kernel: do not release a waiter on a zero-count syncOnAddress wake (linux only) (#168)

* kernel: do not release a waiter on a zero-count syncOnAddress wake

sceKernelSyncOnAddressWake with a count of 0 must release no waiters. On
Linux it released exactly one.

WakeLinux passes the count straight to FUTEX_WAKE, which does not treat 0
as "wake nobody": the kernel's wake loop tests `if (++ret >= nr_wake) break`
after waking a waiter, so any waiter present is released before the loop can
break. A standalone futex program confirms it -- FUTEX_WAKE with val = 0
returns 1, and the waiter observes its wait return 0 with the word unchanged.

WakePortable already short-circuits count == 0, so the two backends of the
same ABI disagreed, with only the Linux one wrong.

TestWakeZeroIsNoOp in SyncOnAddressTests covers exactly this and fails on
Linux without the fix (15 consecutive failures on the assertion "zero-count
wake releases no waiters"); with the fix the suite passes 15 consecutive runs.

* kernel: shorten zero-count futex comment

---------

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

.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:
@@ -73,9 +91,13 @@ jobs:
       - name: Configure
         shell: cmd
         run: |
-          cmake -S src -B _Build/windows ^
+          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 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 "^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:
@@ -147,9 +170,13 @@ jobs:
       - name: Configure
         shell: bash
         run: |
-          cmake -S src -B _Build/macos \
+          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 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 '^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:
@@ -270,9 +298,13 @@ jobs:
         shell: bash
         run: |
           mkdir -p _Build
-          cmake -S src -B _Build/linux \
+          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 \
-              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 '^(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 \

.gitignore

@@ -3,4 +3,5 @@
 .idea/
 build/
 _Build/vscode-clang/
-_Build/
+_Build/
+cmake-build-debug/

.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

.vscode/settings.json

@@ -1,5 +1,5 @@
 {
-    "cmake.sourceDirectory": "${workspaceFolder}/src",
+    "cmake.sourceDirectory": "${workspaceFolder}",
     "cmake.buildDirectory": "${workspaceFolder}/_Build/vscode-clang",
     "cmake.generator": "Ninja",
     "cmake.environment": {

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

@@ -4,8 +4,18 @@ 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 "${CMAKE_CURRENT_SOURCE_DIR}/src")
+set(KYTY_TESTS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/tests")
+set(KYTY_THIRD_PARTY_DIR "${CMAKE_CURRENT_SOURCE_DIR}/3rdparty")
+
 if(CMAKE_SYSTEM_NAME MATCHES ".*Linux")
 	set(LINUX TRUE)
 endif()
@@ -19,12 +29,33 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 set(CMAKE_CXX_SCAN_FOR_MODULES OFF)
 
-include(utils.cmake)
+include("${KYTY_SOURCE_DIR}/utils.cmake")
 include(CTest)
 
 option(KYTY_ENABLE_CLANG_TIDY "Run clang-tidy checks during builds" OFF)
 
-set(KYTY_THIRD_PARTY_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../3rdparty")
+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)
 
@@ -97,36 +128,40 @@ endif()
 
 project(Kyty${KYTY_PROJECT_NAME}${CMAKE_BUILD_TYPE}${KYTY_COMPILER} VERSION 0.2.2)
 
-include_directories(	
+include_directories(
 	${KYTY_THIRD_PARTY_DIR}/gtest/include
 	${KYTY_THIRD_PARTY_DIR}/gtest
 	${KYTY_THIRD_PARTY_DIR}/fmt/include
 	${KYTY_THIRD_PARTY_DIR}/magic_enum/include/magic_enum
 	${PROJECT_BINARY_DIR}
-	${CMAKE_CURRENT_SOURCE_DIR}
+	${KYTY_SOURCE_DIR}
 )
 
 set(KYTY_VERSION "${PROJECT_VERSION}")
 
 configure_file(
-  ${PROJECT_SOURCE_DIR}/cmake_config.h.in
+  ${KYTY_SOURCE_DIR}/cmake_config.h.in
   ${PROJECT_BINARY_DIR}/cmake_config.h
   )
 
 find_package(Git)
-  
-add_custom_target( KytyGitVersion
-    COMMAND ${CMAKE_COMMAND} 
-	-D INPUT_FILE=${CMAKE_CURRENT_SOURCE_DIR}/kytyGitVersion.h.in
-    -D OUTPUT_FILE=${CMAKE_CURRENT_BINARY_DIR}/kytyGitVersion.h
-	-D GIT_EXECUTABLE=${GIT_EXECUTABLE}
-	-D GIT_WORKING_DIRECTORY=${CMAKE_CURRENT_SOURCE_DIR}
-	-P ${CMAKE_CURRENT_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 
+	list(APPEND KYTY_IWYU
 		kyty_emulator
 		common
 		#launcher
@@ -139,55 +174,55 @@ if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 12.0.0)
 endif()
 
 option(KYTY_BUILD_LAUNCHER "Build Qt launcher" ON)
-  
-config_compiler_and_linker()  
-  
+
+config_compiler_and_linker()
+
 add_subdirectory("${KYTY_THIRD_PARTY_DIR}" "${CMAKE_CURRENT_BINARY_DIR}/3rdparty")
-add_subdirectory(common)
+add_subdirectory("${KYTY_SOURCE_DIR}/common" "${CMAKE_CURRENT_BINARY_DIR}/common")
 
 file(GLOB kyty_emulator_src CONFIGURE_DEPENDS
-	libs/*.cpp
-	libs/*.h
-	graphics/*.cpp
-	graphics/*.h
-	graphics/guest_gpu/*.cpp
-	graphics/guest_gpu/*.h
-	graphics/guest_gpu/command_processor/*.cpp
-	graphics/guest_gpu/command_processor/*.h
-	graphics/host_gpu/*.cpp
-	graphics/host_gpu/*.h
-	graphics/host_gpu/renderer/*.cpp
-	graphics/host_gpu/renderer/*.h
-	graphics/host_gpu/renderer/cache/*.cpp
-	graphics/host_gpu/renderer/cache/*.h
-	graphics/host_gpu/renderer/image/*.cpp
-	graphics/host_gpu/renderer/image/*.h
-	graphics/host_gpu/renderer/pipeline/*.cpp
-	graphics/host_gpu/renderer/pipeline/*.h
-	graphics/shader/*.cpp
-	graphics/shader/*.h
-	graphics/shader/recompiler/*.cpp
-	graphics/shader/recompiler/*.h
-	graphics/shader/recompiler/cfg/*.cpp
-	graphics/shader/recompiler/cfg/*.h
-	graphics/shader/recompiler/decompiler/*.cpp
-	graphics/shader/recompiler/decompiler/*.h
-	graphics/shader/recompiler/emitter/*.cpp
-	graphics/shader/recompiler/emitter/*.h
-	graphics/shader/recompiler/ir/*.cpp
-	graphics/shader/recompiler/ir/*.h
-	graphics/presentation/*.cpp
-	graphics/presentation/*.h
-	graphics/presentation/window/*.cpp
-	graphics/presentation/window/*.h
-	kernel/*.cpp
-	kernel/*.h
-	loader/*.cpp
-	loader/*.h
+	"${KYTY_SOURCE_DIR}/libs/*.cpp"
+	"${KYTY_SOURCE_DIR}/libs/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/command_processor/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/command_processor/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/cache/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/cache/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/image/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/image/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/pipeline/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/pipeline/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/cfg/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/cfg/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/decompiler/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/decompiler/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/emitter/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/emitter/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/presentation/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/presentation/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/presentation/window/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/presentation/window/*.h"
+	"${KYTY_SOURCE_DIR}/kernel/*.cpp"
+	"${KYTY_SOURCE_DIR}/kernel/*.h"
+	"${KYTY_SOURCE_DIR}/loader/*.cpp"
+	"${KYTY_SOURCE_DIR}/loader/*.h"
 )
 
 find_program(KYTY_GLSLANG_VALIDATOR glslangValidator REQUIRED)
-set(gpu_tiler_shader_dir "${CMAKE_CURRENT_SOURCE_DIR}/graphics/host_gpu/shaders")
+set(gpu_tiler_shader_dir "${KYTY_SOURCE_DIR}/graphics/host_gpu/shaders")
 set(gpu_tiler_generated_dir "${PROJECT_BINARY_DIR}/gpu_tiler_shaders")
 set(gpu_tiler_shader_names
 	standard256
@@ -215,7 +250,7 @@ foreach(shader_name IN LISTS gpu_tiler_shader_names)
 		COMMAND "${KYTY_GLSLANG_VALIDATOR}" -V --target-env vulkan1.0 -Os
 		        "-I${gpu_tiler_shader_dir}" -o "${shader_spv}" "${shader_source}"
 		COMMAND ${CMAKE_COMMAND} -DINPUT=${shader_spv} -DOUTPUT=${shader_header}
-		        -DSYMBOL=${shader_symbol} -P "${CMAKE_CURRENT_SOURCE_DIR}/embed_spirv.cmake"
+		        -DSYMBOL=${shader_symbol} -P "${KYTY_SOURCE_DIR}/embed_spirv.cmake"
 		DEPENDS "${shader_source}" ${gpu_tiler_shader_includes}
 		VERBATIM
 	)
@@ -239,7 +274,7 @@ foreach(shader_source IN LISTS gpu_blit_shader_sources)
 		COMMAND "${KYTY_GLSLANG_VALIDATOR}" -V --target-env vulkan1.0 -Os
 		        "-I${gpu_tiler_shader_dir}" -o "${shader_spv}" "${shader_source}"
 		COMMAND ${CMAKE_COMMAND} -DINPUT=${shader_spv} -DOUTPUT=${shader_header}
-		        -DSYMBOL=${shader_symbol} -P "${CMAKE_CURRENT_SOURCE_DIR}/embed_spirv.cmake"
+		        -DSYMBOL=${shader_symbol} -P "${KYTY_SOURCE_DIR}/embed_spirv.cmake"
 		DEPENDS "${shader_source}"
 		VERBATIM
 	)
@@ -248,8 +283,8 @@ endforeach()
 list(APPEND kyty_emulator_src ${gpu_blit_shader_headers})
 
 list(APPEND kyty_emulator_src
-	emulator.h
-	emulator.cpp
+	"${KYTY_SOURCE_DIR}/emulator.h"
+	"${KYTY_SOURCE_DIR}/emulator.cpp"
 )
 list(REMOVE_DUPLICATES kyty_emulator_src)
 
@@ -257,7 +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)
@@ -280,7 +315,7 @@ if(LINUX)
 endif()
 
 set(inc_headers
-	${CMAKE_CURRENT_SOURCE_DIR}
+	${KYTY_SOURCE_DIR}
 	${KYTY_THIRD_PARTY_DIR}/SDL2/include
 	${KYTY_THIRD_PARTY_DIR}/VulkanMemoryAllocator/include
 	${KYTY_THIRD_PARTY_DIR}/SPIRV-Tools/include
@@ -292,13 +327,13 @@ set(inc_headers
 
 if (KYTY_CLANG_CL)
 	list(APPEND kyty_emulator_link_libraries winpthread)
-	list(APPEND inc_headers 
+	list(APPEND inc_headers
 		${KYTY_THIRD_PARTY_DIR}/winpthread/include
 	)
 endif()
 
 list(APPEND check_headers
-	${CMAKE_CURRENT_SOURCE_DIR}
+	${KYTY_SOURCE_DIR}
 )
 
 function(add_kyty_full_emulator_test target source)
@@ -320,95 +355,114 @@ endfunction()
 function(configure_macos_guest_address_space target)
 	if(APPLE AND (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR
 			(NOT CMAKE_OSX_ARCHITECTURES AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")))
-		target_sources(${target} PRIVATE kernel/macosGuestAddressSpace.cpp)
+		target_sources(${target} PRIVATE "${KYTY_SOURCE_DIR}/kernel/macosGuestAddressSpace.cpp")
 		target_compile_definitions(${target} PRIVATE KYTY_LINKED_GUEST_ADDRESS_SPACE=1)
 		target_link_options(${target} PRIVATE
 			-Wl,-ld_classic,-no_pie,-no_fixup_chains,-no_huge,-pagezero_size,0x40000,-segaddr,SYSTEM_MANAGED,0x40000,-segaddr,SYSTEM_RESERVED,0x7ffffc000,-segaddr,USER_AREA,0x7000000000,-image_base,0x700000000000)
 	endif()
 endfunction()
 
-add_kyty_full_emulator_test(shader_cfg_tests ../tests/shaderCfgTests.cpp)
+add_kyty_full_emulator_test(shader_cfg_tests "${KYTY_TESTS_DIR}/shaderCfgTests.cpp")
 
 add_executable(scalar_provenance_tests EXCLUDE_FROM_ALL
-	../tests/ScalarProvenanceTests.cpp
-	graphics/host_gpu/hostMemory.cpp
-	graphics/shader/recompiler/ir/ReadLaneElimination.cpp
-	graphics/shader/recompiler/ir/ScalarProvenance.cpp
-	graphics/shader/recompiler/ir/SrtWalker.cpp
+	"${KYTY_TESTS_DIR}/ScalarProvenanceTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ReadLaneElimination.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
 )
 target_link_libraries(scalar_provenance_tests fmt::fmt)
 target_include_directories(scalar_provenance_tests PRIVATE ${inc_headers})
 
 add_executable(page_manager_tests EXCLUDE_FROM_ALL
-	../tests/PageManagerTests.cpp
-	graphics/host_gpu/pageManager.cpp
+	"${KYTY_TESTS_DIR}/PageManagerTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/pageManager.cpp"
 )
 target_include_directories(page_manager_tests PRIVATE ${inc_headers})
 
+add_executable(bit_array_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/BitArrayTests.cpp"
+)
+target_include_directories(bit_array_tests PRIVATE ${inc_headers})
+
+add_executable(ime_dialog_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/ImeDialogTests.cpp"
+	"${KYTY_SOURCE_DIR}/libs/imeDialog.cpp"
+)
+target_include_directories(ime_dialog_tests PRIVATE ${inc_headers})
+
 add_executable(memory_tracker_tests EXCLUDE_FROM_ALL
-	../tests/MemoryTrackerTests.cpp
-	graphics/host_gpu/pageManager.cpp
-	graphics/host_gpu/memoryTracker.cpp
+	"${KYTY_TESTS_DIR}/MemoryTrackerTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/pageManager.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/memoryTracker.cpp"
 )
 target_link_libraries(memory_tracker_tests fmt::fmt common)
 target_include_directories(memory_tracker_tests PRIVATE ${inc_headers})
 
 add_executable(shader_vertex_metadata_tests EXCLUDE_FROM_ALL
-	../tests/ShaderVertexMetadataTests.cpp
-	graphics/host_gpu/hostMemory.cpp
-	graphics/shader/shaderVertexMetadata.cpp
+	"${KYTY_TESTS_DIR}/ShaderVertexMetadataTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/shaderVertexMetadata.cpp"
 )
 target_include_directories(shader_vertex_metadata_tests PRIVATE ${inc_headers})
 
 add_executable(shader_stage_runtime_tests EXCLUDE_FROM_ALL
-	../tests/ShaderStageRuntimeTests.cpp
-	graphics/guest_gpu/gpu_format.cpp
-	graphics/host_gpu/hostMemory.cpp
-	graphics/shader/shaderStageRuntime.cpp
-	graphics/shader/recompiler/ir/ResourceMaterialization.cpp
-	graphics/shader/recompiler/ir/ScalarProvenance.cpp
-	graphics/shader/recompiler/ir/SrtWalker.cpp
+	"${KYTY_TESTS_DIR}/ShaderStageRuntimeTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/shaderStageRuntime.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceMaterialization.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
 )
 target_link_libraries(shader_stage_runtime_tests fmt::fmt)
 target_include_directories(shader_stage_runtime_tests PRIVATE ${inc_headers})
 
 add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
-	../tests/ResourceTrackingTests.cpp
-	graphics/guest_gpu/gpu_format.cpp
-	graphics/host_gpu/hostMemory.cpp
-	graphics/shader/recompiler/ir/ScalarProvenance.cpp
-	graphics/shader/recompiler/ir/SrtWalker.cpp
-	graphics/shader/recompiler/ir/SrtPatcher.cpp
-	graphics/shader/recompiler/ir/ResourceTracking.cpp
-	graphics/shader/recompiler/ir/ResourceMaterialization.cpp
-	graphics/shader/recompiler/ir/ShaderInfoCollection.cpp
-	graphics/shader/recompiler/ir/BindingLayout.cpp
+	"${KYTY_TESTS_DIR}/ResourceTrackingTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtPatcher.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceTracking.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceMaterialization.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ShaderInfoCollection.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/BindingLayout.cpp"
 )
 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
-	../tests/ResourceMutexTests.cpp
-	graphics/host_gpu/renderer/cache/resourceMutex.cpp
+add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/AudioOut2PortTests.cpp"
+	"${KYTY_SOURCE_DIR}/libs/libAudio2.cpp"
+	"${KYTY_SOURCE_DIR}/loader/timer.cpp"
 )
-target_link_libraries(resource_mutex_tests common)
-target_include_directories(resource_mutex_tests PRIVATE ${inc_headers})
+target_link_libraries(audio_out2_port_tests common fmt::fmt)
+target_include_directories(audio_out2_port_tests PRIVATE ${inc_headers})
 
 add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
-	../tests/EventQueueLifetimeTests.cpp
-	kernel/eventQueue.cpp
-	loader/timer.cpp
+	"${KYTY_TESTS_DIR}/EventQueueLifetimeTests.cpp"
+	"${KYTY_SOURCE_DIR}/kernel/eventQueue.cpp"
+	"${KYTY_SOURCE_DIR}/loader/timer.cpp"
 )
 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 ${inc_headers})
+
 add_executable(image_page_table_tests EXCLUDE_FROM_ALL
-	../tests/ImagePageTableTests.cpp
+	"${KYTY_TESTS_DIR}/ImagePageTableTests.cpp"
 )
 target_link_libraries(image_page_table_tests common fmt::fmt)
 target_include_directories(image_page_table_tests PRIVATE ${inc_headers})
 
-add_kyty_full_emulator_test(shader_recompiler_compute_tests ../tests/ShaderRecompilerComputeTests.cpp)
+add_kyty_full_emulator_test(shader_recompiler_compute_tests "${KYTY_TESTS_DIR}/ShaderRecompilerComputeTests.cpp")
 
 set(gpu_test_generated_dir "${PROJECT_BINARY_DIR}/gpu_test_shaders")
 set(gpu_test_ms_depth_source
@@ -424,14 +478,14 @@ add_custom_command(
 	        -o "${gpu_test_ms_depth_spv}" "${gpu_test_ms_depth_source}"
 	COMMAND ${CMAKE_COMMAND} -DINPUT=${gpu_test_ms_depth_spv}
 	        -DOUTPUT=${gpu_test_ms_depth_header} -DSYMBOL=GPU_TEST_MS_DEPTH_SPV
-	        -P "${CMAKE_CURRENT_SOURCE_DIR}/embed_spirv.cmake"
+	        -P "${KYTY_SOURCE_DIR}/embed_spirv.cmake"
 	DEPENDS "${gpu_test_ms_depth_source}"
 	VERBATIM
 )
 target_sources(shader_recompiler_compute_tests PRIVATE
 	"${gpu_test_ms_depth_header}")
 
-add_kyty_full_emulator_test(virtual_memory_allocation_tests ../tests/VirtualMemoryAllocationTests.cpp)
+add_kyty_full_emulator_test(virtual_memory_allocation_tests "${KYTY_TESTS_DIR}/VirtualMemoryAllocationTests.cpp")
 target_compile_definitions(virtual_memory_allocation_tests PRIVATE
 	KYTY_VIRTUAL_MEMORY_ALLOCATION_TESTS=1)
 
@@ -444,21 +498,32 @@ 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>)
 	add_test(NAME memory_tracker COMMAND $<TARGET_FILE:memory_tracker_tests>)
 	add_test(NAME page_manager COMMAND $<TARGET_FILE:page_manager_tests>)
-	add_test(NAME resource_mutex COMMAND $<TARGET_FILE:resource_mutex_tests>)
+	add_test(NAME bit_array COMMAND $<TARGET_FILE:bit_array_tests>)
+	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 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)
 
@@ -482,10 +547,27 @@ if(BUILD_TESTING)
 		add_test(NAME buffer_cache_ranges
 			COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --buffer-cache-range-only)
 	endif()
+
+	add_custom_target(kyty_tests DEPENDS
+		ime_dialog_tests
+		shader_cfg_tests
+		scalar_provenance_tests
+		image_page_table_tests
+		memory_tracker_tests
+		page_manager_tests
+		bit_array_tests
+		shader_vertex_metadata_tests
+		shader_stage_runtime_tests
+		resource_tracking_tests
+		event_queue_lifetime_tests
+		sync_on_address_tests
+		shader_recompiler_compute_tests
+		virtual_memory_allocation_tests
+	)
 endif()
 
 
-add_executable(kyty_emulator main.cpp ${kyty_emulator_src})
+add_executable(kyty_emulator "${KYTY_SOURCE_DIR}/main.cpp" ${kyty_emulator_src})
 configure_macos_guest_address_space(kyty_emulator)
 
 target_link_libraries(kyty_emulator ${kyty_emulator_link_libraries})
@@ -526,8 +608,6 @@ elseif(WIN32 OR LINUX)
 	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${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
@@ -535,14 +615,14 @@ if(APPLE)
 	# reverts to the hardened defaults and aborts when it first executes written code).
 	add_custom_command(TARGET kyty_emulator POST_BUILD
 		COMMAND codesign -s - --force --options runtime
-			--entitlements "${CMAKE_CURRENT_SOURCE_DIR}/macos_jit.entitlements"
+			--entitlements "${KYTY_SOURCE_DIR}/macos_jit.entitlements"
 			$<TARGET_FILE:kyty_emulator>
 		COMMENT "Codesign kyty_emulator with JIT entitlements (macOS)")
 endif()
 
 install(TARGETS kyty_emulator DESTINATION .)
 if(KYTY_BUILD_LAUNCHER)
-	add_subdirectory(launcher)
+	add_subdirectory("${KYTY_SOURCE_DIR}/launcher" "${CMAKE_CURRENT_BINARY_DIR}/launcher")
 endif()
 if(KYTY_CLANG_CL)
 	install(FILES "${KYTY_THIRD_PARTY_DIR}/winpthread/bin/libwinpthread-1.dll" DESTINATION .)

README.md

@@ -141,7 +141,7 @@ git submodule update --init --recursive
 Configure the project. Replace the Qt path with the version installed on your system:
 
 ```powershell
-cmake -S src -B _Build/windows -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_PREFIX_PATH="C:/Qt/6.x.x/msvc2022_64"
+cmake -S . -B _Build/windows -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_PREFIX_PATH="C:/Qt/6.x.x/msvc2022_64"
 ```
 
 Build the launcher and stage a runnable installation:
@@ -174,7 +174,7 @@ Qt 6 (Concurrent, Network, Widgets) is also required — either the distribution
 ```bash
 git submodule update --init --recursive
 
-cmake -S src -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \
+cmake -S . -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \
   -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
   -DCMAKE_PREFIX_PATH="$Qt6_DIR"
 
@@ -188,7 +188,7 @@ The install step copies the Qt libraries and plugins next to the binaries, so
 As on Windows, the MSVC compiler is not used; Clang is required. `cl.exe` is rejected at configure
 time.
 
-Note that the CMake source root is `src`, not the repository root.
+The CMake source root is the repository root.
 
 ### Building on macOS
 
@@ -207,7 +207,7 @@ Requirements:
 ```bash
 git submodule update --init --recursive
 
-cmake -S src -B _Build/macos -G Ninja -DCMAKE_BUILD_TYPE=Release \
+cmake -S . -B _Build/macos -G Ninja -DCMAKE_BUILD_TYPE=Release \
   -DCMAKE_OSX_ARCHITECTURES=x86_64 \
   -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
   -DCMAKE_PREFIX_PATH="$Qt6_DIR"
@@ -229,6 +229,17 @@ codesign --force --sign - _Build/macos/install/libMoltenVK.dylib
 
 Release archives already include a signed `libMoltenVK.dylib`.
 
+### Regression tests
+
+Build every regression executable and run the registered tests with:
+
+```powershell
+cmake --build _Build/windows --target kyty_tests
+ctest --test-dir _Build/windows --output-on-failure
+```
+
+Use `_Build/linux` instead of `_Build/windows` for a Linux build.
+
 ### Visual Studio Code
 
 A ready-made Visual Studio Code setup is included in [`.vscode`](.vscode). It configures CMake

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

@@ -13,7 +13,7 @@ endif()
 add_library(common STATIC ${common_src} ${common_headers})
 
 target_include_directories(common
-	PUBLIC ${PROJECT_SOURCE_DIR}
+	PUBLIC ${KYTY_SOURCE_DIR}
 	PRIVATE
 		${KYTY_THIRD_PARTY_DIR}/cpuinfo/include
 )
@@ -24,6 +24,7 @@ target_link_libraries(common
 		Tracy::TracyClient
 	PRIVATE
 		cpuinfo
+		kyty_git_version
 		spdlog::spdlog
 )
 

src/common/assert.cpp

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

src/common/bitArray.h

@@ -0,0 +1,260 @@
+#ifndef EMULATOR_SRC_COMMON_BITARRAY_H_
+#define EMULATOR_SRC_COMMON_BITARRAY_H_
+
+#include <array>
+#include <bit>
+#include <cstddef>
+#include <cstdint>
+#include <iterator>
+#include <utility>
+
+namespace Common {
+
+template <size_t N>
+class BitArray final {
+	static_assert(N != 0, "BitArray size must be nonzero");
+	static_assert(N % 64 == 0, "BitArray size must be a multiple of 64 bits");
+
+	static constexpr size_t BITS_PER_WORD = 64;
+	static constexpr size_t WORD_COUNT    = N / BITS_PER_WORD;
+
+public:
+	using Range = std::pair<size_t, size_t>;
+
+	class Iterator final {
+	public:
+		using iterator_category = std::forward_iterator_tag;
+		using value_type        = Range;
+		using difference_type   = std::ptrdiff_t;
+		using pointer           = const Range*;
+		using reference         = const Range&;
+
+		Iterator(const BitArray& bits, size_t start)
+		    : m_bits(bits), m_range(bits.FirstRangeFrom(start)) {}
+
+		Iterator& operator++() {
+			m_range = m_bits.FirstRangeFrom(m_range.second);
+			return *this;
+		}
+
+		[[nodiscard]] bool operator==(const Iterator& other) const {
+			return &m_bits == &other.m_bits && m_range == other.m_range;
+		}
+
+		[[nodiscard]] bool operator!=(const Iterator& other) const { return !(*this == other); }
+
+		[[nodiscard]] reference operator*() const { return m_range; }
+		[[nodiscard]] pointer   operator->() const { return &m_range; }
+
+	private:
+		const BitArray& m_bits;
+		Range           m_range;
+	};
+
+	using const_iterator = Iterator;
+
+	constexpr BitArray() = default;
+
+	constexpr BitArray(const BitArray& other, size_t start, size_t end) {
+		if (start >= end || end > N) {
+			return;
+		}
+
+		const auto first_word = start / BITS_PER_WORD;
+		const auto last_word  = (end - 1) / BITS_PER_WORD;
+		const auto start_bit  = start % BITS_PER_WORD;
+		const auto end_bit    = (end - 1) % BITS_PER_WORD;
+		const auto start_mask = ~uint64_t {0} << start_bit;
+		const auto end_mask =
+		    end_bit == BITS_PER_WORD - 1 ? ~uint64_t {0} : (uint64_t {1} << (end_bit + 1)) - 1;
+
+		if (first_word == last_word) {
+			m_data[first_word] = other.m_data[first_word] & start_mask & end_mask;
+			return;
+		}
+
+		m_data[first_word] = other.m_data[first_word] & start_mask;
+		for (auto word = first_word + 1; word < last_word; word++) {
+			m_data[word] = other.m_data[word];
+		}
+		m_data[last_word] = other.m_data[last_word] & end_mask;
+	}
+
+	[[nodiscard]] constexpr bool Get(size_t index) const {
+		return (m_data[index / BITS_PER_WORD] & (uint64_t {1} << (index % BITS_PER_WORD))) != 0;
+	}
+
+	constexpr void Set(size_t index) {
+		m_data[index / BITS_PER_WORD] |= uint64_t {1} << (index % BITS_PER_WORD);
+	}
+
+	constexpr void Unset(size_t index) {
+		m_data[index / BITS_PER_WORD] &= ~(uint64_t {1} << (index % BITS_PER_WORD));
+	}
+
+	constexpr void SetRange(size_t start, size_t end) {
+		if (start >= end || end > N) {
+			return;
+		}
+
+		const auto first_word = start / BITS_PER_WORD;
+		const auto last_word  = (end - 1) / BITS_PER_WORD;
+		const auto start_bit  = start % BITS_PER_WORD;
+		const auto end_bit    = (end - 1) % BITS_PER_WORD;
+		const auto start_mask = ~uint64_t {0} << start_bit;
+		const auto end_mask =
+		    end_bit == BITS_PER_WORD - 1 ? ~uint64_t {0} : (uint64_t {1} << (end_bit + 1)) - 1;
+
+		if (first_word == last_word) {
+			m_data[first_word] |= start_mask & end_mask;
+			return;
+		}
+
+		m_data[first_word] |= start_mask;
+		for (auto word = first_word + 1; word < last_word; word++) {
+			m_data[word] = ~uint64_t {0};
+		}
+		m_data[last_word] |= end_mask;
+	}
+
+	constexpr void UnsetRange(size_t start, size_t end) {
+		if (start >= end || end > N) {
+			return;
+		}
+
+		const auto first_word = start / BITS_PER_WORD;
+		const auto last_word  = (end - 1) / BITS_PER_WORD;
+		const auto start_bit  = start % BITS_PER_WORD;
+		const auto end_bit    = (end - 1) % BITS_PER_WORD;
+		const auto start_mask = (uint64_t {1} << start_bit) - 1;
+		const auto end_mask =
+		    end_bit == BITS_PER_WORD - 1 ? uint64_t {0} : ~((uint64_t {1} << (end_bit + 1)) - 1);
+
+		if (first_word == last_word) {
+			m_data[first_word] &= start_mask | end_mask;
+			return;
+		}
+
+		m_data[first_word] &= start_mask;
+		for (auto word = first_word + 1; word < last_word; word++) {
+			m_data[word] = 0;
+		}
+		m_data[last_word] &= end_mask;
+	}
+
+	constexpr void Clear() { m_data.fill(0); }
+	constexpr void Fill() { m_data.fill(~uint64_t {0}); }
+
+	[[nodiscard]] constexpr bool None() const {
+		uint64_t combined = 0;
+		for (const auto word: m_data) {
+			combined |= word;
+		}
+		return combined == 0;
+	}
+
+	[[nodiscard]] constexpr bool Any() const { return !None(); }
+
+	[[nodiscard]] constexpr Range FirstRangeFrom(size_t start) const {
+		if (start >= N) {
+			return {N, N};
+		}
+
+		auto word_index = start / BITS_PER_WORD;
+		auto word       = m_data[word_index] & (~uint64_t {0} << (start % BITS_PER_WORD));
+		while (word == 0) {
+			word_index++;
+			if (word_index == WORD_COUNT) {
+				return {N, N};
+			}
+			word = m_data[word_index];
+		}
+
+		const auto first     = word_index * BITS_PER_WORD + std::countr_zero(word);
+		const auto first_bit = first % BITS_PER_WORD;
+		const auto first_ones =
+		    static_cast<size_t>(std::countr_one(m_data[word_index] >> first_bit));
+		if (first_bit + first_ones < BITS_PER_WORD) {
+			return {first, first + first_ones};
+		}
+
+		for (word_index++; word_index < WORD_COUNT; word_index++) {
+			word = m_data[word_index];
+			if (word != ~uint64_t {0}) {
+				return {first, word_index * BITS_PER_WORD + std::countr_one(word)};
+			}
+		}
+		return {first, N};
+	}
+
+	[[nodiscard]] constexpr Range FirstRange() const { return FirstRangeFrom(0); }
+
+	[[nodiscard]] constexpr Range LastRangeFrom(size_t end) const {
+		if (end == 0) {
+			return {0, 0};
+		}
+		if (end > N) {
+			end = N;
+		}
+
+		auto       word_index = (end - 1) / BITS_PER_WORD;
+		const auto end_bit    = (end - 1) % BITS_PER_WORD;
+		const auto end_mask =
+		    end_bit == BITS_PER_WORD - 1 ? ~uint64_t {0} : (uint64_t {1} << (end_bit + 1)) - 1;
+		auto word = m_data[word_index] & end_mask;
+		while (word == 0) {
+			if (word_index == 0) {
+				return {0, 0};
+			}
+			word = m_data[--word_index];
+		}
+
+		const auto empty_bits = static_cast<size_t>(std::countl_zero(word));
+		const auto ones       = static_cast<size_t>(std::countl_one(word << empty_bits));
+		const auto last       = (word_index + 1) * BITS_PER_WORD - empty_bits;
+		if (empty_bits + ones < BITS_PER_WORD) {
+			return {last - ones, last};
+		}
+
+		while (word_index != 0) {
+			word = m_data[--word_index];
+			if (word != ~uint64_t {0}) {
+				return {(word_index + 1) * BITS_PER_WORD - std::countl_one(word), last};
+			}
+		}
+		return {0, last};
+	}
+
+	[[nodiscard]] constexpr Range LastRange() const { return LastRangeFrom(N); }
+
+	[[nodiscard]] const_iterator begin() const { return Iterator(*this, 0); }
+	[[nodiscard]] const_iterator end() const { return Iterator(*this, N); }
+
+	constexpr BitArray& operator^=(const BitArray& other) {
+		for (size_t word = 0; word < WORD_COUNT; word++) {
+			m_data[word] ^= other.m_data[word];
+		}
+		return *this;
+	}
+
+	[[nodiscard]] constexpr BitArray operator^(const BitArray& other) const {
+		auto result = *this;
+		result ^= other;
+		return result;
+	}
+
+	[[nodiscard]] constexpr BitArray operator~() const {
+		auto result = *this;
+		for (auto& word: result.m_data) {
+			word = ~word;
+		}
+		return result;
+	}
+
+private:
+	std::array<uint64_t, WORD_COUNT> m_data {};
+};
+
+} // namespace Common
+
+#endif // EMULATOR_SRC_COMMON_BITARRAY_H_

src/common/emulatorConfig.cpp

@@ -33,10 +33,18 @@ 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);
 }
 
+uint32_t GetConsoleLanguage() {
+	return g_config->console_language;
+}
+
 bool VulkanValidationEnabled() {
 	return g_config->vulkan_validation_enabled;
 }
@@ -89,12 +97,18 @@ bool RenderDocEnabled() {
 	return g_config->renderdoc_enabled;
 }
 
-bool NggRectlistDrawEnabled() {
-	return g_config->ngg_rectlist_draw_enabled;
-}
-
 bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled() {
+	return g_config->red_zone_protection_enabled;
+}
+#endif
+
+const Keymap& GetKeymap() {
+	return g_config->keymap;
+}
+
 } // namespace Config

src/common/emulatorConfig.h

@@ -5,6 +5,8 @@
 #include "common/subsystems.h"
 
 #include <filesystem>
+#include <string>
+#include <vector>
 
 namespace Config {
 
@@ -18,10 +20,17 @@ 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;
 	bool                   shader_validation_enabled   = false;
 	ShaderOptimizationType shader_optimization_type    = ShaderOptimizationType::None;
@@ -35,15 +44,20 @@ struct ConfigOptions {
 	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   spirv_debug_printf_enabled  = false;
 	bool                   renderdoc_enabled           = false;
-	bool                   ngg_rectlist_draw_enabled   = true;
 	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();
 
 bool                   ShaderValidationEnabled();
@@ -64,8 +78,12 @@ ProfilerDirection GetProfilerDirection();
 bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
-bool NggRectlistDrawEnabled();
 bool ReadbackLinearImagesEnabled();
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled();
+#endif
+
+const Keymap& GetKeymap();
 
 } // namespace Config
 

src/common/threads.cpp

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

src/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; }
@@ -83,8 +78,7 @@ public:
 	               uint32_t first_instance = 0);
 	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count, uint32_t flags);
 	void DrawIndexAuto(uint32_t index_count, uint32_t flags,
-	                   uint32_t render_target_slice_offset = 0, uint32_t instance_count = 1,
-	                   uint32_t first_vertex = 0, uint32_t first_instance = 0);
+	                   uint32_t render_target_slice_offset = 0);
 	void DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed);
 	void DrawIndirectMulti(uint32_t data_offset, uint32_t max_count_or_count,
 	                       const volatile uint32_t* count_addr, uint32_t stride_in_bytes,
@@ -152,6 +146,9 @@ private:
 	                      uint32_t interrupt_context_id);
 	void ProcessPm4(Pm4Execution& execution, size_t stop_depth);
 	void SuspendPm4();
+	void SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
+	                          uint32_t render_target_slice_offset, uint32_t first_vertex,
+	                          uint32_t first_instance);
 
 	CommandScheduler&    GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	RenderCommandBuffer& CurrentBuffer() { return GetScheduler().Current(); }
@@ -160,6 +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;
@@ -168,12 +167,12 @@ private:
 	uint64_t         m_index_base_addr                  = 0;
 	uint64_t         m_draw_indirect_args_base_addr     = 0;
 	uint64_t         m_dispatch_indirect_args_base_addr = 0;
-	uint32_t         m_num_instances                    = 1;
+	// 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                                                                    \
@@ -895,23 +897,44 @@ KYTY_HW_CTX_PARSER(HwCtxSetModeControl) {
 	return 1;
 }
 
-static void HwCtxIgnorePolyOffsetRegister(uint32_t cmd_offset, uint32_t value) {
-	static std::atomic<uint32_t> log_count = 0;
+static void HwCtxSetPolyOffsetRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_t value) {
+	HW::PolyOffset offset = cp.GetCtx().GetPolyOffset();
 
-	auto count = log_count.fetch_add(1);
-	if (count < 8) {
-		LOGF_COLOR(Log::Color::Red,
-		           "\t temporary: ignoring polygon offset context register 0x%03" PRIx32
-		           " = 0x%08" PRIx32 " (depth bias not implemented)\n",
-		           cmd_offset, value);
+	switch (cmd_offset) {
+		case Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL: {
+			const auto neg_bits =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, NEG_NUM_DB_BITS);
+			offset.neg_num_db_bits = std::bit_cast<int8_t>(static_cast<uint8_t>(neg_bits));
+			offset.db_is_float_fmt =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, DB_IS_FLOAT_FMT) != 0;
+			break;
+		}
+		case Pm4::PA_SU_POLY_OFFSET_CLAMP: offset.clamp = std::bit_cast<float>(value); break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE:
+			offset.front_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET:
+			offset.front_offset = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_SCALE:
+			offset.back_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET:
+			offset.back_offset = std::bit_cast<float>(value);
+			break;
+		default:
+			EXIT("unknown polygon offset context register 0x%03" PRIx32 " = 0x%08" PRIx32 "\n",
+			     cmd_offset, value);
 	}
+
+	cp.GetCtx().SetPolyOffset(offset);
 }
 
 KYTY_HW_CTX_PARSER(HwCtxSetPolyOffsetRegisters) {
 	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
 
 	for (uint32_t i = 0; i < num_values; i++) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset + i, buffer[i]);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset + i, buffer[i]);
 	}
 
 	return num_values;
@@ -1312,26 +1335,36 @@ static void HwShSetCsRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_VGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_VGPRS_MASK;
 			cs_regs.sgprs =
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_SGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_SGPRS_MASK;
-			cs_regs.bulky =
-			    (value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK;
-			cs_regs.wave_size = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
-			                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
-			                        ? 32u
-			                        : 64u;
+			cs_regs.float_mode    = (value >> Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_SHIFT) &
+			                        Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_MASK;
+			cs_regs.dx10_clamp    = ((value >> Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_MASK) != 0u;
+			cs_regs.ieee_mode     = ((value >> Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_MASK) != 0u;
+			cs_regs.fp16_overflow = ((value >> Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_MASK) != 0u;
+			cs_regs.bulky         = ((value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK) != 0u;
+			cs_regs.threadgroup_configuration = ((value >> Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT) &
+			                                     Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_MASK) != 0u;
+			cs_regs.wave_size                 = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
+			                                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
+			                                        ? 32u
+			                                        : 64u;
 			break;
 		case Pm4::COMPUTE_PGM_RSRC2:
-			cs_regs.scratch_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK;
+			cs_regs.scratch_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK) != 0u;
 			cs_regs.user_sgpr      = (value >> Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_MASK;
-			cs_regs.tgid_x_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK;
-			cs_regs.tgid_y_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK;
-			cs_regs.tgid_z_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK;
-			cs_regs.tg_size_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK;
+			cs_regs.tgid_x_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK) != 0u;
+			cs_regs.tgid_y_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK) != 0u;
+			cs_regs.tgid_z_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK) != 0u;
+			cs_regs.tg_size_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK) != 0u;
 			cs_regs.tidig_comp_cnt = (value >> Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_MASK;
 			cs_regs.lds_size       = (value >> Pm4::COMPUTE_PGM_RSRC2_LDS_SIZE_SHIFT) &
@@ -2102,11 +2135,24 @@ 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 +2682,21 @@ 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 +2704,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2665,10 +2714,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2682,17 +2727,6 @@ 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 +2741,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 
 	if (data_sel == 5) {
 		if (gl2_writeback) {
-			cp.SynchronizeGpu();
 			cache_action = 0;
 		}
 
@@ -2718,7 +2751,9 @@ 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;
 	}
@@ -3682,22 +3717,22 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_CLAMP] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_Z_INFO] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -4220,29 +4255,28 @@ 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 +4364,7 @@ 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 +4424,7 @@ 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);
 	};

src/graphics/guest_gpu/gpu_defs.h

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

src/graphics/guest_gpu/graphicsRun.cpp

@@ -32,11 +32,11 @@
 
 namespace Libs::Graphics {
 
-static thread_local CommandProcessor* g_current_processor      = nullptr;
-static thread_local Pm4Execution*     g_current_execution      = nullptr;
-static thread_local uint32_t          g_submission_pause_depth = 0;
-static thread_local bool              g_gpu_mutex_owned        = false;
-static thread_local bool              g_gpu_thread             = false;
+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:
@@ -98,14 +98,11 @@ public:
 	                   bool trigger_agc_interrupt_on_done);
 	void SubmitFlipPreparation(uint64_t request_id);
 	void Done();
-	void PauseSubmissions();
-	void ResumeSubmissions();
 	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:
@@ -127,9 +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);
@@ -142,6 +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;
@@ -156,6 +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() {
@@ -198,9 +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()) {
@@ -215,17 +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);
@@ -270,10 +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++;
@@ -283,11 +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) {
@@ -301,11 +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;
@@ -315,6 +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);
 }
@@ -353,7 +375,6 @@ void CommandProcessor::WaitDeDiff(uint32_t diff) {
 }
 
 void CommandProcessor::IncrementDe() {
-	BufferWait();
 	m_de_count++;
 }
 
@@ -407,7 +428,7 @@ void CommandProcessor::WriteData(uint32_t* dst, const uint32_t* src, uint32_t dw
                                  uint32_t write_control) {
 	const uint32_t dst_sel      = ((write_control >> 30u) & 0x1u) | ((write_control >> 7u) & 0x1eu);
 	const uint32_t cache_policy = (write_control >> 25u) & 0x3u;
-	const uint32_t increment     = (write_control >> 16u) & 0x1u;
+	const uint32_t increment    = (write_control >> 16u) & 0x1u;
 	const uint32_t write_confirm = (write_control >> 20u) & 0x1u;
 
 	switch (dst_sel) {
@@ -512,6 +533,7 @@ 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;
@@ -532,6 +554,7 @@ 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;
@@ -563,6 +586,7 @@ void GpuState::ThreadRun(void* data) {
 		}
 		if (should_stop) {
 			gpu->m_gfx_cp->BufferWait();
+			g_gpu_state  = nullptr;
 			g_gpu_thread = false;
 			return;
 		}
@@ -696,26 +720,6 @@ bool GpuState::Process(Submission& submission) {
 	return complete;
 }
 
-void GpuState::PauseSubmissions() {
-	if (g_gpu_mutex_owned) {
-		EXIT("GPU submissions are already paused by this thread\n");
-	}
-	g_gpu_mutex_owned = true;
-	m_submission_mutex.Lock();
-	if (!IsGpuThread()) {
-		WaitLocked();
-	}
-	m_renderer.GetCommandScheduler().DrainPriorityOperations();
-}
-
-void GpuState::ResumeSubmissions() {
-	if (!g_gpu_mutex_owned) {
-		EXIT("GPU submissions resumed without an active pause\n");
-	}
-	m_submission_mutex.Unlock();
-	g_gpu_mutex_owned = false;
-}
-
 Pm4ProcessResult CommandProcessor::Process(Pm4Execution& execution, uint32_t* buffer,
                                            uint32_t size_dw) {
 	KYTY_PROFILER_BLOCK("CommandProcessor::Process");
@@ -764,6 +768,9 @@ 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) {
@@ -1006,8 +1013,9 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 				     args.start_vertex_location, args.start_instance_location);
 			}
 		}
-		DrawIndexAuto(args.vertex_count_per_instance, 0, 0, args.instance_count,
-		              args.start_vertex_location, args.start_instance_location);
+		m_num_instances = args.instance_count;
+		SubmitNonIndexedDraw(args.vertex_count_per_instance, 0, 0, args.start_vertex_location,
+		                     args.start_instance_location);
 		return;
 	}
 
@@ -1047,6 +1055,7 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 		}
 	}
 
+	m_num_instances = args.instance_count;
 	DrawIndex(index_count, index_addr, 0, 1, args.instance_count, nullptr, 0,
 	          static_cast<int32_t>(args.base_vertex_location), args.start_instance_location);
 }
@@ -1104,8 +1113,9 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 					     args->start_vertex_location, args->start_instance_location);
 				}
 			}
-			DrawIndexAuto(args->vertex_count_per_instance, 0, 0, args->instance_count,
-			              args->start_vertex_location, args->start_instance_location);
+			m_num_instances = args->instance_count;
+			SubmitNonIndexedDraw(args->vertex_count_per_instance, 0, 0, args->start_vertex_location,
+			                     args->start_instance_location);
 			continue;
 		}
 
@@ -1146,6 +1156,7 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 			}
 		}
 
+		m_num_instances = args->instance_count;
 		DrawIndex(index_count, index_addr, 0, 1, args->instance_count, nullptr, 0,
 		          static_cast<int32_t>(args->base_vertex_location), args->start_instance_location);
 	}
@@ -1154,9 +1165,9 @@ 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();
@@ -1180,9 +1191,9 @@ 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)) {
@@ -1198,27 +1209,27 @@ 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) {
@@ -1237,12 +1248,17 @@ void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
 }
 
 void CommandProcessor::DrawIndexAuto(uint32_t index_count, uint32_t flags,
-                                     uint32_t render_target_slice_offset, uint32_t instance_count,
-                                     uint32_t first_vertex, uint32_t first_instance) {
+                                     uint32_t render_target_slice_offset) {
+	SubmitNonIndexedDraw(index_count, flags, render_target_slice_offset, 0, 0);
+}
+
+void CommandProcessor::SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
+                                            uint32_t render_target_slice_offset,
+                                            uint32_t first_vertex, uint32_t first_instance) {
 	CheckBuffer();
 
-	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), index_count, flags,
-	                                        render_target_slice_offset, instance_count,
+	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), vertex_count, flags,
+	                                        render_target_slice_offset, m_num_instances,
 	                                        first_vertex, first_instance);
 }
 
@@ -1325,6 +1341,9 @@ 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) {
@@ -1527,7 +1546,6 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
 				     event_index);
 			}
 			EmitGlobalBarrier();
-			SynchronizeGpu();
 			break;
 		// DbDataWritebackInvalidate, DbMetadataWritebackInvalidate, CbMetadataWritebackInvalidate.
 		case 0x0000002a:
@@ -1664,10 +1682,6 @@ 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);
@@ -1693,46 +1707,8 @@ int Gpu::GetFrameNum() const {
 	return m_state->GetFrameNum();
 }
 
-void Gpu::PauseSubmissions() {
-	m_state->PauseSubmissions();
-}
-
-void Gpu::ResumeSubmissions() {
-	m_state->ResumeSubmissions();
-}
-
-Gpu::SubmissionLock::SubmissionLock(Gpu& gpu): m_gpu(gpu) {
-	if (g_current_processor != nullptr || g_submission_pause_depth == UINT32_MAX) {
-		EXIT("cannot acquire GPU submission lock in the current state\n");
-	}
-	if (g_submission_pause_depth++ == 0) {
-		m_gpu.PauseSubmissions();
-	}
-}
-
-Gpu::SubmissionLock::~SubmissionLock() {
-	if (g_submission_pause_depth == 0) {
-		EXIT("GPU submission lock released without ownership\n");
-	}
-	if (--g_submission_pause_depth == 0) {
-		m_gpu.ResumeSubmissions();
-	}
-}
-
-bool Gpu::IsCommandProcessorThread() noexcept {
-	return g_current_processor != nullptr;
-}
-
-CommandProcessor* Gpu::CurrentCommandProcessor() noexcept {
-	return g_current_processor;
-}
-
-bool Gpu::SubmissionLockHeld() noexcept {
-	return g_submission_pause_depth != 0;
-}
-
-bool Gpu::MutexHeld() noexcept {
-	return g_gpu_mutex_owned;
+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);
@@ -33,27 +31,9 @@ public:
 	void Done();
 	[[nodiscard]] int GetFrameNum() const;
 
-	[[nodiscard]] static bool              IsCommandProcessorThread() noexcept;
-	[[nodiscard]] static CommandProcessor* CurrentCommandProcessor() noexcept;
-	[[nodiscard]] static bool              SubmissionLockHeld() noexcept;
-	[[nodiscard]] static bool              MutexHeld() noexcept;
-
-	class SubmissionLock final {
-	public:
-		explicit SubmissionLock(Gpu& gpu);
-		~SubmissionLock();
-		KYTY_CLASS_NO_COPY(SubmissionLock);
-
-	private:
-		Gpu& m_gpu;
-	};
+	[[nodiscard]] static bool IsGpuThread() noexcept;
 
 private:
-	friend class SubmissionLock;
-
-	void PauseSubmissions();
-	void ResumeSubmissions();
-
 	std::unique_ptr<GpuState> m_state;
 };
 } // namespace Libs::Graphics

src/graphics/guest_gpu/hardwareContext.h

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

src/graphics/guest_gpu/pm4.cpp

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

src/graphics/guest_gpu/pm4.h

@@ -87,6 +87,7 @@ constexpr uint32_t R_WAIT_MEM_64    = 0x16;
 constexpr uint32_t R_FLIP           = 0x17;
 constexpr uint32_t R_RELEASE_MEM    = 0x18;
 constexpr uint32_t R_DMA_DATA       = 0x19;
+constexpr uint32_t R_CONTEXT_STATE  = 0x1A;
 
 constexpr uint32_t R_NUM = 0x3F + 1;
 
@@ -577,24 +578,28 @@ 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 +783,21 @@ 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 +890,101 @@ 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;

src/graphics/guest_gpu/tile.h

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

src/graphics/host_gpu/memoryTracker.cpp

@@ -6,13 +6,7 @@ namespace Libs::Graphics {
 
 static_assert(std::atomic<void*>::is_always_lock_free);
 
-MemoryTracker::MemoryTracker(PageManager& page_manager, PageWatchMode gpu_watch_mode)
-    : m_page_manager(page_manager), m_gpu_watch_mode(gpu_watch_mode) {
-	switch (m_gpu_watch_mode) {
-		case PageWatchMode::Write:
-		case PageWatchMode::ReadWrite: break;
-		default: EXIT("unsupported memory tracker GPU page-watch mode\n");
-	}
+MemoryTracker::MemoryTracker(PageManager& page_manager): m_page_manager(page_manager) {
 	m_regions = std::make_unique<std::atomic<RegionManager*>[]>(REGION_COUNT);
 	for (size_t i = 0; i < REGION_COUNT; i++) {
 		m_regions[i].store(nullptr, std::memory_order_relaxed);
@@ -21,6 +15,7 @@ MemoryTracker::MemoryTracker(PageManager& page_manager, PageWatchMode gpu_watch_
 
 MemoryTracker::~MemoryTracker() = default;
 
+#if KYTY_BUILD == KYTY_BUILD_DEBUG
 void MemoryTracker::ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr, uint64_t size,
                                           const char* operation) const noexcept {
 	if (vaddr == 0 || size == 0 || size > UINT64_MAX - vaddr ||
@@ -58,6 +53,7 @@ void MemoryTracker::ValidateGpuDirtyOwnership(const RangeSet& dirty, uint64_t va
 		}
 	}
 }
+#endif
 
 void MemoryTracker::ValidateRange(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
@@ -83,7 +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);
@@ -92,7 +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);
@@ -101,41 +95,29 @@ 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);
-		const auto       changed =
-		    manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
-		manager->ApplyProtection(changed, false);
+		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
 }
 
 void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
-	Iterate<true>(vaddr, size, [this](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+	Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
-		const auto       changed =
-		    manager->ChangeState<DirtySource::Gpu, true>(manager->GetCpuAddr() + offset, bytes);
-		manager->ApplyGpuProtection(changed, true, m_gpu_watch_mode);
+		manager->ChangeState<DirtySource::Gpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
 }
 
 void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
-	Iterate<true>(vaddr, size, [this](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
-		if (!manager->IsFullyModified<DirtySource::Gpu>(offset, bytes)) {
-			EXIT("cannot clear partially GPU-dirty tracking range\n");
-		}
-		const auto changed =
-		    manager->ChangeState<DirtySource::Gpu, false>(manager->GetCpuAddr() + offset, bytes);
-		manager->ApplyGpuProtection(changed, false, m_gpu_watch_mode);
+		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);
@@ -154,17 +136,14 @@ void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
 		EXIT("cannot untrack GPU-dirty memory\n");
 	}
 	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-		const auto changed =
-		    manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
-		manager->ApplyProtection(changed, false);
+		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
 	locks.clear();
 }
 
 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

@@ -18,8 +18,7 @@ namespace Libs::Graphics {
 
 class MemoryTracker final {
 public:
-	explicit MemoryTracker(PageManager&  page_manager,
-	                       PageWatchMode gpu_watch_mode = PageWatchMode::ReadWrite);
+	explicit MemoryTracker(PageManager& page_manager);
 	~MemoryTracker();
 
 	KYTY_CLASS_NO_COPY(MemoryTracker);
@@ -30,58 +29,45 @@ 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) {
-				               const auto changed = manager->ChangeState<DirtySource::Cpu, true>(
-				                   manager->GetCpuAddr() + offset, bytes);
-				               manager->ApplyProtection(changed, false);
-			               });
-			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,
 	                           const char* operation) const noexcept;
 	void ValidateGpuDirtyOwnership(const RangeSet& dirty, uint64_t vaddr, uint64_t size,
 	                               const char* operation);
+#else
+	void ValidateGpuDirtyPages(const RangeSet&, uint64_t, uint64_t, const char*) const noexcept {}
+	void ValidateGpuDirtyOwnership(const RangeSet&, uint64_t, uint64_t, const char*) {}
+#endif
 
 	template <bool clear, typename Preflight, typename Func>
 	void ForEachDownloadRange(uint64_t vaddr, uint64_t size, Preflight&& preflight, Func&& func) {
 		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);
@@ -104,10 +90,8 @@ public:
 			Iterate<false>(vaddr, size,
 			               [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 				               const auto address = manager->GetCpuAddr() + offset;
-				               const auto changed =
-				                   manager->template ForEachModifiedRange<DirtySource::Gpu, true>(
-				                       address, bytes, [](uint64_t, uint64_t) noexcept {});
-				               manager->ApplyGpuProtection(changed, false, m_gpu_watch_mode);
+				               manager->template ForEachModifiedRange<DirtySource::Gpu, true>(
+				                   address, bytes, [](uint64_t, uint64_t) noexcept {});
 			               });
 		}
 	}
@@ -124,7 +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) {
@@ -137,13 +120,12 @@ public:
 		});
 		upload_func();
 		if (is_written) {
-			Iterate<false>(
-			    vaddr, size, [this](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-				    const auto changed = manager->template ChangeState<DirtySource::Gpu, true>(
-				        manager->GetCpuAddr() + offset, bytes);
-				    manager->ApplyGpuProtection(changed, true, m_gpu_watch_mode);
-				    manager->lock.unlock();
-			    });
+			Iterate<false>(vaddr, size,
+			               [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+				               manager->template ChangeState<DirtySource::Gpu, true>(
+				                   manager->GetCpuAddr() + offset, bytes);
+				               manager->lock.unlock();
+			               });
 		}
 		s_upload_owner = previous_upload_owner;
 	}
@@ -189,15 +171,13 @@ 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;
-	PageWatchMode                                  m_gpu_watch_mode = PageWatchMode::ReadWrite;
 };
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/pageManager.cpp

@@ -9,7 +9,6 @@
 #include <cstdarg>
 #include <cstdio>
 #include <cstdlib>
-#include <limits>
 #include <memory>
 #include <mutex>
 #include <vector>
@@ -22,13 +21,9 @@
 #undef min
 #undef max
 #elif defined(__APPLE__)
-#include <pthread.h>
-#include <sys/mman.h>
 #include <unistd.h>
 #else
 #include <execinfo.h>
-#include <sys/mman.h>
-#include <sys/syscall.h>
 #include <unistd.h>
 #endif
 
@@ -41,9 +36,8 @@ constexpr uint64_t ADDRESS_SIZE = TRACKER_ADDRESS_SIZE;
 constexpr uint64_t REGION_COUNT = ADDRESS_SIZE / REGION_SIZE;
 
 #if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
-// The tracker reuses Win32 memory-protection tags as internal page-state values (on
-// Windows they come from <windows.h> and are what VirtualQuery returns). Mirror the
-// canonical Win32 numeric values so the shared state-machine logic is identical.
+// The tracker reuses Win32 memory-protection tags as internal page-state values.
+// Mirror their canonical numeric values so the shared state-machine logic is identical.
 constexpr uint32_t PAGE_NOACCESS  = 0x01;
 constexpr uint32_t PAGE_READONLY  = 0x02;
 constexpr uint32_t PAGE_READWRITE = 0x04;
@@ -53,8 +47,6 @@ constexpr uint64_t REGION_PAGES = REGION_SIZE / PAGE_SIZE;
 constexpr uint32_t NO_ACCESS_PROTECTION  = PAGE_NOACCESS;
 constexpr uint32_t READ_ONLY_PROTECTION  = PAGE_READONLY;
 constexpr uint32_t READ_WRITE_PROTECTION = PAGE_READWRITE;
-// Zero is the unknown protection sentinel.
-constexpr uint32_t UNKNOWN_PROTECTION = 0;
 
 [[noreturn]] void FailFast(const char* reason = nullptr) noexcept {
 	std::fputs("PageManager fail-fast: ", stderr);
@@ -102,25 +94,6 @@ Common::VirtualMemory::Mode ToMemoryMode(uint32_t protection) {
 	}
 }
 
-uint32_t CurrentThread() noexcept {
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	return GetCurrentThreadId();
-#elif defined(__APPLE__)
-	return static_cast<uint32_t>(pthread_mach_thread_np(pthread_self()));
-#elif defined(__linux__)
-	static thread_local const uint32_t tid = [] {
-		const auto raw = static_cast<uint32_t>(::syscall(SYS_gettid));
-		if (raw == 0) {
-			FailFast("gettid returned the reserved zero owner token");
-		}
-		return raw;
-	}();
-	return tid;
-#else
-	FailFast("page tracking thread identity is unsupported on this platform");
-#endif
-}
-
 class SpinGuard final {
 public:
 	explicit SpinGuard(std::atomic_flag& lock): m_lock(lock) {
@@ -154,38 +127,58 @@ uint64_t PageEnd(uint64_t vaddr, uint64_t size) {
 
 struct PageManager::Impl {
 	struct PageState {
-		std::atomic_flag lock                = ATOMIC_FLAG_INIT;
-		uint32_t         write_watchers      = 0;
-		uint32_t         access_watchers     = 0;
-		uint32_t         original_protection = 0;
-		uint32_t         backing_writer      = 0;
-		// Shadow the protection applied through Protect().
-		uint32_t current_protection = UNKNOWN_PROTECTION;
-		bool     resolving          = false;
-	};
+		uint8_t write_watchers  : 7 = 0;
+		uint8_t access_watchers : 1 = 0;
 
-	struct Region {
-		std::array<PageState, REGION_PAGES> pages;
-	};
+		[[nodiscard]] uint32_t Perms() const noexcept {
+			if (access_watchers != 0) {
+				return NO_ACCESS_PROTECTION;
+			}
+			if (write_watchers != 0) {
+				return READ_ONLY_PROTECTION;
+			}
+			return READ_WRITE_PROTECTION;
+		}
 
-	class PageRangeGuard final {
-	public:
-		explicit PageRangeGuard(std::span<PageState*> pages): m_pages(pages) {
-			for (auto* page: m_pages) {
-				while (page->lock.test_and_set(std::memory_order_acquire)) {
-					std::atomic_signal_fence(std::memory_order_seq_cst);
+		template <int delta, bool is_read>
+		uint32_t AddDelta(uint64_t address) {
+			static_assert(delta >= -1 && delta <= 1);
+			if constexpr (is_read) {
+				if constexpr (delta == 1) {
+					if (access_watchers != 0) {
+						Fatal("read-watcher overflow at 0x%016" PRIx64, address);
+					}
+					return ++access_watchers;
+				} else if constexpr (delta == -1) {
+					if (access_watchers == 0) {
+						Fatal("read-watcher underflow at 0x%016" PRIx64, address);
+					}
+					return --access_watchers;
+				} else {
+					return access_watchers;
+				}
+			} else {
+				if constexpr (delta == 1) {
+					if (write_watchers == 0x7f) {
+						Fatal("write-watcher overflow at 0x%016" PRIx64, address);
+					}
+					return ++write_watchers;
+				} else if constexpr (delta == -1) {
+					if (write_watchers == 0) {
+						Fatal("write-watcher underflow at 0x%016" PRIx64, address);
+					}
+					return --write_watchers;
+				} else {
+					return write_watchers;
 				}
 			}
 		}
-		~PageRangeGuard() {
-			for (auto it = m_pages.rbegin(); it != m_pages.rend(); ++it) {
-				(*it)->lock.clear(std::memory_order_release);
-			}
-		}
-		KYTY_CLASS_NO_COPY(PageRangeGuard);
+	};
+	static_assert(sizeof(PageState) == 1);
 
-	private:
-		std::span<PageState*> m_pages;
+	struct Region {
+		std::atomic_flag                    lock = ATOMIC_FLAG_INIT;
+		std::array<PageState, REGION_PAGES> pages;
 	};
 
 	Impl() {
@@ -215,10 +208,9 @@ struct PageManager::Impl {
 
 	~Impl() {
 		for (const auto& region: region_storage) {
+			SpinGuard lock(region->lock);
 			for (auto& page: region->pages) {
-				SpinGuard lock(page.lock);
-				if (page.write_watchers != 0 || page.access_watchers != 0 ||
-				    page.backing_writer != 0 || page.resolving) {
+				if (page.write_watchers != 0 || page.access_watchers != 0) {
 					FailFast("PageManager destroyed with live page state");
 				}
 			}
@@ -246,56 +238,81 @@ struct PageManager::Impl {
 		return ptr;
 	}
 
-	PageState& GetPage(Region& region, uint64_t vaddr) const {
-		return region.pages[(vaddr % REGION_SIZE) / PAGE_SIZE];
-	}
-
-	static uint32_t WatcherProtection(const PageState& page) {
-		if (page.access_watchers != 0) {
-			return NO_ACCESS_PROTECTION;
-		}
-		if (page.write_watchers != 0) {
-			return READ_ONLY_PROTECTION;
-		}
-		return page.original_protection;
-	}
-
-	static void InitializeProtection(std::span<PageState*> pages) {
-		for (auto* page: pages) {
-			page->original_protection = READ_WRITE_PROTECTION;
-			page->current_protection  = READ_WRITE_PROTECTION;
-		}
-	}
-
-	void ProtectRange(std::span<PageState*> pages, uint64_t vaddr, uint32_t protection,
-	                  std::span<const uint32_t> expected_old) noexcept {
-		const auto size = pages.size() * PAGE_SIZE;
-		if (pages.size() != expected_old.size()) {
-			FailFast("protection range state size mismatch");
-		}
-		for (size_t i = 0; i < pages.size(); i++) {
-			const auto actual = pages[i]->current_protection;
-			if (actual != UNKNOWN_PROTECTION && actual != expected_old[i]) {
-				Fatal("invalid protection transition at 0x%016" PRIx64 ", old=0x%08" PRIx32
-				      ", expected=0x%08" PRIx32 ", new=0x%08" PRIx32,
-				      vaddr + i * PAGE_SIZE, actual, expected_old[i], protection);
-			}
-		}
+	void Protect(uint64_t vaddr, uint64_t size, uint32_t protection) noexcept {
 		if (!Libs::LibKernel::Memory::ProtectGuestHostMemory(vaddr, size,
 		                                                     ToMemoryMode(protection))) {
 			Fatal("address-space protection failed at 0x%016" PRIx64 ", new=0x%08" PRIx32, vaddr,
 			      protection);
 		}
-		for (auto* page: pages) {
-			page->current_protection = protection;
-		}
 	}
 
-	void Protect(PageState& page, uint64_t vaddr, uint32_t protection,
-	             uint32_t expected_old) noexcept {
-		PageState* pages[]    = {&page};
-		uint32_t   expected[] = {expected_old};
-		ProtectRange(pages, vaddr, protection, expected);
+	template <bool track, bool is_read, bool masked>
+	void UpdateRegionWatchers(Region& region, uint64_t base_addr, size_t first, size_t last,
+	                          const RegionBits* mask = nullptr) {
+		SpinGuard lock(region.lock);
+		auto      perms                 = region.pages[first].Perms();
+		uint64_t  range_begin           = 0;
+		uint64_t  range_bytes           = 0;
+		uint64_t  potential_range_bytes = 0;
+
+		const auto release_pending = [&] {
+			if (range_bytes != 0) {
+				Protect(base_addr + range_begin * PAGE_SIZE, range_bytes, perms);
+				range_bytes           = 0;
+				potential_range_bytes = 0;
+			}
+		};
+
+		for (size_t page_index = first; page_index < last; page_index++) {
+			auto&      page    = region.pages[page_index];
+			const auto address = base_addr + page_index * PAGE_SIZE;
+			const bool update  = !masked || mask->Get(page_index);
+
+			const auto old_perms = page.Perms();
+			const auto new_count = update ? page.AddDelta<track ? 1 : -1, is_read>(address)
+			                              : page.AddDelta<0, is_read>(address);
+			const auto new_perms = page.Perms();
+
+			if (new_perms != perms) [[unlikely]] {
+				release_pending();
+				perms = new_perms;
+			} else if (range_bytes != 0) {
+				potential_range_bytes += PAGE_SIZE;
+			}
+
+			if (!update) {
+				continue;
+			}
+
+			const bool watcher_edge = (track && new_count == 1) || (!track && new_count == 0);
+			if (watcher_edge && old_perms != new_perms) {
+				if (range_bytes == 0) {
+					range_begin           = page_index;
+					potential_range_bytes = PAGE_SIZE;
+				}
+				range_bytes = potential_range_bytes;
+			}
+		}
+
+		release_pending();
+	}
+
+	template <bool track, bool is_read>
+	void UpdatePageWatchers(uint64_t vaddr, uint64_t size) {
+		const auto begin = PageStart(vaddr);
+		const auto end   = PageEnd(vaddr, size);
+		for (auto chunk_begin = begin; chunk_begin < end;) {
+			const auto chunk_end   = std::min(end, (chunk_begin / REGION_SIZE + 1) * REGION_SIZE);
+			const auto region_base = chunk_begin / REGION_SIZE * REGION_SIZE;
+			auto*      region = track ? GetOrCreateRegion(chunk_begin) : FindRegion(chunk_begin);
+			if (region == nullptr) {
+				Fatal("untracking unknown page 0x%016" PRIx64, chunk_begin);
+			}
+			const auto first = static_cast<size_t>((chunk_begin - region_base) / PAGE_SIZE);
+			const auto last  = static_cast<size_t>((chunk_end - region_base) / PAGE_SIZE);
+			UpdateRegionWatchers<track, is_read, false>(*region, region_base, first, last);
+			chunk_begin = chunk_end;
+		}
 	}
 
 	std::unique_ptr<std::atomic<Region*>[]> regions;
@@ -313,212 +330,48 @@ uint64_t PageManager::GetPageSize() const {
 	return PAGE_SIZE;
 }
 
-void PageManager::UpdatePageWatchers(bool track, uint64_t vaddr, uint64_t size,
-                                     PageWatchMode mode) {
-	if (mode != PageWatchMode::Write && mode != PageWatchMode::ReadWrite) {
-		Fatal("invalid watcher mode");
-	}
-	const auto begin = PageStart(vaddr);
-	const auto end   = PageEnd(vaddr, size);
-	for (auto chunk_begin = begin; chunk_begin < end;) {
-		const auto chunk_end = std::min(end, (chunk_begin / REGION_SIZE + 1) * REGION_SIZE);
-		auto*      region =
-		    track ? m_impl->GetOrCreateRegion(chunk_begin) : m_impl->FindRegion(chunk_begin);
-		if (region == nullptr) {
-			Fatal("untracking unknown page 0x%016" PRIx64, chunk_begin);
-		}
-
-		const auto page_count = static_cast<size_t>((chunk_end - chunk_begin) / PAGE_SIZE);
-		std::vector<Impl::PageState*> pages;
-		pages.reserve(page_count);
-		for (auto address = chunk_begin; address < chunk_end; address += PAGE_SIZE) {
-			pages.push_back(&m_impl->GetPage(*region, address));
-		}
-		Impl::PageRangeGuard lock(pages);
-
-		std::vector<uint8_t> first_watchers(page_count);
-		for (size_t i = 0; i < page_count; i++) {
-			auto&      page    = *pages[i];
-			const auto address = chunk_begin + i * PAGE_SIZE;
-			if (page.resolving && track) {
-				FailFast("new page watcher raced active fault resolution");
-			}
-			auto& watchers =
-			    (mode == PageWatchMode::ReadWrite ? page.access_watchers : page.write_watchers);
-			if (track) {
-				if (watchers == std::numeric_limits<uint32_t>::max()) {
-					Fatal("watcher overflow at 0x%016" PRIx64, address);
-				}
-				first_watchers[i] = page.write_watchers == 0 && page.access_watchers == 0;
-			} else {
-				if (watchers == 0) {
-					Fatal("watcher underflow at 0x%016" PRIx64, address);
-				}
-				if (page.backing_writer != 0 && page.backing_writer != CurrentThread()) {
-					Fatal("backing write ownership changed at 0x%016" PRIx64, address);
-				}
-			}
-		}
-
-		if (track) {
-			for (size_t first = 0; first < page_count;) {
-				while (first < page_count && first_watchers[first] == 0) {
-					first++;
-				}
-				auto last = first;
-				while (last < page_count && first_watchers[last] != 0) {
-					last++;
-				}
-				if (first != last) {
-					Impl::InitializeProtection(std::span {pages}.subspan(first, last - first));
-				}
-				first = last;
-			}
-		}
-
-		std::vector<uint32_t> old_protections(page_count);
-		std::vector<uint32_t> new_protections(page_count);
-		std::vector<uint8_t>  transitions(page_count);
-		for (size_t i = 0; i < page_count; i++) {
-			auto& page = *pages[i];
-			auto& watchers =
-			    (mode == PageWatchMode::ReadWrite ? page.access_watchers : page.write_watchers);
-			const auto old_protection = Impl::WatcherProtection(page);
-			if (track) {
-				watchers++;
-			} else {
-				watchers--;
-			}
-			const auto new_protection = Impl::WatcherProtection(page);
-			old_protections[i]        = old_protection;
-			new_protections[i]        = new_protection;
-			if (new_protection != old_protection && (track || page.backing_writer == 0)) {
-				transitions[i] = 1;
-			}
-		}
-
-		for (size_t first = 0; first < page_count;) {
-			while (first < page_count && transitions[first] == 0) {
-				first++;
-			}
-			if (first == page_count) {
-				break;
-			}
-			const auto protection = new_protections[first];
-			auto       current    = first + 1;
-			auto       last       = current;
-			for (; current < page_count && new_protections[current] == protection; current++) {
-				if (old_protections[current] != new_protections[current] &&
-				    transitions[current] == 0) {
-					break;
-				}
-				if (transitions[current] != 0) {
-					last = current + 1;
-				}
-			}
-			m_impl->ProtectRange(std::span {pages}.subspan(first, last - first),
-			                     chunk_begin + first * PAGE_SIZE, protection,
-			                     std::span {old_protections}.subspan(first, last - first));
-			first = current;
-		}
-
-		for (auto* page: pages) {
-			if (!track && page->backing_writer == 0 && page->write_watchers == 0 &&
-			    page->access_watchers == 0) {
-				page->original_protection = 0;
-			}
-		}
-		chunk_begin = chunk_end;
-	}
+template <bool track>
+void PageManager::UpdatePageWatchers(uint64_t vaddr, uint64_t size) {
+	m_impl->UpdatePageWatchers<track, false>(vaddr, size);
 }
 
+template void PageManager::UpdatePageWatchers<true>(uint64_t, uint64_t);
+template void PageManager::UpdatePageWatchers<false>(uint64_t, uint64_t);
+
+template <bool track, bool is_read>
+void PageManager::UpdatePageWatchersForRegion(uint64_t base_addr, RegionBits& mask) {
+	if (base_addr % REGION_SIZE != 0 || base_addr >= ADDRESS_SIZE ||
+	    REGION_SIZE > ADDRESS_SIZE - base_addr) {
+		Fatal("invalid tracking region base 0x%016" PRIx64, base_addr);
+	}
+
+	const auto start_range = mask.FirstRange();
+	const auto end_range   = mask.LastRange();
+	if (start_range.first == REGION_PAGES) {
+		FailFast("empty region watcher mask");
+	}
+	const auto first = start_range.first;
+	const auto last  = end_range.second;
+	if (start_range.second == end_range.second) {
+		m_impl->UpdatePageWatchers<track, is_read>(base_addr + first * PAGE_SIZE,
+		                                           (last - first) * PAGE_SIZE);
+		return;
+	}
+
+	auto* region = track ? m_impl->GetOrCreateRegion(base_addr) : m_impl->FindRegion(base_addr);
+	if (region == nullptr) {
+		Fatal("untracking unknown region 0x%016" PRIx64, base_addr);
+	}
+	m_impl->UpdateRegionWatchers<track, is_read, true>(*region, base_addr, first, last, &mask);
+}
+
+template void PageManager::UpdatePageWatchersForRegion<true, true>(uint64_t, RegionBits&);
+template void PageManager::UpdatePageWatchersForRegion<true, false>(uint64_t, RegionBits&);
+template void PageManager::UpdatePageWatchersForRegion<false, true>(uint64_t, RegionBits&);
+template void PageManager::UpdatePageWatchersForRegion<false, false>(uint64_t, RegionBits&);
+
 void PageManager::OnGpuMap(uint64_t, uint64_t) {}
 
 void PageManager::OnGpuUnmap(uint64_t, uint64_t) {}
 
-PageManager::BackingWrite::BackingWrite(PageManager& manager, uint64_t vaddr,
-                                        uint64_t size) noexcept
-    : m_manager(manager), m_vaddr(vaddr), m_size(size) {
-	m_manager.BeginBackingWrite(vaddr, size);
-}
-
-PageManager::BackingWrite::~BackingWrite() {
-	m_manager.EndBackingWrite(m_vaddr, m_size);
-}
-
-std::vector<std::unique_ptr<PageManager::BackingWrite>>
-PageManager::ReserveBackingWrites(std::span<const RangeSet::Range> ranges) {
-	if (ranges.empty()) {
-		Fatal("cannot reserve empty backing-write ranges");
-	}
-	std::vector<std::unique_ptr<BackingWrite>> writes;
-	writes.reserve(ranges.size());
-	uint64_t begin = 0;
-	uint64_t end   = 0;
-	for (const auto& range: ranges) {
-		if (range.address == 0 || range.size == 0 || range.size > UINT64_MAX - range.address ||
-		    range.address + range.size > UINT64_MAX - (PAGE_SIZE - 1)) {
-			Fatal("invalid backing-write range");
-		}
-		const auto page_begin = PageStart(range.address);
-		const auto page_end   = PageStart(range.address + range.size + PAGE_SIZE - 1);
-		if (begin != 0 && page_begin > end) {
-			writes.push_back(std::make_unique<BackingWrite>(*this, begin, end - begin));
-			begin = 0;
-		}
-		if (begin == 0) {
-			begin = page_begin;
-			end   = page_end;
-		} else {
-			end = std::max(end, page_end);
-		}
-	}
-	writes.push_back(std::make_unique<BackingWrite>(*this, begin, end - begin));
-	return writes;
-}
-
-void PageManager::BeginBackingWrite(uint64_t vaddr, uint64_t size) noexcept {
-	const auto end    = PageEnd(vaddr, size);
-	const auto writer = CurrentThread();
-	for (auto address = PageStart(vaddr); address < end; address += PAGE_SIZE) {
-		auto* region = m_impl->FindRegion(address);
-		if (region == nullptr) {
-			Fatal("backing write reserves an unknown page at 0x%016" PRIx64, address);
-		}
-		auto&     page = m_impl->GetPage(*region, address);
-		SpinGuard lock(page.lock);
-		if (page.resolving || page.backing_writer != 0 || page.access_watchers == 0) {
-			Fatal("backing write races page resolution at 0x%016" PRIx64, address);
-		}
-		page.resolving      = true;
-		page.backing_writer = writer;
-	}
-}
-
-void PageManager::EndBackingWrite(uint64_t vaddr, uint64_t size) noexcept {
-	const auto end    = PageEnd(vaddr, size);
-	const auto writer = CurrentThread();
-	for (auto address = PageStart(vaddr); address < end; address += PAGE_SIZE) {
-		auto* region = m_impl->FindRegion(address);
-		if (region == nullptr) {
-			FailFast("backing write ended for an unknown page");
-		}
-		auto&     page = m_impl->GetPage(*region, address);
-		SpinGuard lock(page.lock);
-		if (!page.resolving || page.backing_writer != writer) {
-			FailFast("backing write ended without matching owner and resolving state");
-		}
-		const auto old_protection = NO_ACCESS_PROTECTION;
-		const auto new_protection = Impl::WatcherProtection(page);
-		if (new_protection != old_protection) {
-			m_impl->Protect(page, address, new_protection, old_protection);
-		}
-		if (page.write_watchers == 0 && page.access_watchers == 0) {
-			page.original_protection = 0;
-		}
-		page.backing_writer = 0;
-		page.resolving      = false;
-	}
-}
-
 } // namespace Libs::Graphics

src/graphics/host_gpu/pageManager.h

@@ -2,31 +2,16 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_PAGEMANAGER_H_
 
 #include "common/common.h"
-#include "graphics/host_gpu/rangeSet.h"
+#include "graphics/host_gpu/regionDefinitions.h"
 
 #include <memory>
-#include <span>
-#include <vector>
 
 namespace Libs::Graphics {
 
 enum class PageFaultAccess { Read, Write, Execute, Unknown };
-enum class PageWatchMode { Write, ReadWrite };
 
 class PageManager final {
 public:
-	class BackingWrite final {
-	public:
-		BackingWrite(PageManager& manager, uint64_t vaddr, uint64_t size) noexcept;
-		~BackingWrite();
-		KYTY_CLASS_NO_COPY(BackingWrite);
-
-	private:
-		PageManager& m_manager;
-		uint64_t     m_vaddr = 0;
-		uint64_t     m_size  = 0;
-	};
-
 	PageManager();
 	// The owner must stop all PageManager callers before destruction.
 	~PageManager();
@@ -35,18 +20,14 @@ public:
 
 	[[nodiscard]] uint64_t GetPageSize() const;
 
-	void UpdatePageWatchers(bool track, uint64_t vaddr, uint64_t size,
-	                        PageWatchMode mode = PageWatchMode::Write);
+	template <bool track>
+	void UpdatePageWatchers(uint64_t vaddr, uint64_t size);
+	template <bool track, bool is_read = false>
+	void UpdatePageWatchersForRegion(uint64_t base_addr, RegionBits& mask);
 	void OnGpuMap(uint64_t vaddr, uint64_t size);
 	void OnGpuUnmap(uint64_t vaddr, uint64_t size);
 
-	[[nodiscard]] std::vector<std::unique_ptr<BackingWrite>>
-	ReserveBackingWrites(std::span<const RangeSet::Range> ranges);
-
 private:
-	void BeginBackingWrite(uint64_t vaddr, uint64_t size) noexcept;
-	void EndBackingWrite(uint64_t vaddr, uint64_t size) noexcept;
-
 	struct Impl;
 	std::unique_ptr<Impl> m_impl;
 };

src/graphics/host_gpu/regionDefinitions.h

@@ -1,10 +1,9 @@
 #ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_REGIONDEFINITIONS_H_
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_REGIONDEFINITIONS_H_
 
+#include "common/bitArray.h"
 #include "common/common.h"
 
-#include <bitset>
-
 namespace Libs::Graphics {
 
 constexpr uint64_t TRACKER_PAGE_SIZE    = 4ull * 1024ull;
@@ -13,7 +12,8 @@ constexpr uint64_t TRACKER_ADDRESS_SIZE = 1ull << 40u;
 constexpr size_t   TRACKER_REGION_PAGES = TRACKER_REGION_SIZE / TRACKER_PAGE_SIZE;
 
 enum class DirtySource { Cpu, Gpu };
-using RegionBits = std::bitset<TRACKER_REGION_PAGES>;
+using RegionBits = Common::BitArray<TRACKER_REGION_PAGES>;
+static_assert(sizeof(RegionBits) == TRACKER_REGION_PAGES / 8);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/regionManager.h

@@ -76,8 +76,9 @@ public:
 		if (m_cpu_addr % TRACKER_REGION_SIZE != 0) {
 			EXIT("invalid region tracking manager construction\n");
 		}
-		m_cpu_dirty.set();
-		m_writable.set();
+		m_cpu_dirty.Fill();
+		m_writable.Fill();
+		m_readable.Fill();
 	}
 
 	KYTY_CLASS_NO_COPY(RegionManager);
@@ -87,106 +88,81 @@ public:
 	[[nodiscard]] bool IsModified(uint64_t offset, uint64_t size) const {
 		const auto [start, end] = GetPageRange(m_cpu_addr + offset, size);
 		const auto& bits        = GetBits<source>();
-		for (auto page = start; page < end; page++) {
-			if (bits.test(page)) {
-				return true;
-			}
-		}
-		return false;
-	}
-
-	template <DirtySource source>
-	[[nodiscard]] bool IsFullyModified(uint64_t offset, uint64_t size) const {
-		const auto [start, end] = GetPageRange(m_cpu_addr + offset, size);
-		const auto& bits        = GetBits<source>();
-		for (auto page = start; page < end; page++) {
-			if (!bits.test(page)) {
-				return false;
-			}
-		}
-		return true;
+		return RegionBits(bits, start, end).Any();
 	}
 
 	template <DirtySource source, bool enable>
-	RegionBits ChangeState(uint64_t vaddr, uint64_t size) {
+	void ChangeState(uint64_t vaddr, uint64_t size) {
 		const auto [start, end] = GetPageRange(vaddr, size);
 		if constexpr (source == DirtySource::Cpu && enable) {
-			for (auto page = start; page < end; page++) {
-				if (m_gpu_dirty.test(page)) {
-					EXIT("CPU dirty state conflicts with GPU dirty state\n");
-				}
+			if (RegionBits(m_gpu_dirty, start, end).Any()) {
+				EXIT("CPU dirty state conflicts with GPU dirty state\n");
 			}
 		}
 		if constexpr (source == DirtySource::Gpu && enable) {
-			for (auto page = start; page < end; page++) {
-				if (m_cpu_dirty.test(page)) {
-					EXIT("GPU dirty state conflicts with CPU dirty state\n");
-				}
+			if (RegionBits(m_cpu_dirty, start, end).Any()) {
+				EXIT("GPU dirty state conflicts with CPU dirty state\n");
 			}
 		}
-		auto& bits    = GetBits<source>();
-		auto  changed = bits;
-		for (auto page = start; page < end; page++) {
-			bits.set(page, enable);
+		auto& bits = GetBits<source>();
+		if constexpr (enable) {
+			bits.SetRange(start, end);
+		} else {
+			bits.UnsetRange(start, end);
 		}
-		changed ^= bits;
 		if constexpr (source == DirtySource::Cpu) {
-			changed    = m_cpu_dirty ^ m_writable;
-			m_writable = m_cpu_dirty;
+			UpdateCpuProtection<!enable>();
+		} else {
+			UpdateGpuProtection<enable>();
 		}
-		return changed;
 	}
 
 	template <DirtySource source, bool clear, typename Func>
-	RegionBits ForEachModifiedRange(uint64_t vaddr, uint64_t size, Func&& func) {
+	void ForEachModifiedRange(uint64_t vaddr, uint64_t size, Func&& func) {
 		const auto [start, end] = GetPageRange(vaddr, size);
-		auto mask               = GetBits<source>();
-		for (auto page = 0u; page < start; page++) {
-			mask.reset(page);
-		}
-		for (auto page = end; page < TRACKER_REGION_PAGES; page++) {
-			mask.reset(page);
-		}
+		RegionBits mask(GetBits<source>(), start, end);
 		if constexpr (clear) {
-			auto& bits = GetBits<source>();
-			for (auto page = start; page < end; page++) {
-				if (mask.test(page)) {
-					bits.reset(page);
-				}
-			}
+			GetBits<source>().UnsetRange(start, end);
 		}
 		if constexpr (source == DirtySource::Cpu && clear) {
-			auto changed = m_cpu_dirty ^ m_writable;
-			m_writable   = m_cpu_dirty;
-			ApplyProtection(changed, true);
+			UpdateCpuProtection<true>();
 			ForEachRange(mask, std::forward<Func>(func));
-			return changed;
+			return;
+		}
+		if constexpr (source == DirtySource::Gpu && clear) {
+			UpdateGpuProtection<false>();
 		}
 		ForEachRange(mask, std::forward<Func>(func));
-		if constexpr (clear) {
-			return mask;
-		}
-		return {};
-	}
-
-	void ApplyProtection(const RegionBits& changed, bool track) {
-		ForEachRange(changed, [this, track](uint64_t vaddr, uint64_t size) {
-			m_page_manager.UpdatePageWatchers(track, vaddr, size);
-		});
-	}
-
-	void ApplyGpuProtection(const RegionBits& changed, bool track, PageWatchMode mode) {
-		if (mode != PageWatchMode::Write && mode != PageWatchMode::ReadWrite) {
-			EXIT("unsupported GPU page-watch mode\n");
-		}
-		ForEachRange(changed, [this, track, mode](uint64_t vaddr, uint64_t size) {
-			m_page_manager.UpdatePageWatchers(track, vaddr, size, mode);
-		});
 	}
 
 	TrackingSpinLock lock;
 
 private:
+	template <bool track>
+	void UpdateCpuProtection() {
+		auto mask  = m_cpu_dirty ^ m_writable;
+		m_writable = m_cpu_dirty;
+		if (mask.None()) {
+			return;
+		}
+		m_page_manager.UpdatePageWatchersForRegion<track>(m_cpu_addr, mask);
+	}
+
+	template <bool track>
+	void UpdateGpuProtection() {
+		auto readable = ~m_gpu_dirty;
+		auto mask     = readable ^ m_readable;
+		m_readable    = readable;
+		if (mask.None()) {
+			return;
+		}
+		if constexpr (track) {
+			m_page_manager.UpdatePageWatchersForRegion<true, true>(m_cpu_addr, mask);
+		} else {
+			m_page_manager.UpdatePageWatchersForRegion<false, true>(m_cpu_addr, mask);
+		}
+	}
+
 	template <DirtySource source>
 	RegionBits& GetBits() {
 		if constexpr (source == DirtySource::Cpu) {
@@ -217,18 +193,8 @@ private:
 
 	template <typename Func>
 	void ForEachRange(const RegionBits& bits, Func&& func) const {
-		size_t page = 0;
-		while (page < TRACKER_REGION_PAGES) {
-			while (page < TRACKER_REGION_PAGES && !bits.test(page)) {
-				page++;
-			}
-			const auto start = page;
-			while (page < TRACKER_REGION_PAGES && bits.test(page)) {
-				page++;
-			}
-			if (start != page) {
-				func(m_cpu_addr + start * TRACKER_PAGE_SIZE, (page - start) * TRACKER_PAGE_SIZE);
-			}
+		for (const auto [start, end]: bits) {
+			func(m_cpu_addr + start * TRACKER_PAGE_SIZE, (end - start) * TRACKER_PAGE_SIZE);
 		}
 	}
 
@@ -237,6 +203,7 @@ private:
 	RegionBits   m_cpu_dirty;
 	RegionBits   m_gpu_dirty;
 	RegionBits   m_writable;
+	RegionBits   m_readable;
 };
 
 } // namespace Libs::Graphics

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);
@@ -123,12 +95,6 @@ struct BufferCache::RetiredBuffer {
 	std::shared_ptr<Buffer> owner;
 };
 
-struct BufferCache::PendingBackingPublication {
-	uint64_t address = 0;
-	uint64_t size    = 0;
-	uint64_t tick    = 0;
-};
-
 std::pair<uint64_t, uint64_t> BufferCache::DownloadEnvelope(const DownloadCopy& copy) {
 	if (copy.owner == nullptr || copy.size == 0 || copy.source_offset > copy.owner->Size() ||
 	    copy.size > copy.owner->Size() - copy.source_offset) {
@@ -200,39 +166,27 @@ void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, Ret
 	if (copies.empty()) {
 		return;
 	}
-	auto       downloads = RecordDownloads(copies);
-	const auto tick      = m_scheduler.CurrentTick();
-	BeginBackingPublication(retire.address, retire.size, tick);
-	m_scheduler.DeferOperation([this, downloads = std::move(downloads), retire = std::move(retire),
-	                            tick]() mutable {
-		PublishDownloads(downloads);
-		{
-			FaultSafeCacheLock lock(this, m_mutex);
-			if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size)) {
-				m_memory_tracker.ForEachDownloadRange<true>(
-				    retire.address, retire.size,
-				    [&](uint64_t address, uint64_t size) noexcept {
-					    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, size,
-					                                           "asynchronous garbage retirement");
-				    },
-				    [](uint64_t, uint64_t) noexcept {});
-			}
-			for (const auto& range: downloads) {
-				m_gpu_modified_ranges.Subtract(range.address, range.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);
-		}
-		CompleteBackingPublication(retire.address, retire.size, tick);
-	});
+	auto downloads = RecordDownloads(copies);
+	m_scheduler.DeferOperation(
+	    [this, downloads = std::move(downloads), retire = std::move(retire)]() mutable {
+		    PublishDownloads(downloads);
+		    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),
@@ -240,8 +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_page_manager(page_manager), 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()) {
@@ -262,9 +215,6 @@ BufferCache::~BufferCache() {
 	if (!m_gpu_modified_ranges.Empty()) {
 		EXIT("BufferCache: destroyed with pending GPU-modified ranges\n");
 	}
-	if (!m_pending_backing_publications.empty()) {
-		EXIT("BufferCache: destroyed with pending backing publications\n");
-	}
 	for (const auto& [vaddr, cached]: m_buffers) {
 		(void)vaddr;
 		if (m_memory_tracker.IsRegionGpuModified(cached->vaddr, cached->size)) {
@@ -274,68 +224,6 @@ BufferCache::~BufferCache() {
 	m_buffers.clear();
 }
 
-bool BufferCache::SynchronizeBacking(uint64_t vaddr, uint64_t size) {
-	bool waited = false;
-	for (;;) {
-		uint64_t   tick       = 0;
-		const auto page_begin = vaddr & ~(TRACKER_PAGE_SIZE - 1);
-		const auto page_end   = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-		CacheRange affected {.address = page_begin, .size = page_end - page_begin};
-		{
-			FaultSafeCacheLock lock(this, m_mutex);
-			bool               changed = true;
-			while (changed) {
-				changed = false;
-				for (const auto& [address, cached]: m_buffers) {
-					const CacheRange previous = affected;
-					if (ResolveOverlap(affected, {address, cached->size}) &&
-					    (previous.address != affected.address || previous.size != affected.size)) {
-						changed = true;
-					}
-				}
-			}
-		}
-		{
-			std::lock_guard lock(m_publication_mutex);
-			for (const auto& publication: m_pending_backing_publications) {
-				if (publication.address < affected.address + affected.size &&
-				    affected.address < publication.address + publication.size) {
-					tick = std::max(tick, publication.tick);
-				}
-			}
-		}
-		if (tick == 0) {
-			return waited;
-		}
-		waited = true;
-		m_scheduler.Wait(tick);
-		m_scheduler.WaitPriorityOperations(tick);
-	}
-}
-
-void BufferCache::RefreshInvalidatedRanges(CommandBuffer& command, CachedBuffer& cached,
-                                           uint64_t vaddr, uint64_t size, bool upload) {
-	const auto invalidated = m_image_invalidated_ranges.Intersections(vaddr, size);
-	if (upload) {
-		std::array<uint8_t, 64 * 1024> bytes;
-		for (const auto& range: invalidated) {
-			for (uint64_t copied = 0; copied < range.size;) {
-				const auto chunk = std::min<uint64_t>(range.size - copied, bytes.size());
-				if (!Libs::LibKernel::Memory::TryReadBacking(range.address + copied, bytes.data(),
-				                                             chunk)) {
-					EXIT("BufferCache: failed to refresh an invalidated image alias\n");
-				}
-				Upload(command, *cached.buffer, cached.buffer->Offset(range.address + copied),
-				       bytes.data(), chunk);
-				copied += chunk;
-			}
-		}
-	}
-	if (!invalidated.empty()) {
-		m_image_invalidated_ranges.Subtract(vaddr, size);
-	}
-}
-
 StreamBuffer& BufferCache::GetUtilityBuffer(MemoryUsage usage) noexcept {
 	switch (usage) {
 		case MemoryUsage::Upload: return m_staging_buffer;
@@ -364,65 +252,77 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
 		EXIT("BufferCache: invalid memory-invalidation range\n");
 	}
-	(void)SynchronizeBacking(vaddr, size);
-	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) {
-	(void)SynchronizeBacking(vaddr, 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);
-		m_memory_tracker.ForEachDownloadRange<true>(
-		    vaddr, size,
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-			                                           "memory invalidation completion");
-		    },
-		    [](uint64_t, uint64_t) noexcept {});
-		for (const auto& range: downloads) {
-			m_gpu_modified_ranges.Subtract(range.address, range.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);
 	}
 }
 
@@ -430,54 +330,43 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || size > UINT64_MAX - vaddr) {
 		EXIT("BufferCache: invalid unmap range\n");
 	}
-	(void)SynchronizeBacking(vaddr, size);
-
-	std::vector<DownloadCopy>                               copies;
-	std::vector<RangeSet::Range>                            dirty_ranges;
-	std::vector<std::pair<uint64_t, uint64_t>>              modified_buffers;
-	std::vector<std::unique_ptr<PageManager::BackingWrite>> backing_writes;
-	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);
-			}
+	std::vector<DownloadCopy>                  copies;
+	std::vector<std::pair<uint64_t, uint64_t>> modified_buffers;
+	std::vector<std::pair<uint64_t, uint64_t>> retired_buffers;
+	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");
-			}
-			dirty_ranges.insert(dirty_ranges.end(), dirty.begin(), dirty.end());
-			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;
 		}
-		if (!dirty_ranges.empty()) {
-			backing_writes = m_page_manager.ReserveBackingWrites(dirty_ranges);
+		const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
+		if (dirty.empty()) {
+			EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
 		}
-		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});
-				    }
-			    });
+		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);
@@ -488,43 +377,26 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		// command stream before removing such backing.
 		m_scheduler.FinishCurrent();
 	}
-	backing_writes.clear();
-
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& [begin, bytes]: modified_buffers) {
-			if (!m_memory_tracker.IsRegionGpuModified(begin, bytes)) {
-				continue;
-			}
-			m_memory_tracker.ForEachDownloadRange<true>(
-			    begin, bytes,
-			    [&](uint64_t address, uint64_t download_size) noexcept {
-				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address,
-				                                           download_size, "unmap retirement");
-			    },
-			    [](uint64_t, uint64_t) noexcept {});
-		}
-		for (const auto& [begin, bytes]: modified_buffers) {
-			m_gpu_modified_ranges.Subtract(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_image_invalidated_ranges.Subtract(vaddr, size);
-		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;
 		}
 	}
 }
@@ -606,25 +478,32 @@ 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);
-	(void)SynchronizeBacking(vaddr, size);
 
 	if (is_read && !is_written && size <= CACHING_PAGE_SIZE &&
 	    !m_memory_tracker.IsRegionGpuModified(vaddr, size) &&
 	    m_memory_tracker.IsRegionCpuModified(vaddr, size)) {
-		std::vector<uint8_t> data(size);
-		if (Libs::LibKernel::Memory::TryReadBacking(vaddr, data.data(), size)) {
-			return UploadTransient(data.data(), size, 16);
+		const auto alignment = std::max<uint64_t>(
+		    m_graphics.physical_device_properties.limits.minUniformBufferOffsetAlignment, 1);
+		if (auto [mapped, offset] = m_stream_buffer.Map(size, alignment, false);
+		    mapped != nullptr) {
+			if (Libs::LibKernel::Memory::TryReadBacking(vaddr, mapped, size)) {
+				m_stream_buffer.Commit();
+				return {{}, m_stream_buffer.Handle(), offset};
+			}
+		} else {
+			auto owner = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::Upload, 0,
+			                                      AllFlags, size);
+			if (Libs::LibKernel::Memory::TryReadBacking(vaddr, owner->Mapped().data(), size)) {
+				owner->Flush(0, size);
+				return {owner, owner->Handle(), 0};
+			}
 		}
 	}
 
-	if (is_formatted && is_read && !is_written) {
-		(void)m_texture_cache.SynchronizeImageToBuffer(vaddr, size);
-	} else if (is_formatted && is_written) {
+	if (is_formatted && is_written) {
 		(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(
@@ -636,27 +515,26 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 			           reinterpret_cast<const void*>(address), bytes);
 		    }
 	    });
-	RefreshInvalidatedRanges(command, cached, vaddr, size, is_read);
 	if (is_written) {
 		m_gpu_modified_ranges.Add(vaddr, size);
 	}
-	return {cached.buffer, cached.buffer->Handle(), cached.buffer->Offset(vaddr)};
+	auto     owner  = cached.buffer;
+	uint64_t offset = cached.buffer->Offset(vaddr);
+	if (is_formatted && is_read && !is_written) {
+		(void)SynchronizeBufferFromImage(*owner, vaddr, size);
+	}
+	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) {
@@ -664,7 +542,6 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
 		EXIT("BufferCache: invalid image source\n");
 	}
-	(void)SynchronizeBacking(vaddr, size);
 	auto find_owner = [&]() {
 		auto owner = m_buffers.upper_bound(vaddr);
 		if (owner == m_buffers.begin()) {
@@ -675,17 +552,15 @@ 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 invalidated = !m_image_invalidated_ranges.Intersections(vaddr, size).empty();
-		const bool requested_gpu_owned = !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");
 
 		auto owner = find_owner();
-		if (requested_gpu_owned && owner == m_buffers.end()) {
+		if (has_dirty_buffer_source && owner == m_buffers.end()) {
 			CacheRange merged {.address = AlignDown(vaddr),
 			                   .size    = AlignUp(vaddr + size) - AlignDown(vaddr)};
 			using Iterator = decltype(m_buffers.begin());
@@ -735,42 +610,31 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 				EXIT("BufferCache: merged image source does not contain the requested range\n");
 			}
 		}
-		if (owner != m_buffers.end() && !cpu_modified && !invalidated &&
-		    (!gpu_modified || requested_gpu_owned)) {
-			DiscardGpuDirtyBytesLocked(vaddr, size, "image source transfer");
+		if (owner != m_buffers.end() && !cpu_modified &&
+		    (!gpu_modified || has_dirty_buffer_source)) {
 			owner->second->tick_accessed_last = m_gc_tick;
-			return {owner->second->buffer.get(), owner->second->buffer->Offset(vaddr),
-			        requested_gpu_owned};
+			return {owner->second->buffer.get(), owner->second->buffer->Offset(vaddr)};
 		}
-		if (requested_gpu_owned && owner == m_buffers.end()) {
+		if (has_dirty_buffer_source && owner == m_buffers.end()) {
 			EXIT("BufferCache: GPU-dirty image source could not resolve its native owner\n");
 		}
 	}
 
-	// Direct-memory backing remains readable while PageManager protects the guest mapping. The
-	// fallback exists for plain host mappings used by standalone renderer tests and is deliberately
-	// performed outside the cache lock so a page fault cannot recurse into BufferCache.
-	const auto stage_address = vaddr & ~(TRACKER_PAGE_SIZE - 1);
-	const auto stage_end     = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	const auto stage_size    = stage_end - stage_address;
-	(void)SynchronizeBacking(stage_address, stage_size);
-	std::vector<uint8_t> bytes(stage_size);
-	if (!Libs::LibKernel::Memory::TryReadBacking(stage_address, bytes.data(), stage_size)) {
+	auto [staging, stage_offset] = m_staging_buffer.Map(size, 16);
+	if (staging == nullptr || !Libs::LibKernel::Memory::TryReadBacking(vaddr, staging, size)) {
 		EXIT("BufferCache: failed to read mapped guest image backing\n");
 	}
+	m_staging_buffer.Commit();
 
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         dirty       = m_gpu_modified_ranges.Intersections(vaddr, size);
-	const bool         invalidated = !m_image_invalidated_ranges.Intersections(vaddr, size).empty();
-	const bool         requested_gpu_owned = !dirty.empty();
-	auto               owner               = find_owner();
-	if (requested_gpu_owned && owner == m_buffers.end()) {
+	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");
 	}
-	const auto stage_offset = m_staging_buffer.Copy(bytes.data(), stage_size, 16);
-	if (owner == m_buffers.end() || invalidated ||
-	    (m_memory_tracker.IsRegionGpuModified(vaddr, size) && !requested_gpu_owned)) {
-		return {&m_staging_buffer, stage_offset + vaddr - stage_address, false};
+	if (owner == m_buffers.end() ||
+	    (m_memory_tracker.IsRegionGpuModified(vaddr, size) && !has_dirty_buffer_source)) {
+		return {&m_staging_buffer, stage_offset};
 	}
 
 	auto& cached              = *owner->second;
@@ -784,46 +648,20 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	    [&]() noexcept {
 		    for (const auto& [address, upload_size]: uploads) {
 			    cached.buffer->CopyFrom(
-			        m_scheduler.Current(), m_staging_buffer, stage_offset + address - stage_address,
+			        m_scheduler.Current(), m_staging_buffer, stage_offset + address - vaddr,
 			        cached.buffer->Offset(address), upload_size, vk::AccessFlagBits::eHostWrite);
 		    }
 	    });
-	DiscardGpuDirtyBytesLocked(vaddr, size, "staged image source transfer");
-	return {cached.buffer.get(), cached.buffer->Offset(vaddr), requested_gpu_owned};
-}
-
-void BufferCache::DiscardGpuDirtyBytesLocked(uint64_t vaddr, uint64_t size, const char* operation) {
-	m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size, operation);
-	m_gpu_modified_ranges.Subtract(vaddr, size);
-	const auto page_begin = vaddr & ~(TRACKER_PAGE_SIZE - 1);
-	const auto page_end   = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	for (auto page = page_begin; page < page_end; page += TRACKER_PAGE_SIZE) {
-		if (m_gpu_modified_ranges.Intersections(page, TRACKER_PAGE_SIZE).empty() &&
-		    m_memory_tracker.IsRegionGpuModified(page, TRACKER_PAGE_SIZE)) {
-			m_memory_tracker.UnmarkRegionAsGpuModified(page, TRACKER_PAGE_SIZE);
-		}
-	}
-	m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size, operation);
-}
-
-void BufferCache::DiscardGpuDirtyBytes(uint64_t vaddr, uint64_t size) {
-	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
-	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid dirty-byte discard range\n");
-	}
-	FaultSafeCacheLock lock(this, m_mutex);
-	DiscardGpuDirtyBytesLocked(vaddr, size, "image output supersession");
+	return {cached.buffer.get(), cached.buffer->Offset(vaddr)};
 }
 
 void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data) {
 	if (vaddr == 0 || data.empty() || data.size() > UINT64_MAX - vaddr) {
 		EXIT("BufferCache: invalid host DMA write\n");
 	}
-	(void)SynchronizeBacking(vaddr, data.size());
 	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);
@@ -835,45 +673,6 @@ void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data)
 		       data.data() + begin - vaddr, range_end - begin);
 		cached->tick_accessed_last = m_gc_tick;
 	}
-	m_image_invalidated_ranges.Subtract(vaddr, data.size());
-}
-
-std::pair<std::shared_ptr<Buffer>, uint64_t> BufferCache::ObtainBufferForImageWrite(uint64_t vaddr,
-                                                                                    uint64_t size) {
-	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
-	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid image destination\n");
-	}
-	const auto stage_address = vaddr & ~(TRACKER_PAGE_SIZE - 1);
-	const auto stage_end     = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	const auto stage_size    = stage_end - stage_address;
-	(void)SynchronizeBacking(stage_address, stage_size);
-	std::vector<uint8_t> bytes(stage_size);
-	if (!Libs::LibKernel::Memory::TryReadBacking(stage_address, bytes.data(), stage_size)) {
-		EXIT("BufferCache: failed to preserve guest bytes around an image mirror\n");
-	}
-	FaultSafeCacheLock lock(this, m_mutex);
-	auto&              cached = GetOrCreateBuffer(m_scheduler.Current(), vaddr, size);
-	m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
-	                                           "image destination");
-	if (!m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
-		EXIT("BufferCache: image destination aliases GPU-owned buffer bytes\n");
-	}
-	const auto stage_offset = m_staging_buffer.Copy(bytes.data(), stage_size, 16);
-	std::vector<std::pair<uint64_t, uint64_t>> uploads;
-	m_memory_tracker.ForEachUploadRange(
-	    vaddr, size, false,
-	    [&](uint64_t address, uint64_t upload_size) noexcept {
-		    uploads.emplace_back(address, upload_size);
-	    },
-	    [&]() noexcept {
-		    for (const auto& [address, upload_size]: uploads) {
-			    cached.buffer->CopyFrom(
-			        m_scheduler.Current(), m_staging_buffer, stage_offset + address - stage_address,
-			        cached.buffer->Offset(address), upload_size, vk::AccessFlagBits::eHostWrite);
-		    }
-	    });
-	return {cached.buffer, cached.buffer->Offset(vaddr)};
 }
 
 void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds) {
@@ -892,8 +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);
@@ -932,18 +730,10 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 		EXIT("BufferCache: invalid or overlapping copy range\n");
 	}
 	if (src_memory || dst_memory) {
-		std::lock_guard transaction(m_resource_mutex);
-		if (src_memory) {
-			(void)SynchronizeBacking(src_vaddr, size);
-		}
 		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 {};
-		if (src_memory && src_region.gpu_image_bytes &&
-		    !m_texture_cache.SynchronizeImageToBuffer(src_vaddr, size)) {
-			EXIT("BufferCache: GPU copy source image could not be synchronized\n");
-		}
 		if (src_memory && dst_memory && !HasGpuDirtyBytes(src_vaddr, size) &&
 		    !HasGpuDirtyBytes(dst_vaddr, size) && !src_region.gpu_image_bytes &&
 		    !dst_region.gpu_image_bytes) {
@@ -965,7 +755,7 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	}
 
 	auto& command = m_scheduler.Current();
-	auto  src = src_memory ? ObtainBuffer(command, src_vaddr, size, false, true)
+	auto  src = src_memory ? ObtainBuffer(command, src_vaddr, size, false, true, true)
 	                       : BufferBinding {.buffer = m_gds_buffer.Handle(), .offset = src_vaddr};
 	auto  dst = dst_memory ? ObtainBuffer(command, dst_vaddr, size, true, false, true)
 	                       : BufferBinding {.buffer = m_gds_buffer.Handle(), .offset = dst_vaddr};
@@ -991,18 +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) {
@@ -1010,7 +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();
 }
 
@@ -1018,78 +808,8 @@ bool BufferCache::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 	return m_memory_tracker.IsRegionCpuModified(vaddr, size);
 }
 
-void BufferCache::InvalidateImageAliases(uint64_t vaddr, uint64_t size) {
-	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
-	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid image-alias invalidation\n");
-	}
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         end = vaddr + size;
-	for (const auto& [address, cached]: m_buffers) {
-		const auto cached_end = address + cached->size;
-		const auto begin      = std::max(vaddr, address);
-		const auto range_end  = std::min(end, cached_end);
-		if (begin >= range_end) {
-			continue;
-		}
-		const auto bytes = range_end - begin;
-		if (!m_gpu_modified_ranges.Intersections(begin, bytes).empty()) {
-			EXIT("BufferCache: image ownership overlaps exact dirty buffer bytes\n");
-		}
-		m_image_invalidated_ranges.Add(begin, bytes);
-	}
-}
-
-void BufferCache::BeginBackingPublication(uint64_t vaddr, uint64_t size, uint64_t tick) {
-	if (vaddr == 0 || size == 0 || tick == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
-	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid pending backing publication\n");
-	}
-	std::lock_guard lock(m_publication_mutex);
-	m_pending_backing_publications.push_back({vaddr, size, tick});
-}
-
-void BufferCache::CompleteBackingPublication(uint64_t vaddr, uint64_t size, uint64_t tick) {
-	std::lock_guard lock(m_publication_mutex);
-	const auto      publication =
-	    std::ranges::find_if(m_pending_backing_publications, [&](const auto& pending) {
-		    return pending.address == vaddr && pending.size == size && pending.tick == tick;
-	    });
-	if (publication == m_pending_backing_publications.end()) {
-		EXIT("BufferCache: completed an unknown backing publication\n");
-	}
-	m_pending_backing_publications.erase(publication);
-}
-
-void BufferCache::PublishImageBuffer(uint64_t vaddr, uint64_t size) {
-	FaultSafeCacheLock lock(this, m_mutex);
-	auto               owner = m_buffers.end();
-	for (auto it = m_buffers.begin(); it != m_buffers.end(); ++it) {
-		if (!PageOverlaps(vaddr, size, it->second->vaddr, it->second->size)) {
-			continue;
-		}
-		if (owner != m_buffers.end() || !it->second->buffer->IsInBounds(vaddr, size)) {
-			EXIT("BufferCache: image destination aliases a non-containing cached buffer\n");
-		}
-		owner = it;
-	}
-	m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
-	                                           "image destination publication");
-	if (owner == m_buffers.end() || m_memory_tracker.IsRegionCpuModified(vaddr, size) ||
-	    !m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
-		EXIT("BufferCache: image destination requires clean buffer ownership\n");
-	}
-	m_memory_tracker.MarkRegionAsGpuModified(vaddr, size);
-	m_gpu_modified_ranges.Add(vaddr, size);
-	m_image_invalidated_ranges.Subtract(vaddr, size);
-	m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
-	                                           "published image destination");
-	owner->second->tick_accessed_last = m_gc_tick;
-}
-
 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();
 	}
@@ -1104,7 +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;
@@ -1152,7 +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 ||
@@ -1162,7 +880,6 @@ void BufferCache::RunGarbageCollector() {
 		if (!m_memory_tracker.IsRegionGpuModified(retire.address, retire.size)) {
 			m_memory_tracker.UntrackMemory(retire.address, retire.size);
 		}
-		m_image_invalidated_ranges.Subtract(retire.address, retire.size);
 		if (retire.size > m_total_used_memory) {
 			EXIT("BufferCache: allocation accounting underflow\n");
 		}

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

@@ -3,14 +3,12 @@
 
 #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"
 
 #include <map>
 #include <memory>
-#include <mutex>
 #include <span>
 #include <utility>
 #include <vector>
@@ -21,7 +19,6 @@ struct GraphicContext;
 class CommandBuffer;
 class CommandScheduler;
 class TextureCache;
-class ResourceMutex;
 
 struct BufferBinding {
 	std::shared_ptr<void> owner;
@@ -30,9 +27,8 @@ struct BufferBinding {
 };
 
 struct ImageBufferSource {
-	Buffer*  buffer    = nullptr;
-	uint64_t offset    = 0;
-	bool     gpu_owned = false;
+	Buffer*  buffer = nullptr;
+	uint64_t offset = 0;
 };
 
 class BufferCache {
@@ -43,12 +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,
@@ -60,21 +56,14 @@ public:
 	                                            uint64_t alignment);
 	[[nodiscard]] std::shared_ptr<Buffer> ObtainNullBuffer();
 	[[nodiscard]] ImageBufferSource       ObtainBufferForImage(uint64_t vaddr, uint64_t size);
-	[[nodiscard]] std::pair<std::shared_ptr<Buffer>, uint64_t>
-	     ObtainBufferForImageWrite(uint64_t vaddr, uint64_t size);
-	void DiscardGpuDirtyBytes(uint64_t vaddr, uint64_t size);
 	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);
-	void               InvalidateImageAliases(uint64_t vaddr, uint64_t size);
-	void               BeginBackingPublication(uint64_t vaddr, uint64_t size, uint64_t tick);
-	void               CompleteBackingPublication(uint64_t vaddr, uint64_t size, uint64_t tick);
-	[[nodiscard]] bool SynchronizeBacking(uint64_t vaddr, uint64_t size);
-	void               PublishImageBuffer(uint64_t vaddr, uint64_t size);
 	void               RunGarbageCollector();
 
 private:
@@ -88,47 +77,37 @@ private:
 	struct DownloadCopy;
 	struct DownloadRange;
 	struct RetiredBuffer;
-	struct PendingBackingPublication;
 	static constexpr uint64_t               DOWNLOAD_ALIGNMENT = 64;
 	[[nodiscard]] static uint64_t           AlignDown(uint64_t value) noexcept;
 	[[nodiscard]] static uint64_t           AlignUp(uint64_t value);
 	[[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,
 	            const void* source, uint64_t size);
 	[[nodiscard]] CachedBuffer& GetOrCreateBuffer(CommandBuffer& command, uint64_t vaddr,
 	                                              uint64_t size);
+	[[nodiscard]] bool SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uint64_t size);
 	[[nodiscard]] std::vector<DownloadRange> RecordDownloads(std::span<const DownloadCopy> copies);
 	void PublishDownloads(std::span<const DownloadRange> downloads);
 	void QueueGarbageDownload(std::span<const DownloadCopy> copies, RetiredBuffer retire);
-	void RefreshInvalidatedRanges(CommandBuffer& command, CachedBuffer& cached, uint64_t vaddr,
-	                              uint64_t size, bool upload);
-	void DiscardGpuDirtyBytesLocked(uint64_t vaddr, uint64_t size, const char* operation);
 	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;
-	RangeSet                                          m_image_invalidated_ranges;
-	std::mutex                                        m_publication_mutex;
-	std::vector<PendingBackingPublication>            m_pending_backing_publications;
 	MemoryTracker                                     m_memory_tracker;
 	StreamBuffer                                      m_staging_buffer;
 	StreamBuffer                                      m_stream_buffer;
 	StreamBuffer                                      m_download_buffer;
 	StreamBuffer                                      m_device_buffer;
-	PageManager&                                      m_page_manager;
 	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

@@ -1,14 +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_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;
 
@@ -17,69 +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;
 }
 
@@ -101,7 +52,17 @@ void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size) {
 }
 
 void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
+	if (CommandScheduler::InDeferredOperation()) {
+		EXIT("unsupported memory unmap from an asynchronous GPU completion, "
+		     "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();
+			m_scheduler.FinishCurrent();
+			m_scheduler.WaitPriorityOperations(tick);
+		}
 		m_buffer_cache.UnmapMemory(vaddr, size);
 		m_texture_cache.UnmapMemory(vaddr, size);
 		m_page_manager.OnGpuUnmap(vaddr, size);
@@ -109,13 +70,9 @@ void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		m_mapped_ranges.Subtract(vaddr, size);
 	};
 	if (m_gpu == nullptr) {
-		if (m_resource_mutex.IsOwnedByCurrentThread()) {
-			EXIT("cannot synchronously unmap from a resource transaction\n");
-		}
 		unmap();
 		return;
 	}
-	Gpu::SubmissionLock submissions(*m_gpu);
 	m_gpu->SendCommandSync(unmap);
 }
 

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 +34,7 @@ public:
 
 private:
 	PageManager               m_page_manager;
-	ResourceMutex             m_resource_mutex;
+	CommandScheduler&         m_scheduler;
 	BufferCache               m_buffer_cache;
 	TextureCache              m_texture_cache;
 	mutable std::shared_mutex m_mapped_ranges_mutex;

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

@@ -7,8 +7,9 @@
 #include <array>
 #include <cstddef>
 #include <cstdint>
-#include <list>
+#include <cstring>
 #include <memory>
+#include <type_traits>
 #include <utility>
 #include <vector>
 
@@ -103,6 +104,158 @@ private:
 	size_t                               m_allocated_buckets = 0;
 };
 
+// Inline owner storage for page-table entries. Texture pages normally have only a
+// handful of owners, so avoid a heap allocation on the first 16 registrations.
+//
+// Owners are packed into live uint64_t objects instead of being constructed in
+// raw storage. This keeps sparse bucket construction cheap without relying on
+// subtle implicit-lifetime or pointer-provenance rules.
+template <typename OwnerT, size_t InlineCapacity = 16>
+class InlinePageOwnerList final {
+public:
+	using value_type = OwnerT;
+	using size_type  = size_t;
+
+	class const_iterator final {
+	public:
+		[[nodiscard]] OwnerT operator*() const noexcept { return m_owners->At(m_index); }
+		const_iterator&      operator++() noexcept {
+			++m_index;
+			return *this;
+		}
+		bool operator==(const const_iterator&) const = default;
+
+	private:
+		friend class InlinePageOwnerList;
+		const_iterator(const InlinePageOwnerList* owners, size_type index) noexcept
+		    : m_owners(owners), m_index(index) {}
+
+		const InlinePageOwnerList* m_owners = nullptr;
+		size_type                  m_index  = 0;
+	};
+
+	static_assert(InlineCapacity > 0);
+	static_assert(std::is_default_constructible_v<OwnerT>);
+	static_assert(std::is_trivially_copyable_v<OwnerT>);
+	static_assert(std::has_unique_object_representations_v<OwnerT>);
+	static_assert(sizeof(OwnerT) <= sizeof(uint64_t));
+
+	InlinePageOwnerList() noexcept {}
+	InlinePageOwnerList(const InlinePageOwnerList&)            = delete;
+	InlinePageOwnerList& operator=(const InlinePageOwnerList&) = delete;
+	InlinePageOwnerList(InlinePageOwnerList&& other) noexcept
+	    : m_overflow(std::move(other.m_overflow)),
+	      m_inline_size(std::exchange(other.m_inline_size, 0)) {
+		if (m_overflow == nullptr) {
+			std::copy_n(other.m_inline.begin(), m_inline_size, m_inline.begin());
+		}
+	}
+	InlinePageOwnerList& operator=(InlinePageOwnerList&& other) noexcept {
+		if (this != &other) {
+			m_overflow    = std::move(other.m_overflow);
+			m_inline_size = std::exchange(other.m_inline_size, 0);
+			if (m_overflow == nullptr) {
+				std::copy_n(other.m_inline.begin(), m_inline_size, m_inline.begin());
+			}
+		}
+		return *this;
+	}
+
+	[[nodiscard]] const_iterator begin() const noexcept { return {this, 0}; }
+	[[nodiscard]] const_iterator end() const noexcept { return {this, size()}; }
+
+	[[nodiscard]] bool      empty() const noexcept { return size() == 0; }
+	[[nodiscard]] size_type size() const noexcept {
+		return m_overflow == nullptr ? m_inline_size : m_overflow->size();
+	}
+	[[nodiscard]] OwnerT front() const noexcept { return At(0); }
+	[[nodiscard]] OwnerT operator[](size_type index) const noexcept { return At(index); }
+
+	void push_back(const OwnerT& owner) {
+		const uint64_t packed = Pack(owner);
+		if (m_overflow != nullptr) {
+			m_overflow->push_back(packed);
+			return;
+		}
+		if (m_inline_size < InlineCapacity) {
+			m_inline[m_inline_size] = packed;
+			++m_inline_size;
+			return;
+		}
+		auto overflow = std::make_unique<std::vector<uint64_t>>();
+		overflow->reserve(InlineCapacity * 2);
+		overflow->assign(m_inline.begin(), m_inline.end());
+		overflow->push_back(packed);
+		m_overflow = std::move(overflow);
+	}
+
+	[[nodiscard]] bool Contains(const OwnerT& owner) const noexcept {
+		const uint64_t packed = Pack(owner);
+		if (m_overflow != nullptr) {
+			return std::find(m_overflow->begin(), m_overflow->end(), packed) != m_overflow->end();
+		}
+		return std::find(m_inline.begin(), m_inline.begin() + static_cast<ptrdiff_t>(m_inline_size),
+		                 packed) != m_inline.begin() + static_cast<ptrdiff_t>(m_inline_size);
+	}
+
+	[[nodiscard]] bool Erase(const OwnerT& owner) noexcept {
+		const uint64_t packed = Pack(owner);
+		if (m_overflow != nullptr) {
+			const auto found = std::find(m_overflow->begin(), m_overflow->end(), packed);
+			if (found == m_overflow->end()) {
+				return false;
+			}
+			m_overflow->erase(found);
+			if (m_overflow->size() == InlineCapacity) {
+				std::copy(m_overflow->begin(), m_overflow->end(), m_inline.begin());
+				m_inline_size = InlineCapacity;
+				m_overflow.reset();
+			}
+			return true;
+		}
+		const auto end   = m_inline.begin() + static_cast<ptrdiff_t>(m_inline_size);
+		const auto found = std::find(m_inline.begin(), end, packed);
+		if (found == end) {
+			return false;
+		}
+		std::move(found + 1, end, found);
+		--m_inline_size;
+		return true;
+	}
+
+	template <typename Func>
+	void ForEach(Func&& func) const {
+		if (m_overflow != nullptr) {
+			for (const uint64_t owner: *m_overflow) {
+				func(Unpack(owner));
+			}
+			return;
+		}
+		for (size_type index = 0; index < m_inline_size; ++index) {
+			func(Unpack(m_inline[index]));
+		}
+	}
+
+private:
+	[[nodiscard]] OwnerT At(size_type index) const noexcept {
+		return Unpack(m_overflow == nullptr ? m_inline[index] : (*m_overflow)[index]);
+	}
+	[[nodiscard]] static uint64_t Pack(const OwnerT& owner) noexcept {
+		uint64_t packed = 0;
+		std::memcpy(&packed, &owner, sizeof(owner));
+		return packed;
+	}
+	[[nodiscard]] static OwnerT Unpack(uint64_t packed) noexcept {
+		OwnerT owner {};
+		std::memcpy(&owner, &packed, sizeof(owner));
+		return owner;
+	}
+
+	std::array<uint64_t, InlineCapacity>   m_inline;
+	std::unique_ptr<std::vector<uint64_t>> m_overflow;
+	size_type                              m_inline_size = 0;
+};
+
 // Removes only the requested owner. Other owners in the same page entry remain intact.
 template <typename Container, typename Value>
 [[nodiscard]] bool EraseExact(Container& owners, const Value& owner) {
@@ -114,247 +267,6 @@ template <typename Container, typename Value>
 	return true;
 }
 
-// Multi-range ownership with 1 MiB candidate buckets and precise 4 KiB
-// lifetime accounting. OwnerT only needs equality.
-template <typename OwnerT>
-class MultiRangePageOwnerIndex final {
-private:
-	struct Registration;
-	using MembershipList = std::vector<const Registration*>;
-
-public:
-	struct ByteRange final {
-		uint64_t address = 0;
-		uint64_t size    = 0;
-	};
-
-	using CoarseTable   = MultiLevelPageTable<MembershipList, 20, 40, 10>;
-	using TrackingTable = MultiLevelPageTable<MembershipList, 12, 40, 18>;
-
-	[[nodiscard]] bool Register(const OwnerT& owner, const std::vector<ByteRange>& ranges) {
-		if (FindRegistration(owner) != m_registrations.end()) {
-			return false;
-		}
-		auto normalized = Normalize(ranges);
-		if (normalized.empty()) {
-			return false;
-		}
-		m_registrations.push_back({owner, std::move(normalized)});
-		const Registration* registration = &m_registrations.back();
-		for (const size_t page: CollectPages<20>(registration->ranges)) {
-			m_coarse_pages[page].push_back(registration);
-		}
-		for (const size_t page: CollectPages<12>(registration->ranges)) {
-			m_tracking_pages[page].push_back(registration);
-		}
-		return true;
-	}
-
-	// No state changes if an expected membership is absent. Final releases are
-	// sorted and coalesced 4 KiB spans whose final owner disappeared.
-	[[nodiscard]] bool Unregister(const OwnerT& owner, std::vector<ByteRange>& final_releases) {
-		final_releases.clear();
-		auto registration = FindRegistration(owner);
-		if (registration == m_registrations.end()) {
-			return false;
-		}
-		const auto          coarse_pages     = CollectPages<20>(registration->ranges);
-		const auto          tracking_pages   = CollectPages<12>(registration->ranges);
-		const Registration* registration_ptr = &*registration;
-		if (!HasAllMemberships(m_coarse_pages, coarse_pages, registration_ptr) ||
-		    !HasAllMemberships(m_tracking_pages, tracking_pages, registration_ptr)) {
-			return false;
-		}
-		std::vector<size_t> final_pages;
-		for (const size_t page: coarse_pages) {
-			(void)EraseExact(*m_coarse_pages.Find(page), registration_ptr);
-		}
-		for (const size_t page: tracking_pages) {
-			auto& owners = *m_tracking_pages.Find(page);
-			if (owners.size() == 1) {
-				final_pages.push_back(page);
-			}
-			(void)EraseExact(owners, registration_ptr);
-		}
-		m_registrations.erase(registration);
-		final_releases = CoalesceTrackingPages(final_pages);
-		return true;
-	}
-
-	[[nodiscard]] std::vector<OwnerT> Query(uint64_t address, uint64_t size) const {
-		return Query(address, size, [](const OwnerT&) { return true; });
-	}
-
-	template <typename Predicate>
-	[[nodiscard]] std::vector<OwnerT> Query(uint64_t address, uint64_t size,
-	                                        Predicate&& predicate) const {
-		return QueryImpl(address, size, true, std::forward<Predicate>(predicate));
-	}
-
-	// Fault paths use page candidates: exact byte-disjoint owners sharing a
-	// touched 4 KiB page are intentionally retained.
-	[[nodiscard]] std::vector<OwnerT> QueryCandidates(uint64_t address, uint64_t size) const {
-		return QueryCandidates(address, size, [](const OwnerT&) { return true; });
-	}
-
-	template <typename Predicate>
-	[[nodiscard]] std::vector<OwnerT> QueryCandidates(uint64_t address, uint64_t size,
-	                                                  Predicate&& predicate) const {
-		return QueryImpl(address, size, false, std::forward<Predicate>(predicate));
-	}
-
-	[[nodiscard]] size_t CoarseMembershipCount(size_t page) const {
-		const auto* owners = m_coarse_pages.Find(page);
-		return owners == nullptr ? 0 : owners->size();
-	}
-	[[nodiscard]] size_t TrackingMembershipCount(size_t page) const {
-		const auto* owners = m_tracking_pages.Find(page);
-		return owners == nullptr ? 0 : owners->size();
-	}
-
-private:
-	template <typename Predicate>
-	[[nodiscard]] std::vector<OwnerT> QueryImpl(uint64_t address, uint64_t size, bool strict_bytes,
-	                                            Predicate&& predicate) const {
-		typename CoarseTable::PageRange   coarse_range {};
-		typename TrackingTable::PageRange tracking_range {};
-		if (!CoarseTable::TryGetPageRange(address, size, coarse_range) ||
-		    !TrackingTable::TryGetPageRange(address, size, tracking_range)) {
-			return {};
-		}
-		MembershipList candidates;
-		for (size_t page = coarse_range.first; page < coarse_range.last_exclusive; ++page) {
-			if (const auto* owners = m_coarse_pages.Find(page); owners != nullptr) {
-				AppendUnique(candidates, *owners);
-			}
-		}
-		std::vector<OwnerT> result;
-		for (const Registration* registration: candidates) {
-			if ((!strict_bytes || Overlaps(registration->ranges, address, size)) &&
-			    HasTrackingMembership(registration, tracking_range) &&
-			    predicate(registration->owner)) {
-				result.push_back(registration->owner);
-			}
-		}
-		return result;
-	}
-
-	struct Registration final {
-		OwnerT                 owner;
-		std::vector<ByteRange> ranges;
-	};
-	using RegistrationIterator      = typename std::list<Registration>::iterator;
-	using ConstRegistrationIterator = typename std::list<Registration>::const_iterator;
-
-	[[nodiscard]] RegistrationIterator FindRegistration(const OwnerT& owner) {
-		return std::find_if(m_registrations.begin(), m_registrations.end(),
-		                    [&](const Registration& item) { return item.owner == owner; });
-	}
-	[[nodiscard]] ConstRegistrationIterator FindRegistration(const OwnerT& owner) const {
-		return std::find_if(m_registrations.begin(), m_registrations.end(),
-		                    [&](const Registration& item) { return item.owner == owner; });
-	}
-
-	[[nodiscard]] static std::vector<ByteRange> Normalize(const std::vector<ByteRange>& ranges) {
-		std::vector<ByteRange> sorted;
-		for (const auto& range: ranges) {
-			typename CoarseTable::PageRange ignored {};
-			if (!CoarseTable::TryGetPageRange(range.address, range.size, ignored)) {
-				return {};
-			}
-			sorted.push_back(range);
-		}
-		std::sort(sorted.begin(), sorted.end(), [](const ByteRange& lhs, const ByteRange& rhs) {
-			return lhs.address < rhs.address;
-		});
-		std::vector<ByteRange> merged;
-		for (const auto& range: sorted) {
-			if (merged.empty() || range.address > merged.back().address + merged.back().size) {
-				merged.push_back(range);
-			} else {
-				const uint64_t end = std::max(merged.back().address + merged.back().size,
-				                              range.address + range.size);
-				merged.back().size = end - merged.back().address;
-			}
-		}
-		return merged;
-	}
-
-	template <size_t Bits>
-	[[nodiscard]] static std::vector<size_t> CollectPages(const std::vector<ByteRange>& ranges) {
-		std::vector<size_t> pages;
-		for (const auto& range: ranges) {
-			const size_t first = static_cast<size_t>(range.address >> Bits);
-			const size_t last  = static_cast<size_t>((range.address + range.size - 1) >> Bits);
-			for (size_t page = first; page <= last; ++page) {
-				pages.push_back(page);
-			}
-		}
-		std::sort(pages.begin(), pages.end());
-		pages.erase(std::unique(pages.begin(), pages.end()), pages.end());
-		return pages;
-	}
-
-	template <typename Table>
-	[[nodiscard]] static bool HasAllMemberships(const Table&               table,
-	                                            const std::vector<size_t>& pages,
-	                                            const Registration*        registration) {
-		for (const size_t page: pages) {
-			const auto* owners = table.Find(page);
-			if (owners == nullptr ||
-			    std::find(owners->begin(), owners->end(), registration) == owners->end()) {
-				return false;
-			}
-		}
-		return true;
-	}
-
-	[[nodiscard]] bool HasTrackingMembership(const Registration*                      registration,
-	                                         const typename TrackingTable::PageRange& range) const {
-		for (size_t page = range.first; page < range.last_exclusive; ++page) {
-			const auto* owners = m_tracking_pages.Find(page);
-			if (owners != nullptr &&
-			    std::find(owners->begin(), owners->end(), registration) != owners->end()) {
-				return true;
-			}
-		}
-		return false;
-	}
-
-	[[nodiscard]] static bool Overlaps(const std::vector<ByteRange>& ranges, uint64_t address,
-	                                   uint64_t size) {
-		const uint64_t end = address + size;
-		return std::any_of(ranges.begin(), ranges.end(), [&](const ByteRange& range) {
-			return range.address < end && address < range.address + range.size;
-		});
-	}
-	static void AppendUnique(MembershipList& destination, const MembershipList& source) {
-		for (const Registration* registration: source) {
-			if (std::find(destination.begin(), destination.end(), registration) ==
-			    destination.end()) {
-				destination.push_back(registration);
-			}
-		}
-	}
-	[[nodiscard]] static std::vector<ByteRange>
-	CoalesceTrackingPages(const std::vector<size_t>& pages) {
-		std::vector<ByteRange> result;
-		for (const size_t page: pages) {
-			const uint64_t address = static_cast<uint64_t>(page) << 12;
-			if (!result.empty() && result.back().address + result.back().size == address) {
-				result.back().size += uint64_t {1} << 12;
-			} else {
-				result.push_back({address, uint64_t {1} << 12});
-			}
-		}
-		return result;
-	}
-
-	CoarseTable             m_coarse_pages;
-	TrackingTable           m_tracking_pages;
-	std::list<Registration> m_registrations;
-};
-
 } // namespace Libs::Graphics
 
 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_MULTILEVELPAGETABLE_H_

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;
@@ -195,10 +193,12 @@ void TextureCache::RegisterImage(ImageId id) {
 	if (image.registered || image.info.data.Empty()) {
 		EXIT("TextureCache: invalid image registration\n");
 	}
-	std::vector<ImageOwnerIndex::ByteRange> ranges;
-	ranges.push_back({image.info.data.address, image.info.data.size});
-	if (!m_image_owner_index.Register(id, ranges)) {
-		EXIT("TextureCache: duplicate or invalid image registration\n");
+	ImagePageTable::PageRange pages {};
+	if (!ImagePageTable::TryGetPageRange(image.info.data.address, image.info.data.size, pages)) {
+		EXIT("TextureCache: image registration is outside the guest address space\n");
+	}
+	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+		m_image_page_table[page].push_back(id);
 	}
 	image.registered = true;
 	image.lru_id     = m_lru_cache.Insert(id, m_gc_tick);
@@ -211,9 +211,15 @@ void TextureCache::UnregisterImage(ImageId id) {
 		return;
 	}
 	UntrackImage(id);
-	std::vector<ImageOwnerIndex::ByteRange> releases;
-	if (!m_image_owner_index.Unregister(id, releases)) {
-		EXIT("TextureCache: image missing from owner index\n");
+	ImagePageTable::PageRange pages {};
+	if (!ImagePageTable::TryGetPageRange(image.info.data.address, image.info.data.size, pages)) {
+		EXIT("TextureCache: registered image is outside the guest address space\n");
+	}
+	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+		auto* owners = m_image_page_table.Find(page);
+		if (owners == nullptr || !owners->Erase(id)) {
+			EXIT("TextureCache: image missing from page owner index\n");
+		}
 	}
 	m_lru_cache.Free(image.lru_id);
 	const auto accounted = image.AccountedSize();
@@ -323,7 +329,7 @@ void TextureCache::TrackImage(ImageId id) {
 	if (!image.IsTracked()) {
 		image.track_addr     = image_begin;
 		image.track_addr_end = image_end;
-		m_page_manager.UpdatePageWatchers(true, image_begin, image.info.data.size);
+		m_page_manager.UpdatePageWatchers<true>(image_begin, image.info.data.size);
 		return;
 	}
 	if (image_begin < image.track_addr) {
@@ -348,7 +354,7 @@ void TextureCache::TrackImageHead(ImageId id) {
 	}
 	const auto size  = image.track_addr - image_begin;
 	image.track_addr = image_begin;
-	m_page_manager.UpdatePageWatchers(true, image_begin, size);
+	m_page_manager.UpdatePageWatchers<true>(image_begin, size);
 }
 
 void TextureCache::TrackImageTail(ImageId id) {
@@ -366,7 +372,7 @@ void TextureCache::TrackImageTail(ImageId id) {
 	const auto address   = image.track_addr_end;
 	const auto size      = image_end - address;
 	image.track_addr_end = image_end;
-	m_page_manager.UpdatePageWatchers(true, address, size);
+	m_page_manager.UpdatePageWatchers<true>(address, size);
 }
 
 void TextureCache::UntrackImage(ImageId id) {
@@ -379,7 +385,7 @@ void TextureCache::UntrackImage(ImageId id) {
 	image.track_addr     = 0;
 	image.track_addr_end = 0;
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, address, size);
+		m_page_manager.UpdatePageWatchers<false>(address, size);
 	}
 }
 
@@ -400,7 +406,7 @@ void TextureCache::UntrackImageHead(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, begin, size);
+		m_page_manager.UpdatePageWatchers<false>(begin, size);
 	}
 }
 
@@ -421,14 +427,13 @@ void TextureCache::UntrackImageTail(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, address, size);
+		m_page_manager.UpdatePageWatchers<false>(address, size);
 	}
 }
 
 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);
 }
 
@@ -451,10 +456,48 @@ const Image& TextureCache::GetImage(ImageId id) const {
 	return ResolveImage(id);
 }
 
-std::vector<ImageId> TextureCache::FindImagesInRegion(uint64_t address, uint64_t size,
-                                                      bool page_overlap) const {
-	return page_overlap ? m_image_owner_index.QueryCandidates(address, size)
-	                    : m_image_owner_index.Query(address, size);
+TextureCache::ImageIds TextureCache::FindImagesInRegion(uint64_t address, uint64_t size,
+                                                        bool page_overlap) const {
+	ImagePageTable::PageRange pages {};
+	if (!ImagePageTable::TryGetPageRange(address, size, pages)) {
+		return {};
+	}
+
+	uint32_t query_epoch = ++m_image_query_epoch;
+	if (query_epoch == 0) {
+		for (const auto& slot: m_slots) {
+			if (slot.image != nullptr) {
+				slot.image->query_epoch = 0;
+			}
+		}
+		query_epoch = ++m_image_query_epoch;
+	}
+
+	ImageIds result;
+	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+		const auto* owners = m_image_page_table.Find(page);
+		if (owners == nullptr) {
+			continue;
+		}
+		owners->ForEach([&](ImageId id) {
+			if (!id || id.index >= m_slots.size()) {
+				return;
+			}
+			const auto& slot = m_slots[id.index];
+			if (slot.generation != id.generation || slot.image == nullptr) {
+				return;
+			}
+			auto& image = *slot.image;
+			if (image.query_epoch == query_epoch) {
+				return;
+			}
+			image.query_epoch = query_epoch;
+			if (image.Overlaps(address, size, page_overlap)) {
+				result.push_back(id);
+			}
+		});
+	}
+	return result;
 }
 
 ImageId TextureCache::GetNullImage(const ImageDesc& desc) {
@@ -838,11 +881,20 @@ struct TextureCache::ColorTransferPlan {
 	TextureUploadLayout              layout;
 	std::vector<vk::BufferImageCopy> regions;
 	std::vector<GpuTileInfo>         tiles;
+	uint64_t                         linear_size = 0;
 	bool                             tiled       = false;
 	bool                             swap_bgra16 = false;
 	bool                             valid       = false;
 };
 
+static uint64_t GetLinearSize(std::span<const GpuTileInfo> tiles) {
+	uint64_t size = 0;
+	for (const auto& tile: tiles) {
+		size = std::max(size, tile.linear_offset + tile.linear_size);
+	}
+	return size;
+}
+
 struct TextureCache::DownloadPlan {
 	ColorTransferPlan color;
 	bool              depth = false;
@@ -856,7 +908,6 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 	uint32_t    format           = info.guest_format;
 	uint32_t    layers           = info.TransferLayers();
 	bool        volume           = info.IsVolume();
-	bool        layered          = info.IsLayered();
 	bool        allow_depth_tile = direction == TransferDirection::Upload;
 	const char* owner =
 	    direction == TransferDirection::Upload ? "TextureCache" : "TextureCache readback";
@@ -867,22 +918,27 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 			case BindingType::Texture: break;
 			case BindingType::Storage: owner = "StorageTextureCache"; break;
 			case BindingType::RenderTarget:
+				if (info.resources.layers == 0 || info.data.size % info.resources.layers != 0 ||
+				    info.samples != 1 || image.backing.samples != 1) {
+					EXIT("TextureCache: invalid color-attachment upload\n");
+				}
+				format           = ImageOps::RenderTargetTransferFormat(info.bytes_per_block);
+				allow_depth_tile = false;
+				plan.swap_bgra16 = info.bgra16;
+				owner            = "RenderTarget";
+				break;
 			case BindingType::VideoOut:
 				if (info.resources.layers == 0 || info.data.size % info.resources.layers != 0 ||
 				    info.samples != 1 || image.backing.samples != 1 ||
-				    (binding == BindingType::VideoOut &&
-				     info.metadata.compression != VideoOutCompression::Uncompressed)) {
+				    info.metadata.compression != VideoOutCompression::Uncompressed) {
 					EXIT("TextureCache: invalid color-attachment upload\n");
 				}
-				format           = binding == BindingType::RenderTarget
-				                       ? ImageOps::RenderTargetTransferFormat(info.bytes_per_block)
-				                       : info.guest_format;
+				format           = info.guest_format;
 				layers           = info.resources.layers;
 				volume           = false;
-				layered          = layers > 1;
 				allow_depth_tile = false;
 				plan.swap_bgra16 = info.bgra16;
-				owner = binding == BindingType::RenderTarget ? "RenderTarget" : "VideoOut";
+				owner            = "VideoOut";
 				break;
 			case BindingType::DepthTarget: return plan;
 		}
@@ -897,22 +953,18 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 		plan.swap_bgra16 = info.bgra16;
 	}
 
-	plan.layout = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
-	                                      info.resources.levels, layers, info.pitch, info.tile_mode,
-	                                      info.data.size, allow_depth_tile, volume, owner);
-	plan.regions = TextureBuildImageCopies(plan.layout, info.extent.width, info.extent.height,
-	                                       layers, info.resources.levels, layered, volume);
-	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.tile) != Prospero::TileMode::kLinear;
+	plan.layout  = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
+	                                       info.resources.levels, layers, info.tile_mode,
+	                                       info.data.size, allow_depth_tile, volume, owner);
+	plan.regions = TextureBuildImageCopies(plan.layout);
+	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.surface.description.tile_mode) !=
+	               Prospero::TileMode::kLinear;
 	if (plan.tiled) {
-		// CMASK/FMASK decoding is intentionally deferred.
-		if (plan.layout.tile_family == TileBlockFamily::Depth64KB &&
-		    Prospero::IsFmaskTextureFormat(format)) {
-			return plan;
-		}
-		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout, format, layers,
+		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout,
 		                              info.resources.levels, plan.tiles)) {
 			return plan;
 		}
+		plan.linear_size = GetLinearSize(plan.tiles);
 	}
 	plan.valid = true;
 	return plan;
@@ -950,11 +1002,20 @@ void TextureCache::UploadImage(Image& image, const ImageDesc& desc, Buffer& sour
 
 	if (desc.type != BindingType::DepthTarget) {
 		auto plan = BuildColorTransfer(image, desc.type, TransferDirection::Upload);
-		EXIT_NOT_IMPLEMENTED(!plan.valid);
+		if (!plan.valid) {
+			EXIT("TextureCache: invalid color upload: binding=%u addr=0x%016" PRIx64
+			     " 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.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) {
-			linear = m_tiler->Detile(source.Handle(), source_offset, info.data.size, info.data.size,
-			                         plan.tiles);
+			linear = m_tiler->Detile(source.Handle(), source_offset, info.data.size,
+			                         plan.linear_size, plan.tiles);
 		}
 		if (plan.swap_bgra16) {
 			linear = m_tiler->SwapBgra16(linear);
@@ -1038,25 +1099,20 @@ void TextureCache::InitializeImage(ImageId id, const ImageDesc& desc) {
 	if (image.info.samples > 1) {
 		return;
 	}
-	bool       data_gpu_owned = false;
-	bool       data_imported  = false;
-	const bool upload         = image.IsBufferModified() || image.IsCpuDirty();
+	bool       data_imported = false;
+	const bool upload        = image.IsBufferModified() || image.IsCpuDirty();
 	if (upload) {
 		const auto source =
 		    m_buffer_cache.ObtainBufferForImage(image.info.data.address, image.info.data.size);
 		if (source.buffer == nullptr) {
 			EXIT("TextureCache: failed to obtain image upload source\n");
 		}
-		data_gpu_owned |= source.gpu_owned;
 		data_imported = true;
 		UploadImage(image, desc, *source.buffer, source.offset);
 	}
 	if (data_imported) {
 		image.ClearBufferModified();
 	}
-	if (data_gpu_owned) {
-		image.MarkGpuModified();
-	}
 	if (image.IsCpuDirty()) {
 		image.RefreshComplete();
 	}
@@ -1087,8 +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);
 }
 
@@ -1131,12 +1186,9 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	}
 
 	ImageId result {};
-	bool    replacement_buffer = false;
-	bool    inserted_new       = false;
 	{
-		std::lock_guard            transaction(m_resource_mutex);
-		CacheLock                  lock(*this, m_lock);
-		const std::vector<ImageId> candidates =
+		CacheLock  lock(*this, m_lock);
+		const auto candidates =
 		    FindImagesInRegion(desc.info.data.address, desc.info.data.size, false);
 
 		for (const auto id: candidates) {
@@ -1159,8 +1211,8 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 				if (owner == nullptr) {
 					continue;
 				}
-				const auto merged_info = result ? ResolveImage(result).info : desc.info;
-				const auto overlap     = ResolveOverlap(merged_info, desc.type, candidate, result);
+				const auto& merged_info = result ? ResolveImage(result).info : desc.info;
+				const auto  overlap     = ResolveOverlap(merged_info, desc.type, candidate, result);
 				if (overlap.image) {
 					result     = overlap.image;
 					view_mip   = overlap.mip;
@@ -1174,32 +1226,17 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 			if (exact_format && resolved.info.pixel_format != desc.info.pixel_format) {
 				result = {};
 			} else if (resolved.info.resources < desc.info.resources) {
-				ImageDesc refresh {
-				    .info = resolved.info, .view_info = {}, .type = UploadBinding(resolved)};
-				RefreshImage(result, refresh);
-				if (resolved.IsGpuModified() && !SynchronizeImageToBuffer(result)) {
-					EXIT("TextureCache: cannot preserve an unsupported replacement image\n");
-				}
-				replacement_buffer = resolved.IsBufferModified();
-				DeleteImage(result);
-				result = {};
+				result = ExpandImage(desc.info, result);
 			}
 		}
 		if (!result) {
 			result         = InsertImage(desc.info);
-			inserted_new   = true;
 			auto& inserted = ResolveImage(result);
-			if (replacement_buffer || m_buffer_cache.HasGpuDirtyBytes(inserted.info.data.address,
-			                                                          inserted.info.data.size)) {
+			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) {
@@ -1224,6 +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 {};
@@ -1268,22 +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");
 				}
@@ -1302,16 +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);
@@ -1322,13 +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;
@@ -1336,7 +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);
 	}
@@ -1346,9 +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");
 	}
@@ -1360,11 +1405,6 @@ void TextureCache::CommitGpuWrite(Image& image) {
 	if (image.depth_id || image.backing.image == nullptr) {
 		EXIT("TextureCache: stencil association cannot own image contents\n");
 	}
-	const auto range = image.info.data;
-	if (m_buffer_cache.HasGpuDirtyBytes(range.address, range.size)) {
-		m_buffer_cache.DiscardGpuDirtyBytes(range.address, range.size);
-	}
-	m_buffer_cache.InvalidateImageAliases(range.address, range.size);
 	image.ClearBufferModified();
 	if (image.IsCpuDirty()) {
 		image.RefreshComplete();
@@ -1377,7 +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 {};
@@ -1429,9 +1468,6 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 			return false;
 		}
 	}
-	if (m_buffer_cache.HasGpuDirtyBytes(address, size)) {
-		m_buffer_cache.DiscardGpuDirtyBytes(address, size);
-	}
 	if (image.IsBufferModified() || image.IsCpuDirty()) {
 		ImageDesc refresh {.info = image.info, .view_info = {}, .type = UploadBinding(image)};
 		InitializeImage(selected, refresh);
@@ -1551,11 +1587,14 @@ void TextureCache::DownloadDepth(Image& image, Buffer& destination, uint64_t des
 }
 
 void TextureCache::DownloadImageData(Image& image, Buffer& destination, uint64_t destination_offset,
-                                     DownloadPlan plan) {
+                                     uint64_t destination_size, DownloadPlan plan) {
 	if (!plan.valid) {
 		EXIT("TextureCache: invalid image download plan\n");
 	}
 	if (plan.depth) {
+		if (destination_size != image.info.data.size) {
+			EXIT("TextureCache: partial depth image download is unsupported\n");
+		}
 		DownloadDepth(image, destination, destination_offset);
 		return;
 	}
@@ -1565,22 +1604,114 @@ void TextureCache::DownloadImageData(Image& image, Buffer& destination, uint64_t
 	                                         : TileManager::ColorTransform::None;
 	if (!color.tiled) {
 		if (transform == TileManager::ColorTransform::SwapBgra16) {
-			auto linear = m_tiler->GetScratchBuffer(image.info.data.size);
+			auto linear = m_tiler->GetScratchBuffer(destination_size);
 			image.Download(color.regions, linear.buffer, 0, linear.size);
 			m_tiler->SwapBgra16(linear,
-			                    {destination.Handle(), destination_offset, image.info.data.size});
+			                    {destination.Handle(), destination_offset, destination_size});
 			return;
 		}
 		for (auto& copy: color.regions) {
 			copy.bufferOffset += destination_offset;
 		}
-		image.Download(color.regions, destination.Handle(), destination_offset,
-		               image.info.data.size);
+		image.Download(color.regions, destination.Handle(), destination_offset, destination_size);
 		return;
 	}
 
 	m_tiler->TileImage(image, color.regions, destination.Handle(), destination_offset,
-	                   image.info.data.size, image.info.data.size, color.tiles, transform);
+	                   destination_size, color.linear_size, color.tiles, transform);
+}
+
+bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uint64_t size) {
+	CacheLock            lock(m_texture_cache, m_texture_cache.m_lock);
+	std::vector<ImageId> matches;
+	for (const auto id: m_texture_cache.FindImagesInRegion(vaddr, size, false)) {
+		auto owner = m_texture_cache.ResolveOwner(id);
+		if (owner == nullptr || owner->info.data.address != vaddr) {
+			continue;
+		}
+		if (owner->depth_id) {
+			owner = m_texture_cache.ResolveOwner(owner->depth_id);
+		}
+		if (owner != nullptr && owner->SafeToDownload()) {
+			matches.push_back(id);
+		}
+	}
+
+	ImageId selected {};
+	if (matches.size() == 1) {
+		selected = matches.front();
+	} else {
+		for (const auto id: matches) {
+			const auto& image = m_texture_cache.ResolveImage(id);
+			if (image.info.data.size == size) {
+				selected = id;
+				break;
+			}
+		}
+	}
+	if (!selected) {
+		return false;
+	}
+	if (const auto owner = m_texture_cache.ResolveOwner(selected);
+	    owner != nullptr && owner->depth_id) {
+		selected = owner->depth_id;
+	}
+
+	auto& image = m_texture_cache.ResolveImage(selected);
+	if (!buffer.IsInBounds(image.info.data.address, 1)) {
+		return false;
+	}
+	const auto buf_offset = buffer.Offset(image.info.data.address);
+	const auto available  = buffer.Size() - buf_offset;
+	uint32_t   levels     = 0;
+	uint64_t   copy_size  = 0;
+	if (image.info.IsVolume()) {
+		// Volume mips contain strided block slices, so a mip's linear span cannot prove that
+		// every retained slice fits. Keep volume synchronization whole-image only.
+		if (!buffer.IsInBounds(image.info.data.address, image.info.data.size)) {
+			return false;
+		}
+		levels    = image.info.resources.levels;
+		copy_size = image.info.data.size;
+	} else {
+		for (; levels < image.info.resources.levels; ++levels) {
+			const auto& mip = image.info.mip_layout[levels];
+			if (mip.size == 0 || mip.offset > available || mip.size > available - mip.offset) {
+				break;
+			}
+			copy_size = std::max(copy_size, mip.offset + mip.size);
+		}
+	}
+	if (copy_size == 0) {
+		return false;
+	}
+	auto plan = m_texture_cache.BuildDownload(image);
+	if (!plan.valid) {
+		return false;
+	}
+	if (plan.depth && copy_size != image.info.data.size) {
+		return false;
+	}
+	if (!plan.depth && levels < image.info.resources.levels) {
+		auto& color = plan.color;
+		std::erase_if(color.regions, [levels](const vk::BufferImageCopy& region) {
+			return region.imageSubresource.mipLevel >= levels;
+		});
+		if (color.regions.empty()) {
+			return false;
+		}
+		if (color.tiled) {
+			color.tiles.clear();
+			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, levels,
+			                              color.tiles)) {
+				return false;
+			}
+			color.linear_size = GetLinearSize(color.tiles);
+		}
+	}
+	m_texture_cache.DownloadImageData(image, buffer, buf_offset, copy_size, std::move(plan));
+	m_texture_cache.RetainImage(m_scheduler.Current(), selected);
+	return true;
 }
 
 std::pair<uint8_t*, uint64_t> TextureCache::MapDownload(uint64_t size, uint64_t alignment) {
@@ -1612,12 +1743,9 @@ void TextureCache::QueueDownload(GuestRange range, StreamBuffer& download, uint8
 	m_scheduler.Current().Handle().pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
 	                                               vk::PipelineStageFlagBits::eHost, {}, 0, nullptr,
 	                                               1, &barrier, 0, nullptr);
-	const auto tick = m_scheduler.CurrentTick();
-	m_buffer_cache.BeginBackingPublication(range.address, range.size, tick);
-	m_scheduler.DeferPriorityOperation([this, &download, range, mapped, offset, tick] {
+	m_scheduler.DeferPriorityOperation([&download, range, mapped, offset] {
 		download.Invalidate(offset, range.size);
 		LibKernel::Memory::WriteBacking(range.address, mapped, range.size);
-		m_buffer_cache.CompleteBackingPublication(range.address, range.size, tick);
 	});
 }
 
@@ -1638,7 +1766,7 @@ bool TextureCache::TryDownloadImage(ImageId id) {
 	}
 	download.Flush(offset, range.size);
 
-	DownloadImageData(image, download, offset, std::move(plan));
+	DownloadImageData(image, download, offset, range.size, std::move(plan));
 
 	QueueDownload(range, download, mapped, offset);
 	return true;
@@ -1652,62 +1780,6 @@ void TextureCache::DownloadImage(ImageId id) {
 	m_scheduler.DrainPriorityOperations();
 }
 
-bool TextureCache::SynchronizeImageToBuffer(ImageId id) {
-	auto& image = ResolveImage(id);
-	if (image.depth_id) {
-		return true;
-	}
-	auto plan = BuildDownload(image);
-	if (!plan.valid) {
-		return false;
-	}
-	const auto range = image.info.data;
-	if (image.IsCpuDirty()) {
-		RefreshImage(id,
-		             ImageDesc {.info = image.info, .view_info = {}, .type = UploadBinding(image)});
-	}
-	if (!image.IsGpuModified()) {
-		return true;
-	}
-	if (image.IsDefinitelyCpuDirty() || image.IsBufferModified()) {
-		EXIT("TextureCache: image mirror source is not native-current\n");
-	}
-	auto [destination, offset] =
-	    m_buffer_cache.ObtainBufferForImageWrite(range.address, range.size);
-	if (destination == nullptr) {
-		EXIT("TextureCache: failed to allocate image mirror\n");
-	}
-	DownloadImageData(image, *destination, offset, std::move(plan));
-	m_scheduler.Current().RetainResourceUntilFence(destination);
-	m_buffer_cache.PublishImageBuffer(range.address, range.size);
-	image.MarkBufferModified();
-	RetainImage(m_scheduler.Current(), id);
-	ClearGpuModified(id);
-	return true;
-}
-
-bool TextureCache::SynchronizeImageToBuffer(uint64_t address, uint64_t size) {
-	if (!GuestRange {address, size}.Valid()) {
-		return false;
-	}
-	CacheLock lock(*this, m_lock);
-	ImageId   selected {};
-	for (const auto id: FindImagesInRegion(address, size, true)) {
-		auto owner = ResolveOwner(id);
-		if (owner == nullptr || !owner->GpuOverlaps(address, size)) {
-			continue;
-		}
-		if (selected) {
-			EXIT("TextureCache: ambiguous image-to-buffer synchronization\n");
-		}
-		selected = id;
-	}
-	if (!selected) {
-		return false;
-	}
-	return SynchronizeImageToBuffer(selected);
-}
-
 bool TextureCache::InvalidateMemoryFromGPU(uint64_t address, uint64_t size,
                                            bool formatted_buffer_write) {
 	if (!GuestRange {address, size}.Valid()) {
@@ -1802,9 +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;
 	}
@@ -1830,9 +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();
 	}
@@ -1849,9 +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();
 	}
@@ -1911,8 +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);
@@ -66,7 +66,6 @@ public:
 	[[nodiscard]] bool ClearImageFromBuffer(CommandBuffer& command, uint64_t address, uint64_t size,
 	                                        uint32_t packed_clear);
 	void               InvalidateMemory(uint64_t address, uint64_t size);
-	[[nodiscard]] bool SynchronizeImageToBuffer(uint64_t address, uint64_t size);
 	[[nodiscard]] bool InvalidateMemoryFromGPU(uint64_t address, uint64_t size,
 	                                           bool formatted_buffer_write = false);
 	[[nodiscard]] RegionInfo QueryRegion(uint64_t address, uint64_t size);
@@ -100,7 +99,8 @@ private:
 		int32_t layer = -1;
 	};
 
-	using ImageOwnerIndex = MultiRangePageOwnerIndex<ImageId>;
+	using ImageIds       = InlinePageOwnerList<ImageId, 16>;
+	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
 
 	[[nodiscard]] Image&                 ResolveImage(ImageId id);
 	[[nodiscard]] const Image&           ResolveImage(ImageId id) const;
@@ -126,8 +126,9 @@ private:
 	[[nodiscard]] static BindingType UploadBinding(const Image& image);
 	[[nodiscard]] bool               SafeToDownload(const Image& image);
 
-	[[nodiscard]] std::vector<ImageId> FindImagesInRegion(uint64_t address, uint64_t size,
-	                                                      bool page_overlap) const;
+	// Caller holds m_lock; it also serializes the per-image query epoch.
+	[[nodiscard]] ImageIds      FindImagesInRegion(uint64_t address, uint64_t size,
+	                                               bool page_overlap) const;
 	[[nodiscard]] OverlapResult ResolveOverlap(const ImageInfo& requested, BindingType binding,
 	                                           ImageId cached, ImageId merged);
 	[[nodiscard]] ImageId       ResolveDepthOverlap(const ImageInfo& requested, BindingType binding,
@@ -140,7 +141,7 @@ private:
 	[[nodiscard]] DownloadPlan      BuildDownload(const Image& image) const;
 	void UploadImage(Image& image, const ImageDesc& desc, Buffer& source, uint64_t source_offset);
 	void DownloadImageData(Image& image, Buffer& destination, uint64_t destination_offset,
-	                       DownloadPlan plan);
+	                       uint64_t destination_size, DownloadPlan plan);
 	void DownloadDepth(Image& image, Buffer& destination, uint64_t destination_offset);
 	void CommitGpuWrite(Image& image);
 	void PrepareImageCopy(Image& image);
@@ -155,7 +156,6 @@ private:
 	void InvalidateCpuAliases(uint64_t address, uint64_t size);
 	void ClearGpuModified(ImageId id);
 
-	[[nodiscard]] bool                          SynchronizeImageToBuffer(ImageId id);
 	void                                        DownloadImage(ImageId id);
 	[[nodiscard]] bool                          TryDownloadImage(ImageId id);
 	[[nodiscard]] std::pair<uint8_t*, uint64_t> MapDownload(uint64_t size, uint64_t alignment);
@@ -168,10 +168,9 @@ 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;
-	ImageOwnerIndex                                   m_image_owner_index;
+	ImagePageTable                                    m_image_page_table;
 	std::map<vk::Format, ImageId>                     m_null_images;
 	Common::LeastRecentlyUsedCache<ImageId, uint64_t> m_lru_cache;
 	std::set<ImageId>                                 m_download_images;
@@ -179,11 +178,13 @@ private:
 	uint64_t                                          m_total_used_memory  = 0;
 	uint64_t                                          m_trigger_gc_memory  = 0;
 	uint64_t                                          m_pressure_gc_memory = 1536ull * 1024 * 1024;
-	uint64_t m_critical_gc_memory     = 3ull * 1024 * 1024 * 1024;
-	uint64_t m_gc_tick                = 0;
-	bool     m_readback_linear_images = false;
+	uint64_t         m_critical_gc_memory     = 3ull * 1024 * 1024 * 1024;
+	uint64_t         m_gc_tick                = 0;
+	mutable uint32_t m_image_query_epoch      = 0;
+	bool             m_readback_linear_images = false;
 
 	friend struct TextureCacheTestAccess;
+	friend class BufferCache;
 	friend class RenderExecutor;
 };
 

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -119,7 +119,18 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	uint32_t   pitch  = 0;
 	uint64_t   size   = 0;
 	bool       tile   = false;
-	const bool standard64 =
+	const bool volume = rt.attrib3.dimension == 2;
+	if (rt.attrib3.dimension != 1 && !volume) {
+		EXIT("unsupported render-target dimension: %u\n", rt.attrib3.dimension);
+	}
+	if (!volume && rt.attrib3.depth != 0) {
+		EXIT("2D render target has nonzero depth: %u\n", rt.attrib3.depth);
+	}
+	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);
 
 	switch (rt.attrib3.tile_mode) {
@@ -145,6 +156,7 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	if (bytes_per_element == 0) {
 		EXIT("render-target format has no valid element size\n");
 	}
+	const auto transfer_format = ImageOps::RenderTargetTransferFormat(bytes_per_element);
 	if (standard64 &&
 	    (rt.attrib3.dimension != 1 || rt.attrib3.depth != 0 || levels != 1 ||
 	     rt.view.current_mip_level != 0 || view.base_layer != 0 || view.image_layers != 1 ||
@@ -165,10 +177,14 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	if (rt.pitch.pitch_div8_minus1 != 0) {
 		pitch = (rt.pitch.pitch_div8_minus1 + 1u) << 3u;
 	} else if (tile) {
-		pitch = standard64
-		            ? TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
-		                                  width, levels, rt.attrib3.tile_mode)
-		            : TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
+		if (volume) {
+			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
+		} else if (standard64) {
+			pitch = TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
+			                            width, rt.attrib3.tile_mode);
+		} else {
+			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
+		}
 		if (pitch == 0) {
 			EXIT("unsupported render-target pitch: width=%u bytes=%u\n", width, bytes_per_element);
 		}
@@ -176,14 +192,31 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		pitch = width;
 	}
 
-	TileSizeOffset mip_sizes[16] {};
-	TilePaddedSize mip_padded[16] {};
-	if (tile) {
+	TileSizeOffset    mip_sizes[16] {};
+	TilePaddedSize    mip_padded[16] {};
+	TileSurfaceLayout volume_layout {};
+	uint64_t          backing_size = 0;
+	if (volume) {
+		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);
+		}
+		size         = volume_layout.block_slice_size;
+		backing_size = volume_layout.total_size;
+	} else if (tile) {
 		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,
+			                   height, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
 			                   mip_padded);
 			valid_layout = layout.size != 0 && layout.align == 65536;
 		} else {
@@ -203,12 +236,6 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 			mip_sizes[0]  = {static_cast<uint32_t>(size), 0, 0, 0, 0, 0};
 			mip_padded[0] = {pitch, height};
 		}
-		if (rt.slice.slice_div64_minus1 != 0 &&
-		    (static_cast<uint64_t>(rt.slice.slice_div64_minus1) + 1u) * 64u != size) {
-			EXIT("render-target slice span mismatch: encoded=0x%016" PRIx64 " derived=0x%016" PRIx64
-			     "\n",
-			     (static_cast<uint64_t>(rt.slice.slice_div64_minus1) + 1u) * 64u, size);
-		}
 	} else {
 		size = static_cast<uint64_t>(pitch) * height * bytes_per_element * samples;
 		if (size > UINT32_MAX) {
@@ -217,40 +244,67 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		mip_sizes[0]  = {static_cast<uint32_t>(size), 0, 0, 0, 0, 0};
 		mip_padded[0] = {pitch, height};
 	}
-	if (size == 0 || size > UINT64_MAX / view.image_layers) {
+	if (rt.slice.slice_div64_minus1 != 0 &&
+	    (static_cast<uint64_t>(rt.slice.slice_div64_minus1) + 1u) * 64u != size) {
+		EXIT("render-target slice span mismatch: encoded=0x%016" PRIx64 " derived=0x%016" PRIx64
+		     "\n",
+		     (static_cast<uint64_t>(rt.slice.slice_div64_minus1) + 1u) * 64u, size);
+	}
+	if (size == 0 || (!volume && size > UINT64_MAX / view.image_layers)) {
 		EXIT("render-target memory footprint is invalid\n");
 	}
-	const auto backing_size = size * view.image_layers;
+	if (!volume) {
+		backing_size = size * view.image_layers;
+	}
+	if (backing_size == 0) {
+		EXIT("render-target backing is empty\n");
+	}
 	if (backing_size > TRACKER_ADDRESS_SIZE - rt.base.addr) {
 		EXIT("render-target backing range is invalid\n");
 	}
 
 	const vk::Extent2D view_extent = {std::max(width >> rt.view.current_mip_level, 1u),
 	                                  std::max(height >> rt.view.current_mip_level, 1u)};
+	const uint32_t     view_depth  = std::max(depth >> rt.view.current_mip_level, 1u);
+	if (volume &&
+	    (view.base_layer >= view_depth || view.layer_count > view_depth - view.base_layer)) {
+		EXIT("3D render-target view exceeds mip depth: base=%u count=%u depth=%u mip=%u\n",
+		     view.base_layer, view.layer_count, view_depth, rt.view.current_mip_level);
+	}
 
 	auto decision_log_id = g_render_color_log_count.fetch_add(1);
 	if (decision_log_id < 128) {
 		LOGF("RenderColorTarget: slot=%" PRIu32 " addr=0x%010" PRIx64 " size=0x%016" PRIx64
-		     " extent=%ux%u view_mip=%u view_extent=%ux%u levels=%u pitch=%u"
+		     " 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, rt.view.current_mip_level,
+		     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");
 	}
 
 	TextureCache::ImageDesc desc {};
-	desc.type                 = TextureCache::BindingType::RenderTarget;
-	desc.info.data            = {rt.base.addr, backing_size};
-	desc.info.pixel_format    = target_format.format;
-	desc.info.guest_format    = ImageOps::RenderTargetTransferFormat(bytes_per_element);
-	desc.info.type            = Prospero::ImageType::kColor2D;
-	desc.info.extent          = {width, height, 1};
-	desc.info.resources       = {levels, view.image_layers};
-	desc.info.pitch           = pitch;
+	desc.type              = TextureCache::BindingType::RenderTarget;
+	desc.info.data         = {rt.base.addr, backing_size};
+	desc.info.pixel_format = target_format.format;
+	desc.info.guest_format = transfer_format;
+	desc.info.type         = volume ? Prospero::ImageType::kColor3D : Prospero::ImageType::kColor2D;
+	desc.info.extent       = {width, height, depth};
+	desc.info.resources    = {levels, volume ? 1u : view.image_layers};
+	desc.info.pitch        = pitch;
 	desc.info.bytes_per_block = bytes_per_element;
 	desc.info.samples         = samples;
 	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] = {
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
+			};
+			continue;
+		}
 		const auto level_offset =
 		    mip_sizes[level].src_size != 0 ? mip_sizes[level].src_offset : mip_sizes[level].offset;
 		const auto level_size =

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>
@@ -359,30 +360,12 @@ static void RtCheck(const HW::RenderTarget& rt) {
 				logged = true;
 			}
 		}
-		if (rt.attrib3.depth != 0x00000000) {
-			static bool logged = false;
-			if (!logged) {
-				LOGF("RenderTarget: temporary: ignoring PS5 color target depth_minus1=0x%08" PRIx32
-				     "\n",
-				     rt.attrib3.depth);
-				logged = true;
-			}
-		}
 		if (!RenderIsColorTileMode(rt.attrib3.tile_mode)) {
 			EXIT("unknown PS5 render-target tile mode: 0x%08" PRIx32 "\n", rt.attrib3.tile_mode);
 		}
 		if (!RenderIsColorDimension(rt.attrib3.dimension)) {
 			EXIT("unknown PS5 render-target dimension: 0x%08" PRIx32 "\n", rt.attrib3.dimension);
 		}
-		if (rt.attrib3.dimension != 0x00000001) {
-			static bool logged = false;
-			if (!logged) {
-				LOGF("RenderTarget: temporary: using 2D fallback for PS5 color "
-				     "dimension=0x%08" PRIx32 "\n",
-				     rt.attrib3.dimension);
-				logged = true;
-			}
-		}
 		if (!rt.attrib3.cmask_pipe_aligned) {
 			static bool logged = false;
 			if (!logged) {
@@ -704,14 +687,6 @@ static void McCheck(const HW::ModeControl& c) {
 	}
 	EXIT_NOT_IMPLEMENTED(c.polymode_front_ptype != 0 && c.polymode_front_ptype != 2);
 	EXIT_NOT_IMPLEMENTED(c.polymode_back_ptype != 0 && c.polymode_back_ptype != 2);
-	if (c.poly_offset_front_enable || c.poly_offset_back_enable) {
-		static bool logged = false;
-		if (!logged) {
-			LOGF("\t temporary: PA_SU_SC_MODE_CNTL.POLY_OFFSET_*_ENABLE is not implemented; "
-			     "continuing without depth bias\n");
-			logged = true;
-		}
-	}
 	if (c.vtx_window_offset_enable) {
 		static bool logged = false;
 		if (!logged) {
@@ -1085,18 +1060,24 @@ static ScissorRect ScissorRectClamp(ScissorRect r, uint32_t width, uint32_t heig
 	return r;
 }
 
+static constexpr std::array<uint16_t, 16> MakeScissorIntersectionRules() {
+	std::array<uint16_t, 16> rules {};
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		for (uint32_t combination = 0; combination < 16; combination++) {
+			if ((combination & candidate) == candidate) {
+				rules[candidate] |= static_cast<uint16_t>(1u << combination);
+			}
+		}
+	}
+	return rules;
+}
+
 static bool ScissorClipRuleToIntersectionMask(uint16_t rule, uint8_t* mask) {
 	EXIT_IF(mask == nullptr);
 
-	for (uint32_t candidate = 0; candidate < 16; candidate++) {
-		uint16_t expected = 0;
-		for (uint32_t combination = 0; combination < 16; combination++) {
-			if ((combination & candidate) == candidate) {
-				expected |= static_cast<uint16_t>(1u << combination);
-			}
-		}
-
-		if (expected == rule) {
+	static constexpr auto rules = MakeScissorIntersectionRules();
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		if (rules[candidate] == rule) {
 			*mask = static_cast<uint8_t>(candidate);
 			return true;
 		}

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -38,7 +38,10 @@ namespace Libs::Graphics {
 	EXIT("unsupported render state; details were printed above\n");
 }
 
-static vk::StencilOp ConvertStencilOp(uint8_t value) {
+static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t op_value) {
+	if (write_mask == 0) {
+		return vk::StencilOp::eKeep;
+	}
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
@@ -49,7 +52,17 @@ static vk::StencilOp ConvertStencilOp(uint8_t value) {
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
 		case Prospero::StencilOp::kAddWrap: return vk::StencilOp::eIncrementAndWrap;
 		case Prospero::StencilOp::kSubWrap: return vk::StencilOp::eDecrementAndWrap;
-		default: DepthFatal("unsupported stencil operation");
+		case Prospero::StencilOp::kXor:
+			if ((write_mask & op_value) == 0) {
+				return vk::StencilOp::eKeep;
+			}
+			if ((write_mask & ~op_value) != 0) {
+				DepthFatal("unsupported stencil XOR operands: write mask=0x%02" PRIx8
+				           ", operation value=0x%02" PRIx8,
+				           write_mask, op_value);
+			}
+			return vk::StencilOp::eInvert;
+		default: DepthFatal("unsupported stencil operation: 0x%02" PRIx8, value);
 	}
 }
 
@@ -96,9 +109,11 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	const auto& dc = hw.GetDepthControl();
 	const auto& sc = hw.GetStencilControl();
 	const auto& sm = hw.GetStencilMask();
-	const bool  depth_active =
+	const bool  has_stencil =
+	    z.stencil_info.format != Prospero::GpuEnumValue(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 = dc.stencil_enable || rc.stencil_clear_enable;
+	const bool stencil_active = has_stencil && (dc.stencil_enable || rc.stencil_clear_enable);
 	if (!depth_active && !stencil_active) {
 		return;
 	}
@@ -139,8 +154,6 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		}
 		return;
 	}
-	const bool has_stencil =
-	    z.stencil_info.format != Prospero::GpuEnumValue(Prospero::StencilFormat::kInvalid);
 	const bool has_htile = z.z_info.htile_acceleration;
 	const auto samples   = render_sample_count(z.z_info.num_samples);
 	if (samples == 0) {
@@ -156,8 +169,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 			DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
 			           z.depth_view.slice_max);
 	}
-	if ((stencil_active && !has_stencil) || rc.resummarize_enable || rc.copy_centroid ||
-	    rc.copy_sample != 0 || z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
+	if (rc.resummarize_enable || rc.copy_centroid || rc.copy_sample != 0 ||
+	    z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
 	    z.z_info.partially_resident || z.stencil_info.partially_resident ||
 	    z.z_info.max_mip_level != 0 || z.depth_view.current_mip_level != 0 ||
 	    z.depth_info.addr5_swizzle_mask != 0 || z.depth_info.array_mode != 0 ||
@@ -279,10 +292,12 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	r.depth_min_bounds         = hw.GetDepthBoundsMin();
 	r.depth_max_bounds         = hw.GetDepthBoundsMax();
 
-	r.stencil_clear_enable = rc.stencil_clear_enable;
+	r.stencil_clear_enable = has_stencil && rc.stencil_clear_enable;
 	r.stencil_clear_value  = hw.GetStencilClearValue();
-	r.stencil_test_enable  = dc.stencil_enable;
-	if (dc.stencil_enable) {
+	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 +309,18 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 			DepthFatal("unsupported stencil compare or replacement state");
 		}
 		r.stencil_static_front = {
-		    ConvertStencilOp(sc.stencil_fail), ConvertStencilOp(sc.stencil_zpass),
-		    ConvertStencilOp(sc.stencil_zfail), static_cast<vk::CompareOp>(dc.stencilfunc)};
-		r.stencil_dynamic_front = {sm.stencil_mask,
-		                           rc.stencil_clear_enable ? 0u : sm.stencil_writemask,
-		                           sm.stencil_testval};
+		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
+		    static_cast<vk::CompareOp>(dc.stencilfunc)};
+		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
 		if (dc.backface_enable) {
-			r.stencil_static_back  = {ConvertStencilOp(sc.stencil_fail_bf),
-			                          ConvertStencilOp(sc.stencil_zpass_bf),
-			                          ConvertStencilOp(sc.stencil_zfail_bf),
-			                          static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
-			r.stencil_dynamic_back = {sm.stencil_mask_bf,
-			                          rc.stencil_clear_enable ? 0u : sm.stencil_writemask_bf,
-			                          sm.stencil_testval_bf};
+			r.stencil_static_back = {
+			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
+			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
+			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
 		} else {
 			r.stencil_static_back  = r.stencil_static_front;
 			r.stencil_dynamic_back = r.stencil_dynamic_front;

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

@@ -105,6 +105,10 @@ Image::Barriers Image::GetBarriers(vk::ImageLayout                      destinat
                                    std::optional<ImageSubresourceRange> range) {
 	auto& state              = backing.state;
 	auto& subresource_states = backing.subresource_states;
+	if (range && info.IsVolume()) {
+		range->base_layer  = 0;
+		range->layer_count = 1;
+	}
 
 	const bool partial =
 	    range && (range->base_level != 0 || range->level_count != info.resources.levels ||
@@ -659,7 +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

@@ -154,17 +154,18 @@ public:
 	}
 	[[nodiscard]] uint64_t HashGuestEdges() const;
 
-	ImageInfo      info;
-	VulkanImage    backing;
-	ImageViewCache views;
-	ImageUsage     usage;
-	ImageBinding   binding;
-	bool           registered     = false;
-	uint64_t       track_addr     = 0;
-	uint64_t       track_addr_end = 0;
-	ImageId        depth_id {};
-	uint64_t       tick_accessed_last = 0;
-	size_t         lru_id             = 0;
+	ImageInfo        info;
+	VulkanImage      backing;
+	ImageViewCache   views;
+	ImageUsage       usage;
+	ImageBinding     binding;
+	bool             registered     = false;
+	mutable uint32_t query_epoch    = 0;
+	uint64_t         track_addr     = 0;
+	uint64_t         track_addr_end = 0;
+	ImageId          depth_id {};
+	uint64_t         tick_accessed_last = 0;
+	size_t           lru_id             = 0;
 
 private:
 	friend struct ImageTestAccess;

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

@@ -22,6 +22,10 @@ struct RenderTargetFormatMapping {
 };
 
 constexpr RenderTargetFormatMapping kRenderTargetFormats[] = {
+    {Prospero::ChannelLayout::k8,
+     Prospero::ChannelType::kUInt,
+     Prospero::ChannelOrder::kStandard,
+     {vk::Format::eR8Uint, 1}},
     {Prospero::ChannelLayout::k8_8,
      Prospero::ChannelType::kUNorm,
      Prospero::ChannelOrder::kStandard,
@@ -164,14 +168,6 @@ RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_
 
 namespace {
 
-static uint64_t GetLevelSrcOffset(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_offset : level_size.offset);
-}
-
-static uint64_t GetLevelSrcSize(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_size : level_size.size);
-}
-
 static uint32_t GetTextureLevelDepth(uint32_t depth, uint32_t level, bool volume_texture) {
 	return volume_texture ? std::max(depth >> level, 1u) : depth;
 }
@@ -184,20 +180,6 @@ static size_t GetTextureRegionCount(uint32_t depth, uint64_t levels, bool volume
 	return count;
 }
 
-uint64_t TextureUploadSliceSourceOffset(const TextureUploadLayout& layout, uint32_t level,
-                                        uint32_t slice) {
-	if (level >= 16 || layout.level_sizes[level].size == 0) {
-		EXIT("invalid texture upload slice source, level=%u slice=%u\n", level, slice);
-	}
-	const auto level_offset = GetLevelSrcOffset(layout.level_sizes[level]);
-	const auto slice_stride =
-	    layout.source_slice_stride != 0 ? layout.source_slice_stride : layout.slice_stride;
-	if (slice_stride != 0 && slice > (UINT64_MAX - level_offset) / slice_stride) {
-		EXIT("texture upload slice source offset overflow, level=%u slice=%u\n", level, slice);
-	}
-	return level_offset + static_cast<uint64_t>(slice) * slice_stride;
-}
-
 vk::ComponentSwizzle TextureGetComponentSwizzle(uint8_t s) {
 	switch (static_cast<Prospero::CompSwizzle>(s)) {
 		case Prospero::CompSwizzle::kZero: return vk::ComponentSwizzle::eZero;
@@ -237,12 +219,11 @@ vk::Format TextureGetFormat(uint32_t fmt) {
 
 namespace {
 
-uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t levels,
+uint64_t CalcTextureSliceStride(const TextureUploadMipLayout* mips, uint64_t levels,
                                 uint64_t total_size, uint32_t depth) {
 	uint64_t stride = 0;
 	for (uint32_t i = 0; i < levels; i++) {
-		stride =
-		    std::max(stride, static_cast<uint64_t>(level_sizes[i].offset) + level_sizes[i].size);
+		stride = std::max(stride, mips[i].offset + mips[i].size);
 	}
 
 	if (depth > 1 && total_size != 0 && total_size % depth == 0) {
@@ -255,39 +236,26 @@ uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t leve
 	return stride;
 }
 
-uint64_t CalcLinearUploadLevelSize(uint32_t fmt, uint32_t pitch, uint32_t height) {
-	if (const uint32_t bytes_per_element = Prospero::NumBytesPerElement(fmt);
-	    bytes_per_element != 0) {
-		return static_cast<uint64_t>(pitch) * height * bytes_per_element;
-	}
-
-	if (const uint32_t bytes_per_block = Prospero::BlockCompressedBytesPerBlock(fmt);
-	    bytes_per_block != 0) {
-		const uint32_t blocks_w = std::max((pitch + 3u) / 4u, 1u);
-		const uint32_t blocks_h = std::max((height + 3u) / 4u, 1u);
-		return static_cast<uint64_t>(blocks_w) * blocks_h * bytes_per_block;
-	}
-
-	return 0;
+uint64_t CalcLinearUploadLevelSize(const TileTextureElementLayout& element, uint32_t pitch,
+                                   uint32_t height) {
+	const uint32_t elements_w =
+	    std::max((pitch + element.texel_width - 1u) / element.texel_width, 1u);
+	const uint32_t elements_h =
+	    std::max((height + element.texel_height - 1u) / element.texel_height, 1u);
+	return static_cast<uint64_t>(elements_w) * elements_h * element.bytes;
 }
 
-uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64_t height,
+uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
+                               const TileTextureElementLayout& element, uint64_t height,
                                uint64_t levels, uint32_t base_pitch) {
 	uint64_t offset = 0;
 	auto     pitch  = base_pitch;
 	auto     h      = static_cast<uint32_t>(height);
 
 	for (uint32_t i = 0; i < levels; i++) {
-		const auto size = CalcLinearUploadLevelSize(fmt, pitch, h);
-		EXIT_NOT_IMPLEMENTED(size > 0xffffffffull);
-		EXIT_NOT_IMPLEMENTED(offset > 0xffffffffull);
-
-		level_sizes[i].size       = static_cast<uint32_t>(size);
-		level_sizes[i].offset     = static_cast<uint32_t>(offset);
-		level_sizes[i].src_size   = 0;
-		level_sizes[i].src_offset = 0;
-		level_sizes[i].x          = 0;
-		level_sizes[i].y          = 0;
+		const auto size = CalcLinearUploadLevelSize(element, pitch, h);
+		mips[i].size    = size;
+		mips[i].offset  = offset;
 
 		offset += size;
 		if (pitch > 1) {
@@ -303,141 +271,123 @@ uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64
 
 } // namespace
 
-TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64_t height,
-                                            uint64_t levels, uint32_t depth, uint64_t pitch,
-                                            uint64_t tile, uint64_t upload_size,
-                                            bool allow_depth_tile, bool volume_texture,
-                                            const char* owner) {
+TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
+                                            uint32_t levels, uint32_t depth, uint32_t tile,
+                                            uint64_t upload_size, bool allow_depth_tile,
+                                            bool volume_texture, const char* owner) {
 	TextureUploadLayout layout {};
-	layout.tile           = static_cast<uint32_t>(tile);
-	layout.pitch          = static_cast<uint32_t>(pitch);
-	layout.texel_block    = Prospero::BlockCompressedBytesPerBlock(fmt) != 0 ? 4u : 1u;
-	layout.volume_texture = volume_texture;
+	layout.surface.description = {
+	    fmt,
+	    tile,
+	    volume_texture ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
+	    width,
+	    height,
+	    volume_texture ? depth : 1u,
+	    levels,
+	    volume_texture ? 1u : depth,
+	};
+	const auto&              description = layout.surface.description;
+	const auto               tile_mode   = static_cast<Prospero::TileMode>(description.tile_mode);
+	TileTextureElementLayout element {};
 
-	if (fmt != 0) {
-		if (layout.tile != 0) {
-			const auto tile_mode = static_cast<Prospero::TileMode>(layout.tile);
-			switch (tile_mode) {
-				case Prospero::TileMode::kStandard256B:
-					if (TileIsStandard256BTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard256B;
-					}
-					break;
-				case Prospero::TileMode::kStandard4KB:
-					if (TileIsStandard4KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard4KB;
-					}
-					break;
-				case Prospero::TileMode::kStandard64KB:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard64KB;
-					}
-					break;
-				case Prospero::TileMode::kPrt:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Prt64KB;
-					}
-					break;
-				case Prospero::TileMode::kRenderTarget:
-					if (Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::RenderTarget64KB;
-					}
-					break;
-				case Prospero::TileMode::kDepth:
-					if (allow_depth_tile && Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::Depth64KB;
-					}
-					break;
-				default: break;
-			}
-			if (layout.tile_family == TileBlockFamily::Count) {
-				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
-				     "size=%" PRIu64 " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64
-				     " levels=%" PRIu64 "\n",
-				     owner, static_cast<uint32_t>(fmt), layout.tile, upload_size, width, height,
-				     pitch, levels);
-			}
-		}
-
-		layout.pitch = TileGetTexturePitch(fmt, width, levels, layout.tile);
-
-		TileGetTextureSize(fmt, width, height, layout.pitch, levels, layout.tile, nullptr,
-		                   layout.level_sizes, layout.padded_sizes);
-
-		if (static_cast<Prospero::TileMode>(layout.tile) != Prospero::TileMode::kLinear) {
-			if (layout.volume_texture) {
-				layout.slice_stride = SetLinearUploadLevels(layout.level_sizes, fmt, height, levels,
-				                                            static_cast<uint32_t>(width));
-			} else {
-				TileSizeOffset tiled_levels[16] {};
-				std::copy_n(layout.level_sizes, levels, tiled_levels);
-				layout.source_slice_stride =
-				    CalcTextureSliceStride(tiled_levels, levels, upload_size, depth);
-				SetLinearUploadLevels(layout.level_sizes, fmt, height, levels, layout.pitch);
-				for (uint32_t i = 0; i < levels; ++i) {
-					if (tiled_levels[i].src_size > tiled_levels[i].size) {
-						layout.first_tail_level = std::min(layout.first_tail_level, i);
-					}
-					layout.level_sizes[i].src_offset = GetLevelSrcOffset(tiled_levels[i]);
-					layout.level_sizes[i].src_size   = GetLevelSrcSize(tiled_levels[i]);
-					layout.level_sizes[i].x          = tiled_levels[i].x;
-					layout.level_sizes[i].y          = tiled_levels[i].y;
-				}
-			}
-		} else if (layout.volume_texture) {
-			layout.slice_stride =
-			    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
-		}
-	} else {
+	if (fmt == 0) {
 		EXIT("%s: legacy texture upload format unsupported: fmt=0 tile=%u size=%" PRIu64
-		     " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64 " levels=%" PRIu64 "\n",
-		     owner, layout.tile, upload_size, width, height, pitch, levels);
+		     " extent=%ux%u levels=%u\n",
+		     owner, description.tile_mode, upload_size, width, height, levels);
 	}
 
-	if (!layout.volume_texture) {
-		layout.slice_stride =
-		    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
+	switch (tile_mode) {
+		case Prospero::TileMode::kLinear:
+			if (!TileGetTextureElementLayout(fmt, element)) {
+				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner, fmt);
+			}
+			break;
+		case Prospero::TileMode::kDepth:
+			if (!allow_depth_tile || !TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+		default:
+			if (!TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+	}
+
+	if (tile_mode != Prospero::TileMode::kLinear) {
+		element = {layout.surface.texture.block.bytes_per_element,
+		           layout.surface.texture.texel_width, layout.surface.texture.texel_height};
+	}
+	layout.pitch = TileGetTexturePitch(fmt, width, description.tile_mode);
+	if (tile_mode == Prospero::TileMode::kLinear) {
+		TileSizeOffset level_sizes[16] {};
+		TilePaddedSize padded_sizes[16] {};
+		TileGetTextureSize(fmt, width, height, levels, description.tile_mode, nullptr, level_sizes,
+		                   padded_sizes);
+		for (uint32_t level = 0; level < levels; ++level) {
+			layout.mips[level] = {level_sizes[level].offset, level_sizes[level].size,
+			                      padded_sizes[level].width, padded_sizes[level].height};
+		}
+	} else if (volume_texture) {
+		layout.slice_stride = SetLinearUploadLevels(layout.mips, element, height, levels, width);
+	} else {
+		layout.source_slice_stride = layout.surface.block_slice_size;
+		if (depth > 1 && upload_size != 0 && upload_size % depth == 0) {
+			layout.source_slice_stride = std::max(layout.source_slice_stride, upload_size / depth);
+		}
+		SetLinearUploadLevels(layout.mips, element, height, levels, layout.pitch);
+	}
+
+	if (!volume_texture || tile_mode == Prospero::TileMode::kLinear) {
+		layout.slice_stride = CalcTextureSliceStride(layout.mips, levels, upload_size, depth);
 	}
 	return layout;
 }
 
-std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
-                                                         uint32_t width, uint32_t height,
-                                                         uint32_t depth, uint64_t levels,
-                                                         bool array_texture, bool volume_texture) {
-	uint32_t mip_width  = width;
-	uint32_t mip_height = height;
-	uint32_t mip_pitch  = volume_texture && static_cast<Prospero::TileMode>(layout.tile) !=
-	                                            Prospero::TileMode::kLinear
-	                          ? width
-	                          : layout.pitch;
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout) {
+	const auto& description    = layout.surface.description;
+	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const auto  depth          = volume_texture ? description.depth : description.layers;
+	uint32_t    mip_width      = description.width;
+	uint32_t    mip_height     = description.height;
+	uint32_t mip_pitch = volume_texture && static_cast<Prospero::TileMode>(description.tile_mode) !=
+	                                           Prospero::TileMode::kLinear
+	                         ? description.width
+	                         : layout.pitch;
 
 	std::vector<vk::BufferImageCopy> regions;
-	regions.reserve(GetTextureRegionCount(depth, levels, volume_texture));
-	for (uint32_t i = 0; i < levels; i++) {
-		EXIT_NOT_IMPLEMENTED(layout.level_sizes[i].size == 0);
+	regions.reserve(GetTextureRegionCount(depth, description.levels, volume_texture));
+	for (uint32_t i = 0; i < description.levels; i++) {
+		EXIT_NOT_IMPLEMENTED(layout.mips[i].size == 0);
 
 		const auto mip_depth = GetTextureLevelDepth(depth, i, volume_texture);
 
 		for (uint32_t z = 0; z < mip_depth; z++) {
 			const auto          slice_offset = z * layout.slice_stride;
 			vk::BufferImageCopy region {};
-			region.bufferOffset     = layout.level_sizes[i].offset + slice_offset;
-			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, array_texture ? z : 0,
+			region.bufferOffset     = layout.mips[i].offset + slice_offset;
+			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, volume_texture ? 0u : z,
 			                           1};
 			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
 			region.imageExtent      = {mip_width, mip_height, 1};
-			const bool linear =
-			    static_cast<Prospero::TileMode>(layout.tile) == Prospero::TileMode::kLinear;
+			const bool linear       = static_cast<Prospero::TileMode>(description.tile_mode) ==
+			                          Prospero::TileMode::kLinear;
 			if (linear) {
-				region.bufferRowLength   = layout.padded_sizes[i].width;
-				region.bufferImageHeight = layout.padded_sizes[i].height;
+				region.bufferRowLength   = layout.mips[i].row_length;
+				region.bufferImageHeight = layout.mips[i].image_height;
 			} else {
-				const auto align = [](uint32_t value, uint32_t block) {
+				const auto texel_width = layout.surface.texture.texel_width;
+				const auto align       = [](uint32_t value, uint32_t block) {
 					return ((value + block - 1u) / block) * block;
 				};
-				const auto pitch       = align(mip_pitch, layout.texel_block);
-				region.bufferRowLength = pitch > align(mip_width, layout.texel_block) ? pitch : 0;
+				const auto pitch       = align(mip_pitch, texel_width);
+				region.bufferRowLength = pitch > align(mip_width, texel_width) ? pitch : 0;
 			}
 			regions.push_back(region);
 		}
@@ -456,24 +406,6 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 	return regions;
 }
 
-struct GpuTileElementLayout {
-	uint32_t bytes = 0;
-	uint32_t wide  = 1;
-	uint32_t tall  = 1;
-};
-
-static bool GetGpuTileElementLayout(uint32_t fmt, GpuTileElementLayout& out) {
-	if (const auto bytes = Prospero::NumBytesPerElement(fmt); bytes != 0) {
-		out = {bytes, 1, 1};
-		return true;
-	}
-	if (const auto bytes = Prospero::BlockCompressedBytesPerBlock(fmt); bytes != 0) {
-		out = {bytes, 4, 4};
-		return true;
-	}
-	return false;
-}
-
 static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity, uint64_t& size) {
 	if (offset > capacity || length > capacity - offset) {
 		return false;
@@ -482,43 +414,33 @@ static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity,
 	return true;
 }
 
-bool TextureBuildGpuTileInfos(uint64_t size, const std::vector<vk::BufferImageCopy>& regions,
-                              const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
-                              uint64_t levels, std::vector<GpuTileInfo>& out_infos) {
-	if (size == 0 || levels == 0 || levels > 16 || depth == 0 ||
-	    regions.size() != GetTextureRegionCount(depth, levels, layout.volume_texture) ||
-	    Prospero::IsFmaskTextureFormat(fmt)) {
+bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
+                              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, 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);
@@ -533,49 +455,43 @@ bool TextureBuildGpuTileInfos(uint64_t size, const std::vector<vk::BufferImageCo
 				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;
-				if (!SetGpuTileSize(info.linear_offset, linear_span, size, info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, volume.level_sizes[level], size,
+				    layout.mips[level].size;
+				if (!SetGpuTileSize(info.linear_offset, linear_span, UINT64_MAX,
+				                    info.linear_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 =
@@ -584,38 +500,36 @@ bool TextureBuildGpuTileInfos(uint64_t size, const std::vector<vk::BufferImageCo
 				                                 ? 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, size, info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, GetLevelSrcSize(level_size), size,
-				                    info.tiled_size)) {
+				info.family              = block.family;
+				info.bytes_per_element   = block.bytes_per_element;
+				info.linear_offset       = region.bufferOffset;
+				const auto source_stride = layout.source_slice_stride != 0
+				                               ? layout.source_slice_stride
+				                               : surface.block_slice_size;
+				if (z > (UINT64_MAX - mip.offset) / source_stride) {
 					return false;
 				}
-				info.width =
-				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height    = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
-				info.surface_z = base_family == TileBlockFamily::RenderTarget64KB ||
-				                         base_family == TileBlockFamily::Depth64KB
+				info.tiled_offset = mip.offset + static_cast<uint64_t>(z) * source_stride;
+				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
+				                    info.linear_size) ||
+				    !SetGpuTileSize(info.tiled_offset, mip.size, tiled_size, info.tiled_size)) {
+					return false;
+				}
+				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
+				                           texture.texel_width,
+				                       1u);
+				info.height = std::max(
+				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+				info.surface_z = block.family == TileBlockFamily::RenderTarget64KB ||
+				                         block.family == TileBlockFamily::Depth64KB
 				                     ? region.imageSubresource.baseArrayLayer
 				                     : 0;
-				info.pitch     = std::max((pitch + element.wide - 1u) / element.wide, 1u);
-				info.tail      = tail;
-				info.tail_x    = tail ? level_size.x : 0;
-				info.tail_y    = tail ? level_size.y : 0;
-				info.tiled_width =
-				    layout.padded_sizes[level].width != 0
-				        ? std::max((layout.padded_sizes[level].width + element.wide - 1u) /
-				                       element.wide,
-				                   1u)
-				        : info.pitch;
-				info.tiled_height =
-				    layout.padded_sizes[level].height != 0
-				        ? std::max((layout.padded_sizes[level].height + element.tall - 1u) /
-				                       element.tall,
-				                   1u)
-				        : info.height;
+				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
+				info.tail  = tail;
+				info.tail_x       = tail ? mip.tail_x : 0;
+				info.tail_y       = tail ? mip.tail_y : 0;
+				info.tiled_width  = mip.padded_width;
+				info.tiled_height = mip.padded_height;
 				infos.push_back(info);
 			}
 		}

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

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

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

@@ -239,7 +239,7 @@ bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, c
 	const auto samples      = multisampled ? 1u << descriptor.LastLevel() : 1u;
 	const auto pitch =
 	    multisampled ? TileGetDepthPitch(width, image.info.bytes_per_block, descriptor.LastLevel())
-	                 : TileGetTexturePitch(descriptor.Format(), width, 1, descriptor.TileMode());
+	                 : TileGetTexturePitch(descriptor.Format(), width, descriptor.TileMode());
 	const bool supported_2d    = type == Prospero::ImageType::kColor2D &&
 	                             image.info.resources.layers == 1 && descriptor.Depth() == 0 &&
 	                             descriptor.BaseArray5() == 0;
@@ -321,7 +321,7 @@ static void ValidateDepthTargetBinding(const ShaderRecompiler::IR::ImageResource
 		return;
 	}
 	const auto descriptor_pitch =
-	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u, 1,
+	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u,
 	                        descriptor.TileMode());
 	EXIT("unsupported sampled depth target: resource=%d descriptor=%d encoding=%d format=%d "
 	     "kind=%u dimension=%u mip_mode=%u read=%d written=%d atomic=%d compare=%d "
@@ -373,27 +373,38 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool is_3d = resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D &&
 	                   descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor3D) &&
 	                   descriptor.BaseArray5() == 0;
-	TileBlockLayout depth_block {};
-	const auto      depth_bpe = Prospero::RenderTargetBytesPerElement(descriptor.Format());
-	const bool      supported_depth_tile =
-	    tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) && !resource.read &&
-	    !Prospero::IsFmaskTextureFormat(descriptor.Format()) && (is_2d || is_2d_array) &&
-	    TileGetBlockLayout(TileBlockFamily::Depth64KB, depth_bpe, depth_block);
-	const bool supported_standard_tile =
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB) &&
-	     TileIsStandard4KBTextureSupported(descriptor.Format())) ||
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB) &&
-	     TileIsStandard64KBTextureSupported(descriptor.Format()));
-	const bool supported_tile = tile == Prospero::GpuEnumValue(Prospero::TileMode::kLinear) ||
-	                            tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
-	                            supported_depth_tile || supported_standard_tile;
-	const auto swizzle        = descriptor.DstSelXYZW();
+	TileTextureBlockLayout tile_layout {};
+	bool                   supported_tile = false;
+	switch (static_cast<Prospero::TileMode>(tile)) {
+		case Prospero::TileMode::kLinear: supported_tile = true; break;
+		case Prospero::TileMode::kDepth:
+			supported_tile =
+			    !resource.read && !Prospero::IsFmaskTextureFormat(descriptor.Format()) &&
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard256B:
+			supported_tile =
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, is_3d, tile_layout);
+			break;
+		case Prospero::TileMode::kRenderTarget:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		default: break;
+	}
+	const auto swizzle = descriptor.DstSelXYZW();
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
-	const bool supported_mip_view = descriptor.BaseLevel() == 0 || is_1d || is_2d;
 	return (is_1d || is_1d_array || is_2d || is_2d_array || is_3d) && supported_tile &&
-	       supported_mip_view && descriptor.BaseLevel() == descriptor.LastLevel() &&
+	       descriptor.BaseLevel() == descriptor.LastLevel() &&
 	       descriptor.LastLevel() <= descriptor.MaxMip() && descriptor.MinLod() == 0 &&
 	       supported_swizzle && descriptor.BCSwizzle() == 0 && !descriptor.MsaaDepth();
 }
@@ -494,18 +505,20 @@ static TextureCache::ImageDesc NullTextureDesc(const ShaderRecompiler::IR::Image
 
 static void PopulateTextureMipLayout(ImageInfo& info) {
 	if (info.IsVolume() && info.tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kLinear)) {
-		TileVolumeLayout volume {};
-		if (!TileGetTextureVolumeLayout(info.guest_format, info.extent.width, info.extent.height,
-		                                info.extent.depth, info.resources.levels, info.tile_mode,
-		                                volume)) {
+		TileSurfaceLayout            surface {};
+		const TileSurfaceDescription description {
+		    info.guest_format,  info.tile_mode,    TileSurfaceDimension::Dim3D, info.extent.width,
+		    info.extent.height, info.extent.depth, info.resources.levels,       1};
+		if (!TileGetTiledTextureLayout(description, surface)) {
 			EXIT("unsupported normalized volume texture layout\n");
 		}
 		for (uint32_t level = 0; level < info.resources.levels; level++) {
+			const auto& mip        = surface.mips[level];
 			info.mip_layout[level] = {
-			    volume.level_offsets[level],
-			    volume.level_sizes[level],
-			    volume.level_widths[level],
-			    volume.level_heights[level],
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
 			};
 		}
 		return;
@@ -513,7 +526,7 @@ static void PopulateTextureMipLayout(ImageInfo& info) {
 
 	TileSizeOffset levels[16] {};
 	TilePaddedSize padded[16] {};
-	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height, info.pitch,
+	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height,
 	                   info.resources.levels, info.tile_mode, nullptr, levels, padded);
 	for (uint32_t level = 0; level < info.resources.levels; level++) {
 		const auto offset =
@@ -618,14 +631,15 @@ 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 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;
 	if ((!multisampled && (base_level > last_level || last_level >= levels)) ||
 	    (multisampled &&
 	     (base_level != 0 || last_level == 0 || last_level > 3 ||
-	      descriptor.MaxMip() != last_level || !msaa_tile || (descriptor.MsaaDepth() && !depth_tile) ||
+	      descriptor.MaxMip() != last_level || !msaa_tile ||
+	      (descriptor.MsaaDepth() && !depth_tile) ||
 	      (!msaa_array && (descriptor.Depth() != 0 || descriptor.BaseArray5() != 0))))) {
 		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 "
@@ -634,7 +648,8 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		     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],
-		     descriptor.fields[4], descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
+		     descriptor.fields[4], descriptor.fields[5], descriptor.fields[6],
+		     descriptor.fields[7]);
 	}
 	const auto samples = multisampled ? 1u << last_level : 1u;
 	const auto view_levels =
@@ -661,17 +676,17 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	TileSizeAlign size {};
 	if (multisampled) {
 		const auto bytes = Prospero::NumBytesPerElement(format);
-		pitch = depth_tile ? TileGetDepthPitch(width, bytes, last_level)
-		                   : TileGetRenderTargetPitch(width, bytes, last_level);
+		pitch            = depth_tile ? TileGetDepthPitch(width, bytes, last_level)
+		                              : TileGetRenderTargetPitch(width, bytes, last_level);
 		if (pitch == 0 || !TileGetRenderTargetSize(width, height, pitch, bytes, size, last_level) ||
 		    size.size > UINT32_MAX / image_layers) {
 			EXIT("unsupported multisample texture layout\n");
 		}
 		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);
@@ -679,8 +694,8 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		ValidateStorageTexture(resource, descriptor, size.size);
 	}
 
-	const auto              pixel_format = TextureGetFormat(format);
-	const auto              storage_view_format =
+	const auto pixel_format = TextureGetFormat(format);
+	const auto storage_view_format =
 	    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt)
 	        ? vk::Format::eR32Uint
 	        : SrgbStorageViewFormat(pixel_format);
@@ -918,22 +933,21 @@ RenderExecutor::PrepareGraphicsBindings(CommandBuffer& buffer, const ShaderStage
 	return bindings;
 }
 
-static void RetainBindings(CommandBuffer&                            buffer,
-                           const DescriptorCache::NativeDescriptors& resources) {
+static void RetainBindings(CommandBuffer& buffer, DescriptorCache::NativeDescriptors& resources) {
 	if (resources.flattened_srt.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.flattened_srt.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.flattened_srt.owner));
 	}
 	if (resources.user_data.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.user_data.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.user_data.owner));
 	}
-	for (const auto& view: resources.buffers) {
+	for (auto& view: resources.buffers) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
-	for (const auto& view: resources.addresses) {
+	for (auto& view: resources.addresses) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
 }

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

@@ -154,8 +154,9 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
 		static_params.color_mask[i] = color_mask[i];
 	}
-	static_params.cull_back  = mc.cull_back;
-	static_params.cull_front = mc.cull_front;
+	const bool rect_list     = topology == vk::PrimitiveTopology::ePatchList;
+	static_params.cull_back  = !rect_list && mc.cull_back;
+	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;
 
 	for (uint32_t i = 0; i < color_count; i++) {

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

@@ -14,6 +14,7 @@
 #include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/rectListShader.h"
 #include "graphics/shader/shader.h"
 
 #include <algorithm>
@@ -463,8 +464,12 @@ void CreatePipelineInternal(
     uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active) {
 	EXIT_IF(ps_active && ps_input_info == nullptr);
 
-	vk::ShaderModule vert_shader_module = nullptr;
-	vk::ShaderModule frag_shader_module = nullptr;
+	const bool rect_list = static_params.topology == vk::PrimitiveTopology::ePatchList;
+
+	vk::ShaderModule vert_shader_module         = nullptr;
+	vk::ShaderModule tess_control_shader_module = nullptr;
+	vk::ShaderModule tess_eval_shader_module    = nullptr;
+	vk::ShaderModule frag_shader_module         = nullptr;
 
 	vk::ShaderModuleCreateInfo create_info {};
 
@@ -491,8 +496,33 @@ void CreatePipelineInternal(
 		}
 		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
+	if (rect_list) {
+		const auto shaders =
+		    BuildRectListShaders(vs_input_info, ps_active ? ps_input_info : nullptr);
+		create_info.codeSize = shaders.control.size() * 4;
+		create_info.pCode    = shaders.control.data();
+		result =
+		    graphics.device.createShaderModule(&create_info, nullptr, &tess_control_shader_module);
+		if (graphics_debug_dump_enabled()) {
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
+			     VulkanToString(result).c_str(), static_cast<void*>(tess_control_shader_module));
+		}
+		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
+
+		create_info.codeSize = shaders.evaluation.size() * 4;
+		create_info.pCode    = shaders.evaluation.data();
+		result =
+		    graphics.device.createShaderModule(&create_info, nullptr, &tess_eval_shader_module);
+		if (graphics_debug_dump_enabled()) {
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
+			     VulkanToString(result).c_str(), static_cast<void*>(tess_eval_shader_module));
+		}
+		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
+	}
 
 	EXIT_NOT_IMPLEMENTED(vert_shader_module == nullptr);
+	EXIT_NOT_IMPLEMENTED(
+	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 	EXIT_NOT_IMPLEMENTED(ps_active && frag_shader_module == nullptr);
 
 	vk::PipelineShaderStageCreateInfo                     vert_shader_stage_info {};
@@ -525,9 +555,28 @@ void CreatePipelineInternal(
 		                      frag_shader_stage_info);
 	}
 
-	vk::PipelineShaderStageCreateInfo shader_stages[]    = {vert_shader_stage_info,
-	                                                        frag_shader_stage_info};
-	const uint32_t                    shader_stage_count = ps_active ? 2u : 1u;
+	vk::PipelineShaderStageCreateInfo tess_control_shader_stage_info {};
+	tess_control_shader_stage_info.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
+	tess_control_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationControl;
+	tess_control_shader_stage_info.module = tess_control_shader_module;
+	tess_control_shader_stage_info.pName  = "main";
+
+	vk::PipelineShaderStageCreateInfo tess_eval_shader_stage_info {};
+	tess_eval_shader_stage_info.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
+	tess_eval_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationEvaluation;
+	tess_eval_shader_stage_info.module = tess_eval_shader_module;
+	tess_eval_shader_stage_info.pName  = "main";
+
+	vk::PipelineShaderStageCreateInfo shader_stages[4]   = {};
+	uint32_t                          shader_stage_count = 0;
+	shader_stages[shader_stage_count++]                  = vert_shader_stage_info;
+	if (rect_list) {
+		shader_stages[shader_stage_count++] = tess_control_shader_stage_info;
+		shader_stages[shader_stage_count++] = tess_eval_shader_stage_info;
+	}
+	if (ps_active) {
+		shader_stages[shader_stage_count++] = frag_shader_stage_info;
+	}
 
 	vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX];
 	vk::VertexInputBindingDescription   input_desc[ShaderVertexInputInfo::RES_MAX];
@@ -898,6 +947,8 @@ 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,
@@ -929,10 +980,13 @@ void CreatePipelineInternal(
 	pipeline_info.pStages                  = shader_stages;
 	pipeline_info.pVertexInputState        = &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = &input_assembly;
-	pipeline_info.pTessellationState       = nullptr;
-	pipeline_info.pViewportState           = &viewport_state;
-	pipeline_info.pRasterizationState      = &rasterizer;
-	pipeline_info.pMultisampleState        = &multisampling;
+	vk::PipelineTessellationStateCreateInfo tessellation_state {};
+	tessellation_state.sType              = vk::StructureType::ePipelineTessellationStateCreateInfo;
+	tessellation_state.patchControlPoints = 3;
+	pipeline_info.pTessellationState      = (rect_list ? &tessellation_state : nullptr);
+	pipeline_info.pViewportState          = &viewport_state;
+	pipeline_info.pRasterizationState     = &rasterizer;
+	pipeline_info.pMultisampleState       = &multisampling;
 	pipeline_info.pDepthStencilState = (static_params.with_depth ? &depth_stencil_info : nullptr);
 	pipeline_info.pColorBlendState   = &color_blending;
 	pipeline_info.pDynamicState      = &dynamic_state;
@@ -968,6 +1022,12 @@ void CreatePipelineInternal(
 	if (frag_shader_module != nullptr) {
 		graphics.device.destroyShaderModule(frag_shader_module, nullptr);
 	}
+	if (tess_control_shader_module != nullptr) {
+		graphics.device.destroyShaderModule(tess_control_shader_module, nullptr);
+	}
+	if (tess_eval_shader_module != nullptr) {
+		graphics.device.destroyShaderModule(tess_eval_shader_module, nullptr);
+	}
 	graphics.device.destroyShaderModule(vert_shader_module, nullptr);
 }
 

src/graphics/host_gpu/renderer/renderCompute.cpp

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

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -2,7 +2,6 @@
 
 #include "common/assert.h"
 #include "common/common.h"
-#include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -65,12 +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";
@@ -336,6 +355,22 @@ static PipelineDynamicParameters BuildGraphicsDynamicParams(const RenderCommandB
 	ret.stencil_front      = depth.stencil_dynamic_front;
 	ret.stencil_back       = depth.stencil_dynamic_back;
 
+	const auto& mode        = ctx.GetModeControl();
+	const auto& poly_offset = ctx.GetPolyOffset();
+	const bool  use_front   = mode.poly_offset_front_enable && !mode.cull_front;
+	const bool  use_back    = mode.poly_offset_back_enable && !mode.cull_back;
+	ret.depth_bias_enable   = use_front || use_back;
+	if (ret.depth_bias_enable) {
+		// Vulkan has one bias for both faces. Prefer a visible front face when both are enabled.
+		const float guest_constant_factor =
+		    use_front ? poly_offset.front_offset : poly_offset.back_offset;
+		ret.depth_bias_constant_factor =
+		    ConvertPolygonOffsetConstantFactor(guest_constant_factor, poly_offset, depth.format);
+		ret.depth_bias_clamp = poly_offset.clamp;
+		ret.depth_bias_slope_factor =
+		    (use_front ? poly_offset.front_scale : poly_offset.back_scale) / 16.0f;
+	}
+
 	// CB_COLOR_CONTROL.operation controls special CB operations, not the normal color component
 	// write mask. Use CB_TARGET_MASK here so scanout passes with mode=Disable do not go black.
 	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
@@ -380,6 +415,13 @@ static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffe
 	}
 	vk_buffer.setLineWidth(line_width);
 
+	vk_buffer.setDepthBiasEnable(dynamic_params.depth_bias_enable ? VK_TRUE : VK_FALSE);
+	if (dynamic_params.depth_bias_enable) {
+		vk_buffer.setDepthBias(dynamic_params.depth_bias_constant_factor,
+		                       dynamic_params.depth_bias_clamp,
+		                       dynamic_params.depth_bias_slope_factor);
+	}
+
 	if (dynamic_params.stencil_test_enable) {
 		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
 		                                dynamic_params.stencil_front.compareMask);
@@ -650,11 +692,9 @@ static bool ConsumeMetadataColorOperation(const RenderCommandBuffer& buffer) {
 }
 
 struct DrawEmitInfo {
-	bool     indexed           = false;
-	bool     draw_prim7_as_ngg = false;
-	uint32_t draw_vertex_count = 0;
-	int32_t  vertex_offset     = 0;
-	uint32_t first_vertex      = 0;
+	bool     indexed       = false;
+	int32_t  vertex_offset = 0;
+	uint32_t first_vertex  = 0;
 };
 
 struct DrawIndexBufferSource {
@@ -667,12 +707,10 @@ struct DrawIndexBufferSource {
 
 struct PreparedIndexBuffer {
 	std::shared_ptr<void> owner;
-	vk::Buffer            buffer   = nullptr;
-	uint64_t              address  = 0;
-	uint64_t              size     = 0;
-	vk::DeviceSize        offset   = 0;
-	vk::IndexType         type     = vk::IndexType::eUint16;
-	bool                  streamed = false;
+	vk::Buffer            buffer = nullptr;
+	uint64_t              size   = 0;
+	vk::DeviceSize        offset = 0;
+	vk::IndexType         type   = vk::IndexType::eUint16;
 };
 
 static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer) {
@@ -689,11 +727,24 @@ struct VertexBufferRange {
 	[[nodiscard]] uint64_t RequestedSize() const { return requested_end - base_address; }
 };
 
-static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&         buffer,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+struct PreparedVertexBuffers {
+	static constexpr uint32_t MaxBuffers = ShaderVertexInputInfo::RES_MAX;
+
+	std::array<vk::Buffer, MaxBuffers>            buffers {};
+	std::array<vk::DeviceSize, MaxBuffers>        offsets {};
+	std::array<std::shared_ptr<void>, MaxBuffers> owners {};
+	uint32_t                                      count       = 0;
+	uint32_t                                      owner_count = 0;
+};
+
+static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         buffer,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
+	EXIT_IF(vs_input_info.buffers_num < 0 ||
+	        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX);
+
 	// Collect the non-empty guest vertex ranges.
-	std::vector<VertexBufferRange> ranges;
-	ranges.reserve(vs_input_info.buffers_num);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> ranges {};
+	uint32_t                                                      range_count = 0;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
@@ -704,27 +755,31 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 			EXIT("invalid vertex buffer range: addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 			     vertex.addr, size);
 		}
-		ranges.push_back({vertex.addr, vertex.addr + size});
+		ranges[range_count++] = {vertex.addr, vertex.addr + size};
 	}
 
-	std::ranges::sort(ranges, [](const VertexBufferRange& left, const VertexBufferRange& right) {
-		return left.base_address < right.base_address;
-	});
+	std::sort(ranges.begin(), ranges.begin() + range_count,
+	          [](const VertexBufferRange& left, const VertexBufferRange& right) {
+		          return left.base_address < right.base_address;
+	          });
 
 	// Merge overlapping or touching ranges before acquiring host buffers.
-	std::vector<VertexBufferRange> merged_ranges;
-	merged_ranges.reserve(ranges.size());
-	for (const auto& range: ranges) {
-		if (!merged_ranges.empty() && merged_ranges.back().requested_end >= range.base_address) {
-			merged_ranges.back().requested_end =
-			    std::max(merged_ranges.back().requested_end, range.requested_end);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> merged_ranges {};
+	uint32_t                                                      merged_count = 0;
+	for (uint32_t i = 0; i < range_count; i++) {
+		const auto& range = ranges[i];
+		if (merged_count != 0 &&
+		    merged_ranges[merged_count - 1].requested_end >= range.base_address) {
+			merged_ranges[merged_count - 1].requested_end =
+			    std::max(merged_ranges[merged_count - 1].requested_end, range.requested_end);
 			continue;
 		}
-		merged_ranges.push_back(range);
+		merged_ranges[merged_count++] = {range.base_address, range.requested_end};
 	}
 
 	auto& cache = buffer.GetContext().GetBufferCache();
-	for (auto& range: merged_ranges) {
+	for (uint32_t i = 0; i < merged_count; i++) {
+		auto& range = merged_ranges[i];
 		// PPSA20298
 		const auto size =
 		    Libs::LibKernel::Memory::ClampRangeSize(range.base_address, range.RequestedSize());
@@ -733,30 +788,47 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 	}
 
 	// Rebuild slot bindings, offsetting non-empty slots into their acquired merged range.
-	std::vector<BufferBinding> bindings;
-	bindings.reserve(vs_input_info.buffers_num);
+	PreparedVertexBuffers prepared;
+	prepared.count = static_cast<uint32_t>(vs_input_info.buffers_num);
+	std::shared_ptr<Buffer> null_owner;
+	vk::Buffer              null_buffer = nullptr;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
 		if (size == 0) {
-			auto owner = cache.ObtainNullBuffer();
-			bindings.push_back({owner, owner->Handle(), 0});
+			if (null_owner == nullptr) {
+				null_owner  = cache.ObtainNullBuffer();
+				null_buffer = null_owner->Handle();
+			}
+			prepared.buffers[i] = null_buffer;
+			prepared.offsets[i] = 0;
 			continue;
 		}
 
-		const auto range = std::ranges::find_if(merged_ranges, [&](const VertexBufferRange& value) {
-			return vertex.addr >= value.base_address && vertex.addr < value.acquired_end;
-		});
-		if (range == merged_ranges.end()) {
+		const auto range = std::find_if(merged_ranges.begin(), merged_ranges.begin() + merged_count,
+		                                [&](const VertexBufferRange& value) {
+			                                return vertex.addr >= value.base_address &&
+			                                       vertex.addr < value.acquired_end;
+		                                });
+		if (range == merged_ranges.begin() + merged_count) {
 			EXIT("vertex buffer address is outside the acquired range: addr=0x%016" PRIx64 "\n",
 			     vertex.addr);
 		}
 
-		auto binding = range->binding;
-		binding.offset += vertex.addr - range->base_address;
-		bindings.push_back(std::move(binding));
+		prepared.buffers[i] = range->binding.buffer;
+		prepared.offsets[i] = range->binding.offset + vertex.addr - range->base_address;
 	}
-	return bindings;
+
+	if (null_owner != nullptr) {
+		prepared.owners[prepared.owner_count++] = std::move(null_owner);
+	}
+	for (uint32_t i = 0; i < merged_count; i++) {
+		if (merged_ranges[i].binding.owner != nullptr) {
+			EXIT_IF(prepared.owner_count >= PreparedVertexBuffers::MaxBuffers);
+			prepared.owners[prepared.owner_count++] = std::move(merged_ranges[i].binding.owner);
+		}
+	}
+	return prepared;
 }
 
 static void SetDrawDebugPhase(RenderCommandBuffer& buffer, uint64_t submit_id,
@@ -767,7 +839,7 @@ static void SetDrawDebugPhase(RenderCommandBuffer& buffer, uint64_t submit_id,
 	                    draw.flags, draw.instance_count, draw.first_instance);
 }
 
-static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw, bool use_ngg_rectlist_draw,
+static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
                             vk::PrimitiveTopology& topology) {
 
 	topology = vk::PrimitiveTopology::ePointList;
@@ -791,8 +863,7 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw, bool use
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
 		case Prospero::PrimitiveType::kRectList:
-			topology = (auto_draw && use_ngg_rectlist_draw ? vk::PrimitiveTopology::eTriangleStrip
-			                                               : vk::PrimitiveTopology::eTriangleList);
+			topology = vk::PrimitiveTopology::ePatchList;
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
 			if (!auto_draw) {
@@ -889,10 +960,9 @@ static void RefreshShaders(RenderCommandBuffer& buffer, const DrawCallInfo& draw
 	}
 }
 
-static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t                     submit_id,
-                                                       RenderCommandBuffer&         buffer,
-                                                       const DrawCallInfo&          draw,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+static PreparedVertexBuffers PrepareVertexBuffers(uint64_t submit_id, RenderCommandBuffer& buffer,
+                                                  const DrawCallInfo&          draw,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
 	EXIT_IF(draw.name == nullptr);
 	(void)submit_id;
 
@@ -900,12 +970,6 @@ static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t
 	return AcquireVertexBuffers(buffer, vs_input_info);
 }
 
-static void RebindVertexBuffers(RenderCommandBuffer&         buffer,
-                                const ShaderVertexInputInfo& vs_input_info,
-                                std::vector<BufferBinding>&  bindings) {
-	bindings = AcquireVertexBuffers(buffer, vs_input_info);
-}
-
 static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffer,
                                               const DrawIndexBufferSource& source) {
 	PreparedIndexBuffer prepared;
@@ -913,16 +977,14 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 		return prepared;
 	}
 	EXIT_IF(source.size == 0);
-	prepared.address = source.address;
-	prepared.size    = source.size;
-	prepared.type    = source.type;
+	prepared.size = source.size;
+	prepared.type = source.type;
 	if (source.host_data != nullptr) {
 		auto binding =
 		    buffer.GetContext().GetBufferCache().UploadTransient(source.host_data, source.size, 16);
-		prepared.owner    = std::move(binding.owner);
-		prepared.buffer   = binding.buffer;
-		prepared.offset   = binding.offset;
-		prepared.streamed = true;
+		prepared.owner  = std::move(binding.owner);
+		prepared.buffer = binding.buffer;
+		prepared.offset = binding.offset;
 	} else {
 		auto binding =
 		    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, source.address, source.size);
@@ -933,36 +995,29 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 	return prepared;
 }
 
-static void RebindIndexBuffer(RenderCommandBuffer& buffer, PreparedIndexBuffer& prepared) {
-	if (prepared.size == 0 || prepared.streamed) {
-		return;
-	}
-	auto binding =
-	    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, prepared.address, prepared.size);
-	prepared.owner  = std::move(binding.owner);
-	prepared.buffer = binding.buffer;
-	prepared.offset = binding.offset;
-}
-
 static void CommitVertexBuffers(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                                std::vector<BufferBinding>& bindings) {
-	for (uint32_t slot = 0; slot < bindings.size(); slot++) {
-		auto& binding = bindings[slot];
-		if (binding.owner != nullptr) {
-			buffer.RetainResourceUntilFence(binding.owner);
+                                PreparedVertexBuffers& prepared) {
+	for (uint32_t i = 0; i < prepared.owner_count; i++) {
+		if (prepared.owners[i] != nullptr) {
+			buffer.RetainResourceUntilFence(std::move(prepared.owners[i]));
 		}
-		EXIT_IF(binding.buffer == nullptr);
-		vk_buffer.bindVertexBuffers(slot, 1, &binding.buffer, &binding.offset);
+	}
+	for (uint32_t i = 0; i < prepared.count; i++) {
+		EXIT_IF(prepared.buffers[i] == nullptr);
+	}
+	if (prepared.count != 0) {
+		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
+		                            prepared.offsets.data());
 	}
 }
 
 static void CommitIndexBuffer(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                              const PreparedIndexBuffer& prepared) {
+                              PreparedIndexBuffer& prepared) {
 	if (prepared.size == 0) {
 		return;
 	}
 	if (prepared.owner != nullptr) {
-		buffer.RetainResourceUntilFence(prepared.owner);
+		buffer.RetainResourceUntilFence(std::move(prepared.owner));
 	}
 	EXIT_IF(prepared.buffer == nullptr);
 	vk_buffer.bindIndexBuffer(prepared.buffer, prepared.offset, prepared.type);
@@ -991,20 +1046,6 @@ static void LogDrawStateIfNeeded(const RenderCommandBuffer& buffer, const DrawCa
 	// LogDrawTextureState(draw.name, state.color_info[0], state.ps_input_info);
 }
 
-static bool IsHostExpandedRectListDrawSupported(const ShaderVertexInputInfo& vs_input_info,
-                                                const DrawCallInfo&          draw,
-                                                const DrawEmitInfo&          emit) {
-	if (!emit.draw_prim7_as_ngg) {
-		return true;
-	}
-
-	if (vs_input_info.buffers_num != 0) {
-		return false;
-	}
-
-	return draw.index_count == 3 || draw.index_count == emit.draw_vertex_count;
-}
-
 static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_buffer,
                                const ShaderVertexInputInfo& vs_input_info, const DrawCallInfo& draw,
                                const DrawEmitInfo& emit) {
@@ -1017,22 +1058,12 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriList:
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
-			if (emit.indexed) {
-				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
-				                      draw.first_instance);
-			} else {
-				vk_buffer.draw(draw.index_count, draw.instance_count, emit.first_vertex,
-				               draw.first_instance);
-			}
-			break;
 		case Prospero::PrimitiveType::kRectList:
 			if (emit.indexed) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
 				                      draw.first_instance);
 			} else {
-				EXIT_NOT_IMPLEMENTED(
-				    !IsHostExpandedRectListDrawSupported(vs_input_info, draw, emit));
-				vk_buffer.draw(emit.draw_vertex_count, draw.instance_count, emit.first_vertex,
+				vk_buffer.draw(draw.index_count, draw.instance_count, emit.first_vertex,
 				               draw.first_instance);
 			}
 			break;
@@ -1074,8 +1105,6 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer
 	                                               state.ps_input_info.stage, state.ps_active);
 	auto vertex_bindings = PrepareVertexBuffers(submit_id, buffer, draw, state.vs_input_info);
 	auto index_binding   = PrepareIndexBuffer(buffer, index_source);
-	RebindVertexBuffers(buffer, state.vs_input_info, vertex_bindings);
-	RebindIndexBuffer(buffer, index_binding);
 	state.rendering =
 	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info);
 
@@ -1159,7 +1188,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 	                    reinterpret_cast<uint64_t>(index_addr));
 
 	Common::LockGuard lock(m_context.GetMutex());
-	if (index_count == 0) {
+	if (index_count == 0 || instance_count == 0) {
 		return;
 	}
 
@@ -1201,7 +1230,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 	hw_check(buffer);
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, false, false, topology)) {
+	if (!GetDrawTopology(ucfg, false, topology)) {
 		return;
 	}
 
@@ -1229,10 +1258,6 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 
 	EXIT_NOT_IMPLEMENTED(flags != 0);
 	EXIT_NOT_IMPLEMENTED(type != 1);
-	if (instance_count == 0) {
-		instance_count = 1;
-	}
-
 	const DrawCallInfo    draw {"DrawIndex",    CommandBufferDebugOp::DrawIndex,
 	                            index_count,    flags,
 	                            instance_count, first_instance};
@@ -1292,7 +1317,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	                    index_count, flags, first_vertex, instance_count, first_instance);
 
 	Common::LockGuard lock(m_context.GetMutex());
-	if (index_count == 0) {
+	if (index_count == 0 || instance_count == 0) {
 		return;
 	}
 
@@ -1330,10 +1355,6 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	hw_check(buffer);
 
 	EXIT_NOT_IMPLEMENTED(flags != 0);
-	if (instance_count == 0) {
-		instance_count = 1;
-	}
-
 	const DrawCallInfo draw {"DrawIndexAuto", CommandBufferDebugOp::DrawIndexAuto,
 	                         index_count,     flags,
 	                         instance_count,  first_instance};
@@ -1345,20 +1366,15 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 		return;
 	}
 
-	vk::PrimitiveTopology topology              = vk::PrimitiveTopology::ePointList;
-	const bool            use_ngg_rectlist_draw = Config::NggRectlistDrawEnabled();
-
-	if (!GetDrawTopology(ucfg, true, use_ngg_rectlist_draw, topology)) {
+	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
+	if (!GetDrawTopology(ucfg, true, topology)) {
 		ResetBindings();
 		return;
 	}
-	const bool draw_prim7_as_ngg =
-	    (use_ngg_rectlist_draw &&
-	     ucfg.GetPrimType() == Prospero::GpuEnumValue(Prospero::PrimitiveType::kRectList));
-
 	RefreshShaders(buffer, draw, false, state);
 
-	if (draw_prim7_as_ngg && state.vs_input_info.buffers_num == 0 &&
+	const bool rect_list = topology == vk::PrimitiveTopology::ePatchList;
+	if (rect_list && state.vs_input_info.buffers_num == 0 &&
 	    state.vs_input_info.param_export_mask == 0 && state.ps_input_info.input_num != 0) {
 		if (graphics_debug_dump_enabled()) {
 			LOGF("DrawIndexAuto: skipping rect-list draw with no VS param exports and PS inputs: "
@@ -1375,13 +1391,10 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	                         Prospero::GpuEnumValue(Prospero::PrimitiveType::kRectListLegacy),
 	                     0, nullptr);
 
-	const uint32_t draw_vertex_count = (draw_prim7_as_ngg ? 4u : index_count);
-	const auto     vertex_offset = ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
-	                               static_cast<int32_t>(first_vertex);
-	DrawEmitInfo   emit {};
-	emit.draw_prim7_as_ngg = draw_prim7_as_ngg;
-	emit.draw_vertex_count = draw_vertex_count;
-	emit.first_vertex      = static_cast<uint32_t>(vertex_offset);
+	const auto   vertex_offset = ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
+	                             static_cast<int32_t>(first_vertex);
+	DrawEmitInfo emit {};
+	emit.first_vertex = static_cast<uint32_t>(vertex_offset);
 
 	DrawIndexBufferSource index_source {};
 	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false, false,

src/graphics/host_gpu/renderer/renderDraw.h

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

src/graphics/presentation/imeDialogOverlay.cpp

@@ -0,0 +1,876 @@
+#include "graphics/presentation/imeDialogOverlay.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/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 Ime = Libs::Dialog::ImeDialog;
+
+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 OnVisibilityChanged(bool visible, uint64_t generation) {
+	std::scoped_lock visibility_lock(g_visibility_mutex);
+	if (!g_input_lifecycle_active) {
+		return;
+	}
+	if (generation < g_generation || (generation == g_generation && visible == g_active)) {
+		return;
+	}
+
+	g_generation            = generation;
+	bool capture_controller = false;
+	bool capture_keyboard   = false;
+	bool multiline          = false;
+	if (visible) {
+		Ime::HostSnapshot snapshot;
+		if (Ime::GetHostSnapshot(&snapshot) && snapshot.generation == generation) {
+			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;
+		} else {
+			visible = false;
+		}
+	}
+	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++;
+		}
+	}
+}
+
+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 InitializeImeDialogInput() {
+	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);
+	}
+	Ime::SetVisibilityCallback(OnVisibilityChanged);
+	Ime::HostSnapshot snapshot;
+	if (Ime::GetHostSnapshot(&snapshot)) {
+		OnVisibilityChanged(true, snapshot.generation);
+	}
+}
+
+void ShutdownImeDialogInput() {
+	Ime::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();
+	}
+}
+
+bool ProcessImeDialogInput(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 ImeDialogOverlay::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 ImeDialog 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("##ImeDialog", 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;
+};
+
+ImeDialogOverlay::ImeDialogOverlay(GraphicContext& graphics)
+    : m_impl(std::make_unique<Impl>(graphics)) {}
+
+ImeDialogOverlay::~ImeDialogOverlay() = default;
+
+bool ImeDialogOverlay::PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count) {
+	return m_impl->PrepareFrame(extent, format, image_count);
+}
+
+void ImeDialogOverlay::Record(vk::CommandBuffer command, vk::ImageView target) {
+	m_impl->Record(command, target);
+}
+
+void ImeDialogOverlay::ReleaseVulkan() {
+	m_impl->ReleaseVulkan();
+}
+
+} // namespace Libs::Graphics

src/graphics/presentation/imeDialogOverlay.h

@@ -0,0 +1,36 @@
+#ifndef EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEDIALOGOVERLAY_H_
+#define EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEDIALOGOVERLAY_H_
+
+#include "common/common.h"
+#include "graphics/host_gpu/vulkanCommon.h"
+
+#include <memory>
+
+union SDL_Event;
+
+namespace Libs::Graphics {
+
+struct GraphicContext;
+
+void InitializeImeDialogInput();
+void ShutdownImeDialogInput();
+bool ProcessImeDialogInput(const SDL_Event& event);
+
+class ImeDialogOverlay final {
+public:
+	explicit ImeDialogOverlay(GraphicContext& graphics);
+	~ImeDialogOverlay();
+	KYTY_CLASS_NO_COPY(ImeDialogOverlay);
+
+	[[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_IMEDIALOGOVERLAY_H_

src/graphics/presentation/presenter.h

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

src/graphics/presentation/renderDoc.cpp

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

src/graphics/presentation/videoOut.cpp

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

src/graphics/presentation/window/hostInput.cpp

@@ -0,0 +1,256 @@
+#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 Binding {
+	SDL_Keycode key          = SDLK_UNKNOWN;
+	uint8_t     mouse_button = 0;
+	uint32_t    pad_button   = 0;
+};
+
+static constexpr std::array PAD_BUTTONS = {
+    std::pair {std::string_view("L3"), Controller::PAD_BUTTON_L3},
+    std::pair {std::string_view("R3"), Controller::PAD_BUTTON_R3},
+    std::pair {std::string_view("Options"), Controller::PAD_BUTTON_OPTIONS},
+    std::pair {std::string_view("Up"), Controller::PAD_BUTTON_UP},
+    std::pair {std::string_view("Right"), Controller::PAD_BUTTON_RIGHT},
+    std::pair {std::string_view("Down"), Controller::PAD_BUTTON_DOWN},
+    std::pair {std::string_view("Left"), Controller::PAD_BUTTON_LEFT},
+    std::pair {std::string_view("L2"), Controller::PAD_BUTTON_L2},
+    std::pair {std::string_view("R2"), Controller::PAD_BUTTON_R2},
+    std::pair {std::string_view("L1"), Controller::PAD_BUTTON_L1},
+    std::pair {std::string_view("R1"), Controller::PAD_BUTTON_R1},
+    std::pair {std::string_view("Triangle"), Controller::PAD_BUTTON_TRIANGLE},
+    std::pair {std::string_view("Circle"), Controller::PAD_BUTTON_CIRCLE},
+    std::pair {std::string_view("Cross"), Controller::PAD_BUTTON_CROSS},
+    std::pair {std::string_view("Square"), Controller::PAD_BUTTON_SQUARE},
+    std::pair {std::string_view("TouchPad"), Controller::PAD_BUTTON_TOUCH_PAD},
+};
+
+uint32_t PadButtonFromName(std::string_view name) {
+	const auto button = std::find_if(PAD_BUTTONS.begin(), PAD_BUTTONS.end(),
+	                                 [name](const auto& item) { return item.first == name; });
+	return button != PAD_BUTTONS.end() ? button->second : 0;
+}
+
+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.pad_button == second.pad_button ||
+	       (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.pad_button    = PadButtonFromName(entry.substr(0, split));
+				const auto host_input = entry.substr(split + 1);
+				binding.mouse_button  = MouseButtonFromName(host_input);
+				if (binding.mouse_button == 0) {
+					binding.key = NormalizeKey(SDL_GetKeyFromName(std::string(host_input).c_str()));
+				}
+			}
+
+			const bool reserved =
+			    binding.key == SDLK_ESCAPE || binding.key == SDLK_SPACE || binding.key == SDLK_F1;
+			if (binding.pad_button == 0 || 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]] uint32_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->pad_button : 0;
+	}
+
+	[[nodiscard]] uint32_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->pad_button : 0;
+	}
+
+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, PAD_BUTTONS.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 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()) {
+		SetButton(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) {
+		SetButton(map.FindMouseButton(mouse_button), down);
+	}
+}
+
+} // namespace Libs::Graphics

src/graphics/presentation/window/hostInput.h

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

src/graphics/presentation/window/swapchain.cpp

@@ -31,11 +31,13 @@
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/presentation/imeDialogOverlay.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window/windowInternal.h"
 #include "libs/controller.h"
+#include "libs/imeDialog.h"
 #include "loader/systemContent.h"
 
 #include <algorithm>
@@ -329,8 +331,9 @@ 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 +345,17 @@ 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<ImeDialogOverlay> m_ime_overlay;
+	uint32_t                          m_image_index = static_cast<uint32_t>(-1);
+	uint32_t                          m_frame_index = 0;
 };
 
 struct Presenter::Impl {
@@ -383,17 +387,20 @@ 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 +529,9 @@ 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 +616,15 @@ 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<ImeDialogOverlay>(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 +670,14 @@ 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 +686,22 @@ 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 +806,12 @@ bool Presenter::IsGuestPaused() const noexcept {
 	return m_impl->window.loop.paused.load(std::memory_order_acquire);
 }
 
+bool Presenter::NeedsImeRefresh() const noexcept {
+	const auto visual = Dialog::ImeDialog::GetVisualState();
+	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 +829,22 @@ 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 = Dialog::ImeDialog::GetVisualState();
+	auto&      swapchain  = m_impl->swapchain;
 	for (uint32_t attempt = 0; attempt < 2; attempt++) {
 		auto status = swapchain.AcquireNextImage();
 		if (status != Swapchain::Status::Success) {
@@ -814,8 +857,9 @@ 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 +869,7 @@ void Presenter::Present(Frame& frame, bool reuse) {
 		}
 
 		RenderDocOnPresent();
+		m_impl->presented_ime_revision.store(ime_visual.revision, std::memory_order_release);
 		window.UpdateTitle();
 		m_impl->frames.Release(&frame, true);
 		return;

src/graphics/presentation/window/vulkanWindow.cpp

@@ -31,6 +31,7 @@
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/presentation/imeDialogOverlay.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
@@ -259,6 +260,10 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("sampleRateShading is not supported\n");
 			skip_device = true;
 		}
+		if (device_features2.features.depthBiasClamp != VK_TRUE) {
+			LOGF("depthBiasClamp is not supported\n");
+			skip_device = true;
+		}
 
 		if (device_features2.features.fragmentStoresAndAtomics != VK_TRUE) {
 			LOGF("fragmentStoresAndAtomics is not supported\n");
@@ -548,6 +553,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	EXIT_NOT_IMPLEMENTED(required_features13.synchronization2 == VK_TRUE &&
 	                     supported_features13.synchronization2 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.sampleRateShading != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.depthBiasClamp != VK_TRUE);
 	features12.timelineSemaphore = VK_TRUE;
 
 	vk::PhysicalDeviceFeatures device_features {};
@@ -563,6 +569,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.independentBlend                     = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
+	device_features.depthBiasClamp                       = VK_TRUE;
 	graphics.sample_rate_shading_enabled                 = true;
 	device_features.vertexPipelineStoresAndAtomics =
 	    supported_features2.features.vertexPipelineStoresAndAtomics;
@@ -1008,6 +1015,7 @@ void WindowContext::RecreateSurface() {
 }
 
 WindowContext::~WindowContext() {
+	ShutdownImeDialogInput();
 	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/imeDialogOverlay.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;
@@ -260,16 +241,11 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 		}
 	}
 
-	const auto button = KeyboardKeyToPadButton(key.key_code);
-	if (button != 0 && (key.down || key.up) && !key.repeat) {
-		static bool keyboard_connected = false;
-		if (!keyboard_connected) {
-			Controller::ControllerConnect(KEYBOARD_CONTROLLER_ID);
-			keyboard_connected = true;
-		}
-		Controller::ControllerButton(KEYBOARD_CONTROLLER_ID, button, key.down);
-	}
 #endif
+
+	if ((key.down || key.up) && !key.repeat) {
+		HostInputKey(key.key_code, key.down);
+	}
 }
 
 static void GameEventMouse([[maybe_unused]] const EventMouse& mb) {
@@ -291,6 +267,23 @@ static void GameEventMouse([[maybe_unused]] const EventMouse& mb) {
 		     mb.x, mb.y);
 	}
 #endif
+
+	if (mb.down || mb.up) {
+		uint8_t mouse_button = 0;
+		if (mb.left) {
+			mouse_button = SDL_BUTTON_LEFT;
+		} else if (mb.middle) {
+			mouse_button = SDL_BUTTON_MIDDLE;
+		} else if (mb.right) {
+			mouse_button = SDL_BUTTON_RIGHT;
+		} else if (mb.x1) {
+			mouse_button = SDL_BUTTON_X1;
+		} else if (mb.x2) {
+			mouse_button = SDL_BUTTON_X2;
+		}
+
+		HostInputMouseButton(mouse_button, mb.down);
+	}
 }
 
 static void GameEventFinger([[maybe_unused]] const EventFinger& f) {
@@ -484,6 +477,9 @@ void WindowContext::ProcessEvent(double time_s) {
 	auto& game  = loop;
 	auto* event = &game.event;
 	EXIT_IF(SDL_GetEventState(SDL_DISPLAYEVENT) != SDL_ENABLE);
+	if (ProcessImeDialogInput(*event)) {
+		return;
+	}
 
 	switch (event->type) {
 		case SDL_QUIT: GameEventQuit(game); break;
@@ -756,26 +752,18 @@ void WindowContext::Run() {
 #if defined(__APPLE__)
 		DrainMainThreadTasks();
 #endif
-		if (SDL_PollEvent(&loop.event) != 0) {
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
 		if (loop.paused.load(std::memory_order_acquire)) {
 			if (!timer.IsPaused()) {
 				timer.Pause();
 			}
-			if (SDL_WaitEvent(&loop.event) == 0) {
-				EXIT("%s\n", SDL_GetError());
-			}
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
-		if (timer.IsPaused()) {
+		} else if (timer.IsPaused()) {
 			timer.Resume();
 		}
-		Common::Thread::SleepMicro(1000);
+
+		if (SDL_WaitEvent(&loop.event) == 0) {
+			EXIT("%s\n", SDL_GetError());
+		}
+		ProcessEvent(timer.GetTimeS());
 	}
 }
 
@@ -794,10 +782,15 @@ static void WindowCreate(WindowContext& context) {
 	if (SDL_InitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMECONTROLLER) < 0) {
 		EXIT("%s\n", SDL_GetError());
 	}
+	HostInputInit();
+	InitializeImeDialogInput();
 
 	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 +930,15 @@ 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 +950,11 @@ 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/SpirvBuilder.cpp

@@ -19,7 +19,7 @@ static void AppendInstructionWords(std::vector<uint32_t>& section, const uint32_
 	section.insert(section.end(), words + 1, words + words_num);
 }
 
-Builder::Builder() {
+Builder::Builder(uint32_t version): m_version(version) {
 	m_debug.reserve(InitialSpirvSectionReserve);
 	m_annotations.reserve(InitialSpirvSectionReserve);
 	m_types.reserve(InitialSpirvSectionReserve);
@@ -138,7 +138,7 @@ std::vector<uint32_t> Builder::Build() const {
 	               m_debug.size() + m_annotations.size() + m_types.size() + m_functions.size());
 
 	module.push_back(0x07230203u);
-	module.push_back(0x00010300u);
+	module.push_back(m_version);
 	module.push_back(0u);
 	module.push_back(m_next_id);
 	module.push_back(0u);

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

@@ -10,7 +10,7 @@ namespace Libs::Graphics::ShaderRecompiler::Spirv {
 
 class Builder {
 public:
-	Builder();
+	explicit Builder(uint32_t version = 0x00010300u);
 	~Builder() = default;
 	KYTY_CLASS_DEFAULT_COPY(Builder);
 
@@ -39,6 +39,7 @@ private:
 	static void AppendString(std::vector<uint32_t>& words, const char* text);
 
 	uint32_t              m_next_id = 1;
+	uint32_t              m_version = 0;
 	std::vector<uint32_t> m_capabilities;
 	std::vector<uint32_t> m_extensions;
 	std::vector<uint32_t> m_ext_inst_imports;

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

@@ -7,51 +7,30 @@ namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
 uint32_t PixelParameterMappedLocation(const EmitterState& state, uint32_t attr) {
 	const auto* ps = state.pixel_input_info;
-	if (state.stage != ShaderType::Pixel || ps == nullptr || attr >= ps->input_num) {
+	if (state.stage != ShaderType::Pixel || ps == nullptr) {
 		return attr;
 	}
-	// VINTRP ATTR selects the PS input slot. SPI_PS_INPUT_CNTL maps that slot to a
-	// VS parameter export, which is the SPIR-V location we must link against.
-	return ps->interpolator_settings[attr] & PsInputOffsetMask;
+	return ShaderPixelParameterMappedLocation(*ps, attr);
 }
 
 uint32_t PixelParameterLocation(const EmitterState& state, uint32_t attr) {
-	bool used_locations[32] = {};
-
+	std::array<uint32_t, 32> active_inputs {};
+	uint32_t                 active_count = 0;
 	for (const auto& input: state.inputs) {
-		if (input.kind != IR::StageInputKind::Parameter) {
-			continue;
-		}
-
-		auto location = PixelParameterMappedLocation(state, input.location);
-		if (location < std::size(used_locations) && used_locations[location]) {
-			auto fallback_location = input.location;
-			while (fallback_location < std::size(used_locations) &&
-			       used_locations[fallback_location]) {
-				fallback_location++;
-			}
-			EXIT_NOT_IMPLEMENTED(fallback_location >= std::size(used_locations));
-			location = fallback_location;
-		}
-
-		if (input.location == attr) {
-			return location;
-		}
-
-		if (location < std::size(used_locations)) {
-			used_locations[location] = true;
+		if (input.kind == IR::StageInputKind::Parameter) {
+			active_inputs[active_count++] = input.location;
 		}
 	}
-
-	return PixelParameterMappedLocation(state, attr);
+	return state.stage == ShaderType::Pixel && state.pixel_input_info != nullptr
+	           ? ShaderPixelParameterLocation(*state.pixel_input_info,
+	                                          {active_inputs.data(), active_count}, attr)
+	           : attr;
 }
 
 bool PixelParameterIsFlat(const EmitterState& state, uint32_t attr) {
 	const auto* ps = state.pixel_input_info;
-	if (state.stage != ShaderType::Pixel || ps == nullptr || attr >= ps->input_num) {
-		return false;
-	}
-	return (ps->interpolator_settings[attr] & PsInputFlatShade) != 0;
+	return state.stage == ShaderType::Pixel && ps != nullptr &&
+	       ShaderPixelParameterIsFlat(*ps, attr);
 }
 
 void SetError(std::string* error, const char* message) {
@@ -96,8 +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});
@@ -210,11 +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);
 			}
@@ -405,93 +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) {
@@ -502,7 +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;
@@ -425,9 +428,6 @@ struct EmitterState {
 	std::map<uint32_t, uint32_t>                     float_constants;
 };
 
-constexpr uint32_t PsInputOffsetMask = 0x0000001fu;
-constexpr uint32_t PsInputFlatShade  = 0x00000400u;
-
 enum class VertexInputScalarKind { Float, Sint, Uint };
 
 constexpr uint32_t NoImageComponent = 0xffffffffu;
@@ -614,8 +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,
@@ -654,8 +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);
@@ -1018,6 +1014,9 @@ void EmitDeviceAtomicMemoryBarrier(EmitterState& state);
 
 void EmitAtomicU32(EmitterState& state, const IR::Instruction& inst, uint32_t opcode);
 
+void EmitAtomicFMinF32(EmitterState& state, const IR::Instruction& inst);
+void EmitAtomicFMaxF32(EmitterState& state, const IR::Instruction& inst);
+
 void EmitSLoadDword(EmitterState& state, const IR::Instruction& inst);
 
 void EmitLoadSrtDword(EmitterState& state, const IR::Instruction& inst);
@@ -1348,7 +1347,7 @@ void EmitCompareI16(EmitterState& state, const IR::Instruction& inst, uint32_t o
 
 void EmitBitCompareB32(EmitterState& state, const IR::Instruction& inst, bool bit_set);
 
-void EmitCompareNeU64(EmitterState& state, const IR::Instruction& inst);
+void EmitCompareU64(EmitterState& state, const IR::Instruction& inst);
 
 void EmitCompareConstant(EmitterState& state, const IR::Instruction& inst, bool value);
 
@@ -1580,19 +1579,22 @@ uint32_t EmitValueOrZeroIfCondition(EmitterState& state, uint32_t condition, Fn&
 	}
 
 	const auto then_label  = state.builder.AllocateId();
+	const auto then_exit   = state.builder.AllocateId();
 	const auto else_label  = state.builder.AllocateId();
 	const auto merge_label = state.builder.AllocateId();
 	state.builder.AddFunction({OpSelectionMerge, merge_label, SelectionControlNone});
 	state.builder.AddFunction({OpBranchConditional, condition, then_label, else_label});
 	state.builder.AddFunction({OpLabel, then_label});
 	const auto then_value = fn();
+	state.builder.AddFunction({OpBranch, then_exit});
+	state.builder.AddFunction({OpLabel, then_exit});
 	state.builder.AddFunction({OpBranch, merge_label});
 	state.builder.AddFunction({OpLabel, else_label});
 	state.builder.AddFunction({OpBranch, merge_label});
 	state.builder.AddFunction({OpLabel, merge_label});
 	const auto value = state.builder.AllocateId();
 	state.builder.AddFunction(
-	    {OpPhi, state.uint_type, value, then_value, then_label, ConstantU32(state, 0), else_label});
+	    {OpPhi, state.uint_type, value, then_value, then_exit, ConstantU32(state, 0), else_label});
 	return value;
 }
 

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

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

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

@@ -200,12 +200,12 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 		if (dimension == Decoder::ImageDimension::Unknown || dimension != image.dimension ||
 		    DescriptorIsCube(descriptor) != image.cube) {
 			if (error != nullptr) {
-				*error = fmt::format(
-				    "image descriptor {} no longer matches specialized dimension: "
-				    "{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x}",
-				    i, descriptor.dwords[0], descriptor.dwords[1], descriptor.dwords[2],
-				    descriptor.dwords[3], descriptor.dwords[4], descriptor.dwords[5],
-				    descriptor.dwords[6], descriptor.dwords[7]);
+				*error =
+				    fmt::format("image descriptor {} no longer matches specialized dimension: "
+				                "{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x}",
+				                i, descriptor.dwords[0], descriptor.dwords[1], descriptor.dwords[2],
+				                descriptor.dwords[3], descriptor.dwords[4], descriptor.dwords[5],
+				                descriptor.dwords[6], descriptor.dwords[7]);
 			}
 			return false;
 		}
@@ -224,14 +224,17 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 			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 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("image descriptor {} no longer matches specialized format", i);
+					*error = fmt::format(
+					    "image descriptor {} no longer matches specialized format: "
+					    "{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x}",
+					    i, descriptor.dwords[0], descriptor.dwords[1], descriptor.dwords[2],
+					    descriptor.dwords[3], descriptor.dwords[4], descriptor.dwords[5],
+					    descriptor.dwords[6], descriptor.dwords[7]);
 				}
 				return false;
 			}

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

src/graphics/shader/rectListShader.cpp

@@ -0,0 +1,397 @@
+#include "graphics/shader/rectListShader.h"
+
+#include "common/assert.h"
+#include "graphics/shader/recompiler/emitter/SpirvBuilder.h"
+#include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/shader.h"
+#include "spirv/unified1/spirv.hpp11"
+
+#include <array>
+#include <bit>
+#include <cstdint>
+#include <utility>
+
+namespace Libs::Graphics {
+
+namespace {
+
+using ShaderRecompiler::Spirv::Builder;
+
+constexpr uint32_t SpirvVersion15 = 0x00010500u;
+
+template <typename T>
+constexpr uint32_t Word(T value) {
+	return static_cast<uint32_t>(value);
+}
+
+struct Parameter {
+	uint32_t input_location  = 0;
+	uint32_t output_location = 0;
+	bool     flat            = false;
+};
+
+std::vector<Parameter> GetParameters(const ShaderVertexInputInfo& vertex_info,
+                                     const ShaderPixelInputInfo*  pixel_info) {
+	if (pixel_info == nullptr) {
+		return {};
+	}
+	EXIT_IF(pixel_info->input_num > ShaderVertexInputInfo::RES_MAX);
+	EXIT_IF(pixel_info->stage.program == nullptr);
+
+	std::vector<uint32_t> active_inputs;
+	for (const auto& input: pixel_info->stage.program->info.inputs) {
+		if (input.kind == ShaderRecompiler::IR::StageInputKind::Parameter) {
+			active_inputs.push_back(input.location);
+		}
+	}
+
+	std::vector<Parameter> parameters;
+	for (const auto input: active_inputs) {
+		const auto input_location = ShaderPixelParameterMappedLocation(*pixel_info, input);
+		if ((vertex_info.param_export_mask & (1u << input_location)) != 0) {
+			parameters.push_back({input_location,
+			                      ShaderPixelParameterLocation(*pixel_info, active_inputs, input),
+			                      ShaderPixelParameterIsFlat(*pixel_info, input)});
+		}
+	}
+	return parameters;
+}
+
+class RectListEmitter {
+public:
+	RectListEmitter(const std::vector<Parameter>& parameters_, spv::ExecutionModel model)
+	    : parameters(parameters_) {
+		builder.AddMemoryModel(
+		    {Word(spv::AddressingModel::Logical), Word(spv::MemoryModel::GLSL450)});
+
+		void_type       = Type(spv::Op::OpTypeVoid);
+		uint_type       = Type(spv::Op::OpTypeInt, 32u, 0u);
+		int_type        = Type(spv::Op::OpTypeInt, 32u, 1u);
+		float_type      = Type(spv::Op::OpTypeFloat, 32u);
+		vec4_float_type = Type(spv::Op::OpTypeVector, float_type, 4u);
+		function_type   = Type(spv::Op::OpTypeFunction, void_type);
+
+		per_vertex_type = Type(spv::Op::OpTypeStruct, vec4_float_type);
+		builder.AddAnnotation({Word(spv::Op::OpMemberDecorate), per_vertex_type, 0u,
+		                       Word(spv::Decoration::BuiltIn), Word(spv::BuiltIn::Position)});
+		builder.AddAnnotation(
+		    {Word(spv::Op::OpDecorate), per_vertex_type, Word(spv::Decoration::Block)});
+
+		ptr_input_vec4_float  = Pointer(spv::StorageClass::Input, vec4_float_type);
+		ptr_output_vec4_float = Pointer(spv::StorageClass::Output, vec4_float_type);
+		if (model == spv::ExecutionModel::TessellationControl) {
+			bool_type        = Type(spv::Op::OpTypeBool);
+			vec2_bool_type   = Type(spv::Op::OpTypeVector, bool_type, 2u);
+			vec2_float_type  = Type(spv::Op::OpTypeVector, float_type, 2u);
+			ptr_output_float = Pointer(spv::StorageClass::Output, float_type);
+		} else {
+			vec3_float_type = Type(spv::Op::OpTypeVector, float_type, 3u);
+			ptr_input_float = Pointer(spv::StorageClass::Input, float_type);
+		}
+	}
+
+	std::vector<uint32_t> EmitControl() {
+		DefineEntry(spv::ExecutionModel::TessellationControl);
+		const auto float_one = Constant(float_type, std::bit_cast<uint32_t>(1.0f));
+
+		for (uint32_t i = 0; i < 4; i++) {
+			Store(Access(ptr_output_float, tess_outer, Int(i)), float_one);
+		}
+		for (uint32_t i = 0; i < 2; i++) {
+			Store(Access(ptr_output_float, tess_inner, Int(i)), float_one);
+		}
+
+		std::array<uint32_t, 3> positions {};
+		for (uint32_t i = 0; i < positions.size(); i++) {
+			positions[i] =
+			    Load(vec4_float_type, Access(ptr_input_vec4_float, gl_in, Int(i), Int(0)));
+		}
+
+		std::array<uint32_t, 3> coordinate_equal {};
+		for (uint32_t i = 0; i < coordinate_equal.size(); i++) {
+			const auto left  = Result(spv::Op::OpVectorShuffle, vec2_float_type, positions[i],
+			                          positions[i], 0u, 1u);
+			const auto right = Result(spv::Op::OpVectorShuffle, vec2_float_type,
+			                          positions[(i + 1u) % 3u], positions[(i + 1u) % 3u], 0u, 1u);
+			coordinate_equal[i] = Result(spv::Op::OpFOrdEqual, vec2_bool_type, left, right);
+		}
+
+		std::array<uint32_t, 3> barycentric {};
+		std::array<uint32_t, 3> edge_vertex {};
+		const auto float_minus_one = Constant(float_type, std::bit_cast<uint32_t>(-1.0f));
+		for (uint32_t i = 0; i < edge_vertex.size(); i++) {
+			const auto previous = (i + 2u) % 3u;
+			const auto xy       = Result(
+			    spv::Op::OpLogicalAnd, bool_type,
+			    Result(spv::Op::OpCompositeExtract, bool_type, coordinate_equal[i], 0u),
+			    Result(spv::Op::OpCompositeExtract, bool_type, coordinate_equal[previous], 1u));
+			const auto yx = Result(
+			    spv::Op::OpLogicalAnd, bool_type,
+			    Result(spv::Op::OpCompositeExtract, bool_type, coordinate_equal[i], 1u),
+			    Result(spv::Op::OpCompositeExtract, bool_type, coordinate_equal[previous], 0u));
+			edge_vertex[i] = Result(spv::Op::OpLogicalOr, bool_type, xy, yx);
+			barycentric[i] =
+			    Result(spv::Op::OpSelect, float_type, edge_vertex[i], float_minus_one, float_one);
+		}
+
+		auto vertex_index = Result(spv::Op::OpSelect, int_type, edge_vertex[2], Int(2), Int(0));
+		vertex_index = Result(spv::Op::OpSelect, int_type, edge_vertex[1], Int(1), vertex_index);
+		const auto invocation = Load(int_type, invocation_id);
+		const auto is_fourth  = Result(spv::Op::OpIEqual, bool_type, invocation, Int(3));
+		const auto index =
+		    Result(spv::Op::OpSMod, int_type,
+		           Result(spv::Op::OpIAdd, int_type, vertex_index, invocation), Int(3));
+
+		const auto position3 = Interpolate(positions[0], positions[1], positions[2], barycentric);
+		const auto position =
+		    Result(spv::Op::OpSelect, vec4_float_type, is_fourth, position3,
+		           Load(vec4_float_type, Access(ptr_input_vec4_float, gl_in, index, Int(0))));
+		Store(Access(ptr_output_vec4_float, gl_out, invocation, Int(0)), position);
+
+		for (uint32_t i = 0; i < parameters.size(); i++) {
+			const auto input0 =
+			    Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], Int(0)));
+			if (parameters[i].flat) {
+				Store(Access(ptr_output_vec4_float, outputs[i], invocation), input0);
+				continue;
+			}
+			const auto input1 =
+			    Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], Int(1)));
+			const auto input2 =
+			    Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], Int(2)));
+			const auto input3 = Interpolate(input0, input1, input2, barycentric);
+			const auto value =
+			    Result(spv::Op::OpSelect, vec4_float_type, is_fourth, input3,
+			           Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], index)));
+			Store(Access(ptr_output_vec4_float, outputs[i], invocation), value);
+		}
+
+		Emit(spv::Op::OpReturn);
+		Emit(spv::Op::OpFunctionEnd);
+		return builder.Build();
+	}
+
+	std::vector<uint32_t> EmitEvaluation() {
+		DefineEntry(spv::ExecutionModel::TessellationEvaluation);
+
+		const auto x     = Load(float_type, Access(ptr_input_float, tess_coord, Int(0)));
+		const auto y     = Load(float_type, Access(ptr_input_float, tess_coord, Int(1)));
+		const auto index = Result(
+		    spv::Op::OpIAdd, int_type,
+		    Result(spv::Op::OpIMul, int_type, Result(spv::Op::OpConvertFToS, int_type, y), Int(2)),
+		    Result(spv::Op::OpConvertFToS, int_type, x));
+
+		const auto position =
+		    Load(vec4_float_type, Access(ptr_input_vec4_float, gl_in, index, Int(0)));
+		Store(Access(ptr_output_vec4_float, gl_out, Int(0)), position);
+		for (uint32_t i = 0; i < parameters.size(); i++) {
+			Store(outputs[i],
+			      Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], index)));
+		}
+
+		Emit(spv::Op::OpReturn);
+		Emit(spv::Op::OpFunctionEnd);
+		return builder.Build();
+	}
+
+private:
+	template <typename... Args>
+	uint32_t Type(spv::Op opcode, Args... operands) {
+		const auto id = builder.AllocateId();
+		builder.AddType({Word(opcode), id, Word(operands)...});
+		return id;
+	}
+
+	uint32_t Constant(uint32_t type, uint32_t value) {
+		const auto id = builder.AllocateId();
+		builder.AddType({Word(spv::Op::OpConstant), type, id, value});
+		return id;
+	}
+
+	uint32_t Pointer(spv::StorageClass storage, uint32_t type) {
+		return Type(spv::Op::OpTypePointer, storage, type);
+	}
+
+	uint32_t Array(uint32_t type, uint32_t size) {
+		return Type(spv::Op::OpTypeArray, type, Uint(size));
+	}
+
+	template <typename... Args>
+	uint32_t Result(spv::Op opcode, uint32_t type, Args... operands) {
+		const auto id = builder.AllocateId();
+		builder.AddFunction({Word(opcode), type, id, Word(operands)...});
+		return id;
+	}
+
+	template <typename... Args>
+	uint32_t ResultWithoutType(spv::Op opcode, Args... operands) {
+		const auto id = builder.AllocateId();
+		builder.AddFunction({Word(opcode), id, Word(operands)...});
+		return id;
+	}
+
+	template <typename... Args>
+	void Emit(spv::Op opcode, Args... operands) {
+		builder.AddFunction({Word(opcode), Word(operands)...});
+	}
+
+	template <typename... Args>
+	uint32_t Access(uint32_t pointer_type, uint32_t base, Args... indices) {
+		return Result(spv::Op::OpAccessChain, pointer_type, base, Word(indices)...);
+	}
+
+	uint32_t Load(uint32_t type, uint32_t pointer) {
+		return Result(spv::Op::OpLoad, type, pointer);
+	}
+
+	void Store(uint32_t pointer, uint32_t value) { Emit(spv::Op::OpStore, pointer, value); }
+
+	uint32_t Int(uint32_t value) {
+		auto& id = int_constants[value];
+		if (id == 0) {
+			id = Constant(int_type, value);
+		}
+		return id;
+	}
+
+	uint32_t Uint(uint32_t value) {
+		auto& id = uint_constants[value];
+		if (id == 0) {
+			id = Constant(uint_type, value);
+		}
+		return id;
+	}
+
+	uint32_t AddInterface(spv::StorageClass storage, uint32_t type) {
+		const auto variable = builder.AllocateId();
+		builder.AddType(
+		    {Word(spv::Op::OpVariable), Pointer(storage, type), variable, Word(storage)});
+		interfaces.push_back(variable);
+		return variable;
+	}
+
+	void Decorate(uint32_t target, spv::Decoration decoration, uint32_t value) {
+		builder.AddAnnotation({Word(spv::Op::OpDecorate), target, Word(decoration), value});
+	}
+
+	void DefineEntry(spv::ExecutionModel model) {
+		builder.AddCapability({Word(spv::Capability::Shader)});
+		builder.AddCapability({Word(spv::Capability::Tessellation)});
+		main = Result(spv::Op::OpFunction, void_type, spv::FunctionControlMask::MaskNone,
+		              function_type);
+		if (model == spv::ExecutionModel::TessellationControl) {
+			builder.AddExecutionMode({main, Word(spv::ExecutionMode::OutputVertices), 4u});
+		} else {
+			builder.AddExecutionMode({main, Word(spv::ExecutionMode::Quads)});
+			builder.AddExecutionMode({main, Word(spv::ExecutionMode::SpacingEqual)});
+			builder.AddExecutionMode({main, Word(spv::ExecutionMode::VertexOrderCw)});
+		}
+		DefineInputs(model);
+		DefineOutputs(model);
+		builder.AddEntryPoint(Word(model), main, "main", interfaces);
+		ResultWithoutType(spv::Op::OpLabel);
+	}
+
+	void DefineInputs(spv::ExecutionModel model) {
+		const auto tess_control = model == spv::ExecutionModel::TessellationControl;
+		if (tess_control) {
+			invocation_id = AddInterface(spv::StorageClass::Input, int_type);
+			Decorate(invocation_id, spv::Decoration::BuiltIn, Word(spv::BuiltIn::InvocationId));
+		} else {
+			tess_coord = AddInterface(spv::StorageClass::Input, vec3_float_type);
+			Decorate(tess_coord, spv::Decoration::BuiltIn, Word(spv::BuiltIn::TessCoord));
+		}
+		gl_in =
+		    AddInterface(spv::StorageClass::Input, Array(per_vertex_type, tess_control ? 3u : 4u));
+
+		inputs.resize(parameters.size());
+		std::array<uint32_t, ShaderVertexInputInfo::RES_MAX> locations {};
+		for (uint32_t i = 0; i < parameters.size(); i++) {
+			const auto location =
+			    tess_control ? parameters[i].input_location : parameters[i].output_location;
+			if (tess_control && locations[location] != 0) {
+				inputs[i] = locations[location];
+				continue;
+			}
+			inputs[i] = AddInterface(spv::StorageClass::Input,
+			                         Array(vec4_float_type, tess_control ? 3u : 4u));
+			Decorate(inputs[i], spv::Decoration::Location, location);
+			locations[location] = inputs[i];
+		}
+	}
+
+	void DefineOutputs(spv::ExecutionModel model) {
+		const auto tess_control = model == spv::ExecutionModel::TessellationControl;
+		if (tess_control) {
+			gl_out     = AddInterface(spv::StorageClass::Output, Array(per_vertex_type, 4u));
+			tess_inner = AddInterface(spv::StorageClass::Output, Array(float_type, 2u));
+			Decorate(tess_inner, spv::Decoration::BuiltIn, Word(spv::BuiltIn::TessLevelInner));
+			builder.AddAnnotation(
+			    {Word(spv::Op::OpDecorate), tess_inner, Word(spv::Decoration::Patch)});
+			tess_outer = AddInterface(spv::StorageClass::Output, Array(float_type, 4u));
+			Decorate(tess_outer, spv::Decoration::BuiltIn, Word(spv::BuiltIn::TessLevelOuter));
+			builder.AddAnnotation(
+			    {Word(spv::Op::OpDecorate), tess_outer, Word(spv::Decoration::Patch)});
+		} else {
+			gl_out = AddInterface(spv::StorageClass::Output, per_vertex_type);
+		}
+
+		outputs.resize(parameters.size());
+		for (uint32_t i = 0; i < parameters.size(); i++) {
+			outputs[i] = AddInterface(spv::StorageClass::Output,
+			                          tess_control ? Array(vec4_float_type, 4u) : vec4_float_type);
+			Decorate(outputs[i], spv::Decoration::Location, parameters[i].output_location);
+		}
+	}
+
+	uint32_t Interpolate(uint32_t v0, uint32_t v1, uint32_t v2,
+	                     const std::array<uint32_t, 3>& barycentric) {
+		const auto p0 = Result(spv::Op::OpVectorTimesScalar, vec4_float_type, v0, barycentric[0]);
+		const auto p1 = Result(spv::Op::OpVectorTimesScalar, vec4_float_type, v1, barycentric[1]);
+		const auto p2 = Result(spv::Op::OpVectorTimesScalar, vec4_float_type, v2, barycentric[2]);
+		return Result(spv::Op::OpFAdd, vec4_float_type, p0,
+		              Result(spv::Op::OpFAdd, vec4_float_type, p1, p2));
+	}
+
+	Builder                       builder {SpirvVersion15};
+	const std::vector<Parameter>& parameters;
+	std::vector<uint32_t>         interfaces;
+	std::vector<uint32_t>         inputs;
+	std::vector<uint32_t>         outputs;
+	std::array<uint32_t, 5>       int_constants {};
+	std::array<uint32_t, 5>       uint_constants {};
+	uint32_t                      main                  = 0;
+	uint32_t                      void_type             = 0;
+	uint32_t                      bool_type             = 0;
+	uint32_t                      uint_type             = 0;
+	uint32_t                      int_type              = 0;
+	uint32_t                      float_type            = 0;
+	uint32_t                      vec2_bool_type        = 0;
+	uint32_t                      vec2_float_type       = 0;
+	uint32_t                      vec3_float_type       = 0;
+	uint32_t                      vec4_float_type       = 0;
+	uint32_t                      function_type         = 0;
+	uint32_t                      per_vertex_type       = 0;
+	uint32_t                      ptr_input_float       = 0;
+	uint32_t                      ptr_input_vec4_float  = 0;
+	uint32_t                      ptr_output_float      = 0;
+	uint32_t                      ptr_output_vec4_float = 0;
+	uint32_t                      gl_in                 = 0;
+	uint32_t                      gl_out                = 0;
+	uint32_t                      tess_inner            = 0;
+	uint32_t                      tess_outer            = 0;
+	uint32_t                      tess_coord            = 0;
+	uint32_t                      invocation_id         = 0;
+};
+
+} // namespace
+
+RectListShaders BuildRectListShaders(const ShaderVertexInputInfo& vertex_info,
+                                     const ShaderPixelInputInfo*  pixel_info) {
+	const auto      parameters = GetParameters(vertex_info, pixel_info);
+	RectListEmitter control(parameters, spv::ExecutionModel::TessellationControl);
+	RectListEmitter evaluation(parameters, spv::ExecutionModel::TessellationEvaluation);
+	return {control.EmitControl(), evaluation.EmitEvaluation()};
+}
+
+} // namespace Libs::Graphics

src/graphics/shader/rectListShader.h

@@ -0,0 +1,22 @@
+#ifndef EMULATOR_SRC_GRAPHICS_SHADER_RECTLISTSHADER_H_
+#define EMULATOR_SRC_GRAPHICS_SHADER_RECTLISTSHADER_H_
+
+#include <cstdint>
+#include <vector>
+
+namespace Libs::Graphics {
+
+struct ShaderPixelInputInfo;
+struct ShaderVertexInputInfo;
+
+struct RectListShaders {
+	std::vector<uint32_t> control;
+	std::vector<uint32_t> evaluation;
+};
+
+RectListShaders BuildRectListShaders(const ShaderVertexInputInfo& vertex_info,
+                                     const ShaderPixelInputInfo*  pixel_info);
+
+} // namespace Libs::Graphics
+
+#endif // EMULATOR_SRC_GRAPHICS_SHADER_RECTLISTSHADER_H_

src/graphics/shader/shader.cpp

@@ -45,6 +45,42 @@
 
 namespace Libs::Graphics {
 
+namespace {
+
+constexpr uint32_t PsInputOffsetMask = 0x0000001fu;
+constexpr uint32_t PsInputFlatShade  = 0x00000400u;
+
+} // namespace
+
+uint32_t ShaderPixelParameterMappedLocation(const ShaderPixelInputInfo& info, uint32_t input) {
+	return input < info.input_num ? info.interpolator_settings[input] & PsInputOffsetMask : input;
+}
+
+uint32_t ShaderPixelParameterLocation(const ShaderPixelInputInfo& info,
+                                      std::span<const uint32_t> active_inputs, uint32_t input) {
+	std::array<bool, 32> used_locations {};
+	for (const auto active_input: active_inputs) {
+		auto location = ShaderPixelParameterMappedLocation(info, active_input);
+		if (location < used_locations.size() && used_locations[location]) {
+			location = active_input;
+			while (location < used_locations.size() && used_locations[location]) {
+				location++;
+			}
+			EXIT_NOT_IMPLEMENTED(location >= used_locations.size());
+		}
+
+		if (active_input == input) {
+			return location;
+		}
+		used_locations[location] = true;
+	}
+	return ShaderPixelParameterMappedLocation(info, input);
+}
+
+bool ShaderPixelParameterIsFlat(const ShaderPixelInputInfo& info, uint32_t input) {
+	return input < info.input_num && (info.interpolator_settings[input] & PsInputFlatShade) != 0;
+}
+
 struct ShaderBinaryInfo {
 	uint8_t  signature[7];
 	uint8_t  version;
@@ -266,7 +302,7 @@ static void vs_check(const HW::VertexShaderInfo& vs, const HW::ShaderRegisters&
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.ieee_mode != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.cu_group_enable != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.require_forward_progress != false);
-		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.lds_configuration != false);
+		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.threadgroup_configuration != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.gs_vgpr_component_count != 3);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.fp16_overflow != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc2.scratch_en != false);
@@ -338,7 +374,7 @@ static void ps_check(const HW::PsStageRegisters& ps, const HW::ShaderRegisters&
 			     ps.rsrc2.extra_lds_size);
 		}
 	}
-	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != false);
+	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != 0);
 	EXIT_NOT_IMPLEMENTED(ps.rsrc2.shared_vgprs != 0);
 
 	if (sh.shader_z_format != 0x00000000 && sh.shader_z_format != 0x00000001 &&
@@ -1606,6 +1642,7 @@ ShaderId ShaderGetIdPS(const HW::PixelShaderInfo& regs, const ShaderPixelInputIn
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_pos_z));
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_pos_w));
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_front_face));
+	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_no_perspective));
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_pixel_kill_enable));
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_sample_mask_export_enable));
 	ret.ids.push_back(static_cast<uint32_t>(input_info.ps_early_z));

src/graphics/shader/shader.h

@@ -122,6 +122,11 @@ struct ShaderPixelInputInfo {
 	bool HasPositionInput() const { return ps_pos_x || ps_pos_y || ps_pos_z || ps_pos_w; }
 };
 
+uint32_t ShaderPixelParameterMappedLocation(const ShaderPixelInputInfo& info, uint32_t input);
+uint32_t ShaderPixelParameterLocation(const ShaderPixelInputInfo& info,
+                                      std::span<const uint32_t> active_inputs, uint32_t input);
+bool     ShaderPixelParameterIsFlat(const ShaderPixelInputInfo& info, uint32_t input);
+
 struct ShaderSharp {
 	uint16_t offset_dw : 15;
 	uint16_t size      : 1;