CI - enable LTO, remove OPT:NOREF

Co-authored-by: DeathWhispers <56944463+Death-Whispers-U@users.noreply.github.com>

.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 virtual_memory_allocation_tests --parallel
 
       - name: Test
         shell: cmd
         run: |
-          ctest --test-dir _Build/windows --output-on-failure -R "^virtual_memory_allocation$"
+          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|virtual_memory_allocation)$"
 
       - 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 virtual_memory_allocation_tests \
             --parallel
 
       - name: Test
         shell: bash
         run: |
           ctest --test-dir _Build/macos --output-on-failure \
-            -R '^virtual_memory_allocation$'
+            -R '^(audio_out2_port|virtual_memory_allocation)$'
 
       - 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 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)$'
+            -R '^(audio_out2_port|page_manager|memory_tracker|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -342,10 +374,12 @@ jobs:
   release:
     name: Release KytyPS5
     if: github.event_name == 'push' && github.repository == 'KytyPS5/KytyPS5'
-    needs: [windows, macos, linux]
+    needs: [metadata, windows, macos, linux]
     runs-on: ubuntu-latest
     permissions:
       contents: write
+    env:
+      RELEASE_NAME: ${{ needs.metadata.outputs.release_tag }}
 
     steps:
       - name: Download builds
@@ -353,11 +387,6 @@ jobs:
         with:
           path: artifacts
 
-      - name: Set release name
-        shell: bash
-        run: |
-          echo "RELEASE_NAME=KytyPS5-$(date -u +'%Y-%m-%d')-${GITHUB_SHA::7}" >> "$GITHUB_ENV"
-
       - name: Package builds
         shell: bash
         run: |

.gitignore

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

.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": {

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 xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version)
 
 # 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)
@@ -312,100 +347,116 @@ function(add_kyty_full_emulator_test target source)
 		target_link_libraries(${target} onecore)
 		add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${KYTY_THIRD_PARTY_DIR}/winpthread/bin/libwinpthread-1.dll" $<TARGET_FILE_DIR:${target}>/libwinpthread-1.dll)
 	endif()
+	# The macOS x86_64 guest address space needs its .zerofill segments anchored
+	# by linker flags, or the kernel kills the binary on load (posix_spawn EIO).
+	configure_macos_guest_address_space(${target})
 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 ${inc_headers})
+
 add_executable(memory_tracker_tests EXCLUDE_FROM_ALL
-	../tests/MemoryTrackerTests.cpp
-	graphics/host_gpu/pageManager.cpp
-	graphics/host_gpu/memoryTracker.cpp
+	"${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
+	"${KYTY_TESTS_DIR}/ResourceMutexTests.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/cache/resourceMutex.cpp"
 )
 target_link_libraries(resource_mutex_tests common)
 target_include_directories(resource_mutex_tests PRIVATE ${inc_headers})
 
+add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/AudioOut2PortTests.cpp"
+	"${KYTY_SOURCE_DIR}/libs/libAudio2.cpp"
+	"${KYTY_SOURCE_DIR}/loader/timer.cpp"
+)
+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(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 "${PROJECT_BINARY_DIR}/gpu_test_shaders")
 set(gpu_test_ms_depth_source
@@ -421,17 +472,16 @@ add_custom_command(
 	        -o "${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)
-configure_macos_guest_address_space(virtual_memory_allocation_tests)
 
 # These tests use exceptions.
 if(NOT KYTY_CLANG_CL)
@@ -442,12 +492,18 @@ if(NOT KYTY_CLANG_CL)
 endif()
 
 if(BUILD_TESTING)
+	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 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 resource_mutex COMMAND $<TARGET_FILE:resource_mutex_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_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>)
@@ -457,6 +513,8 @@ if(BUILD_TESTING)
 		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)
 
@@ -480,10 +538,26 @@ 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
+		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
+		resource_mutex_tests
+		event_queue_lifetime_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 ${kyty_emulator_link_libraries})
@@ -513,14 +587,17 @@ set(KYTY_EMULATOR_MAP_LINK_PATH "${CMAKE_CURRENT_BINARY_DIR}/${KYTY_EMULATOR_MAP
 set(KYTY_EMULATOR_PDB_LINK_PATH "${CMAKE_CURRENT_BINARY_DIR}/kyty_emulator.pdb")
 
 if(KYTY_CLANG_CL)
-	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "/DYNAMICBASE:NO /DEBUG:FULL /PDB:${KYTY_EMULATOR_PDB_LINK_PATH} /lldmap:${KYTY_EMULATOR_MAP_LINK_PATH}")
+	target_link_options(kyty_emulator PRIVATE
+		"/DYNAMICBASE:NO"
+		"/DEBUG:FULL"
+		"/PDB:${KYTY_EMULATOR_PDB_LINK_PATH}"
+		"/lldmap:${KYTY_EMULATOR_MAP_LINK_PATH}"
+	)
 	add_custom_command(TARGET kyty_emulator POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${KYTY_THIRD_PARTY_DIR}/winpthread/bin/libwinpthread-1.dll" $<TARGET_FILE_DIR:kyty_emulator>/libwinpthread-1.dll)
 elseif(WIN32 OR LINUX)
 	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${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
@@ -528,14 +605,14 @@ 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
 )

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 +97,12 @@ bool RenderDocEnabled() {
 	return g_config->renderdoc_enabled;
 }
 
-bool NggRectlistDrawEnabled() {
-	return g_config->ngg_rectlist_draw_enabled;
-}
-
 bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+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 +44,17 @@ 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;
+	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();
@@ -64,9 +75,10 @@ ProfilerDirection GetProfilerDirection();
 bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
-bool NggRectlistDrawEnabled();
 bool ReadbackLinearImagesEnabled();
 
+const Keymap& GetKeymap();
+
 } // namespace Config
 
 #endif /* KYTY_COMMON_EMULATOR_CONFIG_H_ */

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

@@ -1917,17 +1917,23 @@ KYTY_CP_OP_PARSER(CpOpCopyData) {
 
 	EXIT_NOT_IMPLEMENTED(cmd_id != KYTY_PM4(6, Pm4::IT_COPY_DATA, 0u));
 
-	const uint32_t control       = buffer[0];
-	const uint32_t src_sel       = ((control & 0xfu) << 1u) | ((control >> 30u) & 0x1u);
-	const uint32_t dst_sel       = ((control >> 8u) & 0xfu) << 1u;
-	const uint8_t  src_cache     = static_cast<uint8_t>((control >> 13u) & 0x3u);
-	const uint8_t  dst_cache     = static_cast<uint8_t>((control >> 25u) & 0x3u);
-	const uint8_t  write_confirm = static_cast<uint8_t>((control >> 20u) & 0x1u);
-	const uint32_t num_bytes     = ((control >> 16u) & 0x1u) != 0 ? 8u : 4u;
-	const uint64_t src           = buffer[1] | (static_cast<uint64_t>(buffer[2]) << 32u);
-	const uint64_t dst           = buffer[3] | (static_cast<uint64_t>(buffer[4]) << 32u);
-	if (src_sel == (9u << 1u)) {
-		if (dst_sel != (2u << 1u) || dst == 0 || (dst & (num_bytes - 1u)) != 0) {
+	const uint32_t control             = buffer[0];
+	const uint32_t src_sel             = ((control & 0xfu) << 1u) | ((control >> 30u) & 0x1u);
+	const uint32_t dst_sel             = ((control >> 8u) & 0xfu) << 1u;
+	const uint8_t  src_cache           = static_cast<uint8_t>((control >> 13u) & 0x3u);
+	const uint8_t  dst_cache           = static_cast<uint8_t>((control >> 25u) & 0x3u);
+	const uint8_t  write_confirm       = static_cast<uint8_t>((control >> 20u) & 0x1u);
+	const uint32_t num_bytes           = ((control >> 16u) & 0x1u) != 0 ? 8u : 4u;
+	const uint64_t src                 = buffer[1] | (static_cast<uint64_t>(buffer[2]) << 32u);
+	const uint64_t dst                 = buffer[3] | (static_cast<uint64_t>(buffer[4]) << 32u);
+	uint32_t       reference_clock_dst = 0;
+	switch (src_sel) {
+		case 9u: reference_clock_dst = 2u; break;
+		case 18u: reference_clock_dst = 4u; break;
+		default: break;
+	}
+	if (reference_clock_dst != 0) {
+		if (dst_sel != reference_clock_dst || dst == 0 || (dst & (num_bytes - 1u)) != 0) {
 			EXIT("unsupported reference-clock copyData, src_sel=0x%02" PRIx32
 			     " dst_sel=0x%02" PRIx32 " dst=0x%016" PRIx64 " size=%u\n",
 			     src_sel, dst_sel, dst, num_bytes);
@@ -2101,6 +2107,19 @@ KYTY_CP_OP_PARSER(CpOpClearState) {
 	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();
 
@@ -3390,6 +3409,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 	g_hw_ctx_indirect_func[Pm4::DB_COUNT_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
+	for (auto cmd_offset = Pm4::DB_SRESULTS_COMPARE_STATE0;
+	     cmd_offset <= Pm4::DB_SRESULTS_COMPARE_STATE1; cmd_offset++) {
+		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
+			HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		};
+	}
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};

src/graphics/guest_gpu/gpu_defs.h

@@ -72,6 +72,7 @@ enum class ChannelLayout : uint32_t {
 	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,

src/graphics/guest_gpu/gpu_format.cpp

@@ -39,6 +39,7 @@ constexpr FormatInfo kFormatInfo[] = {
     {GpuEnumValue(BufferFormat::k16_16Float), 4, 0, 4, true, false},
     {GpuEnumValue(BufferFormat::k11_11_10Float), 4, 0, 4, true, false},
     {GpuEnumValue(BufferFormat::k10_10_10_2UNorm), 4, 0, 4, true, false},
+    {GpuEnumValue(BufferFormat::k10_10_10_2UInt), 4, 0, 4, true, true},
     {GpuEnumValue(BufferFormat::k8_8_8_8UNorm), 4, 0, 4, true, false},
     {GpuEnumValue(BufferFormat::k8_8_8_8SNorm), 4, 0, 4, true, false},
     {GpuEnumValue(BufferFormat::k8_8_8_8UInt), 4, 0, 4, true, true},

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:
@@ -130,6 +127,7 @@ private:
 	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 +140,7 @@ 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 +155,8 @@ 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 +199,25 @@ 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 +232,6 @@ 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);
@@ -306,6 +312,8 @@ void CommandProcessor::Reset() {
 	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 +323,30 @@ 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);
 }
@@ -410,8 +442,13 @@ void CommandProcessor::WriteData(uint32_t* dst, const uint32_t* src, uint32_t dw
 	const uint32_t increment    = (write_control >> 16u) & 0x1u;
 	const uint32_t write_confirm = (write_control >> 20u) & 0x1u;
 
-	if (dst_sel != 0 && dst_sel != 2 && dst_sel != 4 && dst_sel != 5) {
-		EXIT("unsupported writeData destination selector 0x%02" PRIx32 "\n", dst_sel);
+	switch (dst_sel) {
+		case 0:
+		case 2:
+		case 4:
+		case 5:
+		case 6: break;
+		default: EXIT("unsupported writeData destination selector 0x%02" PRIx32 "\n", dst_sel);
 	}
 	EXIT_NOT_IMPLEMENTED(increment != 0);
 
@@ -507,6 +544,7 @@ 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;
@@ -527,6 +565,7 @@ 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;
@@ -558,6 +597,7 @@ void GpuState::ThreadRun(void* data) {
 		}
 		if (should_stop) {
 			gpu->m_gfx_cp->BufferWait();
+			g_gpu_state  = nullptr;
 			g_gpu_thread = false;
 			return;
 		}
@@ -691,26 +731,6 @@ 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");
@@ -759,6 +779,9 @@ 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) {
@@ -1001,8 +1024,9 @@ 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;
 	}
 
@@ -1042,6 +1066,7 @@ 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);
 }
@@ -1099,8 +1124,9 @@ 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;
 		}
 
@@ -1141,6 +1167,7 @@ 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);
 	}
@@ -1232,12 +1259,17 @@ 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);
 }
 
@@ -1659,10 +1691,6 @@ 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);
@@ -1688,46 +1716,8 @@ 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/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;
 }
 
@@ -110,7 +111,6 @@ void DumpPm4PacketStream(Common::File* file, uint32_t* cmd_buffer, uint32_t star
 	auto* cmd = cmd_buffer + start_dw;
 	auto  dw  = num_dw;
 	while (dw != 0) {
-		EXIT_NOT_IMPLEMENTED(dw < 2);
 		EXIT_NOT_IMPLEMENTED(dw > num_dw);
 
 		auto cmd_id = *cmd++;
@@ -120,6 +120,9 @@ void DumpPm4PacketStream(Common::File* file, uint32_t* cmd_buffer, uint32_t star
 		uint32_t len = 0;
 
 		const auto packet_type = static_cast<PacketType>(cmd_id >> 30u);
+		// Type-2 packets are header-only padding; every other packet type requires a body.
+		EXIT_NOT_IMPLEMENTED(dw < 2 && packet_type != PacketType::Type2);
+
 		switch (packet_type) {
 			case PacketType::Type3: {
 				const bool    sh_gx = (cmd_id & 0x2u) == 0;

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;
 

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 ||
@@ -104,34 +100,25 @@ void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 	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);
 	});
 }
 
@@ -154,10 +141,7 @@ 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->Untrack(manager->GetCpuAddr() + offset, bytes);
+		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
 	locks.clear();
 }
@@ -168,93 +152,4 @@ void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
 	UntrackMemoryLocked(vaddr, size);
 }
 
-bool MemoryTracker::InvalidateRegion(uint64_t vaddr, uint64_t size, PageFaultPhase phase) noexcept {
-	switch (phase) {
-		case PageFaultPhase::Release: return true;
-		case PageFaultPhase::Invalidate: {
-			const auto action = BeginCpuFault(vaddr, size);
-			switch (action) {
-				case CpuFaultAction::Untracked: return false;
-				case CpuFaultAction::Continue: return true;
-				case CpuFaultAction::Download:
-					EXIT("generic region invalidation cannot download GPU-dirty memory\n");
-			}
-		}
-		case PageFaultPhase::Complete:
-			return CompleteCpuFault(vaddr, size, PageFaultAccess::Write, false);
-	}
-	EXIT("unsupported region invalidation phase\n");
-}
-
-bool MemoryTracker::InvalidateVirtualGpuWrite(PageFaultAccess access, uint64_t vaddr, uint64_t size,
-                                              PageFaultPhase phase) noexcept {
-	switch (phase) {
-		case PageFaultPhase::Release: return true;
-		case PageFaultPhase::Invalidate: {
-			const bool gpu_modified = 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);
-			    });
-			if (!gpu_modified) {
-				return false;
-			}
-			const auto action = BeginCpuFault(vaddr, size);
-			if (access != PageFaultAccess::Write || action != CpuFaultAction::Download) {
-				EXIT("virtual GPU write fault requires write access to GPU-dirty memory\n");
-			}
-			return true;
-		}
-		case PageFaultPhase::Complete: {
-			if (access != PageFaultAccess::Write) {
-				EXIT("virtual GPU write completion requires write access\n");
-			}
-			bool completed = false;
-			Iterate<false>(
-			    vaddr, size, [&completed](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-				    std::scoped_lock lock(manager->lock);
-				    if (completed) {
-					    EXIT("virtual GPU write fault spans multiple tracked regions\n");
-				    }
-				    completed =
-				        manager->CompleteVirtualGpuWrite(manager->GetCpuAddr() + offset, bytes);
-			    });
-			return completed;
-		}
-	}
-	EXIT("unsupported virtual GPU write invalidation phase\n");
-}
-
-CpuFaultAction MemoryTracker::BeginCpuFault(uint64_t vaddr, uint64_t size,
-                                            PageFaultAccess access) noexcept {
-	CheckNotInUploadCallback();
-	CpuFaultAction action = CpuFaultAction::Untracked;
-	Iterate<false>(
-	    vaddr, size, [&action, access](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-		    std::scoped_lock lock(manager->lock);
-		    if (action != CpuFaultAction::Untracked) {
-			    EXIT("CPU fault spans multiple tracked regions\n");
-		    }
-		    action = manager->BeginCpuFault(manager->GetCpuAddr() + offset, bytes, access);
-	    });
-	return action;
-}
-
-bool MemoryTracker::CompleteCpuFault(uint64_t vaddr, uint64_t size, PageFaultAccess access,
-                                     bool downloaded) noexcept {
-	CheckNotInUploadCallback();
-	bool found = false;
-	Iterate<false>(
-	    vaddr, size,
-	    [&found, access, downloaded](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-		    std::scoped_lock lock(manager->lock);
-		    if (found) {
-			    EXIT("CPU fault completion spans multiple tracked regions\n");
-		    }
-		    found = manager->CompleteCpuFault(manager->GetCpuAddr() + offset, bytes, access,
-		                                      downloaded);
-	    });
-	return found;
-}
-
 } // 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);
@@ -30,60 +29,39 @@ 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);
-	[[nodiscard]] CpuFaultAction
-	                   BeginCpuFault(uint64_t vaddr, uint64_t size,
-	                                 PageFaultAccess access = PageFaultAccess::Write) noexcept;
-	[[nodiscard]] bool CompleteCpuFault(uint64_t vaddr, uint64_t size, PageFaultAccess access,
-	                                    bool downloaded) noexcept;
-	[[nodiscard]] bool InvalidateRegion(uint64_t vaddr, uint64_t size,
-	                                    PageFaultPhase phase) noexcept;
+	// 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");
-		}
+		});
 	}
-	[[nodiscard]] bool InvalidateVirtualGpuWrite(PageFaultAccess access, uint64_t vaddr,
-	                                             uint64_t size, PageFaultPhase phase) noexcept;
-	void               ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr, uint64_t size,
-	                                         const char* operation) const noexcept;
+#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) {
@@ -102,9 +80,6 @@ public:
 		}
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			const auto address = manager->GetCpuAddr() + offset;
-			if (manager->HasPendingFault(address, bytes)) {
-				EXIT("GPU download synchronization raced a pending CPU fault\n");
-			}
 			manager->template ForEachModifiedRange<DirtySource::Gpu, false>(address, bytes,
 			                                                                preflight);
 		});
@@ -116,10 +91,8 @@ 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 {});
 			               });
 		}
 	}
@@ -141,7 +114,6 @@ public:
 		const auto* previous_upload_owner = std::exchange(s_upload_owner, this);
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			manager->lock.lock();
-			manager->Track(manager->GetCpuAddr() + offset, bytes);
 			manager->ForEachModifiedRange<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset,
 			                                                      bytes, range_func);
 			if (!is_written) {
@@ -150,13 +122,12 @@ 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;
 	}
@@ -210,7 +181,6 @@ private:
 	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;
@@ -53,10 +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;
-
-thread_local bool g_in_fault_resolution = false;
 
 [[noreturn]] void FailFast(const char* reason = nullptr) noexcept {
 	std::fputs("PageManager fail-fast: ", stderr);
@@ -104,25 +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) {
@@ -156,47 +127,61 @@ 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;
-		bool     resolving_read_write = false;
-		bool     late_read_pending    = false;
-		bool     late_write_pending   = 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(PageFaultHandler handler, void* context): fault_handler(handler), fault_context(context) {
-		if (fault_handler == nullptr) {
-			Fatal("null page-manager fault callback");
-		}
+	Impl() {
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 		SYSTEM_INFO info {};
 		GetSystemInfo(&info);
@@ -223,10 +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");
 				}
 			}
@@ -254,459 +238,140 @@ 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 PublishDelayedFaults(PageState& page, uint32_t old_protection,
-	                                 uint32_t new_protection) {
-		if (old_protection == NO_ACCESS_PROTECTION && new_protection != NO_ACCESS_PROTECTION) {
-			page.late_read_pending = true;
-		}
-		if ((old_protection == NO_ACCESS_PROTECTION || old_protection == READ_ONLY_PROTECTION) &&
-		    new_protection == READ_WRITE_PROTECTION) {
-			page.late_write_pending = true;
-		}
-	}
-
-	static void InitializeProtection(std::span<PageState*> pages) {
-		for (auto* page: pages) {
-			page->original_protection = READ_WRITE_PROTECTION;
-			page->current_protection  = READ_WRITE_PROTECTION;
-		}
-	}
-
-	static bool AllowsAccess(const PageState& page, [[maybe_unused]] uint64_t vaddr,
-	                         PageFaultAccess access) noexcept {
-		switch (access) {
-			case PageFaultAccess::Read:
-				return page.current_protection == READ_ONLY_PROTECTION ||
-				       page.current_protection == READ_WRITE_PROTECTION;
-			case PageFaultAccess::Write: return page.current_protection == READ_WRITE_PROTECTION;
-			default: return false;
-		}
-	}
-
-	void ProtectRange(std::span<PageState*> pages, uint64_t vaddr, uint32_t protection,
-	                  std::span<const uint32_t> expected_old, bool fault_path) 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]) {
-				if (fault_path) {
-					FailFast("mprotect fault transition did not match expected protection");
-				}
-				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))) {
-			if (fault_path) {
-				FailFast("address-space fault protection transition failed");
-			}
 			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,
-	             bool fault_path) noexcept {
-		PageState* pages[]    = {&page};
-		uint32_t   expected[] = {expected_old};
-		ProtectRange(pages, vaddr, protection, expected, fault_path);
+	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;
 	std::vector<std::unique_ptr<Region>>    region_storage;
 	std::mutex                              region_mutex;
-	PageFaultHandler                        fault_handler = nullptr;
-	void*                                   fault_context = nullptr;
 };
 
 static_assert(std::atomic<void*>::is_always_lock_free);
 
-PageManager::PageManager(PageFaultHandler fault_handler, void* fault_context)
-    : m_impl(std::make_unique<Impl>(fault_handler, fault_context)) {}
+PageManager::PageManager(): m_impl(std::make_unique<Impl>()) {}
 
 PageManager::~PageManager() = default;
 
 uint64_t PageManager::GetPageSize() const {
-	if (g_in_fault_resolution) {
-		FailFast("nested page fault while resolving a watched page");
-	}
 	return PAGE_SIZE;
 }
 
-bool PageManager::IsTracked(uint64_t vaddr) const noexcept {
-	if (g_in_fault_resolution) {
-		FailFast("IsTracked called during fault resolution");
+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);
 	}
-	auto* region = m_impl->FindRegion(vaddr);
+
+	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) {
-		return false;
+		Fatal("untracking unknown region 0x%016" PRIx64, base_addr);
 	}
-	auto&     page = m_impl->GetPage(*region, vaddr);
-	SpinGuard lock(page.lock);
-	return page.write_watchers != 0 || page.access_watchers != 0;
+	m_impl->UpdateRegionWatchers<track, is_read, true>(*region, base_addr, first, last, &mask);
 }
 
-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), false);
-			first = current;
-		}
-
-		for (size_t i = 0; i < page_count; i++) {
-			auto&      page       = *pages[i];
-			const auto protection = new_protections[i];
-			if (track) {
-				switch (protection) {
-					case NO_ACCESS_PROTECTION:
-						page.late_read_pending  = false;
-						page.late_write_pending = false;
-						break;
-					case READ_ONLY_PROTECTION: page.late_write_pending = false; break;
-					default: break;
-				}
-			} else if (page.backing_writer == 0) {
-				Impl::PublishDelayedFaults(page, old_protections[i], protection);
-				if (page.write_watchers == 0 && page.access_watchers == 0) {
-					page.original_protection = 0;
-				}
-			}
-		}
-		chunk_begin = chunk_end;
-	}
-}
+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 {
-	if (g_in_fault_resolution) {
-		FailFast("backing write began during fault resolution");
-	}
-	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.resolving_read_write = true;
-		page.backing_writer       = writer;
-	}
-}
-
-void PageManager::EndBackingWrite(uint64_t vaddr, uint64_t size) noexcept {
-	if (g_in_fault_resolution) {
-		FailFast("backing write ended during fault resolution");
-	}
-	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, false);
-		}
-		Impl::PublishDelayedFaults(page, old_protection, new_protection);
-		if (page.write_watchers == 0 && page.access_watchers == 0) {
-			page.original_protection = 0;
-		}
-		page.backing_writer       = 0;
-		page.resolving            = false;
-		page.resolving_read_write = false;
-	}
-}
-
-bool PageManager::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
-	if (g_in_fault_resolution) {
-		FailFast("nested HandleFault call");
-	}
-	auto* region = m_impl->FindRegion(fault_vaddr);
-	if (region == nullptr) {
-		return false;
-	}
-	auto& page   = m_impl->GetPage(*region, fault_vaddr);
-	bool  waited = false;
-	while (true) {
-		SpinGuard lock(page.lock);
-		if (access == PageFaultAccess::Read && page.late_read_pending &&
-		    Impl::AllowsAccess(page, fault_vaddr, access)) {
-			page.late_read_pending = false;
-			return true;
-		}
-		if (access == PageFaultAccess::Write && page.late_write_pending &&
-		    Impl::AllowsAccess(page, fault_vaddr, access)) {
-			page.late_write_pending = false;
-			return true;
-		}
-		if (page.resolving) {
-			if (page.backing_writer == CurrentThread()) {
-				FailFast("backing writer faulted on its own reserved page");
-			}
-			if ((!page.resolving_read_write && access != PageFaultAccess::Write) ||
-			    (page.resolving_read_write && access != PageFaultAccess::Read &&
-			     access != PageFaultAccess::Write)) {
-				FailFast("fault access is incompatible with the active resolver");
-			}
-			waited = true;
-			continue;
-		}
-		if (page.write_watchers == 0 && page.access_watchers == 0) {
-			if (access != PageFaultAccess::Read && access != PageFaultAccess::Write) {
-				return false;
-			}
-			bool&      pending = (access == PageFaultAccess::Read ? page.late_read_pending
-			                                                      : page.late_write_pending);
-			const bool allowed = Impl::AllowsAccess(page, fault_vaddr, access);
-			pending            = false;
-			if (waited && !allowed) {
-				FailFast("page remained inaccessible after waiting for its resolver");
-			}
-			// More than one CPU can fault before a protection transition becomes visible. The first
-			// delayed fault consumes the hint bit; later faults must also resume once the mapped
-			// page already permits the requested access. A genuinely read-only/no-access page still
-			// falls through to the guest exception path.
-			return allowed;
-		}
-		if ((access != PageFaultAccess::Read && access != PageFaultAccess::Write) ||
-		    (access == PageFaultAccess::Read && page.access_watchers == 0)) {
-			FailFast("fault access is incompatible with active page watchers");
-		}
-		page.resolving            = true;
-		page.resolving_read_write = page.access_watchers != 0;
-		break;
-	}
-	g_in_fault_resolution = true;
-	const bool handled    = m_impl->fault_handler(m_impl->fault_context, access, fault_vaddr, 1,
-	                                              PageFaultPhase::Invalidate);
-	g_in_fault_resolution = false;
-	{
-		SpinGuard lock(page.lock);
-		if (!handled || !page.resolving) {
-			FailFast("fault invalidation did not preserve the resolving state");
-		}
-	}
-	g_in_fault_resolution = true;
-	const bool completed  = m_impl->fault_handler(m_impl->fault_context, access, fault_vaddr, 1,
-	                                              PageFaultPhase::Complete);
-	g_in_fault_resolution = false;
-	{
-		SpinGuard lock(page.lock);
-		if (!completed || !page.resolving) {
-			FailFast("fault completion did not preserve the resolving state");
-		}
-		if (page.write_watchers != 0 || page.access_watchers != 0) {
-			const auto old_protection  = Impl::WatcherProtection(page);
-			const bool read_only_fault = access == PageFaultAccess::Read;
-			if (read_only_fault && page.access_watchers == 0) {
-				FailFast("read fault completed without a read/write watcher");
-			}
-			page.access_watchers = 0;
-			if (!read_only_fault) {
-				page.write_watchers = 0;
-			}
-			const auto restored_protection = Impl::WatcherProtection(page);
-			m_impl->Protect(page, PageStart(fault_vaddr), restored_protection, old_protection,
-			                true);
-			if (page.write_watchers == 0) {
-				page.original_protection = 0;
-			}
-			Impl::PublishDelayedFaults(page, old_protection, restored_protection);
-		} else if (!Impl::AllowsAccess(page, fault_vaddr, access)) {
-			FailFast("fault completion left the page inaccessible");
-		}
-		page.resolving            = false;
-		page.resolving_read_write = false;
-	}
-	g_in_fault_resolution = true;
-	const bool released   = m_impl->fault_handler(m_impl->fault_context, access, fault_vaddr, 1,
-	                                              PageFaultPhase::Release);
-	g_in_fault_resolution = false;
-	if (!released) {
-		FailFast("fault release callback failed");
-	}
-	return true;
-}
-
 } // namespace Libs::Graphics

src/graphics/host_gpu/pageManager.h

@@ -2,56 +2,32 @@
 #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 PageFaultPhase { Invalidate, Complete, Release };
-enum class PageWatchMode { Write, ReadWrite };
 
-using PageFaultHandler = bool (*)(void* context, PageFaultAccess access, uint64_t vaddr,
-                                  uint64_t size, PageFaultPhase phase) noexcept;
 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(PageFaultHandler fault_handler, void* fault_context);
+	PageManager();
 	// The owner must stop all PageManager callers before destruction.
 	~PageManager();
 
 	KYTY_CLASS_NO_COPY(PageManager);
 
 	[[nodiscard]] uint64_t GetPageSize() const;
-	[[nodiscard]] bool     IsTracked(uint64_t vaddr) const noexcept;
 
-	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]] bool HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept;
-	[[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

@@ -25,8 +25,6 @@
 
 namespace Libs::Graphics {
 
-enum class CpuFaultAction { Untracked, Continue, Download };
-
 class TrackingSpinLock final {
 public:
 	void lock() noexcept {
@@ -78,229 +76,93 @@ 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);
 
 	[[nodiscard]] uint64_t GetCpuAddr() const { return m_cpu_addr; }
-	void                   Track(uint64_t vaddr, uint64_t size) {
-		const auto [start, end] = GetPageRange(vaddr, size);
-		for (auto page = start; page < end; page++) {
-			m_tracked.set(page);
-		}
-	}
-	void Untrack(uint64_t vaddr, uint64_t size) {
-		const auto [start, end] = GetPageRange(vaddr, size);
-		for (auto page = start; page < end; page++) {
-			m_tracked.reset(page);
-		}
-	}
 	template <DirtySource source>
 	[[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) || m_fault_pending.test(page)) {
-					EXIT("CPU dirty state conflicts with GPU dirty or pending fault 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) || m_fault_pending.test(page)) {
-					EXIT("GPU dirty state conflicts with CPU dirty or pending fault 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;
-	}
-
-	[[nodiscard]] CpuFaultAction BeginCpuFault(uint64_t vaddr, uint64_t size,
-	                                           PageFaultAccess access = PageFaultAccess::Write) {
-		if (access != PageFaultAccess::Read && access != PageFaultAccess::Write) {
-			EXIT("unsupported CPU fault access while beginning ownership transfer\n");
-		}
-		const auto [start, end] = GetPageRange(vaddr, size);
-		const bool tracked      = m_tracked.test(start);
-		for (auto page = start; page < end; page++) {
-			if (m_tracked.test(page) != tracked) {
-				EXIT("CPU fault spans mixed tracked and untracked pages\n");
-			}
-			if (m_fault_pending.test(page)) {
-				return CpuFaultAction::Untracked;
-			}
-			if (m_cpu_dirty.test(page) != m_writable.test(page) ||
-			    (m_gpu_dirty.test(page) && (m_cpu_dirty.test(page) || m_writable.test(page)))) {
-				EXIT("inconsistent CPU fault page state\n");
-			}
-		}
-		if (!tracked) {
-			return CpuFaultAction::Untracked;
-		}
-		bool gpu_dirty = m_gpu_dirty.test(start);
-		bool writable  = m_writable.test(start);
-		for (auto page = start + 1; page < end; page++) {
-			if (m_gpu_dirty.test(page) != gpu_dirty || m_writable.test(page) != writable) {
-				EXIT("CPU fault spans pages with incompatible dirty or writable state\n");
-			}
-		}
-		for (auto page = start; page < end; page++) {
-			if (!gpu_dirty && access == PageFaultAccess::Write) {
-				m_cpu_dirty.set(page);
-				m_writable.set(page);
-			}
-			m_fault_pending.set(page);
-		}
-		return gpu_dirty ? CpuFaultAction::Download : CpuFaultAction::Continue;
-	}
-
-	[[nodiscard]] bool CompleteCpuFault(uint64_t vaddr, uint64_t size, PageFaultAccess access,
-	                                    bool downloaded) {
-		const auto [start, end] = GetPageRange(vaddr, size);
-		for (auto page = start; page < end; page++) {
-			if (!m_fault_pending.test(page)) {
-				return false;
-			}
-		}
-		for (auto page = start; page < end; page++) {
-			const bool gpu_dirty = m_gpu_dirty.test(page);
-			if (gpu_dirty != downloaded) {
-				EXIT("CPU fault download result disagrees with GPU dirty state\n");
-			}
-			if (gpu_dirty) {
-				m_gpu_dirty.reset(page);
-				switch (access) {
-					case PageFaultAccess::Read: break;
-					case PageFaultAccess::Write:
-						m_cpu_dirty.set(page);
-						m_writable.set(page);
-						break;
-					default: EXIT("unsupported CPU fault access after GPU download\n");
-				}
-			}
-			m_fault_pending.reset(page);
-		}
-		return true;
-	}
-
-	[[nodiscard]] bool HasPendingFault(uint64_t vaddr, uint64_t size) const {
-		const auto [start, end] = GetPageRange(vaddr, size);
-		for (auto page = start; page < end; page++) {
-			if (m_fault_pending.test(page)) {
-				return true;
-			}
-		}
-		return false;
-	}
-
-	[[nodiscard]] bool CompleteVirtualGpuWrite(uint64_t vaddr, uint64_t size) {
-		const auto [start, end] = GetPageRange(vaddr, size);
-		for (auto page = start; page < end; page++) {
-			if (!m_fault_pending.test(page)) {
-				return false;
-			}
-			if (!m_gpu_dirty.test(page)) {
-				EXIT("virtual GPU write completion found a non-GPU-dirty page\n");
-			}
-		}
-		for (auto page = start; page < end; page++) {
-			m_gpu_dirty.reset(page);
-			m_cpu_dirty.set(page);
-			m_writable.set(page);
-			m_fault_pending.reset(page);
-		}
-		return true;
 	}
 
 	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>();
-		if constexpr (source == DirtySource::Cpu) {
-			mask &= ~m_fault_pending;
-		}
-		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) {
@@ -331,18 +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);
 		}
 	}
 
@@ -351,8 +203,7 @@ private:
 	RegionBits   m_cpu_dirty;
 	RegionBits   m_gpu_dirty;
 	RegionBits   m_writable;
-	RegionBits   m_fault_pending;
-	RegionBits   m_tracked;
+	RegionBits   m_readable;
 };
 
 } // namespace Libs::Graphics

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

@@ -3,15 +3,18 @@
 #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>
@@ -123,30 +126,6 @@ struct BufferCache::RetiredBuffer {
 	std::shared_ptr<Buffer> owner;
 };
 
-struct BufferCache::FaultReadback {
-	PageFaultAccess            access = PageFaultAccess::Unknown;
-	uint64_t                   vaddr  = 0;
-	uint64_t                   size   = 0;
-	std::vector<DownloadRange> ranges;
-	bool                       installed = false;
-
-	[[nodiscard]] bool Active() const noexcept { return !ranges.empty(); }
-
-	void Reset() {
-		access    = PageFaultAccess::Unknown;
-		vaddr     = 0;
-		size      = 0;
-		installed = false;
-		ranges.clear();
-	}
-};
-
-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) {
@@ -218,34 +197,26 @@ 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);
+		    {
+			    FaultSafeCacheLock lock(this, m_mutex);
+			    for (const auto& range: downloads) {
+				    m_gpu_modified_ranges.Subtract(range.address, range.size);
+			    }
+			    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
+			    // interval on those pages was downloaded and removed. Clearing the original
+			    // query therefore cannot orphan a dirty sibling on an edge page.
+			    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
+			    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
+			        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
+				    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
+			    }
+			    m_memory_tracker.UntrackMemory(retire.address, retire.size);
+		    }
+	    });
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
@@ -253,13 +224,12 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
                          ResourceMutex& resource_mutex)
     : m_graphics(graphics), m_scheduler(scheduler),
       m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
-      m_fault_readback(std::make_unique<FaultReadback>()), m_memory_tracker(page_manager),
+      m_memory_tracker(page_manager),
       m_staging_buffer(graphics, scheduler, MemoryUsage::Upload, 512 * MiB),
       m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
       m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
       m_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), m_resource_mutex(resource_mutex) {
 	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()) {
@@ -277,15 +247,9 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
 }
 
 BufferCache::~BufferCache() {
-	if (m_fault_readback->Active()) {
-		EXIT("BufferCache: destroyed with an active fault readback\n");
-	}
 	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)) {
@@ -295,68 +259,6 @@ 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;
@@ -385,16 +287,33 @@ 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);
+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);
+	}
+	if (m_resource_mutex.IsOwnedByCurrentThread()) {
+		EXIT("unsupported buffer readback from a pre-owned resource transaction, "
+		     "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) {
 	std::vector<DownloadCopy> copies;
 	{
 		FaultSafeCacheLock lock(this, m_mutex);
@@ -425,126 +344,40 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
 				    }
 			    }
 		    });
-	}
-	if (copies.empty()) {
-		return;
+		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);
 		}
-	}
-}
-
-bool BufferCache::InvalidateMemory(PageFaultAccess access, uint64_t vaddr, uint64_t size,
-                                   PageFaultPhase phase) noexcept {
-	const auto page = vaddr & ~(TRACKER_PAGE_SIZE - 1);
-	if (size == 0 || size > page + TRACKER_PAGE_SIZE - vaddr) {
-		EXIT("BufferCache: invalid page-fault range\n");
-	}
-
-	if (phase == PageFaultPhase::Complete) {
-		FaultSafeCacheLock lock(this, m_mutex);
-		auto&              fault = *m_fault_readback;
-		if (!fault.Active()) {
-			return m_memory_tracker.CompleteCpuFault(vaddr, size, access, false);
-		}
-		if (fault.access != access || fault.vaddr != vaddr || fault.size != size ||
-		    fault.installed) {
-			EXIT("BufferCache: mismatched fault readback completion\n");
-		}
-		PublishDownloads(fault.ranges);
-		if (!m_memory_tracker.CompleteCpuFault(vaddr, size, access, true)) {
-			EXIT("BufferCache: failed to complete downloaded CPU fault\n");
-		}
-		fault.installed = true;
-		return true;
-	}
-
-	if (phase == PageFaultPhase::Release) {
-		FaultSafeCacheLock lock(this, m_mutex);
-		auto&              fault = *m_fault_readback;
-		if (fault.Active()) {
-			if (fault.access != access || fault.vaddr != vaddr || fault.size != size ||
-			    !fault.installed) {
-				EXIT("BufferCache: mismatched fault readback release\n");
-			}
-			for (const auto& range: fault.ranges) {
-				m_gpu_modified_ranges.Subtract(range.address, range.size);
-			}
-			fault.Reset();
-		}
-		return true;
-	}
-
-	if (phase != PageFaultPhase::Invalidate) {
-		EXIT("BufferCache: unsupported page-fault phase\n");
-	}
-
-	const auto action = m_memory_tracker.BeginCpuFault(vaddr, size, access);
-	if (action != CpuFaultAction::Download) {
-		return action == CpuFaultAction::Continue;
-	}
-
-	auto&                     fault = *m_fault_readback;
-	std::vector<DownloadCopy> copies;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		if (fault.Active()) {
-			EXIT("BufferCache: nested fault readback\n");
-		}
-		fault.access = access;
-		fault.vaddr  = vaddr;
-		fault.size   = size;
-
-		m_gpu_modified_ranges.ForEachIntersection(
-		    page, TRACKER_PAGE_SIZE, [&](RangeSet::Range range) {
-			    auto owner = m_buffers.upper_bound(range.address);
-			    if (owner == m_buffers.begin()) {
-				    EXIT("BufferCache: fault readback has no buffer owner\n");
-			    }
-			    --owner;
-			    auto& cached = *owner->second;
-			    if (!cached.buffer->IsInBounds(range.address, range.size)) {
-				    EXIT("BufferCache: fault readback is outside its buffer owner\n");
-			    }
-			    copies.push_back({cached.buffer, cached.buffer->Offset(range.address),
-			                      range.address, range.size});
-		    });
-		if (copies.empty()) {
-			EXIT("BufferCache: GPU-dirty fault page has no dirty byte ranges\n");
+		// 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);
 		}
 	}
-	fault.ranges = RecordDownloads(copies);
-	if (!fault.Active()) {
-		EXIT("BufferCache: GPU-dirty fault page has no dirty byte ranges\n");
-	}
-	m_scheduler.FinishCurrent();
-	return true;
 }
 
 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;
+	std::vector<DownloadCopy>                  copies;
+	std::vector<std::pair<uint64_t, uint64_t>> modified_buffers;
+	std::vector<std::pair<uint64_t, uint64_t>> retired_buffers;
 	{
 		FaultSafeCacheLock lock(this, m_mutex);
 		for (const auto& [begin, cached]: m_buffers) {
@@ -561,12 +394,8 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t 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);
 		}
-		if (!dirty_ranges.empty()) {
-			backing_writes = m_page_manager.ReserveBackingWrites(dirty_ranges);
-		}
 		for (const auto& [begin, bytes]: modified_buffers) {
 			auto owner = m_buffers.find(begin);
 			if (owner == m_buffers.end() || owner->second->size != bytes) {
@@ -596,24 +425,11 @@ 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);
+			m_memory_tracker.UnmarkRegionAsGpuModified(begin, bytes);
 		}
 		for (const auto& [begin, bytes]: retired_buffers) {
 			m_memory_tracker.MarkRegionAsCpuModified(begin, bytes);
@@ -621,7 +437,6 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		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) {
@@ -715,20 +530,29 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 		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);
 	}
 
@@ -744,10 +568,12 @@ 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);
 	}
+	if (is_formatted && is_read && !is_written) {
+		(void)SynchronizeBufferFromImage(*cached.buffer, vaddr, size);
+	}
 	return {cached.buffer, cached.buffer->Handle(), cached.buffer->Offset(vaddr)};
 }
 
@@ -772,7 +598,6 @@ 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()) {
@@ -787,13 +612,12 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 		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         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());
@@ -843,42 +667,32 @@ 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;
@@ -892,42 +706,17 @@ 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);
@@ -943,45 +732,6 @@ 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) {
@@ -1041,17 +791,10 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	}
 	if (src_memory || dst_memory) {
 		std::lock_guard transaction(m_resource_mutex);
-		if (src_memory) {
-			(void)SynchronizeBacking(src_vaddr, size);
-		}
-		const auto src_region =
+		const auto      src_region =
 		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
 		const auto dst_region =
 		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
-		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) {
@@ -1073,7 +816,7 @@ 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};
@@ -1126,82 +869,13 @@ 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++;
 	if (m_graphics.CanReportMemoryUsage()) {
 		m_total_used_memory = m_graphics.GetDeviceMemoryUsage();
 	}
-	if (m_total_used_memory < m_trigger_gc_memory || m_fault_readback->Active()) {
+	if (m_total_used_memory < m_trigger_gc_memory) {
 		return;
 	}
 
@@ -1270,7 +944,6 @@ 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

@@ -10,7 +10,6 @@
 
 #include <map>
 #include <memory>
-#include <mutex>
 #include <span>
 #include <utility>
 #include <vector>
@@ -30,9 +29,8 @@ struct BufferBinding {
 };
 
 struct ImageBufferSource {
-	Buffer*  buffer    = nullptr;
-	uint64_t offset    = 0;
-	bool     gpu_owned = false;
+	Buffer*  buffer = nullptr;
+	uint64_t offset = 0;
 };
 
 class BufferCache {
@@ -47,11 +45,9 @@ public:
 	~BufferCache();
 	KYTY_CLASS_NO_COPY(BufferCache);
 
-	[[nodiscard]] bool InvalidateMemory(PageFaultAccess access, uint64_t vaddr, uint64_t size,
-	                                    PageFaultPhase phase) noexcept;
-	void               InvalidateMemory(uint64_t vaddr, uint64_t size);
-	void               ReadMemory(uint64_t vaddr, uint64_t size);
-	void               UnmapMemory(uint64_t vaddr, uint64_t size);
+	void                        InvalidateMemory(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,
 	                                         bool is_formatted = false);
@@ -62,9 +58,6 @@ 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);
@@ -72,11 +65,6 @@ public:
 	[[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:
@@ -90,8 +78,6 @@ private:
 	struct DownloadCopy;
 	struct DownloadRange;
 	struct RetiredBuffer;
-	struct FaultReadback;
-	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);
@@ -106,13 +92,12 @@ private:
 	            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;
@@ -120,17 +105,12 @@ private:
 	Common::Mutex                                     m_mutex;
 	std::shared_ptr<Buffer>                           m_null_buffer;
 	std::map<uint64_t, std::unique_ptr<CachedBuffer>> m_buffers;
-	std::unique_ptr<FaultReadback>                    m_fault_readback;
 	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;

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

@@ -1,107 +1,37 @@
 #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_page_manager(FaultThunk, this),
+    : m_scheduler(scheduler),
       m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache, m_resource_mutex),
       m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache, m_resource_mutex) {}
 
 GpuResourceManager::~GpuResourceManager() = default;
 
-bool GpuResourceManager::FaultThunk(void* context, PageFaultAccess access, uint64_t vaddr,
-                                    uint64_t size, PageFaultPhase phase) noexcept {
-	return static_cast<GpuResourceManager*>(context)->InvalidateMemory(access, vaddr, size, phase);
-}
-
-bool GpuResourceManager::InvalidateMemory(PageFaultAccess access, uint64_t vaddr, uint64_t size,
-                                          PageFaultPhase phase) noexcept {
-	// Let the authoritative image materialize first. A clean overlapping buffer marks a write
-	// fault CPU-dirty when it begins ownership transfer; doing that before image preflight would
-	// make the image appear to race a real CPU write. Completion and release retain buffer-first
-	// ordering so its pending fault is gone before TextureCache publishes the downloaded backing.
-	if (phase == PageFaultPhase::Invalidate) {
-		const bool image_handled  = m_texture_cache.InvalidateMemory(access, vaddr, size, phase);
-		const bool buffer_handled = m_buffer_cache.InvalidateMemory(access, vaddr, size, phase);
-		return buffer_handled || image_handled;
-	}
-	const bool buffer_handled = m_buffer_cache.InvalidateMemory(access, vaddr, size, phase);
-	const bool image_handled  = m_texture_cache.InvalidateMemory(access, vaddr, size, phase);
-	return buffer_handled || image_handled;
-}
-
 bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
 	constexpr uint64_t fault_size = 8;
 	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;
 }
 
@@ -123,7 +53,22 @@ 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);
+	}
+	if (m_resource_mutex.IsOwnedByCurrentThread()) {
+		EXIT("unsupported memory unmap from a pre-owned resource transaction, "
+		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
+		     vaddr, size);
+	}
 	const auto unmap = [this, vaddr, size] {
+		if (m_scheduler.Active()) {
+			const auto tick = m_scheduler.CurrentTick();
+			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);
@@ -131,13 +76,9 @@ 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

@@ -34,13 +34,9 @@ public:
 	void               RunGarbageCollector();
 
 private:
-	static bool FaultThunk(void* context, PageFaultAccess access, uint64_t vaddr, uint64_t size,
-	                       PageFaultPhase phase) noexcept;
-	[[nodiscard]] bool InvalidateMemory(PageFaultAccess access, uint64_t vaddr, uint64_t size,
-	                                    PageFaultPhase phase) noexcept;
-
 	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

@@ -4,18 +4,10 @@
 
 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");
+	if (m_resource_owner != std::thread::id {}) {
+		EXIT("ResourceMutex destroyed with an active owner\n");
 	}
 }
 
@@ -26,9 +18,6 @@ void ResourceMutex::lock() {
 		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);
@@ -55,39 +44,4 @@ bool ResourceMutex::IsOwnedByCurrentThread() {
 	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

@@ -8,21 +8,9 @@
 
 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.
+// Owner-tracked shared buffer/image 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);
@@ -32,15 +20,9 @@ public:
 	[[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

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

@@ -56,7 +56,10 @@ vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r) {
 			case Prospero::SamplerAnisoRatio::kFour: aniso_ratio = 4.0f; break;
 			case Prospero::SamplerAnisoRatio::kEight: aniso_ratio = 8.0f; break;
 			case Prospero::SamplerAnisoRatio::kSixteen: aniso_ratio = 16.0f; break;
-			default: EXIT("unknown ratio: %d\n", static_cast<int>(r.MaxAnisoRatio()));
+			default:
+				EXIT("unknown ratio: %d dwords=%08x,%08x,%08x,%08x\n",
+				     static_cast<int>(r.MaxAnisoRatio()), r.fields[0], r.fields[1], r.fields[2],
+				     r.fields[3]);
 		}
 	}
 

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

@@ -195,10 +195,12 @@ 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 +213,15 @@ 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 +331,7 @@ 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 +356,7 @@ 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 +374,7 @@ 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 +387,7 @@ 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 +408,7 @@ void TextureCache::UntrackImageHead(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, begin, size);
+		m_page_manager.UpdatePageWatchers<false>(begin, size);
 	}
 }
 
@@ -421,7 +429,7 @@ void TextureCache::UntrackImageTail(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, address, size);
+		m_page_manager.UpdatePageWatchers<false>(address, size);
 	}
 }
 
@@ -451,10 +459,48 @@ 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 +884,20 @@ 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;
@@ -867,22 +922,28 @@ 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;
 		}
@@ -913,6 +974,7 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 		                              info.resources.levels, plan.tiles)) {
 			return plan;
 		}
+		plan.linear_size = GetLinearSize(plan.tiles);
 	}
 	plan.valid = true;
 	return plan;
@@ -950,11 +1012,19 @@ 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.tile_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 +1108,20 @@ 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();
 	}
@@ -1131,12 +1196,11 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	}
 
 	ImageId result {};
-	bool    replacement_buffer = false;
-	bool    inserted_new       = false;
+	bool    inserted_new = false;
 	{
-		std::lock_guard            transaction(m_resource_mutex);
-		CacheLock                  lock(*this, m_lock);
-		const std::vector<ImageId> candidates =
+		std::lock_guard transaction(m_resource_mutex);
+		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 +1223,8 @@ 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;
@@ -1174,23 +1238,15 @@ 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();
 			}
 		}
@@ -1360,11 +1416,6 @@ 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();
@@ -1429,9 +1480,6 @@ 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 +1599,14 @@ 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 +1616,119 @@ 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) {
+			const auto binding = m_texture_cache.UploadBinding(image);
+			const auto format =
+			    binding == TextureCache::BindingType::RenderTarget
+			        ? ImageOps::RenderTargetTransferFormat(image.info.bytes_per_block)
+			        : image.info.guest_format;
+			color.tiles.clear();
+			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, format,
+			                              image.info.TransferLayers(), levels, color.tiles)) {
+				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 +1760,9 @@ 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 +1783,7 @@ 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 +1797,6 @@ 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()) {
@@ -1826,37 +1915,6 @@ bool TextureCache::TouchMeta(uint64_t address, uint32_t slice, bool is_clear) {
 	return true;
 }
 
-bool TextureCache::InvalidateMemory(PageFaultAccess access, uint64_t address, uint64_t size,
-                                    PageFaultPhase phase) noexcept {
-	if ((access != PageFaultAccess::Read && access != PageFaultAccess::Write) ||
-	    !GuestRange {address, size}.Valid()) {
-		return false;
-	}
-	if (access == PageFaultAccess::Read) {
-		return false;
-	}
-	if (phase == PageFaultPhase::Invalidate) {
-		CacheLock  lock(*this, m_lock);
-		const bool tracked =
-		    std::ranges::any_of(FindImagesInRegion(address, size, true), [&](ImageId id) {
-			    const auto owner = ResolveOwner(id);
-			    return owner != nullptr && !owner->depth_id && owner->IsTracked();
-		    });
-		if (tracked) {
-			InvalidateCpuAliases(address, size);
-		}
-		return tracked;
-	}
-	if (phase != PageFaultPhase::Complete && phase != PageFaultPhase::Release) {
-		return false;
-	}
-	CacheLock lock(*this, m_lock);
-	return std::ranges::any_of(FindImagesInRegion(address, size, true), [&](ImageId id) {
-		const auto owner = ResolveOwner(id);
-		return owner != nullptr && !owner->depth_id;
-	});
-}
-
 void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 	if (!GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid unmap range\n");

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

@@ -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);
@@ -76,11 +75,9 @@ public:
 	[[nodiscard]] bool ClearMeta(uint64_t address);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
-	[[nodiscard]] bool InvalidateMemory(PageFaultAccess access, uint64_t address, uint64_t size,
-	                                    PageFaultPhase phase) noexcept;
-	void               UnmapMemory(uint64_t address, uint64_t size);
-	void               ProcessDownloadImages();
-	void               RunGarbageCollector();
+	void UnmapMemory(uint64_t address, uint64_t size);
+	void ProcessDownloadImages();
+	void RunGarbageCollector();
 
 private:
 	enum class TransferDirection { Upload, Download };
@@ -102,7 +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;
@@ -128,8 +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,
@@ -142,7 +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);
@@ -157,7 +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);
@@ -173,7 +171,7 @@ private:
 	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;
@@ -181,11 +179,13 @@ 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, levels, rt.attrib3.tile_mode);
+		} else if (standard64) {
+			pitch = TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
+			                            width, levels, 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);
 		}
@@ -176,9 +192,19 @@ 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] {};
+	TileVolumeLayout volume_layout {};
+	uint64_t         backing_size = 0;
+	if (volume) {
+		if (!tile || !TileGetTextureVolumeLayout(transfer_format, width, height, depth, levels,
+		                                         rt.attrib3.tile_mode, 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) {
@@ -203,12 +229,6 @@ 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 +237,66 @@ 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) {
+			desc.info.mip_layout[level] = {
+			    volume_layout.level_offsets[level],
+			    volume_layout.level_sizes[level],
+			    volume_layout.level_widths[level],
+			    volume_layout.level_heights[level],
+			};
+			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/debug.cpp

@@ -359,30 +359,12 @@ 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) {
@@ -497,7 +479,15 @@ static void ZPrint(const char* func, const HW::DepthRenderTarget& z) {
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-static void ZCheck(const HW::DepthRenderTarget& z) {
+static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
+                   const HW::RenderControl& rc) {
+	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;
+	if (!depth_active && !stencil_active) {
+		return;
+	}
+
 	EXIT_NOT_IMPLEMENTED(!z.z_info.HasValidTextureCompatibility());
 	EXIT_NOT_IMPLEMENTED(!z.stencil_info.HasValidTextureCompatibility());
 	if (z.z_info.format == 0) {
@@ -1206,7 +1196,7 @@ void hw_check(const RenderCommandBuffer& buffer) {
 	log_phase("vp");
 	VpCheck(vp, smc);
 	log_phase("z");
-	ZCheck(z);
+	ZCheck(z, d, rc);
 	log_phase("clip");
 	ClipCheck(c);
 	log_phase("rc");

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -96,9 +96,11 @@ 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 +141,6 @@ 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 +156,8 @@ 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 +279,10 @@ 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) {
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
 		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||

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 ||

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

@@ -70,6 +70,10 @@ constexpr RenderTargetFormatMapping kRenderTargetFormats[] = {
      Prospero::ChannelType::kFloat,
      Prospero::ChannelOrder::kStandard,
      {vk::Format::eB10G11R11UfloatPack32, 4}},
+    {Prospero::ChannelLayout::k5_6_5,
+     Prospero::ChannelType::kUNorm,
+     Prospero::ChannelOrder::kStandard,
+     {vk::Format::eB5G6R5UnormPack16, 2}},
     {Prospero::ChannelLayout::k16,
      Prospero::ChannelType::kUNorm,
      Prospero::ChannelOrder::kStandard,
@@ -478,10 +482,10 @@ 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,
+bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
                               const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
                               uint64_t levels, std::vector<GpuTileInfo>& out_infos) {
-	if (size == 0 || levels == 0 || levels > 16 || depth == 0 ||
+	if (tiled_size == 0 || levels == 0 || levels > 16 || depth == 0 ||
 	    regions.size() != GetTextureRegionCount(depth, levels, layout.volume_texture) ||
 	    Prospero::IsFmaskTextureFormat(fmt)) {
 		return false;
@@ -534,8 +538,9 @@ bool TextureBuildGpuTileInfos(uint64_t size, const std::vector<vk::BufferImageCo
 				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,
+				if (!SetGpuTileSize(info.linear_offset, linear_span, UINT64_MAX,
+				                    info.linear_size) ||
+				    !SetGpuTileSize(info.tiled_offset, volume.level_sizes[level], tiled_size,
 				                    info.tiled_size)) {
 					return false;
 				}
@@ -584,8 +589,9 @@ 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      = TextureUploadSliceSourceOffset(layout, level, z);
-				if (!SetGpuTileSize(info.linear_offset, level_size.size, size, info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, GetLevelSrcSize(level_size), size,
+				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
+				                    info.linear_size) ||
+				    !SetGpuTileSize(info.tiled_offset, GetLevelSrcSize(level_size), tiled_size,
 				                    info.tiled_size)) {
 					return false;
 				}

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

@@ -44,7 +44,7 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
                                                          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,
+bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
                               const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
                               uint64_t levels, std::vector<GpuTileInfo>& infos);
 

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

@@ -270,7 +270,7 @@ bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, c
 	        supported_msaa_array) &&
 	       levels_ok && descriptor.MinLod() == 0 &&
 	       descriptor.TileMode() == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
-	       descriptor.BCSwizzle() == 0 && descriptor.MsaaDepth() == multisampled &&
+	       descriptor.BCSwizzle() == 0 && (!descriptor.MsaaDepth() || multisampled) &&
 	       pitch >= width && pitch == image.info.pitch;
 }
 
@@ -391,9 +391,8 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
-	const 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();
 }
@@ -424,10 +423,14 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	const bool encoding_ok   = IsSupportedStorageTextureEncoding(descriptor);
 	const bool uint_resource =
 	    resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint;
+	const bool raw_sint_storage =
+	    format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) && uint_resource &&
+	    resource.written && !resource.read && !resource.atomic;
 	const bool format_ok =
-	    Prospero::IsSupportedTextureFormat(format) &&
-	    uint_resource == Prospero::IsUintTextureFormat(format) &&
-	    (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt));
+	    raw_sint_storage ||
+	    (Prospero::IsSupportedTextureFormat(format) &&
+	     uint_resource == Prospero::IsUintTextureFormat(format) &&
+	     (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt)));
 	if (resource_ok && descriptor_ok && encoding_ok && format_ok && size != 0) {
 		return;
 	}
@@ -614,17 +617,25 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	const bool multisampled = IsMultisampledTexture(type);
 	const auto levels       = multisampled ? 1u : static_cast<uint32_t>(descriptor.MaxMip()) + 1u;
 	const auto tile         = descriptor.TileMode();
+	const bool depth_tile   = tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth);
 	const bool msaa_tile =
-	    tile == Prospero::GpuEnumValue(descriptor.MsaaDepth() ? Prospero::TileMode::kDepth
-	                                                          : Prospero::TileMode::kRenderTarget);
+	    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) ||
 	      (!msaa_array && (descriptor.Depth() != 0 || descriptor.BaseArray5() != 0))))) {
-		EXIT("unsupported texture mip view: base=%u last=%u levels=%u\n", base_level, last_level,
-		     levels);
+		EXIT("unsupported texture mip view: base=%u last=%u levels=%u max=%u type=%u tile=%u "
+		     "kind=%u dimension=%u mip_mode=%u read=%d written=%d "
+		     "dwords=%08x,%08x,%08x,%08x,%08x,%08x,%08x,%08x\n",
+		     base_level, last_level, levels, descriptor.MaxMip(), descriptor.Type(), tile,
+		     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]);
 	}
 	const auto samples = multisampled ? 1u << last_level : 1u;
 	const auto view_levels =
@@ -651,7 +662,8 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	TileSizeAlign size {};
 	if (multisampled) {
 		const auto bytes = Prospero::NumBytesPerElement(format);
-		pitch            = 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");
@@ -668,8 +680,11 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		ValidateStorageTexture(resource, descriptor, size.size);
 	}
 
-	const auto              pixel_format        = TextureGetFormat(format);
-	const auto              storage_view_format = SrgbStorageViewFormat(pixel_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);
 	const auto              view_format = storage && storage_view_format != vk::Format::eUndefined
 	                                          ? storage_view_format
 	                                          : pixel_format;

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/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];
@@ -929,10 +978,13 @@ 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 +1020,12 @@ 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/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"
@@ -650,11 +649,9 @@ 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 +664,10 @@ 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) {
@@ -767,7 +762,7 @@ 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 +786,7 @@ 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) {
@@ -900,12 +894,6 @@ 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 +901,14 @@ 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);
@@ -933,17 +919,6 @@ 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++) {
@@ -991,20 +966,6 @@ 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 +978,12 @@ 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 +1025,6 @@ 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 +1108,7 @@ 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 +1150,7 @@ 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 +1178,6 @@ 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 +1237,7 @@ 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 +1275,6 @@ 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 +1286,15 @@ 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 +1311,10 @@ 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/vulkanCommon.cpp

@@ -37,6 +37,7 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k16_16Float, vk::Format::eR16G16Sfloat},
     {Prospero::BufferFormat::k11_11_10Float, vk::Format::eB10G11R11UfloatPack32},
     {Prospero::BufferFormat::k10_10_10_2UNorm, vk::Format::eA2B10G10R10UnormPack32},
+    {Prospero::BufferFormat::k10_10_10_2UInt, vk::Format::eA2B10G10R10UintPack32},
     {Prospero::BufferFormat::k8_8_8_8UNorm, vk::Format::eR8G8B8A8Unorm},
     {Prospero::BufferFormat::k8_8_8_8SNorm, vk::Format::eR8G8B8A8Snorm},
     {Prospero::BufferFormat::k8_8_8_8UInt, vk::Format::eR8G8B8A8Uint},

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

@@ -789,12 +789,14 @@ 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;

src/graphics/presentation/window/window.cpp

@@ -32,7 +32,9 @@
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.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 +60,6 @@
 
 namespace Libs::Graphics {
 
-constexpr int KEYBOARD_CONTROLLER_ID = -1000;
-
 struct EventKeyboard {
 	bool     down;
 	bool     up;
@@ -72,26 +72,6 @@ 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 +240,11 @@ 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 +266,23 @@ 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) {
@@ -794,10 +786,14 @@ static void WindowCreate(WindowContext& context) {
 	if (SDL_InitSubSystem(SDL_INIT_VIDEO | SDL_INIT_GAMECONTROLLER) < 0) {
 		EXIT("%s\n", SDL_GetError());
 	}
+	HostInputInit();
 
 	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 +933,22 @@ 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
+
+#if defined(KYTY_OFFICIAL_BUILD)
+	static constexpr auto build_label = "Official build " KYTY_RELEASE_TAG;
+#elif defined(KYTY_FORK_BUILD)
+	static constexpr auto build_label = "Fork build " KYTY_BUILD_REPOSITORY " " KYTY_GIT_HASH;
+#else
+	static constexpr auto build_label = "Source build " KYTY_GIT_HASH;
+#endif
+
 	const auto now       = Common::Timer::QueryPerformanceCounter();
 	const auto frequency = Common::Timer::QueryPerformanceFrequency();
 	frame_num++;
@@ -948,11 +960,11 @@ 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}", 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/cfg/ShaderCFG.cpp

@@ -871,19 +871,19 @@ std::vector<uint32_t> DominatedBlocks(const Graph& graph, uint32_t header,
 	return blocks;
 }
 
-uint32_t AppendSyntheticMergeBlock(Graph& graph, uint32_t old_merge) {
-	const auto* merge = graph.FindBlock(old_merge);
+uint32_t AppendSyntheticBranchBlock(Graph& graph, uint32_t target) {
+	const auto* target_block = graph.FindBlock(target);
 
 	BasicBlock block;
 	block.id                    = static_cast<uint32_t>(graph.blocks.size());
-	block.start_pc              = merge != nullptr ? merge->start_pc : 0u;
+	block.start_pc              = target_block != nullptr ? target_block->start_pc : 0u;
 	block.end_pc                = block.start_pc;
-	block.inst_begin            = merge != nullptr ? merge->inst_begin : 0u;
+	block.inst_begin            = target_block != nullptr ? target_block->inst_begin : 0u;
 	block.inst_end              = block.inst_begin;
-	block.successors            = {old_merge};
+	block.successors            = {target};
 	block.terminator.kind       = TerminatorKind::Branch;
 	block.terminator.condition  = BranchCondition::Always;
-	block.terminator.true_block = old_merge;
+	block.terminator.true_block = target;
 	graph.blocks.push_back(std::move(block));
 	return graph.blocks.back().id;
 }
@@ -897,15 +897,50 @@ bool IsSyntheticMergeForwarder(const Graph& graph, uint32_t block_id, uint32_t m
 	       block->terminator.true_block == merge;
 }
 
-bool IsInsideLoopConstruct(const Graph& graph, const NaturalLoop& loop, uint32_t block_id) {
-	return block_id != UINT32_MAX && block_id != loop.merge && block_id != loop.continue_block &&
-	       graph.Dominates(loop.header, block_id) &&
-	       (loop.merge == UINT32_MAX || !graph.Dominates(loop.merge, block_id));
+const NaturalLoop* FindInnermostContainingLoop(const Graph& graph, uint32_t block_id) {
+	const NaturalLoop* innermost = nullptr;
+	for (const auto& loop: graph.natural_loops) {
+		if (Contains(loop.body_blocks, block_id) &&
+		    (innermost == nullptr || loop.body_blocks.size() < innermost->body_blocks.size())) {
+			innermost = &loop;
+		}
+	}
+	return innermost;
 }
 
-bool SelectionMergeLeavesContainingLoop(const Graph& graph, uint32_t header, uint32_t merge) {
+bool IsInsideLoopConstruct(const Graph& graph, const NaturalLoop& loop, uint32_t block_id) {
+	return block_id != UINT32_MAX && block_id != loop.merge && block_id != loop.continue_block &&
+	       graph.Dominates(loop.header, block_id) && !graph.Dominates(loop.merge, block_id);
+}
+
+bool IsInnermostLoopControlConditional(const Graph& graph, const BasicBlock& block) {
+	if (block.terminator.kind != TerminatorKind::ConditionalBranch) {
+		return false;
+	}
+	const auto* loop = FindInnermostContainingLoop(graph, block.id);
+	if (loop == nullptr || loop->merge == UINT32_MAX || loop->continue_block == UINT32_MAX) {
+		return false;
+	}
+	const auto true_target  = block.terminator.true_block;
+	const auto false_target = block.terminator.false_block;
+	if (block.id == loop->continue_block) {
+		const auto is_repeat_target = [&](uint32_t target) {
+			return target == loop->header || target == loop->merge;
+		};
+		return is_repeat_target(true_target) && is_repeat_target(false_target);
+	}
+	const auto is_control_target = [&](uint32_t target) {
+		return target == loop->merge || target == loop->continue_block;
+	};
+	return (is_control_target(true_target) &&
+	        (is_control_target(false_target) ||
+	         IsInsideLoopConstruct(graph, *loop, false_target))) ||
+	       (is_control_target(false_target) && IsInsideLoopConstruct(graph, *loop, true_target));
+}
+
+bool MergeLeavesContainingLoop(const Graph& graph, uint32_t header, uint32_t merge) {
 	for (const auto& loop: graph.natural_loops) {
-		if (IsInsideLoopConstruct(graph, loop, header) &&
+		if (loop.header != header && IsInsideLoopConstruct(graph, loop, header) &&
 		    !IsInsideLoopConstruct(graph, loop, merge)) {
 			return true;
 		}
@@ -913,6 +948,80 @@ bool SelectionMergeLeavesContainingLoop(const Graph& graph, uint32_t header, uin
 	return false;
 }
 
+bool CanonicalizeNaturalLoops(Graph& graph, std::string* error) {
+	const auto rewrite_budget = graph.blocks.size() * 2u + 16u;
+	for (size_t rewrite = 0; rewrite < rewrite_budget; rewrite++) {
+		bool changed = false;
+		for (const auto& loop: graph.natural_loops) {
+			std::vector<uint32_t> latches;
+			for (const auto& edge: graph.back_edges) {
+				if (edge.to == loop.header) {
+					AddUnique(latches, edge.from);
+				}
+			}
+			if (latches.size() <= 1u) {
+				continue;
+			}
+
+			const auto continue_block = AppendSyntheticBranchBlock(graph, loop.header);
+			for (auto latch: latches) {
+				auto* block = graph.FindBlock(latch);
+				if (block != nullptr) {
+					ReplaceValue(block->successors, loop.header, continue_block);
+					ReplaceTerminatorTarget(block->terminator, loop.header, continue_block);
+				}
+			}
+			RebuildPredecessors(graph);
+			RecomputeAnalyses(graph);
+			changed = true;
+			break;
+		}
+		if (changed) {
+			continue;
+		}
+
+		for (const auto& loop: graph.natural_loops) {
+			const auto* header                 = graph.FindBlock(loop.header);
+			const auto  is_loop_control_target = [&](uint32_t target) {
+				return target == loop.merge || target == loop.continue_block;
+			};
+			if (header == nullptr || header->terminator.kind != TerminatorKind::ConditionalBranch ||
+			    is_loop_control_target(header->terminator.true_block) ||
+			    is_loop_control_target(header->terminator.false_block) ||
+			    !Contains(loop.body_blocks, header->terminator.true_block) ||
+			    !Contains(loop.body_blocks, header->terminator.false_block)) {
+				continue;
+			}
+
+			const auto old_header   = loop.header;
+			const auto predecessors = header->predecessors;
+			const auto new_header   = AppendSyntheticBranchBlock(graph, old_header);
+			for (auto pred: predecessors) {
+				auto* block = graph.FindBlock(pred);
+				if (block != nullptr) {
+					ReplaceValue(block->successors, old_header, new_header);
+					ReplaceTerminatorTarget(block->terminator, old_header, new_header);
+				}
+			}
+			if (graph.entry_block == old_header) {
+				graph.entry_block = new_header;
+			}
+			MoveBlockBefore(graph, new_header, old_header);
+			RebuildPredecessors(graph);
+			RecomputeAnalyses(graph);
+			changed = true;
+			break;
+		}
+		if (!changed) {
+			return true;
+		}
+	}
+
+	SetFailure(graph, FailureKind::StructuredControlFlow, graph.entry_block,
+	           "CFG loop canonicalization exceeded rewrite budget", error);
+	return false;
+}
+
 bool SplitSharedMergeBlock(Graph& graph, uint32_t merge,
                            const std::vector<uint32_t>& construct_blocks,
                            bool                         force_split = false) {
@@ -948,7 +1057,7 @@ bool SplitSharedMergeBlock(Graph& graph, uint32_t merge,
 		return false;
 	}
 
-	const auto synthetic_merge = AppendSyntheticMergeBlock(graph, merge);
+	const auto synthetic_merge = AppendSyntheticBranchBlock(graph, merge);
 	auto*      synthetic_block = graph.FindBlock(synthetic_merge);
 	if (synthetic_block != nullptr) {
 		synthetic_block->predecessors = predecessors_to_split;
@@ -980,14 +1089,111 @@ bool SplitSharedMergeBlock(Graph& graph, uint32_t merge,
 bool SplitOneLoopMerge(Graph& graph) {
 	const auto& loops = graph.natural_loops;
 	for (const auto& loop: loops) {
-		if (SplitSharedMergeBlock(graph, loop.merge, loop.body_blocks)) {
+		const auto construct_blocks = DominatedBlocks(graph, loop.header, loop.merge);
+		const auto force_split      = MergeLeavesContainingLoop(graph, loop.header, loop.merge);
+		if (SplitSharedMergeBlock(graph, loop.merge, construct_blocks, force_split)) {
 			return true;
 		}
 	}
 	return false;
 }
 
-bool SplitOneSelectionMerge(Graph& graph) {
+std::vector<uint32_t> SelectionRegion(const Graph& graph, const BasicBlock& header,
+                                      uint32_t merge) {
+	std::vector<uint32_t> region;
+	std::vector<uint32_t> pending = {header.terminator.true_block,
+	                                 header.terminator.false_block};
+	while (!pending.empty()) {
+		const auto block_id = pending.back();
+		pending.pop_back();
+		if (block_id == merge || Contains(region, block_id)) {
+			continue;
+		}
+		const auto* block = graph.FindBlock(block_id);
+		if (block == nullptr) {
+			continue;
+		}
+		AddUnique(region, block_id);
+		pending.insert(pending.end(), block->successors.begin(), block->successors.end());
+	}
+	SortUnique(region);
+	return region;
+}
+
+bool DuplicateSelectionRegion(Graph& graph, uint32_t header_id, uint32_t merge,
+                              const std::vector<uint32_t>& region, uint32_t block_budget) {
+	std::vector<uint32_t> cloned_blocks;
+	for (auto block_id: region) {
+		if (!graph.Dominates(header_id, block_id)) {
+			cloned_blocks.push_back(block_id);
+		}
+	}
+	if (cloned_blocks.empty() || graph.FindBlock(header_id) == nullptr || header_id >= merge ||
+	    graph.blocks.size() + cloned_blocks.size() + 1u > block_budget) {
+		return false;
+	}
+
+	const auto first_clone = static_cast<uint32_t>(graph.blocks.size());
+	std::map<uint32_t, uint32_t> clones;
+	for (uint32_t i = 0; i < cloned_blocks.size(); i++) {
+		clones.emplace(cloned_blocks[i], first_clone + i);
+	}
+
+	for (auto block_id: cloned_blocks) {
+		BasicBlock clone = *graph.FindBlock(block_id);
+		clone.id         = clones.at(block_id);
+		clone.predecessors.clear();
+		clone.dominators.clear();
+		clone.post_dominators.clear();
+		graph.blocks.push_back(std::move(clone));
+	}
+
+	const auto remap_block = [&](BasicBlock& block) {
+		const auto remap_target = [&](uint32_t& target) {
+			if (const auto it = clones.find(target); it != clones.end()) {
+				target = it->second;
+			}
+		};
+		for (auto& successor: block.successors) {
+			remap_target(successor);
+		}
+		remap_target(block.terminator.true_block);
+		remap_target(block.terminator.false_block);
+		remap_target(block.terminator.merge_block);
+		remap_target(block.terminator.continue_block);
+		for (auto& target: block.terminator.indirect_targets) {
+			remap_target(target);
+		}
+	};
+	for (auto block_id: region) {
+		const auto owned_id = clones.contains(block_id) ? clones.at(block_id) : block_id;
+		remap_block(*graph.FindBlock(owned_id));
+	}
+
+	const auto private_merge = AppendSyntheticBranchBlock(graph, merge);
+	auto&      header        = *graph.FindBlock(header_id);
+	remap_block(header);
+
+	for (auto block_id: region) {
+		const auto owned_id = clones.contains(block_id) ? clones.at(block_id) : block_id;
+		auto*      block    = graph.FindBlock(owned_id);
+		if (block != nullptr) {
+			ReplaceValue(block->successors, merge, private_merge);
+			ReplaceTerminatorTarget(block->terminator, merge, private_merge);
+		}
+	}
+	ReplaceValue(header.successors, merge, private_merge);
+	ReplaceTerminatorTarget(header.terminator, merge, private_merge);
+
+	for (uint32_t i = 0; i <= cloned_blocks.size(); i++) {
+		MoveBlockBefore(graph, first_clone + i, merge + i);
+	}
+	RebuildPredecessors(graph);
+	RecomputeAnalyses(graph);
+	return true;
+}
+
+bool SplitOneSelectionMerge(Graph& graph, uint32_t block_budget) {
 	std::vector<uint32_t> loop_headers;
 	loop_headers.reserve(graph.natural_loops.size());
 	for (const auto& loop: graph.natural_loops) {
@@ -1001,11 +1207,26 @@ bool SplitOneSelectionMerge(Graph& graph) {
 		    Contains(loop_headers, block_id)) {
 			continue;
 		}
+		if (IsInnermostLoopControlConditional(graph, *block)) {
+			continue;
+		}
 
 		const auto merge = graph.FindNearestCommonPostDominator(block->terminator.true_block,
 		                                                        block->terminator.false_block);
+		if (merge == UINT32_MAX || graph.FindBlock(merge) == nullptr) {
+			continue;
+		}
+		const auto region = SelectionRegion(graph, *block, merge);
+		if (std::any_of(region.begin(), region.end(),
+		                [&](uint32_t member) { return !graph.Dominates(block_id, member); })) {
+			if (graph.natural_loops.empty() &&
+			    DuplicateSelectionRegion(graph, block_id, merge, region, block_budget)) {
+				return true;
+			}
+			continue;
+		}
 		const auto construct_blocks = DominatedBlocks(graph, block_id, merge);
-		const auto force_split      = SelectionMergeLeavesContainingLoop(graph, block_id, merge);
+		const auto force_split      = MergeLeavesContainingLoop(graph, block_id, merge);
 		if (SplitSharedMergeBlock(graph, merge, construct_blocks, force_split)) {
 			return true;
 		}
@@ -1015,10 +1236,12 @@ bool SplitOneSelectionMerge(Graph& graph) {
 
 bool SplitSharedMergeBlocks(Graph& graph, std::string* error) {
 	const auto original_block_count = static_cast<uint32_t>(graph.blocks.size());
-	const auto split_budget =
-	    std::max<uint32_t>(16u, std::min<uint32_t>(128u, original_block_count));
+	const auto split_budget = std::max<uint32_t>(
+	    16u, std::min<uint32_t>(128u, original_block_count * 4u));
+	const auto block_budget = std::max<uint32_t>(
+	    32u, std::min<uint32_t>(512u, original_block_count * 8u));
 	for (uint32_t splits = 0; splits < split_budget; splits++) {
-		if (!SplitOneLoopMerge(graph) && !SplitOneSelectionMerge(graph)) {
+		if (!SplitOneLoopMerge(graph) && !SplitOneSelectionMerge(graph, block_budget)) {
 			return true;
 		}
 		RebuildPredecessors(graph);
@@ -1353,6 +1576,9 @@ bool Structurize(Graph& graph, std::string* error) {
 		return false;
 	}
 
+	if (!CanonicalizeNaturalLoops(graph, error)) {
+		return false;
+	}
 	if (!SplitSharedMergeBlocks(graph, error)) {
 		return false;
 	}
@@ -1395,6 +1621,9 @@ bool Structurize(Graph& graph, std::string* error) {
 		    block.terminator.loop_header) {
 			continue;
 		}
+		if (IsInnermostLoopControlConditional(graph, block)) {
+			continue;
+		}
 
 		const auto merge = graph.FindNearestCommonPostDominator(block.terminator.true_block,
 		                                                        block.terminator.false_block);

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/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) {

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

@@ -425,9 +425,6 @@ 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;

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

@@ -47,8 +47,9 @@ bool NullImageDescriptor(const DescriptorValue& descriptor) {
 }
 
 bool ValidImageDescriptor(const DescriptorValue& descriptor) {
-	const auto type = static_cast<Prospero::ImageType>((descriptor.dwords[3] >> 28u) & 0xfu);
-	if (type < Prospero::ImageType::kColor1D) {
+	const auto type   = static_cast<Prospero::ImageType>((descriptor.dwords[3] >> 28u) & 0xfu);
+	const auto format = static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
+	if (type < Prospero::ImageType::kColor1D || format == Prospero::BufferFormat::kInvalid) {
 		return false;
 	}
 	if (type == Prospero::ImageType::kColor2DMsaa ||
@@ -200,7 +201,11 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 		    DescriptorIsCube(descriptor) != image.cube) {
 			if (error != nullptr) {
 				*error =
-				    fmt::format("image descriptor {} no longer matches specialized dimension", i);
+				    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;
 		}
@@ -215,14 +220,21 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 				}
 				return false;
 			}
-			const auto uint_descriptor =
-			    Prospero::IsUintTextureFormat((descriptor.dwords[1] >> 20u) & 0x1ffu);
-			const auto uint_program = image.kind == ResourceKind::ImageUint ||
-			                          image.kind == ResourceKind::StorageImageUint;
+			const auto format = (descriptor.dwords[1] >> 20u) & 0x1ffu;
+			const bool raw_sint_storage =
+			    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) &&
+			    !image.read && !image.atomic;
+			const bool uint_descriptor = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
+			const auto uint_program    = image.kind == ResourceKind::ImageUint ||
+			                             image.kind == ResourceKind::StorageImageUint;
 			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;
 			}
@@ -403,7 +415,14 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 		    image.kind == ResourceKind::StorageImageUint) {
 			image.storage_swizzle = DescriptorImageSwizzle(descriptor);
 		}
-		if (Prospero::IsUintTextureFormat((descriptor.dwords[1] >> 20u) & 0x1ffu)) {
+		const auto format  = (descriptor.dwords[1] >> 20u) & 0x1ffu;
+		const bool storage = image.kind == ResourceKind::StorageImage ||
+		                     image.kind == ResourceKind::StorageImageUint;
+		const bool raw_sint_storage =
+		    storage && format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32SInt) &&
+		    !image.read && !image.atomic;
+		const bool uint_image = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
+		if (uint_image) {
 			switch (image.kind) {
 				case ResourceKind::Image: image.kind = ResourceKind::ImageUint; break;
 				case ResourceKind::StorageImage: image.kind = ResourceKind::StorageImageUint; break;

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

@@ -45,6 +45,8 @@ namespace {
 
 constexpr uint32_t ScalarRegisters = 128;
 constexpr uint32_t VectorRegisters = 256;
+// Clamp X/Y/Z are consecutive three-bit fields; the high bit of each selects a border mode.
+constexpr uint32_t SamplerBorderClampMask = (1u << 2u) | (1u << 5u) | (1u << 8u);
 
 struct ScalarState {
 	std::array<uint32_t, ScalarRegisters> regs          = {};
@@ -647,6 +649,24 @@ private:
 		return AddDescriptor(descriptor);
 	}
 
+	uint32_t AddSamplerDescriptor(const ScalarState& state, uint32_t base) {
+		if (base >= ScalarRegisters || 4u > ScalarRegisters - base) {
+			return ScalarProvenance::Unknown;
+		}
+		DescriptorValue descriptor;
+		descriptor.dword_count = 4;
+		for (uint32_t i = 0; i < 4; i++) {
+			descriptor.dwords[i] = state.regs[base + i];
+		}
+		if (auto d0 = descriptor.dwords[0];
+		    d0 < m_graph.values.size() && m_graph.values[d0].op == ScalarValueOp::Constant &&
+		    (m_graph.values[d0].imm & SamplerBorderClampMask) == 0) {
+			// Without a border clamp, the border color and table index in dword 3 are unused.
+			descriptor.dwords[3] = Constant(0);
+		}
+		return AddDescriptor(descriptor);
+	}
+
 	uint32_t AddFlatAddressDescriptor(const Instruction& inst, const ScalarState& state) {
 		const uint32_t first = FlatStore(inst.op) ? 1u : 0u;
 		if (inst.src_count < first + 2u) {
@@ -704,7 +724,7 @@ private:
 			inst.memory.resource_source = AddDescriptor(state, inst.memory.resource * 4u, 8);
 			if (inst.op == Opcode::ImageSample || inst.op == Opcode::ImageGather4 ||
 			    inst.op == Opcode::ImageGetLod) {
-				inst.memory.sampler_source = AddDescriptor(state, inst.memory.sampler * 4u, 4);
+				inst.memory.sampler_source = AddSamplerDescriptor(state, inst.memory.sampler * 4u);
 			}
 		}
 	}

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;
@@ -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;

src/kernel/fileSystem.cpp

@@ -64,6 +64,7 @@ struct File {
 	std::filesystem::path               real_name;
 	std::atomic_bool                    opened;
 	std::atomic_bool                    directory;
+	std::atomic_bool                    writable;
 	std::atomic_bool                    append;
 	std::atomic_bool                    sync_writes;
 	SpecialFile                         special;
@@ -129,6 +130,7 @@ int FileDescriptors::CreateDescriptor() {
 	auto* file        = new File {};
 	file->opened      = false;
 	file->directory   = false;
+	file->writable    = false;
 	file->append      = false;
 	file->sync_writes = false;
 	file->special     = SpecialFile::None;
@@ -274,6 +276,10 @@ static std::filesystem::path ResolvePathIgnoringCase(const std::filesystem::path
 
 	return resolved;
 }
+#else
+static bool HasWindowsForbiddenFilenameCharacter(const std::string& relative_path) {
+	return relative_path.find_first_of("<>:\"|?*") != std::string::npos;
+}
 #endif
 
 std::filesystem::path MountPoints::GetRealFilename(const std::string& mounted_file_name) {
@@ -293,6 +299,10 @@ std::filesystem::path MountPoints::GetRealFilename(const std::string& mounted_fi
 			rel_path = Common::RemoveFirst(rel_path, 1);
 		}
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		if (HasWindowsForbiddenFilenameCharacter(rel_path)) {
+			::printf("FileSystem: Windows-incompatible guest filename: %s\n",
+			         mounted_file_name.c_str());
+		}
 		return p.dir / rel_path;
 #else
 		return ResolvePathIgnoringCase(p.dir / rel_path);
@@ -408,6 +418,7 @@ int KYTY_SYSV_ABI KernelOpen(const char* path, int flags, uint16_t mode) {
 	EXIT_IF(file == nullptr || file->opened || file->directory);
 
 	file->name        = path;
+	file->writable    = rw_mode != Common::File::Mode::Read;
 	file->append      = append;
 	file->sync_writes = fsync || sync || dsync;
 
@@ -953,6 +964,47 @@ int KYTY_SYSV_ABI KernelFstat(int d, FileStat* sb) {
 	return OK;
 }
 
+int KYTY_SYSV_ABI KernelFtruncate(int d, int64_t length) {
+	PRINT_NAME();
+
+	if (d < DESCRIPTOR_MIN) {
+		return KERNEL_ERROR_EBADF;
+	}
+
+	if (length < 0) {
+		return KERNEL_ERROR_EINVAL;
+	}
+
+	if (::Libs::Network::Net::IsSocket(d)) {
+		return KERNEL_ERROR_EINVAL;
+	}
+
+	auto* file = g_files->GetFile(d);
+
+	if (file == nullptr || !file->opened) {
+		return KERNEL_ERROR_EBADF;
+	}
+
+	if (!file->writable) {
+		return KERNEL_ERROR_EBADF;
+	}
+
+	if (file->directory || file->special != SpecialFile::None) {
+		return KERNEL_ERROR_EINVAL;
+	}
+
+	Common::LockGuard lock(file->mutex);
+
+	if (file->f.IsInvalid() || !file->f.Truncate(static_cast<uint64_t>(length))) {
+		return KERNEL_ERROR_EIO;
+	}
+
+	LOGF("\tFtruncate (size = %" PRId64 ") file: %s\n", length,
+	     Common::PathToString(file->real_name).c_str());
+
+	return OK;
+}
+
 int KYTY_SYSV_ABI KernelUnlink(const char* path) {
 	PRINT_NAME();
 

src/kernel/fileSystem.h

@@ -48,6 +48,7 @@ int64_t KYTY_SYSV_ABI KernelPwrite(int d, const void* buf, size_t nbytes, int64_
 int64_t KYTY_SYSV_ABI KernelLseek(int d, int64_t offset, int whence);
 int KYTY_SYSV_ABI     KernelStat(const char* path, FileStat* sb);
 int KYTY_SYSV_ABI     KernelFstat(int d, FileStat* sb);
+int KYTY_SYSV_ABI     KernelFtruncate(int d, int64_t length);
 int KYTY_SYSV_ABI     KernelUnlink(const char* path);
 int KYTY_SYSV_ABI     KernelRename(const char* from, const char* to);
 int KYTY_SYSV_ABI     KernelGetdirentries(int fd, char* buf, int nbytes, int64_t* basep);

src/kernel/memory.cpp

@@ -18,6 +18,7 @@
 #include <cstdlib>
 #include <cstring>
 #include <map>
+#include <memory>
 #include <mutex>
 #include <vector>
 
@@ -64,7 +65,7 @@ constexpr uint64_t PAGE_TABLE_POOL_SIZE   = 4ull * 1024ull * 1024ull * 1024ull;
 constexpr uint64_t PAGE_TABLE_GRANULARITY = 2ull * 1024ull * 1024ull;
 constexpr int      PAGE_TABLE_POOL_ENTRIES =
     static_cast<int>(PAGE_TABLE_POOL_SIZE / PAGE_TABLE_GRANULARITY);
-constexpr uint64_t DEFAULT_FLEXIBLE_MEMORY_SIZE = 4ull * 1024ull * 1024ull * 1024ull;
+constexpr uint64_t DEFAULT_FLEXIBLE_MEMORY_SIZE = 1ull * 1024ull * 1024ull * 1024ull;
 
 static uint64_t                      g_flexible_memory_size        = DEFAULT_FLEXIBLE_MEMORY_SIZE;
 static bool                          g_flexible_memory_size_frozen = false;
@@ -805,18 +806,23 @@ private:
 	Common::Mutex              m_mutex;
 };
 
-static PhysicalMemory*       g_physical_memory       = nullptr;
-static FlexibleMemory*       g_flexible_memory       = nullptr;
-static PooledMemory*         g_pooled_memory         = nullptr;
-static VirtualRanges*        g_virtual_ranges        = nullptr;
-static GuestAddressSpace*    g_guest_address_space   = nullptr;
-static callback_func_t       g_alloc_callback        = nullptr;
-static callback_func_t       g_free_callback         = nullptr;
-static std::atomic<uint64_t> g_memory_pool_committed = 0;
-static void                  MemoryPoolSubtractCommitted(uint64_t len);
+static std::unique_ptr<PhysicalMemory>    g_physical_memory;
+static std::unique_ptr<FlexibleMemory>    g_flexible_memory;
+static std::unique_ptr<PooledMemory>      g_pooled_memory;
+static std::unique_ptr<VirtualRanges>     g_virtual_ranges;
+static std::unique_ptr<GuestAddressSpace> g_guest_address_space;
+static callback_func_t                    g_alloc_callback        = nullptr;
+static callback_func_t                    g_free_callback         = nullptr;
+static std::atomic<uint64_t>              g_memory_pool_committed = 0;
+static void                               MemoryPoolSubtractCommitted(uint64_t len);
 // Keep host mappings, physical blocks, placeholders, and virtual ranges in step.
 static std::recursive_mutex g_memory_operation_mutex;
 
+// The base address the PS5 kernel hands out for hint-less user mappings. Guest code can
+// assume mappings it did not place explicitly are at or above this (Sony's libc rejects a
+// heap below it), so hint-less searches must not fall back to the low system-managed range.
+static constexpr uint64_t GUEST_DEFAULT_MAP_BASE = 0x200000000ull;
+
 static uint64_t FindGuestFreeRange(uint64_t search_addr, uint64_t size, uint64_t alignment) {
 	EXIT_IF(g_guest_address_space == nullptr || g_virtual_ranges == nullptr);
 
@@ -844,8 +850,11 @@ static uint64_t FindGuestFreeRange(uint64_t search_addr, uint64_t size, uint64_t
 	if (search_addr != 0) {
 		return find_in(search_addr, HOST_USER_MAX + 1u);
 	}
-	auto addr = find_in(HOST_SYSTEM_MANAGED_MIN, HOST_SYSTEM_MANAGED_MAX + 1u);
-	return addr != 0 ? addr : find_in(HOST_USER_MIN, HOST_USER_MAX + 1u);
+	auto addr = find_in(GUEST_DEFAULT_MAP_BASE, HOST_SYSTEM_MANAGED_MAX + 1u);
+	if (addr == 0) {
+		addr = find_in(HOST_USER_MIN, HOST_USER_MAX + 1u);
+	}
+	return addr;
 }
 
 bool TryWriteBacking(uint64_t vaddr, const void* data, uint64_t size) {
@@ -970,11 +979,11 @@ static bool ReplaceFixedRangeWithReserved(uint64_t start, uint64_t size);
 KYTY_SUBSYSTEM_INIT(Memory) {
 	g_flexible_memory_size_frozen = true;
 	VirtualMemory::Init();
-	g_guest_address_space = new GuestAddressSpace(PhysicalMemory::TotalSize());
-	g_physical_memory     = new PhysicalMemory;
-	g_flexible_memory     = new FlexibleMemory;
-	g_pooled_memory       = new PooledMemory;
-	g_virtual_ranges      = new VirtualRanges;
+	g_guest_address_space = std::make_unique<GuestAddressSpace>(PhysicalMemory::TotalSize());
+	g_physical_memory     = std::make_unique<PhysicalMemory>();
+	g_flexible_memory     = std::make_unique<FlexibleMemory>();
+	g_pooled_memory       = std::make_unique<PooledMemory>();
+	g_virtual_ranges      = std::make_unique<VirtualRanges>();
 	EXIT_IF(!g_guest_address_space->SelfTest());
 	EXIT_IF(!SelfTestSub64SharedPlaceholderAlias());
 }
@@ -982,16 +991,11 @@ KYTY_SUBSYSTEM_INIT(Memory) {
 KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Memory) {}
 
 KYTY_SUBSYSTEM_DESTROY(Memory) {
-	delete g_pooled_memory;
-	g_pooled_memory = nullptr;
-	delete g_flexible_memory;
-	g_flexible_memory = nullptr;
-	delete g_physical_memory;
-	g_physical_memory = nullptr;
-	delete g_virtual_ranges;
-	g_virtual_ranges = nullptr;
-	delete g_guest_address_space;
-	g_guest_address_space = nullptr;
+	g_pooled_memory.reset();
+	g_flexible_memory.reset();
+	g_physical_memory.reset();
+	g_virtual_ranges.reset();
+	g_guest_address_space.reset();
 }
 
 struct AlignedPos {

src/kernel/memoryAddressSpace.inc

@@ -1058,6 +1058,76 @@ private:
 		    {HOST_SYSTEM_RESERVED_MIN, HOST_SYSTEM_RESERVED_MAX + 1u},
 		    {HOST_USER_MIN, HOST_USER_MAX + 1u},
 		}};
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+		std::vector<std::pair<uint64_t, uint64_t>> occupied;
+		FILE*                                      maps = fopen("/proc/self/maps", "r");
+		EXIT_IF(maps == nullptr);
+
+		char line[512];
+		while (fgets(line, sizeof(line), maps) != nullptr) {
+			unsigned long long mapping_start = 0;
+			unsigned long long mapping_end   = 0;
+			if (sscanf(line, "%llx-%llx", &mapping_start, &mapping_end) == 2) {
+				occupied.emplace_back(static_cast<uint64_t>(mapping_start),
+				                      static_cast<uint64_t>(mapping_end));
+			}
+		}
+		fclose(maps);
+
+		auto reserve_range = [this](uint64_t start, uint64_t end) {
+			start = AlignUp(start, PageSize());
+			end   = AlignDown(end, PageSize());
+			if (start == 0 || end <= start) {
+				return;
+			}
+
+			const auto size = end - start;
+			int        flags = MAP_PRIVATE | MAP_ANON | MAP_NORESERVE;
+#if defined(KYTY_LINKED_GUEST_ADDRESS_SPACE)
+			flags |= MAP_FIXED;
+#elif defined(MAP_FIXED_NOREPLACE)
+			flags |= MAP_FIXED_NOREPLACE;
+#endif
+
+			void* ptr = mmap(reinterpret_cast<void*>(start), size, PROT_NONE, flags, -1, 0);
+			if (ptr == MAP_FAILED || reinterpret_cast<uint64_t>(ptr) != start) {
+				if (ptr != MAP_FAILED) {
+					munmap(ptr, size);
+				}
+				return;
+			}
+
+			AddFreeUnlocked(start, size);
+			m_owned.emplace_back(start, size);
+		};
+
+		for (const auto& [region_start, region_end]: regions) {
+			auto current = region_start;
+
+			for (const auto& [mapping_start, mapping_end]: occupied) {
+				if (mapping_end <= current) {
+					continue;
+				}
+				if (mapping_start >= region_end) {
+					break;
+				}
+
+				if (mapping_start > current) {
+					reserve_range(current, std::min(mapping_start, region_end));
+				}
+
+				current = std::max(current, mapping_end);
+				if (current >= region_end) {
+					break;
+				}
+			}
+
+			if (current < region_end) {
+				reserve_range(current, region_end);
+			}
+		}
+#else
 		for (const auto& [start, end]: regions) {
 			int flags = MAP_PRIVATE | MAP_ANON | MAP_NORESERVE;
 #if defined(KYTY_LINKED_GUEST_ADDRESS_SPACE)
@@ -1070,13 +1140,14 @@ private:
 				if (ptr != MAP_FAILED) {
 					munmap(ptr, end - start);
 				}
-				EXIT("failed to reserve guest address space at 0x%016" PRIx64 ", size 0x%016" PRIx64
-				     "\n",
+				EXIT("failed to reserve guest address space at 0x%016" PRIx64
+				     ", size 0x%016" PRIx64 "\n",
 				     start, end - start);
 			}
 			AddFreeUnlocked(start, end - start);
 			m_owned.emplace_back(start, end - start);
 		}
+#endif
 #endif
 	}
 

src/kernel/pthread.cpp

@@ -65,6 +65,10 @@
 
 namespace Libs {
 
+namespace LibcInternalExt {
+void RunThreadAtexitDestructors();
+} // namespace LibcInternalExt
+
 namespace LibKernel {
 
 LIB_NAME("libkernel", "libkernel");
@@ -361,6 +365,7 @@ struct PthreadAttrPrivate {
 	uint64_t       stack_map_addr;
 	size_t         stack_map_size;
 	int            policy;
+	int            guest_priority;
 	int            inherit_sched;
 	int            solosched;
 	bool           detached;
@@ -871,9 +876,11 @@ bool TestGuestStackOwnerLifecycle(uint64_t* first_address, uint64_t* second_addr
 
 static KYTY_SYSV_ABI void* RunOnGuestStack(void* arg, pthread_entry_func_t func, void* stack_top) {
 #if defined(__x86_64__) || defined(_M_X64)
-	void*      ret       = nullptr;
-	const auto guest_rsp = reinterpret_cast<uintptr_t>(stack_top) & ~static_cast<uintptr_t>(0x0f);
-	const auto guest_rbp = guest_rsp - 4u * sizeof(uint64_t);
+	void*      ret = nullptr;
+	const auto aligned_stack_top =
+	    reinterpret_cast<uintptr_t>(stack_top) & ~static_cast<uintptr_t>(0x0f);
+	const auto guest_rsp = aligned_stack_top - 2u * sizeof(uintptr_t);
+	const auto guest_rbp = guest_rsp;
 
 	auto* guest_root_frame = reinterpret_cast<uintptr_t*>(guest_rbp);
 	guest_root_frame[0]    = 0;
@@ -2124,13 +2131,8 @@ int KYTY_SYSV_ABI PthreadAttrGetschedparam(const PthreadAttr* attr, KernelSchedP
 
 	int result = pthread_attr_getschedparam(&(*attr)->p, param);
 
-	if (param->sched_priority <= -2) {
-		param->sched_priority = 767;
-	} else if (param->sched_priority >= +2) {
-		param->sched_priority = 256;
-	} else {
-		param->sched_priority = 700;
-	}
+	// Host priority mapping is lossy; return the exact guest value.
+	param->sched_priority = (*attr)->guest_priority;
 
 	if (result == 0) {
 		return OK;
@@ -2294,6 +2296,7 @@ int KYTY_SYSV_ABI PthreadAttrSetschedparam(PthreadAttr* attr, const KernelSchedP
 		return KERNEL_ERROR_EINVAL;
 	}
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	KernelSchedParam pparam {};
 	if (param->sched_priority <= 478) {
 		pparam.sched_priority = +2;
@@ -2303,12 +2306,13 @@ int KYTY_SYSV_ABI PthreadAttrSetschedparam(PthreadAttr* attr, const KernelSchedP
 		pparam.sched_priority = 0;
 	}
 
-	int result = pthread_attr_setschedparam(&attr_value->p, &pparam);
-
-	if (result == 0) {
-		return OK;
+	if (pthread_attr_setschedparam(&attr_value->p, &pparam) != 0) {
+		return KERNEL_ERROR_EINVAL;
 	}
-	return KERNEL_ERROR_EINVAL;
+#endif
+
+	attr_value->guest_priority = param->sched_priority;
+	return OK;
 }
 
 int KYTY_SYSV_ABI PthreadAttrSetschedpolicy(PthreadAttr* attr, int policy) {
@@ -3347,6 +3351,8 @@ int PthreadGetCurrentPriorityForKernel() {
 static void CleanupThread(void* arg) {
 	auto* thread = static_cast<Pthread>(arg);
 
+	LibcInternalExt::RunThreadAtexitDestructors();
+
 	auto thread_dtors = g_pthread_context->GetThreadDtors();
 
 	if (thread_dtors != nullptr) {
@@ -3633,26 +3639,15 @@ int KYTY_SYSV_ABI PthreadGetprio(Pthread thread, int* prio) {
 
 	EXIT_NOT_IMPLEMENTED(prio == nullptr);
 
-	sched_param param {};
-	int         pol = 0;
-
-	int result = pthread_getschedparam(thread->p, &pol, &param);
-
-	if (result == 0) {
-		if (param.sched_priority <= -2) {
-			*prio = 767;
-		} else if (param.sched_priority >= +2) {
-			*prio = 256;
-		} else {
-			*prio = 700;
-		}
-
-		LOGF("\t PthreadGetprio: %d, %d\n", thread->unique_id, *prio);
-
-		return OK;
+	sched_param native_param {};
+	int         native_policy = 0;
+	if (pthread_getschedparam(thread->p, &native_policy, &native_param) != 0) {
+		return KERNEL_ERROR_EINVAL;
 	}
 
-	return KERNEL_ERROR_EINVAL;
+	*prio = thread->attr->guest_priority;
+	LOGF("\t PthreadGetprio: %d, %d\n", thread->unique_id, *prio);
+	return OK;
 }
 
 int KYTY_SYSV_ABI PthreadSetprio(Pthread thread, int prio) {
@@ -3667,25 +3662,27 @@ int KYTY_SYSV_ABI PthreadSetprio(Pthread thread, int prio) {
 
 	int result = pthread_getschedparam(thread->p, &pol, &param);
 
-	if (result == 0) {
-		if (prio <= 478) {
-			param.sched_priority = +2;
-		} else if (prio >= 733) {
-			param.sched_priority = -2;
-		} else {
-			param.sched_priority = 0;
-		}
-
-		result = pthread_setschedparam(thread->p, pol, &param);
-
-		if (result == 0) {
-			LOGF("\t PthreadSetprio: %d, %d\n", thread->unique_id, prio);
-
-			return OK;
-		}
+	if (result != 0) {
+		return KERNEL_ERROR_EINVAL;
 	}
 
-	return KERNEL_ERROR_EINVAL;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	if (prio <= 478) {
+		param.sched_priority = +2;
+	} else if (prio >= 733) {
+		param.sched_priority = -2;
+	} else {
+		param.sched_priority = 0;
+	}
+
+	if (pthread_setschedparam(thread->p, pol, &param) != 0) {
+		return KERNEL_ERROR_EINVAL;
+	}
+#endif
+
+	thread->attr->guest_priority = prio;
+	LOGF("\t PthreadSetprio: %d, %d\n", thread->unique_id, prio);
+	return OK;
 }
 
 void KYTY_SYSV_ABI PthreadTestcancel() {
@@ -3880,8 +3877,12 @@ int KYTY_SYSV_ABI KernelGettimeofday(KernelTimeval* tp) {
 	tp->tv_sec  = static_cast<int64_t>(ticks / 1000000);
 	tp->tv_usec = static_cast<int64_t>(ticks % 1000000);
 #else
-	auto dt = Common::DateTime::FromSystemUTC();
-	sec_to_timeval(tp, dt.ToUnix());
+	struct timespec ts {};
+	result = ::clock_gettime(CLOCK_REALTIME, &ts);
+	if (result == 0) {
+		tp->tv_sec  = static_cast<int64_t>(ts.tv_sec);
+		tp->tv_usec = static_cast<int64_t>(ts.tv_nsec / 1000);
+	}
 #endif
 
 	if (result == 0) {

src/kytyGitVersion.h.in

@@ -1 +1,2 @@
-#define KYTY_GIT_VERSION "@KYTY_GIT_VERSION@"
+#define KYTY_GIT_VERSION "@KYTY_GIT_VERSION@"
+#define KYTY_GIT_HASH "@KYTY_GIT_HASH@"

src/launcher/CMakeLists.txt

@@ -16,6 +16,7 @@ file(GLOB launcher_forms
 add_executable(launcher WIN32
 	${launcher_src}
 	${launcher_forms}
+	resources.qrc
 )
 
 
@@ -37,7 +38,6 @@ elseif(LINUX)
 	set_target_properties(launcher PROPERTIES LINK_FLAGS "${KYTY_LD_OPTIONS}")
 endif()
 
-add_dependencies(launcher KytyGitVersion)  
 add_dependencies(launcher kyty_emulator)
 
 if(KYTY_CLANG_CL)
@@ -141,14 +141,14 @@ elseif(LINUX)
 endif()
 
 
-set(tidy_dirs "${CMAKE_SOURCE_DIR}/launcher/include") 
-set(iwyu_maps "${CMAKE_SOURCE_DIR}/launcher/utils/qt6_16.imp")
+set(tidy_dirs "${CMAKE_CURRENT_SOURCE_DIR}/include")
+set(iwyu_maps "${CMAKE_CURRENT_SOURCE_DIR}/utils/qt6_16.imp")
 
 get_property(inc_headers TARGET launcher PROPERTY INCLUDE_DIRECTORIES)
 
 list(APPEND inc_headers
-	${CMAKE_SOURCE_DIR}/launcher
-	${CMAKE_BINARY_DIR}/launcher/launcher_autogen/include
+	${CMAKE_CURRENT_SOURCE_DIR}
+	${CMAKE_CURRENT_BINARY_DIR}/launcher_autogen/include
 	${Qt6Widgets_INCLUDE_DIRS}
 	${Qt6Core_INCLUDE_DIRS}
 	${Qt6Gui_INCLUDE_DIRS}

src/launcher/forms/configuration_edit_dialog.ui

@@ -7,7 +7,7 @@
     <x>0</x>
     <y>0</y>
     <width>560</width>
-    <height>420</height>
+    <height>450</height>
    </rect>
   </property>
   <property name="windowTitle">
@@ -78,16 +78,13 @@
            </property>
           </widget>
          </item>
-         <item row="2" column="0" colspan="2">
-          <widget class="QCheckBox" name="checkBox_ngg_rectlist_draw">
+         <item row="2" column="0">
+          <widget class="QCheckBox" name="checkBox_fullscreen">
            <property name="toolTip">
-            <string>Use the NGG 4-vertex path for rect-list DrawIndexAuto primitive 7</string>
+            <string>Run in borderless desktop fullscreen</string>
            </property>
            <property name="text">
-            <string>Use NGG rect-list draw</string>
-           </property>
-           <property name="checked">
-            <bool>true</bool>
+            <string>Fullscreen</string>
            </property>
           </widget>
          </item>
@@ -132,41 +129,55 @@
        </widget>
       </item>
       <item row="3" column="0">
+       <widget class="QLabel" name="label_console_language">
+        <property name="text">
+         <string>Console language:</string>
+        </property>
+       </widget>
+      </item>
+      <item row="3" column="1">
+       <widget class="QComboBox" name="comboBox_console_language">
+        <property name="toolTip">
+         <string>Language</string>
+        </property>
+       </widget>
+      </item>
+      <item row="4" column="0">
        <widget class="QLabel" name="label_20">
         <property name="text">
          <string>Shader optimization type:</string>
         </property>
        </widget>
       </item>
-      <item row="3" column="1">
+      <item row="4" column="1">
        <widget class="QComboBox" name="comboBox_shader_optimization_type">
         <property name="toolTip">
          <string>Optimize shaders for code size or performance</string>
         </property>
        </widget>
       </item>
-      <item row="4" column="0">
+      <item row="5" column="0">
        <widget class="QLabel" name="label_21">
         <property name="text">
          <string>Shader log direction:</string>
         </property>
        </widget>
       </item>
-      <item row="4" column="1">
+      <item row="5" column="1">
        <widget class="QComboBox" name="comboBox_shader_log_direction">
         <property name="toolTip">
          <string>Dump shaders to file or console window. If enabled may decrease emulator performance</string>
         </property>
        </widget>
       </item>
-      <item row="5" column="0">
+      <item row="6" column="0">
        <widget class="QLabel" name="label_22">
         <property name="text">
          <string>Shader log folder:</string>
         </property>
        </widget>
       </item>
-      <item row="5" column="1">
+      <item row="6" column="1">
        <widget class="MandatoryLineEdit" name="lineEdit_shader_log_folder">
         <property name="toolTip">
          <string>Specify directory to dump shaders</string>
@@ -176,14 +187,14 @@
         </property>
        </widget>
       </item>
-      <item row="6" column="0">
+      <item row="7" column="0">
        <widget class="QLabel" name="label_24">
         <property name="text">
          <string>Command buffer dump folder:</string>
         </property>
        </widget>
       </item>
-      <item row="6" column="1">
+      <item row="7" column="1">
        <widget class="MandatoryLineEdit" name="lineEdit_cmd_dump_folder">
         <property name="toolTip">
          <string>Specify directory to dump command buffers</string>
@@ -193,28 +204,28 @@
         </property>
        </widget>
       </item>
-      <item row="7" column="0">
+      <item row="8" column="0">
        <widget class="QLabel" name="label_25">
         <property name="text">
          <string>Printf direction:</string>
         </property>
        </widget>
       </item>
-      <item row="7" column="1">
+      <item row="8" column="1">
        <widget class="QComboBox" name="comboBox_printf_direction">
         <property name="toolTip">
          <string>Print logs to file or console window. If enabled may decrease emulator performance</string>
         </property>
        </widget>
       </item>
-      <item row="8" column="0">
+      <item row="9" column="0">
        <widget class="QLabel" name="label_26">
         <property name="text">
          <string>Printf output file:</string>
         </property>
        </widget>
       </item>
-      <item row="8" column="1">
+      <item row="9" column="1">
        <widget class="MandatoryLineEdit" name="lineEdit_printf_file">
         <property name="toolTip">
          <string>Specify file to dump logs</string>
@@ -224,14 +235,14 @@
         </property>
        </widget>
       </item>
-      <item row="9" column="0">
+      <item row="10" column="0">
        <widget class="QLabel" name="label_27">
         <property name="text">
          <string>Profiler direction:</string>
         </property>
        </widget>
       </item>
-      <item row="9" column="1">
+      <item row="10" column="1">
        <widget class="QComboBox" name="comboBox_profiler_direction">
         <property name="toolTip">
          <string>Enable or disable profiler. If enabled may decrease emulator performance</string>

src/launcher/forms/configuration_list_widget.ui

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

src/launcher/icons/edit-configuration.svg

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

src/launcher/icons/global-settings.svg

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

src/launcher/icons/input-mapping.svg

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

src/launcher/icons/remove-configuration.svg

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

src/launcher/include/configuration.h

@@ -48,6 +48,9 @@ class Configuration: public QObject {
 	Q_OBJECT
 
 public:
+	static constexpr int DEFAULT_CONSOLE_LANGUAGE = 1;
+	static constexpr int MAX_CONSOLE_LANGUAGE     = 29;
+
 	enum class Resolution {
 		R1280X720,
 		R1920X1080,
@@ -82,7 +85,9 @@ public:
 	QString    game_comment;
 
 	Resolution             screen_resolution           = Resolution::R1280X720;
+	bool                   fullscreen_enabled          = false;
 	int                    vblank_frequency            = 60;
+	int                    console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = true;
 	bool                   shader_validation_enabled   = true;
 	ShaderOptimizationType shader_optimization_type    = ShaderOptimizationType::Performance;
@@ -94,13 +99,15 @@ public:
 	QString                printf_output_file          = "_kyty.txt";
 	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   renderdoc_enabled           = false;
-	bool                   ngg_rectlist_draw_enabled   = true;
+	QStringList            host_input_mapping;
 
 	QString elf = QStringLiteral("eboot.bin");
 
 	void CopyEmulatorSettingsFrom(const Configuration& other) {
 		screen_resolution           = other.screen_resolution;
+		fullscreen_enabled          = other.fullscreen_enabled;
 		vblank_frequency            = other.vblank_frequency;
+		console_language            = other.console_language;
 		vulkan_validation_enabled   = other.vulkan_validation_enabled;
 		shader_validation_enabled   = other.shader_validation_enabled;
 		shader_optimization_type    = other.shader_optimization_type;
@@ -112,7 +119,7 @@ public:
 		printf_output_file          = other.printf_output_file;
 		profiler_direction          = other.profiler_direction;
 		renderdoc_enabled           = other.renderdoc_enabled;
-		ngg_rectlist_draw_enabled   = other.ngg_rectlist_draw_enabled;
+		host_input_mapping          = other.host_input_mapping;
 	}
 
 	void CopyFrom(const Configuration& other) {
@@ -135,7 +142,9 @@ public:
 		KYTY_CFG_SET(game_path);
 		KYTY_CFG_SET(custom_settings);
 		KYTY_CFG_SET(screen_resolution);
+		KYTY_CFG_SET(fullscreen_enabled);
 		KYTY_CFG_SET(vblank_frequency);
+		KYTY_CFG_SET(console_language);
 		KYTY_CFG_SET(vulkan_validation_enabled);
 		KYTY_CFG_SET(shader_validation_enabled);
 		KYTY_CFG_SET(shader_optimization_type);
@@ -147,7 +156,7 @@ public:
 		KYTY_CFG_SET(printf_output_file);
 		KYTY_CFG_SET(profiler_direction);
 		KYTY_CFG_SET(renderdoc_enabled);
-		KYTY_CFG_SET(ngg_rectlist_draw_enabled);
+		s->setValue("host_input_mapping", host_input_mapping);
 		KYTY_CFG_SET(elf);
 	}
 
@@ -157,7 +166,12 @@ public:
 		KYTY_CFG_GET(game_path);
 		KYTY_CFG_GET(custom_settings);
 		KYTY_CFG_GET(screen_resolution);
+		KYTY_CFG_GET(fullscreen_enabled);
 		vblank_frequency = s->value("vblank_frequency", vblank_frequency).toInt();
+		console_language = s->value("console_language", console_language).toInt();
+		if (console_language < 0 || console_language > MAX_CONSOLE_LANGUAGE) {
+			console_language = DEFAULT_CONSOLE_LANGUAGE;
+		}
 		KYTY_CFG_GET(vulkan_validation_enabled);
 		KYTY_CFG_GET(shader_validation_enabled);
 		KYTY_CFG_GET(shader_optimization_type);
@@ -169,9 +183,8 @@ public:
 		KYTY_CFG_GET(printf_output_file);
 		KYTY_CFG_GET(profiler_direction);
 		KYTY_CFG_GET(renderdoc_enabled);
-		ngg_rectlist_draw_enabled =
-		    s->value("ngg_rectlist_draw_enabled", ngg_rectlist_draw_enabled).toBool();
-		elf = s->value("elf", elf).toString();
+		host_input_mapping = s->value("host_input_mapping", host_input_mapping).toStringList();
+		elf                = s->value("elf", elf).toString();
 	}
 };
 

src/launcher/include/configurationListWidget.h

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

src/launcher/include/inputMappingDialog.h

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

src/launcher/resources.qrc

@@ -0,0 +1,8 @@
+<RCC>
+ <qresource prefix="/icons">
+  <file alias="edit-configuration.svg">icons/edit-configuration.svg</file>
+  <file alias="global-settings.svg">icons/global-settings.svg</file>
+  <file alias="input-mapping.svg">icons/input-mapping.svg</file>
+  <file alias="remove-configuration.svg">icons/remove-configuration.svg</file>
+ </qresource>
+</RCC>

src/launcher/src/configurationEditDialog.cpp

@@ -30,6 +30,39 @@ constexpr char SETTINGS_CFG_DIALOG[]               = "ConfigurationEditDialog";
 constexpr char SETTINGS_CFG_LAST_GEOMETRY[]        = "geometry";
 constexpr int  GLOBAL_SETTINGS_GAME_DIRS_MIN_WIDTH = 560;
 
+const QStringList CONSOLE_LANGUAGE_NAMES = {
+    "Japanese",
+    "English (United States)",
+    "French (France)",
+    "Spanish (Spain)",
+    "German",
+    "Italian",
+    "Dutch",
+    "Portuguese (Portugal)",
+    "Russian",
+    "Korean",
+    "Chinese (Traditional)",
+    "Chinese (Simplified)",
+    "Finnish",
+    "Swedish",
+    "Danish",
+    "Norwegian",
+    "Polish",
+    "Portuguese (Brazil)",
+    "English (United Kingdom)",
+    "Turkish",
+    "Spanish (Latin America)",
+    "Arabic",
+    "French (Canada)",
+    "Czech",
+    "Hungarian",
+    "Greek",
+    "Romanian",
+    "Thai",
+    "Vietnamese",
+    "Indonesian",
+};
+
 static QString NormalizeGameDirectory(const QString& dir) {
 	const auto trimmed = dir.trimmed();
 	if (trimmed.isEmpty()) {
@@ -120,11 +153,17 @@ static void ListInit(QComboBox* combo, T value) {
 
 void ConfigurationEditDialog::Init(const Configuration& info) {
 	ListInit(m_ui->comboBox_screen_resolution, info.screen_resolution);
+	m_ui->checkBox_fullscreen->setChecked(info.fullscreen_enabled);
 	m_ui->spinBox_vblank_frequency->setValue(info.vblank_frequency);
+	m_ui->comboBox_console_language->clear();
+	m_ui->comboBox_console_language->addItems(CONSOLE_LANGUAGE_NAMES);
+	m_ui->comboBox_console_language->setCurrentIndex(
+	    info.console_language >= 0 && info.console_language < CONSOLE_LANGUAGE_NAMES.size()
+	        ? info.console_language
+	        : Configuration::DEFAULT_CONSOLE_LANGUAGE);
 	m_ui->checkBox_shader_validation->setChecked(info.shader_validation_enabled);
 	m_ui->checkBox_vulkan_validation->setChecked(info.vulkan_validation_enabled);
 	m_ui->checkBox_renderdoc_capture->setChecked(info.renderdoc_enabled);
-	m_ui->checkBox_ngg_rectlist_draw->setChecked(info.ngg_rectlist_draw_enabled);
 	ListInit(m_ui->comboBox_shader_optimization_type, info.shader_optimization_type);
 	ListInit(m_ui->comboBox_shader_log_direction, info.shader_log_direction);
 	m_ui->lineEdit_shader_log_folder->setText(info.shader_log_folder);
@@ -241,11 +280,12 @@ void ConfigurationEditDialog::moveEvent(QMoveEvent* event) {
 static void UpdateInfo(Configuration& info, Ui::ConfigurationEditDialog& ui) {
 	info.screen_resolution =
 	    TextToEnum<Configuration::Resolution>(ui.comboBox_screen_resolution->currentText());
+	info.fullscreen_enabled        = ui.checkBox_fullscreen->isChecked();
 	info.vblank_frequency          = ui.spinBox_vblank_frequency->value();
+	info.console_language          = ui.comboBox_console_language->currentIndex();
 	info.vulkan_validation_enabled = ui.checkBox_vulkan_validation->isChecked();
 	info.shader_validation_enabled = ui.checkBox_shader_validation->isChecked();
 	info.renderdoc_enabled         = ui.checkBox_renderdoc_capture->isChecked();
-	info.ngg_rectlist_draw_enabled = ui.checkBox_ngg_rectlist_draw->isChecked();
 	info.shader_optimization_type  = TextToEnum<Configuration::ShaderOptimizationType>(
 	    ui.comboBox_shader_optimization_type->currentText());
 	info.shader_log_direction = TextToEnum<Configuration::ShaderLogDirection>(

src/launcher/src/configurationListWidget.cpp

@@ -6,6 +6,7 @@
 #include "configurationEditDialog.h"
 #include "configurationItem.h"
 #include "gameListTreeWidget.h"
+#include "inputMappingDialog.h"
 #include "mainDialog.h"
 #include "patchesDialog.h"
 #include "trophyViewerDialog.h"
@@ -169,6 +170,7 @@ static void ApplyGameListStyle(Ui::ConfigurationListWidget* ui) {
 	    "QToolButton:disabled { background: transparent; }"));
 	ui->edit_button->setStyleSheet(ui->global_settings_button->styleSheet());
 	ui->delete_button->setStyleSheet(ui->global_settings_button->styleSheet());
+	ui->input_mapping_button->setStyleSheet(ui->global_settings_button->styleSheet());
 }
 
 static void AddSaveDataDir(QStringList* dirs, QSet<QString>* seen, const QString& root,
@@ -226,10 +228,12 @@ ConfigurationListWidget::ConfigurationListWidget(QWidget* parent)
 	m_ui->setupUi(this);
 	ApplyGameListStyle(m_ui);
 
-	m_ui->global_settings_button->setIcon(style()->standardIcon(QStyle::SP_FileDialogDetailedView));
+	m_ui->global_settings_button->setIcon(QIcon(QStringLiteral(":/icons/global-settings.svg")));
 	m_ui->global_settings_button->setToolTip(tr("Edit global settings and game folders"));
-	m_ui->edit_button->setIcon(style()->standardIcon(QStyle::SP_FileIcon));
-	m_ui->delete_button->setIcon(style()->standardIcon(QStyle::SP_DialogDiscardButton));
+	m_ui->input_mapping_button->setIcon(QIcon(QStringLiteral(":/icons/input-mapping.svg")));
+	m_ui->input_mapping_button->setToolTip(tr("Edit global input mapping"));
+	m_ui->edit_button->setIcon(QIcon(QStringLiteral(":/icons/edit-configuration.svg")));
+	m_ui->delete_button->setIcon(QIcon(QStringLiteral(":/icons/remove-configuration.svg")));
 
 	m_ui->delete_button->setEnabled(false);
 	m_ui->edit_button->setEnabled(false);
@@ -249,6 +253,8 @@ ConfigurationListWidget::ConfigurationListWidget(QWidget* parent)
 
 	connect(m_ui->global_settings_button, &QToolButton::clicked, this,
 	        &ConfigurationListWidget::edit_global_settings);
+	connect(m_ui->input_mapping_button, &QToolButton::clicked, this,
+	        &ConfigurationListWidget::edit_input_mapping);
 	connect(m_ui->edit_button, &QToolButton::clicked, this,
 	        &ConfigurationListWidget::edit_configuration);
 	connect(m_ui->delete_button, &QToolButton::clicked, this,
@@ -717,6 +723,14 @@ void ConfigurationListWidget::edit_global_settings() {
 	}
 }
 
+void ConfigurationListWidget::edit_input_mapping() {
+	InputMappingDialog dialog(m_global_info.host_input_mapping, this);
+	if (dialog.exec() == QDialog::Accepted) {
+		m_global_info.host_input_mapping = dialog.Mapping();
+		WriteSettings();
+	}
+}
+
 void ConfigurationListWidget::ClearCustomSettings(ConfigurationItem* item) {
 	delete m_custom_infos.take(item->GetInfo().game_path);
 }

src/launcher/src/inputMappingDialog.cpp

@@ -0,0 +1,311 @@
+#include "inputMappingDialog.h"
+
+#include <QAbstractItemView>
+#include <QDialogButtonBox>
+#include <QHBoxLayout>
+#include <QHash>
+#include <QHeaderView>
+#include <QKeyEvent>
+#include <QKeySequence>
+#include <QLabel>
+#include <QMouseEvent>
+#include <QPushButton>
+#include <QStringList>
+#include <QTreeWidget>
+#include <QVBoxLayout>
+
+namespace {
+
+constexpr int BINDING_COLUMN = 1;
+
+struct PadButton {
+	const char* id;
+	const char* label;
+	const char* default_binding;
+};
+
+constexpr PadButton PAD_BUTTONS[] = {
+    {"Up", "D-pad Up", "Up"},
+    {"Down", "D-pad Down", "Down"},
+    {"Left", "D-pad Left", "Left"},
+    {"Right", "D-pad Right", "Right"},
+    {"Triangle", "Triangle", "I"},
+    {"Circle", "Circle", "L"},
+    {"Cross", "Cross", "J"},
+    {"Square", "Square", "K"},
+    {"L1", "L1", "Q"},
+    {"R1", "R1", "E"},
+    {"L2", "L2", ""},
+    {"R2", "R2", ""},
+    {"L3", "L3", "Left Shift"},
+    {"R3", "R3", "Left Ctrl"},
+    {"Options", "Options", "Return"},
+    {"TouchPad", "Touch pad", "Backspace"},
+};
+
+QString KeypadName(int key) {
+	if (key >= Qt::Key_0 && key <= Qt::Key_9) {
+		return QStringLiteral("Keypad %1").arg(key - Qt::Key_0);
+	}
+
+	switch (key) {
+		case Qt::Key_Return:
+		case Qt::Key_Enter: return QStringLiteral("Keypad Enter");
+		case Qt::Key_Slash: return QStringLiteral("Keypad /");
+		case Qt::Key_Asterisk: return QStringLiteral("Keypad *");
+		case Qt::Key_Minus: return QStringLiteral("Keypad -");
+		case Qt::Key_Plus: return QStringLiteral("Keypad +");
+		case Qt::Key_Period: return QStringLiteral("Keypad .");
+		case Qt::Key_Equal: return QStringLiteral("Keypad =");
+		case Qt::Key_Comma: return QStringLiteral("Keypad ,");
+		default: return {};
+	}
+}
+
+QString KeyName(const QKeyEvent& event) {
+	const int key = event.key();
+	if (event.modifiers().testFlag(Qt::KeypadModifier)) {
+		return KeypadName(key);
+	}
+	if (key >= Qt::Key_A && key <= Qt::Key_Z) {
+		return QChar(key);
+	}
+	if (key >= Qt::Key_0 && key <= Qt::Key_9) {
+		return QChar(key);
+	}
+	if (key >= Qt::Key_F1 && key <= Qt::Key_F24) {
+		return QStringLiteral("F%1").arg(key - Qt::Key_F1 + 1);
+	}
+
+	switch (key) {
+		case Qt::Key_Return:
+		case Qt::Key_Enter: return QStringLiteral("Return");
+		case Qt::Key_Backspace: return QStringLiteral("Backspace");
+		case Qt::Key_Tab: return QStringLiteral("Tab");
+		case Qt::Key_Shift: return QStringLiteral("Left Shift");
+		case Qt::Key_Control: return QStringLiteral("Left Ctrl");
+		case Qt::Key_Alt: return QStringLiteral("Left Alt");
+		case Qt::Key_Meta: return QStringLiteral("Left GUI");
+		case Qt::Key_Insert: return QStringLiteral("Insert");
+		case Qt::Key_Delete: return QStringLiteral("Delete");
+		case Qt::Key_Home: return QStringLiteral("Home");
+		case Qt::Key_End: return QStringLiteral("End");
+		case Qt::Key_PageUp: return QStringLiteral("PageUp");
+		case Qt::Key_PageDown: return QStringLiteral("PageDown");
+		case Qt::Key_Left: return QStringLiteral("Left");
+		case Qt::Key_Right: return QStringLiteral("Right");
+		case Qt::Key_Up: return QStringLiteral("Up");
+		case Qt::Key_Down: return QStringLiteral("Down");
+		case Qt::Key_CapsLock: return QStringLiteral("CapsLock");
+		case Qt::Key_NumLock: return QStringLiteral("Numlock");
+		case Qt::Key_ScrollLock: return QStringLiteral("ScrollLock");
+		case Qt::Key_Pause: return QStringLiteral("Pause");
+		case Qt::Key_Print: return QStringLiteral("PrintScreen");
+		default: break;
+	}
+
+	if (event.modifiers() != Qt::NoModifier) {
+		return {};
+	}
+	const auto name = QKeySequence(key).toString(QKeySequence::PortableText);
+	return name.size() == 1 ? name : QString();
+}
+
+class InputCaptureDialog final: public QDialog {
+public:
+	explicit InputCaptureDialog(QWidget* parent): QDialog(parent) {
+		setWindowTitle(tr("Set Binding"));
+		setModal(true);
+		setMinimumWidth(360);
+
+		auto* layout = new QVBoxLayout(this);
+		m_label      = new QLabel(
+		    tr("Press a key or mouse button.\nSpace and F1 are reserved; Esc cancels."), this);
+		m_label->setAlignment(Qt::AlignCenter);
+		layout->addWidget(m_label);
+	}
+
+	[[nodiscard]] const QString& Binding() const { return m_binding; }
+
+protected:
+	void keyPressEvent(QKeyEvent* event) override {
+		if (event->isAutoRepeat()) {
+			return;
+		}
+		if (event->key() == Qt::Key_Escape) {
+			reject();
+			return;
+		}
+		if (event->key() == Qt::Key_Space || event->key() == Qt::Key_F1) {
+			m_label->setText(tr("That key is reserved by the emulator."));
+			return;
+		}
+
+		m_binding = KeyName(*event);
+		if (!m_binding.isEmpty()) {
+			accept();
+		} else {
+			m_label->setText(tr("That key is not supported."));
+		}
+	}
+
+	void mousePressEvent(QMouseEvent* event) override {
+		switch (event->button()) {
+			case Qt::LeftButton: m_binding = QStringLiteral("Mouse:Left"); break;
+			case Qt::RightButton: m_binding = QStringLiteral("Mouse:Right"); break;
+			case Qt::MiddleButton: m_binding = QStringLiteral("Mouse:Middle"); break;
+			case Qt::BackButton: m_binding = QStringLiteral("Mouse:X1"); break;
+			case Qt::ForwardButton: m_binding = QStringLiteral("Mouse:X2"); break;
+			default: return;
+		}
+		accept();
+	}
+
+private:
+	QLabel* m_label = nullptr;
+	QString m_binding;
+};
+
+QHash<QString, QString> ParseMapping(const QStringList& mapping) {
+	QHash<QString, QString> result;
+	for (const auto& entry: mapping) {
+		const auto separator = entry.indexOf(QLatin1Char('='));
+		if (separator > 0 && separator + 1 < entry.size()) {
+			const auto binding = entry.mid(separator + 1);
+			for (auto item = result.begin(); item != result.end();) {
+				if (item.value().compare(binding, Qt::CaseInsensitive) == 0) {
+					item = result.erase(item);
+				} else {
+					++item;
+				}
+			}
+			result.insert(entry.left(separator), binding);
+		}
+	}
+	return result;
+}
+
+} // namespace
+
+InputMappingDialog::InputMappingDialog(const QStringList& mapping, QWidget* parent)
+    : QDialog(parent), m_customized(!mapping.isEmpty()) {
+	setWindowTitle(tr("Input Mapping"));
+	resize(440, 520);
+
+	auto* layout = new QVBoxLayout(this);
+	layout->addWidget(new QLabel(tr("Map keyboard or mouse input to DualSense buttons.\n"
+	                                "Defaults: left stick W/A/S/D, right stick T/F/G/H."),
+	                             this));
+
+	m_bindings = new QTreeWidget(this);
+	m_bindings->setColumnCount(2);
+	m_bindings->setHeaderLabels({tr("DualSense button"), tr("Host input")});
+	m_bindings->setRootIsDecorated(false);
+	m_bindings->setSelectionMode(QAbstractItemView::SingleSelection);
+	m_bindings->header()->setSectionResizeMode(0, QHeaderView::Stretch);
+	m_bindings->header()->setSectionResizeMode(1, QHeaderView::Stretch);
+	layout->addWidget(m_bindings);
+
+	const auto parsed = ParseMapping(mapping);
+	for (const auto& button: PAD_BUTTONS) {
+		auto* item = new QTreeWidgetItem(m_bindings);
+		item->setText(0, tr(button.label));
+		item->setData(0, Qt::UserRole, QString::fromLatin1(button.id));
+		SetBinding(item, m_customized ? parsed.value(QString::fromLatin1(button.id))
+		                              : QString::fromLatin1(button.default_binding));
+	}
+	m_bindings->setCurrentItem(m_bindings->topLevelItem(0));
+
+	auto* controls  = new QHBoxLayout;
+	m_change_button = new QPushButton(tr("Change..."), this);
+	m_clear_button  = new QPushButton(tr("Clear"), this);
+	auto* defaults  = new QPushButton(tr("Defaults"), this);
+	controls->addWidget(m_change_button);
+	controls->addWidget(m_clear_button);
+	controls->addWidget(defaults);
+	controls->addStretch();
+	layout->addLayout(controls);
+
+	auto* buttons = new QDialogButtonBox(QDialogButtonBox::Ok | QDialogButtonBox::Cancel, this);
+	layout->addWidget(buttons);
+
+	connect(m_bindings, &QTreeWidget::itemDoubleClicked, this,
+	        [this](QTreeWidgetItem*, int) { ChangeBinding(); });
+	connect(m_bindings, &QTreeWidget::itemSelectionChanged, this, [this]() { UpdateButtons(); });
+	connect(m_change_button, &QPushButton::clicked, this, [this]() { ChangeBinding(); });
+	connect(m_clear_button, &QPushButton::clicked, this, [this]() { ClearBinding(); });
+	connect(defaults, &QPushButton::clicked, this, [this]() { RestoreDefaults(); });
+	connect(buttons, &QDialogButtonBox::accepted, this, &QDialog::accept);
+	connect(buttons, &QDialogButtonBox::rejected, this, &QDialog::reject);
+	UpdateButtons();
+}
+
+QStringList InputMappingDialog::Mapping() const {
+	QStringList result;
+	if (!m_customized) {
+		return result;
+	}
+	for (int index = 0; index < m_bindings->topLevelItemCount(); index++) {
+		const auto* item    = m_bindings->topLevelItem(index);
+		const auto  binding = item->data(BINDING_COLUMN, Qt::UserRole).toString();
+		if (!binding.isEmpty()) {
+			result.append(item->data(0, Qt::UserRole).toString() + QLatin1Char('=') + binding);
+		}
+	}
+	return result;
+}
+
+void InputMappingDialog::ChangeBinding() {
+	auto* item = m_bindings->currentItem();
+	if (item == nullptr) {
+		return;
+	}
+
+	InputCaptureDialog dialog(this);
+	if (dialog.exec() != QDialog::Accepted) {
+		return;
+	}
+
+	for (int index = 0; index < m_bindings->topLevelItemCount(); index++) {
+		auto* other = m_bindings->topLevelItem(index);
+		if (other != item && other->data(BINDING_COLUMN, Qt::UserRole)
+		                             .toString()
+		                             .compare(dialog.Binding(), Qt::CaseInsensitive) == 0) {
+			SetBinding(other, {});
+		}
+	}
+	SetBinding(item, dialog.Binding());
+	m_customized = true;
+}
+
+void InputMappingDialog::ClearBinding() {
+	SetBinding(m_bindings->currentItem(), {});
+	m_customized = true;
+}
+
+void InputMappingDialog::RestoreDefaults() {
+	for (int index = 0; index < m_bindings->topLevelItemCount(); index++) {
+		SetBinding(m_bindings->topLevelItem(index),
+		           QString::fromLatin1(PAD_BUTTONS[index].default_binding));
+	}
+	m_customized = false;
+}
+
+void InputMappingDialog::SetBinding(QTreeWidgetItem* item, const QString& binding) {
+	if (item == nullptr) {
+		return;
+	}
+	item->setData(BINDING_COLUMN, Qt::UserRole, binding);
+	item->setText(BINDING_COLUMN, binding.isEmpty() ? tr("None") : binding);
+	UpdateButtons();
+}
+
+void InputMappingDialog::UpdateButtons() {
+	if (m_change_button == nullptr) {
+		return;
+	}
+	const auto* item = m_bindings->currentItem();
+	m_change_button->setEnabled(item != nullptr);
+	m_clear_button->setEnabled(item != nullptr &&
+	                           !item->data(BINDING_COLUMN, Qt::UserRole).toString().isEmpty());
+}

src/launcher/src/mainDialog.cpp

@@ -204,7 +204,11 @@ static QStringList CreateEmulatorArgs(const Configuration& info) {
 
 	args << "--screen-width" << r.at(0);
 	args << "--screen-height" << r.at(1);
+	if (info.fullscreen_enabled) {
+		args << "--fullscreen";
+	}
 	args << "--vblank-frequency" << QString::number(info.vblank_frequency);
+	args << "--console-language" << QString::number(info.console_language);
 	args << "--vulkan-validation" << BoolArg(info.vulkan_validation_enabled);
 	args << "--shader-validation" << BoolArg(info.shader_validation_enabled);
 	args << "--shader-optimization-type" << EnumToText(info.shader_optimization_type);
@@ -216,7 +220,9 @@ static QStringList CreateEmulatorArgs(const Configuration& info) {
 	args << "--printf-output-file" << info.printf_output_file;
 	args << "--profiler-direction" << EnumToText(info.profiler_direction);
 	args << "--spirv-debug-printf" << "false";
-	args << "--ngg-rectlist-draw" << BoolArg(info.ngg_rectlist_draw_enabled);
+	for (const auto& binding: info.host_input_mapping) {
+		args << "--keymap" << binding;
+	}
 	if (info.renderdoc_enabled) {
 		args << "--rd";
 	}
@@ -435,7 +441,10 @@ void MainDialogPrivate::Run() {
 
 	m_running_item->SetRunning(true);
 
-	m_main_dialog->RunInterpreter(&m_process, m_running_item->GetInfo());
+	Configuration info;
+	info.CopyFrom(m_running_item->GetInfo());
+	info.host_input_mapping = m_ui->widget->GetHostInputMapping();
+	m_main_dialog->RunInterpreter(&m_process, info);
 
 	Update();
 }

src/libs/agc.cpp

@@ -6,6 +6,7 @@
 #include "common/logging/log.h"
 #include "common/stringUtils.h"
 #include "common/virtualMemory.h"
+#include "graphics/guest_gpu/command_processor/commandProcessor.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/guest_gpu/hardwareContext.h"
@@ -221,57 +222,6 @@ static RegisterDefaults* get_internal_register_defaults(uint32_t ver) {
 	return get_register_defaults(g_agc_internal_reg_defaults_by_version[index], &storage[index]);
 }
 
-struct PendingGraphicsSegment {
-	uint32_t* start     = nullptr;
-	uint32_t* end       = nullptr;
-	uint32_t* range_end = nullptr;
-};
-
-static std::mutex             g_pending_graphics_segment_mutex;
-static PendingGraphicsSegment g_pending_graphics_segment;
-
-static void track_pending_graphics_segment_after_submit(uint32_t* dcb, uint32_t size_in_dwords) {
-	if (dcb == nullptr || size_in_dwords == 0) {
-		return;
-	}
-
-	auto* segment_start = dcb + size_in_dwords;
-	auto* range_end     = segment_start + 0xfffffu;
-
-	std::lock_guard lock(g_pending_graphics_segment_mutex);
-	g_pending_graphics_segment.start     = segment_start;
-	g_pending_graphics_segment.end       = segment_start;
-	g_pending_graphics_segment.range_end = range_end;
-}
-
-static void track_pending_graphics_allocation(uint32_t* cmd, uint32_t size_dw) {
-	if (cmd == nullptr || size_dw == 0) {
-		return;
-	}
-
-	std::lock_guard lock(g_pending_graphics_segment_mutex);
-	auto*           range_start = g_pending_graphics_segment.start;
-	auto*           range_end   = g_pending_graphics_segment.range_end;
-	if (range_start == nullptr || range_end == nullptr || cmd < range_start || cmd >= range_end) {
-		return;
-	}
-
-	auto* cmd_end = cmd + size_dw;
-	if (cmd > g_pending_graphics_segment.end) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1) < 64) {
-			LOGF("\t pending graphics segment: ignoring non-contiguous allocation cmd = "
-			     "0x%016" PRIx64 ", tracked_end = 0x%016" PRIx64 "\n",
-			     reinterpret_cast<uint64_t>(cmd),
-			     reinterpret_cast<uint64_t>(g_pending_graphics_segment.end));
-		}
-		return;
-	}
-	if (cmd_end > g_pending_graphics_segment.end && cmd_end <= range_end) {
-		g_pending_graphics_segment.end = cmd_end;
-	}
-}
-
 struct CommandBuffer {
 	using Callback = KYTY_SYSV_ABI bool (*)(CommandBuffer*, uint32_t, void*);
 
@@ -370,7 +320,6 @@ struct CommandBuffer {
 		}
 		auto* ret_ptr = cursor_up;
 		cursor_up += size_dw;
-		track_pending_graphics_allocation(ret_ptr, size_dw);
 		return ret_ptr;
 	}
 };
@@ -834,6 +783,104 @@ int KYTY_SYSV_ABI GraphicsUnknownFuseShaderHalves(Shader* fused_result, const Sh
 	return OK;
 }
 
+static void merge_shader_register_max_field(ShaderRegister* dst, const ShaderRegister* src,
+                                            uint32_t shift, uint32_t mask) {
+	const auto dst_field = (dst->value >> shift) & mask;
+	const auto src_field = (src->value >> shift) & mask;
+	const auto field     = std::max(dst_field, src_field);
+
+	dst->value &= ~(mask << shift);
+	dst->value |= field << shift;
+}
+
+int KYTY_SYSV_ABI GraphicsUnknownNApJjpKNBl4(Shader* fused_result, const Shader* front,
+                                             const Shader* back, void* scratch_mem) {
+	PRINT_NAME();
+
+	LOGF("\t fused_result = 0x%016" PRIx64 "\n"
+	     "\t front        = 0x%016" PRIx64 "\n"
+	     "\t back         = 0x%016" PRIx64 "\n"
+	     "\t scratch_mem  = 0x%016" PRIx64 "\n",
+	     reinterpret_cast<uint64_t>(fused_result), reinterpret_cast<uint64_t>(front),
+	     reinterpret_cast<uint64_t>(back), reinterpret_cast<uint64_t>(scratch_mem));
+
+	const auto front_type = static_cast<Prospero::ShaderBinaryType>(front->type);
+	const auto is_gs      = front_type == Prospero::ShaderBinaryType::kGsFront;
+	const auto is_hs      = front_type == Prospero::ShaderBinaryType::kHsFront;
+	if ((!is_gs && !is_hs) ||
+	    (is_gs && back->type != static_cast<uint8_t>(Prospero::ShaderBinaryType::kGsBack)) ||
+	    (is_hs && back->type != static_cast<uint8_t>(Prospero::ShaderBinaryType::kHsBack))) {
+		return GRAPHICS5_ERROR_INVALID_SHADER_HALVES;
+	}
+
+	*fused_result      = *back;
+	fused_result->type = static_cast<uint8_t>(is_gs ? Prospero::ShaderBinaryType::kGs
+	                                                : Prospero::ShaderBinaryType::kHs);
+
+	const auto back_stages  = back->specials->vgt_shader_stages_en.value;
+	const auto front_stages = front->specials->vgt_shader_stages_en.value;
+	const auto mismatch_bit = is_gs ? (1u << 22u) : (1u << 21u);
+	if (((front_stages ^ back_stages) & mismatch_bit) != 0) {
+		return GRAPHICS5_ERROR_INVALID_SHADER_HALVES;
+	}
+
+	if (scratch_mem != nullptr) {
+		auto* sh_registers = static_cast<ShaderRegister*>(scratch_mem);
+		memcpy(sh_registers, back->sh_registers,
+		       static_cast<size_t>(back->num_sh_registers) * sizeof(ShaderRegister));
+		fused_result->sh_registers = sh_registers;
+	}
+
+	auto*      fused_regs      = fused_result->sh_registers;
+	const auto fused_reg_count = static_cast<uint32_t>(fused_result->num_sh_registers);
+	const auto front_reg_count = static_cast<uint32_t>(front->num_sh_registers);
+	const auto checksum_offset =
+	    is_gs ? Pm4::SPI_SHADER_PGM_CHKSUM_GS : Pm4::SPI_SHADER_PGM_CHKSUM_HS;
+	const auto* front_checksum0 =
+	    find_shader_register(front->sh_registers, front_reg_count, checksum_offset, 0);
+	const auto* front_checksum1 =
+	    find_shader_register(front->sh_registers, front_reg_count, checksum_offset, 1);
+	auto* fused_checksum0  = find_shader_register(fused_regs, fused_reg_count, checksum_offset, 0);
+	auto* fused_checksum1  = find_shader_register(fused_regs, fused_reg_count, checksum_offset, 1);
+	fused_checksum0->value = front_checksum0->value;
+	fused_checksum1->value = front_checksum1->value;
+
+	const auto  rsrc1_offset = is_gs ? Pm4::SPI_SHADER_PGM_RSRC1_GS : Pm4::SPI_SHADER_PGM_RSRC1_HS;
+	const auto  rsrc2_offset = is_gs ? Pm4::SPI_SHADER_PGM_RSRC2_GS : Pm4::SPI_SHADER_PGM_RSRC2_HS;
+	const auto* front_rsrc1 =
+	    find_shader_register(front->sh_registers, front_reg_count, rsrc1_offset);
+	const auto* front_rsrc2 =
+	    find_shader_register(front->sh_registers, front_reg_count, rsrc2_offset);
+	auto* fused_rsrc1 = find_shader_register(fused_regs, fused_reg_count, rsrc1_offset);
+	auto* fused_rsrc2 = find_shader_register(fused_regs, fused_reg_count, rsrc2_offset);
+
+	merge_shader_register_max_field(fused_rsrc1, front_rsrc1, 0, 0x3fu);
+	merge_shader_register_max_field(fused_rsrc2, front_rsrc2, 28, 0x0fu);
+	if (is_gs) {
+		merge_shader_register_max_field(fused_rsrc1, front_rsrc1, 29, 0x03u);
+		merge_shader_register_max_field(fused_rsrc2, front_rsrc2, 16, 0x03u);
+		fused_rsrc2->value =
+		    (fused_rsrc2->value & 0xf7ffffc1u) | (front_rsrc2->value & 0x0800003eu);
+		fused_rsrc2->value =
+		    (fused_rsrc2->value & 0xfffbffffu) | (front_rsrc2->value & 0x00040000u);
+	} else {
+		merge_shader_register_max_field(fused_rsrc1, front_rsrc1, 28, 0x03u);
+		fused_rsrc2->value =
+		    (fused_rsrc2->value & 0xf7ffffc1u) | (front_rsrc2->value & 0x0800003eu);
+	}
+
+	const auto program_lo_offset = is_gs ? Pm4::SPI_SHADER_PGM_LO_ES : Pm4::SPI_SHADER_PGM_LO_LS;
+	auto*      program_lo = find_shader_register(fused_regs, fused_reg_count, program_lo_offset);
+	const auto address    = reinterpret_cast<uint64_t>(front->code);
+	program_lo->value     = static_cast<uint32_t>(address >> 8u);
+	(program_lo + 1)->value &= 0xffffff00u;
+	(program_lo + 1)->value |= static_cast<uint32_t>((address >> 40u) & 0xffu);
+
+	fused_result->user_data = front->user_data;
+
+	return OK;
+}
+
 static constexpr int GRAPHICS5_ERROR_INVALID_PACKET = static_cast<int>(0x8a6c000cu);
 
 enum class RegIndirectPacket : uint32_t {
@@ -1347,7 +1394,7 @@ int KYTY_SYSV_ABI GraphicsWriteDataPatchSetAddressOrOffset(uint32_t* cmd,
 		return OK;
 	}
 
-	return static_cast<int>(0x8a6c000cu);
+	return GRAPHICS5_ERROR_INVALID_PACKET;
 }
 
 int KYTY_SYSV_ABI GraphicsUnknownJumpPatchSetTarget(uint32_t* cmd, const volatile uint32_t* target,
@@ -1398,29 +1445,72 @@ int KYTY_SYSV_ABI GraphicsSuspendPoint() {
 	return OK;
 }
 
-uint32_t* KYTY_SYSV_ABI GraphicsUnknownQj7QZpgr9Uw(CommandBuffer* buf, uint32_t mode,
-                                                   uint32_t value) {
-	PRINT_NAME();
-	// ResetContextState?
-	LOGF("\t mode  = 0x%08" PRIx32 "\n"
-	     "\t value = 0x%08" PRIx32 "\n",
-	     mode, value);
+[[nodiscard]] static constexpr uint32_t context_state_op_size_dw(uint32_t operation) {
+	switch (static_cast<ContextStateOperation>(operation)) {
+		case ContextStateOperation::Clear: return 5;
+		case ContextStateOperation::Push:
+		case ContextStateOperation::Pop: return 27;
+		case ContextStateOperation::PushClear: return 32;
+		default: return 0;
+	}
+}
 
-	if (buf == nullptr) {
+uint32_t* KYTY_SYSV_ABI GraphicsDcbContextStateOp(CommandBuffer* buf, uint32_t operation) {
+	PRINT_NAME();
+	LOGF("\t operation = 0x%08" PRIx32 "\n", operation);
+
+	const auto size_dw = context_state_op_size_dw(operation);
+	if (buf == nullptr || size_dw == 0) {
 		return nullptr;
 	}
 
 	buf->DbgDump();
 
-	auto* cmd = buf->AllocateDW(1);
+	uint32_t* first = nullptr;
+	const auto append = [&](uint32_t segment_size_dw) {
+		auto* segment = buf->AllocateDW(segment_size_dw);
+		if (segment == nullptr) {
+			return false;
+		}
+		std::fill_n(segment, segment_size_dw, 0u);
+		if (first == nullptr) {
+			first      = segment;
+			segment[0] = KYTY_PM4(segment_size_dw, Pm4::IT_NOP, Pm4::R_CONTEXT_STATE);
+			segment[1] = operation;
+		} else {
+			segment[0] = KYTY_PM4(segment_size_dw, Pm4::IT_NOP, Pm4::R_ZERO);
+		}
+		return true;
+	};
+	const auto append_save_restore = [&] {
+		// The native helper reserves 22 dwords, then emits 5-, 8-, and 9-dword packets.
+		return buf->ReserveDW(22) && append(5) && append(8) && append(9);
+	};
 
-	if (cmd == nullptr) {
-		return nullptr;
+	// Preserve native packet sizes, allocation boundaries, Push/Pop ordering, and
+	// returned first-packet address. The native packets reference process-private state buffers,
+	// so their net Cx operation is represented by the first HLE packet and type-3 NOPs.
+	bool complete = false;
+	switch (static_cast<ContextStateOperation>(operation)) {
+		case ContextStateOperation::Clear: complete = append(5); break;
+		case ContextStateOperation::Push:
+			complete = append_save_restore() && append(3) && append(2);
+			break;
+		case ContextStateOperation::Pop:
+			complete = append(3) && append_save_restore() && append(2);
+			break;
+		case ContextStateOperation::PushClear:
+			complete = append_save_restore() && append(3) && append(2) && append(5);
+			break;
+		default: break;
 	}
 
-	cmd[0] = 0x80000000u;
+	return complete ? first : nullptr;
+}
 
-	return cmd;
+uint64_t KYTY_SYSV_ABI GraphicsDcbContextStateOpGetSize(uint32_t operation) {
+	PRINT_NAME();
+	return static_cast<uint64_t>(context_state_op_size_dw(operation)) * sizeof(uint32_t);
 }
 
 uint64_t KYTY_SYSV_ABI GraphicsGetIsTrinityMode() {
@@ -1820,7 +1910,6 @@ uint32_t* KYTY_SYSV_ABI GraphicsCbReleaseMem(CommandBuffer* buf, uint8_t action,
 	cmd[5] = static_cast<uint32_t>(packet_data & 0xffffffffu);
 	cmd[6] = static_cast<uint32_t>((packet_data >> 32u) & 0xffffffffu);
 	cmd[7] = interrupt_ctx_id & 0x07ffffffu;
-
 	return cmd;
 }
 
@@ -2380,7 +2469,7 @@ int KYTY_SYSV_ABI GraphicsUnknownIkfdtRIqCE(uint32_t* cmd, uint64_t arg1,
 
 	auto op = (cmd[0] >> 8u) & 0xffu;
 	if (op != Pm4::IT_INDIRECT_BUFFER) {
-		return 0x8a6c000c;
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto vaddr        = reinterpret_cast<uint64_t>(target);
@@ -2800,6 +2889,10 @@ uint32_t KYTY_SYSV_ABI GraphicsAcbCondExecGetSize() {
 	return GraphicsDcbCondExecGetSize();
 }
 
+uint32_t KYTY_SYSV_ABI GraphicsAcbJumpGetSize() {
+	return 0x10u;
+}
+
 uint32_t* KYTY_SYSV_ABI GraphicsAcbWaitRegMem(CommandBuffer* buf, uint8_t size,
                                               uint8_t compare_function, uint8_t cache_policy,
                                               const volatile void* address, uint64_t reference,
@@ -3119,7 +3212,7 @@ int KYTY_SYSV_ABI GraphicsDmaDataPatchSetDstAddressOrOffset(uint32_t* cmd,
 		return OK;
 	}
 
-	return static_cast<int>(0x8a6c000cu);
+	return GRAPHICS5_ERROR_INVALID_PACKET;
 }
 
 int KYTY_SYSV_ABI GraphicsDmaDataPatchSetSrcAddressOrOffsetOrImmediate(
@@ -3133,7 +3226,7 @@ int KYTY_SYSV_ABI GraphicsDmaDataPatchSetSrcAddressOrOffsetOrImmediate(
 		return OK;
 	}
 
-	return static_cast<int>(0x8a6c000cu);
+	return GRAPHICS5_ERROR_INVALID_PACKET;
 }
 
 uint32_t KYTY_SYSV_ABI GraphicsGetPacketSize(uint32_t* packet) {
@@ -3197,6 +3290,15 @@ int KYTY_SYSV_ABI GraphicsSetRangePredication(uint32_t* start, const volatile ui
 	return OK;
 }
 
+int KYTY_SYSV_ABI GraphicsRewindPatchSetRewindState(uint32_t* cmd, uint8_t state) {
+	if (((cmd[0] >> 8u) & 0xffu) != Pm4::IT_REWIND) {
+		return GRAPHICS5_ERROR_INVALID_PACKET;
+	}
+
+	cmd[1] = (cmd[1] & 0x7fffffffu) | (static_cast<uint32_t>(state) << 31u);
+	return OK;
+}
+
 int KYTY_SYSV_ABI GraphicsCondExecPatchSetEnd(uint32_t* cmd, const volatile uint32_t* buffer) {
 	PRINT_NAME();
 
@@ -3205,23 +3307,23 @@ int KYTY_SYSV_ABI GraphicsCondExecPatchSetEnd(uint32_t* cmd, const volatile uint
 	     reinterpret_cast<uint64_t>(cmd), reinterpret_cast<uint64_t>(buffer));
 
 	if (cmd == nullptr || buffer == nullptr) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto op = (cmd[0] >> 8u) & 0xffu;
 	if (op != Pm4::IT_COND_EXEC) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto* packet_end = cmd + 5;
 	auto* range_end  = const_cast<uint32_t*>(reinterpret_cast<const volatile uint32_t*>(buffer));
 	if (range_end < packet_end) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto num_dwords = static_cast<uint64_t>(range_end - packet_end);
 	if (num_dwords > 0x3fffu) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	cmd[4] = (cmd[4] & ~0x3fffu) | static_cast<uint32_t>(num_dwords);
@@ -3237,12 +3339,12 @@ int KYTY_SYSV_ABI GraphicsCondExecPatchSetCommandAddress(uint32_t*
 	     reinterpret_cast<uint64_t>(cmd), reinterpret_cast<uint64_t>(command));
 
 	if (cmd == nullptr || command == nullptr) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto op = (cmd[0] >> 8u) & 0xffu;
 	if (op != Pm4::IT_COND_EXEC || (reinterpret_cast<uintptr_t>(command) & 0x3u) != 0) {
-		return static_cast<int>(0x8a6c000cu);
+		return GRAPHICS5_ERROR_INVALID_PACKET;
 	}
 
 	auto addr = reinterpret_cast<uint64_t>(command);
@@ -3704,150 +3806,6 @@ static void submit_dcb(uint32_t* dcb, uint32_t size_in_dwords) {
 	EXIT_IF(g_renderer == nullptr);
 	g_renderer->GetGpu().Submit(dcb, size_in_dwords, nullptr, 0,
 	                            !dcb_has_queued_interrupt(dcb, size_in_dwords));
-	Gen5::track_pending_graphics_segment_after_submit(dcb, size_in_dwords);
-}
-
-static std::vector<uint64_t> collect_acb_wait_addresses(const uint32_t* acb,
-                                                        uint32_t        size_in_dwords) {
-	std::vector<uint64_t> addresses;
-
-	for (uint32_t offset = 0; offset < size_in_dwords;) {
-		auto cmd_id = acb[offset];
-		auto len    = KYTY_PM4_LEN(cmd_id);
-		if (len == 0 || len > size_in_dwords - offset) {
-			return addresses;
-		}
-
-		auto op = (cmd_id >> 8u) & 0xffu;
-		if (op == Pm4::IT_NOP && KYTY_PM4_R(cmd_id) == Pm4::R_WAIT_MEM_32 && len >= 7) {
-			auto address = static_cast<uint64_t>(acb[offset + 1]) |
-			               (static_cast<uint64_t>(acb[offset + 2]) << 32u);
-			if (address != 0) {
-				addresses.push_back(address);
-			}
-		} else if (op == Pm4::IT_NOP && KYTY_PM4_R(cmd_id) == Pm4::R_WAIT_MEM_64 && len >= 9) {
-			auto address = static_cast<uint64_t>(acb[offset + 1]) |
-			               (static_cast<uint64_t>(acb[offset + 2]) << 32u);
-			if (address != 0) {
-				addresses.push_back(address);
-			}
-		}
-
-		offset += len;
-	}
-
-	return addresses;
-}
-
-static bool acb_waits_for_address(const std::vector<uint64_t>& wait_addresses,
-                                  uint64_t                     release_address) {
-	for (auto address: wait_addresses) {
-		if (address == release_address) {
-			return true;
-		}
-	}
-
-	return false;
-}
-
-static void flush_pending_graphics_segment_before_acb(const uint32_t* acb,
-                                                      uint32_t        acb_size_in_dwords) {
-	uint32_t* dcb            = nullptr;
-	uint32_t  size_in_dwords = 0;
-	auto      wait_addresses = collect_acb_wait_addresses(acb, acb_size_in_dwords);
-
-	{
-		std::lock_guard lock(Gen5::g_pending_graphics_segment_mutex);
-		if (!wait_addresses.empty() && Gen5::g_pending_graphics_segment.start != nullptr) {
-			auto* scan        = Gen5::g_pending_graphics_segment.start;
-			auto* matched_end = Gen5::g_pending_graphics_segment.start;
-			while (scan < Gen5::g_pending_graphics_segment.end) {
-				auto cmd_id = *scan;
-				if (cmd_id == 0x80000000u) {
-					scan++;
-					continue;
-				}
-				if ((cmd_id & 0xC0000000u) != 0xC0000000u) {
-					break;
-				}
-
-				auto len = KYTY_PM4_LEN(cmd_id);
-				if (len == 0 ||
-				    len > static_cast<uint32_t>(Gen5::g_pending_graphics_segment.end - scan)) {
-					break;
-				}
-
-				if (((cmd_id >> 8u) & 0xffu) == Pm4::IT_NOP &&
-				    KYTY_PM4_R(cmd_id) == Pm4::R_RELEASE_MEM && len >= 7) {
-					auto release_addr =
-					    static_cast<uint64_t>(scan[3]) | (static_cast<uint64_t>(scan[4]) << 32u);
-					if (acb_waits_for_address(wait_addresses, release_addr)) {
-						matched_end = scan + len;
-					}
-				}
-
-				scan += len;
-			}
-
-			if (matched_end > Gen5::g_pending_graphics_segment.start) {
-				Gen5::g_pending_graphics_segment.end = matched_end;
-			}
-		}
-
-		if (Gen5::g_pending_graphics_segment.start != nullptr &&
-		    Gen5::g_pending_graphics_segment.end > Gen5::g_pending_graphics_segment.start) {
-			auto* scan      = Gen5::g_pending_graphics_segment.start;
-			auto* valid_end = Gen5::g_pending_graphics_segment.start;
-			while (scan < Gen5::g_pending_graphics_segment.end) {
-				auto cmd_id = *scan;
-				if (cmd_id == 0x80000000u) {
-					scan++;
-					valid_end = scan;
-					continue;
-				}
-				if ((cmd_id & 0xC0000000u) != 0xC0000000u) {
-					break;
-				}
-
-				auto len = KYTY_PM4_LEN(cmd_id);
-				if (len == 0 ||
-				    len > static_cast<uint32_t>(Gen5::g_pending_graphics_segment.end - scan)) {
-					break;
-				}
-
-				scan += len;
-				valid_end = scan;
-			}
-
-			if (valid_end < Gen5::g_pending_graphics_segment.end) {
-				static std::atomic<uint32_t> log_count {0};
-				if (log_count.fetch_add(1) < 64) {
-					LOGF("\t trimming pending graphics segment: addr = 0x%016" PRIx64
-					     ", old_dw = 0x%08" PRIx32 ", new_dw = 0x%08" PRIx32 "\n",
-					     reinterpret_cast<uint64_t>(Gen5::g_pending_graphics_segment.start),
-					     static_cast<uint32_t>(Gen5::g_pending_graphics_segment.end -
-					                           Gen5::g_pending_graphics_segment.start),
-					     static_cast<uint32_t>(valid_end - Gen5::g_pending_graphics_segment.start));
-				}
-				Gen5::g_pending_graphics_segment.end = valid_end;
-			}
-		}
-
-		if (Gen5::g_pending_graphics_segment.start == nullptr ||
-		    Gen5::g_pending_graphics_segment.end <= Gen5::g_pending_graphics_segment.start) {
-			return;
-		}
-
-		dcb            = Gen5::g_pending_graphics_segment.start;
-		size_in_dwords = static_cast<uint32_t>(Gen5::g_pending_graphics_segment.end -
-		                                       Gen5::g_pending_graphics_segment.start);
-	}
-
-	LOGF("\t flushing pending graphics segment before ACB: addr = 0x%016" PRIx64
-	     ", dw_num = 0x%08" PRIx32 "\n",
-	     reinterpret_cast<uint64_t>(dcb), size_in_dwords);
-
-	submit_dcb(dcb, size_in_dwords);
 }
 
 int KYTY_SYSV_ABI GraphicsDriverSubmitDcb(const Packet* packet) {
@@ -3963,8 +3921,6 @@ static void submit_acb(uint32_t queue, uint32_t* acb, uint32_t size_in_dwords) {
 		LOGF("\t acb[%u] = 0x%08" PRIx32 "\n", i, acb[i]);
 	}
 
-	flush_pending_graphics_segment_before_acb(acb, size_in_dwords);
-
 	GraphicsDbgDumpDcb("a", size_in_dwords, acb);
 
 	const bool trigger_interrupt_on_done = !dcb_has_queued_interrupt(acb, size_in_dwords);

src/libs/agc.h

@@ -38,6 +38,8 @@ int KYTY_SYSV_ABI   GraphicsUnknownGetFusedShaderSize(SizeAlign* dst, const Shad
                                                       const Shader* back);
 int KYTY_SYSV_ABI   GraphicsUnknownFuseShaderHalves(Shader* fused_result, const Shader* front,
                                                     const Shader* back, void* scratch_mem);
+int KYTY_SYSV_ABI   GraphicsUnknownNApJjpKNBl4(Shader* fused_result, const Shader* front,
+                                               const Shader* back, void* scratch_mem);
 int KYTY_SYSV_ABI   GraphicsSetCxRegIndirectPatchSetAddress(uint32_t*                      cmd,
                                                             const volatile ShaderRegister* regs);
 int KYTY_SYSV_ABI   GraphicsSetShRegIndirectPatchSetAddress(uint32_t*                      cmd,
@@ -65,8 +67,8 @@ int KYTY_SYSV_ABI GraphicsUnknownJumpPatchSetTarget(uint32_t* cmd, const volatil
 int KYTY_SYSV_ABI GraphicsJumpPatchSetTarget(uint32_t* cmd, const volatile uint32_t* target,
                                              uint32_t size_in_dwords);
 int KYTY_SYSV_ABI GraphicsSuspendPoint();
-uint32_t* KYTY_SYSV_ABI GraphicsUnknownQj7QZpgr9Uw(CommandBuffer* buf, uint32_t mode,
-                                                   uint32_t value);
+uint32_t* KYTY_SYSV_ABI GraphicsDcbContextStateOp(CommandBuffer* buf, uint32_t operation);
+uint64_t KYTY_SYSV_ABI  GraphicsDcbContextStateOpGetSize(uint32_t operation);
 uint64_t KYTY_SYSV_ABI  GraphicsGetIsTrinityMode();
 
 uint32_t KYTY_SYSV_ABI GraphicsDriverGetDefaultOwner();
@@ -183,6 +185,7 @@ uint32_t KYTY_SYSV_ABI  GraphicsAcbAcquireMemGetSize();
 uint32_t* KYTY_SYSV_ABI GraphicsAcbCondExec(CommandBuffer* buf, const volatile uint32_t* address,
                                             uint32_t num_dwords);
 uint32_t KYTY_SYSV_ABI  GraphicsAcbCondExecGetSize();
+uint32_t KYTY_SYSV_ABI  GraphicsAcbJumpGetSize();
 uint32_t* KYTY_SYSV_ABI GraphicsAcbWaitRegMem(CommandBuffer* buf, uint8_t size,
                                               uint8_t compare_function, uint8_t cache_policy,
                                               const volatile void* address, uint64_t reference,
@@ -241,6 +244,7 @@ uint32_t KYTY_SYSV_ABI  GraphicsGetPacketSize(uint32_t* packet);
 int KYTY_SYSV_ABI       GraphicsSetPacketPredication(uint32_t* packet, uint32_t predication);
 int KYTY_SYSV_ABI       GraphicsSetRangePredication(uint32_t* start, const volatile uint32_t* end,
                                                     uint32_t predication);
+int KYTY_SYSV_ABI       GraphicsRewindPatchSetRewindState(uint32_t* cmd, uint8_t state);
 int KYTY_SYSV_ABI       GraphicsCondExecPatchSetEnd(uint32_t* cmd, const volatile uint32_t* buffer);
 int KYTY_SYSV_ABI       GraphicsCondExecPatchSetCommandAddress(uint32_t*                cmd,
                                                                const volatile uint32_t* command);

src/libs/audio.cpp

@@ -364,7 +364,12 @@ bool Audio::QueueSdlAudio(PortOut* port, const void* data, bool blocking) {
 	}
 
 	if (blocking) {
-		const auto min_queued_size = queue_size * 2u;
+		constexpr uint64_t target_latency_us = 40000;
+		const auto buffer_us = port->freq != 0 ? (1000000ULL * port->samples_num) / port->freq : 0;
+		const auto buffers =
+		    buffer_us != 0 ? static_cast<uint32_t>((target_latency_us + buffer_us - 1) / buffer_us)
+		                   : 2u;
+		const auto min_queued_size = queue_size * std::clamp(buffers, 2u, 16u);
 		const auto wait_start      = LibKernel::KernelGetProcessTime();
 		while (SDL_GetQueuedAudioSize(port->audio_device) > min_queued_size) {
 			if (LibKernel::KernelGetProcessTime() - wait_start > 200000) {
@@ -511,7 +516,16 @@ uint32_t Audio::AudioOutOutputs(OutputParam* params, uint32_t num, bool blocking
 		max_wait_time      = (wait_time > max_wait_time ? wait_time : max_wait_time);
 	}
 
-	if (blocking && max_wait_time != 0) {
+	bool all_ports_have_device = true;
+	for (uint32_t i = 0; i < num; i++) {
+		if (m_out_ports[params[i].handle.GetId()].audio_device == 0) {
+			all_ports_have_device = false;
+			break;
+		}
+	}
+
+	// Device-backed ports are paced by the SDL queue above.
+	if (blocking && max_wait_time != 0 && !all_ports_have_device) {
 		Common::Thread::SleepMicro(max_wait_time);
 	}
 
@@ -1416,6 +1430,12 @@ struct Ngs2CustomSubmixerRackOption {
 	uint32_t             max_inputs   = 0;
 };
 
+struct Ngs2CustomMasteringRackOption {
+	Ngs2CustomRackOption custom_rack_option;
+	uint32_t             max_channels = 0;
+	uint32_t             max_inputs   = 0;
+};
+
 struct Ngs2CustomSamplerRackOption {
 	Ngs2CustomRackOption custom_rack_option;
 	uint32_t             max_channel_works        = 0;
@@ -1433,6 +1453,7 @@ union Ngs2RackOptionUnion {
 	Ngs2SubmixerRackOption       submixer;
 	Ngs2ReverbRackOption         reverb;
 	Ngs2CustomSubmixerRackOption custom_submixer;
+	Ngs2CustomMasteringRackOption custom_mastering;
 	Ngs2CustomSamplerRackOption  custom_sampler;
 };
 
@@ -1588,6 +1609,7 @@ enum class Ngs2RackType {
 	Mastering,
 	Reverb,
 	CustomSubmixer,
+	CustomMastering,
 	CustomSampler,
 };
 
@@ -1665,6 +1687,20 @@ struct Ngs2VoiceCallbackParam {
 
 struct Ngs2VoiceState {
 	uint32_t state_flags;
+	int32_t  error_code;
+};
+
+struct Ngs2SubmixerVoiceState {
+	Ngs2VoiceState voice_state;
+	float          envelope_height;
+	float          peak_height;
+	float          compressor_height;
+};
+
+struct Ngs2CustomMasteringVoiceState {
+	Ngs2VoiceState voice_state;
+	uint32_t       reserved;
+	uint32_t       reserved2;
 };
 
 struct Ngs2SamplerVoiceState {
@@ -1683,6 +1719,10 @@ static Ngs2RackInternal* g_racks_list = nullptr;
 
 static_assert(sizeof(Ngs2SystemOption) == 144);
 static_assert(sizeof(Ngs2RackOption) == 176);
+static_assert(sizeof(Ngs2VoiceState) == 8);
+static_assert(sizeof(Ngs2SubmixerVoiceState) == 20);
+static_assert(sizeof(Ngs2CustomMasteringVoiceState) == 16);
+static_assert(sizeof(Ngs2SamplerVoiceState) == 56);
 
 static uint32_t Ngs2GetStateFlags(const Ngs2VoiceInternal* voice) {
 	switch (voice->state) {
@@ -1726,6 +1766,7 @@ static Ngs2Internal* Ngs2CreateSystemInternal(const Ngs2SystemOption* option, vo
 static bool Ngs2RackIsCustom(Ngs2RackType type) {
 	switch (type) {
 		case Ngs2RackType::CustomSubmixer:
+		case Ngs2RackType::CustomMastering:
 		case Ngs2RackType::CustomSampler: return true;
 		default: return false;
 	}
@@ -2037,6 +2078,12 @@ int KYTY_SYSV_ABI Ngs2RackCreate(uintptr_t system_handle, uint32_t rack_id,
 			    *reinterpret_cast<const Ngs2CustomSubmixerRackOption*>(option);
 			rack->type = Ngs2RackType::CustomSubmixer;
 			break;
+		case 0x4003:
+			EXIT_NOT_IMPLEMENTED(option->size != sizeof(Ngs2CustomMasteringRackOption));
+			rack->option.custom_mastering =
+			    *reinterpret_cast<const Ngs2CustomMasteringRackOption*>(option);
+			rack->type = Ngs2RackType::CustomMastering;
+			break;
 		case 0x4001:
 			EXIT_NOT_IMPLEMENTED(option->size != sizeof(Ngs2CustomSamplerRackOption));
 			rack->option.custom_sampler =
@@ -2550,6 +2597,9 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 			case 0x4002:
 				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomSubmixer);
 				break;
+			case 0x4003:
+				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomMastering);
+				break;
 			default: EXIT("unknown rack_id: 0x%" PRIx32 "\n", rack_id);
 		}
 
@@ -2587,6 +2637,25 @@ int KYTY_SYSV_ABI Ngs2VoiceGetState(uintptr_t voice_handle, Ngs2VoiceState* stat
 	Common::LockGuard lock(voice->rack->ngs->mutex);
 
 	switch (voice->rack->type) {
+		case Ngs2RackType::Submixer: {
+			EXIT_NOT_IMPLEMENTED(state_size != sizeof(Ngs2SubmixerVoiceState));
+			auto* submixer = reinterpret_cast<Ngs2SubmixerVoiceState*>(state);
+			*submixer     = {};
+			submixer->voice_state.state_flags = Ngs2GetStateFlags(voice);
+			LOGF("\t state_flags = %u\n", submixer->voice_state.state_flags);
+			break;
+		}
+		case Ngs2RackType::CustomMastering: {
+			const auto configured_size =
+			    voice->rack->option.custom_mastering.custom_rack_option.state_size;
+			EXIT_NOT_IMPLEMENTED(configured_size < sizeof(Ngs2CustomMasteringVoiceState));
+			EXIT_NOT_IMPLEMENTED(state_size != configured_size);
+			std::memset(state, 0, state_size);
+			auto* mastering = reinterpret_cast<Ngs2CustomMasteringVoiceState*>(state);
+			mastering->voice_state.state_flags = Ngs2GetStateFlags(voice);
+			LOGF("\t state_flags = %u\n", mastering->voice_state.state_flags);
+			break;
+		}
 		case Ngs2RackType::Sampler:
 		case Ngs2RackType::CustomSampler: {
 			if (state_size != sizeof(Ngs2SamplerVoiceState)) {

src/libs/controller.cpp

@@ -85,8 +85,6 @@ private:
 
 static GameController* g_controller = nullptr;
 
-static constexpr int KEYBOARD_CONTROLLER_ID = -1000;
-
 static uint8_t pad_connected_count_to_u8(int connected_count) {
 	return static_cast<uint8_t>(connected_count > 255 ? 255 : connected_count);
 }
@@ -117,7 +115,7 @@ KYTY_SUBSYSTEM_INIT(Controller) {
 	EXIT_IF(g_controller != nullptr);
 
 	g_controller = new GameController;
-	g_controller->Connect(KEYBOARD_CONTROLLER_ID);
+	g_controller->Connect(HOST_INPUT_CONTROLLER_ID);
 }
 
 KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Controller) {}
@@ -155,7 +153,7 @@ void GameController::CheckActive() {
 	if (!m_connected_ids.empty()) {
 		new_active_id = m_connected_ids[0];
 		for (const auto id: m_connected_ids) {
-			if (id != KEYBOARD_CONTROLLER_ID) {
+			if (id != HOST_INPUT_CONTROLLER_ID) {
 				new_active_id = id;
 				break;
 			}

src/libs/controller.h

@@ -9,6 +9,8 @@ namespace Libs::Controller {
 
 KYTY_SUBSYSTEM_DEFINE(Controller);
 
+constexpr int HOST_INPUT_CONTROLLER_ID = -1000;
+
 constexpr uint32_t PAD_BUTTON_L3        = 0x00000002;
 constexpr uint32_t PAD_BUTTON_R3        = 0x00000004;
 constexpr uint32_t PAD_BUTTON_OPTIONS   = 0x00000008;

src/libs/libAudio2.cpp

@@ -336,18 +336,6 @@ static AudioOut2PortStateEntry* audioout2_find_port_locked(AudioOut2PortHandle p
 	return nullptr;
 }
 
-static uint32_t audioout2_context_grains(AudioOut2ContextHandle ctx) {
-	uint32_t samples_num = 512;
-
-	g_audioout2_context_mutex.Lock();
-	if (auto* state = audioout2_find_context_locked(ctx); state != nullptr) {
-		samples_num = (state->num_grains == 0 ? 512u : state->num_grains);
-	}
-	g_audioout2_context_mutex.Unlock();
-
-	return samples_num;
-}
-
 static void audioout2_queue_context_audio(AudioOut2ContextHandle ctx, bool blocking) {
 	std::vector<AudioInternal::OutputParam> params;
 	params.reserve(AudioInternal::OUT_PORTS_MAX);
@@ -455,6 +443,12 @@ int KYTY_SYSV_ABI AudioOut2ContextDestroy(AudioOut2ContextHandle ctx) {
 	PRINT_NAME();
 	LOGF("\t ctx = 0x%016" PRIx64 "\n", ctx);
 
+	g_audioout2_context_mutex.Lock();
+	if (auto* state = audioout2_find_context_locked(ctx); state != nullptr) {
+		*state = AudioOut2ContextState {};
+	}
+	g_audioout2_context_mutex.Unlock();
+
 	std::array<int, 256> audio_handles {};
 	size_t               audio_handles_num = 0;
 
@@ -473,12 +467,6 @@ int KYTY_SYSV_ABI AudioOut2ContextDestroy(AudioOut2ContextHandle ctx) {
 		audioout2_close_audio_handle(audio_handles[i]);
 	}
 
-	g_audioout2_context_mutex.Lock();
-	if (auto* state = audioout2_find_context_locked(ctx); state != nullptr) {
-		*state = AudioOut2ContextState {};
-	}
-	g_audioout2_context_mutex.Unlock();
-
 	return OK;
 }
 
@@ -559,30 +547,45 @@ int KYTY_SYSV_ABI AudioOut2PortCreate(AudioOut2ContextHandle ctx, const AudioOut
 	EXIT_NOT_IMPLEMENTED(params == nullptr);
 	EXIT_NOT_IMPLEMENTED(port == nullptr);
 
-	const auto next_port = g_audioout2_next_port.fetch_add(1, std::memory_order_relaxed);
+	const auto next_port    = g_audioout2_next_port.fetch_add(1, std::memory_order_relaxed);
+	const auto audio_format = audioout2_data_format_to_audio_format(params->data_format);
+	const auto audio_type   = audioout2_port_type_to_audio_out_type(params->port_type);
+
+	g_audioout2_context_mutex.Lock();
+	const auto* context_state = audioout2_find_context_locked(ctx);
+	if (context_state == nullptr) {
+		g_audioout2_context_mutex.Unlock();
+		return AUDIO_OUT2_ERROR_INVALID_PARAM;
+	}
+	const auto samples_num = context_state->num_grains == 0 ? 512u : context_state->num_grains;
 
 	g_audioout2_port_mutex.Lock();
-	auto* port_state = audioout2_find_port_locked(0);
-	if (port_state == nullptr) {
-		for (auto& candidate: g_audioout2_ports) {
-			if (!candidate.used) {
-				port_state = &candidate;
-				break;
-			}
+	AudioOut2PortStateEntry* port_state = nullptr;
+	for (auto& candidate: g_audioout2_ports) {
+		if (!candidate.used) {
+			port_state = &candidate;
+			break;
 		}
 	}
+	if (port_state != nullptr) {
+		*port_state               = AudioOut2PortStateEntry {};
+		port_state->used          = true;
+		port_state->handle        = next_port;
+		port_state->context       = ctx;
+		port_state->port_type     = params->port_type;
+		port_state->data_format   = params->data_format;
+		port_state->sampling_freq = params->sampling_freq;
+		port_state->samples_num   = samples_num;
+		port_state->audio_format  = audio_format;
+	}
 	g_audioout2_port_mutex.Unlock();
+	g_audioout2_context_mutex.Unlock();
 
-	if (next_port > g_audioout2_ports.size() || port_state == nullptr) {
+	if (port_state == nullptr) {
 		return AUDIO_OUT2_ERROR_PORT_FULL;
 	}
 
-	*port = next_port;
-
-	const auto samples_num  = audioout2_context_grains(ctx);
-	const auto audio_format = audioout2_data_format_to_audio_format(params->data_format);
-	const auto audio_type   = audioout2_port_type_to_audio_out_type(params->port_type);
-	int        audio_handle = 0;
+	int audio_handle = 0;
 
 	if (audio_format != AudioInternal::Format::Unknown &&
 	    !audioout2_port_type_is_object(params->port_type)) {
@@ -591,17 +594,17 @@ int KYTY_SYSV_ABI AudioOut2PortCreate(AudioOut2ContextHandle ctx, const AudioOut
 	}
 
 	g_audioout2_port_mutex.Lock();
-	*port_state               = AudioOut2PortStateEntry {};
-	port_state->used          = true;
-	port_state->handle        = *port;
-	port_state->context       = ctx;
-	port_state->port_type     = params->port_type;
-	port_state->data_format   = params->data_format;
-	port_state->sampling_freq = params->sampling_freq;
-	port_state->samples_num   = samples_num;
-	port_state->audio_format  = audio_format;
-	port_state->audio_handle  = audio_handle;
+	const bool reserved = port_state->used && port_state->handle == next_port;
+	if (reserved) {
+		port_state->audio_handle = audio_handle;
+	}
 	g_audioout2_port_mutex.Unlock();
+	if (!reserved) {
+		audioout2_close_audio_handle(audio_handle);
+		return AUDIO_OUT2_ERROR_INVALID_PARAM;
+	}
+
+	*port = next_port;
 
 	if (next_port <= 16 || (next_port % 600) == 0) {
 		PRINT_NAME();

src/libs/libC.cpp

@@ -633,6 +633,15 @@ LIB_VERSION("LibcInternalExt", 1, "LibcInternal", 1, 1);
 static uint64_t g_mspace_atomic_id_mask = 0;
 static uint64_t g_mstate_table[64]      = {0};
 
+using thread_atexit_destructor_t = KYTY_SYSV_ABI void (*)(void*);
+
+struct ThreadAtexitDestructor {
+	thread_atexit_destructor_t destructor;
+	void*                      object;
+};
+
+static thread_local std::vector<ThreadAtexitDestructor> g_thread_atexit_destructors;
+
 struct Info {
 	uint64_t  size;
 	uint32_t  unknown1;
@@ -650,25 +659,29 @@ void KYTY_SYSV_ABI LibcHeapGetTraceInfo(Info* info) {
 	info->mstate_table          = g_mstate_table;
 }
 
-uint64_t KYTY_SYSV_ABI LibcInternalExtUnknownQBS714Jr3g(uint64_t arg0, uint64_t arg1, uint64_t arg2,
-                                                        uint64_t arg3, uint64_t arg4,
-                                                        uint64_t arg5) {
+int KYTY_SYSV_ABI LibcInternalExtCxaThreadAtexit(thread_atexit_destructor_t destructor, void* object,
+                                                 void* /*module_id*/) {
 	PRINT_NAME();
 
-	LOGF("\t arg0 = 0x%016" PRIx64 "\n"
-	     "\t arg1 = 0x%016" PRIx64 "\n"
-	     "\t arg2 = 0x%016" PRIx64 "\n"
-	     "\t arg3 = 0x%016" PRIx64 "\n"
-	     "\t arg4 = 0x%016" PRIx64 "\n"
-	     "\t arg5 = 0x%016" PRIx64 "\n",
-	     arg0, arg1, arg2, arg3, arg4, arg5);
+	g_thread_atexit_destructors.push_back({destructor, object});
 
 	return 0;
 }
 
+void RunThreadAtexitDestructors() {
+	while (!g_thread_atexit_destructors.empty()) {
+		auto destructor = g_thread_atexit_destructors.back();
+		g_thread_atexit_destructors.pop_back();
+
+		if (destructor.destructor != nullptr) {
+			destructor.destructor(destructor.object);
+		}
+	}
+}
+
 LIB_DEFINE(InitLibcInternalExt_1) {
 	LIB_FUNC("NWtTN10cJzE", LibcInternalExt::LibcHeapGetTraceInfo);
-	LIB_FUNC("qBS714-Jr3g", LibcInternalExt::LibcInternalExtUnknownQBS714Jr3g);
+	LIB_FUNC("qBS714-Jr3g", LibcInternalExt::LibcInternalExtCxaThreadAtexit);
 }
 
 } // namespace LibcInternalExt

src/libs/libGraphicsDriver.cpp

@@ -159,6 +159,7 @@ LIB_DEFINE(InitGraphicsDriver_1) {
 	LIB_FUNC("f3dg2CSgRKY", Gen5::GraphicsCreateShader);
 	LIB_FUNC("dolOmWH+huQ", Gen5::GraphicsUnknownGetFusedShaderSize);
 	LIB_FUNC("fd5Bp5tGTgo", Gen5::GraphicsUnknownFuseShaderHalves);
+	LIB_FUNC("nApJjpKNBl4", Gen5::GraphicsUnknownNApJjpKNBl4);
 	LIB_FUNC("vcmNN+AAXnY", Gen5::GraphicsSetCxRegIndirectPatchSetAddress);
 	LIB_FUNC("Qrj4c+61z4A", Gen5::GraphicsSetShRegIndirectPatchSetAddress);
 	LIB_FUNC("6lNcCp+fxi4", Gen5::GraphicsSetUcRegIndirectPatchSetAddress);
@@ -177,7 +178,8 @@ LIB_DEFINE(InitGraphicsDriver_1) {
 	LIB_FUNC("bxca0BK4FNg", Gen5::GraphicsUnknownJumpPatchSetTarget);
 	LIB_FUNC("2BS4EtAaF28", Gen5::GraphicsJumpPatchSetTarget);
 	LIB_FUNC("h9z6+0hEydk", Gen5::GraphicsSuspendPoint);
-	LIB_FUNC("qj7QZpgr9Uw", Gen5::GraphicsUnknownQj7QZpgr9Uw);
+	LIB_FUNC("qj7QZpgr9Uw", Gen5::GraphicsDcbContextStateOp);
+	LIB_FUNC("H6vHS5cidSA", Gen5::GraphicsDcbContextStateOpGetSize);
 	LIB_FUNC("BfBDZGbti7A", Gen5::GraphicsGetIsTrinityMode);
 	LIB_FUNC("dbOlWdppb4o", LibGen5::Graphics5UnknownDb);
 
@@ -207,6 +209,7 @@ LIB_DEFINE(InitGraphicsDriver_1) {
 	LIB_FUNC("ewobAQeMo5k", Gen5::GraphicsAcbAcquireMemGetSize);
 	LIB_FUNC("qyM2bxYFPAk", Gen5::GraphicsAcbCondExec);
 	LIB_FUNC("ozKzBP4aki4", Gen5::GraphicsAcbCondExecGetSize);
+	LIB_FUNC("b-oySn+G2tE", Gen5::GraphicsAcbJumpGetSize);
 	LIB_FUNC("htn36gPnBk4", Gen5::GraphicsAcbWaitRegMem);
 	LIB_FUNC("-RnpfpxIhec", Gen5::GraphicsAcbDmaData);
 	LIB_FUNC("qzMN2XKGA4k", Gen5::GraphicsAcbCopyData);
@@ -282,6 +285,7 @@ LIB_DEFINE(InitGraphicsDriver_1) {
 	LIB_FUNC("YWTKOju587o", Gen5::GraphicsCondExecPatchSetCommandAddress);
 	LIB_FUNC("k-JpyR2dYAM", Gen5::GraphicsCondExecPatchSetEnd);
 	LIB_FUNC("3ZWa3AoyWZQ", Gen5::GraphicsCondExecPatchSetCommandAddress);
+	LIB_FUNC("ziVA3whp3p4", Gen5::GraphicsRewindPatchSetRewindState);
 	LIB_FUNC("YUeqkyT7mEQ", Gen5::GraphicsDcbSetFlip);
 }
 

src/libs/libKernel.cpp

@@ -2021,6 +2021,12 @@ int64_t KYTY_SYSV_ABI fstat(int d, LibKernel::FileSystem::FileStat* sb) {
 	return POSIX_N_CALL(LibKernel::FileSystem::KernelFstat(d, sb));
 }
 
+int KYTY_SYSV_ABI ftruncate(int d, int64_t length) {
+	PRINT_NAME();
+
+	return POSIX_CALL(LibKernel::FileSystem::KernelFtruncate(d, length));
+}
+
 int KYTY_SYSV_ABI socket(int family, int type, int protocol) {
 	PRINT_NAME();
 	return Network::Net::Socket(family, type, protocol);
@@ -2159,6 +2165,7 @@ LIB_DEFINE(InitLibKernel_1_Posix) {
 	LIB_FUNC("yS8U2TGCe1A", nanosleep);
 	LIB_FUNC("E6ao34wPw+U", stat);
 	LIB_FUNC("JGMio+21L4c", mkdir);
+	LIB_FUNC("ih4CD9-gghM", Posix::ftruncate);
 	LIB_FUNC("pDuPEf3m4fI", Posix::sem_init);
 	LIB_FUNC("cDW233RAwWo", Posix::sem_destroy);
 	LIB_FUNC("YCV5dGGBcCo", Posix::sem_wait);

src/libs/libNet.cpp

@@ -97,6 +97,10 @@ int KYTY_SYSV_ABI NetShutdown(int s, int how) {
 	return FinishSocketCall(Net::Shutdown(s, how));
 }
 
+int KYTY_SYSV_ABI NetGetsockname(int s, void* addr, uint32_t* addrlen) {
+	return FinishSocketCall(Net::Getsockname(s, addr, addrlen));
+}
+
 int KYTY_SYSV_ABI NetPoolCreate(const char* name, int size, int flags) {
 	return NET_CALL(Net::NetPoolCreate(name, size, flags));
 }
@@ -196,6 +200,7 @@ LIB_DEFINE(InitNet_1_Net) {
 	LIB_FUNC("v6M4txecCuo", LibNet::NetEtherNtostr);
 	LIB_FUNC("6Oc0bLsIYe0", LibNet::NetGetMacAddress);
 	LIB_FUNC("hLuXdjHnhiI", LibNet::NetGetSockInfo);
+	LIB_FUNC("hoOAofhhRvE", LibNet::NetGetsockname);
 	LIB_FUNC("SF47kB2MNTo", LibNet::NetEpollCreate);
 	LIB_FUNC("ZVw46bsasAk", LibNet::NetEpollControl);
 	LIB_FUNC("drjIbDbA7UQ", LibNet::NetEpollWait);

src/libs/libRtc.cpp

@@ -4,6 +4,7 @@
 #include "libs/libs.h"
 #include "loader/symbolDatabase.h"
 
+#include <chrono>
 #include <cstdint>
 #include <cstdio>
 #include <limits>
@@ -247,20 +248,12 @@ static int KYTY_SYSV_ABI RtcGetCurrentTick(RtcTick* tick) {
 		return RTC_ERROR_DATETIME_UNINITIALIZED;
 	}
 
-	const auto now  = Common::DateTime::FromSystemUTC();
-	const auto date = now.GetDate();
-	const auto tod  = now.GetTime();
-
-	RtcDateTime time {};
-	time.year        = static_cast<uint16_t>(date.Year());
-	time.month       = static_cast<uint16_t>(date.Month());
-	time.day         = static_cast<uint16_t>(date.Day());
-	time.hour        = static_cast<uint16_t>(tod.Hour24());
-	time.minute      = static_cast<uint16_t>(tod.Minute());
-	time.second      = static_cast<uint16_t>(tod.Second());
-	time.microsecond = static_cast<uint32_t>(tod.Msec() * 1000);
-
-	return RtcGetTick(&time, tick);
+	const auto now_us =
+	    static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::microseconds>(
+	                              std::chrono::system_clock::now().time_since_epoch())
+	                              .count());
+	tick->tick = RTC_UNIX_EPOCH_TICKS + now_us;
+	return OK;
 }
 
 static int KYTY_SYSV_ABI RtcGetCurrentNetworkTick(RtcTick* tick) {

src/libs/libSaveData.cpp

@@ -375,7 +375,8 @@ int KYTY_SYSV_ABI SaveDataDirNameSearch(const SaveDataDirNameSearchCond* cond,
 
 	if (Common::File::IsDirectoryExisting(root)) {
 		for (const auto& entry: Common::File::GetDirEntries(root)) {
-			if (!entry.is_file && !Common::StartsWith(entry.name, "sce_")) {
+			if (!entry.is_file && entry.name != "." && entry.name != ".." &&
+			    !Common::StartsWith(entry.name, "sce_")) {
 				if (cond->dir_name == nullptr || cond->dir_name->data[0] == '\0' ||
 				    dir_name_match(Common::ToLower(entry.name).c_str(),
 				                   Common::ToLower(std::string(cond->dir_name->data)).c_str())) {

src/libs/libSystemService.cpp

@@ -1,6 +1,7 @@
 #include "common/abi.h"
 #include "common/assert.h"
 #include "common/common.h"
+#include "common/emulatorConfig.h"
 #include "common/logging/log.h"
 #include "common/stringUtils.h"
 #include "libs/errno.h"
@@ -25,37 +26,6 @@ namespace SystemService {
 [[maybe_unused]] constexpr int PARAM_ID_SCREEN_READER       = 208;
 [[maybe_unused]] constexpr int PARAM_ID_ENTER_BUTTON_ASSIGN = 1000;
 
-[[maybe_unused]] constexpr int PARAM_LANG_JAPANESE      = 0;
-[[maybe_unused]] constexpr int PARAM_LANG_ENGLISH_US    = 1;
-[[maybe_unused]] constexpr int PARAM_LANG_FRENCH        = 2;
-[[maybe_unused]] constexpr int PARAM_LANG_SPANISH       = 3;
-[[maybe_unused]] constexpr int PARAM_LANG_GERMAN        = 4;
-[[maybe_unused]] constexpr int PARAM_LANG_ITALIAN       = 5;
-[[maybe_unused]] constexpr int PARAM_LANG_DUTCH         = 6;
-[[maybe_unused]] constexpr int PARAM_LANG_PORTUGUESE_PT = 7;
-[[maybe_unused]] constexpr int PARAM_LANG_RUSSIAN       = 8;
-[[maybe_unused]] constexpr int PARAM_LANG_KOREAN        = 9;
-[[maybe_unused]] constexpr int PARAM_LANG_CHINESE_T     = 10;
-[[maybe_unused]] constexpr int PARAM_LANG_CHINESE_S     = 11;
-[[maybe_unused]] constexpr int PARAM_LANG_FINNISH       = 12;
-[[maybe_unused]] constexpr int PARAM_LANG_SWEDISH       = 13;
-[[maybe_unused]] constexpr int PARAM_LANG_DANISH        = 14;
-[[maybe_unused]] constexpr int PARAM_LANG_NORWEGIAN     = 15;
-[[maybe_unused]] constexpr int PARAM_LANG_POLISH        = 16;
-[[maybe_unused]] constexpr int PARAM_LANG_PORTUGUESE_BR = 17;
-[[maybe_unused]] constexpr int PARAM_LANG_ENGLISH_GB    = 18;
-[[maybe_unused]] constexpr int PARAM_LANG_TURKISH       = 19;
-[[maybe_unused]] constexpr int PARAM_LANG_SPANISH_LA    = 20;
-[[maybe_unused]] constexpr int PARAM_LANG_ARABIC        = 21;
-[[maybe_unused]] constexpr int PARAM_LANG_FRENCH_CA     = 22;
-[[maybe_unused]] constexpr int PARAM_LANG_CZECH         = 23;
-[[maybe_unused]] constexpr int PARAM_LANG_HUNGARIAN     = 24;
-[[maybe_unused]] constexpr int PARAM_LANG_GREEK         = 25;
-[[maybe_unused]] constexpr int PARAM_LANG_ROMANIAN      = 26;
-[[maybe_unused]] constexpr int PARAM_LANG_THAI          = 27;
-[[maybe_unused]] constexpr int PARAM_LANG_VIETNAMESE    = 28;
-[[maybe_unused]] constexpr int PARAM_LANG_INDONESIAN    = 29;
-
 [[maybe_unused]] constexpr int PARAM_DATE_FORMAT_YYYYMMDD = 0;
 [[maybe_unused]] constexpr int PARAM_DATE_FORMAT_DDMMYYYY = 1;
 [[maybe_unused]] constexpr int PARAM_DATE_FORMAT_MMDDYYYY = 2;
@@ -121,7 +91,7 @@ static int KYTY_SYSV_ABI SystemServiceParamGetInt(int param_id, int* value) {
 	int v = 0;
 
 	switch (param_id) {
-		case PARAM_ID_LANG: v = PARAM_LANG_ENGLISH_US; break;
+		case PARAM_ID_LANG: v = static_cast<int>(Config::GetConsoleLanguage()); break;
 		case PARAM_ID_DATE_FORMAT: v = PARAM_DATE_FORMAT_DDMMYYYY; break;
 		case PARAM_ID_TIME_FORMAT: v = PARAM_TIME_FORMAT_24HOUR; break;
 		case PARAM_ID_TIME_ZONE: v = +180; break;

src/loader/runtimeLinker.cpp

@@ -38,7 +38,10 @@
 #include <windows.h>
 #else
 #include <dlfcn.h>
-#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+#if defined(__APPLE__)
+#include <mach/mach.h>
+#include <mach/mach_vm.h>
+#elif KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 #include <sys/uio.h>
 #include <unistd.h>
 #endif
@@ -145,7 +148,7 @@ static bool PatchGuestMemory64(uint64_t vaddr, uint64_t value) {
 }
 
 static uint64_t AllocateUnresolvedImportThunk(uint64_t record_id) {
-	constexpr uint64_t thunk_size = 162;
+	constexpr uint64_t thunk_size = 165;
 
 	if (g_unresolved_stub_thunk_pages.empty() ||
 	    g_unresolved_stub_thunk_offset + thunk_size > UNRESOLVED_STUB_PAGE_SIZE) {
@@ -251,6 +254,10 @@ static uint64_t AllocateUnresolvedImportThunk(uint64_t record_id) {
 	emit(0x41);
 	emit(0xff);
 	emit(0xe3); // jmp r11
+	// Match the integer fallback for floating-point return values.
+	emit(0x0f);
+	emit(0x57);
+	emit(0xc0); // xorps xmm0, xmm0
 	emit(0x31);
 	emit(0xc0); // xor eax, eax
 	emit(0xc3); // ret
@@ -353,9 +360,10 @@ static KYTY_SYSV_ABI void RunEntry(uint64_t addr, EntryParams* params, atexit_fu
 	auto* func = reinterpret_cast<entry_func_t>(addr);
 
 	if (stack_top != nullptr) {
-		const auto guest_rsp =
+		const auto aligned_stack_top =
 		    reinterpret_cast<uintptr_t>(stack_top) & ~static_cast<uintptr_t>(0x0f);
-		const auto guest_rbp = guest_rsp - 4u * sizeof(uint64_t);
+		const auto guest_rsp = aligned_stack_top - 2u * sizeof(uintptr_t);
+		const auto guest_rbp = guest_rsp;
 
 		auto* guest_root_frame = reinterpret_cast<uintptr_t*>(guest_rbp);
 		guest_root_frame[0]    = 0;
@@ -500,6 +508,13 @@ struct MainEntryStackTestState {
 static KYTY_SYSV_ABI void TestMainEntryStackCallback(EntryParams* params,
                                                      atexit_func_t /*atexit_func*/) {
 	auto* state = reinterpret_cast<MainEntryStackTestState*>(const_cast<char*>(params->argv[0]));
+	asm volatile("pushq %%r15\n\t"
+	             "pushq %%r14\n\t"
+	             "popq %%r14\n\t"
+	             "popq %%r15\n\t"
+	             :
+	             :
+	             : "memory");
 	asm volatile("movq %%rsp, %0" : "=r"(state->rsp) : : "memory");
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	asm volatile("movq %%gs:0x08, %0\n\t"
@@ -522,6 +537,8 @@ bool TestMainEntryUsesGuestStack() {
 	MainEntryStackTestState state {};
 	EntryParams             params {};
 	params.argv[0] = reinterpret_cast<const char*>(&state);
+	std::memset(reinterpret_cast<void*>(stack_base), 0xcd, stack_size);
+	auto* root_frame = reinterpret_cast<const uintptr_t*>(stack_base + stack_size) - 2;
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	uintptr_t original_teb_stack_base  = 0;
@@ -551,9 +568,10 @@ bool TestMainEntryUsesGuestStack() {
 	constexpr bool teb_ok = true;
 #endif
 
-	const bool rsp_ok = state.rsp >= stack_base && state.rsp < stack_base + stack_size;
-	const bool freed  = Libs::LibKernel::Memory::FreeGuestMemory(stack_base, stack_size);
-	return state.called && rsp_ok && teb_ok && freed;
+	const bool rsp_ok  = state.rsp >= stack_base && state.rsp < stack_base + stack_size;
+	const bool root_ok = root_frame[0] == 0 && root_frame[1] == 0;
+	const bool freed   = Libs::LibKernel::Memory::FreeGuestMemory(stack_base, stack_size);
+	return state.called && rsp_ok && root_ok && teb_ok && freed;
 }
 
 bool TestModuleRelocationUsesWritableHostMapping() {
@@ -722,7 +740,26 @@ static bool IsReadableRange(uint64_t addr, uint64_t size) {
 		}
 		current = std::min(region_end, end);
 	}
-#elif KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+#elif defined(__APPLE__)
+	// Walk the Mach regions covering the range and require read permission. The fatal
+	// report dumps memory behind raw register values, and a fault inside the reporter
+	// re-enters the signal handler and wedges the reporting thread.
+	uint64_t current = addr;
+	while (current < end) {
+		mach_vm_address_t              region_addr = current;
+		mach_vm_size_t                 region_size = 0;
+		vm_region_basic_info_data_64_t info {};
+		mach_msg_type_number_t         count       = VM_REGION_BASIC_INFO_COUNT_64;
+		mach_port_t                    object_name = MACH_PORT_NULL;
+		if (mach_vm_region(mach_task_self(), &region_addr, &region_size, VM_REGION_BASIC_INFO_64,
+		                   reinterpret_cast<vm_region_info_t>(&info), &count,
+		                   &object_name) != KERN_SUCCESS ||
+		    region_addr > current || (info.protection & VM_PROT_READ) == 0) {
+			return false;
+		}
+		current = region_addr + region_size;
+	}
+#elif KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 	const auto page_size = static_cast<uint64_t>(sysconf(_SC_PAGESIZE));
 	if (page_size == 0) {
 		return false;
@@ -752,7 +789,7 @@ static bool IsReadableRange(uint64_t addr, uint64_t size) {
 }
 
 static bool IsDumpableRange(uint64_t addr, uint64_t size) {
-#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 	return IsReadableRange(addr, size);
 #else
 	(void)size;