launcher: quote emulator paths that contain spaces (#244)

* launcher: quote emulator paths that contain spaces

Run the Linux helper script as a bash file argument instead of bash -c,
and build a properly quoted cmd.exe /K command on Windows so install
folders like "Sony Playstation 5" still launch. Fixes #179.

* launcher: fix cmd command quoting

---------

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

@@ -6,9 +6,27 @@ on:
     branches: [ main, master ]
   pull_request:
 
+env:
+  KYTY_BUILD_ORIGIN: ${{ github.repository != 'KytyPS5/KytyPS5' && 'Fork' || github.event_name == 'push' && 'Official' || 'Source' }}
+  KYTY_BUILD_REPOSITORY: ${{ github.repository }}
+
 jobs:
+  metadata:
+    name: Prepare build metadata
+    runs-on: ubuntu-latest
+    outputs:
+      release_tag: ${{ steps.release_tag.outputs.value }}
+
+    steps:
+      - name: Generate release tag
+        id: release_tag
+        shell: bash
+        run: |
+          echo "value=KytyPS5-$(date -u +'%Y-%m-%d')-${GITHUB_SHA::7}" >> "$GITHUB_OUTPUT"
+
   windows:
     name: Build KytyPS5 (Windows)
+    needs: metadata
     runs-on: windows-2022
 
     steps:
@@ -76,6 +94,10 @@ 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 }} ^
             -DCMAKE_C_COMPILER=clang-cl ^
             -DCMAKE_CXX_COMPILER=clang-cl ^
             -DCMAKE_PREFIX_PATH="%Qt6_DIR%"
@@ -83,12 +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
@@ -104,6 +126,7 @@ jobs:
 
   macos:
     name: Build KytyPS5 (macOS)
+    needs: metadata
     runs-on: macos-15
 
     steps:
@@ -150,6 +173,10 @@ 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 }}" \
             -DCMAKE_OSX_ARCHITECTURES=x86_64 \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
@@ -159,14 +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
@@ -213,6 +240,7 @@ jobs:
 
   linux:
     name: Build KytyPS5 (Linux)
+    needs: metadata
     runs-on: ubuntu-24.04
 
     steps:
@@ -273,6 +301,10 @@ 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="${{ needs.metadata.outputs.release_tag }}" \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
             -DCMAKE_PREFIX_PATH="$Qt6_DIR" 2>&1 | tee _Build/configure.log
@@ -297,14 +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
@@ -341,46 +373,77 @@ jobs:
 
   release:
     name: Release KytyPS5
-    if: github.event_name == 'push' && github.repository == 'KytyPS5/KytyPS5'
-    needs: [windows, macos, linux]
+    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:
       contents: write
+    env:
+      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: Set release name
-        shell: bash
-        run: |
-          echo "RELEASE_NAME=KytyPS5-$(date -u +'%Y-%m-%d')-${GITHUB_SHA::7}" >> "$GITHUB_ENV"
+      - 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: Package builds
+      - 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" .
 
@@ -389,11 +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")
@@ -28,6 +34,29 @@ include(CTest)
 
 option(KYTY_ENABLE_CLANG_TIDY "Run clang-tidy checks during builds" OFF)
 
+set(KYTY_BUILD_ORIGIN "Source" CACHE STRING "Build provenance shown in diagnostics")
+set_property(CACHE KYTY_BUILD_ORIGIN PROPERTY STRINGS Source Official Fork)
+if(NOT KYTY_BUILD_ORIGIN STREQUAL "Source" AND NOT KYTY_BUILD_ORIGIN STREQUAL "Official" AND
+   NOT KYTY_BUILD_ORIGIN STREQUAL "Fork")
+	message(FATAL_ERROR "KYTY_BUILD_ORIGIN must be Source, Official, or Fork")
+endif()
+
+set(KYTY_RELEASE_TAG "" CACHE STRING "Release tag embedded in official builds")
+set(KYTY_BUILD_REPOSITORY "" CACHE STRING "GitHub repository embedded in fork builds")
+set(KYTY_OFFICIAL_BUILD OFF)
+set(KYTY_FORK_BUILD OFF)
+if(KYTY_BUILD_ORIGIN STREQUAL "Official")
+	if(KYTY_RELEASE_TAG STREQUAL "")
+		message(FATAL_ERROR "Official builds require KYTY_RELEASE_TAG")
+	endif()
+	set(KYTY_OFFICIAL_BUILD ON)
+elseif(KYTY_BUILD_ORIGIN STREQUAL "Fork")
+	if(KYTY_BUILD_REPOSITORY STREQUAL "")
+		message(FATAL_ERROR "Fork builds require KYTY_BUILD_REPOSITORY")
+	endif()
+	set(KYTY_FORK_BUILD ON)
+endif()
+
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 
 include(TestBigEndian)
@@ -117,16 +146,20 @@ configure_file(
 
 find_package(Git)
 
-add_custom_target( KytyGitVersion
-    COMMAND ${CMAKE_COMMAND}
-	-D INPUT_FILE=${KYTY_SOURCE_DIR}/kytyGitVersion.h.in
-	-D OUTPUT_FILE=${PROJECT_BINARY_DIR}/kytyGitVersion.h
-	-D GIT_EXECUTABLE=${GIT_EXECUTABLE}
-	-D GIT_WORKING_DIRECTORY=${CMAKE_CURRENT_SOURCE_DIR}
-	-P ${KYTY_SOURCE_DIR}/generate_version.cmake
+add_custom_target(KytyGitVersion
+	COMMAND ${CMAKE_COMMAND}
+		-D INPUT_FILE=${KYTY_SOURCE_DIR}/kytyGitVersion.h.in
+		-D OUTPUT_FILE=${PROJECT_BINARY_DIR}/kytyGitVersion.h
+		-D GIT_EXECUTABLE=${GIT_EXECUTABLE}
+		-D GIT_WORKING_DIRECTORY=${CMAKE_CURRENT_SOURCE_DIR}
+		-P ${KYTY_SOURCE_DIR}/generate_version.cmake
+	BYPRODUCTS ${PROJECT_BINARY_DIR}/kytyGitVersion.h
 	COMMENT "Generate kytyGitVersion.h"
 )
 
+add_library(kyty_git_version INTERFACE)
+add_dependencies(kyty_git_version KytyGitVersion)
+
 if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 12.0.0)
 	list(APPEND KYTY_IWYU
 		kyty_emulator
@@ -259,7 +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)
+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)
@@ -345,6 +378,7 @@ add_executable(page_manager_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/PageManagerTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/pageManager.cpp"
 )
+target_link_libraries(page_manager_tests common)
 target_include_directories(page_manager_tests PRIVATE ${inc_headers})
 
 add_executable(bit_array_tests EXCLUDE_FROM_ALL
@@ -352,6 +386,13 @@ 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/imeCommon.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"
@@ -394,13 +435,6 @@ 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"
@@ -417,6 +451,13 @@ 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"
 )
@@ -459,6 +500,7 @@ 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>)
@@ -468,18 +510,22 @@ 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)
 
@@ -505,6 +551,7 @@ if(BUILD_TESTING)
 	endif()
 
 	add_custom_target(kyty_tests DEPENDS
+		ime_dialog_tests
 		shader_cfg_tests
 		scalar_provenance_tests
 		image_page_table_tests
@@ -514,8 +561,8 @@ 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
 	)
@@ -563,8 +610,6 @@ elseif(WIN32 OR LINUX)
 	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${KYTY_LD_OPTIONS} -Wl,-Map=${KYTY_EMULATOR_MAP_LINK_PATH}")
 endif()
 
-add_dependencies(kyty_emulator KytyGitVersion)
-
 if(APPLE)
 	# The emulator writes x86-64 trampolines/PLT stubs into memory and executes them, and
 	# runs guest code under Rosetta. That requires the JIT / unsigned-executable-memory

README.md

@@ -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,
@@ -52,7 +55,7 @@ graphical glitches, low compatibility, and poor performance.
   <tr>
     <td align="center">
       <strong>Neptunia ReVerse</strong><br>
-      <img src="docs/screenshots/ps5-04.png" width="300" alt="Minecraft Legends running in KytyPS5">
+      <img src="docs/screenshots/ps5-04.png" width="300" alt="Neptunia ReVerse running in KytyPS5">
     </td>
     <td align="center">
       <strong>SILENT HILL: The Short Message</strong><br>
@@ -62,11 +65,11 @@ graphical glitches, low compatibility, and poor performance.
   <tr>
     <td align="center">
       <strong>Hellboy</strong><br>
-      <img src="docs/screenshots/ps5-02.png" width="300" alt="Disgaea 6 running in KytyPS5">
+      <img src="docs/screenshots/ps5-02.png" width="300" alt="Hellboy running in KytyPS5">
     </td>
     <td align="center">
       <strong>Paleo Pines</strong><br>
-      <img src="docs/screenshots/ps5-06.png" width="300" alt="Dreaming Sarah running in KytyPS5">
+      <img src="docs/screenshots/ps5-06.png" width="300" alt="Paleo Pines running in KytyPS5">
     </td>
   </tr>
 </table>

src/cmake_config.h.in

@@ -1,4 +1,8 @@
 #define KYTY_VERSION "@KYTY_VERSION@"
+#define KYTY_BUILD_REPOSITORY "@KYTY_BUILD_REPOSITORY@"
+#define KYTY_RELEASE_TAG "@KYTY_RELEASE_TAG@"
+#cmakedefine KYTY_OFFICIAL_BUILD
+#cmakedefine KYTY_FORK_BUILD
 #define KYTY_ENDIAN @KYTY_ENDIAN@
 #define KYTY_PLATFORM @KYTY_PLATFORM@
 #define KYTY_COMPILER KYTY_COMPILER_@KYTY_COMPILER@

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)));
@@ -34,7 +35,7 @@ static std::string BuildFatalReport(const char* title, std::string_view text, co
 
 static int DbgReport(const char* title, std::string_view text, const char* file, int line) {
 	Log::WriteFatal(BuildFatalReport(title, text, file, line));
-	SubsystemsListSingleton::Instance()->ShutdownAll();
+	Subsystems::EmergencyShutdownActive();
 	return 1;
 }
 

src/common/commonSubsystem.cpp

@@ -1,21 +0,0 @@
-#include "common/commonSubsystem.h"
-
-#include "common/byteBuffer.h" // IWYU pragma: associated
-#include "common/common.h"     // IWYU pragma: associated
-#include "common/debug.h"      // IWYU pragma: associated
-#include "common/hash.h"       // IWYU pragma: associated
-#include "common/magicEnum.h"  // IWYU pragma: associated
-#include "common/singleton.h"  // IWYU pragma: associated
-#include "common/virtualMemory.h"
-
-namespace Common {
-
-KYTY_SUBSYSTEM_INIT(Common) {
-	VirtualMemory::Init();
-}
-
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Common) {}
-
-KYTY_SUBSYSTEM_DESTROY(Common) {}
-
-} // namespace Common

src/common/commonSubsystem.h

@@ -1,12 +0,0 @@
-#ifndef KYTY_COMMON_COMMON_SUBSYSTEM_H_
-#define KYTY_COMMON_COMMON_SUBSYSTEM_H_
-
-#include "common/subsystems.h"
-
-namespace Common {
-
-KYTY_SUBSYSTEM_DEFINE(Common);
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_COMMON_SUBSYSTEM_H_ */

src/common/emulatorConfig.cpp

@@ -9,15 +9,15 @@ namespace Config {
 
 static std::unique_ptr<ConfigOptions> g_config;
 
-KYTY_SUBSYSTEM_INIT(Config) {
+void Initialize() {
 	EXIT_IF(g_config != nullptr);
 
 	g_config = std::make_unique<ConfigOptions>();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Config) {}
-
-KYTY_SUBSYSTEM_DESTROY(Config) {}
+void Shutdown() {
+	g_config.reset();
+}
 
 void Load(const ConfigOptions& cfg) {
 	EXIT_IF(g_config == nullptr);
@@ -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

@@ -2,13 +2,21 @@
 #define KYTY_COMMON_EMULATOR_CONFIG_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <filesystem>
+#include <string>
+#include <vector>
 
 namespace Config {
 
-KYTY_SUBSYSTEM_DEFINE(Config);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name       = "Config";
+	static constexpr auto        initialize = Config::Initialize;
+	static constexpr auto        shutdown   = Config::Shutdown;
+};
 
 enum class ShaderOptimizationType { None, Size, Performance };
 
@@ -18,12 +26,15 @@ 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 +51,17 @@ 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 +85,11 @@ bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled();
+#endif
+
+const Keymap& GetKeymap();
 
 } // namespace Config
 

src/common/logging/log.cpp

@@ -149,7 +149,7 @@ void WriteFatal(fmt::text_style style, std::string_view text) {
 	Flush();
 }
 
-KYTY_SUBSYSTEM_INIT(Log) {
+void Initialize() {
 	g_initialized = true;
 	switch (Config::GetPrintfDirection()) {
 		case Config::OutputDirection::Silent: g_direction = Direction::Silent; break;
@@ -161,13 +161,7 @@ KYTY_SUBSYSTEM_INIT(Log) {
 	SetupLogger();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Log) {
-	Flush();
-	std::lock_guard lock(g_logger_mutex);
-	g_logger.reset();
-}
-
-KYTY_SUBSYSTEM_DESTROY(Log) {
+void Shutdown() {
 	Flush();
 	std::lock_guard lock(g_logger_mutex);
 	g_logger.reset();

src/common/logging/log.h

@@ -2,7 +2,6 @@
 #define KYTY_COMMON_LOGGING_LOG_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <fmt/color.h>
 #include <fmt/printf.h>
@@ -10,7 +9,15 @@
 
 namespace Log {
 
-KYTY_SUBSYSTEM_DEFINE(Log);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name               = "Log";
+	static constexpr auto        initialize         = Log::Initialize;
+	static constexpr auto        shutdown           = Log::Shutdown;
+	static constexpr auto        emergency_shutdown = Log::Shutdown;
+};
 
 enum class Direction { Silent, Console, File };
 

src/common/platform/sysWindowsDbg.cpp

@@ -15,56 +15,8 @@
 #include "common/assert.h"
 #include "common/platform/sysDbg.h"
 
-struct FrameS {
-	struct FrameS* next;
-	void*          ret_addr;
-};
-
-int WalkStack(int z_stack_depth, void** z_stack_trace) {
-	CONTEXT context;
-	//	KNONVOLATILE_CONTEXT_POINTERS NvContext;
-	PRUNTIME_FUNCTION runtime_function  = nullptr;
-	PVOID             handler_data      = nullptr;
-	ULONG64           establisher_frame = 0;
-	ULONG64           image_base        = 0;
-
-	RtlCaptureContext(&context);
-
-	int frame = 0;
-
-	while (true) {
-		if (frame >= z_stack_depth) {
-			break;
-		}
-
-		z_stack_trace[frame] = reinterpret_cast<void*>(context.Rip);
-
-		frame++;
-
-		runtime_function = RtlLookupFunctionEntry(context.Rip, &image_base, nullptr);
-
-		if (runtime_function == nullptr) {
-			break;
-		}
-
-		// RtlZeroMemory(&NvContext, sizeof(KNONVOLATILE_CONTEXT_POINTERS));
-		RtlVirtualUnwind(0, image_base, context.Rip, runtime_function, &context, &handler_data,
-		                 &establisher_frame, nullptr /*&NvContext*/);
-
-		if (context.Rip == 0u) {
-			break;
-		}
-	}
-
-	return frame;
-}
-
 void SysStackWalk(void** stack, int* depth) {
-	int n  = WalkStack(*depth, stack);
-	*depth = n;
-	//	unwind_info_t info = {stack, 0, *depth};
-	//	 _Unwind_Backtrace(&trace_fcn, &info);
-	//	 *depth = info.depth;
+	*depth = static_cast<int>(CaptureStackBackTrace(0, static_cast<DWORD>(*depth), stack, nullptr));
 }
 
 void SysStackUsagePrint(sys_dbg_stack_info_t& stack) {

src/common/profiler.cpp

@@ -1,7 +1,6 @@
 #include "common/profiler.h"
 
 #include "common/emulatorConfig.h"
-#include "common/subsystems.h"
 
 #include <algorithm>
 #include <common/TracyProtocol.hpp>
@@ -55,13 +54,7 @@ void SetThreadName(const char* name) {
 	}
 }
 
-void Close() {
-	if (tracy::ProfilerAvailable()) {
-		tracy::ShutdownProfiler();
-	}
-}
-
-KYTY_SUBSYSTEM_INIT(Profiler) {
+void Initialize() {
 	switch (Config::GetProfilerDirection()) {
 		case Config::ProfilerDirection::Network:
 			if (!tracy::ProfilerAvailable()) {
@@ -78,12 +71,10 @@ KYTY_SUBSYSTEM_INIT(Profiler) {
 	}
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Profiler) {
-	Close();
-}
-
-KYTY_SUBSYSTEM_DESTROY(Profiler) {
-	Close();
+void Shutdown() {
+	if (tracy::ProfilerAvailable()) {
+		tracy::ShutdownProfiler();
+	}
 }
 
 } // namespace Profiler

src/common/profiler.h

@@ -2,7 +2,6 @@
 #define KYTY_COMMON_PROFILER_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <cstdint>
 #include <optional>
@@ -42,7 +41,15 @@ private:
 void EndBlock();
 void SetThreadName(const char* name);
 
-KYTY_SUBSYSTEM_DEFINE(Profiler);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name               = "Profiler";
+	static constexpr auto        initialize         = Profiler::Initialize;
+	static constexpr auto        shutdown           = Profiler::Shutdown;
+	static constexpr auto        emergency_shutdown = Profiler::Shutdown;
+};
 
 } // namespace Profiler
 

src/common/subsystems.cpp

@@ -1,232 +1,43 @@
 #include "common/subsystems.h"
 
-#include "common/assert.h"
-
-#include <cstdarg>
-#include <cstdio>
-#include <cstring>
-#include <memory>
-#include <string>
-
 namespace Common {
 
-class SubsystemPrivate {
-public:
-	SubsystemPrivate()          = default;
-	virtual ~SubsystemPrivate() = default;
+static Subsystems* g_active_subsystems = nullptr;
 
-	KYTY_CLASS_NO_COPY(SubsystemPrivate);
+Subsystems::Subsystems(bool print): m_print(print) {
+	m_active.reserve(16);
+	g_active_subsystems = this;
+}
 
-	bool        failed {false};
-	std::string fail_msg;
-};
-
-class SubsystemsListPrivate {
-public:
-	explicit SubsystemsListPrivate(SubsystemsList& p): parent(p) {}
-
-	virtual ~SubsystemsListPrivate() = default;
-
-	void SetArgs(int argc, char** argv) {
-		this->m_argc = argc;
-		this->m_argv = argv;
+Subsystems::~Subsystems() {
+	Destroy();
+	if (g_active_subsystems == this) {
+		g_active_subsystems = nullptr;
 	}
+}
 
-	void Add(Subsystem* s, std::initializer_list<Subsystem*> deps) {
-		EXIT_IF(!s);
-
-		const char* name = s->Id();
-
-		EXIT_IF(FindByName(name) != nullptr);
-
-		auto nl = std::make_unique<SubsListStruct>();
-
-		nl->s         = s;
-		nl->name      = name;
-		nl->deps      = nullptr;
-		nl->next      = std::move(list);
-		nl->prev_init = nullptr;
-
-		auto* node = nl.get();
-		list       = std::move(nl);
-
-		for (auto* dep: deps) {
-			const char* str = dep->Id();
-
-			auto l      = std::make_unique<DepsListStruct>();
-			l->dep_name = str;
-			l->next     = std::move(node->deps);
-
-			node->deps = std::move(l);
+void Subsystems::Destroy() {
+	for (auto it = m_active.rbegin(); it != m_active.rend(); ++it) {
+		if (it->shutdown != nullptr) {
+			it->shutdown();
 		}
-
-		node->initialized = false;
 	}
+	m_active.clear();
+}
 
-	bool InitAll(bool print_msg) {
-		while (SubsListStruct* n = FindNextToInitialize()) {
-			n->s->Init(&parent);
-
-			if (n->s->m_p->failed) {
-				fail_msg  = n->s->m_p->fail_msg.c_str();
-				fail_name = n->name;
-				return false;
-			}
-
-			if (print_msg) {
-				printf("Initialized: %s\n", n->name);
-			}
-
-			n->initialized = true;
-
-			SubsListStruct* last = last_init;
-			last_init            = n;
-			n->prev_init         = last;
+void Subsystems::EmergencyShutdown() {
+	for (auto it = m_active.rbegin(); it != m_active.rend(); ++it) {
+		if (it->emergency_shutdown != nullptr) {
+			it->emergency_shutdown();
 		}
-
-		return true;
 	}
+	m_active.clear();
+}
 
-	void DestroyAll(bool print_msg) {
-		for (SubsListStruct* n = last_init; n != nullptr; n = n->prev_init) {
-			n->s->Destroy(&parent);
-			n->initialized = false;
-
-			if (print_msg) {
-				printf("Destroyed: %s\n", n->name);
-			}
-		}
-
-		last_init = nullptr;
+void Subsystems::EmergencyShutdownActive() {
+	if (g_active_subsystems != nullptr) {
+		g_active_subsystems->EmergencyShutdown();
 	}
-
-	void ShutdownAll() {
-		for (SubsListStruct* n = last_init; n != nullptr; n = n->prev_init) {
-			n->s->UnexpectedShutdown(&parent);
-			n->initialized = false;
-		}
-
-		last_init = nullptr;
-	}
-
-	struct DepsListStruct {
-		const char*                     dep_name;
-		std::unique_ptr<DepsListStruct> next;
-	};
-
-	struct SubsListStruct {
-		Subsystem*                      s;
-		const char*                     name;
-		std::unique_ptr<DepsListStruct> deps;
-		std::unique_ptr<SubsListStruct> next;
-		SubsListStruct*                 prev_init;
-		bool                            initialized;
-	};
-
-	[[nodiscard]] SubsListStruct* FindByName(const char* name) const {
-		for (SubsListStruct* n = list.get(); n != nullptr; n = n->next.get()) {
-			if (std::strcmp(n->name, name) == 0) {
-				return n;
-			}
-		}
-
-		return nullptr;
-	}
-
-	[[nodiscard]] SubsListStruct* FindNextToInitialize() const {
-		for (SubsListStruct* n = list.get(); n != nullptr; n = n->next.get()) {
-			if (n->initialized) {
-				continue;
-			}
-
-			DepsListStruct* d = n->deps.get();
-			for (; d != nullptr; d = d->next.get()) {
-				SubsListStruct* s = FindByName(d->dep_name);
-				if ((s == nullptr) || !s->initialized) {
-					break;
-				}
-			}
-
-			if (d == nullptr) {
-				return n;
-			}
-		}
-
-		return nullptr;
-	}
-
-	KYTY_CLASS_NO_COPY(SubsystemsListPrivate);
-
-	std::unique_ptr<SubsListStruct> list;
-	SubsListStruct*                 last_init = nullptr;
-	int                             m_argc    = 0;
-	char**                          m_argv    = nullptr;
-	const char*                     fail_msg  = nullptr;
-	const char*                     fail_name = nullptr;
-	SubsystemsList&                 parent;
-};
-
-SubsystemsList::SubsystemsList(): m_p(std::make_unique<SubsystemsListPrivate>(*this)) {}
-
-SubsystemsList::~SubsystemsList() = default;
-
-void SubsystemsList::Add(Subsystem* s, std::initializer_list<Subsystem*> deps) {
-	m_p->Add(s, deps);
-}
-
-bool SubsystemsList::InitAll(bool print_msg) {
-	return m_p->InitAll(print_msg);
-}
-
-void SubsystemsList::DestroyAll(bool print_msg) {
-	m_p->DestroyAll(print_msg);
-}
-
-int* SubsystemsList::GetArgc() {
-	return &m_p->m_argc;
-}
-
-char** SubsystemsList::GetArgv() {
-	return m_p->m_argv;
-}
-
-Subsystem::Subsystem(): m_p(std::make_unique<SubsystemPrivate>()) {}
-
-Subsystem::~Subsystem() = default;
-
-void Subsystem::Fail(const char* format, ...) {
-	va_list args {};
-	va_start(args, format);
-
-	va_list args_copy {};
-	va_copy(args_copy, args);
-	int len = std::vsnprintf(nullptr, 0, format, args_copy);
-	va_end(args_copy);
-
-	if (len > 0) {
-		std::string msg(static_cast<size_t>(len) + 1, '\0');
-		std::vsnprintf(msg.data(), msg.size(), format, args);
-		m_p->fail_msg = msg.c_str();
-		m_p->failed   = true;
-	}
-
-	va_end(args);
-}
-
-const char* SubsystemsList::GetFailName() const {
-	return m_p->fail_name;
-}
-
-const char* SubsystemsList::GetFailMsg() const {
-	return m_p->fail_msg;
-}
-
-void SubsystemsList::SetArgs(int argc, char* argv[]) {
-	m_p->SetArgs(argc, argv);
-}
-
-void SubsystemsList::ShutdownAll() {
-	m_p->ShutdownAll();
 }
 
 } // namespace Common

src/common/subsystems.h

@@ -2,90 +2,61 @@
 #define KYTY_COMMON_SUBSYSTEMS_H_
 
 #include "common/common.h"
-#include "common/singleton.h"
 
-#include <initializer_list>
-#include <memory>
+#include <cstdio>
+#include <vector>
 
 namespace Common {
 
-class Subsystem;
-class SubsystemsListPrivate;
-class SubsystemPrivate;
-
-class SubsystemsList {
+class Subsystems {
 public:
-	SubsystemsList();
-	virtual ~SubsystemsList();
-	void SetArgs(int argc, char* argv[]);
+	using Callback = void (*)();
 
-	// void Add(Subsystem* s, const char* name, ...);
-	void Add(Subsystem* s, std::initializer_list<Subsystem*> deps);
+	explicit Subsystems(bool print = false);
+	~Subsystems();
 
-	bool InitAll(bool print_msg = false);
-	void DestroyAll(bool print_msg = false);
+	template <typename Lifecycle>
+	void Initialize() {
+		Lifecycle::initialize();
+		m_active.push_back({ShutdownCallback<Lifecycle>(), EmergencyCallback<Lifecycle>()});
+		if (m_print) {
+			std::printf("Initialized: %s\n", Lifecycle::name);
+		}
+	}
 
-	int*   GetArgc();
-	char** GetArgv();
+	void Destroy();
+	void EmergencyShutdown();
 
-	[[nodiscard]] const char* GetFailName() const;
-	[[nodiscard]] const char* GetFailMsg() const;
+	static void EmergencyShutdownActive();
 
-	void ShutdownAll();
-
-	static SubsystemsList* Instance() { return Common::Singleton<SubsystemsList>::Instance(); }
-
-	KYTY_CLASS_NO_COPY(SubsystemsList);
+	KYTY_CLASS_NO_COPY(Subsystems);
 
 private:
-	std::unique_ptr<SubsystemsListPrivate> m_p;
+	template <typename Lifecycle>
+	static consteval Callback ShutdownCallback() {
+		if constexpr (requires { Lifecycle::shutdown; }) {
+			return Lifecycle::shutdown;
+		}
+		return nullptr;
+	}
+
+	template <typename Lifecycle>
+	static consteval Callback EmergencyCallback() {
+		if constexpr (requires { Lifecycle::emergency_shutdown; }) {
+			return Lifecycle::emergency_shutdown;
+		}
+		return nullptr;
+	}
+
+	struct Entry {
+		Callback shutdown;
+		Callback emergency_shutdown;
+	};
+
+	std::vector<Entry> m_active;
+	bool               m_print;
 };
 
-using SubsystemsListSingleton = Common::Singleton<SubsystemsList>;
-
-class Subsystem {
-public:
-	Subsystem();
-	virtual ~Subsystem();
-
-	virtual const char* Id()                                       = 0;
-	virtual void        Init(SubsystemsList* parent)               = 0;
-	virtual void        Destroy(SubsystemsList* parent)            = 0;
-	virtual void        UnexpectedShutdown(SubsystemsList* parent) = 0;
-
-	friend class SubsystemsListPrivate;
-
-	KYTY_CLASS_NO_COPY(Subsystem);
-
-protected:
-	void Fail(const char* format, ...) KYTY_FORMAT_PRINTF(2, 3);
-
-private:
-	std::unique_ptr<SubsystemPrivate> m_p;
-};
-
-#define KYTY_SUBSYSTEM_DEFINE(s)                                                                   \
-	class s##Subsystem: public Common::Subsystem {                                                 \
-	public:                                                                                        \
-		static Subsystem* Instance() { return Common::Singleton<s##Subsystem>::Instance(); }       \
-		const char*       Id() { return #s; }                                                      \
-		void              Init(Common::SubsystemsList* parent);                                    \
-		void              Destroy(Common::SubsystemsList* parent);                                 \
-		void              UnexpectedShutdown(Common::SubsystemsList* parent);                      \
-	};                                                                                             \
-	typedef Common::Singleton<s##Subsystem> s##SubsystemSingleton;
-
-#define KYTY_SUBSYSTEM_INIT(s)                                                                     \
-	void s##Subsystem::Init([[maybe_unused]] Common::SubsystemsList* parent)
-#define KYTY_SUBSYSTEM_DESTROY(s)                                                                  \
-	void s##Subsystem::Destroy([[maybe_unused]] Common::SubsystemsList* parent)
-#define KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(s)                                                      \
-	void s##Subsystem::UnexpectedShutdown([[maybe_unused]] Common::SubsystemsList* parent)
-
-// #define KYTY_SUBSYSTEM_ADD(list, s, ...) list.push_back(s##SubsystemSingleton::Instance(), #s,
-// __VA_ARGS__); #define KYTY_SUBSYSTEM_ADD2(list, ...) list.Add2(s##SubsystemSingleton::Instance(),
-// __VA_ARGS__);
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_SUBSYSTEMS_H_ */

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 &&
@@ -248,15 +252,11 @@ static thread_id_t      g_main_thread;
 static int              g_main_thread_int;
 static std::atomic<int> g_thread_counter = 0;
 
-KYTY_SUBSYSTEM_INIT(Threads) {
+void InitializeThreads() {
 	g_main_thread     = std::this_thread::get_id();
 	g_main_thread_int = Thread::GetThreadIdUnique();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Threads) {}
-
-KYTY_SUBSYSTEM_DESTROY(Threads) {}
-
 Thread::Thread(thread_func_t func, void* arg)
     : m_thread(std::make_unique<ThreadPrivate>(func, arg)) {
 	while (!m_thread->started) {

src/common/threads.h

@@ -2,14 +2,13 @@
 #define KYTY_COMMON_THREADS_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <memory>
 #include <string>
 
 namespace Common {
 
-KYTY_SUBSYSTEM_DEFINE(Threads);
+void InitializeThreads();
 
 using thread_func_t    = void (*)(void*);
 using wait_poll_func_t = void (*)();

src/emulator.cpp

@@ -2,7 +2,6 @@
 
 #include "common/abi.h"
 #include "common/assert.h"
-#include "common/commonSubsystem.h"
 #include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
@@ -43,7 +42,7 @@ static void KytyClose() {
 
 	LOGF("done!\n");
 
-	Common::SubsystemsListSingleton::Instance()->ShutdownAll();
+	Common::Subsystems::EmergencyShutdownActive();
 }
 
 static void MountOrCreateDir(const std::filesystem::path& dir, const std::string& point) {
@@ -105,30 +104,13 @@ static void ClearDebugTextureFolder() {
 	}
 }
 
-static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path& param_json) {
+static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path& param_json,
+                 Common::Subsystems& subsystems) {
 	EXIT_IF(!Common::Thread::IsMainThread());
 
-	auto* slist = Common::SubsystemsList::Instance();
-
-	auto* audio       = Libs::Audio::AudioSubsystem::Instance();
-	auto* config      = Config::ConfigSubsystem::Instance();
-	auto* controller  = Libs::Controller::ControllerSubsystem::Instance();
-	auto* core        = Common::CommonSubsystem::Instance();
-	auto* file_system = Libs::LibKernel::FileSystem::FileSystemSubsystem::Instance();
-	auto* graphics    = Libs::Graphics::GraphicsSubsystem::Instance();
-	auto* log         = Log::LogSubsystem::Instance();
-	auto* memory      = Libs::LibKernel::Memory::MemorySubsystem::Instance();
-	auto* network     = Libs::Network::NetworkSubsystem::Instance();
-	auto* profiler    = Profiler::ProfilerSubsystem::Instance();
-	auto* pthread     = Libs::LibKernel::PthreadSubsystem::Instance();
-	auto* timer       = Loader::Timer::TimerSubsystem::Instance();
-
-	slist->Add(config, {core});
-	slist->InitAll(true);
-
+	subsystems.Initialize<Config::Lifecycle>();
 	Config::Load(cfg);
-	slist->Add(log, {core, config});
-	slist->InitAll(true);
+	subsystems.Initialize<Log::Lifecycle>();
 
 	if (Common::File::IsFileExisting(param_json)) {
 		Loader::SystemContentLoadParamSfo(param_json);
@@ -138,17 +120,16 @@ static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path&
 		}
 	}
 
-	slist->Add(audio, {core, log, pthread, memory});
-	slist->Add(controller, {core, log, config});
-	slist->Add(file_system, {core, log, pthread});
-	slist->Add(graphics, {core, log, pthread, memory, config, profiler, controller});
-	slist->Add(memory, {core, log});
-	slist->Add(network, {core, log, pthread});
-	slist->Add(profiler, {core, config});
-	slist->Add(pthread, {core, log, timer});
-	slist->Add(timer, {core, log});
-
-	slist->InitAll(true);
+	// Initialization order is explicit; destruction is automatic and reversed.
+	subsystems.Initialize<Loader::Timer::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::PthreadLifecycle>();
+	subsystems.Initialize<Profiler::Lifecycle>();
+	subsystems.Initialize<Libs::Network::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::Memory::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::FileSystem::Lifecycle>();
+	subsystems.Initialize<Libs::Controller::Lifecycle>();
+	subsystems.Initialize<Libs::Audio::Lifecycle>();
+	subsystems.Initialize<Libs::Graphics::Lifecycle>();
 }
 
 static void LoadElf(const std::filesystem::path& elf, bool dbg_print_reloc = false,
@@ -189,8 +170,9 @@ void Run(const RunOptions& options) {
 		EXIT("ELF is required\n");
 	}
 
-	const auto param_json = options.app0_dir / "sce_sys" / "param.json";
-	Init(options.config, param_json);
+	const auto         param_json = options.app0_dir / "sce_sys" / "param.json";
+	Common::Subsystems subsystems(true);
+	Init(options.config, param_json, subsystems);
 
 	ClearDebugTextureFolder();
 

src/generate_version.cmake

@@ -1,7 +1,8 @@
 set(KYTY_GIT_VERSION "unknown")
+set(KYTY_GIT_HASH "unknown")
 if(GIT_EXECUTABLE)
 	execute_process(
-		COMMAND "${GIT_EXECUTABLE}" describe --tags --always
+		COMMAND "${GIT_EXECUTABLE}" describe --tags --always --dirty
 		WORKING_DIRECTORY "${GIT_WORKING_DIRECTORY}"
 		OUTPUT_VARIABLE KYTY_GIT_VERSION
 		OUTPUT_STRIP_TRAILING_WHITESPACE
@@ -11,5 +12,27 @@ if(GIT_EXECUTABLE)
 	if(NOT GIT_RESULT EQUAL 0)
 		set(KYTY_GIT_VERSION "unknown")
 	endif()
+
+	execute_process(
+		COMMAND "${GIT_EXECUTABLE}" rev-parse --short=7 HEAD
+		WORKING_DIRECTORY "${GIT_WORKING_DIRECTORY}"
+		OUTPUT_VARIABLE KYTY_GIT_HASH
+		OUTPUT_STRIP_TRAILING_WHITESPACE
+		RESULT_VARIABLE GIT_HASH_RESULT
+		ERROR_QUIET
+	)
+	if(NOT GIT_HASH_RESULT EQUAL 0)
+		set(KYTY_GIT_HASH "unknown")
+	else()
+		execute_process(
+			COMMAND "${GIT_EXECUTABLE}" diff-index --quiet HEAD --
+			WORKING_DIRECTORY "${GIT_WORKING_DIRECTORY}"
+			RESULT_VARIABLE GIT_DIRTY_RESULT
+			ERROR_QUIET
+		)
+		if(NOT GIT_DIRTY_RESULT EQUAL 0)
+			string(APPEND KYTY_GIT_HASH "-dirty")
+		endif()
+	endif()
 endif()
 configure_file("${INPUT_FILE}" "${OUTPUT_FILE}")

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                                                                    \
@@ -101,6 +103,128 @@ void LogUnknownReleaseMemGcr(uint32_t gcr_cntl) {
 
 } // namespace
 
+static HW::BlendControl DecodeBlendControl(uint32_t value) {
+	HW::BlendControl r {};
+
+	r.color_srcblend       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_SRCBLEND);
+	r.color_comb_fcn       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_COMB_FCN);
+	r.color_destblend      = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_DESTBLEND);
+	r.alpha_srcblend       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_SRCBLEND);
+	r.alpha_comb_fcn       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_COMB_FCN);
+	r.alpha_destblend      = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_DESTBLEND);
+	r.separate_alpha_blend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, SEPARATE_ALPHA_BLEND) != 0;
+	r.enable               = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ENABLE) != 0;
+
+	return r;
+}
+
+static HW::ClipControl DecodeClipControl(uint32_t value) {
+	HW::ClipControl r {};
+
+	r.user_clip_planes          = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_ENA);
+	r.user_clip_plane_mode      = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_MODE);
+	r.dx_clip_space             = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
+	r.vertex_kill_any           = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
+	r.min_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
+	r.max_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
+	r.user_clip_plane_negate_y  = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
+	r.clip_disable              = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
+	r.user_clip_plane_cull_only = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
+	r.cull_on_clipping_error_disable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
+	r.linear_attribute_clip_enable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
+	r.force_viewport_index_from_vs_enable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
+
+	return r;
+}
+
+static HW::ColorControl DecodeColorControl(uint32_t value) {
+	HW::ColorControl r {};
+	r.mode = KYTY_PM4_GET(value, CB_COLOR_CONTROL, MODE);
+	r.op   = KYTY_PM4_GET(value, CB_COLOR_CONTROL, ROP3);
+	return r;
+}
+
+static HW::DepthControl DecodeDepthControl(uint32_t value) {
+	HW::DepthControl r {};
+
+	r.stencil_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
+	r.z_enable            = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_ENABLE) != 0;
+	r.z_write_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
+	r.depth_bounds_enable = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
+	r.zfunc               = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ZFUNC);
+	r.backface_enable     = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
+	r.stencilfunc         = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC);
+	r.stencilfunc_bf      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC_BF);
+	r.color_writes_on_depth_fail_enable =
+	    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
+	r.color_writes_on_depth_pass_disable =
+	    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
+
+	return r;
+}
+
+static HW::ModeControl DecodeModeControl(uint32_t value) {
+	HW::ModeControl r {};
+
+	r.cull_front           = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
+	r.cull_back            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
+	r.face                 = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, FACE) != 0;
+	r.poly_mode            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_MODE);
+	r.polymode_front_ptype = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
+	r.polymode_back_ptype  = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
+	r.poly_offset_front_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
+	r.poly_offset_back_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
+	r.vtx_window_offset_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
+	r.provoking_vtx_last = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
+	r.persp_corr_dis     = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
+
+	return r;
+}
+
+static HW::RenderControl DecodeRenderControl(uint32_t value) {
+	HW::RenderControl r {};
+
+	r.depth_clear_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
+	r.stencil_clear_enable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
+	r.resummarize_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
+	r.stencil_compress_disable =
+	    KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
+	r.depth_compress_disable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
+	r.copy_centroid          = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_CENTROID) != 0;
+	r.copy_sample            = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_SAMPLE);
+
+	return r;
+}
+
+static HW::ScanModeControl DecodeScanModeControl(uint32_t value) {
+	HW::ScanModeControl r {};
+
+	r.msaa_enable          = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
+	r.vport_scissor_enable = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
+	r.line_stipple_enable  = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
+
+	return r;
+}
+
+static HW::StencilControl DecodeStencilControl(uint32_t value) {
+	HW::StencilControl r {};
+
+	r.stencil_fail     = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL);
+	r.stencil_zpass    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS);
+	r.stencil_zfail    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL);
+	r.stencil_fail_bf  = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL_BF);
+	r.stencil_zpass_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS_BF);
+	r.stencil_zfail_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL_BF);
+
+	return r;
+}
+
 static HW::AaConfig ParseAaConfig(uint32_t value) {
 	HW::AaConfig r;
 
@@ -181,6 +305,10 @@ 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) {}
@@ -423,26 +551,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetBlendControl) {
 
 	uint32_t param = (cmd_offset - Pm4::CB_BLEND0_CONTROL) / 1;
 
-	HW::BlendControl r;
-
-	r.color_srcblend       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_SRCBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_SRCBLEND_MASK;
-	r.color_comb_fcn       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_COMB_FCN_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_COMB_FCN_MASK;
-	r.color_destblend      = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_DESTBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_DESTBLEND_MASK;
-	r.alpha_srcblend       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_SRCBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_SRCBLEND_MASK;
-	r.alpha_comb_fcn       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_COMB_FCN_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_COMB_FCN_MASK;
-	r.alpha_destblend      = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_DESTBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_DESTBLEND_MASK;
-	r.separate_alpha_blend = ((buffer[0] >> Pm4::CB_BLEND0_CONTROL_SEPARATE_ALPHA_BLEND_SHIFT) &
-	                          Pm4::CB_BLEND0_CONTROL_SEPARATE_ALPHA_BLEND_MASK) != 0;
-	r.enable               = ((buffer[0] >> Pm4::CB_BLEND0_CONTROL_ENABLE_SHIFT) &
-	                          Pm4::CB_BLEND0_CONTROL_ENABLE_MASK) != 0;
-
-	cp.GetCtx().SetBlendControl(param, r);
+	cp.GetCtx().SetBlendControl(param, DecodeBlendControl(buffer[0]));
 
 	return 1;
 }
@@ -451,38 +560,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetClipControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_CL_CLIP_CNTL);
 
-	HW::ClipControl r;
-
-	//	r.user_clip_planes                    = buffer[0] & 0x3fu;
-	//	r.user_clip_plane_mode                = (buffer[0] >> 14u) & 0x3u;
-	//	r.clip_space                          = (buffer[0] >> 19u) & 0x1u;
-	//	r.vertex_kill_mode                    = (buffer[0] >> 21u) & 0x1u;
-	//	r.min_z_clip_enable                   = (buffer[0] >> 26u) & 0x1u;
-	//	r.max_z_clip_enable                   = (buffer[0] >> 27u) & 0x1u;
-	//	r.user_clip_plane_negate_y            = (buffer[0] & 0x00002000u) != 0;
-	//	r.clip_enable                         = (buffer[0] & 0x00010000u) != 0;
-	//	r.user_clip_plane_cull_only           = (buffer[0] & 0x00020000u) != 0;
-	//	r.cull_on_clipping_error_disable      = (buffer[0] & 0x00100000u) != 0;
-	//	r.linear_attribute_clip_enable        = (buffer[0] & 0x01000000u) != 0;
-	//	r.force_viewport_index_from_vs_enable = (buffer[0] & 0x02000000u) != 0;
-
-	r.user_clip_planes          = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, UCP_ENA);
-	r.user_clip_plane_mode      = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, PS_UCP_MODE);
-	r.dx_clip_space             = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
-	r.vertex_kill_any           = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
-	r.min_z_clip_disable        = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
-	r.max_z_clip_disable        = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
-	r.user_clip_plane_negate_y  = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
-	r.clip_disable              = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
-	r.user_clip_plane_cull_only = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
-	r.cull_on_clipping_error_disable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
-	r.linear_attribute_clip_enable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
-	r.force_viewport_index_from_vs_enable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
-
-	cp.GetCtx().SetClipControl(r);
+	cp.GetCtx().SetClipControl(DecodeClipControl(buffer[0]));
 
 	return 1;
 }
@@ -491,12 +569,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetColorControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::CB_COLOR_CONTROL);
 
-	HW::ColorControl r;
-
-	r.mode = KYTY_PM4_GET(buffer[0], CB_COLOR_CONTROL, MODE);
-	r.op   = KYTY_PM4_GET(buffer[0], CB_COLOR_CONTROL, ROP3);
-
-	cp.GetCtx().SetColorControl(r);
+	cp.GetCtx().SetColorControl(DecodeColorControl(buffer[0]));
 
 	return 1;
 }
@@ -520,9 +593,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetColorInfo) {
 	//	r.format                         = (buffer[4] >> 2u) & 0x1fu;
 	//	r.channel_type                   = (buffer[4] >> 8u) & 0x7u;
 	//	r.channel_order                  = (buffer[4] >> 11u) & 0x3u;
-	r.format                   = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FORMAT);
-	r.channel_type             = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, NUMBER_TYPE);
-	r.channel_order            = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMP_SWAP);
+	r.format =
+	    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FORMAT));
+	r.channel_type =
+	    static_cast<Prospero::ChannelType>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, NUMBER_TYPE));
+	r.channel_order =
+	    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMP_SWAP));
 	r.cmask_fast_clear_enable  = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FAST_CLEAR) != 0;
 	r.fmask_compression_enable = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMPRESSION) != 0;
 	r.blend_clamp              = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -554,22 +630,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetDepthControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_DEPTH_CONTROL);
 
-	HW::DepthControl r;
-
-	r.stencil_enable      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
-	r.z_enable            = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, Z_ENABLE) != 0;
-	r.z_write_enable      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
-	r.depth_bounds_enable = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
-	r.zfunc               = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, ZFUNC);
-	r.backface_enable     = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
-	r.stencilfunc         = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCILFUNC);
-	r.stencilfunc_bf      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCILFUNC_BF);
-	r.color_writes_on_depth_fail_enable =
-	    KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
-	r.color_writes_on_depth_pass_disable =
-	    KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
-
-	cp.GetCtx().SetDepthControl(r);
+	cp.GetCtx().SetDepthControl(DecodeDepthControl(buffer[0]));
 
 	return 1;
 }
@@ -873,45 +934,49 @@ KYTY_HW_CTX_PARSER(HwCtxSetModeControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_SU_SC_MODE_CNTL);
 
-	HW::ModeControl r;
-
-	r.cull_front           = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
-	r.cull_back            = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
-	r.face                 = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, FACE) != 0;
-	r.poly_mode            = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_MODE);
-	r.polymode_front_ptype = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
-	r.polymode_back_ptype  = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
-	r.poly_offset_front_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
-	r.poly_offset_back_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
-	r.vtx_window_offset_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
-	r.provoking_vtx_last = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
-	r.persp_corr_dis     = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
-
-	cp.GetCtx().SetModeControl(r);
+	cp.GetCtx().SetModeControl(DecodeModeControl(buffer[0]));
 
 	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;
@@ -935,19 +1000,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_RENDER_CONTROL);
 
-	HW::RenderControl r;
-
-	r.depth_clear_enable   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
-	r.stencil_clear_enable = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
-	r.resummarize_enable   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
-	r.stencil_compress_disable =
-	    KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
-	r.depth_compress_disable =
-	    KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
-	r.copy_centroid = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, COPY_CENTROID) != 0;
-	r.copy_sample   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, COPY_SAMPLE);
-
-	cp.GetCtx().SetRenderControl(r);
+	cp.GetCtx().SetRenderControl(DecodeRenderControl(buffer[0]));
 
 	return 1;
 }
@@ -1015,9 +1068,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderTarget) {
 	//	info.format                  = (buffer[4] >> 2u) & 0x1fu;
 	//	info.channel_type            = (buffer[4] >> 8u) & 0x7u;
 	//	info.channel_order           = (buffer[4] >> 11u) & 0x3u;
-	info.format                   = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FORMAT);
-	info.channel_type             = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, NUMBER_TYPE);
-	info.channel_order            = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMP_SWAP);
+	info.format =
+	    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FORMAT));
+	info.channel_type =
+	    static_cast<Prospero::ChannelType>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, NUMBER_TYPE));
+	info.channel_order =
+	    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMP_SWAP));
 	info.cmask_fast_clear_enable  = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FAST_CLEAR) != 0;
 	info.fmask_compression_enable = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMPRESSION) != 0;
 	info.blend_clamp              = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -1039,10 +1095,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderTarget) {
 	//	attrib.num_fragments            = (buffer[5] >> 15u) & 0x3u;
 	attrib.force_dest_alpha_to_one =
 	    KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FORCE_DST_ALPHA_1) != 0;
-	attrib.tile_mode       = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, TILE_MODE_INDEX);
-	attrib.fmask_tile_mode = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX);
-	attrib.num_samples     = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_SAMPLES);
-	attrib.num_fragments   = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
+	attrib.tile_mode =
+	    static_cast<Prospero::TileMode>(KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, TILE_MODE_INDEX));
+	attrib.fmask_tile_mode = static_cast<Prospero::TileMode>(
+	    KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX));
+	attrib.num_samples   = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_SAMPLES);
+	attrib.num_fragments = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
 
 	//	dcc.max_uncompressed_block_size = (buffer[6] >> 2u) & 0x3u;
 	//	dcc.max_compressed_block_size   = (buffer[6] >> 5u) & 0x3u;
@@ -1135,13 +1193,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetScanModeControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_SC_MODE_CNTL_0);
 
-	HW::ScanModeControl r;
-
-	r.msaa_enable          = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
-	r.vport_scissor_enable = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
-	r.line_stipple_enable  = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
-
-	cp.GetCtx().SetScanModeControl(r);
+	cp.GetCtx().SetScanModeControl(DecodeScanModeControl(buffer[0]));
 
 	return 1;
 }
@@ -1181,16 +1233,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetStencilControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_STENCIL_CONTROL);
 
-	HW::StencilControl r;
-
-	r.stencil_fail     = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILFAIL);
-	r.stencil_zpass    = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZPASS);
-	r.stencil_zfail    = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZFAIL);
-	r.stencil_fail_bf  = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILFAIL_BF);
-	r.stencil_zpass_bf = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZPASS_BF);
-	r.stencil_zfail_bf = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZFAIL_BF);
-
-	cp.GetCtx().SetStencilControl(r);
+	cp.GetCtx().SetStencilControl(DecodeStencilControl(buffer[0]));
 
 	return 1;
 }
@@ -1312,26 +1355,36 @@ 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) &
@@ -1531,7 +1584,7 @@ KYTY_HW_UC_PARSER(HwUcSetPrimitiveType) {
 
 	uint32_t prim_type = KYTY_PM4_GET(buffer[0], VGT_PRIMITIVE_TYPE, PRIM_TYPE);
 
-	cp.GetUcfg().SetPrimitiveType(prim_type);
+	cp.GetUcfg().SetPrimitiveType(static_cast<Prospero::PrimitiveType>(prim_type));
 
 	return 1;
 }
@@ -2102,11 +2155,24 @@ 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 +2702,21 @@ 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 +2724,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2665,10 +2734,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2682,17 +2747,6 @@ 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 +2761,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 
 	if (data_sel == 5) {
 		if (gl2_writeback) {
-			cp.SynchronizeGpu();
 			cache_action = 0;
 		}
 
@@ -2718,7 +2771,9 @@ 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;
 	}
@@ -3067,9 +3122,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t      slot = (cmd_offset - Pm4::CB_COLOR0_INFO) / 15;
 			HW::ColorInfo info;
-			info.format                   = KYTY_PM4_GET(value, CB_COLOR0_INFO, FORMAT);
-			info.channel_type             = KYTY_PM4_GET(value, CB_COLOR0_INFO, NUMBER_TYPE);
-			info.channel_order            = KYTY_PM4_GET(value, CB_COLOR0_INFO, COMP_SWAP);
+			info.format =
+			    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(value, CB_COLOR0_INFO, FORMAT));
+			info.channel_type = static_cast<Prospero::ChannelType>(
+			    KYTY_PM4_GET(value, CB_COLOR0_INFO, NUMBER_TYPE));
+			info.channel_order =
+			    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(value, CB_COLOR0_INFO, COMP_SWAP));
 			info.cmask_fast_clear_enable  = KYTY_PM4_GET(value, CB_COLOR0_INFO, FAST_CLEAR) != 0;
 			info.fmask_compression_enable = KYTY_PM4_GET(value, CB_COLOR0_INFO, COMPRESSION) != 0;
 			info.blend_clamp              = KYTY_PM4_GET(value, CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -3094,10 +3152,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 			HW::ColorAttrib attrib;
 			attrib.force_dest_alpha_to_one =
 			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FORCE_DST_ALPHA_1) != 0;
-			attrib.tile_mode       = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, TILE_MODE_INDEX);
-			attrib.fmask_tile_mode = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX);
-			attrib.num_samples     = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_SAMPLES);
-			attrib.num_fragments   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
+			attrib.tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, TILE_MODE_INDEX));
+			attrib.fmask_tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX));
+			attrib.num_samples   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_SAMPLES);
+			attrib.num_fragments = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
 			cp.GetCtx().SetColorAttrib(slot, attrib);
 		};
 	}
@@ -3251,8 +3311,9 @@ void GraphicsInitJmpTablesCxIndirect() {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t         slot = (cmd_offset - Pm4::CB_COLOR0_ATTRIB3);
 			HW::ColorAttrib3 attrib3;
-			attrib3.depth              = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, MIP0_DEPTH);
-			attrib3.tile_mode          = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, COLOR_SW_MODE);
+			attrib3.depth     = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, MIP0_DEPTH);
+			attrib3.tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, COLOR_SW_MODE));
 			attrib3.dimension          = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, RESOURCE_TYPE);
 			attrib3.cmask_pipe_aligned = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, CMASK_PIPE_ALIGNED);
 			attrib3.dcc_pipe_aligned   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, DCC_PIPE_ALIGNED);
@@ -3308,6 +3369,12 @@ 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) {
@@ -3646,58 +3713,31 @@ void GraphicsInitJmpTablesCxIndirect() {
 	for (uint32_t slot = 0; slot < 8; slot++) {
 		g_hw_ctx_indirect_func[Pm4::CB_BLEND0_CONTROL + slot] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t param = cmd_offset - Pm4::CB_BLEND0_CONTROL;
-
-			HW::BlendControl r;
-			r.color_srcblend  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_SRCBLEND);
-			r.color_comb_fcn  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_COMB_FCN);
-			r.color_destblend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_DESTBLEND);
-			r.alpha_srcblend  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_SRCBLEND);
-			r.alpha_comb_fcn  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_COMB_FCN);
-			r.alpha_destblend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_DESTBLEND);
-			r.separate_alpha_blend =
-			    KYTY_PM4_GET(value, CB_BLEND0_CONTROL, SEPARATE_ALPHA_BLEND) != 0;
-			r.enable = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ENABLE) != 0;
-
-			cp.GetCtx().SetBlendControl(param, r);
+			cp.GetCtx().SetBlendControl(param, DecodeBlendControl(value));
 		};
 	}
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_SC_MODE_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ModeControl r;
-		r.cull_front           = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
-		r.cull_back            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
-		r.face                 = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, FACE) != 0;
-		r.poly_mode            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_MODE);
-		r.polymode_front_ptype = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
-		r.polymode_back_ptype  = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
-		r.poly_offset_front_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
-		r.poly_offset_back_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
-		r.vtx_window_offset_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
-		r.provoking_vtx_last = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
-		r.persp_corr_dis     = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
-		cp.GetCtx().SetModeControl(r);
+		cp.GetCtx().SetModeControl(DecodeModeControl(value));
 	};
 
 	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) {
@@ -3846,28 +3886,11 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_STENCIL_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::StencilControl r;
-		r.stencil_fail     = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL);
-		r.stencil_zpass    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS);
-		r.stencil_zfail    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL);
-		r.stencil_fail_bf  = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL_BF);
-		r.stencil_zpass_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS_BF);
-		r.stencil_zfail_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL_BF);
-		cp.GetCtx().SetStencilControl(r);
+		cp.GetCtx().SetStencilControl(DecodeStencilControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::RenderControl r;
-		r.depth_clear_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
-		r.stencil_clear_enable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
-		r.resummarize_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
-		r.stencil_compress_disable =
-		    KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
-		r.depth_compress_disable =
-		    KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
-		r.copy_centroid = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_CENTROID) != 0;
-		r.copy_sample   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_SAMPLE);
-		cp.GetCtx().SetRenderControl(r);
+		cp.GetCtx().SetRenderControl(DecodeRenderControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_STENCILREFMASK] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3889,23 +3912,7 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_CL_CLIP_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ClipControl r;
-		r.user_clip_planes          = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_ENA);
-		r.user_clip_plane_mode      = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_MODE);
-		r.dx_clip_space             = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
-		r.vertex_kill_any           = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
-		r.min_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
-		r.max_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
-		r.user_clip_plane_negate_y  = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
-		r.clip_disable              = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
-		r.user_clip_plane_cull_only = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
-		r.cull_on_clipping_error_disable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
-		r.linear_attribute_clip_enable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
-		r.force_viewport_index_from_vs_enable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
-		cp.GetCtx().SetClipControl(r);
+		cp.GetCtx().SetClipControl(DecodeClipControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_LINE_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3933,11 +3940,7 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SC_MODE_CNTL_0] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ScanModeControl r;
-		r.msaa_enable          = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
-		r.vport_scissor_enable = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
-		r.line_stipple_enable  = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
-		cp.GetCtx().SetScanModeControl(r);
+		cp.GetCtx().SetScanModeControl(DecodeScanModeControl(value));
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SC_MODE_CNTL_1] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreScanModeControl1(value);
@@ -3996,27 +3999,11 @@ void GraphicsInitJmpTablesCxIndirect() {
 		HwCtxIgnoreAlphaToMaskRegister(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::CB_COLOR_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ColorControl r;
-		r.mode = KYTY_PM4_GET(value, CB_COLOR_CONTROL, MODE);
-		r.op   = KYTY_PM4_GET(value, CB_COLOR_CONTROL, ROP3);
-		cp.GetCtx().SetColorControl(r);
+		cp.GetCtx().SetColorControl(DecodeColorControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_DEPTH_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::DepthControl r;
-		r.stencil_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
-		r.z_enable            = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_ENABLE) != 0;
-		r.z_write_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
-		r.depth_bounds_enable = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
-		r.zfunc               = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ZFUNC);
-		r.backface_enable     = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
-		r.stencilfunc         = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC);
-		r.stencilfunc_bf      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC_BF);
-		r.color_writes_on_depth_fail_enable =
-		    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
-		r.color_writes_on_depth_pass_disable =
-		    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
-		cp.GetCtx().SetDepthControl(r);
+		cp.GetCtx().SetDepthControl(DecodeDepthControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_EQAA] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -4220,29 +4207,28 @@ 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 +4316,7 @@ 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 +4376,7 @@ 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 +4393,11 @@ 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;
@@ -4418,7 +4409,7 @@ void GraphicsInitJmpTablesUcIndirect() {
 
 	g_hw_uc_indirect_func[Pm4::VGT_PRIMITIVE_TYPE] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		uint32_t prim_type = KYTY_PM4_GET(value, VGT_PRIMITIVE_TYPE, PRIM_TYPE);
-		cp.GetUcfg().SetPrimitiveType(prim_type);
+		cp.GetUcfg().SetPrimitiveType(static_cast<Prospero::PrimitiveType>(prim_type));
 	};
 	g_hw_uc_indirect_func[Pm4::VGT_INDEX_TYPE] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		cp.SetIndexType(value & 0x3u);

src/graphics/guest_gpu/gpu_defs.h

@@ -3,15 +3,8 @@
 
 #include "common/common.h"
 
-#include <type_traits>
-
 namespace Libs::Graphics::Prospero {
 
-template <typename T>
-constexpr auto GpuEnumValue(T value) {
-	return static_cast<std::underlying_type_t<T>>(value);
-}
-
 enum class PrimitiveType : uint32_t {
 	kNone               = 0,
 	kPointList          = 1,
@@ -60,24 +53,34 @@ 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 {
@@ -105,32 +108,52 @@ enum class ChannelOrder : uint32_t {
 	kAltReversed = 3,
 };
 
-// Maps a shader's logical RGBA export to the host color-attachment components. Two bits per
-// component keep this cheap to carry in shader and pipeline keys.
+// Selects the logical shader component written to each physical color-attachment component.
+// Two bits per component keep this cheap to carry in shader and pipeline keys.
 struct ColorComponentMapping {
-	uint8_t packed = 0xe4u; // RGBA: 0, 1, 2, 3
+	static constexpr uint8_t kIdentity = 0xe4u;
+
+	uint8_t packed = kIdentity;
 
 	[[nodiscard]] constexpr uint32_t Map(uint32_t component) const {
 		return component < 4u ? (packed >> (component * 2u)) & 0x3u : component;
 	}
 
-	[[nodiscard]] constexpr uint32_t ApplyMask(uint32_t mask) const {
+	[[nodiscard]] constexpr uint32_t ApplyMask(uint32_t logical_mask) const {
 		uint32_t mapped = 0;
-		for (uint32_t component = 0; component < 4u; component++) {
-			mapped |= ((mask >> component) & 1u) << Map(component);
+		for (uint32_t physical_component = 0; physical_component < 4u; physical_component++) {
+			mapped |= ((logical_mask >> Map(physical_component)) & 1u) << physical_component;
 		}
 		return mapped;
 	}
 
-	[[nodiscard]] constexpr bool IsIdentity() const { return packed == 0xe4u; }
+	// This mapping selects an intermediate component; next selects the final logical component.
+	[[nodiscard]] constexpr ColorComponentMapping Then(ColorComponentMapping next) const {
+		ColorComponentMapping result {0u};
+		for (uint32_t physical_component = 0; physical_component < 4u; physical_component++) {
+			result.packed |= static_cast<uint8_t>(next.Map(Map(physical_component))
+			                                      << (physical_component * 2u));
+		}
+		return result;
+	}
+
+	[[nodiscard]] constexpr bool IsIdentity() const { return packed == kIdentity; }
 
 	bool operator==(const ColorComponentMapping&) const = default;
 };
 
 static_assert(sizeof(ColorComponentMapping) == sizeof(uint8_t));
 
-inline constexpr ColorComponentMapping ColorMappingAbgr {0x1bu}; // 3, 2, 1, 0
+inline constexpr ColorComponentMapping ColorMappingRgba {ColorComponentMapping::kIdentity};
+inline constexpr ColorComponentMapping ColorMappingGr {0xe1u};   // 1, 0, 2, 3
+inline constexpr ColorComponentMapping ColorMappingRabg {0x6cu}; // 0, 3, 2, 1
+inline constexpr ColorComponentMapping ColorMappingRgab {0xb4u}; // 0, 1, 3, 2
 inline constexpr ColorComponentMapping ColorMappingBgra {0xc6u}; // 2, 1, 0, 3
+inline constexpr ColorComponentMapping ColorMappingAbgr {0x1bu}; // 3, 2, 1, 0
+inline constexpr ColorComponentMapping ColorMappingAgba {0x27u}; // 3, 1, 2, 0
+inline constexpr ColorComponentMapping ColorMappingArbg {0x63u}; // 3, 0, 2, 1
+inline constexpr ColorComponentMapping ColorMappingAgbr {0x87u}; // 3, 1, 0, 2
+inline constexpr ColorComponentMapping ColorMappingArgb {0x93u}; // 3, 0, 1, 2
 
 enum class DepthFormat : uint32_t {
 	kInvalid = 0,
@@ -378,9 +401,11 @@ enum class BufferFormat : uint32_t {
 	k8Srgb              = 128,
 	k8_8Srgb            = 129,
 	k8_8_8_8Srgb        = 130,
+	k10_10_10_2Float    = 131,
 	k9_9_9_5Float       = 132,
 	k5_6_5UNorm         = 133,
 	k5_5_5_1UNorm       = 134,
+	k1_5_5_5UNorm       = 135,
 	k4_4_4_4UNorm       = 136,
 	kFmask8_S4_F4       = 161,
 	kBc1UNorm           = 169,

src/graphics/guest_gpu/gpu_format.cpp

@@ -3,129 +3,216 @@
 #include "graphics/guest_gpu/gpu_defs.h"
 
 #include <array>
+#include <bit>
 
 namespace Libs::Graphics::Prospero {
 namespace {
 
 struct FormatInfo {
-	uint32_t format;
-	uint32_t bytes_per_element;
-	uint32_t block_compressed_bytes_per_block;
-	uint32_t render_target_bytes_per_element;
-	bool     supported_sampled_texture_format;
-	bool     unsigned_integer_texture_format;
+	BufferFormat format;
+	uint32_t     bytes_per_element;
+	uint32_t     block_compressed_bytes_per_block;
+	uint32_t     render_target_bytes_per_element;
+	bool         supported_sampled_texture_format;
+	bool         unsigned_integer_texture_format;
 };
 
 constexpr FormatInfo kFormatInfo[] = {
-    {GpuEnumValue(BufferFormat::k8UNorm), 1, 0, 1, true, false},
-    {GpuEnumValue(BufferFormat::k8SNorm), 0, 0, 1, false, false},
-    {GpuEnumValue(BufferFormat::k8UInt), 1, 0, 1, true, true},
-    {GpuEnumValue(BufferFormat::k16UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k16SNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k16UInt), 2, 0, 2, true, true},
-    {GpuEnumValue(BufferFormat::k16SInt), 2, 0, 2, false, false},
-    {GpuEnumValue(BufferFormat::k16Float), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8SNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8UInt), 2, 0, 2, true, true},
-    {GpuEnumValue(BufferFormat::k8_8SInt), 2, 0, 2, false, false},
-    {GpuEnumValue(BufferFormat::k32UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k32SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k32Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16SNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k16_16SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k16_16Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k11_11_10Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k10_10_10_2UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k10_10_10_2UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k8_8_8_8UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8SNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k8_8_8_8SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k32_32UInt), 8, 0, 8, true, true},
-    {GpuEnumValue(BufferFormat::k32_32SInt), 8, 0, 8, false, false},
-    {GpuEnumValue(BufferFormat::k32_32Float), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16UNorm), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16SNorm), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16UInt), 8, 0, 8, true, true},
-    {GpuEnumValue(BufferFormat::k16_16_16_16SInt), 8, 0, 8, false, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16Float), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k32_32_32UInt), 12, 0, 12, true, true},
-    {GpuEnumValue(BufferFormat::k32_32_32SInt), 12, 0, 12, false, false},
-    {GpuEnumValue(BufferFormat::k32_32_32Float), 12, 0, 12, true, false},
-    {GpuEnumValue(BufferFormat::k32_32_32_32UInt), 16, 0, 16, true, true},
-    {GpuEnumValue(BufferFormat::k32_32_32_32SInt), 16, 0, 16, false, false},
-    {GpuEnumValue(BufferFormat::k32_32_32_32Float), 16, 0, 16, true, false},
-    {GpuEnumValue(BufferFormat::k8Srgb), 1, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k8_8Srgb), 2, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8Srgb), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k9_9_9_5Float), 4, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k5_6_5UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k5_5_5_1UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k4_4_4_4UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::kFmask8_S4_F4), 1, 0, 1, true, false},
-    {GpuEnumValue(BufferFormat::kBc1UNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc1Srgb), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc2UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc2Srgb), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc3UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc3Srgb), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc4UNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc4SNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc5UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc5SNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc6UFloat), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc6SFloat), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc7UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc7Srgb), 0, 16, 0, true, false},
+    {BufferFormat::k8UNorm, 1, 0, 1, true, false},
+    {BufferFormat::k8SNorm, 0, 0, 1, false, false},
+    {BufferFormat::k8UInt, 1, 0, 1, true, true},
+    {BufferFormat::k16UNorm, 2, 0, 2, true, false},
+    {BufferFormat::k16SNorm, 2, 0, 2, true, false},
+    {BufferFormat::k16UInt, 2, 0, 2, true, true},
+    {BufferFormat::k16SInt, 2, 0, 2, false, false},
+    {BufferFormat::k16Float, 2, 0, 2, true, false},
+    {BufferFormat::k8_8UNorm, 2, 0, 2, true, false},
+    {BufferFormat::k8_8SNorm, 2, 0, 2, true, false},
+    {BufferFormat::k8_8UInt, 2, 0, 2, true, true},
+    {BufferFormat::k8_8SInt, 2, 0, 2, false, false},
+    {BufferFormat::k32UInt, 4, 0, 4, true, true},
+    {BufferFormat::k32SInt, 4, 0, 4, false, false},
+    {BufferFormat::k32Float, 4, 0, 4, true, false},
+    {BufferFormat::k16_16UNorm, 4, 0, 4, true, false},
+    {BufferFormat::k16_16SNorm, 4, 0, 4, true, false},
+    {BufferFormat::k16_16UInt, 4, 0, 4, true, true},
+    {BufferFormat::k16_16SInt, 4, 0, 4, false, false},
+    {BufferFormat::k16_16Float, 4, 0, 4, true, false},
+    {BufferFormat::k11_11_10Float, 4, 0, 4, true, false},
+    {BufferFormat::k10_10_10_2UNorm, 4, 0, 4, true, false},
+    {BufferFormat::k10_10_10_2UInt, 4, 0, 4, true, true},
+    {BufferFormat::k8_8_8_8UNorm, 4, 0, 4, true, false},
+    {BufferFormat::k8_8_8_8SNorm, 4, 0, 4, true, false},
+    {BufferFormat::k8_8_8_8UInt, 4, 0, 4, true, true},
+    {BufferFormat::k8_8_8_8SInt, 4, 0, 4, false, false},
+    {BufferFormat::k32_32UInt, 8, 0, 8, true, true},
+    {BufferFormat::k32_32SInt, 8, 0, 8, false, false},
+    {BufferFormat::k32_32Float, 8, 0, 8, true, false},
+    {BufferFormat::k16_16_16_16UNorm, 8, 0, 8, true, false},
+    {BufferFormat::k16_16_16_16SNorm, 8, 0, 8, true, false},
+    {BufferFormat::k16_16_16_16UInt, 8, 0, 8, true, true},
+    {BufferFormat::k16_16_16_16SInt, 8, 0, 8, false, false},
+    {BufferFormat::k16_16_16_16Float, 8, 0, 8, true, false},
+    {BufferFormat::k32_32_32UInt, 12, 0, 12, true, true},
+    {BufferFormat::k32_32_32SInt, 12, 0, 12, false, false},
+    {BufferFormat::k32_32_32Float, 12, 0, 12, true, false},
+    {BufferFormat::k32_32_32_32UInt, 16, 0, 16, true, true},
+    {BufferFormat::k32_32_32_32SInt, 16, 0, 16, false, false},
+    {BufferFormat::k32_32_32_32Float, 16, 0, 16, true, false},
+    {BufferFormat::k8Srgb, 1, 0, 0, true, false},
+    {BufferFormat::k8_8Srgb, 2, 0, 0, true, false},
+    {BufferFormat::k8_8_8_8Srgb, 4, 0, 4, true, false},
+    {BufferFormat::k9_9_9_5Float, 4, 0, 0, true, false},
+    {BufferFormat::k5_6_5UNorm, 2, 0, 2, true, false},
+    {BufferFormat::k5_5_5_1UNorm, 2, 0, 2, true, false},
+    {BufferFormat::k1_5_5_5UNorm, 0, 0, 2, false, false},
+    {BufferFormat::k4_4_4_4UNorm, 2, 0, 2, true, false},
+    {BufferFormat::kFmask8_S4_F4, 1, 0, 1, true, false},
+    {BufferFormat::kBc1UNorm, 0, 8, 0, true, false},
+    {BufferFormat::kBc1Srgb, 0, 8, 0, true, false},
+    {BufferFormat::kBc2UNorm, 0, 16, 0, true, false},
+    {BufferFormat::kBc2Srgb, 0, 16, 0, true, false},
+    {BufferFormat::kBc3UNorm, 0, 16, 0, true, false},
+    {BufferFormat::kBc3Srgb, 0, 16, 0, true, false},
+    {BufferFormat::kBc4UNorm, 0, 8, 0, true, false},
+    {BufferFormat::kBc4SNorm, 0, 8, 0, true, false},
+    {BufferFormat::kBc5UNorm, 0, 16, 0, true, false},
+    {BufferFormat::kBc5SNorm, 0, 16, 0, true, false},
+    {BufferFormat::kBc6UFloat, 0, 16, 0, true, false},
+    {BufferFormat::kBc6SFloat, 0, 16, 0, true, false},
+    {BufferFormat::kBc7UNorm, 0, 16, 0, true, false},
+    {BufferFormat::kBc7Srgb, 0, 16, 0, true, false},
 };
 
+constexpr uint8_t RenderTargetTypeBit(ChannelType type) {
+	return static_cast<uint8_t>(1u << static_cast<uint32_t>(type));
+}
+
+constexpr auto kUNormType = RenderTargetTypeBit(ChannelType::kUNorm);
+constexpr auto kIntegerTypes =
+    RenderTargetTypeBit(ChannelType::kUInt) | RenderTargetTypeBit(ChannelType::kSInt);
+constexpr auto kNormalizedIntegerTypes =
+    kUNormType | RenderTargetTypeBit(ChannelType::kSNorm) | kIntegerTypes;
+
+struct RenderTargetLayoutInfo {
+	ChannelLayout       layout;
+	BufferFormat        first_linear_format;
+	uint8_t             linear_type_mask;
+	uint8_t             components;
+	BufferFormat        extra_format  = BufferFormat::kInvalid;
+	ChannelType         extra_type    = ChannelType::kUNorm;
+	ChannelOrderSupport order_support = ChannelOrderSupport::kAll;
+};
+
+constexpr RenderTargetLayoutInfo kRenderTargetLayouts[] = {
+    {ChannelLayout::k8, BufferFormat::k8UNorm, kNormalizedIntegerTypes, 1, BufferFormat::k8Srgb,
+     ChannelType::kSrgb},
+    {ChannelLayout::k16, BufferFormat::k16UNorm, kNormalizedIntegerTypes, 1, BufferFormat::k16Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k8_8, BufferFormat::k8_8UNorm, kNormalizedIntegerTypes, 2,
+     BufferFormat::k8_8Srgb, ChannelType::kSrgb},
+    {ChannelLayout::k32, BufferFormat::k32UInt, kIntegerTypes, 1, BufferFormat::k32Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k16_16, BufferFormat::k16_16UNorm, kNormalizedIntegerTypes, 2,
+     BufferFormat::k16_16Float, ChannelType::kFloat},
+    {ChannelLayout::k11_11_10, BufferFormat::k11_11_10UNorm, kNormalizedIntegerTypes, 3,
+     BufferFormat::k11_11_10Float, ChannelType::kFloat},
+    {ChannelLayout::k10_11_11, BufferFormat::k10_11_11UNorm, kNormalizedIntegerTypes, 3,
+     BufferFormat::k10_11_11Float, ChannelType::kFloat},
+    {ChannelLayout::k2_10_10_10, BufferFormat::k2_10_10_10UNorm, kNormalizedIntegerTypes, 4},
+    {ChannelLayout::k10_10_10_2, BufferFormat::k10_10_10_2UNorm, kNormalizedIntegerTypes, 4},
+    {ChannelLayout::k8_8_8_8, BufferFormat::k8_8_8_8UNorm, kNormalizedIntegerTypes, 4,
+     BufferFormat::k8_8_8_8Srgb, ChannelType::kSrgb},
+    {ChannelLayout::k32_32, BufferFormat::k32_32UInt, kIntegerTypes, 2, BufferFormat::k32_32Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k16_16_16_16, BufferFormat::k16_16_16_16UNorm, kNormalizedIntegerTypes, 4,
+     BufferFormat::k16_16_16_16Float, ChannelType::kFloat},
+    {ChannelLayout::k32_32_32_32, BufferFormat::k32_32_32_32UInt, kIntegerTypes, 4,
+     BufferFormat::k32_32_32_32Float, ChannelType::kFloat},
+    {ChannelLayout::k5_6_5, BufferFormat::k5_6_5UNorm, kUNormType, 3},
+    {ChannelLayout::k5_5_5_1, BufferFormat::k5_5_5_1UNorm, kUNormType, 4},
+    {ChannelLayout::k1_5_5_5, BufferFormat::k1_5_5_5UNorm, kUNormType, 4},
+    {ChannelLayout::k4_4_4_4, BufferFormat::k4_4_4_4UNorm, kUNormType, 4},
+    {ChannelLayout::k10_10_10_2Float, BufferFormat::kInvalid, 0, 4, BufferFormat::k10_10_10_2Float,
+     ChannelType::kFloat, ChannelOrderSupport::kStandardOnly},
+};
+
+const RenderTargetLayoutInfo* FindRenderTargetLayout(ChannelLayout layout) {
+	for (const auto& info: kRenderTargetLayouts) {
+		if (info.layout == layout) {
+			return &info;
+		}
+	}
+	return nullptr;
+}
+
 constexpr auto MakeFormatInfoLookup() {
-	constexpr uint32_t                            kMaxFormat = GpuEnumValue(BufferFormat::kBc7Srgb);
+	constexpr size_t kMaxFormat = static_cast<size_t>(BufferFormat::kBc7Srgb);
 	std::array<const FormatInfo*, kMaxFormat + 1> lookup {};
 	for (const auto& info: kFormatInfo) {
-		lookup[info.format] = &info;
+		lookup[static_cast<size_t>(info.format)] = &info;
 	}
 	return lookup;
 }
 
 constexpr auto kFormatInfoLookup = MakeFormatInfoLookup();
 
-const FormatInfo* FindFormatInfo(uint32_t format) {
-	return format < kFormatInfoLookup.size() ? kFormatInfoLookup[format] : nullptr;
+const FormatInfo* FindFormatInfo(BufferFormat format) {
+	const auto index = static_cast<size_t>(format);
+	return index < kFormatInfoLookup.size() ? kFormatInfoLookup[index] : nullptr;
 }
 
 } // namespace
 
-uint32_t NumBytesPerElement(uint32_t format) {
+RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type) {
+	const auto* layout_info = FindRenderTargetLayout(layout);
+	const auto  raw_type    = static_cast<uint32_t>(type);
+	if (layout_info == nullptr || raw_type >= 8u) {
+		return {};
+	}
+
+	auto format = BufferFormat::kInvalid;
+	if (layout_info->extra_format != BufferFormat::kInvalid && type == layout_info->extra_type) {
+		format = layout_info->extra_format;
+	} else if ((layout_info->linear_type_mask & (1u << raw_type)) != 0) {
+		const auto first_type =
+		    std::countr_zero(static_cast<uint32_t>(layout_info->linear_type_mask));
+		format = static_cast<BufferFormat>(static_cast<uint32_t>(layout_info->first_linear_format) +
+		                                   raw_type - first_type);
+	}
+	if (format == BufferFormat::kInvalid) {
+		return {};
+	}
+	return {format, layout_info->components, layout_info->order_support};
+}
+
+uint32_t NumBytesPerElement(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->bytes_per_element : 0;
 }
 
-uint32_t BlockCompressedBytesPerBlock(uint32_t format) {
+uint32_t BlockCompressedBytesPerBlock(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->block_compressed_bytes_per_block : 0;
 }
 
-uint32_t RenderTargetBytesPerElement(uint32_t format) {
+uint32_t RenderTargetBytesPerElement(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->render_target_bytes_per_element : 0;
 }
 
-bool IsSupportedTextureFormat(uint32_t format) {
+bool IsSupportedTextureFormat(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr && info->supported_sampled_texture_format;
 }
 
-bool IsUintTextureFormat(uint32_t format) {
+bool IsUintTextureFormat(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr && info->unsigned_integer_texture_format;
 }
 
-bool IsFmaskTextureFormat(uint32_t format) {
-	return format == GpuEnumValue(BufferFormat::kFmask8_S4_F4);
+bool IsFmaskTextureFormat(BufferFormat format) {
+	return format == BufferFormat::kFmask8_S4_F4;
 }
 
 } // namespace Libs::Graphics::Prospero

src/graphics/guest_gpu/gpu_format.h

@@ -1,16 +1,49 @@
 #ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_GUEST_GPU_GPU_FORMAT_H_
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_GUEST_GPU_GPU_FORMAT_H_
 
-#include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 
 namespace Libs::Graphics::Prospero {
 
-uint32_t NumBytesPerElement(uint32_t format);
-uint32_t BlockCompressedBytesPerBlock(uint32_t format);
-uint32_t RenderTargetBytesPerElement(uint32_t format);
-bool     IsSupportedTextureFormat(uint32_t format);
-bool     IsUintTextureFormat(uint32_t format);
-bool     IsFmaskTextureFormat(uint32_t format);
+enum class ChannelOrderSupport : uint8_t {
+	kNone,
+	kStandardOnly,
+	kAll,
+};
+
+struct RenderTargetFormatEncoding {
+	BufferFormat        buffer_format = BufferFormat::kInvalid;
+	uint8_t             components    = 0;
+	ChannelOrderSupport order_support = ChannelOrderSupport::kNone;
+
+	[[nodiscard]] constexpr bool IsValid() const {
+		return buffer_format != BufferFormat::kInvalid && components >= 1u && components <= 4u &&
+		       order_support != ChannelOrderSupport::kNone;
+	}
+
+	[[nodiscard]] constexpr bool SupportsOrder(ChannelOrder order) const {
+		switch (order_support) {
+			case ChannelOrderSupport::kStandardOnly: return order == ChannelOrder::kStandard;
+			case ChannelOrderSupport::kAll:
+				switch (order) {
+					case ChannelOrder::kStandard:
+					case ChannelOrder::kAlt:
+					case ChannelOrder::kReversed:
+					case ChannelOrder::kAltReversed: return true;
+				}
+				return false;
+			default: return false;
+		}
+	}
+};
+
+RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type);
+uint32_t                   NumBytesPerElement(BufferFormat format);
+uint32_t                   BlockCompressedBytesPerBlock(BufferFormat format);
+uint32_t                   RenderTargetBytesPerElement(BufferFormat format);
+bool                       IsSupportedTextureFormat(BufferFormat format);
+bool                       IsUintTextureFormat(BufferFormat format);
+bool                       IsFmaskTextureFormat(BufferFormat format);
 
 } // namespace Libs::Graphics::Prospero
 

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) {
@@ -1352,9 +1384,14 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					case 0x00:
 						switch (eop_event_type) {
 							case 0x04:
+								if (event_index == 0x05) {
+									write64(false);
+									return;
+								}
+								break;
+							case 0x14:
 							case 0x28:
-								if ((eop_event_type == 0x04 && event_index == 0x05) ||
-								    (eop_event_type == 0x28 && event_index == 0x00)) {
+								if (event_index == 0x00) {
 									write64(false);
 									return;
 								}
@@ -1513,7 +1550,6 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
 				     event_index);
 			}
 			EmitGlobalBarrier();
-			SynchronizeGpu();
 			break;
 		// DbDataWritebackInvalidate, DbMetadataWritebackInvalidate, CbMetadataWritebackInvalidate.
 		case 0x0000002a:
@@ -1650,10 +1686,6 @@ 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 +1711,8 @@ 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

@@ -29,27 +29,27 @@ struct ColorView {
 struct ColorInfo {
 	bool fmask_compression_enable = false;
 	// uint32_t fmask_compression_mode   = 0;
-	bool     fmask_data_compression_disable = false;
-	bool     fmask_one_frag_mode            = false;
-	bool     cmask_fast_clear_enable        = false;
-	bool     dcc_compression_enable         = false;
-	bool     blend_clamp                    = false;
-	bool     blend_bypass                   = false;
-	bool     round_mode                     = false;
-	uint32_t cmask_is_linear                = 0;
-	uint32_t cmask_addr_type                = 0;
-	bool     alt_tile_mode                  = false;
-	uint32_t format                         = 0;
-	uint32_t channel_type                   = 0;
-	uint32_t channel_order                  = 0;
+	bool                    fmask_data_compression_disable = false;
+	bool                    fmask_one_frag_mode            = false;
+	bool                    cmask_fast_clear_enable        = false;
+	bool                    dcc_compression_enable         = false;
+	bool                    blend_clamp                    = false;
+	bool                    blend_bypass                   = false;
+	bool                    round_mode                     = false;
+	uint32_t                cmask_is_linear                = 0;
+	uint32_t                cmask_addr_type                = 0;
+	bool                    alt_tile_mode                  = false;
+	Prospero::ChannelLayout format                         = Prospero::ChannelLayout::kInvalid;
+	Prospero::ChannelType   channel_type                   = Prospero::ChannelType::kUNorm;
+	Prospero::ChannelOrder  channel_order                  = Prospero::ChannelOrder::kStandard;
 };
 
 struct ColorAttrib {
-	bool     force_dest_alpha_to_one = false;
-	uint32_t tile_mode               = 0;
-	uint32_t fmask_tile_mode         = 0;
-	uint32_t num_samples             = 0;
-	uint32_t num_fragments           = 0;
+	bool               force_dest_alpha_to_one = false;
+	Prospero::TileMode tile_mode               = Prospero::TileMode::kLinear;
+	Prospero::TileMode fmask_tile_mode         = Prospero::TileMode::kLinear;
+	uint32_t           num_samples             = 0;
+	uint32_t           num_fragments           = 0;
 };
 
 struct ColorAttrib2 {
@@ -59,11 +59,11 @@ struct ColorAttrib2 {
 };
 
 struct ColorAttrib3 {
-	uint32_t depth              = 0;
-	uint32_t tile_mode          = 0;
-	uint32_t dimension          = 0;
-	bool     cmask_pipe_aligned = false;
-	bool     dcc_pipe_aligned   = false;
+	uint32_t           depth              = 0;
+	Prospero::TileMode tile_mode          = Prospero::TileMode::kLinear;
+	uint32_t           dimension          = 0;
+	bool               cmask_pipe_aligned = false;
+	bool               dcc_pipe_aligned   = false;
 };
 
 struct ColorDccControl {
@@ -132,7 +132,7 @@ struct RenderTarget {
 };
 
 struct DepthZInfo {
-	uint32_t                                    format      = 0;
+	Prospero::DepthFormat                       format      = Prospero::DepthFormat::kInvalid;
 	uint32_t                                    num_samples = 0;
 	Prospero::TextureCompatiblePlaneCompression texture_compatibility =
 	    Prospero::TextureCompatiblePlaneCompression::kDisable;
@@ -144,16 +144,16 @@ struct DepthZInfo {
 
 	[[nodiscard]] static DepthZInfo Decode(uint32_t value) {
 		DepthZInfo info;
-		info.format                = value & 0x3u;
+		info.format                = static_cast<Prospero::DepthFormat>(value & 0x3u);
 		info.num_samples           = (value >> 2u) & 0x3u;
 		info.texture_compatibility = static_cast<Prospero::TextureCompatiblePlaneCompression>(
-		    value & Prospero::GpuEnumValue(Prospero::TextureCompatiblePlaneCompression::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::TextureCompatiblePlaneCompression::kBitMask));
 		info.partially_resident = (value & 0x00001000u) != 0;
 		info.max_mip_level      = static_cast<uint8_t>((value >> 16u) & 0x0fu);
 		info.expclear_enabled   = (value & 0x08000000u) != 0;
 		info.htile_acceleration = (value & 0x20000000u) != 0;
 		info.z_compare_base     = static_cast<Prospero::ZCompareBase>(
-		    value & Prospero::GpuEnumValue(Prospero::ZCompareBase::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::ZCompareBase::kBitMask));
 		return info;
 	}
 
@@ -167,7 +167,7 @@ struct DepthZInfo {
 };
 
 struct DepthStencilInfo {
-	uint32_t                           format = 0;
+	Prospero::StencilFormat            format = Prospero::StencilFormat::kInvalid;
 	Prospero::TextureCompatibleStencil texture_compatibility =
 	    Prospero::TextureCompatibleStencil::kDisable;
 	bool expclear_enabled       = false;
@@ -176,9 +176,9 @@ struct DepthStencilInfo {
 
 	[[nodiscard]] static DepthStencilInfo Decode(uint32_t value) {
 		DepthStencilInfo info;
-		info.format                = value & 0x1u;
+		info.format                = static_cast<Prospero::StencilFormat>(value & 0x1u);
 		info.texture_compatibility = static_cast<Prospero::TextureCompatibleStencil>(
-		    value & Prospero::GpuEnumValue(Prospero::TextureCompatibleStencil::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::TextureCompatibleStencil::kBitMask));
 		info.partially_resident     = (value & 0x00001000u) != 0;
 		info.expclear_enabled       = (value & 0x08000000u) != 0;
 		info.htile_stencil_disabled = (value & 0x20000000u) != 0;
@@ -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;
@@ -991,8 +1029,8 @@ public:
 
 	void Reset() { *this = UserConfig(); }
 
-	void                   SetPrimitiveType(uint32_t prim_type) { m_prim_type = prim_type; }
-	[[nodiscard]] uint32_t GetPrimType() const { return m_prim_type; }
+	void SetPrimitiveType(Prospero::PrimitiveType prim_type) { m_prim_type = prim_type; }
+	[[nodiscard]] Prospero::PrimitiveType GetPrimType() const { return m_prim_type; }
 	void                   SetIndexOffset(uint32_t index_offset) { m_index_offset = index_offset; }
 	[[nodiscard]] uint32_t GetIndexOffset() const { return m_index_offset; }
 	void                   SetObjectId(uint32_t object_id) { m_object_id = object_id; }
@@ -1011,9 +1049,9 @@ public:
 	void SetGdsOaAddress(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].address = value; }
 
 private:
-	uint32_t m_prim_type    = 0;
-	uint32_t m_index_offset = 0;
-	uint32_t m_object_id    = 0;
+	Prospero::PrimitiveType m_prim_type    = Prospero::PrimitiveType::kNone;
+	uint32_t                m_index_offset = 0;
+	uint32_t                m_object_id    = 0;
 
 	GeControl    m_ge_cntl;
 	GeUserVgprEn m_ge_user_vgpr_en;
@@ -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

@@ -3,6 +3,7 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 
 namespace Libs::Graphics {
 
@@ -39,7 +40,7 @@ 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;
@@ -47,42 +48,73 @@ 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 {
+	Prospero::BufferFormat format    = Prospero::BufferFormat::kInvalid;
+	Prospero::TileMode     tile_mode = Prospero::TileMode::kLinear;
+	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(Prospero::BufferFormat format, TileTextureElementLayout& layout);
+bool TileGetTextureBlockLayout(Prospero::BufferFormat format, Prospero::TileMode 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,
-                          uint32_t num_fragments_log2 = 0);
+bool     TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch,
+                          Prospero::DepthFormat z_format, Prospero::StencilFormat stencil_format,
+                          bool htile, TileSizeAlign& stencil_size, TileSizeAlign& htile_size,
+                          TileSizeAlign& depth_size, uint32_t num_fragments_log2 = 0);
 uint32_t TileGetRenderTargetPitch(uint32_t width, uint32_t bytes_per_element,
                                   uint32_t num_fragments_log2 = 0);
 uint32_t TileGetDepthPitch(uint32_t width, uint32_t bytes_per_element,
@@ -94,13 +126,14 @@ 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,
+void     TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                            uint32_t levels, Prospero::TileMode 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     TileGetTextureTotalSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                                 uint32_t depth, uint32_t levels, Prospero::TileMode tile,
+                                 bool volume_texture, TileSizeAlign& total_size);
+uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
+                             Prospero::TileMode 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/pageManager.cpp

@@ -1,5 +1,6 @@
 #include "graphics/host_gpu/pageManager.h"
 
+#include "common/platform/sysDbg.h"
 #include "graphics/host_gpu/regionDefinitions.h"
 #include "kernel/memory.h"
 
@@ -52,20 +53,20 @@ constexpr uint32_t READ_WRITE_PROTECTION = PAGE_READWRITE;
 	std::fputs("PageManager fail-fast: ", stderr);
 	std::fputs(reason != nullptr ? reason : "invalid page state", stderr);
 	std::fputc('\n', stderr);
+#if !defined(__APPLE__)
+	void* frames[16] {};
+	int   frame_count = static_cast<int>(std::size(frames));
+	SysStackWalk(frames, &frame_count);
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	void*      frames[16] {};
-	const auto frame_count =
-	    CaptureStackBackTrace(0, static_cast<DWORD>(std::size(frames)), frames, nullptr);
 	const auto image_base = reinterpret_cast<uintptr_t>(GetModuleHandleW(nullptr));
-	for (uint16_t i = 0; i < frame_count; i++) {
+	for (int i = 0; i < frame_count; i++) {
 		const auto address = reinterpret_cast<uintptr_t>(frames[i]);
-		std::fprintf(stderr, "  frame[%u]=0x%016" PRIxPTR " image_rva=0x%016" PRIxPTR "\n", i,
+		std::fprintf(stderr, "  frame[%d]=0x%016" PRIxPTR " image_rva=0x%016" PRIxPTR "\n", i,
 		             address, address >= image_base ? address - image_base : 0);
 	}
-#elif !defined(__APPLE__)
-	void*     frames[16] {};
-	const int frame_count = ::backtrace(frames, static_cast<int>(std::size(frames)));
+#else
 	::backtrace_symbols_fd(frames, frame_count, STDERR_FILENO);
+#endif
 #endif
 	std::fflush(stderr);
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS

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);
 }
 
@@ -519,7 +516,7 @@ ImageId TextureCache::GetNullImage(const ImageDesc& desc) {
 	info.pitch           = 1;
 	info.bytes_per_block = std::max(desc.info.bytes_per_block, 1u);
 	info.samples         = 1;
-	info.tile_mode       = Prospero::GpuEnumValue(Prospero::TileMode::kLinear);
+	info.tile_mode       = Prospero::TileMode::kLinear;
 	info.mip_layout[0]   = {0, info.bytes_per_block, 1, 1};
 	const auto id        = InsertImage(info);
 	m_null_images.emplace(format, id);
@@ -825,7 +822,8 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		     requested.data.address, requested.resources.levels, requested.resources.layers,
 		     cached.info.resources.levels, cached.info.resources.layers, requested.data.size,
 		     cached.info.data.size, static_cast<uint32_t>(requested.type),
-		     static_cast<uint32_t>(cached.info.type), requested.tile_mode, cached.info.tile_mode);
+		     static_cast<uint32_t>(cached.info.type), static_cast<uint32_t>(requested.tile_mode),
+		     static_cast<uint32_t>(cached.info.tile_mode));
 	}
 
 	if (requested.data.address > cached.info.data.address) {
@@ -908,10 +906,9 @@ TextureCache::ColorTransferPlan
 TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
                                  TransferDirection direction) const {
 	const auto& info             = image.info;
-	uint32_t    format           = info.guest_format;
+	auto        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 +937,6 @@ 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 +954,13 @@ 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   = 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);
+			     static_cast<uint32_t>(info.guest_format), static_cast<uint32_t>(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) {
@@ -1054,7 +1045,7 @@ void TextureCache::UploadImage(Image& image, const ImageDesc& desc, Buffer& sour
 		copy.bufferImageHeight = info.extent.height;
 		copy.imageSubresource  = {vk::ImageAspectFlagBits::eDepth, 0, layer, 1};
 		copy.imageExtent       = {info.extent.width, info.extent.height, 1};
-		if (static_cast<Prospero::TileMode>(info.tile_mode) != Prospero::TileMode::kLinear) {
+		if (info.tile_mode != Prospero::TileMode::kLinear) {
 			tiles.push_back({block.family, block.bytes_per_element, offset, full_slice_size, offset,
 			                 full_slice_size, 0, info.extent.width, info.extent.height, 1,
 			                 info.pitch});
@@ -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

@@ -101,8 +101,9 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 			     " attrib3_tile=0x%08" PRIx32 " attrib3_dim=0x%08" PRIx32 " fmt=0x%08" PRIx32
 			     " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 "\n",
 			     rt_slot, rt.base.addr, mask, rt.attrib2.width, rt.attrib2.height,
-			     rt.attrib3.tile_mode, rt.attrib3.dimension, rt.info.format, rt.info.channel_type,
-			     rt.info.channel_order);
+			     static_cast<uint32_t>(rt.attrib3.tile_mode), rt.attrib3.dimension,
+			     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+			     static_cast<uint32_t>(rt.info.channel_order));
 		}
 	}
 
@@ -129,17 +130,16 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	if (volume && samples != 1) {
 		EXIT("multisampled 3D render targets are unsupported\n");
 	}
-	const uint32_t depth = volume ? rt.attrib3.depth + 1u : 1u;
-	const bool     standard64 =
-	    rt.attrib3.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB);
+	const uint32_t depth      = volume ? rt.attrib3.depth + 1u : 1u;
+	const bool     standard64 = rt.attrib3.tile_mode == Prospero::TileMode::kStandard64KB;
 
 	switch (rt.attrib3.tile_mode) {
-		case Prospero::GpuEnumValue(Prospero::TileMode::kLinear):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget):
+		case Prospero::TileMode::kLinear:
+		case Prospero::TileMode::kStandard64KB:
+		case Prospero::TileMode::kRenderTarget:
 			tile = !RenderIsColorTileModeLinear(rt.attrib3.tile_mode);
 			break;
-		default: EXIT("unknown tile mode: %u\n", rt.attrib3.tile_mode);
+		default: EXIT("unknown tile mode: %u\n", static_cast<uint32_t>(rt.attrib3.tile_mode));
 	}
 	if (!tile && levels > 1) {
 		EXIT("linear mipmapped render targets are unsupported\n");
@@ -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);
+			pitch =
+			    TileGetTexturePitch(Prospero::BufferFormat::k32Float, width, rt.attrib3.tile_mode);
 		} else {
 			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
 		}
@@ -192,15 +192,22 @@ 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);
+			     depth, levels, static_cast<uint32_t>(rt.attrib3.tile_mode));
 		}
 		size         = volume_layout.block_slice_size;
 		backing_size = volume_layout.total_size;
@@ -208,9 +215,8 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		TileSizeAlign layout {};
 		bool          valid_layout = false;
 		if (standard64) {
-			TileGetTextureSize(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float), width,
-			                   height, pitch, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
-			                   mip_padded);
+			TileGetTextureSize(Prospero::BufferFormat::k32Float, width, height, levels,
+			                   rt.attrib3.tile_mode, &layout, mip_sizes, mip_padded);
 			valid_layout = layout.size != 0 && layout.align == 65536;
 		} else {
 			valid_layout =
@@ -271,8 +277,9 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		     " extent=%ux%ux%u view_mip=%u view_extent=%ux%u levels=%u pitch=%u"
 		     " fmt=0x%08" PRIx32 " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 " samples=%u tile=%s\n",
 		     rt_slot, rt.base.addr, backing_size, width, height, depth, rt.view.current_mip_level,
-		     view_extent.width, view_extent.height, levels, pitch, rt.info.format,
-		     rt.info.channel_type, rt.info.channel_order, samples, tile ? "tiled" : "linear");
+		     view_extent.width, view_extent.height, levels, pitch,
+		     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+		     static_cast<uint32_t>(rt.info.channel_order), samples, tile ? "tiled" : "linear");
 	}
 
 	TextureCache::ImageDesc desc {};
@@ -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>
@@ -75,8 +76,8 @@ void uc_print(const char* func, const HW::UserConfig& uc) {
 	     "\t en_user_vgpr1         = %s\n"
 	     "\t en_user_vgpr2         = %s\n"
 	     "\t en_user_vgpr3         = %s\n",
-	     uc.GetPrimType(), uc.GetIndexOffset(), uc.GetObjectId(), ge_cntl.primitive_group_size,
-	     ge_cntl.vertex_group_size, user_en.vgpr1 ? "true" : "false",
+	     static_cast<uint32_t>(uc.GetPrimType()), uc.GetIndexOffset(), uc.GetObjectId(),
+	     ge_cntl.primitive_group_size, ge_cntl.vertex_group_size, user_en.vgpr1 ? "true" : "false",
 	     user_en.vgpr2 ? "true" : "false", user_en.vgpr3 ? "true" : "false");
 }
 
@@ -140,11 +141,12 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	    fmt::format("\t info.cmask_addr_type            = 0x{:08x}\n", rt.info.cmask_addr_type));
 	dst.push_back(fmt::format("\t info.alt_tile_mode              = {}\n",
 	                          rt.info.alt_tile_mode ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n", rt.info.format));
-	dst.push_back(
-	    fmt::format("\t info.channel_type               = 0x{:08x}\n", rt.info.channel_type));
-	dst.push_back(
-	    fmt::format("\t info.channel_order              = 0x{:08x}\n", rt.info.channel_order));
+	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.format)));
+	dst.push_back(fmt::format("\t info.channel_type               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_type)));
+	dst.push_back(fmt::format("\t info.channel_order              = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_order)));
 	dst.push_back(fmt::format("\t info.blend_bypa                 = {}\n",
 	                          rt.info.blend_bypass ? "true" : "false"));
 	dst.push_back(fmt::format("\t info.blend_clamp                = {}\n",
@@ -153,10 +155,10 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	                          rt.info.round_mode ? "true" : "false"));
 	dst.push_back(fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
 	                          rt.attrib.force_dest_alpha_to_one ? "true" : "false"));
-	dst.push_back(
-	    fmt::format("\t attrib.tile_mode                = 0x{:08x}\n", rt.attrib.tile_mode));
-	dst.push_back(
-	    fmt::format("\t attrib.fmask_tile_mode          = 0x{:08x}\n", rt.attrib.fmask_tile_mode));
+	dst.push_back(fmt::format("\t attrib.tile_mode                = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib.tile_mode)));
+	dst.push_back(fmt::format("\t attrib.fmask_tile_mode          = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib.fmask_tile_mode)));
 	dst.push_back(
 	    fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples));
 	dst.push_back(
@@ -167,8 +169,8 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	dst.push_back(
 	    fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels));
 	dst.push_back(fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth));
-	dst.push_back(
-	    fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n", rt.attrib3.tile_mode));
+	dst.push_back(fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib3.tile_mode)));
 	dst.push_back(
 	    fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension));
 	dst.push_back(fmt::format("\t attrib3.cmask_pipe_aligned      = {}\n",
@@ -211,22 +213,22 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	return dst;
 }
 
-static bool RenderIsColorTileMode(uint32_t tile_mode) {
+static bool RenderIsColorTileMode(Prospero::TileMode tile_mode) {
 	// AGC CxRenderTarget::TileMode shifted by CB_COLOR*_ATTRIB3.COLOR_SW_MODE.
 	switch (tile_mode) {
-		case Prospero::GpuEnumValue(Prospero::TileMode::kLinear):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard256B):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kPrt):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kDepth):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget): return true;
+		case Prospero::TileMode::kLinear:
+		case Prospero::TileMode::kStandard256B:
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+		case Prospero::TileMode::kPrt:
+		case Prospero::TileMode::kDepth:
+		case Prospero::TileMode::kRenderTarget: return true;
 		default: return false;
 	}
 }
 
-bool RenderIsColorTileModeLinear(uint32_t tile_mode) {
-	return tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kLinear);
+bool RenderIsColorTileModeLinear(Prospero::TileMode tile_mode) {
+	return tile_mode == Prospero::TileMode::kLinear;
 }
 
 static bool RenderIsColorDimension(uint32_t dimension) {
@@ -360,7 +362,8 @@ static void RtCheck(const HW::RenderTarget& rt) {
 			}
 		}
 		if (!RenderIsColorTileMode(rt.attrib3.tile_mode)) {
-			EXIT("unknown PS5 render-target tile mode: 0x%08" PRIx32 "\n", rt.attrib3.tile_mode);
+			EXIT("unknown PS5 render-target tile mode: 0x%08" PRIx32 "\n",
+			     static_cast<uint32_t>(rt.attrib3.tile_mode));
 		}
 		if (!RenderIsColorDimension(rt.attrib3.dimension)) {
 			EXIT("unknown PS5 render-target dimension: 0x%08" PRIx32 "\n", rt.attrib3.dimension);
@@ -455,13 +458,14 @@ static void ZPrint(const char* func, const HW::DepthRenderTarget& z) {
 	     "\t height                                = 0x%08" PRIx32 "\n"
 	     "\t size.x_max                            = 0x%04" PRIx16 "\n"
 	     "\t size.y_max                            = 0x%04" PRIx16 "\n",
-	     z.z_info.format, z.z_info.num_samples,
-	     Prospero::GpuEnumValue(z.z_info.texture_compatibility),
+	     static_cast<uint32_t>(z.z_info.format), z.z_info.num_samples,
+	     static_cast<uint32_t>(z.z_info.texture_compatibility),
 	     z.z_info.htile_acceleration ? "true" : "false",
 	     z.z_info.expclear_enabled ? "true" : "false",
-	     Prospero::GpuEnumValue(z.z_info.z_compare_base),
+	     static_cast<uint32_t>(z.z_info.z_compare_base),
 	     z.z_info.partially_resident ? "true" : "false", z.z_info.max_mip_level,
-	     z.stencil_info.format, Prospero::GpuEnumValue(z.stencil_info.texture_compatibility),
+	     static_cast<uint32_t>(z.stencil_info.format),
+	     static_cast<uint32_t>(z.stencil_info.texture_compatibility),
 	     z.stencil_info.htile_stencil_disabled ? "true" : "false",
 	     z.stencil_info.expclear_enabled ? "true" : "false",
 	     z.stencil_info.partially_resident ? "true" : "false", z.depth_info.addr5_swizzle_mask,
@@ -490,15 +494,16 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 
 	EXIT_NOT_IMPLEMENTED(!z.z_info.HasValidTextureCompatibility());
 	EXIT_NOT_IMPLEMENTED(!z.stencil_info.HasValidTextureCompatibility());
-	if (z.z_info.format == 0) {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != 0);
+	if (z.z_info.format == Prospero::DepthFormat::kInvalid) {
+		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kInvalid);
 		EXIT_NOT_IMPLEMENTED(z.z_info.num_samples != 0);
 		EXIT_NOT_IMPLEMENTED(z.z_info.htile_acceleration != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.partially_resident != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
 	} else {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != 0x00000001 && z.z_info.format != 0x00000003);
+		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kZ16 &&
+		                     z.z_info.format != Prospero::DepthFormat::kZ32F);
 		if (z.z_info.num_samples != 0x00000000) {
 			static bool logged = false;
 			if (!logged) {
@@ -512,16 +517,17 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
 	}
 
-	if (z.stencil_info.format == 0) {
+	if (z.stencil_info.format == Prospero::StencilFormat::kInvalid) {
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
 	} else {
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.format != 0x00000001);
+		EXIT_NOT_IMPLEMENTED(z.stencil_info.format != Prospero::StencilFormat::k8UInt);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
 	}
 
-	if (z.z_info.format != 0 || z.stencil_info.format != 0) {
+	if (z.z_info.format != Prospero::DepthFormat::kInvalid ||
+	    z.stencil_info.format != Prospero::StencilFormat::kInvalid) {
 
 		EXIT_NOT_IMPLEMENTED(z.depth_info.addr5_swizzle_mask != 0x00000000);
 		EXIT_NOT_IMPLEMENTED(z.depth_info.array_mode != 0x00000000);
@@ -556,7 +562,8 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 		EXIT_NOT_IMPLEMENTED(z.z_read_base_addr != z.z_write_base_addr);
 		EXIT_NOT_IMPLEMENTED(z.stencil_read_base_addr != z.stencil_write_base_addr);
 		EXIT_NOT_IMPLEMENTED(z.z_write_base_addr == 0);
-		if (z.stencil_info.format != 0 && z.stencil_write_base_addr == 0) {
+		if (z.stencil_info.format != Prospero::StencilFormat::kInvalid &&
+		    z.stencil_write_base_addr == 0) {
 			LOGF("\t warning: stencil format is set without a stencil base address, continuing "
 			     "without stencil attachment\n");
 		}
@@ -686,14 +693,6 @@ 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 +1066,24 @@ 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;
 		}
@@ -1219,8 +1224,9 @@ void hw_check(const RenderCommandBuffer& buffer) {
 			const auto& mrt = hw.GetRenderTarget(i);
 			LOGF("\t RT%u addr=0x%010" PRIx64 " mask=0x%x fmt=0x%08" PRIx32
 			     " width=%u height=%u tile=%u\n",
-			     i, mrt.base.addr, (hw.GetRenderTargetMask() >> (i * 4u)) & 0x0fu, mrt.info.format,
-			     mrt.attrib2.width + 1, mrt.attrib2.height + 1, mrt.attrib3.tile_mode);
+			     i, mrt.base.addr, (hw.GetRenderTargetMask() >> (i * 4u)) & 0x0fu,
+			     static_cast<uint32_t>(mrt.info.format), mrt.attrib2.width + 1,
+			     mrt.attrib2.height + 1, static_cast<uint32_t>(mrt.attrib3.tile_mode));
 		}
 	}
 	if (rc.depth_clear_enable && hw.GetDepthClearValue() != 0.0f &&

src/graphics/host_gpu/renderer/debug.h

@@ -34,7 +34,7 @@ void                     uc_print(const char* func, const HW::UserConfig& uc);
 void                     uc_check(const HW::UserConfig& uc);
 void                     sh_print(const char* func, const HW::Shader& uc);
 std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt);
-bool                     RenderIsColorTileModeLinear(uint32_t tile_mode);
+bool                     RenderIsColorTileModeLinear(Prospero::TileMode tile_mode);
 void                     hw_print(const RenderCommandBuffer& buffer);
 void                     hw_check(const RenderCommandBuffer& buffer);
 void                     LogDrawPhase(const char* draw_name, const char* phase);

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;
@@ -49,14 +52,23 @@ 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);
 	}
 }
 
 static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
-	return fail == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp) ||
-	       pass == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp) ||
-	       depth_fail == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp);
+	constexpr auto replace_op = static_cast<uint8_t>(Prospero::StencilOp::kReplaceOp);
+	return fail == replace_op || pass == replace_op || depth_fail == replace_op;
 }
 
 [[nodiscard]] static vk::Format ResolveHostDepthAttachmentFormat(const RenderCommandBuffer& buffer,
@@ -90,15 +102,14 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
                                               RenderDepthInfo& r) {
 	KYTY_PROFILER_FUNCTION();
 	(void)submit_id;
-	const auto& hw = buffer.GetRegisters();
-	const auto& z  = hw.GetDepthRenderTarget();
-	const auto& rc = hw.GetRenderControl();
-	const auto& dc = hw.GetDepthControl();
-	const auto& sc = hw.GetStencilControl();
-	const auto& sm = hw.GetStencilMask();
-	const bool  has_stencil =
-	    z.stencil_info.format != Prospero::GpuEnumValue(Prospero::StencilFormat::kInvalid);
-	const bool depth_active =
+	const auto& hw          = buffer.GetRegisters();
+	const auto& z           = hw.GetDepthRenderTarget();
+	const auto& rc          = hw.GetRenderControl();
+	const auto& dc          = hw.GetDepthControl();
+	const auto& sc          = hw.GetStencilControl();
+	const auto& sm          = hw.GetStencilMask();
+	const bool  has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
+	const bool  depth_active =
 	    dc.z_enable || dc.z_write_enable || dc.depth_bounds_enable || rc.depth_clear_enable;
 	const bool stencil_active = has_stencil && (dc.stencil_enable || rc.stencil_clear_enable);
 	if (!depth_active && !stencil_active) {
@@ -109,9 +120,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		DepthFatal("invalid PS5 depth texture-compatibility encoding");
 	}
 	const bool attachment_unbound =
-	    z.z_info.format == Prospero::GpuEnumValue(Prospero::DepthFormat::kInvalid) &&
-	    z.stencil_info.format == Prospero::GpuEnumValue(Prospero::StencilFormat::kInvalid) &&
-	    z.z_info.num_samples == 0 &&
+	    z.z_info.format == Prospero::DepthFormat::kInvalid &&
+	    z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
 	    z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
 	    !z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
 	    z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
@@ -173,8 +183,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		DepthFatal("unsupported depth register state");
 	}
 	if (has_stencil) {
-		if (z.stencil_info.format != Prospero::GpuEnumValue(Prospero::StencilFormat::k8UInt) ||
-		    !htile_stencil_compat || z.stencil_read_base_addr == 0 ||
+		if (z.stencil_info.format != Prospero::StencilFormat::k8UInt || !htile_stencil_compat ||
+		    z.stencil_read_base_addr == 0 ||
 		    z.stencil_write_base_addr != z.stencil_read_base_addr ||
 		    (z.stencil_read_base_addr & 0xffffu) != 0 || z.depth_view.stencil_write_disable) {
 			DepthFatal("unsupported stencil attachment state");
@@ -192,8 +202,6 @@ 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;
@@ -218,7 +226,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		DepthFatal("no host depth/stencil format supports required usage for %s",
 		           VulkanToString(ideal_format).c_str());
 	}
-	const uint32_t guest_format = Prospero::GpuEnumValue(policy->guest_format);
+	const auto     guest_format = policy->guest_format;
 	const uint32_t bytes        = policy->bytes_per_element;
 	const auto     pitch        = TileGetDepthPitch(width, bytes, z.z_info.num_samples);
 	if (z.pitch_height_valid && ((static_cast<uint64_t>(z.pitch_div8_minus1) + 1u) * 8u != pitch ||
@@ -283,6 +291,8 @@ 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 +304,18 @@ 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;
@@ -331,7 +340,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	desc.info.pitch           = pitch;
 	desc.info.bytes_per_block = bytes;
 	desc.info.samples         = samples;
-	desc.info.tile_mode       = Prospero::GpuEnumValue(Prospero::TileMode::kDepth);
+	desc.info.tile_mode       = Prospero::TileMode::kDepth;
 	desc.info.mip_layout[0]   = {0, r.depth_buffer_size, pitch, height};
 	desc.info.metadata.range  = {r.htile_buffer_vaddr, r.htile_buffer_size};
 	desc.info.metadata.kind   = has_htile ? ImageMetadataKind::Htile : ImageMetadataKind::None;

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

@@ -646,13 +646,13 @@ void Validate(const ImageInfo& info) {
 	}
 }
 
-uint32_t RenderTargetTransferFormat(uint32_t bytes_per_element) {
+Prospero::BufferFormat RenderTargetTransferFormat(uint32_t bytes_per_element) {
 	switch (bytes_per_element) {
-		case 1: return Prospero::GpuEnumValue(Prospero::BufferFormat::k8UNorm);
-		case 2: return Prospero::GpuEnumValue(Prospero::BufferFormat::k16UNorm);
-		case 4: return Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float);
-		case 8: return Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float);
-		case 16: return Prospero::GpuEnumValue(Prospero::BufferFormat::k32_32_32_32Float);
+		case 1: return Prospero::BufferFormat::k8UNorm;
+		case 2: return Prospero::BufferFormat::k16UNorm;
+		case 4: return Prospero::BufferFormat::k32Float;
+		case 8: return Prospero::BufferFormat::k16_16_16_16Float;
+		case 16: return Prospero::BufferFormat::k32_32_32_32Float;
 		default: EXIT("unsupported render-target element size: %u\n", bytes_per_element);
 	}
 }
@@ -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/image.h

@@ -188,8 +188,8 @@ private:
 
 namespace ImageOps {
 
-void                   Validate(const ImageInfo& info);
-[[nodiscard]] uint32_t RenderTargetTransferFormat(uint32_t bytes_per_element);
+void                                 Validate(const ImageInfo& info);
+[[nodiscard]] Prospero::BufferFormat RenderTargetTransferFormat(uint32_t bytes_per_element);
 
 } // namespace ImageOps
 

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

@@ -67,14 +67,14 @@ struct ImageInfo {
 	ImageMetadataInfo            metadata;
 	uint32_t                     htile_clear_mask = UINT32_MAX;
 	vk::Format                   pixel_format     = vk::Format::eUndefined;
-	uint32_t                     guest_format     = 0;
+	Prospero::BufferFormat       guest_format     = Prospero::BufferFormat::kInvalid;
 	Prospero::ImageType          type             = Prospero::ImageType::kColor2D;
 	vk::Extent3D                 extent           = {1, 1, 1};
 	ImageSubresources            resources;
 	uint32_t                     pitch           = 0;
 	uint32_t                     bytes_per_block = 0;
 	uint32_t                     samples         = 1;
-	uint32_t                     tile_mode       = 0;
+	Prospero::TileMode           tile_mode       = Prospero::TileMode::kLinear;
 	bool                         bgra16          = false;
 	std::array<ImageMipInfo, 16> mip_layout {};
 
@@ -86,9 +86,7 @@ struct ImageInfo {
 	[[nodiscard]] bool IsBlock() const noexcept {
 		return Prospero::BlockCompressedBytesPerBlock(guest_format) != 0;
 	}
-	[[nodiscard]] bool IsTiled() const noexcept {
-		return tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kLinear);
-	}
+	[[nodiscard]] bool IsTiled() const noexcept { return tile_mode != Prospero::TileMode::kLinear; }
 	[[nodiscard]] constexpr bool IsVolume() const noexcept {
 		return type == Prospero::ImageType::kColor3D;
 	}
@@ -235,9 +233,9 @@ inline constexpr std::array<DepthFormatPolicy, 2> DEPTH_FORMAT_POLICIES {{
 }};
 
 [[nodiscard]] inline constexpr const DepthFormatPolicy*
-FindDepthFormatPolicy(uint32_t depth_format) noexcept {
+FindDepthFormatPolicy(Prospero::DepthFormat depth_format) noexcept {
 	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (Prospero::GpuEnumValue(policy.depth_format) == depth_format) {
+		if (policy.depth_format == depth_format) {
 			return &policy;
 		}
 	}
@@ -245,9 +243,9 @@ FindDepthFormatPolicy(uint32_t depth_format) noexcept {
 }
 
 [[nodiscard]] inline constexpr const DepthFormatPolicy*
-FindGuestDepthFormatPolicy(uint32_t guest_format) noexcept {
+FindGuestDepthFormatPolicy(Prospero::BufferFormat guest_format) noexcept {
 	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (Prospero::GpuEnumValue(policy.guest_format) == guest_format) {
+		if (policy.guest_format == guest_format) {
 			return &policy;
 		}
 	}
@@ -269,10 +267,11 @@ FindGuestDepthFormatPolicy(uint32_t guest_format) noexcept {
 	return has_stencil ? policy.stencil_attachment_formats.front() : policy.depth_attachment_format;
 }
 
-[[nodiscard]] inline constexpr vk::Format DepthAttachmentFormat(uint32_t depth_format,
-                                                                uint32_t stencil_format) noexcept {
+[[nodiscard]] inline constexpr vk::Format
+DepthAttachmentFormat(Prospero::DepthFormat   depth_format,
+                      Prospero::StencilFormat stencil_format) noexcept {
 	bool has_stencil = false;
-	switch (static_cast<Prospero::StencilFormat>(stencil_format)) {
+	switch (stencil_format) {
 		case Prospero::StencilFormat::kInvalid: break;
 		case Prospero::StencilFormat::k8UInt: has_stencil = true; break;
 		default: return vk::Format::eUndefined;
@@ -318,9 +317,9 @@ inline bool ImageInfo::IsDepth() const noexcept {
 	return std::bit_cast<uint32_t>(static_cast<float>(value) / 65535.0f);
 }
 
-[[nodiscard]] inline constexpr bool IsSupportedSampledDepthFormat(vk::Format image_format,
-                                                                  uint32_t   guest_format,
-                                                                  vk::Format view_format) noexcept {
+[[nodiscard]] inline constexpr bool
+IsSupportedSampledDepthFormat(vk::Format image_format, Prospero::BufferFormat guest_format,
+                              vk::Format view_format) noexcept {
 	const auto* policy = FindGuestDepthFormatPolicy(guest_format);
 	return policy != nullptr && view_format == policy->sampled_view_format &&
 	       (image_format == policy->depth_attachment_format ||
@@ -330,8 +329,7 @@ inline bool ImageInfo::IsDepth() const noexcept {
 [[nodiscard]] inline constexpr bool IsSupportedSampledDepthFormat(vk::Format image_format,
                                                                   vk::Format view_format) noexcept {
 	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (IsSupportedSampledDepthFormat(image_format, Prospero::GpuEnumValue(policy.guest_format),
-		                                  view_format)) {
+		if (IsSupportedSampledDepthFormat(image_format, policy.guest_format, view_format)) {
 			return true;
 		}
 	}
@@ -383,10 +381,10 @@ CanUseVideoOutNativeWithoutUpload(VideoOutCompression compression, bool render_t
 }
 
 struct VideoOutPixelFormatInfo {
-	vk::Format format            = vk::Format::eUndefined;
-	uint32_t   guest_format      = 0;
-	uint32_t   bytes_per_element = 0;
-	bool       bgra16            = false;
+	vk::Format             format            = vk::Format::eUndefined;
+	Prospero::BufferFormat guest_format      = Prospero::BufferFormat::kInvalid;
+	uint32_t               bytes_per_element = 0;
+	bool                   bgra16            = false;
 };
 
 struct VideoOutFormatPolicy {
@@ -396,23 +394,17 @@ struct VideoOutFormatPolicy {
 
 inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {{
     {0x8000000022000000ull,
-     {vk::Format::eR8G8B8A8Srgb, Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8Srgb), 4,
-      false}},
+     {vk::Format::eR8G8B8A8Srgb, Prospero::BufferFormat::k8_8_8_8Srgb, 4, false}},
     {0x8000000000000000ull,
-     {vk::Format::eB8G8R8A8Srgb, Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8Srgb), 4,
-      false}},
+     {vk::Format::eB8G8R8A8Srgb, Prospero::BufferFormat::k8_8_8_8Srgb, 4, false}},
     {0x8100000022000000ull,
-     {vk::Format::eA2B10G10R10UnormPack32,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k10_10_10_2UNorm), 4, false}},
+     {vk::Format::eA2B10G10R10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0x8100000000000000ull,
-     {vk::Format::eA2R10G10B10UnormPack32,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k10_10_10_2UNorm), 4, false}},
+     {vk::Format::eA2R10G10B10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0xc001000622000000ull,
-     {vk::Format::eR16G16B16A16Sfloat,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float), 8, false}},
+     {vk::Format::eR16G16B16A16Sfloat, Prospero::BufferFormat::k16_16_16_16Float, 8, false}},
     {0xc001000600000000ull,
-     {vk::Format::eR16G16B16A16Sfloat,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float), 8, true}},
+     {vk::Format::eR16G16B16A16Sfloat, Prospero::BufferFormat::k16_16_16_16Float, 8, true}},
 }};
 
 [[nodiscard]] inline bool DecodeVideoOutPixelFormat(uint64_t                 pixel_format,
@@ -439,15 +431,15 @@ inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {
 }
 
 [[nodiscard]] inline constexpr bool
-IsSupportedDisplayRenderTargetTileMode(uint32_t tile_mode) noexcept {
-	return tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
+IsSupportedDisplayRenderTargetTileMode(Prospero::TileMode tile_mode) noexcept {
+	return tile_mode == Prospero::TileMode::kRenderTarget;
 }
 
 [[nodiscard]] inline constexpr bool IsSupportedStandard64RenderTarget(const ImageInfo& info) {
-	if (info.tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB) ||
-	    info.data.address == 0 || (info.data.address & 0xffffu) != 0 || info.extent.width == 0 ||
-	    info.extent.height == 0 || info.bytes_per_block != 4 || info.resources.levels != 1 ||
-	    info.resources.layers != 1 || info.samples != 1) {
+	if (info.tile_mode != Prospero::TileMode::kStandard64KB || info.data.address == 0 ||
+	    (info.data.address & 0xffffu) != 0 || info.extent.width == 0 || info.extent.height == 0 ||
+	    info.bytes_per_block != 4 || info.resources.levels != 1 || info.resources.layers != 1 ||
+	    info.samples != 1) {
 		return false;
 	}
 	const auto expected_pitch =
@@ -460,7 +452,7 @@ IsSupportedDisplayRenderTargetTileMode(uint32_t tile_mode) noexcept {
 }
 
 [[nodiscard]] inline constexpr bool IsTiledRenderTarget(const ImageInfo& info) noexcept {
-	return info.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
+	return info.tile_mode == Prospero::TileMode::kRenderTarget ||
 	       IsSupportedStandard64RenderTarget(info);
 }
 

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

@@ -14,164 +14,71 @@
 namespace Libs::Graphics {
 namespace {
 
-struct RenderTargetFormatMapping {
-	Prospero::ChannelLayout layout;
-	Prospero::ChannelType   type;
-	Prospero::ChannelOrder  order;
-	RenderTargetFormatInfo  info;
+// Rows are channel order; columns are the number of physical components minus one. Unused
+// selectors complete each entry to a permutation so logical write masks can be inverted.
+constexpr Prospero::ColorComponentMapping kRenderTargetColorMappings[4][4] = {
+    {Prospero::ColorMappingRgba, Prospero::ColorMappingRgba, Prospero::ColorMappingRgba,
+     Prospero::ColorMappingRgba},
+    {Prospero::ColorMappingGr, Prospero::ColorMappingRabg, Prospero::ColorMappingRgab,
+     Prospero::ColorMappingBgra},
+    {Prospero::ColorMappingBgra, Prospero::ColorMappingGr, Prospero::ColorMappingBgra,
+     Prospero::ColorMappingAbgr},
+    {Prospero::ColorMappingAgba, Prospero::ColorMappingArbg, Prospero::ColorMappingAgbr,
+     Prospero::ColorMappingArgb},
 };
 
-constexpr RenderTargetFormatMapping kRenderTargetFormats[] = {
-    {Prospero::ChannelLayout::k8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8Unorm, 2}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Unorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Snorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSrgb,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Srgb, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Unorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Snorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSrgb,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Srgb, 4}},
-    {Prospero::ChannelLayout::k5_5_5_1,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR5G5B5A1UnormPack16, 2}},
-    {Prospero::ChannelLayout::k4_4_4_4,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kReversed,
-     {vk::Format::eB4G4R4A4UnormPack16, 2}},
-    {Prospero::ChannelLayout::k10_10_10_2,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eA2B10G10R10UnormPack32, 4}},
-    {Prospero::ChannelLayout::k10_10_10_2,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eA2R10G10B10UnormPack32, 4}},
-    {Prospero::ChannelLayout::k11_11_10,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eB10G11R11UfloatPack32, 4}},
-    {Prospero::ChannelLayout::k5_6_5,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eB5G6R5UnormPack16, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Unorm, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Uint, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Sfloat, 2}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Unorm, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Snorm, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Uint, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Sfloat, 4}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Unorm, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Uint, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Sfloat, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eR16G16B16A16Sfloat, 8, Prospero::ColorMappingBgra}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kReversed,
-     {vk::Format::eR16G16B16A16Sfloat, 8, Prospero::ColorMappingAbgr}},
-    {Prospero::ChannelLayout::k32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32Sfloat, 4}},
-    {Prospero::ChannelLayout::k32_32,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32Uint, 8}},
-    {Prospero::ChannelLayout::k32_32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32Sfloat, 8}},
-    {Prospero::ChannelLayout::k32_32_32_32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32B32A32Sfloat, 16}},
+struct RenderTargetHostFormatInfo {
+	vk::Format                      format = vk::Format::eUndefined;
+	Prospero::ColorComponentMapping host_to_storage;
 };
 
+RenderTargetHostFormatInfo ResolveRenderTargetHostFormat(Prospero::BufferFormat guest_format,
+                                                         Prospero::ChannelOrder order) {
+	if (order == Prospero::ChannelOrder::kAlt) {
+		switch (guest_format) {
+			case Prospero::BufferFormat::k8_8_8_8UNorm:
+				return {vk::Format::eB8G8R8A8Unorm, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k8_8_8_8SNorm:
+				return {vk::Format::eB8G8R8A8Snorm, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k8_8_8_8Srgb:
+				return {vk::Format::eB8G8R8A8Srgb, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k10_10_10_2UNorm:
+				return {vk::Format::eA2R10G10B10UnormPack32, Prospero::ColorMappingBgra};
+			default: break;
+		}
+	}
+	switch (guest_format) {
+		case Prospero::BufferFormat::k5_5_5_1UNorm:
+			return {vk::Format::eA1R5G5B5UnormPack16, Prospero::ColorMappingBgra};
+		case Prospero::BufferFormat::k1_5_5_5UNorm:
+			return {vk::Format::eR5G5B5A1UnormPack16, Prospero::ColorMappingAbgr};
+		case Prospero::BufferFormat::k4_4_4_4UNorm:
+			return {vk::Format::eR4G4B4A4UnormPack16, Prospero::ColorMappingAbgr};
+		default: return {VulkanFormat(guest_format), {}};
+	}
+}
+
 } // namespace
 
-// TODO: cleanup!
-RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_t raw_type,
-                                                    uint32_t raw_order) {
-	const auto layout = static_cast<Prospero::ChannelLayout>(raw_layout);
-	const auto type   = static_cast<Prospero::ChannelType>(raw_type);
-	const auto order  = static_cast<Prospero::ChannelOrder>(raw_order);
-
-	if (layout == Prospero::ChannelLayout::k8 && type == Prospero::ChannelType::kUNorm &&
-	    raw_order <= Prospero::GpuEnumValue(Prospero::ChannelOrder::kAltReversed)) {
-		return {vk::Format::eR8Unorm, 1};
-	}
-	for (const auto& mapping: kRenderTargetFormats) {
-		if (mapping.layout == layout && mapping.type == type && mapping.order == order) {
-			return mapping.info;
+RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
+                                                    Prospero::ChannelType   type,
+                                                    Prospero::ChannelOrder  order) {
+	const auto encoding = Prospero::ResolveRenderTargetFormat(layout, type);
+	if (encoding.IsValid() && encoding.SupportsOrder(order)) {
+		const auto host  = ResolveRenderTargetHostFormat(encoding.buffer_format, order);
+		const auto bytes = Prospero::RenderTargetBytesPerElement(encoding.buffer_format);
+		if (host.format != vk::Format::eUndefined && bytes != 0) {
+			const auto order_mapping =
+			    kRenderTargetColorMappings[static_cast<size_t>(order)][encoding.components - 1u];
+			return {host.format, bytes, host.host_to_storage.Then(order_mapping)};
 		}
 	}
-	EXIT("unsupported render-target format combination: layout=%u type=%u order=%u\n", raw_layout,
-	     raw_type, raw_order);
+	EXIT("unsupported render-target format combination: layout=%u type=%u order=%u\n",
+	     static_cast<uint32_t>(layout), static_cast<uint32_t>(type), static_cast<uint32_t>(order));
 }
 
 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 +91,6 @@ 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;
@@ -226,23 +119,22 @@ vk::ComponentMapping TextureGetComponentMapping(uint32_t swizzle) {
 	return components;
 }
 
-vk::Format TextureGetFormat(uint32_t fmt) {
-	const auto vk_format = VulkanFormat(fmt);
+vk::Format TextureGetFormat(Prospero::BufferFormat format) {
+	const auto vk_format = VulkanFormat(format);
 	if (vk_format != vk::Format::eUndefined) {
 		return vk_format;
 	}
-	EXIT("unknown format: fmt = %u\n", fmt);
+	EXIT("unknown format: fmt = %u\n", static_cast<uint32_t>(format));
 	return vk::Format::eUndefined;
 }
 
 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 +147,26 @@ 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 +182,126 @@ 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,
+TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
+                                            uint32_t height, uint32_t levels, uint32_t depth,
+                                            Prospero::TileMode 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 = {
+	    format,
+	    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   = 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 (format == Prospero::BufferFormat::kInvalid) {
 		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, static_cast<uint32_t>(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(format, element)) {
+				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner,
+				     static_cast<uint32_t>(format));
+			}
+			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>(format),
+				     static_cast<uint32_t>(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>(format),
+				     static_cast<uint32_t>(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(format, width, description.tile_mode);
+	if (tile_mode == Prospero::TileMode::kLinear) {
+		TileSizeOffset level_sizes[16] {};
+		TilePaddedSize padded_sizes[16] {};
+		TileGetTextureSize(format, 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 && 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       = 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 +320,6 @@ 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 +329,32 @@ 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 +369,43 @@ 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 +414,36 @@ 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 +19,35 @@ 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,
+vk::Format             TextureGetFormat(Prospero::BufferFormat format);
+RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
+                                                    Prospero::ChannelType   type,
+                                                    Prospero::ChannelOrder  order);
+TextureUploadLayout    TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
+                                               uint32_t height, uint32_t levels, uint32_t depth,
+                                               Prospero::TileMode 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);
+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

@@ -59,7 +59,7 @@ static void CopyNativeDescriptor(const ShaderRecompiler::IR::DescriptorValue& so
 }
 
 static Prospero::ImageType TextureType(const ShaderTextureResource& descriptor) {
-	const auto type = static_cast<Prospero::ImageType>(descriptor.Type());
+	const auto type = descriptor.Type();
 	return type == Prospero::ImageType::kCube ? Prospero::ImageType::kColor2DArray : type;
 }
 
@@ -227,19 +227,19 @@ bool IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& r
 	return image.usage.video_out && image.info.resources.layers == 1 &&
 	       IsSupportedSampledColorResource(resource) &&
 	       resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D &&
-	       descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor2D) &&
-	       descriptor.Depth() == 0 && descriptor.BaseArray5() == 0;
+	       descriptor.Type() == Prospero::ImageType::kColor2D && descriptor.Depth() == 0 &&
+	       descriptor.BaseArray5() == 0;
 }
 
 bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, const Image& image) {
 	const auto width        = static_cast<uint32_t>(descriptor.Width5()) + 1u;
 	const auto height       = static_cast<uint32_t>(descriptor.Height5()) + 1u;
-	const auto type         = static_cast<Prospero::ImageType>(descriptor.Type());
+	const auto type         = descriptor.Type();
 	const bool multisampled = IsMultisampledTexture(type);
 	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;
@@ -269,9 +269,8 @@ bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, c
 	       (supported_2d || supported_array || supported_cube || supported_msaa_2d ||
 	        supported_msaa_array) &&
 	       levels_ok && descriptor.MinLod() == 0 &&
-	       descriptor.TileMode() == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
-	       descriptor.BCSwizzle() == 0 && (!descriptor.MsaaDepth() || multisampled) &&
-	       pitch >= width && pitch == image.info.pitch;
+	       descriptor.TileMode() == Prospero::TileMode::kDepth && descriptor.BCSwizzle() == 0 &&
+	       (!descriptor.MsaaDepth() || multisampled) && pitch >= width && pitch == image.info.pitch;
 }
 
 bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, const Image& image) {
@@ -283,13 +282,13 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, co
 	                                     (static_cast<uint32_t>(descriptor.TileMode()) << 20u) |
 	                                     (static_cast<uint32_t>(descriptor.Type()) << 28u);
 	const uint32_t     field4_expected = descriptor.Depth() | (descriptor.BaseArray5() << 16u);
-	const uint32_t     field5_expected =
-	    0x00700000u | (static_cast<uint32_t>(descriptor.MaxMip()) << 4u);
-	const bool common = (descriptor.fields[1] & field1_reserved_mask) == 0 &&
-	                    (descriptor.fields[2] & field2_reserved_mask) == 0 &&
-	                    descriptor.fields[3] == field3_expected &&
-	                    descriptor.fields[4] == field4_expected &&
-	                    descriptor.fields[5] == field5_expected;
+	const uint32_t     field5_expected = (static_cast<uint32_t>(descriptor.PerfMod5()) << 20u) |
+	                                     (static_cast<uint32_t>(descriptor.MaxMip()) << 4u);
+	const bool         common          = (descriptor.fields[1] & field1_reserved_mask) == 0 &&
+	                                     (descriptor.fields[2] & field2_reserved_mask) == 0 &&
+	                                     descriptor.fields[3] == field3_expected &&
+	                                     descriptor.fields[4] == field4_expected &&
+	                                     descriptor.fields[5] == field5_expected;
 	if (!common || (descriptor.fields[6] == 0 && descriptor.fields[7] != 0)) {
 		return false;
 	}
@@ -300,8 +299,8 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, co
 	const uint32_t expected_control  = htile_control | (descriptor.MsaaDepth() ? (1u << 10u) : 0u);
 	const auto     metadata_addr     = descriptor.MetaAddr() << 8u;
 	return (descriptor.fields[6] & 0x00ffffffu) == expected_control && metadata_addr != 0 &&
-	       descriptor.TileMode() == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
-	       image.info.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
+	       descriptor.TileMode() == Prospero::TileMode::kDepth &&
+	       image.info.tile_mode == Prospero::TileMode::kDepth &&
 	       image.info.metadata.kind == ImageMetadataKind::Htile &&
 	       image.info.metadata.range.Valid() && image.info.metadata.range.address == metadata_addr;
 }
@@ -321,7 +320,7 @@ 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 "
@@ -332,23 +331,21 @@ static void ValidateDepthTargetBinding(const ShaderRecompiler::IR::ImageResource
 	     resource_ok, descriptor_ok, encoding_ok, format_ok, static_cast<uint32_t>(resource.kind),
 	     static_cast<uint32_t>(resource.dimension), static_cast<uint32_t>(resource.mip_mode),
 	     resource.read, resource.written, resource.atomic, resource.depth_compare,
-	     descriptor.Format(), descriptor.DstSelXYZW(),
+	     static_cast<uint32_t>(descriptor.Format()), descriptor.DstSelXYZW(),
 	     image == nullptr ? static_cast<int>(vk::Format::eUndefined)
 	                      : static_cast<int>(image->info.pixel_format),
 	     static_cast<int>(view_format), image == nullptr ? 0u : image->info.resources.layers,
-	     descriptor.Type(), descriptor.BaseArray5(), descriptor.Depth(), descriptor_pitch,
-	     image == nullptr ? 0u : image->info.pitch, descriptor.Base40(), size, descriptor.fields[0],
-	     descriptor.fields[1], descriptor.fields[2], descriptor.fields[3], descriptor.fields[4],
-	     descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
+	     static_cast<uint32_t>(descriptor.Type()), descriptor.BaseArray5(), descriptor.Depth(),
+	     descriptor_pitch, image == nullptr ? 0u : image->info.pitch, descriptor.Base40(), size,
+	     descriptor.fields[0], descriptor.fields[1], descriptor.fields[2], descriptor.fields[3],
+	     descriptor.fields[4], descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
 }
 
 static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::ImageResource& resource,
                                                 const ShaderTextureResource& descriptor) {
-	const auto tile = descriptor.TileMode();
-	const bool is_color_1d =
-	    descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor1D);
-	const bool is_color_1d_array =
-	    descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor1DArray);
+	const auto tile              = descriptor.TileMode();
+	const bool is_color_1d       = descriptor.Type() == Prospero::ImageType::kColor1D;
+	const bool is_color_1d_array = descriptor.Type() == Prospero::ImageType::kColor1DArray;
 	const bool valid_1d_slice =
 	    (is_color_1d && descriptor.Depth() == 0 && descriptor.BaseArray5() == 0) ||
 	    (is_color_1d_array && descriptor.BaseArray5() <= descriptor.Depth());
@@ -358,36 +355,51 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim1DArray &&
 	    is_color_1d_array && descriptor.Height5() == 0 &&
 	    descriptor.BaseArray5() <= descriptor.Depth();
-	const bool is_color_2d =
-	    descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor2D);
-	const bool is_color_2d_array =
-	    descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor2DArray);
+	const bool is_color_2d       = descriptor.Type() == Prospero::ImageType::kColor2D;
+	const bool is_color_2d_array = descriptor.Type() == Prospero::ImageType::kColor2DArray;
 	const bool valid_2d_slice =
 	    (is_color_2d && descriptor.Depth() == 0 && descriptor.BaseArray5() == 0) ||
 	    (is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth());
 	const bool is_2d =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D && valid_2d_slice;
+	const bool is_cube = resource.cube && descriptor.Type() == Prospero::ImageType::kCube &&
+	                     descriptor.Width5() == descriptor.Height5() &&
+	                     descriptor.BaseArray5() <= descriptor.Depth() &&
+	                     (descriptor.Depth() - descriptor.BaseArray5() + 1u) % 6u == 0;
 	const bool is_2d_array =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray &&
-	    is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth();
+	    ((!resource.cube && is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth()) ||
+	     is_cube);
 	const bool is_3d = resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D &&
-	                   descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor3D) &&
+	                   descriptor.Type() == 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 (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);
@@ -423,14 +435,12 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	const bool encoding_ok   = IsSupportedStorageTextureEncoding(descriptor);
 	const bool uint_resource =
 	    resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint;
-	const bool raw_sint_storage =
-	    format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) && uint_resource &&
-	    resource.written && !resource.read && !resource.atomic;
+	const bool raw_sint_storage = format == Prospero::BufferFormat::k32SInt && uint_resource &&
+	                              resource.written && !resource.read && !resource.atomic;
 	const bool format_ok =
-	    raw_sint_storage ||
-	    (Prospero::IsSupportedTextureFormat(format) &&
-	     uint_resource == Prospero::IsUintTextureFormat(format) &&
-	     (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt)));
+	    raw_sint_storage || (Prospero::IsSupportedTextureFormat(format) &&
+	                         uint_resource == Prospero::IsUintTextureFormat(format) &&
+	                         (!resource.atomic || format == Prospero::BufferFormat::k32UInt));
 	if (resource_ok && descriptor_ok && encoding_ok && format_ok && size != 0) {
 		return;
 	}
@@ -449,24 +459,23 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	     IsValidImageSwizzle(descriptor.DstSelXYZW()), descriptor.Base40(), size,
 	     static_cast<uint32_t>(descriptor.Width5()) + 1u,
 	     static_cast<uint32_t>(descriptor.Height5()) + 1u,
-	     static_cast<uint32_t>(descriptor.Depth()) + 1u, descriptor.Type(), format,
-	     descriptor.TileMode(), descriptor.DstSelXYZW(), resource.read, resource.written,
-	     descriptor.fields[0], descriptor.fields[1], descriptor.fields[2], descriptor.fields[3],
-	     descriptor.fields[4], descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
+	     static_cast<uint32_t>(descriptor.Depth()) + 1u, static_cast<uint32_t>(descriptor.Type()),
+	     static_cast<uint32_t>(format), static_cast<uint32_t>(descriptor.TileMode()),
+	     descriptor.DstSelXYZW(), resource.read, resource.written, descriptor.fields[0],
+	     descriptor.fields[1], descriptor.fields[2], descriptor.fields[3], descriptor.fields[4],
+	     descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
 }
 
 struct NullImageSpec {
-	vk::Format format;
-	uint32_t   guest_format;
+	vk::Format             format;
+	Prospero::BufferFormat guest_format;
 };
 
 static NullImageSpec NullTextureSpec(const ShaderRecompiler::IR::ImageResource& resource) {
 	const bool uint_image = resource.kind == ShaderRecompiler::IR::ResourceKind::ImageUint ||
 	                        resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint;
-	return uint_image ? NullImageSpec {vk::Format::eR32Uint,
-	                                   Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt)}
-	                  : NullImageSpec {vk::Format::eR32Sfloat,
-	                                   Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float)};
+	return uint_image ? NullImageSpec {vk::Format::eR32Uint, Prospero::BufferFormat::k32UInt}
+	                  : NullImageSpec {vk::Format::eR32Sfloat, Prospero::BufferFormat::k32Float};
 }
 
 static TextureCache::ImageDesc NullTextureDesc(const ShaderRecompiler::IR::ImageResource& resource,
@@ -492,19 +501,21 @@ 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)) {
+	if (info.IsVolume() && info.tile_mode != Prospero::TileMode::kLinear) {
+		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 +523,7 @@ 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 =
@@ -617,10 +628,9 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	const bool multisampled = IsMultisampledTexture(type);
 	const auto levels       = multisampled ? 1u : static_cast<uint32_t>(descriptor.MaxMip()) + 1u;
 	const auto tile         = descriptor.TileMode();
-	const bool depth_tile   = tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth);
-	const bool msaa_tile =
-	    depth_tile || tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
-	const bool msaa_array = type == Prospero::ImageType::kColor2DMsaaArray;
+	const bool depth_tile   = tile == Prospero::TileMode::kDepth;
+	const bool msaa_tile    = depth_tile || tile == Prospero::TileMode::kRenderTarget;
+	const bool msaa_array   = type == Prospero::ImageType::kColor2DMsaaArray;
 	if ((!multisampled && (base_level > last_level || last_level >= levels)) ||
 	    (multisampled &&
 	     (base_level != 0 || last_level == 0 || last_level > 3 ||
@@ -630,7 +640,8 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		EXIT("unsupported texture mip view: base=%u last=%u levels=%u max=%u type=%u tile=%u "
 		     "kind=%u dimension=%u mip_mode=%u read=%d written=%d "
 		     "dwords=%08x,%08x,%08x,%08x,%08x,%08x,%08x,%08x\n",
-		     base_level, last_level, levels, descriptor.MaxMip(), descriptor.Type(), tile,
+		     base_level, last_level, levels, descriptor.MaxMip(),
+		     static_cast<uint32_t>(descriptor.Type()), static_cast<uint32_t>(tile),
 		     static_cast<uint32_t>(resource.kind), static_cast<uint32_t>(resource.dimension),
 		     static_cast<uint32_t>(resource.mip_mode), resource.read, resource.written,
 		     descriptor.fields[0], descriptor.fields[1], descriptor.fields[2], descriptor.fields[3],
@@ -650,7 +661,7 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	     resource.kind == ShaderRecompiler::IR::ResourceKind::ImageUint) &&
 	    !sampled_numeric_class) {
 		EXIT("sampled image numeric class mismatch: kind=%u format=%u addr=0x%016" PRIx64 "\n",
-		     static_cast<uint32_t>(resource.kind), format, address);
+		     static_cast<uint32_t>(resource.kind), static_cast<uint32_t>(format), address);
 	}
 
 	const bool    volume       = type == Prospero::ImageType::kColor3D;
@@ -670,9 +681,9 @@ 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);
@@ -680,15 +691,14 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		ValidateStorageTexture(resource, descriptor, size.size);
 	}
 
-	const auto pixel_format = TextureGetFormat(format);
-	const auto storage_view_format =
-	    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt)
-	        ? vk::Format::eR32Uint
-	        : SrgbStorageViewFormat(pixel_format);
-	const auto              view_format = storage && storage_view_format != vk::Format::eUndefined
-	                                          ? storage_view_format
-	                                          : pixel_format;
-	const auto              block_bytes = Prospero::BlockCompressedBytesPerBlock(format);
+	const auto pixel_format        = TextureGetFormat(format);
+	const auto storage_view_format = storage && format == Prospero::BufferFormat::k32SInt
+	                                     ? vk::Format::eR32Uint
+	                                     : SrgbStorageViewFormat(pixel_format);
+	const auto view_format         = storage && storage_view_format != vk::Format::eUndefined
+	                                     ? storage_view_format
+	                                     : pixel_format;
+	const auto block_bytes         = Prospero::BlockCompressedBytesPerBlock(format);
 	TextureCache::ImageDesc desc {};
 	desc.info.data         = {address, size.size};
 	desc.info.pixel_format = pixel_format;
@@ -919,22 +929,21 @@ 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

@@ -24,6 +24,8 @@
 namespace Libs::Graphics {
 
 // IDK: maybe we can remove it?
+constexpr uint8_t kTemporaryVertexAttribFormat113 =
+    static_cast<uint8_t>(Prospero::VertexAttribFormat::k16_16SInt);
 constexpr uint32_t kTemporaryPs5BufferFormat121 = 121u;
 
 static bool NarrowInputFormat(vk::Format& format, uint32_t& size, uint32_t used_components) {
@@ -88,8 +90,9 @@ static bool NarrowInputFormat(vk::Format& format, uint32_t& size, uint32_t used_
 
 static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format, uint32_t& size,
                            uint32_t used_components) {
-	auto fmt = res.Format();
-	if (fmt == Prospero::GpuEnumValue(Prospero::VertexAttribFormat::k16_16SInt)) {
+	const auto fmt        = res.Format();
+	const auto raw_format = res.RawFormat();
+	if (raw_format == kTemporaryVertexAttribFormat113) {
 		static bool logged_113 = false;
 		if (!logged_113) {
 			LOGF("InputFormat: temporary: accepting invalid PS5 buffer format 113 as "
@@ -99,12 +102,12 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		format = vk::Format::eR32G32B32A32Sfloat;
 		size   = 4;
 		if (NarrowInputFormat(format, size, used_components)) {
-			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", fmt,
+			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
 			     VulkanToString(format).c_str(), used_components);
 		}
 		return;
 	}
-	if (fmt == kTemporaryPs5BufferFormat121) {
+	if (raw_format == kTemporaryPs5BufferFormat121) {
 		static bool logged_121 = false;
 		if (!logged_121) {
 			LOGF("InputFormat: accepting PS5 buffer format 121 as vk::Format::eR16G16Sfloat\n");
@@ -113,13 +116,13 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		format = vk::Format::eR16G16Sfloat;
 		size   = 2;
 		if (NarrowInputFormat(format, size, used_components)) {
-			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", fmt,
+			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
 			     VulkanToString(format).c_str(), used_components);
 		}
 		return;
 	}
 
-	switch (static_cast<Prospero::BufferFormat>(fmt)) {
+	switch (fmt) {
 		case Prospero::BufferFormat::k32_32_32_32Float:
 			format = vk::Format::eR32G32B32A32Sfloat;
 			size   = 4;
@@ -329,7 +332,7 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 			size   = 1;
 			break;
 		default:
-			EXIT("unknown format: fmt = %u\n", fmt);
+			EXIT("unknown format: fmt = %u\n", raw_format);
 			format = vk::Format::eUndefined;
 			size   = 4;
 			break;
@@ -947,15 +950,23 @@ 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

@@ -93,7 +93,7 @@ bool ResolveComputeImageClear(const ShaderComputeInputInfo& input, uint32_t grou
 	const auto  descriptor = DecodeNativeDescriptor<ShaderBufferResource>(raw);
 	if (!resource.formatted || !resource.written || resource.read || resource.atomic ||
 	    resource.scalar || resource.max_byte_extent != 16 || descriptor.Stride() != 16 ||
-	    descriptor.Format() != Prospero::GpuEnumValue(Prospero::BufferFormat::k32_32_32_32UInt) ||
+	    descriptor.Format() != Prospero::BufferFormat::k32_32_32_32UInt ||
 	    descriptor.SwizzleEnabled() || descriptor.IndexStride() != 0 || descriptor.AddTid() ||
 	    resource.packed_stride != descriptor.PackedStride() || raw.dword_count != 4 ||
 	    program.user_data_base != 0 || resources.user_data.size() != 8) {
@@ -246,7 +246,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 			LOGF("  CS buffer[%u]: source=%u usage=%s addr=0x%012" PRIx64
 			     " stride=%u records=%u format=%u\n",
 			     i, buffer.source, buffer.written ? "read-write" : "read-only", r.Base48(),
-			     r.Stride(), r.NumRecords(), r.Format());
+			     r.Stride(), r.NumRecords(), r.RawFormat());
 		}
 		for (uint32_t i = 0; i < program.info.images.size(); i++) {
 			const auto& image = program.info.images[i];
@@ -258,13 +258,13 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 			      image.kind == ShaderRecompiler::IR::ResourceKind::ImageUint)
 			         ? "true"
 			         : "false",
-			     r.Base40(), static_cast<uint32_t>(r.Type()), r.Format(),
+			     r.Base40(), static_cast<uint32_t>(r.Type()), static_cast<uint32_t>(r.Format()),
 			     static_cast<uint32_t>(r.Width5()) + 1u, static_cast<uint32_t>(r.Height5()) + 1u,
 			     static_cast<uint32_t>(r.Depth()) + 1u,
 			     std::max<uint32_t>(static_cast<uint32_t>(r.LastLevel()),
 			                        static_cast<uint32_t>(r.MaxMip())) +
 			         1u,
-			     r.TileMode());
+			     static_cast<uint32_t>(r.TileMode()));
 		}
 		for (uint32_t i = 0; i < program.info.samplers.size(); i++) {
 			const auto r = DecodeNativeDescriptor<ShaderSamplerResource>(resources.samplers[i]);
@@ -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";
@@ -103,7 +123,7 @@ static void LogFramebufferSkip(const char* draw_name, const RenderColorInfo& col
 	    log_id, draw_name, RenderColorTypeName(color.type), color.base_addr, color.buffer_size,
 	    color.image_id ? "yes" : "no", VulkanToString(depth.format).c_str(),
 	    depth.image_id ? "yes" : "no", depth.vaddr_num, ctx.GetRenderTargetMask(),
-	    ucfg.GetPrimType(), index_count, flags);
+	    static_cast<uint32_t>(ucfg.GetPrimType()), index_count, flags);
 }
 
 static void LogMrtState(const char* draw_name, const RenderCommandBuffer& buffer,
@@ -159,10 +179,12 @@ static void LogMrtState(const char* draw_name, const RenderCommandBuffer& buffer
 		     " fmt=0x%08" PRIx32 " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32
 		     " width=%u height=%u tile=%u"
 		     " blend=%s src=%u dst=%u alpha_src=%u alpha_dst=%u\n",
-		     log_id, i, rt.base.addr, ctm, csm, ps_input_info.target_output_mode[i], rt.info.format,
-		     rt.info.channel_type, rt.info.channel_order, rt.attrib2.width + 1,
-		     rt.attrib2.height + 1, rt.attrib3.tile_mode, bc.enable ? "true" : "false",
-		     bc.color_srcblend, bc.color_destblend, bc.alpha_srcblend, bc.alpha_destblend);
+		     log_id, i, rt.base.addr, ctm, csm, ps_input_info.target_output_mode[i],
+		     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+		     static_cast<uint32_t>(rt.info.channel_order), rt.attrib2.width + 1,
+		     rt.attrib2.height + 1, static_cast<uint32_t>(rt.attrib3.tile_mode),
+		     bc.enable ? "true" : "false", bc.color_srcblend, bc.color_destblend, bc.alpha_srcblend,
+		     bc.alpha_destblend);
 	}
 }
 
@@ -205,7 +227,7 @@ static void LogDrawTargetState(const char* draw_name, const RenderColorInfo& col
 	    "scissor=(%d,%d)-(%d,%d)\n",
 	    log_id, buffer.GetContext().GetGpu().GetFrameNum(), draw_name,
 	    RenderColorTypeName(color.type), color.base_addr, extent.width, extent.height,
-	    ucfg.GetPrimType(), index_count, flags, ctx.GetRenderTargetMask(),
+	    static_cast<uint32_t>(ucfg.GetPrimType()), index_count, flags, ctx.GetRenderTargetMask(),
 	    color.color_clear_enable ? "true" : "false", color.color_clear_value.float32[0],
 	    color.color_clear_value.float32[1], color.color_clear_value.float32[2],
 	    color.color_clear_value.float32[3], cc.mode, cc.op, bc.enable ? "true" : "false",
@@ -269,8 +291,7 @@ static void LogDrawInputState(const RenderCommandBuffer& buffer, const RenderCol
 					const auto& rd        = vs_input_info.resources_dst[res_index];
 					const auto  offset    = b.attr_offsets[ai];
 					if (offset + 4u <= b.stride &&
-					    r.Format() ==
-					        Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8UNorm)) {
+					    r.Format() == Prospero::BufferFormat::k8_8_8_8UNorm) {
 						uint32_t packed = 0;
 						std::memcpy(&packed, rec_bytes + offset, sizeof(packed));
 						const auto r8 = (packed >> 0u) & 0xffu;
@@ -335,6 +356,22 @@ 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 +416,13 @@ 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 +708,10 @@ 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 +728,24 @@ 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 +756,31 @@ 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 +789,47 @@ 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,
@@ -769,7 +845,7 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 
 	topology = vk::PrimitiveTopology::ePointList;
 
-	switch (static_cast<Prospero::PrimitiveType>(ucfg.GetPrimType())) {
+	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kNone: return false;
 		case Prospero::PrimitiveType::kPointList:
 			topology = vk::PrimitiveTopology::ePointList;
@@ -792,14 +868,14 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
 			if (!auto_draw) {
-				EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 			}
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
 			break;
-		default: EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+		default: EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 	}
 
 	return true;
@@ -885,10 +961,9 @@ 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 +971,6 @@ 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 +978,14 @@ 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 +996,29 @@ 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);
@@ -992,7 +1052,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
                                const DrawEmitInfo& emit) {
 	EXIT_IF(draw.name == nullptr);
 
-	switch (static_cast<Prospero::PrimitiveType>(ucfg.GetPrimType())) {
+	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kPointList:
 		case Prospero::PrimitiveType::kLineList:
 		case Prospero::PrimitiveType::kLineStrip:
@@ -1010,7 +1070,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
 			if (emit.indexed) {
-				EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 			}
 			// Sarah
 			EXIT_NOT_IMPLEMENTED(!(draw.index_count == 3 && vs_input_info.buffers_num == 0));
@@ -1028,7 +1088,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 				}
 			}
 			break;
-		default: EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+		default: EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 	}
 }
 
@@ -1046,8 +1106,6 @@ 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);
 
@@ -1330,9 +1388,8 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	}
 
 	LogDrawStateIfNeeded(buffer, draw, state, false,
-	                     ucfg.GetPrimType() ==
-	                         Prospero::GpuEnumValue(Prospero::PrimitiveType::kRectListLegacy),
-	                     0, nullptr);
+	                     ucfg.GetPrimType() == Prospero::PrimitiveType::kRectListLegacy, 0,
+	                     nullptr);
 
 	const auto   vertex_offset = ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
 	                             static_cast<int32_t>(first_vertex);

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/host_gpu/vulkanCommon.cpp

@@ -83,11 +83,11 @@ constexpr FormatMapping kFormatMappings[] = {
 };
 
 constexpr auto MakeFormatLookup() {
-	constexpr auto kMaxFormat = Prospero::GpuEnumValue(Prospero::BufferFormat::kBc7Srgb);
+	constexpr auto kMaxFormat = static_cast<size_t>(Prospero::BufferFormat::kBc7Srgb);
 	std::array<vk::Format, kMaxFormat + 1> lookup {};
 	lookup.fill(vk::Format::eUndefined);
 	for (const auto& mapping: kFormatMappings) {
-		lookup[Prospero::GpuEnumValue(mapping.guest)] = mapping.host;
+		lookup[static_cast<size_t>(mapping.guest)] = mapping.host;
 	}
 	return lookup;
 }
@@ -112,9 +112,9 @@ std::string VulkanToString(vk::QueueFlags value) {
 	return vk::to_string(value);
 }
 
-vk::Format VulkanFormat(uint32_t guest_format) {
-	return guest_format < kFormatLookup.size() ? kFormatLookup[guest_format]
-	                                           : vk::Format::eUndefined;
+vk::Format VulkanFormat(Prospero::BufferFormat guest_format) {
+	const auto index = static_cast<size_t>(guest_format);
+	return index < kFormatLookup.size() ? kFormatLookup[index] : vk::Format::eUndefined;
 }
 
 void RequireVulkanSuccess(vk::Result result, const char* operation) {

src/graphics/host_gpu/vulkanCommon.h

@@ -10,6 +10,8 @@
 #define VMA_STATIC_VULKAN_FUNCTIONS  0
 #define VMA_DYNAMIC_VULKAN_FUNCTIONS 1
 
+#include "graphics/guest_gpu/gpu_defs.h"
+
 #include <cstdint>
 #include <string>
 #include <vector>
@@ -29,7 +31,7 @@ std::string VulkanToString(vk::Result value);
 std::string VulkanToString(vk::Format value);
 std::string VulkanToString(vk::ImageLayout value);
 std::string VulkanToString(vk::QueueFlags value);
-vk::Format  VulkanFormat(uint32_t guest_format);
+vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
 
 template <typename T, typename Enumerator>

src/graphics/presentation/imeOverlay.cpp

@@ -0,0 +1,958 @@
+#include "graphics/presentation/imeOverlay.h"
+
+#include "SDL.h"
+#include "common/assert.h"
+#include "common/stringUtils.h"
+#include "graphics/host_gpu/graphicContext.h"
+#include "imgui.h"
+#include "imgui_impl_vulkan.h"
+#include "libs/controller.h"
+#include "libs/ime.h"
+#include "libs/imeDialog.h"
+
+#include <algorithm>
+#include <array>
+#include <atomic>
+#include <chrono>
+#include <cmath>
+#include <cstdint>
+#include <deque>
+#include <mutex>
+#include <span>
+#include <string>
+#include <string_view>
+#include <utility>
+
+namespace Libs::Graphics {
+
+namespace {
+
+namespace CoreIme   = Libs::Ime;
+namespace DialogIme = Libs::Dialog::ImeDialog;
+
+namespace Ime {
+
+using Type           = ImeCommon::Type;
+using EnterLabel     = ImeCommon::EnterLabel;
+using Alignment      = ImeCommon::Alignment;
+using ExternalAction = ImeCommon::ExternalAction;
+using ExternalInput  = ImeCommon::ExternalInput;
+using HostSnapshot   = ImeCommon::HostSnapshot;
+
+constexpr uint32_t OPTION_MULTILINE            = ImeCommon::OPTION_MULTILINE;
+constexpr uint32_t OPTION_NO_AUTO_CAPITALIZE   = ImeCommon::OPTION_NO_AUTO_CAPITALIZE;
+constexpr uint32_t OPTION_PASSWORD             = ImeCommon::OPTION_PASSWORD;
+constexpr uint32_t OPTION_FIXED_POSITION       = ImeCommon::OPTION_FIXED_POSITION;
+constexpr uint32_t OPTION_DISABLE_POSITION_ADJ = ImeCommon::OPTION_DISABLE_POSITION_ADJUST;
+constexpr uint32_t OPTION_USE_OVER_2K          = ImeCommon::OPTION_USE_OVER_2K;
+constexpr uint32_t DISABLE_DEVICE_CONTROLLER   = ImeCommon::DISABLE_DEVICE_CONTROLLER;
+constexpr uint32_t DISABLE_DEVICE_EXT_KEYBOARD = ImeCommon::DISABLE_DEVICE_EXT_KEYBOARD;
+constexpr uint64_t CORE_GENERATION_BIT         = uint64_t {1} << 63;
+
+uint64_t PackCoreGeneration(uint64_t generation) {
+	return generation | CORE_GENERATION_BIT;
+}
+
+bool IsCoreGeneration(uint64_t generation) {
+	return (generation & CORE_GENERATION_BIT) != 0;
+}
+
+uint64_t UnpackGeneration(uint64_t generation) {
+	return generation & ~CORE_GENERATION_BIT;
+}
+
+bool GetHostSnapshot(HostSnapshot* snapshot) {
+	if (snapshot == nullptr) {
+		return false;
+	}
+	if (CoreIme::GetHostSnapshot(snapshot)) {
+		snapshot->generation = PackCoreGeneration(snapshot->generation);
+		return true;
+	}
+	return DialogIme::GetHostSnapshot(snapshot);
+}
+
+bool HostInsertText(uint64_t generation, std::u16string_view text) {
+	return IsCoreGeneration(generation)
+	           ? CoreIme::HostInsertText(UnpackGeneration(generation), text)
+	           : DialogIme::HostInsertText(generation, text);
+}
+
+bool HostBackspace(uint64_t generation) {
+	return IsCoreGeneration(generation) ? CoreIme::HostBackspace(UnpackGeneration(generation))
+	                                    : DialogIme::HostBackspace(generation);
+}
+
+bool HostAccept(uint64_t generation) {
+	return IsCoreGeneration(generation) ? CoreIme::HostAccept(UnpackGeneration(generation))
+	                                    : DialogIme::HostAccept(generation);
+}
+
+bool HostCancel(uint64_t generation) {
+	return IsCoreGeneration(generation) ? CoreIme::HostCancel(UnpackGeneration(generation))
+	                                    : DialogIme::HostCancel(generation);
+}
+
+bool HostQueueExternalInput(uint64_t generation, ExternalInput input) {
+	return IsCoreGeneration(generation)
+	           ? CoreIme::HostQueueExternalInput(UnpackGeneration(generation), std::move(input))
+	           : DialogIme::HostQueueExternalInput(generation, std::move(input));
+}
+
+} // namespace Ime
+
+constexpr size_t INPUT_QUEUE_CAPACITY = 128;
+
+enum class InputKind : uint8_t {
+	Button,
+	Axis,
+	MousePosition,
+	MouseButton,
+	MouseWheel,
+	ResetController
+};
+
+struct InputEvent {
+	InputKind kind;
+	uint64_t  generation;
+	int       id;
+	float     x;
+	float     y;
+};
+
+struct VisibilityUpdate {
+	uint64_t generation;
+	bool     visible;
+	bool     capture_controller;
+	bool     capture_keyboard;
+	bool     multiline;
+};
+
+std::atomic<Uint32>          g_visibility_event {static_cast<Uint32>(-1)};
+std::mutex                   g_visibility_mutex;
+std::mutex                   g_input_mutex;
+std::deque<InputEvent>       g_input_events;
+std::deque<VisibilityUpdate> g_visibility_updates;
+size_t                       g_missing_visibility_wakeups = 0;
+bool                         g_input_reset_requested      = false;
+uint16_t                     g_last_external_keycode      = 0;
+uint32_t                     g_last_external_status       = 0;
+uint64_t                     g_input_generation           = 0;
+bool                         g_input_active               = false;
+bool                         g_input_controller           = false;
+bool                         g_input_keyboard             = false;
+bool                         g_input_multiline            = false;
+bool                         g_input_lifecycle_active     = false;
+bool                         g_active                     = false;
+bool                         g_controller_captured        = false;
+uint64_t                     g_generation                 = 0;
+
+void ClearInputEvents() {
+	std::scoped_lock lock(g_input_mutex);
+	g_input_events.clear();
+	g_input_reset_requested = false;
+}
+
+void QueueInput(InputEvent event) {
+	std::scoped_lock lock(g_input_mutex);
+	const bool       replaceable =
+	    event.kind == InputKind::Axis || event.kind == InputKind::MousePosition;
+	if (replaceable && !g_input_events.empty()) {
+		auto& last = g_input_events.back();
+		if (last.generation == event.generation && last.kind == event.kind && last.id == event.id) {
+			last = event;
+			return;
+		}
+	}
+	if (g_input_events.size() == INPUT_QUEUE_CAPACITY) {
+		g_input_events.clear();
+		g_input_reset_requested = true;
+	}
+	g_input_events.push_back(event);
+}
+
+void RetryVisibilityWakeup() {
+	const Uint32 type = g_visibility_event.load(std::memory_order_acquire);
+	if (type == static_cast<Uint32>(-1)) {
+		return;
+	}
+	std::scoped_lock lock(g_input_mutex);
+	if (g_missing_visibility_wakeups == 0) {
+		return;
+	}
+	SDL_Event event {};
+	event.type = type;
+	if (SDL_PushEvent(&event) > 0) {
+		g_missing_visibility_wakeups--;
+	}
+}
+
+void RefreshVisibility() {
+	std::scoped_lock visibility_lock(g_visibility_mutex);
+	if (!g_input_lifecycle_active) {
+		return;
+	}
+	Ime::HostSnapshot snapshot;
+	const bool        visible    = Ime::GetHostSnapshot(&snapshot);
+	const uint64_t    generation = visible ? snapshot.generation : g_generation;
+	if (visible == g_active && (!visible || generation == g_generation)) {
+		return;
+	}
+	g_generation            = generation;
+	bool capture_controller = false;
+	bool capture_keyboard   = false;
+	bool multiline          = false;
+	if (visible) {
+		capture_controller = (snapshot.disable_device & Ime::DISABLE_DEVICE_CONTROLLER) == 0;
+		capture_keyboard   = (snapshot.disable_device & Ime::DISABLE_DEVICE_EXT_KEYBOARD) == 0;
+		multiline          = (snapshot.option & Ime::OPTION_MULTILINE) != 0;
+	}
+	const bool was_controller = std::exchange(g_controller_captured, capture_controller);
+	if (capture_controller || was_controller) {
+		Controller::ControllerResetInputState();
+	}
+	g_active = visible;
+
+	const Uint32 type = g_visibility_event.load(std::memory_order_acquire);
+	if (type != static_cast<Uint32>(-1)) {
+		SDL_Event event {};
+		event.type = type;
+		std::scoped_lock input_lock(g_input_mutex);
+		g_visibility_updates.push_back(
+		    {generation, visible, capture_controller, capture_keyboard, multiline});
+		if (g_missing_visibility_wakeups != 0 || SDL_PushEvent(&event) <= 0) {
+			g_missing_visibility_wakeups++;
+		}
+	}
+}
+
+void OnCoreVisibilityChanged(bool, uint64_t) {
+	RefreshVisibility();
+}
+
+void OnDialogVisibilityChanged(bool, uint64_t) {
+	RefreshVisibility();
+}
+
+uint32_t ExternalKeyStatus(SDL_Keymod modifiers, bool character_valid) {
+	uint32_t status = 0x00000001 | (character_valid ? 0x00000002 : 0);
+	if ((modifiers & KMOD_LCTRL) != 0) status |= 0x00000100;
+	if ((modifiers & KMOD_LSHIFT) != 0) status |= 0x00000200;
+	if ((modifiers & KMOD_LALT) != 0) status |= 0x00000400;
+	if ((modifiers & KMOD_LGUI) != 0) status |= 0x00000800;
+	if ((modifiers & KMOD_RCTRL) != 0) status |= 0x00001000;
+	if ((modifiers & KMOD_RSHIFT) != 0) status |= 0x00002000;
+	if ((modifiers & KMOD_RALT) != 0) status |= 0x00004000;
+	if ((modifiers & KMOD_RGUI) != 0) status |= 0x00008000;
+	if ((modifiers & KMOD_NUM) != 0) status |= 0x00010000;
+	if ((modifiers & KMOD_CAPS) != 0) status |= 0x00020000;
+	return status;
+}
+
+Ime::ExternalInput MakeExternalInput(Ime::ExternalAction action, uint16_t keycode,
+                                     uint32_t status) {
+	Ime::ExternalInput input {};
+	input.key.keycode = keycode;
+	input.key.status  = status;
+	input.key.type    = 4;
+	input.action      = action;
+	return input;
+}
+
+std::u16string Utf8ToUtf16(std::string_view text) {
+	std::u16string result;
+	result.reserve(text.size());
+	for (size_t i = 0; i < text.size();) {
+		const auto first     = static_cast<uint8_t>(text[i++]);
+		uint32_t   codepoint = 0;
+		uint32_t   remaining = 0;
+		if (first < 0x80) {
+			codepoint = first;
+		} else if ((first & 0xe0) == 0xc0) {
+			codepoint = first & 0x1f;
+			remaining = 1;
+		} else if ((first & 0xf0) == 0xe0) {
+			codepoint = first & 0x0f;
+			remaining = 2;
+		} else if ((first & 0xf8) == 0xf0) {
+			codepoint = first & 0x07;
+			remaining = 3;
+		} else {
+			continue;
+		}
+		if (i + remaining > text.size()) {
+			break;
+		}
+		bool valid = true;
+		for (uint32_t j = 0; j < remaining; j++) {
+			const auto next = static_cast<uint8_t>(text[i++]);
+			if ((next & 0xc0) != 0x80) {
+				valid = false;
+				break;
+			}
+			codepoint = (codepoint << 6) | (next & 0x3f);
+		}
+		if (!valid || codepoint > 0x10ffff || (codepoint >= 0xd800 && codepoint <= 0xdfff)) {
+			continue;
+		}
+		if (codepoint <= 0xffff) {
+			result.push_back(static_cast<char16_t>(codepoint));
+		} else {
+			codepoint -= 0x10000;
+			result.push_back(static_cast<char16_t>(0xd800 + (codepoint >> 10)));
+			result.push_back(static_cast<char16_t>(0xdc00 + (codepoint & 0x3ff)));
+		}
+	}
+	return result;
+}
+
+std::string VisibleText(const Ime::HostSnapshot& snapshot, size_t max_units) {
+	const size_t cursor        = std::min<size_t>(snapshot.cursor, snapshot.text.size());
+	const size_t payload_units = max_units > 4 ? max_units - 4 : 1;
+	size_t       begin         = cursor > payload_units / 2 ? cursor - payload_units / 2 : 0;
+	size_t       end           = std::min(snapshot.text.size(), begin + payload_units);
+	if (end == snapshot.text.size() && end - begin < payload_units) {
+		begin = end > payload_units ? end - payload_units : 0;
+	}
+	if (begin > 0 && snapshot.text[begin] >= 0xdc00 && snapshot.text[begin] <= 0xdfff) {
+		begin--;
+	}
+	if (end < snapshot.text.size() && end > begin && snapshot.text[end - 1] >= 0xd800 &&
+	    snapshot.text[end - 1] <= 0xdbff) {
+		end--;
+	}
+
+	std::u16string visible;
+	if (begin != 0) {
+		visible.push_back(u'\u2026');
+	}
+	const size_t caret = visible.size() + cursor - begin;
+	if ((snapshot.option & Ime::OPTION_PASSWORD) != 0) {
+		visible.append(end - begin, u'*');
+	} else {
+		visible.append(snapshot.text, begin, end - begin);
+		std::replace(visible.begin(), visible.end(), u'\n', u'\u21b5');
+		std::replace(visible.begin(), visible.end(), u'\r', u'\u21b5');
+	}
+	visible.insert(std::min(caret, visible.size()), 1, u'|');
+	if (end != snapshot.text.size()) {
+		visible.push_back(u'\u2026');
+	}
+	return Common::Utf16ToUtf8(visible.c_str());
+}
+
+const char* EnterLabel(Ime::EnterLabel label) {
+	switch (label) {
+		case Ime::EnterLabel::Send: return "Send";
+		case Ime::EnterLabel::Search: return "Search";
+		case Ime::EnterLabel::Go: return "Go";
+		default: return "Done";
+	}
+}
+
+float AlignmentPivot(Ime::Alignment alignment) {
+	switch (alignment) {
+		case Ime::Alignment::Start: return 0.0f;
+		case Ime::Alignment::End: return 1.0f;
+		default: return 0.5f;
+	}
+}
+
+ImGuiKey ControllerButtonToKey(int button) {
+	switch (button) {
+		case SDL_CONTROLLER_BUTTON_A: return ImGuiKey_GamepadFaceDown;
+		case SDL_CONTROLLER_BUTTON_B: return ImGuiKey_GamepadFaceRight;
+		case SDL_CONTROLLER_BUTTON_X: return ImGuiKey_GamepadFaceLeft;
+		case SDL_CONTROLLER_BUTTON_Y: return ImGuiKey_GamepadFaceUp;
+		case SDL_CONTROLLER_BUTTON_DPAD_LEFT: return ImGuiKey_GamepadDpadLeft;
+		case SDL_CONTROLLER_BUTTON_DPAD_RIGHT: return ImGuiKey_GamepadDpadRight;
+		case SDL_CONTROLLER_BUTTON_DPAD_UP: return ImGuiKey_GamepadDpadUp;
+		case SDL_CONTROLLER_BUTTON_DPAD_DOWN: return ImGuiKey_GamepadDpadDown;
+		default: return ImGuiKey_None;
+	}
+}
+
+PFN_vkVoidFunction LoadVulkanFunction(const char* name, void* user_data) {
+	auto& graphics = *static_cast<GraphicContext*>(user_data);
+	return graphics.instance.getProcAddr(name);
+}
+
+void CheckVulkanResult(VkResult result) {
+	EXIT_IF(result != VK_SUCCESS);
+}
+
+} // namespace
+
+void InitializeImeInput() {
+	const Uint32 type = SDL_RegisterEvents(1);
+	EXIT_IF(type == static_cast<Uint32>(-1));
+	{
+		std::scoped_lock lock(g_visibility_mutex);
+		g_input_lifecycle_active = true;
+		g_visibility_event.store(type, std::memory_order_release);
+	}
+	CoreIme::SetVisibilityCallback(OnCoreVisibilityChanged);
+	DialogIme::SetVisibilityCallback(OnDialogVisibilityChanged);
+	RefreshVisibility();
+}
+
+void ShutdownImeInput() {
+	CoreIme::SetVisibilityCallback(nullptr);
+	DialogIme::SetVisibilityCallback(nullptr);
+	{
+		std::scoped_lock lock(g_visibility_mutex);
+		g_input_lifecycle_active = false;
+		g_active                 = false;
+		if (std::exchange(g_controller_captured, false)) {
+			Controller::ControllerResetInputState();
+		}
+		g_visibility_event.store(static_cast<Uint32>(-1), std::memory_order_release);
+	}
+	ClearInputEvents();
+	{
+		std::scoped_lock lock(g_input_mutex);
+		g_visibility_updates.clear();
+		g_missing_visibility_wakeups = 0;
+	}
+	g_input_active     = false;
+	g_input_controller = false;
+	g_input_keyboard   = false;
+	g_input_multiline  = false;
+	if (SDL_IsTextInputActive() == SDL_TRUE) {
+		SDL_StopTextInput();
+	}
+}
+
+ImeVisualState GetImeVisualState() noexcept {
+	const auto core   = CoreIme::GetVisualState();
+	const auto dialog = DialogIme::GetVisualState();
+	return {core.active || dialog.active, core.revision + dialog.revision};
+}
+
+bool ProcessImeInput(const SDL_Event& event) {
+	RetryVisibilityWakeup();
+	if (event.type == g_visibility_event.load(std::memory_order_acquire)) {
+		VisibilityUpdate update {};
+		{
+			std::scoped_lock lock(g_input_mutex);
+			if (g_visibility_updates.empty()) {
+				return true;
+			}
+			update = g_visibility_updates.front();
+			g_visibility_updates.pop_front();
+		}
+		ClearInputEvents();
+		g_last_external_keycode = 0;
+		g_last_external_status  = 0;
+		g_input_generation      = update.generation;
+		g_input_active          = update.visible;
+		g_input_controller      = update.capture_controller;
+		g_input_keyboard        = update.capture_keyboard;
+		g_input_multiline       = update.multiline;
+		if (g_input_keyboard) {
+			SDL_StartTextInput();
+		} else if (SDL_IsTextInputActive() == SDL_TRUE) {
+			SDL_StopTextInput();
+		}
+		return true;
+	}
+	if (event.type == SDL_CONTROLLERDEVICEREMOVED) {
+		if (g_input_active && g_input_controller) {
+			QueueInput({InputKind::ResetController, g_input_generation, 0, 0.0f, 0.0f});
+		}
+		return false;
+	}
+	if (!g_input_active) {
+		return false;
+	}
+
+	const uint64_t generation       = g_input_generation;
+	const bool     controller_event = event.type == SDL_CONTROLLERBUTTONDOWN ||
+	                                  event.type == SDL_CONTROLLERBUTTONUP ||
+	                                  event.type == SDL_CONTROLLERAXISMOTION;
+	if (controller_event && !g_input_controller) {
+		return false;
+	}
+	const bool keyboard_event = event.type == SDL_TEXTINPUT || event.type == SDL_TEXTEDITING ||
+	                            event.type == SDL_KEYDOWN || event.type == SDL_KEYUP;
+	if (keyboard_event && !g_input_keyboard) {
+		return false;
+	}
+	switch (event.type) {
+		case SDL_TEXTINPUT: {
+			const auto text = Utf8ToUtf16(event.text.text);
+			if (!text.empty()) {
+				auto input = MakeExternalInput(Ime::ExternalAction::Text, g_last_external_keycode,
+				                               g_last_external_status | 0x00000002);
+				input.key.character = text.front();
+				input.text          = text;
+				Ime::HostQueueExternalInput(generation, std::move(input));
+			}
+			return true;
+		}
+		case SDL_TEXTEDITING: return true;
+		case SDL_KEYDOWN: {
+			g_last_external_keycode = static_cast<uint16_t>(event.key.keysym.scancode);
+			g_last_external_status =
+			    ExternalKeyStatus(static_cast<SDL_Keymod>(event.key.keysym.mod), false);
+			auto action = Ime::ExternalAction::Text;
+			bool queue  = true;
+			if (event.key.keysym.sym == SDLK_BACKSPACE) {
+				action = Ime::ExternalAction::Backspace;
+			} else if (event.key.keysym.sym == SDLK_LEFT) {
+				action = Ime::ExternalAction::MoveLeft;
+			} else if (event.key.keysym.sym == SDLK_RIGHT) {
+				action = Ime::ExternalAction::MoveRight;
+			} else if (event.key.keysym.sym == SDLK_ESCAPE) {
+				action = Ime::ExternalAction::Cancel;
+			} else if (event.key.keysym.sym == SDLK_RETURN ||
+			           event.key.keysym.sym == SDLK_KP_ENTER) {
+				action =
+				    g_input_multiline ? Ime::ExternalAction::Newline : Ime::ExternalAction::Accept;
+			} else if (event.key.keysym.sym == SDLK_TAB) {
+				action = Ime::ExternalAction::None;
+			} else {
+				queue = false;
+			}
+			if (queue) {
+				Ime::HostQueueExternalInput(
+				    generation,
+				    MakeExternalInput(action, g_last_external_keycode, g_last_external_status));
+			}
+			return true;
+		}
+		case SDL_KEYUP: return true;
+		case SDL_CONTROLLERBUTTONDOWN:
+		case SDL_CONTROLLERBUTTONUP:
+			QueueInput({InputKind::Button, generation, event.cbutton.button,
+			            event.type == SDL_CONTROLLERBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
+			return true;
+		case SDL_CONTROLLERAXISMOTION:
+			QueueInput({InputKind::Axis, generation, event.caxis.axis,
+			            static_cast<float>(event.caxis.value), 0.0f});
+			return true;
+		case SDL_MOUSEMOTION:
+			QueueInput({InputKind::MousePosition, generation, 0, static_cast<float>(event.motion.x),
+			            static_cast<float>(event.motion.y)});
+			return true;
+		case SDL_MOUSEBUTTONDOWN:
+		case SDL_MOUSEBUTTONUP:
+			QueueInput({InputKind::MouseButton, generation, event.button.button,
+			            event.type == SDL_MOUSEBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
+			return true;
+		case SDL_MOUSEWHEEL:
+			QueueInput({InputKind::MouseWheel, generation, 0, static_cast<float>(event.wheel.x),
+			            static_cast<float>(event.wheel.y)});
+			return true;
+		default: return false;
+	}
+}
+
+struct ImeOverlay::Impl {
+	explicit Impl(GraphicContext& context): graphics(context) {}
+
+	~Impl() {
+		ReleaseVulkan();
+		if (imgui_context != nullptr) {
+			ImGui::DestroyContext(imgui_context);
+		}
+	}
+
+	void EnsureContext() {
+		if (imgui_context != nullptr) {
+			ImGui::SetCurrentContext(imgui_context);
+			return;
+		}
+		IMGUI_CHECKVERSION();
+		imgui_context = ImGui::CreateContext();
+		ImGui::SetCurrentContext(imgui_context);
+		auto& io       = ImGui::GetIO();
+		io.IniFilename = nullptr;
+		io.LogFilename = nullptr;
+		io.ConfigFlags |= ImGuiConfigFlags_NavEnableGamepad;
+		io.BackendFlags |= ImGuiBackendFlags_HasGamepad;
+		io.BackendPlatformName = "Kyty IME input";
+		ImGui::StyleColorsDark();
+		auto& style          = ImGui::GetStyle();
+		style.WindowRounding = 10.0f;
+		style.FrameRounding  = 6.0f;
+		style.ItemSpacing    = {8.0f, 8.0f};
+	}
+
+	void EnsureVulkan(vk::Format format, uint32_t image_count) {
+		EnsureContext();
+		if (vulkan_initialized) {
+			return;
+		}
+		EXIT_IF(image_count < 2);
+		EXIT_IF(!ImGui_ImplVulkan_LoadFunctions(VK_API_VERSION_1_3, LoadVulkanFunction, &graphics));
+
+		const VkFormat            color_format = static_cast<VkFormat>(format);
+		ImGui_ImplVulkan_InitInfo info {};
+		info.ApiVersion                   = VK_API_VERSION_1_3;
+		info.Instance                     = static_cast<VkInstance>(graphics.instance);
+		info.PhysicalDevice               = static_cast<VkPhysicalDevice>(graphics.physical_device);
+		info.Device                       = static_cast<VkDevice>(graphics.device);
+		info.QueueFamily                  = graphics.queue_family;
+		info.Queue                        = static_cast<VkQueue>(graphics.queue);
+		info.DescriptorPoolSize           = IMGUI_IMPL_VULKAN_MINIMUM_SAMPLED_IMAGE_POOL_SIZE;
+		info.MinImageCount                = image_count;
+		info.ImageCount                   = image_count;
+		info.UseDynamicRendering          = true;
+		info.PipelineInfoMain.MSAASamples = VK_SAMPLE_COUNT_1_BIT;
+		info.PipelineInfoMain.PipelineRenderingCreateInfo.sType =
+		    VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO;
+		info.PipelineInfoMain.PipelineRenderingCreateInfo.colorAttachmentCount    = 1;
+		info.PipelineInfoMain.PipelineRenderingCreateInfo.pColorAttachmentFormats = &color_format;
+		info.CheckVkResultFn = CheckVulkanResult;
+		EXIT_IF(!ImGui_ImplVulkan_Init(&info));
+		vulkan_initialized = true;
+	}
+
+	void DrainInput(uint64_t generation) {
+		std::deque<InputEvent> events;
+		bool                   reset = false;
+		{
+			std::scoped_lock lock(g_input_mutex);
+			events.swap(g_input_events);
+			reset                   = g_input_reset_requested;
+			g_input_reset_requested = false;
+		}
+		auto& io = ImGui::GetIO();
+		if (reset) {
+			io.ClearEventsQueue();
+			io.ClearInputKeys();
+			io.ClearInputMouse();
+			right_stick = {};
+		}
+		for (const auto& event: events) {
+			if (event.generation != generation) {
+				continue;
+			}
+			switch (event.kind) {
+				case InputKind::Button: {
+					const bool down = event.x != 0.0f;
+					if (down && event.id == SDL_CONTROLLER_BUTTON_B) {
+						Ime::HostCancel(generation);
+					} else if (down && event.id == SDL_CONTROLLER_BUTTON_Y) {
+						Ime::HostBackspace(generation);
+					}
+					const ImGuiKey key = ControllerButtonToKey(event.id);
+					if (key != ImGuiKey_None) {
+						io.AddKeyEvent(key, down);
+					}
+					break;
+				}
+				case InputKind::Axis: {
+					const float value = event.x / (event.x < 0.0f ? 32768.0f : 32767.0f);
+					if (event.id == SDL_CONTROLLER_AXIS_LEFTX) {
+						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickLeft, value < -0.25f,
+						                     std::max(-value, 0.0f));
+						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickRight, value > 0.25f,
+						                     std::max(value, 0.0f));
+					} else if (event.id == SDL_CONTROLLER_AXIS_LEFTY) {
+						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickUp, value < -0.25f,
+						                     std::max(-value, 0.0f));
+						io.AddKeyAnalogEvent(ImGuiKey_GamepadLStickDown, value > 0.25f,
+						                     std::max(value, 0.0f));
+					} else if (event.id == SDL_CONTROLLER_AXIS_RIGHTX) {
+						right_stick.x = std::abs(value) > 0.2f ? value : 0.0f;
+					} else if (event.id == SDL_CONTROLLER_AXIS_RIGHTY) {
+						right_stick.y = std::abs(value) > 0.2f ? value : 0.0f;
+					}
+					break;
+				}
+				case InputKind::MousePosition: io.AddMousePosEvent(event.x, event.y); break;
+				case InputKind::MouseButton: {
+					int button = -1;
+					if (event.id == SDL_BUTTON_LEFT) button = 0;
+					if (event.id == SDL_BUTTON_RIGHT) button = 1;
+					if (event.id == SDL_BUTTON_MIDDLE) button = 2;
+					if (button >= 0) io.AddMouseButtonEvent(button, event.x != 0.0f);
+					break;
+				}
+				case InputKind::MouseWheel: io.AddMouseWheelEvent(event.x, event.y); break;
+				case InputKind::ResetController:
+					io.ClearEventsQueue();
+					io.ClearInputKeys();
+					right_stick = {};
+					break;
+			}
+		}
+	}
+
+	void KeyButton(std::string_view label, char16_t value, uint64_t generation, float width,
+	               bool default_focus) {
+		const bool pressed = ImGui::Button(label.data(), {width, button_height});
+		if (default_focus) {
+			ImGui::SetItemDefaultFocus();
+		}
+		if (pressed) {
+			Ime::HostInsertText(generation, std::u16string_view(&value, 1));
+		}
+	}
+
+	void DrawKeyRows(const Ime::HostSnapshot& snapshot, float width) {
+		static constexpr std::array<std::string_view, 4> lower   = {"1234567890", "qwertyuiop",
+		                                                            "asdfghjkl", "zxcvbnm"};
+		static constexpr std::array<std::string_view, 4> upper   = {"1234567890", "QWERTYUIOP",
+		                                                            "ASDFGHJKL", "ZXCVBNM"};
+		static constexpr std::array<std::string_view, 3> symbols = {"1234567890", "!@#$%^&*()",
+		                                                            "-_=+/:?."};
+		static constexpr std::array<std::string_view, 4> number  = {"123", "456", "789", "-0."};
+
+		std::span<const std::string_view> rows;
+		if (snapshot.type == Ime::Type::Number) {
+			rows = number;
+		} else if (symbol_mode) {
+			rows = symbols;
+		} else {
+			rows = shift ? std::span<const std::string_view>(upper)
+			             : std::span<const std::string_view>(lower);
+		}
+		bool   first     = focus_pending;
+		size_t row_index = 0;
+		for (const auto row: rows) {
+			const float key_width =
+			    std::min(56.0f * ui_scale, (width - 8.0f * (row.size() - 1)) / row.size());
+			const float row_width = key_width * row.size() + 8.0f * (row.size() - 1);
+			ImGui::SetCursorPosX((width - row_width) * 0.5f);
+			ImGui::PushID(static_cast<int>(row_index++));
+			for (size_t i = 0; i < row.size(); i++) {
+				ImGui::PushID(static_cast<int>(i));
+				const char label[2] = {row[i], '\0'};
+				KeyButton(label, static_cast<char16_t>(row[i]), snapshot.generation, key_width,
+				          first);
+				first = false;
+				ImGui::PopID();
+				if (i + 1 < row.size()) ImGui::SameLine();
+			}
+			ImGui::PopID();
+		}
+		focus_pending = false;
+	}
+
+	void DrawDialog(const Ime::HostSnapshot& snapshot, vk::Extent2D extent) {
+		const ImVec2 display(static_cast<float>(extent.width), static_cast<float>(extent.height));
+		const bool   over_2k          = (snapshot.option & Ime::OPTION_USE_OVER_2K) != 0;
+		const float  reference_width  = over_2k ? 3840.0f : 1920.0f;
+		const float  reference_height = over_2k ? 2160.0f : 1080.0f;
+		const float  scale_x          = display.x / reference_width;
+		const float  scale_y          = display.y / reference_height;
+		const float  width            = static_cast<float>(snapshot.panel_width) * scale_x;
+		const float  height           = static_cast<float>(snapshot.panel_height) * scale_y;
+		ui_scale                      = std::min(scale_x, scale_y) * (over_2k ? 2.0f : 1.0f);
+		button_height                 = std::max(28.0f, 42.0f * ui_scale);
+
+		const ImVec2 pivot {AlignmentPivot(snapshot.horizontal_alignment),
+		                    AlignmentPivot(snapshot.vertical_alignment)};
+		if ((snapshot.option & Ime::OPTION_FIXED_POSITION) == 0) {
+			const float movement = 600.0f * ImGui::GetIO().DeltaTime;
+			panel_offset.x += right_stick.x * movement;
+			panel_offset.y += right_stick.y * movement;
+		}
+		const ImVec2 base_position {snapshot.posx * scale_x - pivot.x * width,
+		                            snapshot.posy * scale_y - pivot.y * height};
+		ImVec2       position {base_position.x + panel_offset.x, base_position.y + panel_offset.y};
+		if ((snapshot.option & Ime::OPTION_DISABLE_POSITION_ADJ) == 0) {
+			position.x   = std::clamp(position.x, 0.0f, std::max(display.x - width, 0.0f));
+			position.y   = std::clamp(position.y, 0.0f, std::max(display.y - height, 0.0f));
+			panel_offset = {position.x - base_position.x, position.y - base_position.y};
+		}
+		ImGui::SetNextWindowPos(position, ImGuiCond_Always);
+		ImGui::SetNextWindowSize({width, height}, ImGuiCond_Always);
+		constexpr ImGuiWindowFlags flags = ImGuiWindowFlags_NoDecoration | ImGuiWindowFlags_NoMove |
+		                                   ImGuiWindowFlags_NoSavedSettings;
+		ImGui::Begin("##Ime", nullptr, flags);
+
+		const std::u16string title = snapshot.title.empty() ? u"Enter text" : snapshot.title;
+		if (!snapshot.key_panel_visible) {
+			std::string compact     = Common::Utf16ToUtf8(title.c_str()) + ": ";
+			const float glyph_width = std::max(ImGui::CalcTextSize("M").x, 1.0f);
+			const float text_width =
+			    std::max(ImGui::GetContentRegionAvail().x - ImGui::CalcTextSize(compact.c_str()).x,
+			             glyph_width);
+			const size_t visible_units =
+			    std::max<size_t>(6, static_cast<size_t>(text_width / glyph_width));
+			compact += snapshot.text.empty() && !snapshot.placeholder.empty()
+			               ? Common::Utf16ToUtf8(snapshot.placeholder.c_str())
+			               : VisibleText(snapshot, visible_units);
+			ImGui::TextUnformatted(compact.c_str());
+			const float compact_height = std::min(button_height, 28.0f * ui_scale);
+			if (ImGui::Button("Cancel", {90.0f * ui_scale, compact_height})) {
+				Ime::HostCancel(snapshot.generation);
+			}
+			ImGui::SameLine();
+			if (ImGui::Button(EnterLabel(snapshot.enter_label),
+			                  {90.0f * ui_scale, compact_height})) {
+				Ime::HostAccept(snapshot.generation);
+			}
+			ImGui::End();
+			return;
+		}
+
+		ImGui::TextUnformatted(Common::Utf16ToUtf8(title.c_str()).c_str());
+		ImGui::Separator();
+		const float text_height = std::max(
+		    32.0f, ((snapshot.option & Ime::OPTION_MULTILINE) != 0 ? 80.0f : 48.0f) * ui_scale);
+		ImGui::BeginChild("##ImeText", {0.0f, text_height}, true);
+		if (snapshot.text.empty() && !snapshot.placeholder.empty()) {
+			ImGui::TextDisabled("%s", Common::Utf16ToUtf8(snapshot.placeholder.c_str()).c_str());
+		} else {
+			const float  glyph_width = std::max(ImGui::CalcTextSize("M").x, 1.0f);
+			const size_t columns     = std::max<size_t>(
+			    8, static_cast<size_t>(ImGui::GetContentRegionAvail().x / glyph_width));
+			const size_t lines =
+			    std::max<size_t>(1, static_cast<size_t>(ImGui::GetContentRegionAvail().y /
+			                                            ImGui::GetTextLineHeightWithSpacing()));
+			const auto text = VisibleText(snapshot, columns * lines);
+			ImGui::TextWrapped("%s", text.c_str());
+		}
+		ImGui::EndChild();
+		ImGui::TextDisabled("%zu / %u", snapshot.text.size(), snapshot.max_text_length);
+		ImGui::Separator();
+
+		const float content_width = ImGui::GetContentRegionAvail().x;
+		DrawKeyRows(snapshot, content_width);
+		if (snapshot.type != Ime::Type::Number) {
+			if (ImGui::Button(shift ? "Lower" : "Shift", {84.0f * ui_scale, button_height})) {
+				shift = !shift;
+			}
+			ImGui::SameLine();
+			if (ImGui::Button(symbol_mode ? "ABC" : "Symbols", {84.0f * ui_scale, button_height})) {
+				symbol_mode = !symbol_mode;
+			}
+			ImGui::SameLine();
+			if (ImGui::Button("Space", {120.0f * ui_scale, button_height})) {
+				Ime::HostInsertText(snapshot.generation, u" ");
+			}
+			ImGui::SameLine();
+		}
+		const float action_width = snapshot.type == Ime::Type::Number
+		                               ? std::max((content_width - 16.0f) / 3.0f, 48.0f)
+		                               : 100.0f * ui_scale;
+		if (ImGui::Button("Backspace", {action_width, button_height})) {
+			Ime::HostBackspace(snapshot.generation);
+		}
+		ImGui::SameLine();
+		if ((snapshot.option & Ime::OPTION_MULTILINE) != 0) {
+			if (ImGui::Button("Newline", {action_width, button_height})) {
+				Ime::HostInsertText(snapshot.generation, u"\n");
+			}
+			ImGui::SameLine();
+		}
+		if (ImGui::Button("Cancel", {action_width, button_height})) {
+			Ime::HostCancel(snapshot.generation);
+		}
+		ImGui::SameLine();
+		if (ImGui::Button(EnterLabel(snapshot.enter_label), {action_width, button_height})) {
+			Ime::HostAccept(snapshot.generation);
+		}
+		ImGui::End();
+	}
+
+	bool PrepareFrame(vk::Extent2D frame_extent, vk::Format format, uint32_t image_count) {
+		Ime::HostSnapshot snapshot;
+		if (!Ime::GetHostSnapshot(&snapshot)) {
+			return false;
+		}
+		const uint64_t prepared_generation = snapshot.generation;
+		EnsureVulkan(format, image_count);
+		if (generation != snapshot.generation) {
+			generation    = snapshot.generation;
+			focus_pending = true;
+			shift         = (snapshot.option & Ime::OPTION_NO_AUTO_CAPITALIZE) == 0;
+			symbol_mode   = false;
+			panel_offset  = {};
+			right_stick   = {};
+			auto& io      = ImGui::GetIO();
+			io.ClearEventsQueue();
+			io.ClearInputKeys();
+			io.ClearInputMouse();
+		}
+		DrainInput(snapshot.generation);
+
+		auto& io       = ImGui::GetIO();
+		io.DisplaySize = {static_cast<float>(frame_extent.width),
+		                  static_cast<float>(frame_extent.height)};
+		const auto now = std::chrono::steady_clock::now();
+		io.DeltaTime   = last_frame == std::chrono::steady_clock::time_point {}
+		                     ? 1.0f / 60.0f
+		                     : std::clamp(std::chrono::duration<float>(now - last_frame).count(),
+		                                  1.0f / 1000.0f, 0.1f);
+		last_frame     = now;
+		ImGui_ImplVulkan_NewFrame();
+		ImGui::NewFrame();
+		if (!Ime::GetHostSnapshot(&snapshot) || snapshot.generation != prepared_generation) {
+			ImGui::EndFrame();
+			return false;
+		}
+		DrawDialog(snapshot, frame_extent);
+		ImGui::Render();
+		extent = frame_extent;
+		return true;
+	}
+
+	void Record(vk::CommandBuffer command, vk::ImageView target) {
+		vk::RenderingAttachmentInfo color {};
+		color.sType       = vk::StructureType::eRenderingAttachmentInfo;
+		color.imageView   = target;
+		color.imageLayout = vk::ImageLayout::eColorAttachmentOptimal;
+		color.loadOp      = vk::AttachmentLoadOp::eLoad;
+		color.storeOp     = vk::AttachmentStoreOp::eStore;
+		vk::RenderingInfo rendering {};
+		rendering.sType                = vk::StructureType::eRenderingInfo;
+		rendering.renderArea.extent    = extent;
+		rendering.layerCount           = 1;
+		rendering.colorAttachmentCount = 1;
+		rendering.pColorAttachments    = &color;
+		command.beginRendering(rendering);
+		{
+			Common::LockGuard queue_lock(graphics.queue_mutex);
+			ImGui_ImplVulkan_RenderDrawData(ImGui::GetDrawData(),
+			                                static_cast<VkCommandBuffer>(command));
+		}
+		command.endRendering();
+	}
+
+	void ReleaseVulkan() {
+		if (!vulkan_initialized) {
+			return;
+		}
+		ImGui::SetCurrentContext(imgui_context);
+		ImGui_ImplVulkan_Shutdown();
+		vulkan_initialized = false;
+	}
+
+	GraphicContext&                       graphics;
+	ImGuiContext*                         imgui_context      = nullptr;
+	bool                                  vulkan_initialized = false;
+	bool                                  shift              = false;
+	bool                                  symbol_mode        = false;
+	bool                                  focus_pending      = true;
+	float                                 ui_scale           = 1.0f;
+	float                                 button_height      = 42.0f;
+	ImVec2                                panel_offset {};
+	ImVec2                                right_stick {};
+	uint64_t                              generation = 0;
+	vk::Extent2D                          extent {};
+	std::chrono::steady_clock::time_point last_frame;
+};
+
+ImeOverlay::ImeOverlay(GraphicContext& graphics): m_impl(std::make_unique<Impl>(graphics)) {}
+
+ImeOverlay::~ImeOverlay() = default;
+
+bool ImeOverlay::PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count) {
+	return m_impl->PrepareFrame(extent, format, image_count);
+}
+
+void ImeOverlay::Record(vk::CommandBuffer command, vk::ImageView target) {
+	m_impl->Record(command, target);
+}
+
+void ImeOverlay::ReleaseVulkan() {
+	m_impl->ReleaseVulkan();
+}
+
+} // namespace Libs::Graphics

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

@@ -534,13 +534,12 @@ Graphics::ImageInfo BufferAttributeGroup::ImageInfo(const VideoOutBuffer& buffer
 	if (!Graphics::DecodeVideoOutPixelFormat(attribute.pixel_format, pixel_format)) {
 		EXIT("unsupported video-out pixel format: 0x%016" PRIx64 "\n", attribute.pixel_format);
 	}
-	const auto tile_mode =
-	    Graphics::Prospero::GpuEnumValue(Graphics::Prospero::TileMode::kRenderTarget);
+	const auto tile_mode = 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 +787,10 @@ 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 +812,24 @@ 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 +1,306 @@
+#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 ||
+			    binding.key == SDLK_F11;
+			if (binding.control == INVALID_CONTROL || reserved ||
+			    (binding.key == SDLK_UNKNOWN && binding.mouse_button == 0)) {
+				EXIT("Invalid input mapping: %s\n", value.c_str());
+			}
+			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

@@ -22,7 +22,6 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "common/stringUtils.h"
-#include "common/subsystems.h"
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
@@ -31,6 +30,7 @@
 #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 +329,9 @@ 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 +343,17 @@ 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 +385,20 @@ 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 +527,9 @@ 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 +614,15 @@ 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 +668,14 @@ 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 +684,22 @@ 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 +804,12 @@ 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 +827,22 @@ 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 +855,9 @@ 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 +867,7 @@ 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

@@ -22,7 +22,6 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "common/stringUtils.h"
-#include "common/subsystems.h"
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
@@ -31,6 +30,7 @@
 #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 +259,10 @@ 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 +552,7 @@ 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 +568,7 @@ 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 +606,6 @@ 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 +1010,7 @@ void WindowContext::RecreateSurface() {
 }
 
 WindowContext::~WindowContext() {
+	ShutdownImeInput();
 	presenter.reset();
 	LibKernel::Memory::InstallGpuResources(nullptr);
 	render_context.reset();

src/graphics/presentation/window/window.cpp

@@ -31,8 +31,11 @@
 #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"
 #include "loader/systemContent.h"
 
@@ -58,8 +61,6 @@
 
 namespace Libs::Graphics {
 
-constexpr int KEYBOARD_CONTROLLER_ID = -1000;
-
 struct EventKeyboard {
 	bool     down;
 	bool     up;
@@ -72,26 +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;
@@ -238,7 +219,24 @@ static void GameEventTerminate(WindowLoopState& game) {
 	game.need_exit = true;
 }
 
+static void ToggleDesktopFullscreen() {
+	if (g_window == nullptr || g_window->window == nullptr) {
+		return;
+	}
+
+	const auto flags = static_cast<uint32_t>(SDL_GetWindowFlags(g_window->window));
+	const bool fullscreen =
+	    (flags & static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP)) != 0u;
+	const auto mode =
+	    fullscreen ? 0u : static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP);
+	if (SDL_SetWindowFullscreen(g_window->window, mode) != 0) {
+		LOGF("Toggle fullscreen failed: %s\n", SDL_GetError());
+	}
+}
+
 static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
+	static SDL_Keycode fullscreen_key = SDLK_UNKNOWN;
+
 #ifdef KYTY_DBG_INPUT
 	LOGF("Key: time = %.04f, %s%s, %s%s, %s, scan = %d, key = %d, mod = %04" PRIx16 "\n",
 	     key.timestamp_seconds, (key.down ? "down" : ""), (key.up ? "up" : ""),
@@ -256,20 +254,32 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 					RenderDocRequestCapture();
 				}
 				break;
+			case SDLK_F11:
+				if (!key.repeat) {
+					ToggleDesktopFullscreen();
+				}
+				break;
+			case SDLK_RETURN:
+			case SDLK_KP_ENTER:
+				if (!key.repeat && (key.mod & KMOD_ALT) != 0) {
+					ToggleDesktopFullscreen();
+					fullscreen_key = key.key_code;
+				}
+				break;
 			default: break;
 		}
 	}
 
-	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
+
+	const bool fullscreen_key_event =
+	    fullscreen_key != SDLK_UNKNOWN && key.key_code == fullscreen_key;
+	if (fullscreen_key_event && key.up) {
+		fullscreen_key = SDLK_UNKNOWN;
+	}
+	if ((key.down || key.up) && !key.repeat && !fullscreen_key_event) {
+		HostInputKey(key.key_code, key.down);
+	}
 }
 
 static void GameEventMouse([[maybe_unused]] const EventMouse& mb) {
@@ -291,6 +301,23 @@ 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) {
@@ -484,6 +511,9 @@ 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;
@@ -756,26 +786,18 @@ 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());
 	}
 }
 
@@ -794,10 +816,15 @@ 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
@@ -937,6 +964,14 @@ void WindowContext::UpdateTitle() {
 	static uint64_t fps_frames  = 0;
 	static double   current_fps = 0.0;
 
+#if KYTY_BUILD == KYTY_BUILD_DEBUG
+	static constexpr auto build_type = "Debug";
+#elif KYTY_BUILD == KYTY_BUILD_RELEASE
+	static constexpr auto build_type = "Release";
+#else
+	static constexpr auto build_type = "Unknown";
+#endif
+
 	const auto now       = Common::Timer::QueryPerformanceCounter();
 	const auto frequency = Common::Timer::QueryPerformanceFrequency();
 	frame_num++;
@@ -948,11 +983,11 @@ void WindowContext::UpdateTitle() {
 		fps_frames  = 0;
 	}
 
-	auto fps =
-	    fmt::format("{}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", (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);
@@ -894,7 +893,8 @@ uint32_t EmitByteAddress(EmitterState& state, const IR::Instruction& inst, uint3
 uint32_t StorageBufferPackedStride(const EmitterState& state, const IR::MemoryInfo& mem,
                                    uint32_t use_pc);
 
-uint32_t StorageBufferFormat(const EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc);
+Prospero::BufferFormat StorageBufferFormat(const EmitterState& state, const IR::MemoryInfo& mem,
+                                           uint32_t use_pc);
 
 bool ShouldApplyBufferAddTid(const IR::Instruction& inst);
 
@@ -1015,6 +1015,9 @@ 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 +1348,7 @@ 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 +1580,22 @@ 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

@@ -153,8 +153,8 @@ uint32_t StorageBufferPackedStride(const EmitterState& state, const IR::MemoryIn
 	return state.program.info.buffers[mem.resource].packed_stride;
 }
 
-uint32_t StorageBufferFormat(const EmitterState& state, const IR::MemoryInfo& mem,
-                             uint32_t use_pc) {
+Prospero::BufferFormat StorageBufferFormat(const EmitterState& state, const IR::MemoryInfo& mem,
+                                           uint32_t use_pc) {
 	(void)use_pc;
 	if (mem.resource >= state.program.info.buffers.size()) {
 		ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::Buffers, mem.resource,
@@ -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,
@@ -748,8 +749,7 @@ Prospero::BufferFormat FormattedBufferFormat(const EmitterState&    state,
                                              const IR::Instruction& inst) {
 	return inst.memory.typed
 	           ? Format::DecodeTBufferFormat(inst.memory.data_format, inst.memory.number_format)
-	           : static_cast<Prospero::BufferFormat>(
-	                 StorageBufferFormat(state, inst.memory, inst.pc));
+	           : StorageBufferFormat(state, inst.memory, inst.pc);
 }
 
 IR::Instruction WithFormatComponentByteOffset(const IR::Instruction& inst,
@@ -1181,6 +1181,79 @@ 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/emitter/spirvEmitterModule.cpp

@@ -64,26 +64,26 @@ VertexInputScalarKind VertexParameterScalarKind(const EmitterState& state, uint3
 	}
 
 	switch (state.vertex_input_info->resources[location].Format()) {
-		case 5:
-		case 11:
-		case 18:
-		case 20:
-		case 27:
-		case 60:
-		case 62:
-		case 69:
-		case 72:
-		case 75: return VertexInputScalarKind::Uint;
-		case 6:
-		case 12:
-		case 19:
-		case 21:
-		case 28:
-		case 61:
-		case 63:
-		case 70:
-		case 73:
-		case 76: return VertexInputScalarKind::Sint;
+		case Prospero::BufferFormat::k8UInt:
+		case Prospero::BufferFormat::k16UInt:
+		case Prospero::BufferFormat::k8_8UInt:
+		case Prospero::BufferFormat::k32UInt:
+		case Prospero::BufferFormat::k16_16UInt:
+		case Prospero::BufferFormat::k8_8_8_8UInt:
+		case Prospero::BufferFormat::k32_32UInt:
+		case Prospero::BufferFormat::k16_16_16_16UInt:
+		case Prospero::BufferFormat::k32_32_32UInt:
+		case Prospero::BufferFormat::k32_32_32_32UInt: return VertexInputScalarKind::Uint;
+		case Prospero::BufferFormat::k8SInt:
+		case Prospero::BufferFormat::k16SInt:
+		case Prospero::BufferFormat::k8_8SInt:
+		case Prospero::BufferFormat::k32SInt:
+		case Prospero::BufferFormat::k16_16SInt:
+		case Prospero::BufferFormat::k8_8_8_8SInt:
+		case Prospero::BufferFormat::k32_32SInt:
+		case Prospero::BufferFormat::k16_16_16_16SInt:
+		case Prospero::BufferFormat::k32_32_32SInt:
+		case Prospero::BufferFormat::k32_32_32_32SInt: return VertexInputScalarKind::Sint;
 		default: return VertexInputScalarKind::Float;
 	}
 }

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

@@ -220,13 +220,13 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 				}
 				return false;
 			}
-			const auto format = (descriptor.dwords[1] >> 20u) & 0x1ffu;
-			const bool raw_sint_storage =
-			    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) &&
-			    !image.read && !image.atomic;
-			const bool uint_descriptor = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
-			const auto uint_program    = image.kind == ResourceKind::ImageUint ||
-			                             image.kind == ResourceKind::StorageImageUint;
+			const auto format =
+			    static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
+			const bool raw_sint_storage = storage && format == Prospero::BufferFormat::k32SInt &&
+			                              !image.read && !image.atomic;
+			const bool uint_descriptor  = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
+			const auto uint_program     = image.kind == ResourceKind::ImageUint ||
+			                              image.kind == ResourceKind::StorageImageUint;
 			if (uint_descriptor != uint_program && !(image.atomic && uint_program)) {
 				if (error != nullptr) {
 					*error = fmt::format(
@@ -415,12 +415,12 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 		    image.kind == ResourceKind::StorageImageUint) {
 			image.storage_swizzle = DescriptorImageSwizzle(descriptor);
 		}
-		const auto format  = (descriptor.dwords[1] >> 20u) & 0x1ffu;
+		const auto format =
+		    static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
 		const bool storage = image.kind == ResourceKind::StorageImage ||
 		                     image.kind == ResourceKind::StorageImageUint;
 		const bool raw_sint_storage =
-		    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) &&
-		    !image.read && !image.atomic;
+		    storage && format == Prospero::BufferFormat::k32SInt && !image.read && !image.atomic;
 		const bool uint_image = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
 		if (uint_image) {
 			switch (image.kind) {

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

@@ -3,355 +3,54 @@
 
 #include "common/common.h"
 #include "common/stringUtils.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/shader/recompiler/cfg/ShaderCFG.h"
 #include "graphics/shader/recompiler/decompiler/ShaderDecoder.h"
 #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 {
@@ -578,17 +277,17 @@ struct SrtPlan {
 struct BufferResource {
 	static constexpr uint32_t NoImageAlias = UINT32_MAX;
 
-	uint32_t source            = 0;
-	uint32_t first_use_pc      = 0;
-	uint32_t max_byte_extent   = 0;
-	uint32_t packed_stride     = 0;
-	uint32_t descriptor_format = 0;
-	uint32_t image_alias       = NoImageAlias;
-	bool     read              = false;
-	bool     written           = false;
-	bool     atomic            = false;
-	bool     formatted         = false;
-	bool     scalar            = false;
+	uint32_t               source            = 0;
+	uint32_t               first_use_pc      = 0;
+	uint32_t               max_byte_extent   = 0;
+	uint32_t               packed_stride     = 0;
+	Prospero::BufferFormat descriptor_format = Prospero::BufferFormat::kInvalid;
+	uint32_t               image_alias       = NoImageAlias;
+	bool                   read              = false;
+	bool                   written           = false;
+	bool                   atomic            = false;
+	bool                   formatted         = false;
+	bool                   scalar            = false;
 
 	bool operator==(const BufferResource& other) const = default;
 };

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);