = inst.pc;
+			}
+			break;
 	}
 }
 

src/graphics/shader/recompiler/emitter/spirvEmitterInternal.h

@@ -414,6 +414,9 @@ struct EmitterState {
 	bool                                             needs_image_gather_extended           = false;
 	bool                                             needs_function_lds                    = false;
 	bool                                             needs_pixel_valid_mask                = false;
+	bool                                             unsupported_ir_instruction            = false;
+	IR::Opcode                                       unsupported_ir_opcode = IR::Opcode::Count;
+	uint32_t                                         unsupported_ir_pc     = 0;
 	std::vector<RegisterBinding>                     registers;
 	std::vector<InputBinding>                        inputs;
 	std::vector<OutputBinding>                       outputs;
@@ -611,8 +614,6 @@ bool IsMaskRegisterFile(IR::RegisterFile file);
 
 bool IsSccOperand(const IR::Operand& operand);
 
-bool IsCompareOpcode(IR::Opcode op);
-
 void CollectMaskStateRegisters(std::vector<RegisterBinding>& registers);
 
 void CollectSequentialRegisters(std::vector<RegisterBinding>& registers, const IR::Operand& base,
@@ -651,8 +652,6 @@ bool ProgramNeedsSubgroupBallot(const IR::Program& program);
 
 bool ProgramNeedsSubgroupShuffle(const IR::Program& program);
 
-bool IsCompareOpcode(IR::Opcode op);
-
 bool ProgramNeedsSubgroupLocalInvocationId(const IR::Program& program);
 
 uint32_t PointerForRegister(const EmitterState& state, IR::Register reg);
@@ -1015,6 +1014,9 @@ void EmitDeviceAtomicMemoryBarrier(EmitterState& state);
 
 void EmitAtomicU32(EmitterState& state, const IR::Instruction& inst, uint32_t opcode);
 
+void EmitAtomicFMinF32(EmitterState& state, const IR::Instruction& inst);
+void EmitAtomicFMaxF32(EmitterState& state, const IR::Instruction& inst);
+
 void EmitSLoadDword(EmitterState& state, const IR::Instruction& inst);
 
 void EmitLoadSrtDword(EmitterState& state, const IR::Instruction& inst);
@@ -1345,7 +1347,7 @@ void EmitCompareI16(EmitterState& state, const IR::Instruction& inst, uint32_t o
 
 void EmitBitCompareB32(EmitterState& state, const IR::Instruction& inst, bool bit_set);
 
-void EmitCompareNeU64(EmitterState& state, const IR::Instruction& inst);
+void EmitCompareU64(EmitterState& state, const IR::Instruction& inst);
 
 void EmitCompareConstant(EmitterState& state, const IR::Instruction& inst, bool value);
 
@@ -1577,19 +1579,22 @@ uint32_t EmitValueOrZeroIfCondition(EmitterState& state, uint32_t condition, Fn&
 	}
 
 	const auto then_label  = state.builder.AllocateId();
+	const auto then_exit   = state.builder.AllocateId();
 	const auto else_label  = state.builder.AllocateId();
 	const auto merge_label = state.builder.AllocateId();
 	state.builder.AddFunction({OpSelectionMerge, merge_label, SelectionControlNone});
 	state.builder.AddFunction({OpBranchConditional, condition, then_label, else_label});
 	state.builder.AddFunction({OpLabel, then_label});
 	const auto then_value = fn();
+	state.builder.AddFunction({OpBranch, then_exit});
+	state.builder.AddFunction({OpLabel, then_exit});
 	state.builder.AddFunction({OpBranch, merge_label});
 	state.builder.AddFunction({OpLabel, else_label});
 	state.builder.AddFunction({OpBranch, merge_label});
 	state.builder.AddFunction({OpLabel, merge_label});
 	const auto value = state.builder.AllocateId();
 	state.builder.AddFunction(
-	    {OpPhi, state.uint_type, value, then_value, then_label, ConstantU32(state, 0), else_label});
+	    {OpPhi, state.uint_type, value, then_value, then_exit, ConstantU32(state, 0), else_label});
 	return value;
 }
 

src/graphics/shader/recompiler/emitter/spirvEmitterMemory.cpp

@@ -640,8 +640,8 @@ void EmitMemoryStoreU32(EmitterState& state, const IR::Instruction& inst, IR::Re
 }
 
 template <typename Fn>
-void EmitAtomicUpdateU32(EmitterState& state, uint32_t pointer, IR::ResourceKind kind,
-                         Fn&& desired_value) {
+uint32_t EmitAtomicUpdateU32(EmitterState& state, uint32_t pointer, IR::ResourceKind kind,
+                             Fn&& desired_value) {
 	const auto scope          = kind == IR::ResourceKind::Lds ? ScopeWorkgroup : ScopeDevice;
 	const auto memory         = kind == IR::ResourceKind::Lds ? MemorySemanticsWorkgroupMemory
 	                                                          : MemorySemanticsUniformMemory;
@@ -675,6 +675,7 @@ void EmitAtomicUpdateU32(EmitterState& state, uint32_t pointer, IR::ResourceKind
 	const auto semantics = MemorySemanticsAcquireRelease | memory;
 	state.builder.AddFunction(
 	    {OpMemoryBarrier, ConstantU32(state, scope), ConstantU32(state, semantics)});
+	return observed;
 }
 
 void EmitMemoryStoreSubDwordU32(EmitterState& state, const IR::Instruction& inst,
@@ -1181,6 +1182,79 @@ void EmitAtomicU32(EmitterState& state, const IR::Instruction& inst, uint32_t op
 	EmitStoreU32(state, inst.dst, old);
 }
 
+namespace {
+
+uint32_t EmitF32BitsOrderedCompare(EmitterState& state, uint32_t lhs_bits, uint32_t rhs_bits,
+                                   uint32_t comparison_opcode) {
+	struct ClassifiedBits {
+		uint32_t nan  = 0;
+		uint32_t zero = 0;
+		uint32_t key  = 0;
+	};
+	const auto Classify = [&](uint32_t bits) {
+		ClassifiedBits cls;
+		const auto     abs_bits      = EmitAndConstant(state, bits, 0x7fffffffu);
+		const auto     exponent_bits = EmitAndConstant(state, abs_bits, 0x7f800000u);
+		const auto     mantissa_bits = EmitAndConstant(state, abs_bits, 0x007fffffu);
+		const auto     exponent_max =
+		    EmitCompareU32Constant(state, OpIEqual, exponent_bits, 0x7f800000u);
+		const auto mantissa_nonzero = EmitCompareU32Constant(state, OpINotEqual, mantissa_bits, 0);
+		const auto negative = EmitCompareU32Constant(state, OpINotEqual,
+		                                             EmitAndConstant(state, bits, 0x80000000u), 0);
+		const auto negative_key = state.builder.AllocateId();
+		const auto positive_key = state.builder.AllocateId();
+		// Map all non-NaN IEEE-754 encodings to monotonically increasing unsigned keys.
+		state.builder.AddFunction({OpNot, state.uint_type, negative_key, bits});
+		state.builder.AddFunction(
+		    {OpBitwiseXor, state.uint_type, positive_key, bits, ConstantU32(state, 0x80000000u)});
+		cls.nan  = EmitLogicalAndBool(state, exponent_max, mantissa_nonzero);
+		cls.zero = EmitCompareU32Constant(state, OpIEqual, abs_bits, 0);
+		cls.key  = EmitSelectValueU32(state, negative, negative_key, positive_key);
+		return cls;
+	};
+
+	const auto lhs     = Classify(lhs_bits);
+	const auto rhs     = Classify(rhs_bits);
+	const auto any_nan = EmitLogicalOrBool(state, lhs.nan, rhs.nan);
+	// The key transform orders -0 below +0, while the floating comparison treats them as equal.
+	const auto both_zero = EmitLogicalAndBool(state, lhs.zero, rhs.zero);
+	const auto ordered_nonzero =
+	    EmitLogicalNotBool(state, EmitLogicalOrBool(state, any_nan, both_zero));
+	const auto comparison = state.builder.AllocateId();
+	state.builder.AddFunction({comparison_opcode, state.bool_type, comparison, lhs.key, rhs.key});
+	return EmitLogicalAndBool(state, ordered_nonzero, comparison);
+}
+
+void EmitAtomicFMinMaxF32(EmitterState& state, const IR::Instruction& inst,
+                          uint32_t comparison_opcode) {
+	const auto index =
+	    EmitMemoryDwordIndex(state, inst, inst.memory, 1, AddressSourceCount(inst, 1));
+	const auto in_bounds = EmitStorageBufferElementInBounds(state, inst.memory, index, inst.pc);
+	const auto src_u32   = EmitValueLoad(state, inst.src[0]);
+	const auto old       = EmitValueOrZeroIfCondition(state, in_bounds, [&]() {
+		const auto pointer = EmitStorageBufferElementPointer(state, inst.memory, index, inst.pc);
+		return EmitAtomicUpdateU32(state, pointer, inst.memory.kind, [&](uint32_t old_u32) {
+			const auto replace_old = state.builder.AllocateId();
+			const auto replace =
+			    EmitF32BitsOrderedCompare(state, src_u32, old_u32, comparison_opcode);
+			state.builder.AddFunction(
+			    {OpSelect, state.uint_type, replace_old, replace, src_u32, old_u32});
+			return replace_old;
+		});
+	});
+	EmitStoreU32(state, inst.dst, old);
+}
+
+} // namespace
+
+void EmitAtomicFMinF32(EmitterState& state, const IR::Instruction& inst) {
+	EmitAtomicFMinMaxF32(state, inst, OpULessThan);
+}
+
+void EmitAtomicFMaxF32(EmitterState& state, const IR::Instruction& inst) {
+	EmitAtomicFMinMaxF32(state, inst, OpUGreaterThan);
+}
+
 void EmitSLoadDword(EmitterState& state, const IR::Instruction& inst) {
 	if (state.address_memory_variable == 0) {
 		ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::AddressMemory,

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

@@ -30,7 +30,9 @@ bool IsAtomic(Opcode op) {
 		case Opcode::AtomicUMaxU32:
 		case Opcode::AtomicAndU32:
 		case Opcode::AtomicOrU32:
-		case Opcode::AtomicXorU32: return true;
+		case Opcode::AtomicXorU32:
+		case Opcode::AtomicFMinF32:
+		case Opcode::AtomicFMaxF32: return true;
 		default: return false;
 	}
 }

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

@@ -490,13 +490,26 @@ bool IsScalarMinMaxOpcode(Decoder::Opcode opcode) {
 	}
 }
 
+bool ResolveIrOpcode(Decoder::Opcode decoded, Opcode& lowered, std::string* error) {
+	const auto opcode = LookupIrOpcode(decoded);
+	if (!opcode.has_value()) {
+		if (error != nullptr) {
+			*error = fmt::format("decoded opcode has no IR lowering: {}",
+			                     Decoder::OpcodeToString(decoded));
+		}
+		return false;
+	}
+	lowered = *opcode;
+	return true;
+}
+
 bool LowerScalarMinMaxWithScc(const Decoder::Instruction& decoded, BasicBlock& block,
                               std::string* error) {
 	Instruction result;
 	result.pc        = decoded.pc;
-	result.op        = LookupIrOpcode(decoded.opcode);
 	result.src_count = 2;
-	if (!LowerRegisterOperand(decoded.dst, result.dst, error) ||
+	if (!ResolveIrOpcode(decoded.opcode, result.op, error) ||
+	    !LowerRegisterOperand(decoded.dst, result.dst, error) ||
 	    !LowerSourceOperand(decoded.src0, result.src[0], error) ||
 	    !LowerSourceOperand(decoded.src1, result.src[1], error)) {
 		return false;
@@ -962,7 +975,12 @@ bool LowerControlInstruction(const Decoder::Instruction& decoded, BasicBlock& bl
 			return LowerControlMarker(decoded, block, Opcode::TtraceData, true, error);
 		case Decoder::Opcode::SInstPrefetch:
 			return LowerControlMarker(decoded, block, Opcode::InstPrefetch, true, error);
-		default: return false;
+		default:
+			if (error != nullptr) {
+				*error = fmt::format("control opcode has no specialized IR lowering: {}",
+				                     Decoder::OpcodeToString(decoded.opcode));
+			}
+			return false;
 	}
 }
 
@@ -1051,6 +1069,18 @@ bool IsTerminatorOpcode(Decoder::Opcode opcode) {
 
 bool LowerImplemented(const Decoder::Instruction& decoded, BasicBlock& block, std::string* error);
 
+bool IsMemoryFamily(Decoder::Family family) {
+	switch (family) {
+		case Decoder::Family::SMEM:
+		case Decoder::Family::MUBUF:
+		case Decoder::Family::MTBUF:
+		case Decoder::Family::FLAT:
+		case Decoder::Family::DS:
+		case Decoder::Family::MIMG: return true;
+		default: return false;
+	}
+}
+
 bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block,
                              std::string* error) {
 	switch (inst.opcode) {
@@ -1104,6 +1134,15 @@ bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block
 	if (IsMemoryOpcode(inst.opcode)) {
 		return LowerMemoryInstruction(inst, block, error);
 	}
+	// Memory instructions require specialized lowering to populate MemoryInfo. Never let a
+	// newly decoded memory-family opcode fall through to generic register-only IR.
+	if (IsMemoryFamily(inst.family)) {
+		if (error != nullptr) {
+			*error = fmt::format("decoded memory-family opcode has no specialized IR lowering: {}",
+			                     Decoder::OpcodeToString(inst.opcode));
+		}
+		return false;
+	}
 	if (ScalarShiftLeftAddAmount(inst.opcode) != 0) {
 		return LowerScalarShiftLeftAdd(inst, block, error);
 	}
@@ -1116,23 +1155,16 @@ bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block
 	if (VectorByteConvertIndex(inst.opcode) <= 3u) {
 		return LowerVectorByteConvert(inst, block, error);
 	}
-	if (!IsImplemented(inst.opcode)) {
-		if (error != nullptr) {
-			*error = fmt::format("decoded opcode has no IR lowering yet: {}",
-			                     Decoder::InstructionToString(inst).c_str());
-		}
-		return false;
-	}
 	return LowerImplemented(inst, block, error);
 }
 
 bool LowerImplemented(const Decoder::Instruction& decoded, BasicBlock& block, std::string* error) {
 	Instruction inst;
 	inst.pc        = decoded.pc;
-	inst.op        = LookupIrOpcode(decoded.opcode);
 	inst.src_count = decoded.src_count;
 
-	if (!LowerRegisterOperand(decoded.dst, inst.dst, error)) {
+	if (!ResolveIrOpcode(decoded.opcode, inst.op, error) ||
+	    !LowerRegisterOperand(decoded.dst, inst.dst, error)) {
 		return false;
 	}
 	ApplyDppDestinationMask(decoded, inst.dst);

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

@@ -8,350 +8,48 @@
 #include "graphics/shader/shader.h"
 
 #include <array>
+#include <string_view>
 #include <vector>
 
 namespace Libs::Graphics::ShaderRecompiler::IR {
 
 enum class Opcode {
-	ControlNop,
-	Waitcnt,
-	Barrier,
-	Sendmsg,
-	TtraceData,
-	InstPrefetch,
-	MoveU32,
-	MoveF32Bits,
-	MoveRelDestU32,
-	MoveRelSourceU32,
-	MoveU64,
-	WqmB64,
-	SaveexecB32,
-	SaveexecB64,
-	ReadFirstLaneU32,
-	ReadLaneU32,
-	WriteLaneU32,
-	Permlane16B32,
-	Permlanex16B32,
-	AbsI32,
-	IAddU32,
-	IAddCarryU32,
-	ISubBorrowU32,
-	ScalarAddCarryU32,
-	ScalarSubBorrowU32,
-	ScalarSubBorrowCarryU32,
-	ScalarSignedAddOverflowI32,
-	ScalarSignedSubOverflowI32,
-	ScalarShiftLeftAddCarryU32,
-	ISubU32,
-	IMulU32,
-	UMulHighU32,
-	SMulHighI32,
-	IMadI24U32,
-	UMadU24U32,
-	UMadU64U32,
-	SadU32,
-	IAdd3U32,
-	IMulI24U32,
-	UMulU24U32,
-	IMinI32,
-	IMaxI32,
-	IMin3I32,
-	IMax3I32,
-	IMed3I32,
-	UMinU32,
-	UMaxU32,
-	UMin3U32,
-	UMax3U32,
-	UMed3U32,
-	BitwiseAndU32,
-	BitwiseAndU64,
-	BitwiseAndNotU32,
-	BitwiseAndNotU64,
-	BitwiseOrU32,
-	BitwiseOrU64,
-	BitwiseOrNotU32,
-	BitwiseOrNotU64,
-	BitwiseAndOrU32,
-	BitwiseOr3U32,
-	BitwiseXorU32,
-	BitwiseXorU64,
-	BitwiseXor3U32,
-	BitwiseNandU32,
-	BitwiseNandU64,
-	BitwiseNorU32,
-	BitwiseNorU64,
-	BitwiseXnorU32,
-	BitwiseXnorU64,
-	BitwiseNotU32,
-	BitwiseNotU64,
-	BitClearU32,
-	BitSetU32,
-	BitReverseU32,
-	BitCountU32,
-	BitCountU64,
-	BitReplicateB64B32,
-	BitCountAddU32,
-	MaskedBitCountLowU32,
-	MaskedBitCountHighU32,
-	FindLsbU32,
-	FindMsbFromHighU32,
-	FindMsbFromHighU64,
-	BitFieldMaskU32,
-	BitFieldMaskU64,
-	BitFieldExtractU32,
-	BitFieldExtractU64,
-	BitFieldExtract3U32,
-	BitFieldExtract3I32,
-	BitFieldInsertSelectU32,
-	BitCompare0B32,
-	BitCompare1B32,
-	AlignBitU32,
-	ShiftLeftAddU32,
-	AddShiftLeftU32,
-	XorAddU32,
-	ShiftLeftOrU32,
-	ShiftLeftLogicalU32,
-	ShiftLeftLogicalU64,
-	ShiftLeftLogicalU16,
-	ShiftRightLogicalU32,
-	ShiftRightLogicalU64,
-	ShiftRightLogicalU16,
-	ShiftRightArithmeticI32,
-	ShiftRightArithmeticI16,
-	SelectU32,
-	SelectMaskU32,
-	SelectF32Bits,
-	SelectMaskF32Bits,
-	SelectU64,
-	PackLowLowU16,
-	PackLowHighU16,
-	PackHighHighU16,
-	CompareFalse,
-	CompareTrue,
-	CompareEqU32,
-	CompareNeU32,
-	CompareGtU32,
-	CompareGeU32,
-	CompareLtU32,
-	CompareLeU32,
-	CompareNeU64,
-	CompareMaskEqU32,
-	CompareMaskNeU32,
-	CompareMaskGtU32,
-	CompareMaskGeU32,
-	CompareMaskLtU32,
-	CompareMaskLeU32,
-	CompareEqI32,
-	CompareNeI32,
-	CompareGtI32,
-	CompareGeI32,
-	CompareLtI32,
-	CompareLeI32,
-	CompareEqI16,
-	CompareNeI16,
-	CompareGtI16,
-	CompareGeI16,
-	CompareLtI16,
-	CompareLeI16,
-	CompareMaskEqI32,
-	CompareMaskNeI32,
-	CompareMaskGtI32,
-	CompareMaskGeI32,
-	CompareMaskLtI32,
-	CompareMaskLeI32,
-	CompareEqU16,
-	CompareNeU16,
-	CompareGtU16,
-	CompareGeU16,
-	CompareLtU16,
-	CompareLeU16,
-	CompareEqF32,
-	CompareNeF32,
-	CompareGtF32,
-	CompareGeF32,
-	CompareLtF32,
-	CompareLeF32,
-	CompareOrderedF32,
-	CompareUnorderedF32,
-	CompareUnordEqF32,
-	CompareUnordNeF32,
-	CompareUnordGtF32,
-	CompareUnordGeF32,
-	CompareUnordLtF32,
-	CompareUnordLeF32,
-	CompareClassF32,
-	CompareEqF16,
-	CompareNeF16,
-	CompareGtF16,
-	CompareGeF16,
-	CompareLtF16,
-	CompareLeF16,
-	CompareUnordNeF16,
-	CompareMaskEqF16,
-	CompareMaskNeF16,
-	CompareMaskGtF16,
-	CompareMaskGeF16,
-	CompareMaskLtF16,
-	CompareMaskLeF16,
-	CompareMaskUnordNeF16,
-	CompareMaskUnordGeF16,
-	CompareMaskEqF32,
-	CompareMaskNeF32,
-	CompareMaskGtF32,
-	CompareMaskGeF32,
-	CompareMaskLtF32,
-	CompareMaskLeF32,
-	CompareMaskUnordEqF32,
-	CompareMaskUnordNeF32,
-	CompareMaskUnordGtF32,
-	CompareMaskUnordGeF32,
-	CompareMaskUnordLtF32,
-	CompareMaskUnordLeF32,
-	ConvertByteU32ToF32,
-	ConvertU32ToF32,
-	ConvertI32ToF32,
-	ConvertF32ToU32,
-	ConvertF32ToI32,
-	ConvertF32ToF16,
-	ConvertF16ToF32,
-	ConvertU16ToF16,
-	ConvertF16ToU16,
-	ConvertI16ToF16,
-	ConvertF16ToI16,
-	ConvertRoundPlusInfF32ToI32,
-	ConvertFloorF32ToI32,
-	ConvertI4ToOffsetF32,
-	LdexpF32,
-	PackF32ToF16Rtz,
-	PackSnorm2x16F32,
-	PackUnorm2x16F32,
-	PackU16U32,
-	PackU8F32,
-	PackB32F16,
-	PackedMadI16,
-	PackedMulLoU16,
-	PackedAddI16,
-	PackedSubI16,
-	PackedLshlrevB16,
-	PackedLshrrevB16,
-	PackedAshrrevI16,
-	PackedMaxI16,
-	PackedMinI16,
-	PackedMadU16,
-	PackedAddU16,
-	PackedSubU16,
-	PackedMaxU16,
-	PackedMinU16,
-	PackedAddF16,
-	PackedMulF16,
-	PackedMinF16,
-	PackedMaxF16,
-	PackedFmaF16,
-	AddF16,
-	SubF16,
-	MulF16,
-	MinF16,
-	MaxF16,
-	FmaF16,
-	MadMixF16,
-	IAddU16,
-	ISubI16,
-	IMinI16,
-	IMaxI16,
-	UMinU16,
-	UMaxU16,
-	RcpF32,
-	FractF32,
-	TruncF32,
-	CeilF32,
-	RoundEvenF32,
-	FloorF32,
-	Exp2F32,
-	Log2F32,
-	InverseSqrtF32,
-	SqrtF32,
-	RcpF16,
-	SqrtF16,
-	InverseSqrtF16,
-	Log2F16,
-	Exp2F16,
-	FloorF16,
-	CeilF16,
-	TruncF16,
-	RoundEvenF16,
-	SinF32,
-	CosF32,
-	CubeIdF32,
-	CubeScF32,
-	CubeTcF32,
-	CubeMaF32,
-	FAddF32,
-	FSubF32,
-	FMulF32,
-	FMinF32,
-	FMaxF32,
-	FMadF32,
-	Dot2AccF32F16,
-	FMin3F32,
-	FMax3F32,
-	FMed3F32,
-	Min3F16,
-	Max3F16,
-	Med3F16,
-	LoadSrtDword,
-	SLoadDword,
-	SBufferLoadDword,
-	BufferLoadUbyte,
-	BufferLoadSbyte,
-	BufferLoadUshort,
-	BufferLoadSshort,
-	BufferLoadDword,
-	BufferStoreByte,
-	BufferStoreShort,
-	BufferStoreDword,
-	AtomicSwapU32,
-	AtomicAddU32,
-	AtomicSubU32,
-	AtomicSMinI32,
-	AtomicUMinU32,
-	AtomicSMaxI32,
-	AtomicUMaxU32,
-	AtomicAndU32,
-	AtomicOrU32,
-	AtomicXorU32,
-	FlatLoadUbyte,
-	FlatLoadSbyte,
-	FlatLoadUshort,
-	FlatLoadSshort,
-	FlatLoadDword,
-	FlatStoreByte,
-	FlatStoreShort,
-	FlatStoreDword,
-	DsReadUbyte,
-	DsReadSbyte,
-	DsReadUshort,
-	DsReadSshort,
-	DsReadB32,
-	DsWriteByte,
-	DsWriteShort,
-	DsWriteB32,
-	DsMinF32,
-	DsMaxF32,
-	DsSwizzleB32,
-	DsConsume,
-	DsAppend,
-	DsWriteAddtidB32,
-	DsReadAddtidB32,
-	ImageGetResinfo,
-	ImageGetLod,
-	ImageLoad,
-	ImageStore,
-	ImageSample,
-	ImageGather4,
-	LoadInputF32,
-	Export,
+#define IR_OPCODE(name, category) name,
+#include "graphics/shader/recompiler/ir/ShaderIROpcodes.inc"
+#undef IR_OPCODE
+	Count,
 };
 
+enum class OpcodeClass { General, Compare, Compare64 };
+
+struct OpcodeInfo {
+	std::string_view name;
+	OpcodeClass      opcode_class = OpcodeClass::General;
+};
+
+inline constexpr OpcodeInfo OPCODE_INFO[] = {
+#define IR_OPCODE(name, category) {#name, OpcodeClass::category},
+#include "graphics/shader/recompiler/ir/ShaderIROpcodes.inc"
+#undef IR_OPCODE
+};
+static_assert(sizeof(OPCODE_INFO) / sizeof(OPCODE_INFO[0]) == static_cast<size_t>(Opcode::Count));
+
+constexpr const OpcodeInfo& GetOpcodeInfo(Opcode opcode) {
+	return OPCODE_INFO[static_cast<size_t>(opcode)];
+}
+
+constexpr std::string_view OpcodeName(Opcode opcode) {
+	return GetOpcodeInfo(opcode).name;
+}
+
+constexpr bool IsCompareOpcode(Opcode opcode) {
+	return GetOpcodeInfo(opcode).opcode_class != OpcodeClass::General;
+}
+
+constexpr bool IsCompare64Opcode(Opcode opcode) {
+	return GetOpcodeInfo(opcode).opcode_class == OpcodeClass::Compare64;
+}
+
 enum class RegisterFile { Scalar, Vector, Vcc, Exec, Scc, M0 };
 
 struct Register {

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

@@ -3,16 +3,17 @@
 
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 
+#include <optional>
+
 namespace Libs::Graphics::ShaderRecompiler::IR {
 
 void SetError(std::string* error, const char* message);
 
-Opcode LookupIrOpcode(Decoder::Opcode opcode);
-bool   IsImplemented(Decoder::Opcode opcode);
-bool   IsReversedBinary(Decoder::Opcode opcode);
-bool   IsVectorCarryOutOpcode(Decoder::Opcode opcode);
-bool   ScalarResultWritesSccNonZero(Decoder::Opcode opcode);
-bool   ScalarResultIs64Bit(Decoder::Opcode opcode);
+std::optional<Opcode> LookupIrOpcode(Decoder::Opcode opcode);
+bool                  IsReversedBinary(Decoder::Opcode opcode);
+bool                  IsVectorCarryOutOpcode(Decoder::Opcode opcode);
+bool                  ScalarResultWritesSccNonZero(Decoder::Opcode opcode);
+bool                  ScalarResultIs64Bit(Decoder::Opcode opcode);
 
 Operand  MakeSccOperand();
 Operand  MakeImmediateU32(uint32_t value);

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

@@ -157,345 +157,7 @@ std::string InstructionToString(const Instruction& inst) {
 		text += ExportInfoToString(inst.export_info);
 		return text;
 	}
-	switch (inst.op) {
-		case Opcode::ControlNop: text += "ControlNop"; break;
-		case Opcode::Waitcnt: text += "Waitcnt"; break;
-		case Opcode::Barrier: text += "Barrier"; break;
-		case Opcode::Sendmsg: text += "Sendmsg"; break;
-		case Opcode::TtraceData: text += "TtraceData"; break;
-		case Opcode::InstPrefetch: text += "InstPrefetch"; break;
-		case Opcode::MoveU32: text += "MoveU32"; break;
-		case Opcode::MoveF32Bits: text += "MoveF32Bits"; break;
-		case Opcode::MoveRelDestU32: text += "MoveRelDestU32"; break;
-		case Opcode::MoveRelSourceU32: text += "MoveRelSourceU32"; break;
-		case Opcode::MoveU64: text += "MoveU64"; break;
-		case Opcode::WqmB64: text += "WqmB64"; break;
-		case Opcode::SaveexecB32: text += "SaveexecB32"; break;
-		case Opcode::SaveexecB64: text += "SaveexecB64"; break;
-		case Opcode::ReadFirstLaneU32: text += "ReadFirstLaneU32"; break;
-		case Opcode::ReadLaneU32: text += "ReadLaneU32"; break;
-		case Opcode::WriteLaneU32: text += "WriteLaneU32"; break;
-		case Opcode::Permlane16B32: text += "Permlane16B32"; break;
-		case Opcode::Permlanex16B32: text += "Permlanex16B32"; break;
-		case Opcode::AbsI32: text += "AbsI32"; break;
-		case Opcode::IAddU32: text += "IAddU32"; break;
-		case Opcode::IAddCarryU32: text += "IAddCarryU32"; break;
-		case Opcode::ISubBorrowU32: text += "ISubBorrowU32"; break;
-		case Opcode::ScalarAddCarryU32: text += "ScalarAddCarryU32"; break;
-		case Opcode::ScalarSubBorrowU32: text += "ScalarSubBorrowU32"; break;
-		case Opcode::ScalarSubBorrowCarryU32: text += "ScalarSubBorrowCarryU32"; break;
-		case Opcode::ScalarSignedAddOverflowI32: text += "ScalarSignedAddOverflowI32"; break;
-		case Opcode::ScalarSignedSubOverflowI32: text += "ScalarSignedSubOverflowI32"; break;
-		case Opcode::ScalarShiftLeftAddCarryU32: text += "ScalarShiftLeftAddCarryU32"; break;
-		case Opcode::ISubU32: text += "ISubU32"; break;
-		case Opcode::IMulU32: text += "IMulU32"; break;
-		case Opcode::UMulHighU32: text += "UMulHighU32"; break;
-		case Opcode::SMulHighI32: text += "SMulHighI32"; break;
-		case Opcode::IMadI24U32: text += "IMadI24U32"; break;
-		case Opcode::UMadU24U32: text += "UMadU24U32"; break;
-		case Opcode::UMadU64U32: text += "UMadU64U32"; break;
-		case Opcode::SadU32: text += "SadU32"; break;
-		case Opcode::IAdd3U32: text += "IAdd3U32"; break;
-		case Opcode::IMulI24U32: text += "IMulI24U32"; break;
-		case Opcode::UMulU24U32: text += "UMulU24U32"; break;
-		case Opcode::IMinI32: text += "IMinI32"; break;
-		case Opcode::IMaxI32: text += "IMaxI32"; break;
-		case Opcode::IMin3I32: text += "IMin3I32"; break;
-		case Opcode::IMax3I32: text += "IMax3I32"; break;
-		case Opcode::IMed3I32: text += "IMed3I32"; break;
-		case Opcode::UMinU32: text += "UMinU32"; break;
-		case Opcode::UMaxU32: text += "UMaxU32"; break;
-		case Opcode::UMin3U32: text += "UMin3U32"; break;
-		case Opcode::UMax3U32: text += "UMax3U32"; break;
-		case Opcode::UMed3U32: text += "UMed3U32"; break;
-		case Opcode::BitwiseAndU32: text += "BitwiseAndU32"; break;
-		case Opcode::BitwiseAndU64: text += "BitwiseAndU64"; break;
-		case Opcode::BitwiseAndNotU32: text += "BitwiseAndNotU32"; break;
-		case Opcode::BitwiseAndNotU64: text += "BitwiseAndNotU64"; break;
-		case Opcode::BitwiseOrU32: text += "BitwiseOrU32"; break;
-		case Opcode::BitwiseOrU64: text += "BitwiseOrU64"; break;
-		case Opcode::BitwiseOrNotU32: text += "BitwiseOrNotU32"; break;
-		case Opcode::BitwiseOrNotU64: text += "BitwiseOrNotU64"; break;
-		case Opcode::BitwiseAndOrU32: text += "BitwiseAndOrU32"; break;
-		case Opcode::BitwiseOr3U32: text += "BitwiseOr3U32"; break;
-		case Opcode::BitwiseXorU32: text += "BitwiseXorU32"; break;
-		case Opcode::BitwiseXorU64: text += "BitwiseXorU64"; break;
-		case Opcode::BitwiseXor3U32: text += "BitwiseXor3U32"; break;
-		case Opcode::BitwiseNandU32: text += "BitwiseNandU32"; break;
-		case Opcode::BitwiseNandU64: text += "BitwiseNandU64"; break;
-		case Opcode::BitwiseNorU32: text += "BitwiseNorU32"; break;
-		case Opcode::BitwiseNorU64: text += "BitwiseNorU64"; break;
-		case Opcode::BitwiseXnorU32: text += "BitwiseXnorU32"; break;
-		case Opcode::BitwiseXnorU64: text += "BitwiseXnorU64"; break;
-		case Opcode::BitwiseNotU32: text += "BitwiseNotU32"; break;
-		case Opcode::BitwiseNotU64: text += "BitwiseNotU64"; break;
-		case Opcode::BitClearU32: text += "BitClearU32"; break;
-		case Opcode::BitSetU32: text += "BitSetU32"; break;
-		case Opcode::BitReverseU32: text += "BitReverseU32"; break;
-		case Opcode::BitCountU32: text += "BitCountU32"; break;
-		case Opcode::BitCountU64: text += "BitCountU64"; break;
-		case Opcode::BitReplicateB64B32: text += "BitReplicateB64B32"; break;
-		case Opcode::BitCountAddU32: text += "BitCountAddU32"; break;
-		case Opcode::MaskedBitCountLowU32: text += "MaskedBitCountLowU32"; break;
-		case Opcode::MaskedBitCountHighU32: text += "MaskedBitCountHighU32"; break;
-		case Opcode::FindLsbU32: text += "FindLsbU32"; break;
-		case Opcode::FindMsbFromHighU32: text += "FindMsbFromHighU32"; break;
-		case Opcode::FindMsbFromHighU64: text += "FindMsbFromHighU64"; break;
-		case Opcode::BitFieldMaskU32: text += "BitFieldMaskU32"; break;
-		case Opcode::BitFieldMaskU64: text += "BitFieldMaskU64"; break;
-		case Opcode::BitFieldExtractU32: text += "BitFieldExtractU32"; break;
-		case Opcode::BitFieldExtractU64: text += "BitFieldExtractU64"; break;
-		case Opcode::BitFieldExtract3U32: text += "BitFieldExtract3U32"; break;
-		case Opcode::BitFieldExtract3I32: text += "BitFieldExtract3I32"; break;
-		case Opcode::BitFieldInsertSelectU32: text += "BitFieldInsertSelectU32"; break;
-		case Opcode::BitCompare0B32: text += "BitCompare0B32"; break;
-		case Opcode::BitCompare1B32: text += "BitCompare1B32"; break;
-		case Opcode::AlignBitU32: text += "AlignBitU32"; break;
-		case Opcode::ShiftLeftAddU32: text += "ShiftLeftAddU32"; break;
-		case Opcode::AddShiftLeftU32: text += "AddShiftLeftU32"; break;
-		case Opcode::XorAddU32: text += "XorAddU32"; break;
-		case Opcode::ShiftLeftOrU32: text += "ShiftLeftOrU32"; break;
-		case Opcode::ShiftLeftLogicalU32: text += "ShiftLeftLogicalU32"; break;
-		case Opcode::ShiftLeftLogicalU64: text += "ShiftLeftLogicalU64"; break;
-		case Opcode::ShiftLeftLogicalU16: text += "ShiftLeftLogicalU16"; break;
-		case Opcode::ShiftRightLogicalU32: text += "ShiftRightLogicalU32"; break;
-		case Opcode::ShiftRightLogicalU64: text += "ShiftRightLogicalU64"; break;
-		case Opcode::ShiftRightLogicalU16: text += "ShiftRightLogicalU16"; break;
-		case Opcode::ShiftRightArithmeticI32: text += "ShiftRightArithmeticI32"; break;
-		case Opcode::ShiftRightArithmeticI16: text += "ShiftRightArithmeticI16"; break;
-		case Opcode::SelectU32: text += "SelectU32"; break;
-		case Opcode::SelectMaskU32: text += "SelectMaskU32"; break;
-		case Opcode::SelectF32Bits: text += "SelectF32Bits"; break;
-		case Opcode::SelectMaskF32Bits: text += "SelectMaskF32Bits"; break;
-		case Opcode::SelectU64: text += "SelectU64"; break;
-		case Opcode::PackLowLowU16: text += "PackLowLowU16"; break;
-		case Opcode::PackLowHighU16: text += "PackLowHighU16"; break;
-		case Opcode::PackHighHighU16: text += "PackHighHighU16"; break;
-		case Opcode::CompareFalse: text += "CompareFalse"; break;
-		case Opcode::CompareTrue: text += "CompareTrue"; break;
-		case Opcode::CompareEqU32: text += "CompareEqU32"; break;
-		case Opcode::CompareNeU32: text += "CompareNeU32"; break;
-		case Opcode::CompareGtU32: text += "CompareGtU32"; break;
-		case Opcode::CompareGeU32: text += "CompareGeU32"; break;
-		case Opcode::CompareLtU32: text += "CompareLtU32"; break;
-		case Opcode::CompareLeU32: text += "CompareLeU32"; break;
-		case Opcode::CompareNeU64: text += "CompareNeU64"; break;
-		case Opcode::CompareMaskEqU32: text += "CompareMaskEqU32"; break;
-		case Opcode::CompareMaskNeU32: text += "CompareMaskNeU32"; break;
-		case Opcode::CompareMaskGtU32: text += "CompareMaskGtU32"; break;
-		case Opcode::CompareMaskGeU32: text += "CompareMaskGeU32"; break;
-		case Opcode::CompareMaskLtU32: text += "CompareMaskLtU32"; break;
-		case Opcode::CompareMaskLeU32: text += "CompareMaskLeU32"; break;
-		case Opcode::CompareEqI32: text += "CompareEqI32"; break;
-		case Opcode::CompareNeI32: text += "CompareNeI32"; break;
-		case Opcode::CompareGtI32: text += "CompareGtI32"; break;
-		case Opcode::CompareGeI32: text += "CompareGeI32"; break;
-		case Opcode::CompareLtI32: text += "CompareLtI32"; break;
-		case Opcode::CompareLeI32: text += "CompareLeI32"; break;
-		case Opcode::CompareEqI16: text += "CompareEqI16"; break;
-		case Opcode::CompareNeI16: text += "CompareNeI16"; break;
-		case Opcode::CompareGtI16: text += "CompareGtI16"; break;
-		case Opcode::CompareGeI16: text += "CompareGeI16"; break;
-		case Opcode::CompareLtI16: text += "CompareLtI16"; break;
-		case Opcode::CompareLeI16: text += "CompareLeI16"; break;
-		case Opcode::CompareMaskEqI32: text += "CompareMaskEqI32"; break;
-		case Opcode::CompareMaskNeI32: text += "CompareMaskNeI32"; break;
-		case Opcode::CompareMaskGtI32: text += "CompareMaskGtI32"; break;
-		case Opcode::CompareMaskGeI32: text += "CompareMaskGeI32"; break;
-		case Opcode::CompareMaskLtI32: text += "CompareMaskLtI32"; break;
-		case Opcode::CompareMaskLeI32: text += "CompareMaskLeI32"; break;
-		case Opcode::CompareEqU16: text += "CompareEqU16"; break;
-		case Opcode::CompareNeU16: text += "CompareNeU16"; break;
-		case Opcode::CompareGtU16: text += "CompareGtU16"; break;
-		case Opcode::CompareGeU16: text += "CompareGeU16"; break;
-		case Opcode::CompareLtU16: text += "CompareLtU16"; break;
-		case Opcode::CompareLeU16: text += "CompareLeU16"; break;
-		case Opcode::CompareEqF32: text += "CompareEqF32"; break;
-		case Opcode::CompareNeF32: text += "CompareNeF32"; break;
-		case Opcode::CompareGtF32: text += "CompareGtF32"; break;
-		case Opcode::CompareGeF32: text += "CompareGeF32"; break;
-		case Opcode::CompareLtF32: text += "CompareLtF32"; break;
-		case Opcode::CompareLeF32: text += "CompareLeF32"; break;
-		case Opcode::CompareOrderedF32: text += "CompareOrderedF32"; break;
-		case Opcode::CompareUnorderedF32: text += "CompareUnorderedF32"; break;
-		case Opcode::CompareUnordEqF32: text += "CompareUnordEqF32"; break;
-		case Opcode::CompareUnordNeF32: text += "CompareUnordNeF32"; break;
-		case Opcode::CompareUnordGtF32: text += "CompareUnordGtF32"; break;
-		case Opcode::CompareUnordGeF32: text += "CompareUnordGeF32"; break;
-		case Opcode::CompareUnordLtF32: text += "CompareUnordLtF32"; break;
-		case Opcode::CompareUnordLeF32: text += "CompareUnordLeF32"; break;
-		case Opcode::CompareClassF32: text += "CompareClassF32"; break;
-		case Opcode::CompareEqF16: text += "CompareEqF16"; break;
-		case Opcode::CompareNeF16: text += "CompareNeF16"; break;
-		case Opcode::CompareGtF16: text += "CompareGtF16"; break;
-		case Opcode::CompareGeF16: text += "CompareGeF16"; break;
-		case Opcode::CompareLtF16: text += "CompareLtF16"; break;
-		case Opcode::CompareLeF16: text += "CompareLeF16"; break;
-		case Opcode::CompareUnordNeF16: text += "CompareUnordNeF16"; break;
-		case Opcode::CompareMaskEqF16: text += "CompareMaskEqF16"; break;
-		case Opcode::CompareMaskNeF16: text += "CompareMaskNeF16"; break;
-		case Opcode::CompareMaskGtF16: text += "CompareMaskGtF16"; break;
-		case Opcode::CompareMaskGeF16: text += "CompareMaskGeF16"; break;
-		case Opcode::CompareMaskLtF16: text += "CompareMaskLtF16"; break;
-		case Opcode::CompareMaskLeF16: text += "CompareMaskLeF16"; break;
-		case Opcode::CompareMaskUnordNeF16: text += "CompareMaskUnordNeF16"; break;
-		case Opcode::CompareMaskUnordGeF16: text += "CompareMaskUnordGeF16"; break;
-		case Opcode::CompareMaskEqF32: text += "CompareMaskEqF32"; break;
-		case Opcode::CompareMaskNeF32: text += "CompareMaskNeF32"; break;
-		case Opcode::CompareMaskGtF32: text += "CompareMaskGtF32"; break;
-		case Opcode::CompareMaskGeF32: text += "CompareMaskGeF32"; break;
-		case Opcode::CompareMaskLtF32: text += "CompareMaskLtF32"; break;
-		case Opcode::CompareMaskLeF32: text += "CompareMaskLeF32"; break;
-		case Opcode::CompareMaskUnordEqF32: text += "CompareMaskUnordEqF32"; break;
-		case Opcode::CompareMaskUnordNeF32: text += "CompareMaskUnordNeF32"; break;
-		case Opcode::CompareMaskUnordGtF32: text += "CompareMaskUnordGtF32"; break;
-		case Opcode::CompareMaskUnordGeF32: text += "CompareMaskUnordGeF32"; break;
-		case Opcode::CompareMaskUnordLtF32: text += "CompareMaskUnordLtF32"; break;
-		case Opcode::CompareMaskUnordLeF32: text += "CompareMaskUnordLeF32"; break;
-		case Opcode::ConvertByteU32ToF32: text += "ConvertByteU32ToF32"; break;
-		case Opcode::ConvertU32ToF32: text += "ConvertU32ToF32"; break;
-		case Opcode::ConvertI32ToF32: text += "ConvertI32ToF32"; break;
-		case Opcode::ConvertF32ToU32: text += "ConvertF32ToU32"; break;
-		case Opcode::ConvertF32ToI32: text += "ConvertF32ToI32"; break;
-		case Opcode::ConvertF32ToF16: text += "ConvertF32ToF16"; break;
-		case Opcode::ConvertF16ToF32: text += "ConvertF16ToF32"; break;
-		case Opcode::ConvertU16ToF16: text += "ConvertU16ToF16"; break;
-		case Opcode::ConvertF16ToU16: text += "ConvertF16ToU16"; break;
-		case Opcode::ConvertI16ToF16: text += "ConvertI16ToF16"; break;
-		case Opcode::ConvertF16ToI16: text += "ConvertF16ToI16"; break;
-		case Opcode::ConvertRoundPlusInfF32ToI32: text += "ConvertRoundPlusInfF32ToI32"; break;
-		case Opcode::ConvertFloorF32ToI32: text += "ConvertFloorF32ToI32"; break;
-		case Opcode::ConvertI4ToOffsetF32: text += "ConvertI4ToOffsetF32"; break;
-		case Opcode::LdexpF32: text += "LdexpF32"; break;
-		case Opcode::PackF32ToF16Rtz: text += "PackF32ToF16Rtz"; break;
-		case Opcode::PackSnorm2x16F32: text += "PackSnorm2x16F32"; break;
-		case Opcode::PackUnorm2x16F32: text += "PackUnorm2x16F32"; break;
-		case Opcode::PackU16U32: text += "PackU16U32"; break;
-		case Opcode::PackU8F32: text += "PackU8F32"; break;
-		case Opcode::PackB32F16: text += "PackB32F16"; break;
-		case Opcode::PackedMadI16: text += "PackedMadI16"; break;
-		case Opcode::PackedMulLoU16: text += "PackedMulLoU16"; break;
-		case Opcode::PackedAddI16: text += "PackedAddI16"; break;
-		case Opcode::PackedSubI16: text += "PackedSubI16"; break;
-		case Opcode::PackedLshlrevB16: text += "PackedLshlrevB16"; break;
-		case Opcode::PackedLshrrevB16: text += "PackedLshrrevB16"; break;
-		case Opcode::PackedAshrrevI16: text += "PackedAshrrevI16"; break;
-		case Opcode::PackedMaxI16: text += "PackedMaxI16"; break;
-		case Opcode::PackedMinI16: text += "PackedMinI16"; break;
-		case Opcode::PackedMadU16: text += "PackedMadU16"; break;
-		case Opcode::PackedAddU16: text += "PackedAddU16"; break;
-		case Opcode::PackedSubU16: text += "PackedSubU16"; break;
-		case Opcode::PackedMaxU16: text += "PackedMaxU16"; break;
-		case Opcode::PackedMinU16: text += "PackedMinU16"; break;
-		case Opcode::PackedAddF16: text += "PackedAddF16"; break;
-		case Opcode::PackedMulF16: text += "PackedMulF16"; break;
-		case Opcode::PackedMinF16: text += "PackedMinF16"; break;
-		case Opcode::PackedMaxF16: text += "PackedMaxF16"; break;
-		case Opcode::PackedFmaF16: text += "PackedFmaF16"; break;
-		case Opcode::AddF16: text += "AddF16"; break;
-		case Opcode::SubF16: text += "SubF16"; break;
-		case Opcode::MulF16: text += "MulF16"; break;
-		case Opcode::MinF16: text += "MinF16"; break;
-		case Opcode::MaxF16: text += "MaxF16"; break;
-		case Opcode::FmaF16: text += "FmaF16"; break;
-		case Opcode::MadMixF16: text += "MadMixF16"; break;
-		case Opcode::IAddU16: text += "IAddU16"; break;
-		case Opcode::ISubI16: text += "ISubI16"; break;
-		case Opcode::IMinI16: text += "IMinI16"; break;
-		case Opcode::IMaxI16: text += "IMaxI16"; break;
-		case Opcode::UMinU16: text += "UMinU16"; break;
-		case Opcode::UMaxU16: text += "UMaxU16"; break;
-		case Opcode::RcpF32: text += "RcpF32"; break;
-		case Opcode::FractF32: text += "FractF32"; break;
-		case Opcode::TruncF32: text += "TruncF32"; break;
-		case Opcode::CeilF32: text += "CeilF32"; break;
-		case Opcode::RoundEvenF32: text += "RoundEvenF32"; break;
-		case Opcode::FloorF32: text += "FloorF32"; break;
-		case Opcode::Exp2F32: text += "Exp2F32"; break;
-		case Opcode::Log2F32: text += "Log2F32"; break;
-		case Opcode::InverseSqrtF32: text += "InverseSqrtF32"; break;
-		case Opcode::SqrtF32: text += "SqrtF32"; break;
-		case Opcode::RcpF16: text += "RcpF16"; break;
-		case Opcode::SqrtF16: text += "SqrtF16"; break;
-		case Opcode::InverseSqrtF16: text += "InverseSqrtF16"; break;
-		case Opcode::Log2F16: text += "Log2F16"; break;
-		case Opcode::Exp2F16: text += "Exp2F16"; break;
-		case Opcode::FloorF16: text += "FloorF16"; break;
-		case Opcode::CeilF16: text += "CeilF16"; break;
-		case Opcode::TruncF16: text += "TruncF16"; break;
-		case Opcode::RoundEvenF16: text += "RoundEvenF16"; break;
-		case Opcode::SinF32: text += "SinF32"; break;
-		case Opcode::CosF32: text += "CosF32"; break;
-		case Opcode::CubeIdF32: text += "CubeIdF32"; break;
-		case Opcode::CubeScF32: text += "CubeScF32"; break;
-		case Opcode::CubeTcF32: text += "CubeTcF32"; break;
-		case Opcode::CubeMaF32: text += "CubeMaF32"; break;
-		case Opcode::FAddF32: text += "FAddF32"; break;
-		case Opcode::FSubF32: text += "FSubF32"; break;
-		case Opcode::FMulF32: text += "FMulF32"; break;
-		case Opcode::FMinF32: text += "FMinF32"; break;
-		case Opcode::FMaxF32: text += "FMaxF32"; break;
-		case Opcode::FMadF32: text += "FMadF32"; break;
-		case Opcode::Dot2AccF32F16: text += "Dot2AccF32F16"; break;
-		case Opcode::FMin3F32: text += "FMin3F32"; break;
-		case Opcode::FMax3F32: text += "FMax3F32"; break;
-		case Opcode::FMed3F32: text += "FMed3F32"; break;
-		case Opcode::Min3F16: text += "Min3F16"; break;
-		case Opcode::Max3F16: text += "Max3F16"; break;
-		case Opcode::Med3F16: text += "Med3F16"; break;
-		case Opcode::LoadSrtDword: text += "LoadSrtDword"; break;
-		case Opcode::SLoadDword: text += "SLoadDword"; break;
-		case Opcode::SBufferLoadDword: text += "SBufferLoadDword"; break;
-		case Opcode::BufferLoadUbyte: text += "BufferLoadUbyte"; break;
-		case Opcode::BufferLoadSbyte: text += "BufferLoadSbyte"; break;
-		case Opcode::BufferLoadUshort: text += "BufferLoadUshort"; break;
-		case Opcode::BufferLoadSshort: text += "BufferLoadSshort"; break;
-		case Opcode::BufferLoadDword: text += "BufferLoadDword"; break;
-		case Opcode::BufferStoreByte: text += "BufferStoreByte"; break;
-		case Opcode::BufferStoreShort: text += "BufferStoreShort"; break;
-		case Opcode::BufferStoreDword: text += "BufferStoreDword"; break;
-		case Opcode::AtomicSwapU32: text += "AtomicSwapU32"; break;
-		case Opcode::AtomicAddU32: text += "AtomicAddU32"; break;
-		case Opcode::AtomicSubU32: text += "AtomicSubU32"; break;
-		case Opcode::AtomicSMinI32: text += "AtomicSMinI32"; break;
-		case Opcode::AtomicUMinU32: text += "AtomicUMinU32"; break;
-		case Opcode::AtomicSMaxI32: text += "AtomicSMaxI32"; break;
-		case Opcode::AtomicUMaxU32: text += "AtomicUMaxU32"; break;
-		case Opcode::AtomicAndU32: text += "AtomicAndU32"; break;
-		case Opcode::AtomicOrU32: text += "AtomicOrU32"; break;
-		case Opcode::AtomicXorU32: text += "AtomicXorU32"; break;
-		case Opcode::FlatLoadUbyte: text += "FlatLoadUbyte"; break;
-		case Opcode::FlatLoadSbyte: text += "FlatLoadSbyte"; break;
-		case Opcode::FlatLoadUshort: text += "FlatLoadUshort"; break;
-		case Opcode::FlatLoadSshort: text += "FlatLoadSshort"; break;
-		case Opcode::FlatLoadDword: text += "FlatLoadDword"; break;
-		case Opcode::FlatStoreByte: text += "FlatStoreByte"; break;
-		case Opcode::FlatStoreShort: text += "FlatStoreShort"; break;
-		case Opcode::FlatStoreDword: text += "FlatStoreDword"; break;
-		case Opcode::DsReadUbyte: text += "DsReadUbyte"; break;
-		case Opcode::DsReadSbyte: text += "DsReadSbyte"; break;
-		case Opcode::DsReadUshort: text += "DsReadUshort"; break;
-		case Opcode::DsReadSshort: text += "DsReadSshort"; break;
-		case Opcode::DsReadB32: text += "DsReadB32"; break;
-		case Opcode::DsWriteByte: text += "DsWriteByte"; break;
-		case Opcode::DsWriteShort: text += "DsWriteShort"; break;
-		case Opcode::DsWriteB32: text += "DsWriteB32"; break;
-		case Opcode::DsMinF32: text += "DsMinF32"; break;
-		case Opcode::DsMaxF32: text += "DsMaxF32"; break;
-		case Opcode::DsSwizzleB32: text += "DsSwizzleB32"; break;
-		case Opcode::DsConsume: text += "DsConsume"; break;
-		case Opcode::DsAppend: text += "DsAppend"; break;
-		case Opcode::DsWriteAddtidB32: text += "DsWriteAddtidB32"; break;
-		case Opcode::DsReadAddtidB32: text += "DsReadAddtidB32"; break;
-		case Opcode::ImageGetResinfo: text += "ImageGetResinfo"; break;
-		case Opcode::ImageGetLod: text += "ImageGetLod"; break;
-		case Opcode::ImageLoad: text += "ImageLoad"; break;
-		case Opcode::ImageStore: text += "ImageStore"; break;
-		case Opcode::ImageSample: text += "ImageSample"; break;
-		case Opcode::ImageGather4: text += "ImageGather4"; break;
-		case Opcode::LoadInputF32: text += "LoadInputF32"; break;
-		case Opcode::Export: text += "Export"; break;
-	}
+	text += OpcodeName(inst.op);
 	text += " ";
 	text += OperandToString(inst.dst);
 	if (inst.dst2.kind != OperandKind::Null) {

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

@@ -569,6 +569,10 @@ bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& blo
 			return LowerBufferAtomicDword(decoded, block, Opcode::AtomicOrU32, error);
 		case Decoder::Opcode::BufferAtomicXor:
 			return LowerBufferAtomicDword(decoded, block, Opcode::AtomicXorU32, error);
+		case Decoder::Opcode::BufferAtomicFMin:
+			return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMinF32, error);
+		case Decoder::Opcode::BufferAtomicFMax:
+			return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMaxF32, error);
 		case Decoder::Opcode::FlatLoadUbyte:
 		case Decoder::Opcode::FlatLoadSbyte:
 		case Decoder::Opcode::FlatLoadUshort:
@@ -673,7 +677,12 @@ bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& blo
 		case Decoder::Opcode::ImageAtomicAnd:
 		case Decoder::Opcode::ImageAtomicOr:
 		case Decoder::Opcode::ImageAtomicXor: return LowerImageAtomicU32(decoded, block, error);
-		default: return false;
+		default:
+			if (error != nullptr) {
+				*error = fmt::format("memory opcode has no specialized IR lowering: {}",
+				                     Decoder::OpcodeToString(decoded.opcode));
+			}
+			return false;
 	}
 }
 
@@ -729,6 +738,8 @@ bool IsMemoryOpcode(Decoder::Opcode opcode) {
 		case Decoder::Opcode::BufferAtomicAnd:
 		case Decoder::Opcode::BufferAtomicOr:
 		case Decoder::Opcode::BufferAtomicXor:
+		case Decoder::Opcode::BufferAtomicFMin:
+		case Decoder::Opcode::BufferAtomicFMax:
 		case Decoder::Opcode::FlatLoadUbyte:
 		case Decoder::Opcode::FlatLoadSbyte:
 		case Decoder::Opcode::FlatLoadUshort:

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

@@ -82,6 +82,7 @@ constexpr LowerMap LOWER_OPS[] = {
     {Decoder::Opcode::SCmpGeU32, Opcode::CompareGeU32},
     {Decoder::Opcode::SCmpLtU32, Opcode::CompareLtU32},
     {Decoder::Opcode::SCmpLeU32, Opcode::CompareLeU32},
+    {Decoder::Opcode::SCmpEqU64, Opcode::CompareEqU64},
     {Decoder::Opcode::SCmpLgU64, Opcode::CompareNeU64},
     {Decoder::Opcode::VReadfirstlaneB32, Opcode::ReadFirstLaneU32},
     {Decoder::Opcode::VReadlaneB32, Opcode::ReadLaneU32},
@@ -325,6 +326,7 @@ constexpr LowerMap LOWER_OPS[] = {
     {Decoder::Opcode::VCmpNeU32, Opcode::CompareNeU32},
     {Decoder::Opcode::VCmpGeU32, Opcode::CompareGeU32},
     {Decoder::Opcode::VCmpTU32, Opcode::CompareTrue},
+    {Decoder::Opcode::VCmpEqI64, Opcode::CompareEqU64},
     {Decoder::Opcode::VCmpNeU64, Opcode::CompareNeU64},
     {Decoder::Opcode::VCmpxLtU32, Opcode::CompareMaskLtU32},
     {Decoder::Opcode::VCmpxEqU32, Opcode::CompareMaskEqU32},
@@ -350,6 +352,8 @@ constexpr LowerMap LOWER_OPS[] = {
     {Decoder::Opcode::BufferAtomicAnd, Opcode::AtomicAndU32},
     {Decoder::Opcode::BufferAtomicOr, Opcode::AtomicOrU32},
     {Decoder::Opcode::BufferAtomicXor, Opcode::AtomicXorU32},
+    {Decoder::Opcode::BufferAtomicFMin, Opcode::AtomicFMinF32},
+    {Decoder::Opcode::BufferAtomicFMax, Opcode::AtomicFMaxF32},
     {Decoder::Opcode::FlatLoadUbyte, Opcode::FlatLoadUbyte},
     {Decoder::Opcode::FlatLoadSbyte, Opcode::FlatLoadSbyte},
     {Decoder::Opcode::FlatLoadSshort, Opcode::FlatLoadSshort},
@@ -432,22 +436,13 @@ void SetError(std::string* error, const char* message) {
 	}
 }
 
-Opcode LookupIrOpcode(Decoder::Opcode opcode) {
+std::optional<Opcode> LookupIrOpcode(Decoder::Opcode opcode) {
 	for (const auto& op: LOWER_OPS) {
 		if (op.decoded == opcode) {
 			return op.ir;
 		}
 	}
-	return Opcode::MoveU32;
-}
-
-bool IsImplemented(Decoder::Opcode opcode) {
-	for (const auto& op: LOWER_OPS) {
-		if (op.decoded == opcode) {
-			return true;
-		}
-	}
-	return false;
+	return std::nullopt;
 }
 
 bool IsReversedBinary(Decoder::Opcode opcode) {

src/graphics/shader/recompiler/ir/ShaderIROpcodes.inc

@@ -0,0 +1,342 @@
+// Central IR opcode catalog. Keep semantic categories here so enum identity, logging,
+// and generic analysis cannot drift apart.
+IR_OPCODE(ControlNop, General)
+IR_OPCODE(Waitcnt, General)
+IR_OPCODE(Barrier, General)
+IR_OPCODE(Sendmsg, General)
+IR_OPCODE(TtraceData, General)
+IR_OPCODE(InstPrefetch, General)
+IR_OPCODE(MoveU32, General)
+IR_OPCODE(MoveF32Bits, General)
+IR_OPCODE(MoveRelDestU32, General)
+IR_OPCODE(MoveRelSourceU32, General)
+IR_OPCODE(MoveU64, General)
+IR_OPCODE(WqmB64, General)
+IR_OPCODE(SaveexecB32, General)
+IR_OPCODE(SaveexecB64, General)
+IR_OPCODE(ReadFirstLaneU32, General)
+IR_OPCODE(ReadLaneU32, General)
+IR_OPCODE(WriteLaneU32, General)
+IR_OPCODE(Permlane16B32, General)
+IR_OPCODE(Permlanex16B32, General)
+IR_OPCODE(AbsI32, General)
+IR_OPCODE(IAddU32, General)
+IR_OPCODE(IAddCarryU32, General)
+IR_OPCODE(ISubBorrowU32, General)
+IR_OPCODE(ScalarAddCarryU32, General)
+IR_OPCODE(ScalarSubBorrowU32, General)
+IR_OPCODE(ScalarSubBorrowCarryU32, General)
+IR_OPCODE(ScalarSignedAddOverflowI32, General)
+IR_OPCODE(ScalarSignedSubOverflowI32, General)
+IR_OPCODE(ScalarShiftLeftAddCarryU32, General)
+IR_OPCODE(ISubU32, General)
+IR_OPCODE(IMulU32, General)
+IR_OPCODE(UMulHighU32, General)
+IR_OPCODE(SMulHighI32, General)
+IR_OPCODE(IMadI24U32, General)
+IR_OPCODE(UMadU24U32, General)
+IR_OPCODE(UMadU64U32, General)
+IR_OPCODE(SadU32, General)
+IR_OPCODE(IAdd3U32, General)
+IR_OPCODE(IMulI24U32, General)
+IR_OPCODE(UMulU24U32, General)
+IR_OPCODE(IMinI32, General)
+IR_OPCODE(IMaxI32, General)
+IR_OPCODE(IMin3I32, General)
+IR_OPCODE(IMax3I32, General)
+IR_OPCODE(IMed3I32, General)
+IR_OPCODE(UMinU32, General)
+IR_OPCODE(UMaxU32, General)
+IR_OPCODE(UMin3U32, General)
+IR_OPCODE(UMax3U32, General)
+IR_OPCODE(UMed3U32, General)
+IR_OPCODE(BitwiseAndU32, General)
+IR_OPCODE(BitwiseAndU64, General)
+IR_OPCODE(BitwiseAndNotU32, General)
+IR_OPCODE(BitwiseAndNotU64, General)
+IR_OPCODE(BitwiseOrU32, General)
+IR_OPCODE(BitwiseOrU64, General)
+IR_OPCODE(BitwiseOrNotU32, General)
+IR_OPCODE(BitwiseOrNotU64, General)
+IR_OPCODE(BitwiseAndOrU32, General)
+IR_OPCODE(BitwiseOr3U32, General)
+IR_OPCODE(BitwiseXorU32, General)
+IR_OPCODE(BitwiseXorU64, General)
+IR_OPCODE(BitwiseXor3U32, General)
+IR_OPCODE(BitwiseNandU32, General)
+IR_OPCODE(BitwiseNandU64, General)
+IR_OPCODE(BitwiseNorU32, General)
+IR_OPCODE(BitwiseNorU64, General)
+IR_OPCODE(BitwiseXnorU32, General)
+IR_OPCODE(BitwiseXnorU64, General)
+IR_OPCODE(BitwiseNotU32, General)
+IR_OPCODE(BitwiseNotU64, General)
+IR_OPCODE(BitClearU32, General)
+IR_OPCODE(BitSetU32, General)
+IR_OPCODE(BitReverseU32, General)
+IR_OPCODE(BitCountU32, General)
+IR_OPCODE(BitCountU64, General)
+IR_OPCODE(BitReplicateB64B32, General)
+IR_OPCODE(BitCountAddU32, General)
+IR_OPCODE(MaskedBitCountLowU32, General)
+IR_OPCODE(MaskedBitCountHighU32, General)
+IR_OPCODE(FindLsbU32, General)
+IR_OPCODE(FindMsbFromHighU32, General)
+IR_OPCODE(FindMsbFromHighU64, General)
+IR_OPCODE(BitFieldMaskU32, General)
+IR_OPCODE(BitFieldMaskU64, General)
+IR_OPCODE(BitFieldExtractU32, General)
+IR_OPCODE(BitFieldExtractU64, General)
+IR_OPCODE(BitFieldExtract3U32, General)
+IR_OPCODE(BitFieldExtract3I32, General)
+IR_OPCODE(BitFieldInsertSelectU32, General)
+IR_OPCODE(BitCompare0B32, General)
+IR_OPCODE(BitCompare1B32, General)
+IR_OPCODE(AlignBitU32, General)
+IR_OPCODE(ShiftLeftAddU32, General)
+IR_OPCODE(AddShiftLeftU32, General)
+IR_OPCODE(XorAddU32, General)
+IR_OPCODE(ShiftLeftOrU32, General)
+IR_OPCODE(ShiftLeftLogicalU32, General)
+IR_OPCODE(ShiftLeftLogicalU64, General)
+IR_OPCODE(ShiftLeftLogicalU16, General)
+IR_OPCODE(ShiftRightLogicalU32, General)
+IR_OPCODE(ShiftRightLogicalU64, General)
+IR_OPCODE(ShiftRightLogicalU16, General)
+IR_OPCODE(ShiftRightArithmeticI32, General)
+IR_OPCODE(ShiftRightArithmeticI16, General)
+IR_OPCODE(SelectU32, General)
+IR_OPCODE(SelectMaskU32, General)
+IR_OPCODE(SelectF32Bits, General)
+IR_OPCODE(SelectMaskF32Bits, General)
+IR_OPCODE(SelectU64, General)
+IR_OPCODE(PackLowLowU16, General)
+IR_OPCODE(PackLowHighU16, General)
+IR_OPCODE(PackHighHighU16, General)
+IR_OPCODE(CompareFalse, Compare)
+IR_OPCODE(CompareTrue, Compare)
+IR_OPCODE(CompareEqU32, Compare)
+IR_OPCODE(CompareNeU32, Compare)
+IR_OPCODE(CompareGtU32, Compare)
+IR_OPCODE(CompareGeU32, Compare)
+IR_OPCODE(CompareLtU32, Compare)
+IR_OPCODE(CompareLeU32, Compare)
+IR_OPCODE(CompareEqU64, Compare64)
+IR_OPCODE(CompareNeU64, Compare64)
+IR_OPCODE(CompareMaskEqU32, Compare)
+IR_OPCODE(CompareMaskNeU32, Compare)
+IR_OPCODE(CompareMaskGtU32, Compare)
+IR_OPCODE(CompareMaskGeU32, Compare)
+IR_OPCODE(CompareMaskLtU32, Compare)
+IR_OPCODE(CompareMaskLeU32, Compare)
+IR_OPCODE(CompareEqI32, Compare)
+IR_OPCODE(CompareNeI32, Compare)
+IR_OPCODE(CompareGtI32, Compare)
+IR_OPCODE(CompareGeI32, Compare)
+IR_OPCODE(CompareLtI32, Compare)
+IR_OPCODE(CompareLeI32, Compare)
+IR_OPCODE(CompareEqI16, Compare)
+IR_OPCODE(CompareNeI16, Compare)
+IR_OPCODE(CompareGtI16, Compare)
+IR_OPCODE(CompareGeI16, Compare)
+IR_OPCODE(CompareLtI16, Compare)
+IR_OPCODE(CompareLeI16, Compare)
+IR_OPCODE(CompareMaskEqI32, Compare)
+IR_OPCODE(CompareMaskNeI32, Compare)
+IR_OPCODE(CompareMaskGtI32, Compare)
+IR_OPCODE(CompareMaskGeI32, Compare)
+IR_OPCODE(CompareMaskLtI32, Compare)
+IR_OPCODE(CompareMaskLeI32, Compare)
+IR_OPCODE(CompareEqU16, Compare)
+IR_OPCODE(CompareNeU16, Compare)
+IR_OPCODE(CompareGtU16, Compare)
+IR_OPCODE(CompareGeU16, Compare)
+IR_OPCODE(CompareLtU16, Compare)
+IR_OPCODE(CompareLeU16, Compare)
+IR_OPCODE(CompareEqF32, Compare)
+IR_OPCODE(CompareNeF32, Compare)
+IR_OPCODE(CompareGtF32, Compare)
+IR_OPCODE(CompareGeF32, Compare)
+IR_OPCODE(CompareLtF32, Compare)
+IR_OPCODE(CompareLeF32, Compare)
+IR_OPCODE(CompareOrderedF32, Compare)
+IR_OPCODE(CompareUnorderedF32, Compare)
+IR_OPCODE(CompareUnordEqF32, Compare)
+IR_OPCODE(CompareUnordNeF32, Compare)
+IR_OPCODE(CompareUnordGtF32, Compare)
+IR_OPCODE(CompareUnordGeF32, Compare)
+IR_OPCODE(CompareUnordLtF32, Compare)
+IR_OPCODE(CompareUnordLeF32, Compare)
+IR_OPCODE(CompareClassF32, Compare)
+IR_OPCODE(CompareEqF16, Compare)
+IR_OPCODE(CompareNeF16, Compare)
+IR_OPCODE(CompareGtF16, Compare)
+IR_OPCODE(CompareGeF16, Compare)
+IR_OPCODE(CompareLtF16, Compare)
+IR_OPCODE(CompareLeF16, Compare)
+IR_OPCODE(CompareUnordNeF16, Compare)
+IR_OPCODE(CompareMaskEqF16, Compare)
+IR_OPCODE(CompareMaskNeF16, Compare)
+IR_OPCODE(CompareMaskGtF16, Compare)
+IR_OPCODE(CompareMaskGeF16, Compare)
+IR_OPCODE(CompareMaskLtF16, Compare)
+IR_OPCODE(CompareMaskLeF16, Compare)
+IR_OPCODE(CompareMaskUnordNeF16, Compare)
+IR_OPCODE(CompareMaskUnordGeF16, Compare)
+IR_OPCODE(CompareMaskEqF32, Compare)
+IR_OPCODE(CompareMaskNeF32, Compare)
+IR_OPCODE(CompareMaskGtF32, Compare)
+IR_OPCODE(CompareMaskGeF32, Compare)
+IR_OPCODE(CompareMaskLtF32, Compare)
+IR_OPCODE(CompareMaskLeF32, Compare)
+IR_OPCODE(CompareMaskUnordEqF32, Compare)
+IR_OPCODE(CompareMaskUnordNeF32, Compare)
+IR_OPCODE(CompareMaskUnordGtF32, Compare)
+IR_OPCODE(CompareMaskUnordGeF32, Compare)
+IR_OPCODE(CompareMaskUnordLtF32, Compare)
+IR_OPCODE(CompareMaskUnordLeF32, Compare)
+IR_OPCODE(ConvertByteU32ToF32, General)
+IR_OPCODE(ConvertU32ToF32, General)
+IR_OPCODE(ConvertI32ToF32, General)
+IR_OPCODE(ConvertF32ToU32, General)
+IR_OPCODE(ConvertF32ToI32, General)
+IR_OPCODE(ConvertF32ToF16, General)
+IR_OPCODE(ConvertF16ToF32, General)
+IR_OPCODE(ConvertU16ToF16, General)
+IR_OPCODE(ConvertF16ToU16, General)
+IR_OPCODE(ConvertI16ToF16, General)
+IR_OPCODE(ConvertF16ToI16, General)
+IR_OPCODE(ConvertRoundPlusInfF32ToI32, General)
+IR_OPCODE(ConvertFloorF32ToI32, General)
+IR_OPCODE(ConvertI4ToOffsetF32, General)
+IR_OPCODE(LdexpF32, General)
+IR_OPCODE(PackF32ToF16Rtz, General)
+IR_OPCODE(PackSnorm2x16F32, General)
+IR_OPCODE(PackUnorm2x16F32, General)
+IR_OPCODE(PackU16U32, General)
+IR_OPCODE(PackU8F32, General)
+IR_OPCODE(PackB32F16, General)
+IR_OPCODE(PackedMadI16, General)
+IR_OPCODE(PackedMulLoU16, General)
+IR_OPCODE(PackedAddI16, General)
+IR_OPCODE(PackedSubI16, General)
+IR_OPCODE(PackedLshlrevB16, General)
+IR_OPCODE(PackedLshrrevB16, General)
+IR_OPCODE(PackedAshrrevI16, General)
+IR_OPCODE(PackedMaxI16, General)
+IR_OPCODE(PackedMinI16, General)
+IR_OPCODE(PackedMadU16, General)
+IR_OPCODE(PackedAddU16, General)
+IR_OPCODE(PackedSubU16, General)
+IR_OPCODE(PackedMaxU16, General)
+IR_OPCODE(PackedMinU16, General)
+IR_OPCODE(PackedAddF16, General)
+IR_OPCODE(PackedMulF16, General)
+IR_OPCODE(PackedMinF16, General)
+IR_OPCODE(PackedMaxF16, General)
+IR_OPCODE(PackedFmaF16, General)
+IR_OPCODE(AddF16, General)
+IR_OPCODE(SubF16, General)
+IR_OPCODE(MulF16, General)
+IR_OPCODE(MinF16, General)
+IR_OPCODE(MaxF16, General)
+IR_OPCODE(FmaF16, General)
+IR_OPCODE(MadMixF16, General)
+IR_OPCODE(IAddU16, General)
+IR_OPCODE(ISubI16, General)
+IR_OPCODE(IMinI16, General)
+IR_OPCODE(IMaxI16, General)
+IR_OPCODE(UMinU16, General)
+IR_OPCODE(UMaxU16, General)
+IR_OPCODE(RcpF32, General)
+IR_OPCODE(FractF32, General)
+IR_OPCODE(TruncF32, General)
+IR_OPCODE(CeilF32, General)
+IR_OPCODE(RoundEvenF32, General)
+IR_OPCODE(FloorF32, General)
+IR_OPCODE(Exp2F32, General)
+IR_OPCODE(Log2F32, General)
+IR_OPCODE(InverseSqrtF32, General)
+IR_OPCODE(SqrtF32, General)
+IR_OPCODE(RcpF16, General)
+IR_OPCODE(SqrtF16, General)
+IR_OPCODE(InverseSqrtF16, General)
+IR_OPCODE(Log2F16, General)
+IR_OPCODE(Exp2F16, General)
+IR_OPCODE(FloorF16, General)
+IR_OPCODE(CeilF16, General)
+IR_OPCODE(TruncF16, General)
+IR_OPCODE(RoundEvenF16, General)
+IR_OPCODE(SinF32, General)
+IR_OPCODE(CosF32, General)
+IR_OPCODE(CubeIdF32, General)
+IR_OPCODE(CubeScF32, General)
+IR_OPCODE(CubeTcF32, General)
+IR_OPCODE(CubeMaF32, General)
+IR_OPCODE(FAddF32, General)
+IR_OPCODE(FSubF32, General)
+IR_OPCODE(FMulF32, General)
+IR_OPCODE(FMinF32, General)
+IR_OPCODE(FMaxF32, General)
+IR_OPCODE(FMadF32, General)
+IR_OPCODE(Dot2AccF32F16, General)
+IR_OPCODE(FMin3F32, General)
+IR_OPCODE(FMax3F32, General)
+IR_OPCODE(FMed3F32, General)
+IR_OPCODE(Min3F16, General)
+IR_OPCODE(Max3F16, General)
+IR_OPCODE(Med3F16, General)
+IR_OPCODE(LoadSrtDword, General)
+IR_OPCODE(SLoadDword, General)
+IR_OPCODE(SBufferLoadDword, General)
+IR_OPCODE(BufferLoadUbyte, General)
+IR_OPCODE(BufferLoadSbyte, General)
+IR_OPCODE(BufferLoadUshort, General)
+IR_OPCODE(BufferLoadSshort, General)
+IR_OPCODE(BufferLoadDword, General)
+IR_OPCODE(BufferStoreByte, General)
+IR_OPCODE(BufferStoreShort, General)
+IR_OPCODE(BufferStoreDword, General)
+IR_OPCODE(AtomicSwapU32, General)
+IR_OPCODE(AtomicAddU32, General)
+IR_OPCODE(AtomicSubU32, General)
+IR_OPCODE(AtomicSMinI32, General)
+IR_OPCODE(AtomicUMinU32, General)
+IR_OPCODE(AtomicSMaxI32, General)
+IR_OPCODE(AtomicUMaxU32, General)
+IR_OPCODE(AtomicAndU32, General)
+IR_OPCODE(AtomicOrU32, General)
+IR_OPCODE(AtomicXorU32, General)
+IR_OPCODE(AtomicFMinF32, General)
+IR_OPCODE(AtomicFMaxF32, General)
+IR_OPCODE(FlatLoadUbyte, General)
+IR_OPCODE(FlatLoadSbyte, General)
+IR_OPCODE(FlatLoadUshort, General)
+IR_OPCODE(FlatLoadSshort, General)
+IR_OPCODE(FlatLoadDword, General)
+IR_OPCODE(FlatStoreByte, General)
+IR_OPCODE(FlatStoreShort, General)
+IR_OPCODE(FlatStoreDword, General)
+IR_OPCODE(DsReadUbyte, General)
+IR_OPCODE(DsReadSbyte, General)
+IR_OPCODE(DsReadUshort, General)
+IR_OPCODE(DsReadSshort, General)
+IR_OPCODE(DsReadB32, General)
+IR_OPCODE(DsWriteByte, General)
+IR_OPCODE(DsWriteShort, General)
+IR_OPCODE(DsWriteB32, General)
+IR_OPCODE(DsMinF32, General)
+IR_OPCODE(DsMaxF32, General)
+IR_OPCODE(DsSwizzleB32, General)
+IR_OPCODE(DsConsume, General)
+IR_OPCODE(DsAppend, General)
+IR_OPCODE(DsWriteAddtidB32, General)
+IR_OPCODE(DsReadAddtidB32, General)
+IR_OPCODE(ImageGetResinfo, General)
+IR_OPCODE(ImageGetLod, General)
+IR_OPCODE(ImageLoad, General)
+IR_OPCODE(ImageStore, General)
+IR_OPCODE(ImageSample, General)
+IR_OPCODE(ImageGather4, General)
+IR_OPCODE(LoadInputF32, General)
+IR_OPCODE(Export, General)

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

@@ -562,8 +562,7 @@ static bool EvaluateRuntimeSourcesImpl(const Program&
 		const auto  dynamic =
 		    std::find(program.srt.dynamic_sources.begin(), program.srt.dynamic_sources.end(),
 		              request.source) != program.srt.dynamic_sources.end();
-		if (descriptor == nullptr || dynamic ||
-		    !DescriptorSourceResolved(program, request.source)) {
+		if (descriptor == nullptr || dynamic) {
 			if (error != nullptr) {
 				*error = Diagnostic(program, request.use_pc,
 				                    fmt::format("descriptor source {} is {}", request.source,

src/graphics/shader/shader.cpp

@@ -302,7 +302,7 @@ static void vs_check(const HW::VertexShaderInfo& vs, const HW::ShaderRegisters&
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.ieee_mode != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.cu_group_enable != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.require_forward_progress != false);
-		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.lds_configuration != false);
+		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.threadgroup_configuration != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.gs_vgpr_component_count != 3);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.fp16_overflow != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc2.scratch_en != false);
@@ -374,7 +374,7 @@ static void ps_check(const HW::PsStageRegisters& ps, const HW::ShaderRegisters&
 			     ps.rsrc2.extra_lds_size);
 		}
 	}
-	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != false);
+	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != 0);
 	EXIT_NOT_IMPLEMENTED(ps.rsrc2.shared_vgprs != 0);
 
 	if (sh.shader_z_format != 0x00000000 && sh.shader_z_format != 0x00000001 &&

src/kernel/fileSystem.cpp

@@ -276,6 +276,10 @@ static std::filesystem::path ResolvePathIgnoringCase(const std::filesystem::path
 
 	return resolved;
 }
+#else
+static bool HasWindowsForbiddenFilenameCharacter(const std::string& relative_path) {
+	return relative_path.find_first_of("<>:\"|?*") != std::string::npos;
+}
 #endif
 
 std::filesystem::path MountPoints::GetRealFilename(const std::string& mounted_file_name) {
@@ -295,6 +299,10 @@ std::filesystem::path MountPoints::GetRealFilename(const std::string& mounted_fi
 			rel_path = Common::RemoveFirst(rel_path, 1);
 		}
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		if (HasWindowsForbiddenFilenameCharacter(rel_path)) {
+			::printf("FileSystem: Windows-incompatible guest filename: %s\n",
+			         mounted_file_name.c_str());
+		}
 		return p.dir / rel_path;
 #else
 		return ResolvePathIgnoringCase(p.dir / rel_path);
@@ -795,7 +803,7 @@ int64_t KYTY_SYSV_ABI KernelLseek(int d, int64_t offset, int whence) {
 		return KERNEL_ERROR_ESPIPE;
 	}
 
-	file->mutex.Lock();
+	Common::LockGuard lock(file->mutex);
 
 	bool is_invalid = file->f.IsInvalid();
 
@@ -820,8 +828,6 @@ int64_t KYTY_SYSV_ABI KernelLseek(int d, int64_t offset, int whence) {
 
 	EXIT_IF(pos != offset);
 
-	file->mutex.Unlock();
-
 	if (is_invalid) {
 		LOGF("\tfile is invalid\n");
 		return KERNEL_ERROR_EIO;

src/kernel/memoryAddressSpace.inc

@@ -22,6 +22,13 @@ static bool GuestFreeRangeContains(uint64_t base, uint64_t range_size, uint64_t
 	return candidate <= range_end && size <= range_end - candidate;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+static bool CoalescePlaceholders(uint64_t vaddr, uint64_t size) {
+	return size != 0 && VirtualFree(reinterpret_cast<void*>(vaddr), size,
+	                                MEM_RELEASE | MEM_COALESCE_PLACEHOLDERS) != 0;
+}
+#endif
+
 class GuestBackingStore {
 public:
 	enum class FailureReason {
@@ -264,6 +271,19 @@ public:
 			removed.push_back(entry);
 		}
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		// Each unmapped file view becomes a separate Windows placeholder. The address-space owner
+		// records this contiguous guest span as one free range, so normalize fragmented backing
+		// views into one placeholder before publishing it there.
+		if (removed.size() > 1 && !CoalescePlaceholders(vaddr, size)) {
+			for (auto it = removed.rbegin(); it != removed.rend(); ++it) {
+				EXIT_IF(!MapExistingPlaceholderFixed(it->vaddr, it->size, it->backing_offset,
+				                                     it->mode));
+			}
+			return false;
+		}
+#endif
+
 		if (placeholder_preserved != nullptr) {
 			*placeholder_preserved = true;
 		}
@@ -832,6 +852,8 @@ public:
 			EXIT_IF(VirtualFree(reinterpret_cast<void*>(region_vaddr), region_size,
 			                    MEM_RELEASE | MEM_PRESERVE_PLACEHOLDER) == 0);
 		}
+		// ReleaseCommitted also publishes the aggregate span as one logical free range.
+		EXIT_IF(regions.size() > 1 && !CoalescePlaceholders(vaddr, size));
 #else
 		for (const auto& [region_vaddr, region_size]: regions) {
 			RestorePlaceholder(region_vaddr, region_size);
@@ -1252,8 +1274,7 @@ private:
 
 		if (base != it->first || end != it->first + it->second) {
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-			if (VirtualFree(reinterpret_cast<void*>(base), end - base,
-			                MEM_RELEASE | MEM_COALESCE_PLACEHOLDERS) != 0) {
+			if (CoalescePlaceholders(base, end - base)) {
 				m_free.erase(first, last);
 				m_free.emplace(base, end - base);
 			}

src/kernel/pthread.cpp

@@ -1849,8 +1849,9 @@ static int PthreadMutexInitNamed(PthreadMutex* mutex, const PthreadMutexattr* at
 	}
 }
 
-int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr) {
-	return PthreadMutexInitNamed(mutex, attr, "");
+int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr,
+                                   const char* name) {
+	return PthreadMutexInitNamed(mutex, attr, name);
 }
 
 int KYTY_SYSV_ABI PthreadMutexDestroy(PthreadMutex* mutex) {
@@ -2443,8 +2444,9 @@ static int PthreadRwlockInitNamed(PthreadRwlock* rwlock, const PthreadRwlockattr
 	return OK;
 }
 
-int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr) {
-	return PthreadRwlockInitNamed(rwlock, attr, "");
+int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr,
+                                    const char* name) {
+	return PthreadRwlockInitNamed(rwlock, attr, name);
 }
 
 static PthreadRwlockPrivate::Reader* RwlockFindReader(PthreadRwlock rwlock, Pthread thread) {
@@ -2938,8 +2940,9 @@ static int PthreadCondInitNamed(PthreadCond* cond, const PthreadCondattr* attr,
 	}
 }
 
-int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr) {
-	return PthreadCondInitNamed(cond, attr, "");
+int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr,
+                                  const char* name) {
+	return PthreadCondInitNamed(cond, attr, name);
 }
 
 int KYTY_SYSV_ABI PthreadCondSignal(PthreadCond* cond) {
@@ -4417,6 +4420,13 @@ int KYTY_SYSV_ABI pthread_cond_signal(LibKernel::PthreadCond* cond) {
 	return POSIX_PTHREAD_CALL(LibKernel::PthreadCondSignal(cond));
 }
 
+int KYTY_SYSV_ABI pthread_cond_init(LibKernel::PthreadCond*           cond,
+                                    const LibKernel::PthreadCondattr* attr) {
+	PRINT_NAME();
+
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadCondInit(cond, attr, nullptr));
+}
+
 int KYTY_SYSV_ABI pthread_condattr_setclock(LibKernel::PthreadCondattr* attr,
                                             LibKernel::KernelClockid    clock_id) {
 	PRINT_NAME();
@@ -4525,6 +4535,13 @@ int KYTY_SYSV_ABI pthread_rwlock_destroy(LibKernel::PthreadRwlock* rwlock) {
 	return POSIX_PTHREAD_CALL(LibKernel::PthreadRwlockDestroy(rwlock));
 }
 
+int KYTY_SYSV_ABI pthread_rwlock_init(LibKernel::PthreadRwlock*           rwlock,
+                                      const LibKernel::PthreadRwlockattr* attr) {
+	PRINT_NAME();
+
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadRwlockInit(rwlock, attr, nullptr));
+}
+
 int KYTY_SYSV_ABI pthread_key_create(LibKernel::PthreadKey*                   key,
                                      LibKernel::pthread_key_destructor_func_t destructor) {
 	PRINT_NAME();
@@ -4560,7 +4577,7 @@ int KYTY_SYSV_ABI pthread_mutex_init(LibKernel::PthreadMutex*           mutex,
                                      const LibKernel::PthreadMutexattr* attr) {
 	PRINT_NAME();
 
-	return POSIX_PTHREAD_CALL(LibKernel::PthreadMutexInit(mutex, attr));
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadMutexInit(mutex, attr, nullptr));
 }
 
 int KYTY_SYSV_ABI pthread_mutexattr_init(LibKernel::PthreadMutexattr* attr) {

src/kernel/pthread.h

@@ -77,7 +77,8 @@ int KYTY_SYSV_ABI PthreadMutexattrInit(PthreadMutexattr* attr);
 int KYTY_SYSV_ABI PthreadMutexattrDestroy(PthreadMutexattr* attr);
 int KYTY_SYSV_ABI PthreadMutexattrSettype(PthreadMutexattr* attr, int type);
 int KYTY_SYSV_ABI PthreadMutexattrSetprotocol(PthreadMutexattr* attr, int protocol);
-int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr);
+int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr,
+                                   const char* name);
 int KYTY_SYSV_ABI PthreadMutexDestroy(PthreadMutex* mutex);
 int KYTY_SYSV_ABI PthreadMutexLock(PthreadMutex* mutex);
 int KYTY_SYSV_ABI PthreadMutexTrylock(PthreadMutex* mutex);
@@ -159,7 +160,8 @@ int KYTY_SYSV_ABI PthreadAttrSetstackaddr(PthreadAttr* attr, void* addr);
 int KYTY_SYSV_ABI PthreadAttrSetstacksize(PthreadAttr* attr, size_t stack_size);
 
 int KYTY_SYSV_ABI PthreadRwlockDestroy(PthreadRwlock* rwlock);
-int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr);
+int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr,
+                                    const char* name);
 int KYTY_SYSV_ABI PthreadRwlockRdlock(PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI PthreadRwlockTimedrdlock(PthreadRwlock* rwlock, KernelUseconds usec);
 int KYTY_SYSV_ABI PthreadRwlockTimedwrlock(PthreadRwlock* rwlock, KernelUseconds usec);
@@ -177,7 +179,7 @@ int KYTY_SYSV_ABI PthreadCondattrInit(PthreadCondattr* attr);
 int KYTY_SYSV_ABI PthreadCondattrSetclock(PthreadCondattr* attr, KernelClockid clock_id);
 int KYTY_SYSV_ABI PthreadCondBroadcast(PthreadCond* cond);
 int KYTY_SYSV_ABI PthreadCondDestroy(PthreadCond* cond);
-int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr);
+int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr, const char* name);
 int KYTY_SYSV_ABI PthreadCondSignal(PthreadCond* cond);
 int KYTY_SYSV_ABI PthreadCondSignalto(PthreadCond* cond, Pthread thread);
 int KYTY_SYSV_ABI PthreadCondTimedwait(PthreadCond* cond, PthreadMutex* mutex, KernelUseconds usec);
@@ -267,6 +269,8 @@ int KYTY_SYSV_ABI pthread_rwlock_rdlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_unlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_wrlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_destroy(LibKernel::PthreadRwlock* rwlock);
+int KYTY_SYSV_ABI pthread_rwlock_init(LibKernel::PthreadRwlock*           rwlock,
+                                      const LibKernel::PthreadRwlockattr* attr);
 int KYTY_SYSV_ABI pthread_key_create(LibKernel::PthreadKey*                   key,
                                      LibKernel::pthread_key_destructor_func_t destructor);
 int KYTY_SYSV_ABI pthread_key_delete(LibKernel::PthreadKey key);

src/kernel/syncOnAddress.cpp

@@ -0,0 +1,292 @@
+#include "kernel/syncOnAddress.h"
+
+#include "common/threads.h"
+#include "libs/errno.h"
+
+#include <algorithm>
+#include <chrono>
+#include <climits>
+#include <cstdint>
+#include <list>
+#include <memory>
+#include <mutex>
+#include <unordered_map>
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+#include <cerrno>
+#include <linux/futex.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#endif
+
+namespace Libs::LibKernel::SyncOnAddress {
+
+namespace {
+
+constexpr uint32_t SIGNAL_POLL_MICROS = 10000;
+
+using Clock = std::chrono::steady_clock;
+
+template <typename T>
+[[nodiscard]] bool IsValidWaitAddress(const volatile T* address) {
+	return address != nullptr && (reinterpret_cast<uintptr_t>(address) & (alignof(T) - 1u)) == 0;
+}
+
+[[nodiscard]] bool IsValidWakeAddress(const volatile void* address) {
+	return address != nullptr &&
+	       (reinterpret_cast<uintptr_t>(address) & (alignof(uint32_t) - 1u)) == 0;
+}
+
+template <typename T>
+[[nodiscard]] T ReadWord(const volatile T* address) {
+	return __atomic_load_n(address, __ATOMIC_ACQUIRE);
+}
+
+struct WaitDeadline {
+	bool              finite = false;
+	Clock::time_point end {};
+};
+
+[[nodiscard]] WaitDeadline MakeDeadline(const uint32_t* timeout_micros) {
+	if (timeout_micros == nullptr) {
+		return {};
+	}
+	return {true, Clock::now() + std::chrono::microseconds(*timeout_micros)};
+}
+
+[[nodiscard]] uint32_t GetWaitSliceMicros(const WaitDeadline& deadline, bool first_wait) {
+	if (!deadline.finite) {
+		return SIGNAL_POLL_MICROS;
+	}
+
+	const auto now = Clock::now();
+	if (now >= deadline.end) {
+		return first_wait ? 0u : UINT32_MAX;
+	}
+
+	const auto remaining =
+	    std::chrono::duration_cast<std::chrono::microseconds>(deadline.end - now).count();
+	return static_cast<uint32_t>(std::min<int64_t>(remaining, SIGNAL_POLL_MICROS));
+}
+
+void PollSignals(signal_poll_func_t signal_poll) {
+	if (signal_poll != nullptr) {
+		signal_poll();
+	}
+}
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+
+template <typename T>
+int WaitLinux(volatile T* address, T expected, const uint32_t* timeout_micros,
+              signal_poll_func_t signal_poll) {
+	const auto deadline   = MakeDeadline(timeout_micros);
+	bool       first_wait = true;
+
+	for (;;) {
+		if (ReadWord(address) != expected) {
+			return OK;
+		}
+		const auto slice_micros = GetWaitSliceMicros(deadline, first_wait);
+		if (slice_micros == UINT32_MAX) {
+			return ReadWord(address) == expected ? KERNEL_ERROR_ETIMEDOUT : OK;
+		}
+		const timespec timeout = {
+		    .tv_sec  = static_cast<time_t>(slice_micros / 1000000u),
+		    .tv_nsec = static_cast<long>(slice_micros % 1000000u) * 1000L,
+		};
+
+		long result     = 0;
+		int  wait_error = 0;
+		result          = syscall(SYS_futex, const_cast<T*>(address), FUTEX_WAIT_PRIVATE,
+		                          static_cast<uint32_t>(expected), &timeout, nullptr, 0);
+		if (result != 0) {
+			wait_error = errno;
+		}
+		if (result == 0 || wait_error == EAGAIN) {
+			return OK;
+		}
+		if (wait_error != ETIMEDOUT && wait_error != EINTR) {
+			return KERNEL_ERROR_EINVAL;
+		}
+
+		PollSignals(signal_poll);
+		if (deadline.finite && Clock::now() >= deadline.end) {
+			return ReadWord(address) == expected ? KERNEL_ERROR_ETIMEDOUT : OK;
+		}
+		first_wait = false;
+	}
+}
+
+int WakeLinux(volatile void* address, int32_t count) {
+	// The legacy FUTEX_WAKE path may wake one waiter when count is zero.
+	if (count == 0) {
+		return OK;
+	}
+
+	const auto result = syscall(SYS_futex, const_cast<void*>(address), FUTEX_WAKE_PRIVATE, count,
+	                            nullptr, nullptr, 0);
+	return result < 0 ? KERNEL_ERROR_EINVAL : OK;
+}
+
+#endif
+
+struct PortableWaiter {
+	Common::CondVar condition;
+	bool            wake_requested = false;
+};
+
+struct PortableAddressEntry {
+	Common::Mutex              mutex;
+	std::list<PortableWaiter*> waiters;
+};
+
+struct PortableAddressRegistry {
+	std::mutex                                                           mutex;
+	std::unordered_map<uintptr_t, std::shared_ptr<PortableAddressEntry>> entries;
+};
+
+PortableAddressRegistry& GetPortableRegistry() {
+	static PortableAddressRegistry registry;
+	return registry;
+}
+
+std::shared_ptr<PortableAddressEntry> RegisterPortableWaiter(volatile void*  address,
+                                                             PortableWaiter* waiter) {
+	auto&           registry = GetPortableRegistry();
+	std::lock_guard registry_lock(registry.mutex);
+	auto&           entry = registry.entries[reinterpret_cast<uintptr_t>(address)];
+	if (!entry) {
+		entry = std::make_shared<PortableAddressEntry>();
+	}
+	entry->mutex.Lock();
+	entry->waiters.push_back(waiter);
+	return entry;
+}
+
+void UnregisterPortableWaiter(volatile void*                               address,
+                              const std::shared_ptr<PortableAddressEntry>& entry,
+                              PortableWaiter*                              waiter) {
+	entry->waiters.remove(waiter);
+	const bool empty = entry->waiters.empty();
+	entry->mutex.Unlock();
+	if (!empty) {
+		return;
+	}
+
+	auto&           registry = GetPortableRegistry();
+	std::lock_guard registry_lock(registry.mutex);
+	entry->mutex.Lock();
+	const auto it = registry.entries.find(reinterpret_cast<uintptr_t>(address));
+	if (it != registry.entries.end() && it->second == entry && entry->waiters.empty()) {
+		registry.entries.erase(it);
+	}
+	entry->mutex.Unlock();
+}
+
+template <typename T>
+int WaitPortable(volatile T* address, T expected, const uint32_t* timeout_micros,
+                 signal_poll_func_t signal_poll) {
+	PortableWaiter waiter;
+	auto           entry      = RegisterPortableWaiter(address, &waiter);
+	const auto     deadline   = MakeDeadline(timeout_micros);
+	bool           first_wait = true;
+	int            result     = OK;
+
+	while (ReadWord(address) == expected && !waiter.wake_requested) {
+		const auto slice_micros = GetWaitSliceMicros(deadline, first_wait);
+		if (slice_micros == UINT32_MAX) {
+			result = KERNEL_ERROR_ETIMEDOUT;
+			break;
+		}
+		if (slice_micros == 0) {
+			result = KERNEL_ERROR_ETIMEDOUT;
+			break;
+		}
+
+		(void)waiter.condition.WaitFor(&entry->mutex, slice_micros);
+		entry->mutex.Unlock();
+		PollSignals(signal_poll);
+		entry->mutex.Lock();
+
+		if (deadline.finite && Clock::now() >= deadline.end && ReadWord(address) == expected &&
+		    !waiter.wake_requested) {
+			result = KERNEL_ERROR_ETIMEDOUT;
+			break;
+		}
+		first_wait = false;
+	}
+
+	UnregisterPortableWaiter(address, entry, &waiter);
+	return result;
+}
+
+int WakePortable(volatile void* address, int32_t count) {
+	if (count == 0) {
+		return OK;
+	}
+	auto&            registry = GetPortableRegistry();
+	std::unique_lock registry_lock(registry.mutex);
+	const auto       it = registry.entries.find(reinterpret_cast<uintptr_t>(address));
+	if (it == registry.entries.end()) {
+		return OK;
+	}
+	auto entry = it->second;
+	entry->mutex.Lock();
+	registry_lock.unlock();
+
+	int32_t remaining = count;
+	for (auto* waiter: entry->waiters) {
+		if (!waiter->wake_requested) {
+			waiter->wake_requested = true;
+			waiter->condition.Signal();
+			if (remaining != INT_MAX && --remaining == 0) {
+				break;
+			}
+		}
+	}
+	entry->mutex.Unlock();
+	return OK;
+}
+
+template <typename T>
+int WaitImpl(volatile T* address, T expected, const uint32_t* timeout_micros,
+             signal_poll_func_t signal_poll) {
+	if (!IsValidWaitAddress(address)) {
+		return KERNEL_ERROR_EINVAL;
+	}
+
+	int result = OK;
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+	result = WaitLinux(address, expected, timeout_micros, signal_poll);
+#else
+	result = WaitPortable(address, expected, timeout_micros, signal_poll);
+#endif
+	PollSignals(signal_poll);
+	return result;
+}
+
+} // namespace
+
+int Wait32(volatile uint32_t* address, uint32_t expected, const uint32_t* timeout_micros,
+           signal_poll_func_t signal_poll) {
+	return WaitImpl(address, expected, timeout_micros, signal_poll);
+}
+
+int Wait64(volatile uint64_t* address, uint64_t expected, const uint32_t* timeout_micros,
+           signal_poll_func_t signal_poll) {
+	return WaitImpl(address, expected, timeout_micros, signal_poll);
+}
+
+int Wake(volatile void* address, int32_t count) {
+	if (!IsValidWakeAddress(address) || count < 0) {
+		return KERNEL_ERROR_EINVAL;
+	}
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+	return WakeLinux(address, count);
+#endif
+	return WakePortable(address, count);
+}
+
+} // namespace Libs::LibKernel::SyncOnAddress

src/kernel/syncOnAddress.h

@@ -0,0 +1,19 @@
+#ifndef EMULATOR_INCLUDE_EMULATOR_KERNEL_SYNC_ON_ADDRESS_H_
+#define EMULATOR_INCLUDE_EMULATOR_KERNEL_SYNC_ON_ADDRESS_H_
+
+#include "common/abi.h"
+#include "common/common.h"
+
+namespace Libs::LibKernel::SyncOnAddress {
+
+using signal_poll_func_t = void (*)();
+
+int Wait32(volatile uint32_t* address, uint32_t expected, const uint32_t* timeout_micros,
+           signal_poll_func_t signal_poll = nullptr);
+int Wait64(volatile uint64_t* address, uint64_t expected, const uint32_t* timeout_micros,
+           signal_poll_func_t signal_poll = nullptr);
+int Wake(volatile void* address, int32_t count);
+
+} // namespace Libs::LibKernel::SyncOnAddress
+
+#endif /* EMULATOR_INCLUDE_EMULATOR_KERNEL_SYNC_ON_ADDRESS_H_ */

src/kytyGitVersion.h.in

@@ -1,2 +1,13 @@
+#include "cmake_config.h"
+
 #define KYTY_GIT_VERSION "@KYTY_GIT_VERSION@"
 #define KYTY_GIT_HASH "@KYTY_GIT_HASH@"
+
+/* Build provenance, shown in the window title and in crash reports. */
+#if defined(KYTY_OFFICIAL_BUILD)
+#define KYTY_BUILD_LABEL "Official build " KYTY_RELEASE_TAG
+#elif defined(KYTY_FORK_BUILD)
+#define KYTY_BUILD_LABEL "Fork build " KYTY_BUILD_REPOSITORY " " KYTY_GIT_HASH
+#else
+#define KYTY_BUILD_LABEL "Source build " KYTY_GIT_HASH
+#endif

src/launcher/CMakeLists.txt

@@ -16,6 +16,7 @@ file(GLOB launcher_forms
 add_executable(launcher WIN32
 	${launcher_src}
 	${launcher_forms}
+	resources.qrc
 )
 
 

src/launcher/forms/configuration_edit_dialog.ui

@@ -78,6 +78,26 @@
            </property>
           </widget>
          </item>
+         <item row="2" column="0">
+          <widget class="QCheckBox" name="checkBox_fullscreen">
+           <property name="toolTip">
+            <string>Run in borderless desktop fullscreen</string>
+           </property>
+           <property name="text">
+            <string>Fullscreen</string>
+           </property>
+          </widget>
+         </item>
+         <item row="2" column="1">
+          <widget class="QCheckBox" name="checkBox_red_zone_protection">
+           <property name="toolTip">
+            <string>Helps prevent crashes in some games on Windows</string>
+           </property>
+           <property name="text">
+            <string>Windows SysV red zone crash protection (Experimental)</string>
+           </property>
+          </widget>
+         </item>
         </layout>
        </widget>
       </item>

src/launcher/forms/configuration_list_widget.ui

@@ -69,6 +69,31 @@
            </property>
           </widget>
          </item>
+         <item>
+          <widget class="QToolButton" name="input_mapping_button">
+           <property name="minimumSize">
+            <size>
+             <width>36</width>
+             <height>36</height>
+            </size>
+           </property>
+           <property name="toolTip">
+            <string>Edit input mapping</string>
+           </property>
+           <property name="text">
+            <string/>
+           </property>
+           <property name="iconSize">
+            <size>
+             <width>32</width>
+             <height>32</height>
+            </size>
+           </property>
+           <property name="autoRaise">
+            <bool>true</bool>
+           </property>
+          </widget>
+         </item>
          <item>
           <widget class="QToolButton" name="edit_button">
            <property name="minimumSize">

src/launcher/icons/edit-configuration.svg

@@ -0,0 +1,7 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32">
+  <g fill="none" stroke="#e8f1ff" stroke-linecap="round" stroke-linejoin="round" stroke-width="2">
+    <path d="M8 4h11l5 5v7M24 9h-5V4M16 27H8a2 2 0 0 1-2-2V6a2 2 0 0 1 2-2"/>
+    <path d="m15 24 1-4 8-8 4 4-8 8-4 1Z"/>
+    <path d="m22 14 4 4" opacity=".8"/>
+  </g>
+</svg>

src/launcher/icons/global-settings.svg

@@ -0,0 +1,8 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32">
+  <g fill="none" stroke="#e8f1ff" stroke-linecap="round" stroke-linejoin="round" stroke-width="2">
+    <path d="M5 8h3M14 8h13M5 16h13M24 16h3M5 24h6M17 24h10"/>
+    <circle cx="11" cy="8" r="3"/>
+    <circle cx="21" cy="16" r="3"/>
+    <circle cx="14" cy="24" r="3"/>
+  </g>
+</svg>

src/launcher/icons/input-mapping.svg

@@ -0,0 +1,9 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32">
+  <g fill="none" stroke="#e8f1ff" stroke-linecap="round" stroke-linejoin="round" stroke-width="2">
+    <path d="M10 10h12c3.6 0 5.8 2.3 6.6 6.8l.9 5.1c.4 2.4-.8 4.1-2.8 4.1-1.4 0-2.5-.7-3.6-2.3L21.3 21H10.7l-1.8 2.7C7.8 25.3 6.7 26 5.3 26c-2 0-3.2-1.7-2.8-4.1l.9-5.1C4.2 12.3 6.4 10 10 10Z"/>
+    <path d="M8 16h6M11 13v6"/>
+    <circle cx="22" cy="15" r="1" fill="#e8f1ff" stroke="none"/>
+    <circle cx="25" cy="18" r="1" fill="#e8f1ff" stroke="none"/>
+    <path d="M13 7h6M16 4v6" opacity=".8"/>
+  </g>
+</svg>

src/launcher/icons/remove-configuration.svg

@@ -0,0 +1,6 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32">
+  <g fill="none" stroke="#e8f1ff" stroke-linecap="round" stroke-linejoin="round" stroke-width="2">
+    <path d="M6 9h20M12 9V5h8v4M9 9l1 18h12l1-18"/>
+    <path d="M14 14v8M18 14v8" opacity=".8"/>
+  </g>
+</svg>

src/launcher/include/configuration.h

@@ -85,9 +85,10 @@ public:
 	QString    game_comment;
 
 	Resolution             screen_resolution           = Resolution::R1280X720;
+	bool                   fullscreen_enabled          = false;
 	int                    vblank_frequency            = 60;
 	int                    console_language            = DEFAULT_CONSOLE_LANGUAGE;
-	bool                   vulkan_validation_enabled   = true;
+	bool                   vulkan_validation_enabled   = false;
 	bool                   shader_validation_enabled   = true;
 	ShaderOptimizationType shader_optimization_type    = ShaderOptimizationType::Performance;
 	ShaderLogDirection     shader_log_direction        = ShaderLogDirection::Silent;
@@ -98,11 +99,16 @@ public:
 	QString                printf_output_file          = "_kyty.txt";
 	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   renderdoc_enabled           = false;
+#if defined(_WIN32)
+	bool red_zone_protection_enabled = false;
+#endif
+	QStringList host_input_mapping;
 
 	QString elf = QStringLiteral("eboot.bin");
 
 	void CopyEmulatorSettingsFrom(const Configuration& other) {
 		screen_resolution           = other.screen_resolution;
+		fullscreen_enabled          = other.fullscreen_enabled;
 		vblank_frequency            = other.vblank_frequency;
 		console_language            = other.console_language;
 		vulkan_validation_enabled   = other.vulkan_validation_enabled;
@@ -116,6 +122,10 @@ public:
 		printf_output_file          = other.printf_output_file;
 		profiler_direction          = other.profiler_direction;
 		renderdoc_enabled           = other.renderdoc_enabled;
+#if defined(_WIN32)
+		red_zone_protection_enabled = other.red_zone_protection_enabled;
+#endif
+		host_input_mapping = other.host_input_mapping;
 	}
 
 	void CopyFrom(const Configuration& other) {
@@ -138,6 +148,7 @@ public:
 		KYTY_CFG_SET(game_path);
 		KYTY_CFG_SET(custom_settings);
 		KYTY_CFG_SET(screen_resolution);
+		KYTY_CFG_SET(fullscreen_enabled);
 		KYTY_CFG_SET(vblank_frequency);
 		KYTY_CFG_SET(console_language);
 		KYTY_CFG_SET(vulkan_validation_enabled);
@@ -151,6 +162,10 @@ public:
 		KYTY_CFG_SET(printf_output_file);
 		KYTY_CFG_SET(profiler_direction);
 		KYTY_CFG_SET(renderdoc_enabled);
+#if defined(_WIN32)
+		KYTY_CFG_SET(red_zone_protection_enabled);
+#endif
+		s->setValue("host_input_mapping", host_input_mapping);
 		KYTY_CFG_SET(elf);
 	}
 
@@ -160,6 +175,7 @@ public:
 		KYTY_CFG_GET(game_path);
 		KYTY_CFG_GET(custom_settings);
 		KYTY_CFG_GET(screen_resolution);
+		KYTY_CFG_GET(fullscreen_enabled);
 		vblank_frequency = s->value("vblank_frequency", vblank_frequency).toInt();
 		console_language = s->value("console_language", console_language).toInt();
 		if (console_language < 0 || console_language > MAX_CONSOLE_LANGUAGE) {
@@ -176,7 +192,12 @@ public:
 		KYTY_CFG_GET(printf_output_file);
 		KYTY_CFG_GET(profiler_direction);
 		KYTY_CFG_GET(renderdoc_enabled);
-		elf = s->value("elf", elf).toString();
+#if defined(_WIN32)
+		red_zone_protection_enabled =
+		    s->value("red_zone_protection_enabled", red_zone_protection_enabled).toBool();
+#endif
+		host_input_mapping = s->value("host_input_mapping", host_input_mapping).toStringList();
+		elf                = s->value("elf", elf).toString();
 	}
 };
 

src/launcher/include/configurationListWidget.h

@@ -38,7 +38,10 @@ public:
 
 	[[nodiscard]] const QString&     GetSettingsFile() const { return m_settings_file; }
 	[[nodiscard]] const QStringList& GetGameDirectories() const { return m_game_dirs; }
-	[[nodiscard]] bool               CanViewSelectedTrophies() const;
+	[[nodiscard]] const QStringList& GetHostInputMapping() const {
+		return m_global_info.host_input_mapping;
+	}
+	[[nodiscard]] bool CanViewSelectedTrophies() const;
 
 	bool EnsureGameDirectory();
 	void ScanGameDirectory();
@@ -58,6 +61,7 @@ protected slots:
 	void edit_configuration();
 	void delete_configuartion();
 	void edit_global_settings();
+	void edit_input_mapping();
 	void run_configuration();
 	void list_currentItemChanged(QTreeWidgetItem* current, QTreeWidgetItem* previous);
 	void list_itemDoubleClicked(QTreeWidgetItem* witem, int column);

src/launcher/include/inputMappingDialog.h

@@ -0,0 +1,31 @@
+#ifndef LAUNCHER_INCLUDE_INPUT_MAPPING_DIALOG_H_
+#define LAUNCHER_INCLUDE_INPUT_MAPPING_DIALOG_H_
+
+#include <QDialog>
+#include <QStringList>
+
+class QPushButton;
+class QTreeWidget;
+class QTreeWidgetItem;
+class QWidget;
+
+class InputMappingDialog final: public QDialog {
+public:
+	explicit InputMappingDialog(const QStringList& mapping, QWidget* parent = nullptr);
+
+	[[nodiscard]] QStringList Mapping() const;
+
+private:
+	void ChangeBinding();
+	void ClearBinding();
+	void RestoreDefaults();
+	void SetBinding(QTreeWidgetItem* item, const QString& binding);
+	void UpdateButtons();
+
+	QTreeWidget* m_bindings      = nullptr;
+	QPushButton* m_change_button = nullptr;
+	QPushButton* m_clear_button  = nullptr;
+	bool         m_customized    = false;
+};
+
+#endif /* LAUNCHER_INCLUDE_INPUT_MAPPING_DIALOG_H_ */