loader: preserve guest root stack frame

Extracted from Stefanos Costa's pthread fix in #147.

.github/workflows/build.yml

@@ -73,7 +73,7 @@ jobs:
       - name: Configure
         shell: cmd
         run: |
-          cmake -S src -B _Build/windows ^
+          cmake -S . -B _Build/windows ^
             -G Ninja ^
             -DCMAKE_BUILD_TYPE=Release ^
             -DCMAKE_C_COMPILER=clang-cl ^
@@ -83,7 +83,12 @@ jobs:
       - name: Build
         shell: cmd
         run: |
-          cmake --build _Build/windows --target launcher --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 "^(audio_out2_port|virtual_memory_allocation)$"
 
       - name: Install
         shell: cmd
@@ -142,7 +147,7 @@ jobs:
       - name: Configure
         shell: bash
         run: |
-          cmake -S src -B _Build/macos \
+          cmake -S . -B _Build/macos \
             -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
             -DCMAKE_OSX_ARCHITECTURES=x86_64 \
@@ -153,7 +158,15 @@ jobs:
       - name: Build
         shell: bash
         run: |
-          cmake --build _Build/macos --target launcher --parallel
+          cmake --build _Build/macos \
+            --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 '^(audio_out2_port|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -257,7 +270,7 @@ 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_C_COMPILER=clang \
@@ -284,13 +297,14 @@ jobs:
         run: |
           cmake --build _Build/linux \
             --target launcher page_manager_tests memory_tracker_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)$'
+            -R '^(audio_out2_port|page_manager|memory_tracker|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash

.gitignore

@@ -4,3 +4,4 @@
 build/
 _Build/vscode-clang/
 _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

@@ -6,6 +6,10 @@ 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()
@@ -19,13 +23,11 @@ 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(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 
 include(TestBigEndian)
@@ -103,13 +105,13 @@ include_directories(
 	${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
   )
 
@@ -117,11 +119,11 @@ find_package(Git)
 
 add_custom_target( KytyGitVersion
     COMMAND ${CMAKE_COMMAND}
-	-D INPUT_FILE=${CMAKE_CURRENT_SOURCE_DIR}/kytyGitVersion.h.in
-    -D OUTPUT_FILE=${CMAKE_CURRENT_BINARY_DIR}/kytyGitVersion.h
+	-D INPUT_FILE=${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 ${CMAKE_CURRENT_SOURCE_DIR}/generate_version.cmake
+	-P ${KYTY_SOURCE_DIR}/generate_version.cmake
 	COMMENT "Generate kytyGitVersion.h"
 )
 
@@ -143,51 +145,51 @@ option(KYTY_BUILD_LAUNCHER "Build Qt launcher" ON)
 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 "${PROJECT_BINARY_DIR}/gpu_tiler_shaders")
 set(gpu_tiler_shader_names
 	standard256
@@ -215,7 +217,7 @@ foreach(shader_name IN LISTS gpu_tiler_shader_names)
 		COMMAND "${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}" ${gpu_tiler_shader_includes}
 		VERBATIM
 	)
@@ -239,7 +241,7 @@ foreach(shader_source IN LISTS gpu_blit_shader_sources)
 		COMMAND "${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 +250,8 @@ 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)
 
@@ -280,7 +282,7 @@ 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
@@ -298,7 +300,7 @@ if (KYTY_CLANG_CL)
 endif()
 
 list(APPEND check_headers
-	${CMAKE_CURRENT_SOURCE_DIR}
+	${KYTY_SOURCE_DIR}
 )
 
 function(add_kyty_full_emulator_test target source)
@@ -312,90 +314,116 @@ 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()
 
-add_kyty_full_emulator_test(shader_cfg_tests ../tests/shaderCfgTests.cpp)
+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 "${KYTY_SOURCE_DIR}/kernel/macosGuestAddressSpace.cpp")
+		target_compile_definitions(${target} PRIVATE KYTY_LINKED_GUEST_ADDRESS_SPACE=1)
+		target_link_options(${target} PRIVATE
+			-Wl,-ld_classic,-no_pie,-no_fixup_chains,-no_huge,-pagezero_size,0x40000,-segaddr,SYSTEM_MANAGED,0x40000,-segaddr,SYSTEM_RESERVED,0x7ffffc000,-segaddr,USER_AREA,0x7000000000,-image_base,0x700000000000)
+	endif()
+endfunction()
+
+add_kyty_full_emulator_test(shader_cfg_tests "${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/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})
-target_compile_definitions(memory_tracker_tests PRIVATE KYTY_MEMORY_TRACKER_TESTS=1)
 
 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 "${PROJECT_BINARY_DIR}/gpu_test_shaders")
 set(gpu_test_ms_depth_source
@@ -411,14 +439,14 @@ add_custom_command(
 	        -o "${gpu_test_ms_depth_spv}" "${gpu_test_ms_depth_source}"
 	COMMAND ${CMAKE_COMMAND} -DINPUT=${gpu_test_ms_depth_spv}
 	        -DOUTPUT=${gpu_test_ms_depth_header} -DSYMBOL=GPU_TEST_MS_DEPTH_SPV
-	        -P "${CMAKE_CURRENT_SOURCE_DIR}/embed_spirv.cmake"
+	        -P "${KYTY_SOURCE_DIR}/embed_spirv.cmake"
 	DEPENDS "${gpu_test_ms_depth_source}"
 	VERBATIM
 )
 target_sources(shader_recompiler_compute_tests PRIVATE
 	"${gpu_test_ms_depth_header}")
 
-add_kyty_full_emulator_test(virtual_memory_allocation_tests ../tests/VirtualMemoryAllocationTests.cpp)
+add_kyty_full_emulator_test(virtual_memory_allocation_tests "${KYTY_TESTS_DIR}/VirtualMemoryAllocationTests.cpp")
 target_compile_definitions(virtual_memory_allocation_tests PRIVATE
 	KYTY_VIRTUAL_MEMORY_ALLOCATION_TESTS=1)
 
@@ -431,12 +459,21 @@ 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>)
 	add_test(NAME command_scheduler_timeline
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --scheduler-only)
 	add_test(NAME stream_buffer_ring
@@ -466,10 +503,27 @@ if(BUILD_TESTING)
 		add_test(NAME buffer_cache_ranges
 			COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --buffer-cache-range-only)
 	endif()
+
+	add_custom_target(kyty_tests DEPENDS
+		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 ${kyty_emulator_link_libraries})
 if (WIN32)
@@ -498,7 +552,12 @@ set(KYTY_EMULATOR_MAP_LINK_PATH "${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=${KYTY_EMULATOR_MAP_LINK_PATH}")
@@ -513,14 +572,14 @@ if(APPLE)
 	# reverts to the hardened defaults and aborts when it first executes written code).
 	add_custom_command(TARGET kyty_emulator POST_BUILD
 		COMMAND codesign -s - --force --options runtime
-			--entitlements "${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

@@ -27,8 +27,9 @@ Development is focused on compatibility and boot reliability.
 Windows is the primary platform and receives the most testing. Linux builds and runs; see
 [Building on Linux](#building-on-linux).
 
-macOS support is experimental. Compatibility with the same games on Windows and macOS has not yet
-been tested.
+macOS support is experimental. The emulator is built for x86-64 and runs on Apple Silicon under
+Rosetta 2, with Vulkan provided by MoltenVK. A small number of titles have been verified in-game
+on Apple Silicon hardware; see [Building on macOS](#building-on-macos).
 
 ## Bugs and Issues
 
@@ -50,7 +51,7 @@ graphical glitches, low compatibility, and poor performance.
   </tr>
   <tr>
     <td align="center">
-      <strong>Minecraft Legends</strong><br>
+      <strong>Neptunia ReVerse</strong><br>
       <img src="docs/screenshots/ps5-04.png" width="300" alt="Minecraft Legends running in KytyPS5">
     </td>
     <td align="center">
@@ -58,8 +59,20 @@ graphical glitches, low compatibility, and poor performance.
       <img src="docs/screenshots/ps5-05.png" width="300" alt="SILENT HILL: The Short Message running in KytyPS5">
     </td>
   </tr>
+  <tr>
+    <td align="center">
+      <strong>Hellboy</strong><br>
+      <img src="docs/screenshots/ps5-02.png" width="300" alt="Disgaea 6 running in KytyPS5">
+    </td>
+    <td align="center">
+      <strong>Paleo Pines</strong><br>
+      <img src="docs/screenshots/ps5-06.png" width="300" alt="Dreaming Sarah running in KytyPS5">
+    </td>
+  </tr>
 </table>
 
+<p align="center"><em>And many more...</em></p>
+
 ## Contributing
 
 Testing games and submitting detailed bug reports are useful ways to contribute. Search existing
@@ -102,9 +115,10 @@ the Vulkan/SPIR-V validation rules.
 
 ### System requirements
 
-- Windows 10 version 1803, or a current Linux distribution
-- A 64-bit x86 processor
-- A Vulkan 1.3-capable GPU with current drivers
+- Windows 10 version 1803, a current Linux distribution, or macOS on Apple Silicon
+- A 64-bit x86 processor (on macOS, an Apple Silicon processor with Rosetta 2)
+- A Vulkan 1.3-capable GPU with current drivers (on macOS, Vulkan is provided by the bundled
+  MoltenVK)
 
 ### Build requirements (Windows)
 
@@ -127,7 +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:
@@ -160,7 +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"
 
@@ -174,7 +188,57 @@ 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
+
+macOS builds target x86-64 and run under Rosetta 2 on Apple Silicon, so the PS5's x86-64 game
+code executes through the same translation layer as the emulator itself. Prebuilt archives are
+attached to releases; the steps below are for building from source.
+
+Requirements:
+
+- An Apple Silicon Mac with Rosetta 2 installed (`softwareupdate --install-rosetta`)
+- Xcode (or the Command Line Tools)
+- Homebrew packages: `brew install cmake ninja glslang`
+- Qt 6 (Concurrent, Network, Widgets) with x86-64 support. The official Qt installation is
+  universal and works; Homebrew's Qt is arm64-only and will not link
+
+```bash
+git submodule update --init --recursive
+
+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"
+
+cmake --build _Build/macos --target launcher --parallel
+cmake --install _Build/macos --prefix _Build/macos/install
+```
+
+The build re-signs `kyty_emulator` with the JIT entitlements it needs to execute translated
+guest code; no manual signing step is required.
+
+Vulkan comes from MoltenVK. Download `MoltenVK-macos.tar` from the
+[MoltenVK releases](https://github.com/KhronosGroup/MoltenVK/releases), then copy
+`MoltenVK/dynamic/dylib/macOS/libMoltenVK.dylib` next to `kyty_emulator` and ad-hoc sign it:
+
+```bash
+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
 
@@ -221,6 +285,14 @@ The emulator can also be started directly with a legally obtained game directory
 ./_Build/linux/install/kyty_emulator --game "/games/ExampleGame"
 ```
 
+On macOS, point SDL at the MoltenVK library explicitly; the hardened runtime prevents it from
+being picked up from the executable's directory:
+
+```bash
+cd _Build/macos/install
+SDL_VULKAN_LIBRARY="$PWD/libMoltenVK.dylib" ./kyty_emulator --game "/games/ExampleGame"
+```
+
 Run `kyty_emulator --help` to see the available graphics, logging, validation, profiling, and
 debugging options.
 

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

@@ -37,6 +37,10 @@ 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;
 }
@@ -89,10 +93,6 @@ bool RenderDocEnabled() {
 	return g_config->renderdoc_enabled;
 }
 
-bool NggRectlistDrawEnabled() {
-	return g_config->ngg_rectlist_draw_enabled;
-}
-
 bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }

src/common/emulatorConfig.h

@@ -18,10 +18,14 @@ enum class ProfilerDirection { None, Network };
 
 enum class OutputDirection { Silent, Console, File };
 
+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;
 	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;
@@ -35,7 +39,6 @@ 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;
 };
 
@@ -44,6 +47,7 @@ void Load(const ConfigOptions& cfg);
 uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 uint32_t GetVblankFrequency();
+uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();
 
 bool                   ShaderValidationEnabled();
@@ -64,7 +68,6 @@ ProfilerDirection GetProfilerDirection();
 bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
-bool NggRectlistDrawEnabled();
 bool ReadbackLinearImagesEnabled();
 
 } // namespace Config

src/common/hostException.cpp

@@ -324,6 +324,10 @@ bool InstallHandler(Handler handler) {
 	sa.sa_sigaction = SignalHandler;
 	sa.sa_flags     = SA_SIGINFO;
 	sigemptyset(&sa.sa_mask);
+	// The guest signal-dispatch path (KernelRaiseException) interrupts threads with
+	// SIGUSR1; block it while a fault is being resolved so a stop-the-world request
+	// cannot preempt the handler between the protection fix and the retry.
+	sigaddset(&sa.sa_mask, SIGUSR1);
 
 	// macOS raises SIGBUS for protection faults on some paths and SIGSEGV on others;
 	// SIGILL covers instructions the host cannot execute (routed to the x64 emulator).

src/common/lruCache.h

@@ -49,8 +49,7 @@ public:
 
 	template <typename Function>
 	void ForEachItemBelow(Tick tick, Function&& function) {
-		constexpr bool ReturnsBool =
-		    std::is_same_v<std::invoke_result_t<Function, Object>, bool>;
+		constexpr bool ReturnsBool = std::is_same_v<std::invoke_result_t<Function, Object>, bool>;
 		for (auto* item = m_first; item != nullptr;) {
 			if (item->tick > tick) {
 				return;

src/common/platform/sysLinuxDbg.cpp

@@ -33,8 +33,7 @@ static bool OnOwnStack() {
 	}
 	void*      base = nullptr;
 	size_t     size = 0;
-	const bool ok =
-	    pthread_attr_getstack(&attr, &base, &size) == 0 && base != nullptr && size != 0;
+	const bool ok = pthread_attr_getstack(&attr, &base, &size) == 0 && base != nullptr && size != 0;
 	pthread_attr_destroy(&attr);
 	if (!ok) {
 		return false;

src/common/platform/sysLinuxFileIO.cpp

@@ -172,8 +172,7 @@ sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	return ret;
 }
 
-sys_file_t* SysFileOpenR(const std::filesystem::path& file_name,
-                         sys_file_cache_type_t cache_type) {
+sys_file_t* SysFileOpenR(const std::filesystem::path& file_name, sys_file_cache_type_t cache_type) {
 	auto* ret = new sys_file_t;
 
 	ret->type = SYS_FILE_FILE;
@@ -218,8 +217,7 @@ sys_file_t* SysFileCreate() {
 	return ret;
 }
 
-sys_file_t* SysFileOpenW(const std::filesystem::path& file_name,
-                         sys_file_cache_type_t cache_type) {
+sys_file_t* SysFileOpenW(const std::filesystem::path& file_name, sys_file_cache_type_t cache_type) {
 	auto* ret = new sys_file_t;
 
 	auto real_name     = get_internal_name(file_name);

src/common/platform/sysLinuxVirtual.cpp

@@ -136,8 +136,8 @@ static void* map_anonymous(uintptr_t addr, size_t size, int protect, int flags)
 			break;
 		}
 		const auto hint = (top - step) & ~(LOW_ARENA_GRAIN - 1);
-		void*      ptr  = mmap(reinterpret_cast<void*>(hint), size, protect,
-		                       flags | MAP_FIXED_NOREPLACE, -1, 0); // NOLINT
+		void* ptr = mmap(reinterpret_cast<void*>(hint), size, protect, flags | MAP_FIXED_NOREPLACE,
+		                 -1, 0); // NOLINT
 		if (ptr != MAP_FAILED) {
 			return ptr;
 		}
@@ -194,8 +194,8 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 	if (ptr != MAP_FAILED && ((ret_addr & (alignment - 1)) != 0)) {
 		munmap(ptr, size);
 
-		ptr      = map_anonymous(addr, size + alignment, protect,
-		                         MAP_PRIVATE | MAP_ANON | MAP_NORESERVE);
+		ptr =
+		    map_anonymous(addr, size + alignment, protect, MAP_PRIVATE | MAP_ANON | MAP_NORESERVE);
 		ret_addr = reinterpret_cast<uintptr_t>(ptr);
 		if (ptr != MAP_FAILED) {
 #if defined(__APPLE__)

src/common/threads.cpp

@@ -5,9 +5,9 @@
 
 #include <algorithm>
 #include <atomic>
+#include <cerrno>
 #include <chrono>             // IWYU pragma: keep
 #include <condition_variable> // IWYU pragma: keep
-#include <cerrno>
 #include <mutex>
 #include <vector>
 

src/common/uniqueFunction.h

@@ -20,9 +20,7 @@ class UniqueFunction {
 	public:
 		explicit Callable(Function function): m_function(std::move(function)) {}
 
-		Result Invoke(Args&&... args) override {
-			return m_function(std::forward<Args>(args)...);
-		}
+		Result Invoke(Args&&... args) override { return m_function(std::forward<Args>(args)...); }
 
 	private:
 		Function m_function;

src/common/virtualMemory.cpp

@@ -58,25 +58,6 @@ bool FlushInstructionCache(uint64_t address, uint64_t size) {
 	return SysVirtualFlushInstructionCache(address, size);
 }
 
-bool PatchReplace(uint64_t vaddr, uint64_t value) {
-	Mode old_mode {};
-	Protect(vaddr, 8, Mode::ReadWrite, &old_mode);
-
-	auto* ptr = reinterpret_cast<uint64_t*>(vaddr);
-
-	bool ret = (*ptr != value);
-
-	*ptr = value;
-
-	Protect(vaddr, 8, old_mode);
-
-	if (IsExecute(old_mode)) {
-		FlushInstructionCache(vaddr, 8);
-	}
-
-	return ret;
-}
-
 } // namespace VirtualMemory
 
 } // namespace Common

src/common/virtualMemory.h

@@ -37,7 +37,6 @@ bool     Free(uint64_t address);
 bool     FreeRange(uint64_t address, uint64_t size);
 bool     Protect(uint64_t address, uint64_t size, Mode mode, Mode* old_mode = nullptr);
 bool     FlushInstructionCache(uint64_t address, uint64_t size);
-bool     PatchReplace(uint64_t vaddr, uint64_t value);
 
 } // namespace VirtualMemory
 

src/emulator.cpp

@@ -105,7 +105,7 @@ static void ClearDebugTextureFolder() {
 	}
 }
 
-static void Init(const Config::ConfigOptions& cfg) {
+static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path& param_json) {
 	EXIT_IF(!Common::Thread::IsMainThread());
 
 	auto* slist = Common::SubsystemsList::Instance();
@@ -127,12 +127,21 @@ static void Init(const Config::ConfigOptions& cfg) {
 	slist->InitAll(true);
 
 	Config::Load(cfg);
+	slist->Add(log, {core, config});
+	slist->InitAll(true);
+
+	if (Common::File::IsFileExisting(param_json)) {
+		Loader::SystemContentLoadParamSfo(param_json);
+		if (const auto flexible_memory_size = Loader::SystemContentGetFlexibleMemorySize();
+		    flexible_memory_size != 0) {
+			Libs::LibKernel::Memory::SetFlexibleMemorySize(flexible_memory_size);
+		}
+	}
 
 	slist->Add(audio, {core, log, pthread, memory});
 	slist->Add(controller, {core, log, config});
 	slist->Add(file_system, {core, log, pthread});
 	slist->Add(graphics, {core, log, pthread, memory, config, profiler, controller});
-	slist->Add(log, {core, config});
 	slist->Add(memory, {core, log});
 	slist->Add(network, {core, log, pthread});
 	slist->Add(profiler, {core, config});
@@ -180,7 +189,8 @@ void Run(const RunOptions& options) {
 		EXIT("ELF is required\n");
 	}
 
-	Init(options.config);
+	const auto param_json = options.app0_dir / "sce_sys" / "param.json";
+	Init(options.config, param_json);
 
 	ClearDebugTextureFolder();
 
@@ -192,15 +202,6 @@ void Run(const RunOptions& options) {
 	Libs::LibKernel::FileSystem::Mount(options.app0_dir, "/app0");
 	Libs::LibKernel::FileSystem::Mount(options.app0_dir, "/hostapp");
 
-	auto param_json = options.app0_dir / "sce_sys" / "param.json";
-	if (Common::File::IsFileExisting(param_json)) {
-		Loader::SystemContentLoadParamSfo(param_json);
-		if (auto flexible_memory_size = Loader::SystemContentGetFlexibleMemorySize();
-		    flexible_memory_size != 0) {
-			Libs::LibKernel::Memory::SetFlexibleMemorySize(flexible_memory_size);
-		}
-	}
-
 	MountSandboxDirs();
 
 	auto* rt = Common::Singleton<Loader::RuntimeLinker>::Instance();

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -83,8 +83,7 @@ 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 +151,9 @@ 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(); }
@@ -168,6 +170,7 @@ 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;
+	// Persistent draw state: indirect draws update it for subsequent draws.
 	uint32_t m_num_instances = 1;
 
 	uint32_t m_de_count        = 0;

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -1926,8 +1926,14 @@ KYTY_CP_OP_PARSER(CpOpCopyData) {
 	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) {
+	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);
@@ -3390,6 +3396,12 @@ 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,
@@ -374,6 +375,8 @@ enum class BufferFormat : uint32_t {
 	k32_32_32_32UInt    = 75,
 	k32_32_32_32SInt    = 76,
 	k32_32_32_32Float   = 77,
+	k8Srgb              = 128,
+	k8_8Srgb            = 129,
 	k8_8_8_8Srgb        = 130,
 	k9_9_9_5Float       = 132,
 	k5_6_5UNorm         = 133,

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},
@@ -57,6 +58,8 @@ constexpr FormatInfo kFormatInfo[] = {
     {GpuEnumValue(BufferFormat::k32_32_32_32UInt), 16, 0, 16, true, true},
     {GpuEnumValue(BufferFormat::k32_32_32_32SInt), 16, 0, 16, false, false},
     {GpuEnumValue(BufferFormat::k32_32_32_32Float), 16, 0, 16, true, false},
+    {GpuEnumValue(BufferFormat::k8Srgb), 1, 0, 0, true, false},
+    {GpuEnumValue(BufferFormat::k8_8Srgb), 2, 0, 0, true, false},
     {GpuEnumValue(BufferFormat::k8_8_8_8Srgb), 4, 0, 4, true, false},
     {GpuEnumValue(BufferFormat::k9_9_9_5Float), 4, 0, 0, true, false},
     {GpuEnumValue(BufferFormat::k5_6_5UNorm), 2, 0, 2, true, false},

src/graphics/guest_gpu/graphicsRun.cpp

@@ -34,7 +34,6 @@ 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;
 
@@ -98,8 +97,6 @@ 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);
@@ -410,8 +407,13 @@ 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);
 
@@ -691,26 +693,6 @@ 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");
@@ -962,9 +944,8 @@ void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_cou
 	auto* index_addr = reinterpret_cast<const void*>(
 	    m_index_base_addr + static_cast<uint64_t>(index_offset) * index_size);
 
-	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(),
-	                                             m_index_type_and_size, index_count, index_addr,
-	                                             flags, 1, m_num_instances);
+	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(), m_index_type_and_size,
+	                                         index_count, index_addr, flags, 1, m_num_instances);
 }
 
 void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed) {
@@ -1002,8 +983,9 @@ 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;
 	}
 
@@ -1043,6 +1025,7 @@ 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);
 }
@@ -1100,8 +1083,9 @@ 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;
 		}
 
@@ -1142,6 +1126,7 @@ 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);
 	}
@@ -1190,8 +1175,8 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 			}
 		}
 
-		m_renderer.GetRenderExecutor().DispatchDirect(
-		    m_submit_id, CurrentBuffer(), thread_group_x, thread_group_y, thread_group_z, mode);
+		m_renderer.GetRenderExecutor().DispatchDirect(m_submit_id, CurrentBuffer(), thread_group_x,
+		                                              thread_group_y, thread_group_z, mode);
 	}
 
 	constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
@@ -1233,13 +1218,18 @@ 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 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, first_vertex, first_instance);
+	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), vertex_count, flags,
+	                                        render_target_slice_offset, m_num_instances,
+	                                        first_vertex, first_instance);
 }
 
 void CommandProcessor::WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index) {
@@ -1317,8 +1307,8 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 				if (eop_event_type == 0x2f && cache_action == 0x00 && event_index == 0x06) {
 					auto* dst = static_cast<uint32_t*>(dst_gpu_addr);
 					SynchronizeGpu();
-					Sync::ReadGds(m_renderer.GetBufferCache().GetGdsBuffer(), dst,
-					              value & 0xffffu, value >> 16u);
+					Sync::ReadGds(m_renderer.GetBufferCache().GetGdsBuffer(), dst, value & 0xffffu,
+					              value >> 16u);
 					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
 					                            value >> 16u);
 					return;
@@ -1486,8 +1476,7 @@ void CommandProcessor::EmitGlobalBarrier() {
 	barrier.srcStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
 	barrier.srcAccessMask = vk::AccessFlagBits2::eMemoryWrite;
 	barrier.dstStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
-	barrier.dstAccessMask =
-	    vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite;
 
 	vk::DependencyInfo dependency {};
 	dependency.memoryBarrierCount = 1;
@@ -1690,32 +1679,6 @@ 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;
 }
@@ -1724,12 +1687,4 @@ 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;
-}
-
 } // namespace Libs::Graphics

src/graphics/guest_gpu/graphicsRun.h

@@ -35,25 +35,8 @@ public:
 
 	[[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;
-	};
 
 private:
-	friend class SubmissionLock;
-
-	void PauseSubmissions();
-	void ResumeSubmissions();
-
 	std::unique_ptr<GpuState> m_state;
 };
 } // namespace Libs::Graphics

src/graphics/guest_gpu/pm4.cpp

@@ -110,7 +110,6 @@ 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 +119,9 @@ 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/host_gpu/memoryTracker.cpp

@@ -4,25 +4,9 @@
 
 namespace Libs::Graphics {
 
-#if defined(KYTY_MEMORY_TRACKER_TESTS)
-namespace {
-std::atomic<MemoryTracker::UnmapContentionHook> g_unmap_contention_hook {nullptr};
-}
-
-void MemoryTracker::SetUnmapContentionHook(UnmapContentionHook hook) noexcept {
-	g_unmap_contention_hook.store(hook, std::memory_order_release);
-}
-#endif
-
 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);
@@ -31,6 +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 ||
@@ -68,6 +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 ||
@@ -94,7 +80,6 @@ RegionManager* MemoryTracker::GetOrCreateRegion(uint64_t index) {
 bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
 	std::lock_guard access(m_access_mutex);
-	RequireMapped(vaddr, size);
 	return Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Cpu>(offset, bytes);
@@ -104,7 +89,6 @@ bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
 	std::lock_guard access(m_access_mutex);
-	RequireMapped(vaddr, size);
 	return Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Gpu>(offset, bytes);
@@ -114,45 +98,31 @@ bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
 	std::lock_guard access(m_access_mutex);
-	RequireMapped(vaddr, size);
 	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);
 	});
 }
 
 void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
 	std::lock_guard access(m_access_mutex);
-	RequireMapped(vaddr, size);
-	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);
 	});
 }
 
 void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
 	std::lock_guard access(m_access_mutex);
-	RequireMapped(vaddr, size);
-	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);
 	});
 }
 
 void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
-	RequireMapped(vaddr, size);
-
 	std::vector<RegionManager*> managers;
 	managers.reserve((vaddr % TRACKER_REGION_SIZE + size + TRACKER_REGION_SIZE - 1) /
 	                 TRACKER_REGION_SIZE);
@@ -171,10 +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);
 	});
 	locks.clear();
 }
@@ -185,109 +152,4 @@ void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
 	UntrackMemoryLocked(vaddr, size);
 }
 
-void MemoryTracker::UnmapMemory(uint64_t vaddr, uint64_t size) {
-	CheckNotInUploadCallback();
-	std::unique_lock access(m_access_mutex, std::try_to_lock);
-	if (!access.owns_lock()) {
-#if defined(KYTY_MEMORY_TRACKER_TESTS)
-		if (const auto hook = g_unmap_contention_hook.load(std::memory_order_acquire);
-		    hook != nullptr) {
-			hook();
-		}
-#endif
-		access.lock();
-	}
-	UntrackMemoryLocked(vaddr, size);
-	m_page_manager.OnGpuUnmap(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);
@@ -30,14 +29,6 @@ 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);
-	void               UnmapMemory(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;
 	template <typename Flush>
 	void InvalidateRegion(uint64_t vaddr, uint64_t size, Flush&& on_flush) {
 		static_assert(std::is_invocable_v<Flush&>);
@@ -64,9 +55,8 @@ public:
 			}
 			Iterate<false>(vaddr, size,
 			               [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-				               const auto changed = manager->ChangeState<DirtySource::Cpu, true>(
+				               manager->ChangeState<DirtySource::Cpu, true>(
 				                   manager->GetCpuAddr() + offset, bytes);
-				               manager->ApplyProtection(changed, false);
 			               });
 			return false;
 		};
@@ -79,12 +69,15 @@ public:
 			EXIT("memory invalidation retained GPU-owned pages\n");
 		}
 	}
-	[[nodiscard]] bool InvalidateVirtualGpuWrite(PageFaultAccess access, uint64_t vaddr,
-	                                             uint64_t size, PageFaultPhase phase) 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) {
@@ -92,7 +85,6 @@ public:
 		static_assert(std::is_nothrow_invocable_v<Func&, uint64_t, uint64_t>);
 		CheckNotInUploadCallback();
 		std::lock_guard             access(m_access_mutex);
-		RequireMapped(vaddr, size);
 		std::vector<RegionManager*> managers;
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t, uint64_t) {
 			managers.push_back(manager);
@@ -104,9 +96,6 @@ 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);
 		});
@@ -118,10 +107,8 @@ 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);
 			               });
 		}
 	}
@@ -132,11 +119,6 @@ public:
 		    vaddr, size, [](uint64_t, uint64_t) noexcept {}, std::forward<Func>(func));
 	}
 
-#if defined(KYTY_MEMORY_TRACKER_TESTS)
-	using UnmapContentionHook = void (*)() noexcept;
-	static void SetUnmapContentionHook(UnmapContentionHook hook) noexcept;
-#endif
-
 	template <typename RangeFunc, typename UploadFunc>
 	void ForEachUploadRange(uint64_t vaddr, uint64_t size, bool is_written, RangeFunc&& range_func,
 	                        UploadFunc&& upload_func) {
@@ -144,12 +126,10 @@ public:
 		static_assert(std::is_nothrow_invocable_v<UploadFunc&>);
 		CheckNotInUploadCallback();
 		std::unique_lock access(m_access_mutex);
-		RequireMapped(vaddr, size);
 		Iterate<true>(vaddr, size, [](RegionManager*, uint64_t, uint64_t) {});
 		const auto* previous_upload_owner = std::exchange(s_upload_owner, this);
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			manager->lock.lock();
-			manager->Track(manager->GetCpuAddr() + offset, bytes);
 			manager->ForEachModifiedRange<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset,
 			                                                      bytes, range_func);
 			if (!is_written) {
@@ -158,11 +138,10 @@ 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>(
+			Iterate<false>(vaddr, size,
+			               [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+				               manager->template ChangeState<DirtySource::Gpu, true>(
 				                   manager->GetCpuAddr() + offset, bytes);
-				    manager->ApplyGpuProtection(changed, true, m_gpu_watch_mode);
 				               manager->lock.unlock();
 			               });
 		}
@@ -211,14 +190,6 @@ private:
 
 	static void    ValidateRange(uint64_t vaddr, uint64_t size);
 	void           UntrackMemoryLocked(uint64_t vaddr, uint64_t size);
-	void        RequireMapped(uint64_t vaddr, uint64_t size) const {
-		ValidateRange(vaddr, size);
-		if (!m_page_manager.IsMapped(vaddr, size)) {
-			EXIT("memory tracker range [0x%llx, 0x%llx) is not mapped\n",
-			     static_cast<unsigned long long>(vaddr),
-			     static_cast<unsigned long long>(vaddr + size));
-		}
-	}
 	RegionManager* GetOrCreateRegion(uint64_t index);
 
 	std::unique_ptr<std::atomic<RegionManager*>[]> m_regions;
@@ -226,7 +197,6 @@ 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.h

@@ -2,60 +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 };
-enum class GpuAccess { Read, 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;
-	[[nodiscard]] bool     IsMapped(uint64_t vaddr, uint64_t size) const noexcept;
-	[[nodiscard]] bool HasGpuAccess(uint64_t vaddr, uint64_t size, GpuAccess access) const noexcept;
 
-	void UpdatePageWatchers(bool track, uint64_t vaddr, uint64_t size,
-	                        PageWatchMode mode = PageWatchMode::Write);
-	void OnGpuMap(uint64_t vaddr, uint64_t size, GpuAccess access = GpuAccess::ReadWrite);
-	void OnGpuUnmap(uint64_t vaddr, uint64_t size, GpuAccess access = GpuAccess::ReadWrite);
-
-	[[nodiscard]] bool HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept;
-	[[nodiscard]] std::vector<std::unique_ptr<BackingWrite>>
-	ReserveBackingWrites(std::span<const RangeSet::Range> ranges);
+	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);
 
 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);
+		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

@@ -123,30 +123,6 @@ 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) {
@@ -219,32 +195,24 @@ void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, Ret
 		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 {
+	m_scheduler.DeferOperation(
+	    [this, downloads = std::move(downloads), retire = std::move(retire)]() 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);
 			    }
+			    // 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);
 		    }
-		CompleteBackingPublication(retire.address, retire.size, tick);
 	    });
 }
 
@@ -253,13 +221,12 @@ 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 +244,9 @@ 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 +256,6 @@ 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;
@@ -385,7 +284,6 @@ 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;
 	}
@@ -394,7 +292,6 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 }
 
 void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
-	(void)SynchronizeBacking(vaddr, size);
 	std::vector<DownloadCopy> copies;
 	{
 		FaultSafeCacheLock lock(this, m_mutex);
@@ -434,116 +331,20 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
 	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);
 		}
+		// The enumeration above covered whole dirty pages and every exact interval on them.
+		m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, 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");
-		}
-	}
-	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;
 	{
 		FaultSafeCacheLock lock(this, m_mutex);
@@ -561,12 +362,8 @@ 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 +393,11 @@ 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 +405,6 @@ 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) {
@@ -714,22 +497,30 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 	if (command.IsInvalid() || command.IsExecute()) {
 		EXIT("BufferCache: buffer request requires a recording command buffer\n");
 	}
-	ValidateGpuAccess(vaddr, size, is_read, is_written);
 	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);
 	}
 
@@ -745,10 +536,12 @@ 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)};
 }
 
@@ -773,7 +566,6 @@ 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()) {
@@ -788,13 +580,12 @@ 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());
@@ -844,42 +635,32 @@ 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();
+	const bool         has_dirty_buffer_source = !dirty.empty();
 	auto               owner                   = find_owner();
-	if (requested_gpu_owned && owner == m_buffers.end()) {
+	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;
@@ -893,42 +674,17 @@ 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);
@@ -944,45 +700,6 @@ 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) {
@@ -999,7 +716,6 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 	if (vaddr == 0) {
 		EXIT("BufferCache: invalid fill memory address\n");
 	}
-	ValidateGpuAccess(vaddr, size, false, true);
 	(void)m_texture_cache.ClearMeta(vaddr);
 	{
 		std::lock_guard transaction(m_resource_mutex);
@@ -1041,25 +757,12 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	    (src_gds && (src_vaddr > m_gds_buffer.Size() || size > m_gds_buffer.Size() - src_vaddr))) {
 		EXIT("BufferCache: invalid or overlapping copy range\n");
 	}
-	if (src_memory) {
-		ValidateGpuAccess(src_vaddr, size, true, false);
-	}
-	if (dst_memory) {
-		ValidateGpuAccess(dst_vaddr, size, false, true);
-	}
 	if (src_memory || dst_memory) {
 		std::lock_guard transaction(m_resource_mutex);
-		if (src_memory) {
-			(void)SynchronizeBacking(src_vaddr, size);
-		}
 		const auto      src_region =
 		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
 		const auto dst_region =
 		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
-		if (src_memory && src_region.gpu_image_bytes &&
-		    !m_texture_cache.SynchronizeImageToBuffer(src_vaddr, size)) {
-			EXIT("BufferCache: GPU copy source image could not be synchronized\n");
-		}
 		if (src_memory && dst_memory && !HasGpuDirtyBytes(src_vaddr, size) &&
 		    !HasGpuDirtyBytes(dst_vaddr, size) && !src_region.gpu_image_bytes &&
 		    !dst_region.gpu_image_bytes) {
@@ -1081,7 +784,7 @@ 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};
@@ -1134,95 +837,13 @@ 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::ValidateGpuAccess(uint64_t vaddr, uint64_t size, bool is_read,
-                                    bool is_written) const {
-	if ((!is_read && !is_written) || vaddr == 0 || size == 0 || size > UINT64_MAX - vaddr) {
-		EXIT("BufferCache: invalid GPU access request\n");
-	}
-	if (is_read && !m_page_manager.HasGpuAccess(vaddr, size, GpuAccess::Read)) {
-		EXIT("BufferCache: GPU-read access denied\n");
-	}
-	if (is_written && !m_page_manager.HasGpuAccess(vaddr, size, GpuAccess::Write)) {
-		EXIT("BufferCache: GPU-write access denied\n");
-	}
-}
-
 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;
 	}
 
@@ -1291,7 +912,6 @@ 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>
@@ -32,7 +31,6 @@ struct BufferBinding {
 struct ImageBufferSource {
 	Buffer*  buffer = nullptr;
 	uint64_t offset = 0;
-	bool     gpu_owned = false;
 };
 
 class BufferCache {
@@ -47,8 +45,6 @@ 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);
@@ -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);
@@ -72,12 +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 ValidateGpuAccess(uint64_t vaddr, uint64_t size, bool is_read, bool is_written) const;
 	void               RunGarbageCollector();
 
 private:
@@ -91,8 +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);
@@ -107,12 +92,10 @@ 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);
 
 	GraphicContext&                                   m_graphics;
@@ -121,17 +104,12 @@ 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

@@ -4,37 +4,15 @@
 #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)) {
@@ -115,33 +93,41 @@ bool GpuResourceManager::IsMapped(uint64_t vaddr, uint64_t size) const noexcept
 	return m_mapped_ranges.Contains(vaddr, size);
 }
 
-void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size, GpuAccess access) {
+void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size) {
 	{
 		std::lock_guard lock(m_mapped_ranges_mutex);
 		m_mapped_ranges.Add(vaddr, size);
 	}
-	m_page_manager.OnGpuMap(vaddr, size, access);
+	m_page_manager.OnGpuMap(vaddr, size);
 }
 
-void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size, GpuAccess access) {
-	if (!IsMapped(vaddr, size)) {
-		EXIT("cannot unmap an unmapped GPU resource range\n");
+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);
 		}
-	const auto unmap = [this, vaddr, size, access] {
-		m_texture_cache.UnmapMemory(vaddr, size);
 		m_buffer_cache.UnmapMemory(vaddr, size);
-		m_page_manager.OnGpuUnmap(vaddr, size, access);
+		m_texture_cache.UnmapMemory(vaddr, size);
+		m_page_manager.OnGpuUnmap(vaddr, size);
 		std::lock_guard lock(m_mapped_ranges_mutex);
 		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

@@ -29,18 +29,14 @@ public:
 	[[nodiscard]] bool HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept;
 	[[nodiscard]] bool InvalidateMemory(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool IsMapped(uint64_t vaddr, uint64_t size) const noexcept;
-	void               MapMemory(uint64_t vaddr, uint64_t size, GpuAccess access);
-	void               UnmapMemory(uint64_t vaddr, uint64_t size, GpuAccess access);
+	void               MapMemory(uint64_t vaddr, uint64_t size);
+	void               UnmapMemory(uint64_t vaddr, uint64_t size);
 	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

@@ -219,7 +219,7 @@ private:
 		typename CoarseTable::PageRange   coarse_range {};
 		typename TrackingTable::PageRange tracking_range {};
 		if (!CoarseTable::TryGetPageRange(address, size, coarse_range) ||
-		    !TrackingTable::TryGetPageRange(address, size, tracking_range)) {
+		    (!strict_bytes && !TrackingTable::TryGetPageRange(address, size, tracking_range))) {
 			return {};
 		}
 		MembershipList candidates;
@@ -230,8 +230,8 @@ private:
 		}
 		std::vector<OwnerT> result;
 		for (const Registration* registration: candidates) {
-			if ((!strict_bytes || Overlaps(registration->ranges, address, size)) &&
-			    HasTrackingMembership(registration, tracking_range) &&
+			if ((strict_bytes ? Overlaps(registration->ranges, address, size)
+			                  : HasTrackingMembership(registration, tracking_range)) &&
 			    predicate(registration->owner)) {
 				result.push_back(registration->owner);
 			}

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/streamBuffer.cpp

@@ -135,8 +135,8 @@ void Buffer::Write(uint64_t offset, const void* source, uint64_t size) {
 void Buffer::Flush(uint64_t offset, uint64_t size) {
 	EXIT_IF(m_mapped.empty() || offset > m_size || size > m_size - offset);
 	if (!m_is_coherent && size != 0) {
-		const auto result = vmaFlushAllocation(m_graphics->allocator, m_buffer->memory.allocation,
-		                                      offset, size);
+		const auto result =
+		    vmaFlushAllocation(m_graphics->allocator, m_buffer->memory.allocation, offset, size);
 		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}
 }
@@ -144,8 +144,8 @@ void Buffer::Flush(uint64_t offset, uint64_t size) {
 vk::BufferMemoryBarrier Buffer::Barrier(uint64_t offset, uint64_t size, vk::AccessFlags source,
                                         vk::AccessFlags destination) const {
 	if (Handle() == nullptr || size == 0 || offset > m_size || size > m_size - offset) {
-		EXIT("Buffer: invalid DMA barrier, handle=%p offset=0x%016" PRIx64
-		     " size=0x%016" PRIx64 " capacity=0x%016" PRIx64 "\n",
+		EXIT("Buffer: invalid DMA barrier, handle=%p offset=0x%016" PRIx64 " size=0x%016" PRIx64
+		     " capacity=0x%016" PRIx64 "\n",
 		     static_cast<const void*>(Handle()), offset, size, m_size);
 	}
 	vk::BufferMemoryBarrier barrier {};
@@ -175,8 +175,7 @@ void Buffer::CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t sou
 	command.EndRendering();
 	const vk::BufferMemoryBarrier before[] = {
 	    source.Barrier(source_offset, size, source_before, vk::AccessFlagBits::eTransferRead),
-	    Barrier(destination_offset, size, destination_before,
-	            vk::AccessFlagBits::eTransferWrite),
+	    Barrier(destination_offset, size, destination_before, vk::AccessFlagBits::eTransferWrite),
 	};
 	const auto host_access  = vk::AccessFlagBits::eHostRead | vk::AccessFlagBits::eHostWrite;
 	auto       before_stage = vk::PipelineStageFlags {vk::PipelineStageFlagBits::eAllCommands};
@@ -214,8 +213,7 @@ void Buffer::Fill(uint64_t offset, uint64_t size, uint32_t value) {
 	                       vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion,
 	                       0, nullptr, 1, &before, 0, nullptr);
 	native.fillBuffer(Handle(), offset, size, value);
-	const auto after =
-	    Barrier(offset, size, vk::AccessFlagBits::eTransferWrite,
+	const auto after = Barrier(offset, size, vk::AccessFlagBits::eTransferWrite,
 	                           vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite);
 	native.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
 	                       vk::PipelineStageFlagBits::eAllCommands,

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

@@ -54,16 +54,15 @@ public:
 	[[nodiscard]] bool IsInBounds(uint64_t address, uint64_t size) const noexcept;
 	void               Write(uint64_t offset, const void* source, uint64_t size);
 	void               Flush(uint64_t offset, uint64_t size);
-	void               CopyFrom(
-	                  CommandBuffer& command, const Buffer& source, uint64_t source_offset,
+	void CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t source_offset,
 	              uint64_t destination_offset, uint64_t size,
 	              vk::AccessFlags source_before      = vk::AccessFlagBits::eMemoryWrite,
-	                  vk::AccessFlags destination_before =
-	                      vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
-	                  vk::AccessFlags source_after =
-	                      vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
-	                  vk::AccessFlags destination_after =
-	                      vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite);
+	              vk::AccessFlags destination_before = vk::AccessFlagBits::eMemoryRead |
+	                                                   vk::AccessFlagBits::eMemoryWrite,
+	              vk::AccessFlags source_after       = vk::AccessFlagBits::eMemoryRead |
+	                                                   vk::AccessFlagBits::eMemoryWrite,
+	              vk::AccessFlags destination_after  = vk::AccessFlagBits::eMemoryRead |
+	                                                   vk::AccessFlagBits::eMemoryWrite);
 	void Fill(uint64_t offset, uint64_t size, uint32_t value);
 
 protected:

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

@@ -88,15 +88,34 @@ TextureCache::~TextureCache() {
 
 bool TextureCache::SameBacking(const ImageInfo& cached, const ImageInfo& requested,
                                bool exact_format) {
-	const bool unit_extent =
-	    requested.extent.width == 1 && requested.extent.height == 1 && requested.extent.depth == 1;
-	return cached.data == requested.data && cached.extent == requested.extent &&
-	       cached.samples == requested.samples &&
-	       cached.bytes_per_block == requested.bytes_per_block &&
-	       (cached.type == requested.type || unit_extent) &&
-	       (exact_format
-	            ? cached.pixel_format == requested.pixel_format
-	            : ImageViewOps::FormatsCompatible(cached.pixel_format, requested.pixel_format));
+	if (cached.data.address != requested.data.address) {
+		return false;
+	}
+	if (cached.data.size != requested.data.size) {
+		return false;
+	}
+	if (cached.extent != requested.extent) {
+		return false;
+	}
+	if (cached.samples != requested.samples) {
+		return false;
+	}
+	if (cached.bytes_per_block != requested.bytes_per_block) {
+		return false;
+	}
+	if (cached.tile_mode != requested.tile_mode) {
+		return false;
+	}
+	if (!ImageViewOps::FormatsCompatible(cached.pixel_format, requested.pixel_format)) {
+		return false;
+	}
+	if (cached.type != requested.type && requested.extent != vk::Extent3D {1, 1, 1}) {
+		return false;
+	}
+	if (exact_format && cached.pixel_format != requested.pixel_format) {
+		return false;
+	}
+	return true;
 }
 
 TextureCache::BindingType TextureCache::UploadBinding(const Image& image) {
@@ -304,7 +323,7 @@ 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) {
@@ -329,7 +348,7 @@ 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) {
@@ -347,7 +366,7 @@ 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) {
@@ -360,7 +379,7 @@ 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);
 	}
 }
 
@@ -381,7 +400,7 @@ void TextureCache::UntrackImageHead(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, begin, size);
+		m_page_manager.UpdatePageWatchers<false>(begin, size);
 	}
 }
 
@@ -402,7 +421,7 @@ void TextureCache::UntrackImageTail(ImageId id) {
 		UntrackImage(id);
 	}
 	if (size != 0) {
-		m_page_manager.UpdatePageWatchers(false, address, size);
+		m_page_manager.UpdatePageWatchers<false>(address, size);
 	}
 }
 
@@ -735,6 +754,12 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		    (requested.IsVolume() || cached.info.IsVolume())) {
 			return {ExpandImage(requested, cached_id)};
 		}
+		if (requested.tile_mode != cached.info.tile_mode) {
+			if (safe_to_delete) {
+				DeleteImages(std::array {cached_id}, cached_id);
+			}
+			return {merged_id};
+		}
 		if (requested.pixel_format != cached.info.pixel_format ||
 		    requested.data.size <= cached.info.data.size) {
 			const auto result_id = merged_id ? merged_id : cached_id;
@@ -747,12 +772,6 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		if (requested.type == cached.info.type && requested.resources > cached.info.resources) {
 			return {ExpandImage(requested, cached_id)};
 		}
-		if (requested.tile_mode != cached.info.tile_mode) {
-			if (safe_to_delete) {
-				DeleteImages(std::array {cached_id}, cached_id);
-			}
-			return {merged_id};
-		}
 		EXIT("TextureCache: unresolvable equal-address image overlap, address=0x%016" PRIx64
 		     " requested=%ux%u "
 		     "cached=%ux%u requested_size=0x%016" PRIx64 " cached_size=0x%016" PRIx64
@@ -819,11 +838,20 @@ 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;
@@ -848,22 +876,28 @@ 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;
 		}
@@ -894,6 +928,7 @@ 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;
@@ -931,11 +966,19 @@ 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);
@@ -1019,7 +1062,6 @@ 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();
 	if (upload) {
@@ -1028,16 +1070,12 @@ void TextureCache::InitializeImage(ImageId id, const ImageDesc& desc) {
 		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();
 	}
@@ -1112,7 +1150,6 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	}
 
 	ImageId result {};
-	bool    replacement_buffer = false;
 	bool    inserted_new = false;
 	{
 		std::lock_guard            transaction(m_resource_mutex);
@@ -1122,7 +1159,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 
 		for (const auto id: candidates) {
 			const auto owner = ResolveOwner(id);
-			if (owner == nullptr || owner->info.data != desc.info.data) {
+			if (owner == nullptr) {
 				continue;
 			}
 			if (SameBacking(owner->info, desc.info, exact_format)) {
@@ -1140,7 +1177,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 				if (owner == nullptr) {
 					continue;
 				}
-				const auto merged_info = result ? ResolveImage(result).info : desc.info;
+				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;
@@ -1155,22 +1192,14 @@ 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,
+			if (m_buffer_cache.HasGpuDirtyBytes(inserted.info.data.address,
 			                                    inserted.info.data.size)) {
 				inserted.MarkBufferModified();
 			}
@@ -1341,11 +1370,6 @@ 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();
@@ -1358,7 +1382,6 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 	if (command.IsInvalid() || !GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid image clear\n");
 	}
-	m_buffer_cache.ValidateGpuAccess(address, size, false, true);
 	std::lock_guard      transaction(m_resource_mutex);
 	CacheLock            lock(*this, m_lock);
 	ImageId              selected {};
@@ -1411,9 +1434,6 @@ 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);
@@ -1533,11 +1553,14 @@ 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;
 	}
@@ -1547,22 +1570,119 @@ 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) {
@@ -1594,12 +1714,9 @@ 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);
 	});
 }
 
@@ -1620,7 +1737,7 @@ 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;
@@ -1634,62 +1751,6 @@ 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()) {
@@ -1808,37 +1869,6 @@ 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);
@@ -76,8 +75,6 @@ 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();
@@ -142,7 +139,7 @@ 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 +154,6 @@ 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);
@@ -186,6 +182,7 @@ private:
 	bool     m_readback_linear_images = false;
 
 	friend struct TextureCacheTestAccess;
+	friend class BufferCache;
 	friend class RenderExecutor;
 };
 

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -7,8 +7,8 @@
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/debug.h"
+#include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
@@ -79,10 +79,8 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	const auto view = ResolveTargetViewInfo(
 	    rt.view.base_array_slice_index, rt.view.last_array_slice_index, render_target_slice_offset);
 	switch (view.type) {
-		case TargetViewType::Image2D: break;
-		case TargetViewType::Image2DArray:
-			EXIT("layered render-target views are unsupported: base=%u count=%u\n", view.base_layer,
-			     view.layer_count);
+		case TargetViewType::Image2D:
+		case TargetViewType::Image2DArray: break;
 		case TargetViewType::Unsupported:
 			EXIT("invalid render-target view: base=%u last=%u draw_offset=%u\n",
 			     rt.view.base_array_slice_index, rt.view.last_array_slice_index,
@@ -121,6 +119,17 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	uint32_t   pitch  = 0;
 	uint64_t   size   = 0;
 	bool       tile   = false;
+	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);
 
@@ -147,6 +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 ||
@@ -167,10 +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);
 		}
@@ -180,7 +194,17 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 
 	TileSizeOffset   mip_sizes[16] {};
 	TilePaddedSize   mip_padded[16] {};
-	if (tile) {
+	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) {
@@ -205,12 +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) {
@@ -219,23 +237,40 @@ 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");
 	}
@@ -244,15 +279,24 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	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.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/colorRenderTarget.h

@@ -2,8 +2,8 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_COLORRENDERTARGET_H_
 
 #include "graphics/guest_gpu/gpu_defs.h"
-#include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
+#include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <cstdint>

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -331,8 +331,7 @@ void CommandScheduler::WaitPriorityOperations(uint64_t tick) {
 	EXIT_IF(g_deferred_callback_scheduler == this);
 	std::unique_lock lock(m_operation_mutex);
 	m_operation_available.wait(lock, [this, tick] {
-		const bool active_before_or_at =
-		    m_priority_active && m_priority_active_tick <= tick;
+		const bool active_before_or_at = m_priority_active && m_priority_active_tick <= tick;
 		const bool queued_before_or_at =
 		    !m_priority_operations.empty() && m_priority_operations.front().tick <= tick;
 		return !active_before_or_at && !queued_before_or_at;

src/graphics/host_gpu/renderer/context.cpp

@@ -8,8 +8,8 @@
 #include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/depthRenderTarget.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
@@ -270,8 +270,8 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 		colors[i].sType        = vk::StructureType::eRenderingAttachmentInfo;
 		colors[i].imageView    = attachment.image_view;
 		colors[i].imageLayout  = attachment.image_layout;
-		colors[i].loadOp = attachment.is_clear ? vk::AttachmentLoadOp::eClear
-		                                      : vk::AttachmentLoadOp::eLoad;
+		colors[i].loadOp =
+		    attachment.is_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad;
 		colors[i].storeOp                 = vk::AttachmentStoreOp::eStore;
 		colors[i].clearValue.color.uint32 = attachment.clear_value;
 	}
@@ -281,8 +281,8 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 	depth.sType       = vk::StructureType::eRenderingAttachmentInfo;
 	depth.imageView   = depth_stencil.image_view;
 	depth.imageLayout = depth_stencil.image_layout;
-	depth.loadOp = depth_stencil.depth_clear ? vk::AttachmentLoadOp::eClear
-	                                        : vk::AttachmentLoadOp::eLoad;
+	depth.loadOp =
+	    depth_stencil.depth_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad;
 	depth.storeOp                       = vk::AttachmentStoreOp::eStore;
 	depth.clearValue.depthStencil.depth = std::bit_cast<float>(depth_stencil.clear_value[0]);
 
@@ -290,8 +290,8 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 	stencil.sType       = vk::StructureType::eRenderingAttachmentInfo;
 	stencil.imageView   = depth_stencil.image_view;
 	stencil.imageLayout = depth_stencil.image_layout;
-	stencil.loadOp = depth_stencil.stencil_clear ? vk::AttachmentLoadOp::eClear
-	                                            : vk::AttachmentLoadOp::eLoad;
+	stencil.loadOp =
+	    depth_stencil.stencil_clear ? vk::AttachmentLoadOp::eClear : vk::AttachmentLoadOp::eLoad;
 	stencil.storeOp                         = vk::AttachmentStoreOp::eStore;
 	stencil.clearValue.depthStencil.stencil = depth_stencil.clear_value[1];
 

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) {
@@ -548,14 +538,6 @@ static void ZCheck(const HW::DepthRenderTarget& z) {
 		EXIT_NOT_IMPLEMENTED(z.htile_surface.prefetch_height != 0x00000000);
 		EXIT_NOT_IMPLEMENTED(z.htile_surface.dst_outside_zero_to_one != 0x00000000);
 
-		if (z.depth_view.slice_start != 0x00000000 || z.depth_view.slice_max != 0x00000000) {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
-				LOGF("DepthTarget: temporary: ignoring PS5 array slice view start=0x%08" PRIx32
-				     ", max=0x%08" PRIx32 "\n",
-				     z.depth_view.slice_start, z.depth_view.slice_max);
-			}
-		}
 		if (z.depth_view.current_mip_level != 0x00000000) {
 			static std::atomic<uint32_t> log_count {0};
 			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
@@ -1214,7 +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

@@ -10,10 +10,10 @@
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/debug.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
+#include "graphics/host_gpu/renderer/image/textureCommon.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
@@ -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  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) {
@@ -150,16 +150,14 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	    has_stencil, has_htile, z.stencil_info.htile_stencil_disabled);
 	const auto view = ResolveTargetViewInfo(z.depth_view.slice_start, z.depth_view.slice_max);
 	switch (view.type) {
-		case TargetViewType::Image2D: break;
-		case TargetViewType::Image2DArray:
-			DepthFatal("layered depth views are unsupported: base=%u count=%u", view.base_layer,
-			           view.layer_count);
+		case TargetViewType::Image2D:
+		case TargetViewType::Image2DArray: break;
 		case TargetViewType::Unsupported:
 			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 ||
@@ -281,10 +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/depthRenderTarget.h

@@ -2,9 +2,9 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DEPTHRENDERTARGET_H_
 
 #include "common/assert.h"
+#include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
-#include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <cstdint>

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

@@ -182,8 +182,8 @@ void BlitHelper::ReinterpretColorAsMsDepth(Image& source, Image& destination) {
 	auto  command        = command_buffer.Handle();
 	source.Transit(vk::ImageLayout::eShaderReadOnlyOptimal, vk::AccessFlagBits2::eShaderRead, {},
 	               command);
-	destination.Transit(ColorToMsDepthLayout,
-	                    vk::AccessFlagBits2::eDepthStencilAttachmentWrite, {}, command);
+	destination.Transit(ColorToMsDepthLayout, vk::AccessFlagBits2::eDepthStencilAttachmentWrite, {},
+	                    command);
 
 	vk::RenderingAttachmentInfo depth_attachment {};
 	depth_attachment.sType                   = vk::StructureType::eRenderingAttachmentInfo;

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/imageInfo.h

@@ -21,8 +21,7 @@ struct GuestRange {
 
 	[[nodiscard]] constexpr bool Empty() const noexcept { return address == 0 || size == 0; }
 	[[nodiscard]] constexpr bool Valid() const noexcept {
-		return !Empty() && address < TRACKER_ADDRESS_SIZE &&
-		       size <= TRACKER_ADDRESS_SIZE - address;
+		return !Empty() && address < TRACKER_ADDRESS_SIZE && size <= TRACKER_ADDRESS_SIZE - address;
 	}
 	[[nodiscard]] constexpr uint64_t End() const noexcept { return address + size; }
 	auto                             operator<=>(const GuestRange&) const = default;
@@ -352,8 +351,7 @@ inline bool ImageInfo::IsDepth() const noexcept {
 	}
 	const auto transfer_bytes = DepthAspectTransferBytes(info.pixel_format);
 	return transfer_bytes == info.bytes_per_block ||
-	       (info.bytes_per_block == sizeof(uint16_t) &&
-	        transfer_bytes == sizeof(uint32_t));
+	       (info.bytes_per_block == sizeof(uint16_t) && transfer_bytes == sizeof(uint32_t));
 }
 
 [[nodiscard]] inline VideoOutCompression
@@ -472,16 +470,11 @@ IsSupportedDisplayRenderTargetTileMode(uint32_t tile_mode) noexcept {
 	const auto unorm8 = [](uint32_t value) { return static_cast<float>(value & 0xffu) / 255.0f; };
 	const auto srgb8  = [](uint32_t value) {
 		const auto encoded = static_cast<float>(value & 0xffu) / 255.0f;
-		return encoded <= 0.04045f ? encoded / 12.92f
-		                           : std::pow((encoded + 0.055f) / 1.055f, 2.4f);
+		return encoded <= 0.04045f ? encoded / 12.92f : std::pow((encoded + 0.055f) / 1.055f, 2.4f);
 	};
 	switch (format) {
-		case vk::Format::eR32Uint:
-			next.uint32[0] = packed;
-			break;
-		case vk::Format::eR32Sint:
-			next.int32[0] = static_cast<int32_t>(packed);
-			break;
+		case vk::Format::eR32Uint: next.uint32[0] = packed; break;
+		case vk::Format::eR32Sint: next.int32[0] = static_cast<int32_t>(packed); break;
 		case vk::Format::eR8G8B8A8Srgb:
 			next.float32[0] = srgb8(packed);
 			next.float32[1] = srgb8(packed >> 8u);

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

@@ -70,15 +70,14 @@ namespace {
 			}
 		case vk::ImageType::e3D:
 			switch (info.type) {
-				case vk::ImageViewType::e3D:
-					return info.base_layer == 0 && info.layer_count == 1;
+				case vk::ImageViewType::e3D: return info.base_layer == 0 && info.layer_count == 1;
 				case vk::ImageViewType::e2D:
-					return static_cast<bool>(
-					           image.flags & vk::ImageCreateFlagBits::e2DArrayCompatible) &&
+					return static_cast<bool>(image.flags &
+					                         vk::ImageCreateFlagBits::e2DArrayCompatible) &&
 					       info.level_count == 1 && info.layer_count == 1;
 				case vk::ImageViewType::e2DArray:
-					return static_cast<bool>(
-					           image.flags & vk::ImageCreateFlagBits::e2DArrayCompatible) &&
+					return static_cast<bool>(image.flags &
+					                         vk::ImageCreateFlagBits::e2DArrayCompatible) &&
 					       info.level_count == 1;
 				default: return false;
 			}
@@ -327,8 +326,7 @@ bool FormatsCompatible(vk::Format base, vk::Format view) noexcept {
 vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	const auto& image      = backing;
 	auto        normalized = view_info;
-	const bool  is_storage =
-	    static_cast<bool>(normalized.usage & vk::ImageUsageFlagBits::eStorage);
+	const bool  is_storage = static_cast<bool>(normalized.usage & vk::ImageUsageFlagBits::eStorage);
 	normalized.aspect      = FullAspectMask(image.format);
 	if (normalized.aspect & vk::ImageAspectFlagBits::eDepth &&
 	    IsDepthViewFormat(normalized.format)) {
@@ -340,12 +338,11 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		normalized.format = image.format;
 		normalized.aspect = vk::ImageAspectFlagBits::eStencil;
 	}
-	normalized.usage =
-	    is_storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlags {};
+	normalized.usage = is_storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlags {};
 	const bool format_compatible = normalized.format != vk::Format::eUndefined &&
 	                               IsCompatibleViewFormat(image.format, normalized.format);
-	const bool slice_view = image.image_type == vk::ImageType::e3D &&
-	                        (normalized.type == vk::ImageViewType::e2D ||
+	const bool slice_view =
+	    image.image_type == vk::ImageType::e3D && (normalized.type == vk::ImageViewType::e2D ||
 	                                               normalized.type == vk::ImageViewType::e2DArray);
 	const bool levels_valid = normalized.level_count != 0 &&
 	                          normalized.base_level < image.mip_levels &&
@@ -353,15 +350,14 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	const auto view_layers  = slice_view && levels_valid
 	                              ? std::max(image.extent.depth >> normalized.base_level, 1u)
 	                              : image.layers;
-	const bool ranges_valid = levels_valid &&
-	                          normalized.layer_count != 0 && normalized.base_layer < view_layers &&
+	const bool ranges_valid = levels_valid && normalized.layer_count != 0 &&
+	                          normalized.base_layer < view_layers &&
 	                          normalized.layer_count <= view_layers - normalized.base_layer;
 	const bool mapping_valid =
 	    IsComponentSwizzle(normalized.mapping.r) && IsComponentSwizzle(normalized.mapping.g) &&
 	    IsComponentSwizzle(normalized.mapping.b) && IsComponentSwizzle(normalized.mapping.a);
 	if (image.image == nullptr || !format_compatible || !ranges_valid || !mapping_valid ||
-	    !IsValidViewType(image, normalized) ||
-	    !IsValidAspect(image, normalized.aspect)) {
+	    !IsValidViewType(image, normalized) || !IsValidAspect(image, normalized.aspect)) {
 		EXIT("invalid image view: image_format=%d view_format=%d type=%d aspect=0x%x "
 		     "mip=%u+%u layer=%u+%u usage=0x%x image_levels=%u image_layers=%u\n",
 		     static_cast<int>(image.format), static_cast<int>(normalized.format),

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

@@ -88,7 +88,9 @@ SelectSampledDepthView(vk::Format image_format, vk::Format view_format, uint32_t
 IsSupportedSampledDepthResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
 	return resource.kind == ShaderRecompiler::IR::ResourceKind::Image &&
 	       (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
-	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray) &&
+	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray ||
+	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa ||
+	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray) &&
 	       resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.read &&
 	       !resource.written && !resource.atomic;
 }

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,
@@ -397,10 +401,10 @@ TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64
 	return layout;
 }
 
-std::vector<vk::BufferImageCopy>
-TextureBuildImageCopies(const TextureUploadLayout& layout, uint32_t width, uint32_t height,
-                        uint32_t depth, uint64_t levels, bool array_texture,
-                        bool volume_texture) {
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
+                                                         uint32_t width, uint32_t height,
+                                                         uint32_t depth, uint64_t levels,
+                                                         bool array_texture, bool volume_texture) {
 	uint32_t mip_width  = width;
 	uint32_t mip_height = height;
 	uint32_t mip_pitch  = volume_texture && static_cast<Prospero::TileMode>(layout.tile) !=
@@ -418,10 +422,9 @@ TextureBuildImageCopies(const TextureUploadLayout& layout, uint32_t width, uint3
 		for (uint32_t z = 0; z < mip_depth; z++) {
 			const auto          slice_offset = z * layout.slice_stride;
 			vk::BufferImageCopy region {};
-			region.bufferOffset =
-			    layout.level_sizes[i].offset + slice_offset;
-			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i,
-			                           array_texture ? z : 0, 1};
+			region.bufferOffset     = layout.level_sizes[i].offset + slice_offset;
+			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, array_texture ? z : 0,
+			                           1};
 			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
 			region.imageExtent      = {mip_width, mip_height, 1};
 			const bool linear =
@@ -434,8 +437,7 @@ TextureBuildImageCopies(const TextureUploadLayout& layout, uint32_t width, uint3
 					return ((value + block - 1u) / block) * block;
 				};
 				const auto pitch       = align(mip_pitch, layout.texel_block);
-				region.bufferRowLength =
-				    pitch > align(mip_width, layout.texel_block) ? pitch : 0;
+				region.bufferRowLength = pitch > align(mip_width, layout.texel_block) ? pitch : 0;
 			}
 			regions.push_back(region);
 		}
@@ -480,11 +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;
@@ -522,9 +523,8 @@ bool TextureBuildGpuTileInfos(uint64_t size,
 			for (uint32_t z = 0; z < mip_depth; z += block.block_depth) {
 				const uint32_t copy_depth = std::min(block.block_depth, mip_depth - z);
 				const auto&    region     = regions[region_base + z];
-				const auto     pitch = region.bufferRowLength != 0
-				                           ? region.bufferRowLength
-				                           : region.imageExtent.width;
+				const auto     pitch =
+				    region.bufferRowLength != 0 ? region.bufferRowLength : region.imageExtent.width;
 				const auto  logical_height = region.bufferImageHeight != 0
 				                                 ? region.bufferImageHeight
 				                                 : region.imageExtent.height;
@@ -538,22 +538,20 @@ bool TextureBuildGpuTileInfos(uint64_t size,
 				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;
 				}
 				info.linear_slice_stride = linear_stride;
-				info.width = std::max(
-				    (region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height = std::max(
-				    (logical_height + element.tall - 1u) / element.tall, 1u);
+				info.width =
+				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
+				info.height = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
 				info.depth  = copy_depth;
-				info.surface_z = block.block_depth == 1
-				                     ? static_cast<uint32_t>(region.imageOffset.z)
-				                     : 0;
-				info.pitch =
-				    std::max((pitch + element.wide - 1u) / element.wide, 1u);
+				info.surface_z =
+				    block.block_depth == 1 ? static_cast<uint32_t>(region.imageOffset.z) : 0;
+				info.pitch        = std::max((pitch + element.wide - 1u) / element.wide, 1u);
 				info.tail_x       = tail ? volume.tail_x[level] : 0;
 				info.tail_y       = tail ? volume.tail_y[level] : 0;
 				info.tail         = tail;
@@ -581,9 +579,8 @@ bool TextureBuildGpuTileInfos(uint64_t size,
 			const auto level_depth = GetTextureLevelDepth(depth, level, layout.volume_texture);
 			for (uint32_t z = 0; z < level_depth; z++) {
 				const auto& region = regions[region_index++];
-				const auto pitch = region.bufferRowLength != 0
-				                       ? region.bufferRowLength
-				                       : region.imageExtent.width;
+				const auto  pitch =
+				    region.bufferRowLength != 0 ? region.bufferRowLength : region.imageExtent.width;
 				const auto  logical_height = region.bufferImageHeight != 0
 				                                 ? region.bufferImageHeight
 				                                 : region.imageExtent.height;
@@ -592,21 +589,20 @@ bool TextureBuildGpuTileInfos(uint64_t size,
 				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;
 				}
-				info.width = std::max(
-				    (region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height = std::max(
-				    (logical_height + element.tall - 1u) / element.tall, 1u);
+				info.width =
+				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
+				info.height    = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
 				info.surface_z = base_family == TileBlockFamily::RenderTarget64KB ||
 				                         base_family == TileBlockFamily::Depth64KB
 				                     ? region.imageSubresource.baseArrayLayer
 				                     : 0;
-				info.pitch =
-				    std::max((pitch + element.wide - 1u) / element.wide, 1u);
+				info.pitch     = std::max((pitch + element.wide - 1u) / element.wide, 1u);
 				info.tail      = tail;
 				info.tail_x    = tail ? level_size.x : 0;
 				info.tail_y    = tail ? level_size.y : 0;

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

@@ -40,12 +40,11 @@ TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64
                                                uint64_t tile, uint64_t upload_size,
                                                bool allow_depth_tile, bool volume_texture,
                                                const char* owner);
-std::vector<vk::BufferImageCopy>
-TextureBuildImageCopies(const TextureUploadLayout& layout, uint32_t width, uint32_t height,
-                        uint32_t depth, uint64_t levels, bool array_texture,
-                        bool volume_texture);
-bool TextureBuildGpuTileInfos(uint64_t size,
-                              const std::vector<vk::BufferImageCopy>& regions,
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
+                                                         uint32_t width, uint32_t height,
+                                                         uint32_t depth, uint64_t levels,
+                                                         bool array_texture, bool volume_texture);
+bool TextureBuildGpuTileInfos(uint64_t 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/image/tiler.cpp

@@ -14,9 +14,9 @@
 #include "gpu_tiler_shaders/gpu_tiler_standard64_spv.h"
 #include "gpu_tiler_shaders/gpu_tiler_swap_bgra16_spv.h"
 #include "graphics/host_gpu/graphicContext.h"
+#include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/image/image.h"
-#include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 
 #include <algorithm>
 #include <array>

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

@@ -26,11 +26,15 @@ bool IsSampledImage(BindingKind kind) {
 		case BindingKind::Sampled1DArray:
 		case BindingKind::Sampled2D:
 		case BindingKind::Sampled2DArray:
+		case BindingKind::Sampled2DMsaa:
+		case BindingKind::Sampled2DMsaaArray:
 		case BindingKind::Sampled3D:
 		case BindingKind::SampledUint1D:
 		case BindingKind::SampledUint1DArray:
 		case BindingKind::SampledUint2D:
 		case BindingKind::SampledUint2DArray:
+		case BindingKind::SampledUint2DMsaa:
+		case BindingKind::SampledUint2DMsaaArray:
 		case BindingKind::SampledUint3D: return true;
 		default: return false;
 	}

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

@@ -73,6 +73,11 @@ static Prospero::ImageType TextureBaseType(Prospero::ImageType type) {
 	}
 }
 
+static bool IsMultisampledTexture(Prospero::ImageType type) {
+	return type == Prospero::ImageType::kColor2DMsaa ||
+	       type == Prospero::ImageType::kColor2DMsaaArray;
+}
+
 static BufferView NativeStorageBuffer(RenderContext& context, CommandBuffer& command_buffer,
                                       const ShaderBufferResource&                 descriptor,
                                       const ShaderRecompiler::IR::BufferResource& resource,
@@ -159,6 +164,8 @@ static bool IsSupportedSampledColorResource(const ShaderRecompiler::IR::ImageRes
 		case ShaderRecompiler::Decoder::ImageDimension::Dim1DArray:
 		case ShaderRecompiler::Decoder::ImageDimension::Dim2D:
 		case ShaderRecompiler::Decoder::ImageDimension::Dim2DArray:
+		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa:
+		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray:
 			supported_dimension = true;
 			break;
 		default: break;
@@ -195,6 +202,22 @@ TargetTextureViewInfo ResolveTargetTextureView(const ShaderRecompiler::IR::Image
 			           ? TargetTextureViewInfo {vk::ImageViewType::e2DArray, base_layer,
 			                                    image_layers - base_layer}
 			           : TargetTextureViewInfo {};
+		case Prospero::ImageType::kColor2DMsaa:
+			return resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa &&
+			               base_layer == 0 && image_layers == 1
+			           ? TargetTextureViewInfo {vk::ImageViewType::e2D, 0, 1}
+			           : TargetTextureViewInfo {};
+		case Prospero::ImageType::kColor2DMsaaArray:
+			if (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa &&
+			    base_layer == 0 && image_layers == 1) {
+				return {vk::ImageViewType::e2D, 0, 1};
+			}
+			return resource.dimension ==
+			                   ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray &&
+			               base_layer < image_layers
+			           ? TargetTextureViewInfo {vk::ImageViewType::e2DArray, base_layer,
+			                                    image_layers - base_layer}
+			           : TargetTextureViewInfo {};
 		default: return {};
 	}
 }
@@ -211,34 +234,57 @@ bool IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& r
 bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, const Image& image) {
 	const auto width        = static_cast<uint32_t>(descriptor.Width5()) + 1u;
 	const auto height       = static_cast<uint32_t>(descriptor.Height5()) + 1u;
-	const auto pitch  = TileGetTexturePitch(descriptor.Format(), width, 1, descriptor.TileMode());
 	const auto type         = static_cast<Prospero::ImageType>(descriptor.Type());
-	const bool supported_single_layer =
+	const bool multisampled = IsMultisampledTexture(type);
+	const auto samples      = multisampled ? 1u << descriptor.LastLevel() : 1u;
+	const auto pitch =
+	    multisampled ? TileGetDepthPitch(width, image.info.bytes_per_block, descriptor.LastLevel())
+	                 : TileGetTexturePitch(descriptor.Format(), width, 1, descriptor.TileMode());
+	const bool supported_2d    = type == Prospero::ImageType::kColor2D &&
 	                             image.info.resources.layers == 1 && descriptor.Depth() == 0 &&
-	    descriptor.BaseArray5() == 0 &&
-	    (type == Prospero::ImageType::kColor2D || type == Prospero::ImageType::kColor2DArray);
+	                             descriptor.BaseArray5() == 0;
+	const bool supported_array = type == Prospero::ImageType::kColor2DArray &&
+	                             descriptor.BaseArray5() <= descriptor.Depth() &&
+	                             descriptor.Depth() < image.info.resources.layers;
 	const bool supported_cube =
 	    type == Prospero::ImageType::kCube && width == height && image.info.resources.layers >= 6 &&
 	    image.info.resources.layers % 6u == 0 &&
 	    static_cast<uint32_t>(descriptor.Depth()) + 1u == image.info.resources.layers &&
 	    descriptor.BaseArray5() == 0;
+	const bool supported_msaa_2d    = type == Prospero::ImageType::kColor2DMsaa &&
+	                                  image.info.resources.layers == 1 && descriptor.Depth() == 0 &&
+	                                  descriptor.BaseArray5() == 0;
+	const bool supported_msaa_array = type == Prospero::ImageType::kColor2DMsaaArray &&
+	                                  descriptor.BaseArray5() <= descriptor.Depth() &&
+	                                  descriptor.Depth() < image.info.resources.layers;
+	const bool levels_ok =
+	    multisampled
+	        ? descriptor.BaseLevel() == 0 && descriptor.LastLevel() >= 1 &&
+	              descriptor.LastLevel() <= 3 && descriptor.MaxMip() == descriptor.LastLevel() &&
+	              image.info.resources.levels == 1 && image.info.samples == samples
+	        : descriptor.BaseLevel() == 0 && descriptor.LastLevel() == 0 &&
+	              descriptor.MaxMip() == 0 && image.info.samples == 1;
 	return image.info.IsDepth() && width == image.info.extent.width &&
-	       height == image.info.extent.height && (supported_single_layer || supported_cube) &&
-	       descriptor.BaseLevel() == 0 && descriptor.LastLevel() == 0 && descriptor.MaxMip() == 0 &&
-	       descriptor.MinLod() == 0 && descriptor.BaseArray5() == 0 &&
+	       height == image.info.extent.height &&
+	       (supported_2d || supported_array || supported_cube || supported_msaa_2d ||
+	        supported_msaa_array) &&
+	       levels_ok && descriptor.MinLod() == 0 &&
 	       descriptor.TileMode() == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
-	       descriptor.BCSwizzle() == 0 && !descriptor.MsaaDepth() && pitch >= width &&
-	       pitch == image.info.pitch;
+	       descriptor.BCSwizzle() == 0 && (!descriptor.MsaaDepth() || multisampled) &&
+	       pitch >= width && pitch == image.info.pitch;
 }
 
 bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, const Image& image) {
 	constexpr uint32_t field1_reserved_mask = 0x200fff00u;
 	constexpr uint32_t field2_reserved_mask = 0xf0003000u;
-	constexpr uint32_t field3_common        = 0x01800000u;
-	constexpr uint32_t field5_expected      = 0x00700000u;
-	const uint32_t     field3_expected =
-	    (descriptor.Type() << 28u) | field3_common | descriptor.DstSelXYZW();
+	const uint32_t     field3_expected = descriptor.DstSelXYZW() |
+	                                     (static_cast<uint32_t>(descriptor.BaseLevel()) << 12u) |
+	                                     (static_cast<uint32_t>(descriptor.LastLevel()) << 16u) |
+	                                     (static_cast<uint32_t>(descriptor.TileMode()) << 20u) |
+	                                     (static_cast<uint32_t>(descriptor.Type()) << 28u);
 	const uint32_t     field4_expected = descriptor.Depth() | (descriptor.BaseArray5() << 16u);
+	const uint32_t     field5_expected =
+	    0x00700000u | (static_cast<uint32_t>(descriptor.MaxMip()) << 4u);
 	const bool common = (descriptor.fields[1] & field1_reserved_mask) == 0 &&
 	                    (descriptor.fields[2] & field2_reserved_mask) == 0 &&
 	                    descriptor.fields[3] == field3_expected &&
@@ -251,8 +297,9 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, co
 		return true;
 	}
 	constexpr uint32_t htile_control = 0x00280000u;
+	const uint32_t expected_control  = htile_control | (descriptor.MsaaDepth() ? (1u << 10u) : 0u);
 	const auto     metadata_addr     = descriptor.MetaAddr() << 8u;
-	return (descriptor.fields[6] & 0x00ffffffu) == htile_control && metadata_addr != 0 &&
+	return (descriptor.fields[6] & 0x00ffffffu) == expected_control && metadata_addr != 0 &&
 	       descriptor.TileMode() == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
 	       image.info.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) &&
 	       image.info.metadata.kind == ImageMetadataKind::Htile &&
@@ -344,9 +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();
 }
@@ -377,10 +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) &&
+	    raw_sint_storage ||
+	    (Prospero::IsSupportedTextureFormat(format) &&
 	     uint_resource == Prospero::IsUintTextureFormat(format) &&
-	    (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt));
+	     (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt)));
 	if (resource_ok && descriptor_ok && encoding_ok && format_ok && size != 0) {
 		return;
 	}
@@ -518,6 +568,7 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 			view.layer_count = 1;
 			break;
 		case ShaderRecompiler::Decoder::ImageDimension::Dim2DArray:
+		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray:
 			view.type       = vk::ImageViewType::e2DArray;
 			view.base_layer = descriptor.BaseArray5();
 			if (view.base_layer >= image_layers) {
@@ -526,6 +577,7 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 			view.layer_count = image_layers - view.base_layer;
 			break;
 		case ShaderRecompiler::Decoder::ImageDimension::Dim2D:
+		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa:
 			view.type       = vk::ImageViewType::e2D;
 			view.base_layer = descriptor.BaseArray5();
 			if (view.base_layer >= image_layers) {
@@ -562,19 +614,28 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	const auto base_level   = descriptor.BaseLevel();
 	const auto last_level   = descriptor.LastLevel();
 	const auto type         = TextureType(descriptor);
-	const bool multisampled =
-	    type == Prospero::ImageType::kColor2DMsaa || type == Prospero::ImageType::kColor2DMsaaArray;
+	const bool multisampled = IsMultisampledTexture(type);
 	const auto levels       = multisampled ? 1u : static_cast<uint32_t>(descriptor.MaxMip()) + 1u;
 	const auto tile         = descriptor.TileMode();
-	const bool msaa_tile  = tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
+	const bool depth_tile   = tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth);
+	const bool msaa_tile =
+	    depth_tile || tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
 	const bool msaa_array = type == Prospero::ImageType::kColor2DMsaaArray;
 	if ((!multisampled && (base_level > last_level || last_level >= levels)) ||
 	    (multisampled &&
 	     (base_level != 0 || last_level == 0 || last_level > 3 ||
-	      descriptor.MaxMip() != last_level || !msaa_tile || descriptor.MsaaDepth() ||
+	      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 =
@@ -601,7 +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");
@@ -616,12 +678,13 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 	                     (address & (static_cast<uint64_t>(size.align) - 1u)) != 0);
 	if (storage) {
 		ValidateStorageTexture(resource, descriptor, size.size);
-		m_context.GetBufferCache().ValidateGpuAccess(address, size.size, resource.read,
-		                                             resource.written);
 	}
 
 	const auto pixel_format = TextureGetFormat(format);
-	const auto              storage_view_format = SrgbStorageViewFormat(pixel_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

@@ -116,8 +116,8 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 		if (attachment_samples == 0) {
 			attachment_samples = depth.samples;
 		} else if (attachment_samples != depth.samples) {
-			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n",
-			     attachment_samples, depth.samples);
+			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n", attachment_samples,
+			     depth.samples);
 		}
 	}
 	EXIT_IF(attachment_samples == 0 ||
@@ -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++) {
@@ -203,9 +204,8 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id.hash0, vs_id.crc32, ps_id.hash0,
 	                 ps_id.crc32);
 	CreatePipelineInternal(m_graphics, m_descriptor_cache, *cached, rendering, vs_input_info,
-	                       vs_spirv, ps_input_info,
-	                       ps_spirv, static_params, vs_id.hash0, vs_id.crc32, ps_id.hash0,
-	                       ps_id.crc32, ps_active);
+	                       vs_spirv, ps_input_info, ps_spirv, static_params, vs_id.hash0,
+	                       vs_id.crc32, ps_id.hash0, ps_id.crc32, ps_active);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id.hash0, vs_id.crc32, ps_id.hash0,
 	                 ps_id.crc32);
 

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

@@ -118,11 +118,12 @@ public:
 		ShaderId cs_shader_id;
 	};
 
-	GraphicsPipeline& CreateGraphicsPipeline(
-	    RenderColorInfo* colors, uint32_t color_count, RenderDepthInfo& depth,
+	GraphicsPipeline&
+	CreateGraphicsPipeline(RenderColorInfo* colors, uint32_t color_count, RenderDepthInfo& depth,
 	                       ShaderVertexInputInfo& vs_input_info, RenderCommandBuffer& command,
-	    ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology, bool ps_active,
-	    std::span<const uint32_t> vs_spirv, std::span<const uint32_t> ps_spirv);
+	                       ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
+	                       bool ps_active, std::span<const uint32_t> vs_spirv,
+	                       std::span<const uint32_t> ps_spirv);
 	ComputePipeline& CreateComputePipeline(ShaderComputeInputInfo&      input_info,
 	                                       const HW::ComputeShaderInfo& cs_regs,
 	                                       std::span<const uint32_t>    cs_spirv);
@@ -211,16 +212,13 @@ private:
 
 void LogPipelineTrace(const char* phase, uint32_t vs_hash0, uint32_t vs_crc32, uint32_t ps_hash0,
                       uint32_t ps_crc32);
-void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descriptor_cache,
-                            PipelineCache::GraphicsPipeline& pipeline,
-                            const PipelineRenderingState&   rendering,
-                            const ShaderVertexInputInfo&    vs_input_info,
-                            std::span<const uint32_t>       vs_shader,
-                            const ShaderPixelInputInfo*     ps_input_info,
-                            std::span<const uint32_t>       ps_shader,
-                            const PipelineStaticParameters& static_params, uint32_t vs_hash0,
-                            uint32_t vs_crc32, uint32_t ps_hash0, uint32_t ps_crc32,
-                            bool ps_active);
+void CreatePipelineInternal(
+    GraphicContext& graphics, DescriptorCache& descriptor_cache,
+    PipelineCache::GraphicsPipeline& pipeline, const PipelineRenderingState& rendering,
+    const ShaderVertexInputInfo& vs_input_info, std::span<const uint32_t> vs_shader,
+    const ShaderPixelInputInfo* ps_input_info, std::span<const uint32_t> ps_shader,
+    const PipelineStaticParameters& static_params, uint32_t vs_hash0, uint32_t vs_crc32,
+    uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active);
 void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descriptor_cache,
                             PipelineCache::ComputePipeline& pipeline,
                             const ShaderComputeInputInfo&   input_info,

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

@@ -8,12 +8,13 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
+#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.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>
@@ -386,8 +387,7 @@ static vk::BlendOp GetBlendOp(uint32_t op) {
 }
 
 static void CreateLayout(DescriptorCache&                   descriptor_cache,
-                         std::span<vk::DescriptorSetLayout> set_layouts,
-                         uint32_t& set_layouts_num,
+                         std::span<vk::DescriptorSetLayout> set_layouts, uint32_t& set_layouts_num,
                          std::span<vk::PushConstantRange>     push_constant_info,
                          uint32_t&                            push_constant_info_num,
                          const ShaderRecompiler::IR::Program& program,
@@ -412,8 +412,7 @@ static void CreateLayout(DescriptorCache& descriptor_cache,
 	}
 }
 
-static void ConfigureSubgroupSize(const GraphicContext&                                  graphics,
-                                  vk::ShaderStageFlagBits                                vk_stage,
+static void ConfigureSubgroupSize(const GraphicContext& graphics, vk::ShaderStageFlagBits vk_stage,
                                   const ShaderRecompiler::IR::Program&                   program,
                                   vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo& required,
                                   vk::PipelineShaderStageCreateInfo&                     stage) {
@@ -456,19 +455,20 @@ static void ConfigureSubgroupSize(const GraphicContext&
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descriptor_cache,
-                            PipelineCache::GraphicsPipeline& pipeline,
-                            const PipelineRenderingState&   rendering,
-                            const ShaderVertexInputInfo&    vs_input_info,
-                            std::span<const uint32_t>       vs_shader,
-                            const ShaderPixelInputInfo*     ps_input_info,
-                            std::span<const uint32_t>       ps_shader,
-                            const PipelineStaticParameters& static_params, uint32_t vs_hash0,
-                            uint32_t vs_crc32, uint32_t ps_hash0, uint32_t ps_crc32,
-                            bool ps_active) {
+void CreatePipelineInternal(
+    GraphicContext& graphics, DescriptorCache& descriptor_cache,
+    PipelineCache::GraphicsPipeline& pipeline, const PipelineRenderingState& rendering,
+    const ShaderVertexInputInfo& vs_input_info, std::span<const uint32_t> vs_shader,
+    const ShaderPixelInputInfo* ps_input_info, std::span<const uint32_t> ps_shader,
+    const PipelineStaticParameters& static_params, uint32_t vs_hash0, uint32_t vs_crc32,
+    uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active) {
 	EXIT_IF(ps_active && ps_input_info == 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 {};
@@ -496,8 +496,33 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 		}
 		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 {};
@@ -511,8 +536,7 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	vert_shader_stage_info.pName               = "main";
 	vert_shader_stage_info.pSpecializationInfo = nullptr;
 	EXIT_IF(!vs_input_info.stage);
-	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eVertex,
-	                      *vs_input_info.stage.program,
+	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eVertex, *vs_input_info.stage.program,
 	                      vert_subgroup_size, vert_shader_stage_info);
 
 	vk::PipelineShaderStageCreateInfo                     frag_shader_stage_info {};
@@ -527,13 +551,32 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eFragment,
-		                      *ps_input_info->stage.program,
-		                      frag_subgroup_size, frag_shader_stage_info);
+		                      *ps_input_info->stage.program, frag_subgroup_size,
+		                      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];
@@ -838,15 +881,13 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 
 	EXIT_IF(!vs_input_info.stage);
 	CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-	             push_constant_info_num,
-	             *vs_input_info.stage.program, vk::ShaderStageFlagBits::eVertex,
-	             DescriptorCache::Stage::Vertex);
+	             push_constant_info_num, *vs_input_info.stage.program,
+	             vk::ShaderStageFlagBits::eVertex, DescriptorCache::Stage::Vertex);
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-		             push_constant_info_num,
-		             *ps_input_info->stage.program, vk::ShaderStageFlagBits::eFragment,
-		             DescriptorCache::Stage::Pixel);
+		             push_constant_info_num, *ps_input_info->stage.program,
+		             vk::ShaderStageFlagBits::eFragment, DescriptorCache::Stage::Pixel);
 	}
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
@@ -937,7 +978,10 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	pipeline_info.pStages                  = shader_stages;
 	pipeline_info.pVertexInputState        = &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = &input_assembly;
-	pipeline_info.pTessellationState  = nullptr;
+	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;
@@ -976,6 +1020,12 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	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);
 }
 
@@ -1012,8 +1062,7 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	comp_shader_stage_info.pName               = "main";
 	comp_shader_stage_info.pSpecializationInfo = nullptr;
 	EXIT_IF(!input_info.stage);
-	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eCompute,
-	                      *input_info.stage.program,
+	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eCompute, *input_info.stage.program,
 	                      comp_subgroup_size, comp_shader_stage_info);
 
 	vk::DescriptorSetLayout set_layouts[1]  = {};
@@ -1024,9 +1073,8 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 
 	EXIT_IF(!input_info.stage);
 	CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-	             push_constant_info_num,
-	             *input_info.stage.program, vk::ShaderStageFlagBits::eCompute,
-	             DescriptorCache::Stage::Compute);
+	             push_constant_info_num, *input_info.stage.program,
+	             vk::ShaderStageFlagBits::eCompute, DescriptorCache::Stage::Compute);
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
 	pipeline_layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -10,14 +10,14 @@
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/host_gpu/graphicContext.h"
+#include "graphics/host_gpu/renderer/image/imageInfo.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
-#include "graphics/host_gpu/renderer/image/imageInfo.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
-#include "graphics/host_gpu/renderer/render.h"
-#include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
+#include "graphics/host_gpu/renderer/render.h"
+#include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"

src/graphics/host_gpu/renderer/renderContext.cpp

@@ -14,8 +14,7 @@ namespace Libs::Graphics {
 RenderContext::RenderContext(GraphicContext& graphics)
     : m_graphics(graphics), m_render_executor(*this), m_command_scheduler(*this, graphics),
       m_descriptor_cache(graphics), m_pipeline_cache(graphics, m_descriptor_cache),
-      m_sampler_cache(graphics),
-      m_gpu_resources(graphics, m_command_scheduler) {
+      m_sampler_cache(graphics), m_gpu_resources(graphics, m_command_scheduler) {
 	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
 }
 
@@ -99,8 +98,7 @@ void RenderContext::TriggerEopEvent(uint32_t context_id) {
 		    registration.eq, static_cast<uintptr_t>(registration.id),
 		    LibKernel::EventQueue::KERNEL_EVFILT_GRAPHICS,
 		    reinterpret_cast<void*>(static_cast<uintptr_t>(context_id)));
-		if (result == LibKernel::KERNEL_ERROR_EBADF ||
-		    result == LibKernel::KERNEL_ERROR_ENOENT) {
+		if (result == LibKernel::KERNEL_ERROR_EBADF || result == LibKernel::KERNEL_ERROR_ENOENT) {
 			DeleteEopEq(registration.eq, registration.id);
 			continue;
 		}

src/graphics/host_gpu/renderer/renderContext.h

@@ -6,12 +6,12 @@
 #include "common/common.h"
 #include "common/threads.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
-#include "graphics/host_gpu/renderer/commandScheduler.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/cache/gpuResourceManager.h"
-#include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/cache/samplerCache.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
+#include "graphics/host_gpu/renderer/commandScheduler.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
+#include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "kernel/eventQueue.h"
 
 #include <memory>

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"
@@ -651,8 +650,6 @@ 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;
 };
@@ -767,7 +764,7 @@ 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 +788,7 @@ 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) {
@@ -991,20 +987,6 @@ 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 +999,12 @@ 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;
@@ -1159,7 +1131,7 @@ 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 +1173,7 @@ 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 +1201,6 @@ 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 +1260,7 @@ 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 +1298,6 @@ 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};
@@ -1346,19 +1310,14 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	}
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	const bool            use_ngg_rectlist_draw = Config::NggRectlistDrawEnabled();
-
-	if (!GetDrawTopology(ucfg, true, use_ngg_rectlist_draw, topology)) {
+	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: "
@@ -1375,12 +1334,9 @@ 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);
 
 	DrawIndexBufferSource index_source {};

src/graphics/host_gpu/renderer/sync.cpp

@@ -252,8 +252,8 @@ uint64_t PrepareVideoOutFlip(CommandBuffer& buffer, int handle, int index, int f
 	for (;;) {
 		uint64_t   request_id = 0;
 		auto&      video_out  = buffer.GetContext().GetVideoOut();
-		const auto result = video_out.SubmitFlipFromGpu(
-		    buffer, handle, index, flip_mode, flip_arg, request_id);
+		const auto result =
+		    video_out.SubmitFlipFromGpu(buffer, handle, index, flip_mode, flip_arg, request_id);
 		if (result == OK) {
 			EXIT_IF(request_id == 0);
 			return request_id;

src/graphics/host_gpu/vma.cpp

@@ -122,8 +122,8 @@ uint64_t GraphicContext::GetDeviceMemoryUsage() const {
 	    physical_device_properties.deviceType == vk::PhysicalDeviceType::eDiscreteGpu;
 	uint64_t usage = 0;
 	for (uint32_t heap = 0; heap < physical_device_memory_properties.memoryHeapCount; heap++) {
-		const bool device_local = static_cast<bool>(
-		    physical_device_memory_properties.memoryHeaps[heap].flags &
+		const bool device_local =
+		    static_cast<bool>(physical_device_memory_properties.memoryHeaps[heap].flags &
 		                      vk::MemoryHeapFlagBits::eDeviceLocal);
 		if (!discrete || device_local) {
 			usage += budgets[heap].usage;
@@ -160,8 +160,7 @@ uint64_t GraphicContext::GetTotalMemoryBudget() const {
 	}
 	constexpr uint64_t system_reserve = 8ull * 1024 * 1024 * 1024;
 	const auto         available      = budget > usage ? budget - usage : uint64_t {0};
-	return std::max(local,
-	                available > system_reserve ? available - system_reserve : uint64_t {0});
+	return std::max(local, available > system_reserve ? available - system_reserve : uint64_t {0});
 }
 
 void GraphicContext::CreateBuffer(uint64_t size, VulkanBuffer& buffer) {

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},
@@ -55,6 +56,10 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k32_32_32_32UInt, vk::Format::eR32G32B32A32Uint},
     {Prospero::BufferFormat::k32_32_32_32SInt, vk::Format::eR32G32B32A32Sint},
     {Prospero::BufferFormat::k32_32_32_32Float, vk::Format::eR32G32B32A32Sfloat},
+    // Narrow-channel sRGB formats are optional in Vulkan. Keep a same-width fallback until
+    // sampler-aware sRGB emulation is available.
+    {Prospero::BufferFormat::k8Srgb, vk::Format::eR8Unorm},
+    {Prospero::BufferFormat::k8_8Srgb, vk::Format::eR8G8Unorm},
     {Prospero::BufferFormat::k8_8_8_8Srgb, vk::Format::eR8G8B8A8Srgb},
     {Prospero::BufferFormat::k9_9_9_5Float, vk::Format::eE5B9G9R9UfloatPack32},
     {Prospero::BufferFormat::k5_6_5UNorm, vk::Format::eB5G6R5UnormPack16},

src/graphics/presentation/videoOut.cpp

@@ -359,9 +359,9 @@ static void TriggerVideoOutEvents(VideoOutConfig& video_out, VideoOutEventKind k
 		if (!registration || registration->generation != video_out.generation) {
 			continue;
 		}
-		const auto result = EventQueue::KernelTriggerEvent(
-		    registration->handle, VideoOutEventId(kind), EventQueue::KERNEL_EVFILT_VIDEO_OUT,
-		    trigger_data);
+		const auto result =
+		    EventQueue::KernelTriggerEvent(registration->handle, VideoOutEventId(kind),
+		                                   EventQueue::KERNEL_EVFILT_VIDEO_OUT, trigger_data);
 		EXIT_NOT_IMPLEMENTED(result != OK && result != LibKernel::KERNEL_ERROR_EBADF &&
 		                     result != LibKernel::KERNEL_ERROR_ENOENT);
 	}
@@ -372,9 +372,8 @@ static void DeleteVideoOutEvents(const VideoOutEventQueues& queues, VideoOutEven
 		if (!registration) {
 			continue;
 		}
-		const auto result =
-		    EventQueue::KernelDeleteEvent(registration->handle, VideoOutEventId(kind),
-		                                  EventQueue::KERNEL_EVFILT_VIDEO_OUT);
+		const auto result = EventQueue::KernelDeleteEvent(
+		    registration->handle, VideoOutEventId(kind), EventQueue::KERNEL_EVFILT_VIDEO_OUT);
 		EXIT_NOT_IMPLEMENTED(result != OK && result != LibKernel::KERNEL_ERROR_EBADF &&
 		                     result != LibKernel::KERNEL_ERROR_ENOENT);
 	}
@@ -425,17 +424,15 @@ static int RegisterVideoOutEvent(int handle, EventQueue::KernelEqueue eq, VideoO
 	bool                         add_queue = false;
 	{
 		Common::LockGuard event_lock(event_state->mutex);
-		const auto existing = std::find_if(queues.begin(), queues.end(), [&](const auto& candidate) {
+		const auto        existing =
+		    std::find_if(queues.begin(), queues.end(), [&](const auto& candidate) {
 			    return candidate->handle == eq && candidate->generation == generation;
 		    });
 		if (existing != queues.end()) {
 			registration = *existing;
 		} else {
-			registration = std::make_shared<VideoOutEventRegistration>(
-			    VideoOutEventRegistration {.handle     = eq,
-			                               .state      = event_state,
-			                               .generation = generation,
-			                               .kind       = kind});
+			registration = std::make_shared<VideoOutEventRegistration>(VideoOutEventRegistration {
+			    .handle = eq, .state = event_state, .generation = generation, .kind = kind});
 			queues.push_back(registration);
 			add_queue = true;
 		}
@@ -814,8 +811,8 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 				m_presenter.Present(*frame, true);
 			}
 			const auto frame_end = Common::Timer::QueryPerformanceCounter();
-			total_wait += static_cast<int64_t>(period) -
-			              static_cast<int64_t>(frame_end - frame_begin);
+			total_wait +=
+			    static_cast<int64_t>(period) - static_cast<int64_t>(frame_end - frame_begin);
 			continue;
 		}
 
@@ -841,8 +838,7 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 		VblankEnd();
 
 		const auto frame_end = Common::Timer::QueryPerformanceCounter();
-		total_wait += static_cast<int64_t>(period) -
-		              static_cast<int64_t>(frame_end - frame_begin);
+		total_wait += static_cast<int64_t>(period) - static_cast<int64_t>(frame_end - frame_begin);
 	}
 }
 
@@ -1000,8 +996,8 @@ void FlipQueue::Prepare(uint64_t request_id, Graphics::CommandBuffer& buffer) {
 	}
 	Graphics::Presenter::Frame* frame = nullptr;
 	if (special) {
-		frame = &m_presenter.PrepareBlankFrame(width, height,
-		                                       index == VIDEO_OUT_BUFFER_INDEX_BLACK, &buffer);
+		frame = &m_presenter.PrepareBlankFrame(width, height, index == VIDEO_OUT_BUFFER_INDEX_BLACK,
+		                                       &buffer);
 	} else {
 		frame = &m_presenter.PrepareFrame(buffer, source_info);
 	}

src/graphics/presentation/window/swapchain.cpp

@@ -206,21 +206,18 @@ private:
 	vk::Format                                     m_format     = vk::Format::eUndefined;
 };
 
-void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent,
-                                 vk::Format format) {
+void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent, vk::Format format) {
 	if (extent.width == 0 || extent.height == 0 || format == vk::Format::eUndefined) {
 		EXIT("unsupported prepared frame, extent=%ux%u format=%d\n", extent.width, extent.height,
 		     static_cast<int>(format));
 	}
 	const auto features = graphics.GetFormatProperties(format).optimalTilingFeatures;
-	const auto required = vk::FormatFeatureFlagBits::eBlitSrc |
-	                      vk::FormatFeatureFlagBits::eSampledImageFilterLinear |
-	                      vk::FormatFeatureFlagBits::eTransferSrc |
-	                      vk::FormatFeatureFlagBits::eTransferDst;
+	const auto required =
+	    vk::FormatFeatureFlagBits::eBlitSrc | vk::FormatFeatureFlagBits::eSampledImageFilterLinear |
+	    vk::FormatFeatureFlagBits::eTransferSrc | vk::FormatFeatureFlagBits::eTransferDst;
 	if ((features & required) != required) {
 		EXIT("prepared presentation format lacks optimal blit support: format=%d features=0x%x\n",
-		     static_cast<int>(format),
-		     static_cast<vk::FormatFeatureFlags::MaskType>(features));
+		     static_cast<int>(format), static_cast<vk::FormatFeatureFlags::MaskType>(features));
 	}
 
 	auto&      dst        = image;
@@ -267,8 +264,7 @@ void Presenter::Frame::Transit(vk::CommandBuffer command, vk::ImageLayout layout
 	                            ? vk::PipelineStageFlagBits2::eTransfer
 	                            : vk::PipelineStageFlagBits2::eAllCommands;
 	constexpr auto writes = vk::AccessFlagBits2::eTransferWrite |
-	                        vk::AccessFlagBits2::eShaderWrite |
-	                        vk::AccessFlagBits2::eMemoryWrite;
+	                        vk::AccessFlagBits2::eShaderWrite | vk::AccessFlagBits2::eMemoryWrite;
 	if (image.state.layout == layout && image.state.access_mask == access &&
 	    !static_cast<bool>(image.state.access_mask & writes)) {
 		return;
@@ -299,35 +295,27 @@ void Presenter::Frame::Transit(vk::CommandBuffer command, vk::ImageLayout layout
 void Presenter::Frame::CopyFrom(CommandBuffer& command_buffer, Image& source) {
 	command_buffer.EndRendering();
 	auto command = command_buffer.Handle();
-	source.Transit(vk::ImageLayout::eTransferSrcOptimal,
-	               vk::AccessFlagBits2::eTransferRead, {}, command);
-	Transit(command, vk::ImageLayout::eTransferDstOptimal,
-	        vk::AccessFlagBits2::eTransferWrite);
+	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {},
+	               command);
+	Transit(command, vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite);
 	vk::ImageCopy copy {};
-	copy.srcSubresource = {vk::ImageAspectFlagBits::eColor, 0, 0,
-	                       source.backing.layers};
+	copy.srcSubresource = {vk::ImageAspectFlagBits::eColor, 0, 0, source.backing.layers};
 	copy.dstSubresource = {vk::ImageAspectFlagBits::eColor, 0, 0, image.layers};
 	copy.extent         = {std::min(source.backing.extent.width, image.extent.width),
 	                       std::min(source.backing.extent.height, image.extent.height), 1};
 	EXIT_IF(copy.srcSubresource.layerCount != copy.dstSubresource.layerCount);
-	command.copyImage(source.backing.image, vk::ImageLayout::eTransferSrcOptimal,
-	                  image.image, vk::ImageLayout::eTransferDstOptimal, copy);
-	Transit(command, vk::ImageLayout::eTransferSrcOptimal,
-	        vk::AccessFlagBits2::eTransferRead);
+	command.copyImage(source.backing.image, vk::ImageLayout::eTransferSrcOptimal, image.image,
+	                  vk::ImageLayout::eTransferDstOptimal, copy);
+	Transit(command, vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead);
 }
 
-void Presenter::Frame::Clear(CommandBuffer& command_buffer,
-                             const vk::ClearColorValue& color) {
+void Presenter::Frame::Clear(CommandBuffer& command_buffer, const vk::ClearColorValue& color) {
 	command_buffer.EndRendering();
 	auto command = command_buffer.Handle();
-	Transit(command, vk::ImageLayout::eTransferDstOptimal,
-	        vk::AccessFlagBits2::eTransferWrite);
-	const vk::ImageSubresourceRange range {
-	    vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1};
-	command.clearColorImage(image.image, vk::ImageLayout::eTransferDstOptimal, &color, 1,
-	                        &range);
-	Transit(command, vk::ImageLayout::eTransferSrcOptimal,
-	        vk::AccessFlagBits2::eTransferRead);
+	Transit(command, vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite);
+	const vk::ImageSubresourceRange range {vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1};
+	command.clearColorImage(image.image, vk::ImageLayout::eTransferDstOptimal, &color, 1, &range);
+	Transit(command, vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead);
 }
 
 class Swapchain final {
@@ -441,16 +429,12 @@ void Swapchain::Create() {
 	        ? vk::CompositeAlphaFlagBitsKHR::eOpaque
 	        : vk::CompositeAlphaFlagBitsKHR::eInherit;
 
-	vk::SurfaceFormatKHR format {vk::Format::eR8G8B8A8Unorm,
-	                             vk::ColorSpaceKHR::eSrgbNonlinear};
-	if (surface.formats.size() != 1 ||
-	    surface.formats.front().format != vk::Format::eUndefined) {
+	vk::SurfaceFormatKHR format {vk::Format::eR8G8B8A8Unorm, vk::ColorSpaceKHR::eSrgbNonlinear};
+	if (surface.formats.size() != 1 || surface.formats.front().format != vk::Format::eUndefined) {
 		const auto it = std::find_if(surface.formats.begin(), surface.formats.end(),
 		                             [](const vk::SurfaceFormatKHR& candidate) {
-			                             return candidate.format ==
-			                                        vk::Format::eB8G8R8A8Unorm ||
-			                                    candidate.format ==
-			                                        vk::Format::eR8G8B8A8Unorm;
+			                             return candidate.format == vk::Format::eB8G8R8A8Unorm ||
+			                                    candidate.format == vk::Format::eR8G8B8A8Unorm;
 		                             });
 		if (it == surface.formats.end()) {
 			EXIT("no supported UNORM swapchain format\n");
@@ -458,8 +442,7 @@ void Swapchain::Create() {
 		format = *it;
 	}
 	m_format                      = format.format;
-	const auto swapchain_features =
-	    graphics.GetFormatProperties(m_format).optimalTilingFeatures;
+	const auto swapchain_features = graphics.GetFormatProperties(m_format).optimalTilingFeatures;
 	if (!static_cast<bool>(swapchain_features & vk::FormatFeatureFlagBits::eBlitDst)) {
 		EXIT("swapchain format cannot be a blit destination: format=%d\n",
 		     static_cast<int>(m_format));
@@ -503,8 +486,7 @@ void Swapchain::Create() {
 		view.subresourceRange.baseMipLevel   = 0;
 		view.subresourceRange.layerCount     = 1;
 		view.subresourceRange.levelCount     = 1;
-		RequireVulkanSuccess(
-		    graphics.device.createImageView(&view, nullptr, &m_image_views[i]),
+		RequireVulkanSuccess(graphics.device.createImageView(&view, nullptr, &m_image_views[i]),
 		                     "vkCreateImageView");
 		EXIT_IF(m_image_views[i] == nullptr);
 	}
@@ -683,10 +665,9 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	to_present.subresourceRange.levelCount     = 1;
 	to_present.subresourceRange.baseArrayLayer = 0;
 	to_present.subresourceRange.layerCount     = 1;
-	vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
-	                           vk::PipelineStageFlagBits::eAllCommands,
-	                           vk::DependencyFlagBits::eByRegion, 0,
-	                           nullptr, 0, nullptr, 1, &to_present);
+	vk_command.pipelineBarrier(
+	    vk::PipelineStageFlagBits::eAllCommands, vk::PipelineStageFlagBits::eAllCommands,
+	    vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1, &to_present);
 	command.End();
 }
 
@@ -752,8 +733,7 @@ Presenter::Frame& Presenter::PrepareFrame(CommandBuffer& buffer, const ImageInfo
 		default: break;
 	}
 	frame->Configure(m_impl->window.graphic_ctx,
-	                 {image.backing.extent.width, image.backing.extent.height},
-	                 frame_format);
+	                 {image.backing.extent.width, image.backing.extent.height}, frame_format);
 	frame->CopyFrom(buffer, image);
 	return *frame;
 }
@@ -772,8 +752,7 @@ Presenter::Frame& Presenter::PrepareBlankFrame(uint32_t width, uint32_t height,
 		frame->Clear(*producer, clear);
 	} else {
 		if (frame->present_commands == nullptr) {
-			frame->present_commands =
-			    std::make_unique<CommandBuffer>(m_impl->present_scheduler);
+			frame->present_commands = std::make_unique<CommandBuffer>(m_impl->present_scheduler);
 		}
 		auto& command = *frame->present_commands;
 		command.WaitForFenceAndReset();
@@ -830,8 +809,7 @@ void Presenter::Present(Frame& frame, bool reuse) {
 			continue;
 		}
 		if (frame.present_commands == nullptr) {
-			frame.present_commands =
-			    std::make_unique<CommandBuffer>(m_impl->present_scheduler);
+			frame.present_commands = std::make_unique<CommandBuffer>(m_impl->present_scheduler);
 		}
 		{
 			Common::LockGuard render_lock(m_impl->renderer.GetMutex());

src/graphics/presentation/window/vulkanWindow.cpp

@@ -32,11 +32,11 @@
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/presenter.h"
-#include "kernel/memory.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/presentation/window/windowInternal.h"
+#include "kernel/memory.h"
 #include "libs/controller.h"
 #include "loader/systemContent.h"
 
@@ -909,10 +909,9 @@ void WindowContext::CreateVulkan() {
 	}
 	surface = native_surface;
 
-	std::vector<const char*> device_extensions = {VK_KHR_SWAPCHAIN_EXTENSION_NAME,
-	                                              VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME,
-	                                              VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
-	                                              "VK_KHR_maintenance1"};
+	std::vector<const char*> device_extensions = {
+	    VK_KHR_SWAPCHAIN_EXTENSION_NAME, VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME,
+	    VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME, "VK_KHR_maintenance1"};
 
 #if defined(__APPLE__)
 	// MoltenVK lacks VK_EXT_depth_clip_enable and VK_EXT_color_write_enable; the renderer
@@ -932,8 +931,8 @@ void WindowContext::CreateVulkan() {
 
 	uint32_t queue_family = static_cast<uint32_t>(-1);
 
-	VulkanFindPhysicalDevice(graphic_ctx.instance, surface, device_extensions,
-	                         surface_capabilities, graphic_ctx.physical_device, queue_family);
+	VulkanFindPhysicalDevice(graphic_ctx.instance, surface, device_extensions, surface_capabilities,
+	                         graphic_ctx.physical_device, queue_family);
 
 	if (graphic_ctx.physical_device == nullptr) {
 		EXIT("Could not find suitable device");
@@ -949,9 +948,8 @@ void WindowContext::CreateVulkan() {
 		auto available_extensions = EnumerateVulkan<vk::ExtensionProperties>(
 		    "vkEnumerateDeviceExtensionProperties",
 		    [&](uint32_t* count, vk::ExtensionProperties* values) {
-			    return graphic_ctx.physical_device.enumerateDeviceExtensionProperties(nullptr,
-			                                                                           count,
-			                                                                           values);
+			    return graphic_ctx.physical_device.enumerateDeviceExtensionProperties(
+			        nullptr, count, values);
 		    });
 
 		if (HasExtension(available_extensions, VK_EXT_MEMORY_BUDGET_EXTENSION_NAME)) {

src/graphics/presentation/window/window.cpp

@@ -1,7 +1,5 @@
 #include "graphics/presentation/window.h"
 
-#include <cstdlib>
-
 #include "SDL.h"
 #include "SDL_error.h"
 #include "SDL_events.h"
@@ -40,6 +38,7 @@
 
 #include <algorithm>
 #include <cstdio>
+#include <cstdlib>
 #include <cstring>
 #include <memory>
 #include <string>
@@ -251,9 +250,7 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 	if (key.down) {
 		switch (key.key_code) {
 			case SDLK_ESCAPE: game.need_exit = true; break;
-			case SDLK_SPACE:
-				SetPause(game, !game.paused.load(std::memory_order_acquire));
-				break;
+			case SDLK_SPACE: SetPause(game, !game.paused.load(std::memory_order_acquire)); break;
 			case SDLK_F1:
 				if (!key.repeat) {
 					RenderDocRequestCapture();
@@ -390,7 +387,9 @@ void WindowContext::Resize(uint32_t new_width, uint32_t new_height) {
 void WindowContext::ProcessWindowEvent(const SDL_WindowEvent& event) {
 	const auto& window_event = event;
 	switch (window_event.event) {
-		case SDL_WINDOWEVENT_SHOWN: LOGF("Window %" PRIu32 " shown\n", window_event.windowID); break;
+		case SDL_WINDOWEVENT_SHOWN:
+			LOGF("Window %" PRIu32 " shown\n", window_event.windowID);
+			break;
 
 		case SDL_WINDOWEVENT_HIDDEN:
 			LOGF("Window %" PRIu32 " hidden\n", window_event.windowID);
@@ -401,13 +400,13 @@ void WindowContext::ProcessWindowEvent(const SDL_WindowEvent& event) {
 			break;
 
 		case SDL_WINDOWEVENT_MOVED:
-			LOGF("Window %" PRIu32 " moved to %" PRId32 ",%" PRId32 "\n",
-			     window_event.windowID, window_event.data1, window_event.data2);
+			LOGF("Window %" PRIu32 " moved to %" PRId32 ",%" PRId32 "\n", window_event.windowID,
+			     window_event.data1, window_event.data2);
 			break;
 
 		case SDL_WINDOWEVENT_RESIZED:
-			LOGF("Window %" PRIu32 " resized to %" PRId32 "x%" PRId32 "\n",
-			     window_event.windowID, window_event.data1, window_event.data2);
+			LOGF("Window %" PRIu32 " resized to %" PRId32 "x%" PRId32 "\n", window_event.windowID,
+			     window_event.data1, window_event.data2);
 
 			LOGF("m: %d\n", static_cast<int>(SDL_ThreadID()));
 			Resize(window_event.data1, window_event.data2);
@@ -807,8 +806,7 @@ static void WindowCreate(WindowContext& context) {
 		window_flags |= static_cast<uint32_t>(SDL_WINDOW_BORDERLESS);
 	}
 #endif
-	context.window =
-	    SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, KYTY_SDL_WINDOWPOS_CENTERED,
+	context.window = SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, KYTY_SDL_WINDOWPOS_CENTERED,
 	                                  KYTY_SDL_WINDOWPOS_CENTERED, width, height, window_flags);
 
 	context.window_hidden = true;
@@ -950,11 +948,11 @@ void WindowContext::UpdateTitle() {
 		fps_frames  = 0;
 	}
 
-	auto fps = fmt::format("{}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", (has_title ? title : ""),
+	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);
+	                (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/presentation/window/windowInternal.h

@@ -38,8 +38,7 @@ struct WindowContext {
 	~WindowContext();
 	KYTY_CLASS_NO_COPY(WindowContext);
 
-	[[nodiscard]] static vk::PhysicalDeviceVulkan13Features
-	RequiredVulkan13Features() noexcept;
+	[[nodiscard]] static vk::PhysicalDeviceVulkan13Features RequiredVulkan13Features() noexcept;
 	void                                                    CreateVulkan();
 	void                                                    RecreateSurface();
 	void                                                    RefreshSurfaceCapabilities();

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -2,15 +2,16 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "graphics/shader/recompiler/cfg/ShaderCFG.h"
+#include "graphics/shader/recompiler/decompiler/ShaderDecoder.h"
+#include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
 #include "graphics/shader/recompiler/ir/BindingLayout.h"
+#include "graphics/shader/recompiler/ir/ReadLaneElimination.h"
 #include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ResourceTracking.h"
 #include "graphics/shader/recompiler/ir/ScalarProvenance.h"
-#include "graphics/shader/recompiler/cfg/ShaderCFG.h"
-#include "graphics/shader/recompiler/decompiler/ShaderDecoder.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/recompiler/ir/ShaderInfoCollection.h"
-#include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
 #include "graphics/shader/recompiler/ir/SrtPatcher.h"
 #include "graphics/shader/recompiler/ir/SrtWalker.h"
 
@@ -838,6 +839,11 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	if (!IR::AllocateBindings(ir, layout_options, error)) {
 		return false;
 	}
+	const auto read_lane_stats = IR::EliminateReadLane(ir);
+	if (read_lane_stats.rewritten_reads != 0) {
+		LOGF("%s read-lane elimination: reads=%" PRIu32 " shadow_writes=%" PRIu32 "\n",
+		     GetDumpLabel(options), read_lane_stats.rewritten_reads, read_lane_stats.shadow_writes);
+	}
 	std::string ir_dump;
 	if (options.dump_ir) {
 		ir_dump = MakeIrDump(cfg, ir);

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/decompiler/ImageOps.cpp

@@ -35,9 +35,9 @@ constexpr ImageDimension DecodeImageDimension(uint32_t dim) {
 		case 2u: return ImageDimension::Dim3D;
 		case 3u: return ImageDimension::Dim2DArray;
 		case 4u: return ImageDimension::Dim1DArray;
-		case 5u:
-		case 7u: return ImageDimension::Dim2DArray;
-		case 6u: return ImageDimension::Dim2D;
+		case 5u: return ImageDimension::Dim2DArray;
+		case 6u: return ImageDimension::Dim2DMsaa;
+		case 7u: return ImageDimension::Dim2DMsaaArray;
 		default: return ImageDimension::Unknown;
 	}
 }
@@ -46,8 +46,10 @@ constexpr uint32_t ImageCoordComponents(ImageDimension dimension) {
 	switch (dimension) {
 		case ImageDimension::Dim1D: return 1u;
 		case ImageDimension::Dim1DArray: return 2u;
+		case ImageDimension::Dim2DMsaa:
 		case ImageDimension::Dim3D:
 		case ImageDimension::Dim2DArray: return 3u;
+		case ImageDimension::Dim2DMsaaArray: return 4u;
 		default: return 2u;
 	}
 }

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

@@ -442,11 +442,6 @@ bool DecodeDs(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
 	     inst.opcode == Opcode::DsReadAddtidB32)) {
 		SetUnsupported(inst, Family::DS, opcode, "DS swizzle/addtid is available only for LDS");
 	}
-	if (inst.gds && (inst.opcode == Opcode::DsAppend || inst.opcode == Opcode::DsConsume) &&
-	    inst.offset != 0u) {
-		SetUnsupported(inst, Family::DS, opcode,
-		               "GDS append/consume requires a zero instruction offset");
-	}
 	if (inst.opcode == Opcode::DsWriteAddtidB32 && data1 != 0u) {
 		SetUnsupported(inst, Family::DS, opcode,
 		               "DS write addtid data1 operand is not implemented");

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

@@ -194,6 +194,8 @@ const char* ImageDimensionToString(ImageDimension dimension) {
 		case ImageDimension::Dim2D: return "2d";
 		case ImageDimension::Dim3D: return "3d";
 		case ImageDimension::Dim2DArray: return "2d_array";
+		case ImageDimension::Dim2DMsaa: return "2d_msaa";
+		case ImageDimension::Dim2DMsaaArray: return "2d_msaa_array";
 		default: return "unknown";
 	}
 }

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

@@ -575,6 +575,8 @@ enum class ImageDimension : uint32_t {
 	Dim2D,
 	Dim3D,
 	Dim2DArray,
+	Dim2DMsaa,
+	Dim2DMsaaArray,
 };
 
 constexpr uint32_t MaxInstructionRawWords       = 5u;

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

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

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

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

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

@@ -1,7 +1,7 @@
 #include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
 
-#include "graphics/shader/recompiler/ir/SrtWalker.h"
 #include "graphics/shader/recompiler/emitter/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/ir/SrtWalker.h"
 
 #include <algorithm>
 #include <array>
@@ -30,10 +30,16 @@ bool ImageBinding(const IR::ImageResource& image, IR::DescriptorBindingKind& kin
 				kind = integer ? Kind::SampledUint1DArray : Kind::Sampled1DArray;
 				return true;
 			case Dim::Dim2D: kind = integer ? Kind::SampledUint2D : Kind::Sampled2D; return true;
+			case Dim::Dim2DMsaa:
+				kind = integer ? Kind::SampledUint2DMsaa : Kind::Sampled2DMsaa;
+				return true;
 			case Dim::Dim3D: kind = integer ? Kind::SampledUint3D : Kind::Sampled3D; return true;
 			case Dim::Dim2DArray:
 				kind = integer ? Kind::SampledUint2DArray : Kind::Sampled2DArray;
 				return true;
+			case Dim::Dim2DMsaaArray:
+				kind = integer ? Kind::SampledUint2DMsaaArray : Kind::Sampled2DMsaaArray;
+				return true;
 			case Dim::Unknown: return false;
 		}
 	}
@@ -51,6 +57,8 @@ bool ImageBinding(const IR::ImageResource& image, IR::DescriptorBindingKind& kin
 		case Dim::Dim2DArray:
 			kind = uint_image ? Kind::StorageUint2DArray : Kind::Storage2DArray;
 			return true;
+		case Dim::Dim2DMsaa:
+		case Dim::Dim2DMsaaArray: return false;
 		case Dim::Unknown: return false;
 	}
 	return false;
@@ -160,8 +168,7 @@ bool ValidateInstructionContract(const IR::Instruction& inst, std::string* error
 	                            inst.dst.kind != IR::OperandKind::Null)) ||
 	    ((inst.op == IR::Opcode::DsAppend || inst.op == IR::Opcode::DsConsume) &&
 	     (!ds_kind || !ds_resource || inst.src_count != 1 ||
-	      inst.dst.kind != IR::OperandKind::Register ||
-	      (kind == IR::ResourceKind::Gds && inst.memory.offset != 0))) ||
+	      inst.dst.kind != IR::OperandKind::Register)) ||
 	    ((inst.op == IR::Opcode::DsMinF32 || inst.op == IR::Opcode::DsMaxF32) &&
 	     (!ds_kind || !ds_resource || inst.src_count != 3 ||
 	      inst.dst.kind != IR::OperandKind::Null)) ||

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) {
@@ -276,8 +255,7 @@ void CopyProgramInputsAndOutputs(EmitterState& state, const IR::Program& program
 		if (HasOutput(state.outputs, output.kind, output.index)) {
 			continue;
 		}
-		state.outputs.push_back(
-		    {output.kind, output.index, output.location, 0, output.debug_name});
+		state.outputs.push_back({output.kind, output.index, output.location, 0, output.debug_name});
 	}
 }
 
@@ -576,6 +554,8 @@ ImageViewKind ImageViewKindFromDimension(Decoder::ImageDimension dimension) {
 		case Decoder::ImageDimension::Dim1DArray: return ImageViewKind::Dim1DArray;
 		case Decoder::ImageDimension::Dim2DArray: return ImageViewKind::Dim2DArray;
 		case Decoder::ImageDimension::Dim3D: return ImageViewKind::Dim3D;
+		case Decoder::ImageDimension::Dim2DMsaa: return ImageViewKind::Dim2DMsaa;
+		case Decoder::ImageDimension::Dim2DMsaaArray: return ImageViewKind::Dim2DMsaaArray;
 		default: return ImageViewKind::Dim2D;
 	}
 }
@@ -601,7 +581,9 @@ uint32_t ImageViewCoordinateComponents(ImageViewKind view) {
 		case ImageViewKind::Dim1DArray:
 		case ImageViewKind::Dim2D: return 2u;
 		case ImageViewKind::Dim2DArray:
+		case ImageViewKind::Dim2DMsaaArray:
 		case ImageViewKind::Dim3D: return 3u;
+		case ImageViewKind::Dim2DMsaa: return 2u;
 		default: return 0u;
 	}
 }
@@ -611,7 +593,9 @@ uint32_t ImageViewSpatialComponents(ImageViewKind view) {
 		case ImageViewKind::Dim1D:
 		case ImageViewKind::Dim1DArray: return 1u;
 		case ImageViewKind::Dim2D:
-		case ImageViewKind::Dim2DArray: return 2u;
+		case ImageViewKind::Dim2DArray:
+		case ImageViewKind::Dim2DMsaa:
+		case ImageViewKind::Dim2DMsaaArray: return 2u;
 		case ImageViewKind::Dim3D: return 3u;
 		default: return 0u;
 	}
@@ -663,8 +647,7 @@ uint32_t LoadSampledImageDescriptor(EmitterState& state, const IR::MemoryInfo& m
 
 uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler, uint32_t use_pc) {
 	(void)use_pc;
-	const auto binding =
-	    ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
+	const auto binding = ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
 	const auto pointer = DescriptorElementPointer(
 	    state, state.ptr_uniform_sampler, state.sampler_variable, binding.array_index,
 	    IR::DescriptorBindingKind::Samplers, sampler, "sampler descriptor array was not emitted");

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

@@ -1025,15 +1025,47 @@ void EmitDispatcherSwitch(EmitterState& state, const IR::Program& program) {
 	EmitDispatcherExit(state);
 }
 
+size_t BufferLoadGroupSize(const IR::BasicBlock& block, size_t first_index) {
+	const auto& first = block.instructions[first_index];
+	if (first.op != IR::Opcode::BufferLoadDword || first.memory.component_index != 0u ||
+	    first.memory.component_count <= 1u) {
+		return 1u;
+	}
+
+	size_t count = 1u;
+	while (first_index + count < block.instructions.size() &&
+	       count < first.memory.component_count) {
+		const auto& next = block.instructions[first_index + count];
+		if (next.op != IR::Opcode::BufferLoadDword || next.pc != first.pc ||
+		    next.memory.component_index != count ||
+		    next.memory.component_count != first.memory.component_count) {
+			break;
+		}
+		count++;
+	}
+	return count;
+}
+
+void EmitBlockInstructions(EmitterState& state, const IR::BasicBlock& block) {
+	for (size_t i = 0; i < block.instructions.size();) {
+		const auto count = BufferLoadGroupSize(block, i);
+		if (count > 1u) {
+			EmitBufferLoadDwordGroup(state, block.instructions.data() + i,
+			                         static_cast<uint32_t>(count));
+		} else {
+			EmitInstruction(state, block.instructions[i]);
+		}
+		i += count;
+	}
+}
+
 void EmitDispatcherBlocks(EmitterState& state, const IR::Program& program) {
 	for (const auto& block: program.blocks) {
 		if (block.id >= state.reachable_blocks.size() || !state.reachable_blocks[block.id]) {
 			continue;
 		}
 		state.builder.AddFunction({OpLabel, BlockLabel(state, block.id)});
-		for (const auto& inst: block.instructions) {
-			EmitInstruction(state, inst);
-		}
+		EmitBlockInstructions(state, block);
 		EmitDispatcherTerminator(state, block.terminator);
 	}
 }
@@ -1083,9 +1115,7 @@ void EmitFunction(EmitterState& state, const IR::Program& program) {
 			continue;
 		}
 		state.builder.AddFunction({OpLabel, BlockLabel(state, block.id)});
-		for (const auto& inst: block.instructions) {
-			EmitInstruction(state, inst);
-		}
+		EmitBlockInstructions(state, block);
 		EmitTerminator(state, block.terminator);
 	}
 

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

@@ -36,8 +36,8 @@ uint32_t EmitExportVec4F32(EmitterState& state, const IR::Instruction& inst) {
 			}
 		}
 		const auto vec = state.builder.AllocateId();
-		state.builder.AddFunction({OpCompositeConstruct, state.vec4_float_type, vec,
-		                            components[0], components[1], components[2], components[3]});
+		state.builder.AddFunction({OpCompositeConstruct, state.vec4_float_type, vec, components[0],
+		                           components[1], components[2], components[3]});
 		return vec;
 	}
 
@@ -50,7 +50,59 @@ uint32_t EmitExportVec4F32(EmitterState& state, const IR::Instruction& inst) {
 	return vec;
 }
 
-uint32_t ApplyMrtExportMapping(EmitterState& state, const IR::Instruction& inst, uint32_t value) {
+uint32_t EmitExportComponentU32(EmitterState& state, const IR::Instruction& inst,
+                                uint32_t component) {
+	const bool enabled = ((inst.export_info.en >> component) & 1u) != 0;
+	if (!enabled || component >= inst.src_count || component >= 4u) {
+		return ConstantU32(state, component == 3u ? 1u : 0u);
+	}
+	return EmitValueLoad(state, inst.src[component]);
+}
+
+uint32_t EmitExportVec4U32(EmitterState& state, const IR::Instruction& inst) {
+	uint32_t components[4] = {
+	    ConstantU32(state, 0u),
+	    ConstantU32(state, 0u),
+	    ConstantU32(state, 0u),
+	    ConstantU32(state, 1u),
+	};
+
+	if (inst.export_info.compr) {
+		for (uint32_t pair_index = 0; pair_index < 2u && pair_index < inst.src_count;
+		     pair_index++) {
+			const auto raw = EmitValueLoad(state, inst.src[pair_index]);
+			for (uint32_t lane = 0; lane < 2u; lane++) {
+				const auto component = pair_index * 2u + lane;
+				if (((inst.export_info.en >> component) & 1u) == 0) {
+					continue;
+				}
+				components[component] = state.builder.AllocateId();
+				state.builder.AddFunction(
+				    {OpBitFieldUExtract, state.uint_type, components[component], raw,
+				     ConstantU32(state, lane * 16u), ConstantU32(state, 16u)});
+			}
+		}
+	} else {
+		for (uint32_t component = 0; component < 4u; component++) {
+			components[component] = EmitExportComponentU32(state, inst, component);
+		}
+	}
+
+	const auto vec = state.builder.AllocateId();
+	state.builder.AddFunction({OpCompositeConstruct, state.vec4_uint_type, vec, components[0],
+	                           components[1], components[2], components[3]});
+	return vec;
+}
+
+static bool MrtUsesUintOutput(const EmitterState& state, const IR::Instruction& inst) {
+	return inst.export_info.kind == IR::ExportTargetKind::Mrt &&
+	       state.pixel_input_info != nullptr &&
+	       inst.export_info.index < std::size(state.pixel_input_info->target_output_mode) &&
+	       state.pixel_input_info->target_output_mode[inst.export_info.index] == 7u;
+}
+
+uint32_t ApplyMrtExportMapping(EmitterState& state, const IR::Instruction& inst, uint32_t value,
+                               uint32_t vector_type) {
 	if (inst.export_info.kind != IR::ExportTargetKind::Mrt || state.pixel_input_info == nullptr ||
 	    inst.export_info.index >= state.pixel_input_info->target_export_mapping.size()) {
 		return value;
@@ -62,8 +114,8 @@ uint32_t ApplyMrtExportMapping(EmitterState& state, const IR::Instruction& inst,
 	}
 
 	const auto mapped = state.builder.AllocateId();
-	state.builder.AddFunction({OpVectorShuffle, state.vec4_float_type, mapped, value, value,
-	                            mapping.Map(0), mapping.Map(1), mapping.Map(2), mapping.Map(3)});
+	state.builder.AddFunction({OpVectorShuffle, vector_type, mapped, value, value, mapping.Map(0),
+	                           mapping.Map(1), mapping.Map(2), mapping.Map(3)});
 	return mapped;
 }
 
@@ -114,11 +166,15 @@ void EmitExport(EmitterState& state, const IR::Instruction& inst) {
 		return;
 	}
 
-	const auto value = ApplyMrtExportMapping(state, inst, EmitExportVec4F32(state, inst));
+	const auto uint_output = MrtUsesUintOutput(state, inst);
+	const auto vector_type = uint_output ? state.vec4_uint_type : state.vec4_float_type;
+	const auto value       = ApplyMrtExportMapping(
+	    state, inst, uint_output ? EmitExportVec4U32(state, inst) : EmitExportVec4F32(state, inst),
+	    vector_type);
 	if (inst.export_info.kind == IR::ExportTargetKind::Position) {
 		const auto pointer = state.builder.AllocateId();
-		state.builder.AddFunction({OpAccessChain, state.ptr_output_vec4_float, pointer, variable,
-		                            ConstantU32(state, 0)});
+		state.builder.AddFunction(
+		    {OpAccessChain, state.ptr_output_vec4_float, pointer, variable, ConstantU32(state, 0)});
 		state.builder.AddFunction({OpStore, pointer, value});
 		return;
 	}

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

@@ -150,8 +150,8 @@ void EmitWqmB64(EmitterState& state, const IR::Instruction& inst) {
 		EmitPerInvocationMask(state, inst.dst, active);
 	} else {
 		const auto result = state.builder.AllocateId();
-		state.builder.AddFunction({OpSelect, state.uint_type, result, active,
-		                            ConstantU32(state, 1), ConstantU32(state, 0)});
+		state.builder.AddFunction({OpSelect, state.uint_type, result, active, ConstantU32(state, 1),
+		                           ConstantU32(state, 0)});
 		EmitStoreU32(state, inst.dst, result);
 		EmitStoreU32(state, OffsetRegisterOperand(inst.dst, 1), ConstantU32(state, 0));
 	}
@@ -205,8 +205,7 @@ void EmitSaveexecB32(EmitterState& state, const IR::Instruction& inst) {
 
 	const auto cond = state.builder.AllocateId();
 	const auto scc  = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpINotEqual, state.bool_type, cond, new_low, ConstantU32(state, 0)});
+	state.builder.AddFunction({OpINotEqual, state.bool_type, cond, new_low, ConstantU32(state, 0)});
 	state.builder.AddFunction(
 	    {OpSelect, state.uint_type, scc, cond, ConstantU32(state, 1), ConstantU32(state, 0)});
 	EmitStoreU32(state, SccOperand(), scc);
@@ -279,8 +278,9 @@ void EmitReadFirstLaneU32(EmitterState& state, const IR::Instruction& inst) {
 
 uint32_t EmitLaneIndex(EmitterState& state, const IR::Operand& operand) {
 	const auto lane = state.builder.AllocateId();
+	const auto mask = state.wave_size == 32u ? 31u : 63u;
 	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, lane, EmitValueLoad(state, operand),
-	                            ConstantU32(state, 63)});
+	                           ConstantU32(state, mask)});
 	return lane;
 }
 
@@ -336,10 +336,8 @@ void EmitPermlaneB32(EmitterState& state, const IR::Instruction& inst, bool x16)
 		state.builder.AddFunction(
 		    {OpBitwiseXor, state.uint_type, row_value, row, ConstantU32(state, 16)});
 	}
-	state.builder.AddFunction(
-	    {OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 15)});
-	state.builder.AddFunction(
-	    {OpBitwiseAnd, state.uint_type, lane8, lane, ConstantU32(state, 7)});
+	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 15)});
+	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, lane8, lane, ConstantU32(state, 7)});
 	state.builder.AddFunction(
 	    {OpShiftLeftLogical, state.uint_type, shift, lane8, ConstantU32(state, 2)});
 	state.builder.AddFunction(

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

@@ -2,6 +2,66 @@
 #include "graphics/shader/recompiler/emitter/spirvEmitterInternal.h"
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+namespace {
+
+uint32_t EmitCubeAxisF32(EmitterState& state, uint32_t value) {
+	const auto normalized = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpFSub, state.float_type, normalized, value, ConstantF32(state, 0x3f800000u)});
+	return normalized;
+}
+
+uint32_t EmitCubeLayerF32(EmitterState& state, uint32_t face_id) {
+	// Sampled RDNA2 cubemaps encode face_id as slice * 8 + face. The native
+	// 2D-array view stores six contiguous faces per slice, so remove the two
+	// reserved face IDs from every preceding slice.
+	const auto guest_layer = state.builder.AllocateId();
+	const auto slice       = state.builder.AllocateId();
+	const auto padding     = state.builder.AllocateId();
+	const auto host_layer  = state.builder.AllocateId();
+	const auto result      = state.builder.AllocateId();
+	state.builder.AddFunction({OpConvertFToU, state.uint_type, guest_layer, face_id});
+	state.builder.AddFunction(
+	    {OpShiftRightLogical, state.uint_type, slice, guest_layer, ConstantU32(state, 3)});
+	state.builder.AddFunction(
+	    {OpShiftLeftLogical, state.uint_type, padding, slice, ConstantU32(state, 1)});
+	state.builder.AddFunction({OpISub, state.uint_type, host_layer, guest_layer, padding});
+	state.builder.AddFunction({OpConvertUToF, state.float_type, result, host_layer});
+	return result;
+}
+
+uint32_t EmitImageCoordF32Impl(EmitterState& state, const IR::Instruction& inst,
+                               const IR::Operand& address, uint32_t first_component,
+                               uint32_t components) {
+	auto x = EmitImageAddressFloatLoad(state, inst, address, first_component);
+	if (components == 1u) {
+		return x;
+	}
+	auto y = inst.memory.image_address_components > first_component + 1u
+	             ? EmitImageAddressFloatLoad(state, inst, address, first_component + 1u)
+	             : EmitZeroF32(state);
+	if (inst.memory.image_cube) {
+		// RDNA2 sampled cubemap S/T coordinates are biased by +1 relative to
+		// normalized 2D-array coordinates.
+		x = EmitCubeAxisF32(state, x);
+		y = EmitCubeAxisF32(state, y);
+	}
+	const auto coord = state.builder.AllocateId();
+	if (components == 3u) {
+		auto z = inst.memory.image_address_components > first_component + 2u
+		             ? EmitImageAddressFloatLoad(state, inst, address, first_component + 2u)
+		             : EmitZeroF32(state);
+		if (inst.memory.image_cube) {
+			z = EmitCubeLayerF32(state, z);
+		}
+		state.builder.AddFunction({OpCompositeConstruct, state.vec3_float_type, coord, x, y, z});
+	} else {
+		state.builder.AddFunction({OpCompositeConstruct, state.vec2_float_type, coord, x, y});
+	}
+	return coord;
+}
+
+} // namespace
 
 bool HasImageSampleFlag(const IR::Instruction& inst, uint32_t flag) {
 	return (inst.memory.image_sample_flags & flag) != 0;
@@ -36,24 +96,8 @@ ImageSampleLayout MakeImageSampleLayout(const IR::Instruction& inst, ImageViewKi
 
 uint32_t EmitImageCoordF32(EmitterState& state, const IR::Instruction& inst,
                            const ImageSampleLayout& layout, ImageViewKind view) {
-	const auto x          = EmitImageAddressFloatLoad(state, inst, inst.src[0], layout.coord);
-	const auto components = ImageViewCoordinateComponents(view);
-	if (components == 1u) {
-		return x;
-	}
-	const auto y = inst.memory.image_address_components > layout.coord + 1u
-	                       ? EmitImageAddressFloatLoad(state, inst, inst.src[0], layout.coord + 1u)
-	                       : EmitZeroF32(state);
-	const auto coord = state.builder.AllocateId();
-	if (components == 3u) {
-		const auto z = inst.memory.image_address_components > layout.coord + 2u
-		                   ? EmitImageAddressFloatLoad(state, inst, inst.src[0], layout.coord + 2u)
-		                   : EmitZeroF32(state);
-		state.builder.AddFunction({OpCompositeConstruct, state.vec3_float_type, coord, x, y, z});
-	} else {
-		state.builder.AddFunction({OpCompositeConstruct, state.vec2_float_type, coord, x, y});
-	}
-	return coord;
+	return EmitImageCoordF32Impl(state, inst, inst.src[0], layout.coord,
+	                             ImageViewCoordinateComponents(view));
 }
 
 uint32_t EmitImageLodF32(EmitterState& state, const IR::Instruction& inst,
@@ -95,9 +139,9 @@ uint32_t EmitImageGradientF32(EmitterState& state, const IR::Instruction& inst,
 	                   : EmitZeroF32(state);
 	const auto grad = state.builder.AllocateId();
 	if (components == 3u) {
-		const auto z = inst.memory.image_address_components > first_component + 2u
-		                   ? EmitImageAddressFloatLoad(state, inst, inst.src[0],
-		                                               first_component + 2u)
+		const auto z =
+		    inst.memory.image_address_components > first_component + 2u
+		        ? EmitImageAddressFloatLoad(state, inst, inst.src[0], first_component + 2u)
 		        : EmitZeroF32(state);
 		state.builder.AddFunction({OpCompositeConstruct, state.vec3_float_type, grad, x, y, z});
 	} else {
@@ -120,8 +164,7 @@ uint32_t EmitImagePackedOffsetI32(EmitterState& state, const IR::Instruction& in
 			state.builder.AddFunction(
 			    {OpCompositeConstruct, state.vec3_int_type, ret, zero, zero, zero});
 		} else {
-			state.builder.AddFunction(
-			    {OpCompositeConstruct, state.vec2_int_type, ret, zero, zero});
+			state.builder.AddFunction({OpCompositeConstruct, state.vec2_int_type, ret, zero, zero});
 		}
 		return ret;
 	}
@@ -209,24 +252,8 @@ uint32_t EmitImageMipLodU32(EmitterState& state, const IR::Instruction& inst,
 
 uint32_t EmitImageQueryCoordF32(EmitterState& state, const IR::Instruction& inst,
                                 ImageViewKind view) {
-	const auto x          = EmitImageAddressFloatLoad(state, inst, inst.src[0], 0);
-	const auto components = ImageViewCoordinateComponents(view);
-	if (components == 1u) {
-		return x;
-	}
-	const auto y = inst.memory.image_address_components > 1u
-	                       ? EmitImageAddressFloatLoad(state, inst, inst.src[0], 1)
-	                       : EmitZeroF32(state);
-	const auto coord = state.builder.AllocateId();
-	if (components == 3u) {
-		const auto z = inst.memory.image_address_components > 2u
-		                   ? EmitImageAddressFloatLoad(state, inst, inst.src[0], 2)
-		                   : EmitZeroF32(state);
-		state.builder.AddFunction({OpCompositeConstruct, state.vec3_float_type, coord, x, y, z});
-	} else {
-		state.builder.AddFunction({OpCompositeConstruct, state.vec2_float_type, coord, x, y});
-	}
-	return coord;
+	// OpImageQueryLod takes only the spatial coordinates, even for arrayed images.
+	return EmitImageCoordF32Impl(state, inst, inst.src[0], 0, ImageViewSpatialComponents(view));
 }
 
 uint32_t DmaskComponentIndex(uint32_t dmask, uint32_t component) {

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

@@ -133,10 +133,18 @@ void EmitImageLoad(EmitterState& state, const IR::Instruction& inst) {
 	const bool integer = inst.memory.kind == IR::ResourceKind::ImageUint;
 
 	const auto color = state.builder.AllocateId();
-	state.builder.AddFunction({OpImageFetch, integer ? state.vec4_uint_type : state.vec4_float_type,
-	                           color, image, EmitImageLoadCoordU32(state, inst, view),
-	                           ImageOperandsLodMask,
-	                           EmitImageMipLodU32(state, inst, inst.src[0], view)});
+	const auto coord = EmitImageLoadCoordU32(state, inst, view);
+	if (ImageSpirvMultisampled(view) != 0) {
+		const auto sample = EmitImageAddressValueLoad(state, inst, inst.src[0],
+		                                              ImageViewCoordinateComponents(view));
+		state.builder.AddFunction({OpImageFetch,
+		                           integer ? state.vec4_uint_type : state.vec4_float_type, color,
+		                           image, coord, ImageOperandsSampleMask, sample});
+	} else {
+		state.builder.AddFunction(
+		    {OpImageFetch, integer ? state.vec4_uint_type : state.vec4_float_type, color, image,
+		     coord, ImageOperandsLodMask, EmitImageMipLodU32(state, inst, inst.src[0], view)});
+	}
 
 	const auto dmask     = inst.memory.dmask != 0 ? inst.memory.dmask : 1u;
 	uint32_t   dst_index = 0;
@@ -158,8 +166,8 @@ void EmitImageLoad(EmitterState& state, const IR::Instruction& inst) {
 void EmitImageStore(EmitterState& state, const IR::Instruction& inst) {
 	const auto uint_image = inst.memory.kind == IR::ResourceKind::StorageImageUint;
 	const auto view       = StorageImageViewKind(state, inst.memory, uint_image, inst.pc);
-	const auto binding    = ResourceForDescriptor(state, StorageBindingKind(uint_image, view),
-	                                              inst.memory.resource);
+	const auto binding =
+	    ResourceForDescriptor(state, StorageBindingKind(uint_image, view), inst.memory.resource);
 	const auto image = LoadStorageImageDescriptorAtIndex(state, inst.memory.resource,
 	                                                     binding.array_index, uint_image, view);
 

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

@@ -3,11 +3,11 @@
 
 #include "common/common.h"
 #include "common/stringUtils.h"
-#include "graphics/shader/recompiler/ir/BindingLayout.h"
 #include "graphics/shader/recompiler/BufferFormat.h"
+#include "graphics/shader/recompiler/emitter/SpirvBuilder.h"
+#include "graphics/shader/recompiler/ir/BindingLayout.h"
 #include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/recompiler/emitter/SpirvBuilder.h"
 
 #include <algorithm>
 #include <array>
@@ -99,6 +99,7 @@ enum : uint32_t {
 	ImageOperandsGradMask         = 0x00000004u,
 	ImageOperandsOffsetMask       = 0x00000010u,
 	ImageOperandsConstOffsetsMask = 0x00000020u,
+	ImageOperandsSampleMask       = 0x00000040u,
 };
 
 enum : uint32_t {
@@ -150,7 +151,6 @@ enum : uint32_t {
 	OpImageGather                  = 96,
 	OpImageDrefGather              = 97,
 	OpImageWrite                   = 99,
-	OpImage                        = 100,
 	OpImageQuerySizeLod            = 103,
 	OpImageQueryLod                = 105,
 	OpImageQueryLevels             = 106,
@@ -382,7 +382,7 @@ struct EmitterState {
 	uint32_t                                         ptr_workgroup_array       = 0;
 	uint32_t                                         ptr_workgroup_uint        = 0;
 	uint32_t                                         lds_variable              = 0;
-	std::array<SampledImageDescriptors, 10> sampled_images;
+	std::array<SampledImageDescriptors, 14>          sampled_images;
 	std::array<StorageImageDescriptors, 10>          storage_images;
 	uint32_t                                         sampler_type                          = 0;
 	uint32_t                                         sampler_array_type                    = 0;
@@ -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;
@@ -453,17 +450,20 @@ enum class ImageViewKind {
 	Dim2D,
 	Dim2DArray,
 	Dim3D,
+	Dim2DMsaa,
+	Dim2DMsaaArray,
 	Count,
 };
 
-constexpr uint32_t ImageViewKindCount = static_cast<uint32_t>(ImageViewKind::Count);
+constexpr uint32_t SampledImageViewKindCount = static_cast<uint32_t>(ImageViewKind::Count);
+constexpr uint32_t StorageImageViewKindCount = static_cast<uint32_t>(ImageViewKind::Dim2DMsaa);
 
 constexpr uint32_t SampledImageIndex(bool integer, ImageViewKind view) {
-	return static_cast<uint32_t>(view) + (integer ? ImageViewKindCount : 0u);
+	return static_cast<uint32_t>(view) + (integer ? SampledImageViewKindCount : 0u);
 }
 
 constexpr uint32_t StorageImageIndex(bool integer, ImageViewKind view) {
-	return static_cast<uint32_t>(view) + (integer ? ImageViewKindCount : 0u);
+	return static_cast<uint32_t>(view) + (integer ? StorageImageViewKindCount : 0u);
 }
 
 constexpr IR::DescriptorBindingKind SampledBindingKind(bool integer, ImageViewKind view) {
@@ -474,6 +474,9 @@ constexpr IR::DescriptorBindingKind SampledBindingKind(bool integer, ImageViewKi
 			case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::SampledUint2D;
 			case ImageViewKind::Dim2DArray: return IR::DescriptorBindingKind::SampledUint2DArray;
 			case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::SampledUint3D;
+			case ImageViewKind::Dim2DMsaa: return IR::DescriptorBindingKind::SampledUint2DMsaa;
+			case ImageViewKind::Dim2DMsaaArray:
+				return IR::DescriptorBindingKind::SampledUint2DMsaaArray;
 			default: break;
 		}
 	}
@@ -483,6 +486,8 @@ constexpr IR::DescriptorBindingKind SampledBindingKind(bool integer, ImageViewKi
 		case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::Sampled2D;
 		case ImageViewKind::Dim2DArray: return IR::DescriptorBindingKind::Sampled2DArray;
 		case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::Sampled3D;
+		case ImageViewKind::Dim2DMsaa: return IR::DescriptorBindingKind::Sampled2DMsaa;
+		case ImageViewKind::Dim2DMsaaArray: return IR::DescriptorBindingKind::Sampled2DMsaaArray;
 		default: break;
 	}
 	return IR::DescriptorBindingKind::Count;
@@ -516,6 +521,8 @@ constexpr uint32_t ImageSpirvDimension(ImageViewKind view) {
 		case ImageViewKind::Dim1DArray: return Dim1D;
 		case ImageViewKind::Dim2D:
 		case ImageViewKind::Dim2DArray:
+		case ImageViewKind::Dim2DMsaa:
+		case ImageViewKind::Dim2DMsaaArray:
 		case ImageViewKind::Count: return Dim2D;
 		case ImageViewKind::Dim3D: return Dim3D;
 	}
@@ -523,7 +530,14 @@ constexpr uint32_t ImageSpirvDimension(ImageViewKind view) {
 }
 
 constexpr uint32_t ImageSpirvArrayed(ImageViewKind view) {
-	return view == ImageViewKind::Dim1DArray || view == ImageViewKind::Dim2DArray ? 1u : 0u;
+	return view == ImageViewKind::Dim1DArray || view == ImageViewKind::Dim2DArray ||
+	               view == ImageViewKind::Dim2DMsaaArray
+	           ? 1u
+	           : 0u;
+}
+
+constexpr uint32_t ImageSpirvMultisampled(ImageViewKind view) {
+	return view == ImageViewKind::Dim2DMsaa || view == ImageViewKind::Dim2DMsaaArray ? 1u : 0u;
 }
 
 struct AddCarryResult {
@@ -990,11 +1004,6 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 uint32_t UnpackTBufferFormat(EmitterState& state, const IR::Instruction& inst,
                              const Format::BufferFormatInfo& info);
 
-bool EmitTypedTBufferLoad(EmitterState& state, const IR::Instruction& inst,
-                          const Format::BufferFormatInfo& info);
-
-bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst);
-
 uint32_t FormattedBufferDwordStoreComponentCount(Prospero::BufferFormat format,
                                                  uint32_t               opcode_components);
 
@@ -1020,6 +1029,9 @@ void EmitBufferLoadSshort(EmitterState& state, const IR::Instruction& inst);
 
 void EmitBufferLoadDword(EmitterState& state, const IR::Instruction& inst);
 
+void EmitBufferLoadDwordGroup(EmitterState& state, const IR::Instruction* instructions,
+                              uint32_t count);
+
 void EmitBufferStoreDword(EmitterState& state, const IR::Instruction& inst);
 
 void EmitFlatLoadUbyte(EmitterState& state, const IR::Instruction& inst);

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

@@ -34,16 +34,15 @@ uint32_t EmitDppWriteActiveBool(EmitterState& state, const IR::Operand& dst) {
 	    {OpShiftLeftLogical, state.uint_type, bank_bit, ConstantU32(state, 1), bank});
 	state.builder.AddFunction(
 	    {OpShiftLeftLogical, state.uint_type, row_bit, ConstantU32(state, 1), row});
-	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, bank_hit,
-	                            ConstantU32(state, dst.dpp_bank_mask), bank_bit});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, state.uint_type, bank_hit, ConstantU32(state, dst.dpp_bank_mask), bank_bit});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, row_hit, ConstantU32(state, dst.dpp_row_mask), row_bit});
 	state.builder.AddFunction(
 	    {OpINotEqual, state.bool_type, bank_active, bank_hit, ConstantU32(state, 0)});
 	state.builder.AddFunction(
 	    {OpINotEqual, state.bool_type, row_active, row_hit, ConstantU32(state, 0)});
-	state.builder.AddFunction(
-	    {OpLogicalAnd, state.bool_type, dpp_active, bank_active, row_active});
+	state.builder.AddFunction({OpLogicalAnd, state.bool_type, dpp_active, bank_active, row_active});
 	uint32_t write_active = dpp_active;
 	if (!dst.dpp_bound_ctrl) {
 		const auto target  = EmitDppTargetLane(state, dst.dpp_ctrl);
@@ -131,8 +130,7 @@ uint32_t EmitNotEqualZeroBool(EmitterState& state, uint32_t value) {
 uint32_t EmitSelectU32Value(EmitterState& state, uint32_t condition, uint32_t true_value,
                             uint32_t false_value) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpSelect, state.uint_type, ret, condition, true_value, false_value});
+	state.builder.AddFunction({OpSelect, state.uint_type, ret, condition, true_value, false_value});
 	return ret;
 }
 
@@ -212,12 +210,12 @@ bool IsStorageBufferMemoryKind(IR::ResourceKind kind) {
 
 void EmitStorageBufferOffsets(EmitterState& state) {
 	for (uint32_t i = 0; i < state.program.bindings.buffer_offset_count; i++) {
-		const auto word = EmitShaderDataDwordLoad(
-		    state, state.program.bindings.buffer_offset_dword + i / 4u);
+		const auto word =
+		    EmitShaderDataDwordLoad(state, state.program.bindings.buffer_offset_dword + i / 4u);
 		const auto shift                = ConstantU32(state, (i % 4u) * 8u + 2u);
 		state.storage_buffer_offsets[i] = EmitBinaryU32(
-		    state, OpBitwiseAnd,
-		    EmitBinaryU32(state, OpShiftRightLogical, word, shift), ConstantU32(state, 0x3fu));
+		    state, OpBitwiseAnd, EmitBinaryU32(state, OpShiftRightLogical, word, shift),
+		    ConstantU32(state, 0x3fu));
 	}
 }
 
@@ -329,8 +327,7 @@ uint32_t EmitRelativeAddress(EmitterState& state, const IR::Instruction& inst, u
 
 uint32_t EmitFlatVirtualAddress(EmitterState& state, const IR::Instruction& inst,
                                 uint32_t first_src, uint32_t src_count) {
-	if (inst.memory.resource >= state.resources.addresses.size() ||
-	    src_count < 2) {
+	if (inst.memory.resource >= state.resources.addresses.size() || src_count < 2) {
 		ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::AddressMemory,
 		                             inst.memory.resource, "flat address snapshot is missing");
 	}
@@ -659,8 +656,7 @@ void EmitAtomicUpdateU32(EmitterState& state, uint32_t pointer, IR::ResourceKind
 	state.builder.AddFunction({OpBranch, preheader});
 	state.builder.AddFunction({OpLabel, preheader});
 	state.builder.AddFunction({OpAtomicLoad, state.uint_type, initial, pointer,
-	                            ConstantU32(state, scope),
-	                            ConstantU32(state, MemorySemanticsNone)});
+	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone)});
 	state.builder.AddFunction({OpBranch, header});
 	state.builder.AddFunction({OpLabel, header});
 	state.builder.AddFunction(
@@ -793,8 +789,7 @@ uint32_t EmitTBufferBitcastU32ToI32(EmitterState& state, uint32_t value) {
 uint32_t EmitTBufferCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
                                        uint32_t constant) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {opcode, state.bool_type, ret, value, ConstantU32(state, constant)});
+	state.builder.AddFunction({opcode, state.bool_type, ret, value, ConstantU32(state, constant)});
 	return ret;
 }
 
@@ -961,18 +956,8 @@ uint32_t UnpackTBufferFormat(EmitterState& state, const IR::Instruction& inst,
 	return NormalizeFormatComponent(state, info, inst.memory.component_index, raw);
 }
 
-bool EmitTypedTBufferLoad(EmitterState& state, const IR::Instruction& inst,
-                          const Format::BufferFormatInfo& info) {
-	if (!Format::CanUseTypedBufferLoad(info.format)) {
-		return false;
-	}
-	const auto value = EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
-	                                               AddressSourceCount(inst, 0));
-	EmitStoreU32(state, inst.dst, value);
-	return true;
-}
-
-bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst) {
+bool EmitFormattedBufferLoadValueU32(EmitterState& state, const IR::Instruction& inst,
+                                     uint32_t& value) {
 	if (!IsFormattedBufferComponent(inst)) {
 		return false;
 	}
@@ -984,18 +969,29 @@ bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst) {
 
 	const auto info = Format::GetFormatInfo(format);
 	if (inst.memory.component_index >= info.component_count) {
-		EmitStoreU32(state, inst.dst, ConstantU32(state, 0));
+		value = ConstantU32(state, 0);
 		return true;
 	}
 
-	if (EmitTypedTBufferLoad(state, inst, info)) {
+	if (Format::CanUseTypedBufferLoad(info.format)) {
+		value = EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
+		                                    AddressSourceCount(inst, 0));
 		return true;
 	}
 
-	EmitStoreU32(state, inst.dst, UnpackTBufferFormat(state, inst, info));
+	value = UnpackTBufferFormat(state, inst, info);
 	return true;
 }
 
+uint32_t EmitBufferLoadDwordValueU32(EmitterState& state, const IR::Instruction& inst) {
+	uint32_t value = 0;
+	if (EmitFormattedBufferLoadValueU32(state, inst, value)) {
+		return value;
+	}
+	return EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
+	                                   AddressSourceCount(inst, 0));
+}
+
 uint32_t FormattedBufferDwordStoreComponentCount(Prospero::BufferFormat format,
                                                  uint32_t               opcode_components) {
 	switch (format) {
@@ -1262,11 +1258,27 @@ void EmitBufferLoadSshort(EmitterState& state, const IR::Instruction& inst) {
 }
 
 void EmitBufferLoadDword(EmitterState& state, const IR::Instruction& inst) {
-	EmitGuardedByExec(state, [&]() {
-		if (EmitFormattedBufferLoad(state, inst)) {
+	EmitGuardedByExec(
+	    state, [&]() { EmitStoreU32(state, inst.dst, EmitBufferLoadDwordValueU32(state, inst)); });
+}
+
+void EmitBufferLoadDwordGroup(EmitterState& state, const IR::Instruction* instructions,
+                              uint32_t count) {
+	if (instructions == nullptr || count == 0u) {
 		return;
 	}
-		EmitMemoryLoadU32(state, inst, IR::ResourceKind::Buffer, 0, AddressSourceCount(inst, 0));
+
+	EmitGuardedByExec(state, [&]() {
+		// RDNA VMEM captures every VADDR component before making overlapping VDATA writes
+		// visible. Keep the split IR components instruction-atomic by deferring all stores.
+		std::vector<uint32_t> values;
+		values.reserve(count);
+		for (uint32_t i = 0; i < count; i++) {
+			values.push_back(EmitBufferLoadDwordValueU32(state, instructions[i]));
+		}
+		for (uint32_t i = 0; i < count; i++) {
+			EmitStoreU32(state, instructions[i].dst, values[i]);
+		}
 	});
 }
 
@@ -1482,8 +1494,7 @@ void EmitDsAppendConsume(EmitterState& state, const IR::Instruction& inst, uint3
 	const auto do_atomic  = state.builder.AllocateId();
 	state.builder.AddFunction({OpIEqual, state.bool_type, first_lane, subid, exec.first_lane});
 	const auto first_active = EmitLogicalAndBool(state, first_lane, exec.any_active);
-	state.builder.AddFunction(
-	    {OpLogicalAnd, state.bool_type, do_atomic, first_active, in_bounds});
+	state.builder.AddFunction({OpLogicalAnd, state.bool_type, do_atomic, first_active, in_bounds});
 
 	const auto atomic_value = EmitValueOrZeroIfCondition(state, do_atomic, [&]() {
 		const auto pointer = gds ? EmitGdsElementPointer(state, address.index)
@@ -1575,14 +1586,13 @@ uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t c
 	const auto xored  = state.builder.AllocateId();
 	const auto base   = state.builder.AllocateId();
 	const auto target = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 31)});
+	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 31)});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, masked, lane, ConstantU32(state, control & 0x1fu)});
 	state.builder.AddFunction(
 	    {OpBitwiseOr, state.uint_type, ored, masked, ConstantU32(state, (control >> 5u) & 0x1fu)});
-	state.builder.AddFunction({OpBitwiseXor, state.uint_type, xored, ored,
-	                            ConstantU32(state, (control >> 10u) & 0x1fu)});
+	state.builder.AddFunction(
+	    {OpBitwiseXor, state.uint_type, xored, ored, ConstantU32(state, (control >> 10u) & 0x1fu)});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, base, subid, ConstantU32(state, 0xffffffe0u)});
 	state.builder.AddFunction({OpBitwiseOr, state.uint_type, target, base, xored});

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

@@ -174,6 +174,12 @@ uint32_t VertexParameterInputPointerType(const EmitterState& state, VertexInputS
 	}
 }
 
+static bool MrtUsesUintOutput(const EmitterState& state, uint32_t index) {
+	return state.stage == ShaderType::Pixel && state.pixel_input_info != nullptr &&
+	       index < std::size(state.pixel_input_info->target_output_mode) &&
+	       state.pixel_input_info->target_output_mode[index] == 7u;
+}
+
 void AllocateInputVariables(EmitterState& state) {
 	for (auto& binding: state.inputs) {
 		binding.variable_id = state.builder.AllocateId();
@@ -323,23 +329,39 @@ void AddDescriptorAnnotationsAndNames(EmitterState& state) {
 		Decorate(state.address_memory_variable, "address_memory",
 		         IR::DescriptorBindingKind::AddressMemory);
 	}
-	constexpr const char* SampledNames[] = {
-	    "sampled_1d",      "sampled_1d_array",      "sampled_2d",      "sampled_2d_array",
-	    "sampled_3d",      "sampled_uint_1d",       "sampled_uint_1d_array",
-	    "sampled_uint_2d", "sampled_uint_2d_array", "sampled_uint_3d"};
+	constexpr const char* SampledNames[] = {"sampled_1d",
+	                                        "sampled_1d_array",
+	                                        "sampled_2d",
+	                                        "sampled_2d_array",
+	                                        "sampled_3d",
+	                                        "sampled_2d_msaa",
+	                                        "sampled_2d_msaa_array",
+	                                        "sampled_uint_1d",
+	                                        "sampled_uint_1d_array",
+	                                        "sampled_uint_2d",
+	                                        "sampled_uint_2d_array",
+	                                        "sampled_uint_3d",
+	                                        "sampled_uint_2d_msaa",
+	                                        "sampled_uint_2d_msaa_array"};
 	for (uint32_t i = 0; i < state.sampled_images.size(); i++) {
-		const auto view = static_cast<ImageViewKind>(i % ImageViewKindCount);
+		const auto view = static_cast<ImageViewKind>(i % SampledImageViewKindCount);
 		Decorate(state.sampled_images[i].variable, SampledNames[i],
-		         SampledBindingKind(i >= ImageViewKindCount, view));
+		         SampledBindingKind(i >= SampledImageViewKindCount, view));
 	}
-	constexpr const char* StorageNames[] = {
-	    "storage_1d",      "storage_1d_array",      "storage_2d",      "storage_2d_array",
-	    "storage_3d",      "storage_uint_1d",       "storage_uint_1d_array",
-	    "storage_uint_2d", "storage_uint_2d_array", "storage_uint_3d"};
+	constexpr const char* StorageNames[] = {"storage_1d",
+	                                        "storage_1d_array",
+	                                        "storage_2d",
+	                                        "storage_2d_array",
+	                                        "storage_3d",
+	                                        "storage_uint_1d",
+	                                        "storage_uint_1d_array",
+	                                        "storage_uint_2d",
+	                                        "storage_uint_2d_array",
+	                                        "storage_uint_3d"};
 	for (uint32_t i = 0; i < state.storage_images.size(); i++) {
-		const auto view = static_cast<ImageViewKind>(i % ImageViewKindCount);
+		const auto view = static_cast<ImageViewKind>(i % StorageImageViewKindCount);
 		Decorate(state.storage_images[i].variable, StorageNames[i],
-		         StorageBindingKind(i >= ImageViewKindCount, view));
+		         StorageBindingKind(i >= StorageImageViewKindCount, view));
 	}
 	if (state.sampler_variable != 0) {
 		Decorate(state.sampler_variable, "samplers", IR::DescriptorBindingKind::Samplers);
@@ -409,6 +431,7 @@ void EmitHeaderAndTypes(EmitterState& state) {
 	state.ptr_output_sample_mask_array = state.builder.AllocateId();
 	state.ptr_output_float             = state.builder.AllocateId();
 	state.ptr_output_vec4_float        = state.builder.AllocateId();
+	const auto ptr_output_vec4_uint    = state.builder.AllocateId();
 	state.per_vertex_type              = state.builder.AllocateId();
 	state.ptr_output_per_vertex        = state.builder.AllocateId();
 	state.storage_runtime_array_type   = state.builder.AllocateId();
@@ -462,7 +485,7 @@ void EmitHeaderAndTypes(EmitterState& state) {
 		state.builder.AddCapability({CapabilityImageGatherExtended});
 	}
 	if (std::any_of(state.storage_images.begin(),
-	                state.storage_images.begin() + ImageViewKindCount,
+	                state.storage_images.begin() + StorageImageViewKindCount,
 	                [](const auto& image) { return image.variable != 0; })) {
 		state.builder.AddCapability({CapabilityStorageImageReadWithoutFormat});
 		state.builder.AddCapability({CapabilityStorageImageWriteWithoutFormat});
@@ -605,6 +628,8 @@ void EmitHeaderAndTypes(EmitterState& state) {
 	    {OpTypePointer, state.ptr_output_int, StorageClassOutput, state.int_type});
 	state.builder.AddType(
 	    {OpTypePointer, state.ptr_output_vec4_float, StorageClassOutput, state.vec4_float_type});
+	state.builder.AddType(
+	    {OpTypePointer, ptr_output_vec4_uint, StorageClassOutput, state.vec4_uint_type});
 	if (state.per_vertex_variable != 0) {
 		state.builder.AddType({OpTypeStruct, state.per_vertex_type, state.vec4_float_type});
 		state.builder.AddType({OpTypePointer, state.ptr_output_per_vertex, StorageClassOutput,
@@ -615,8 +640,12 @@ void EmitHeaderAndTypes(EmitterState& state) {
 	for (const auto& binding: state.outputs) {
 		if (binding.kind == IR::StageOutputKind::Parameter ||
 		    binding.kind == IR::StageOutputKind::Mrt) {
+			const auto pointer_type =
+			    binding.kind == IR::StageOutputKind::Mrt && MrtUsesUintOutput(state, binding.index)
+			        ? ptr_output_vec4_uint
+			        : state.ptr_output_vec4_float;
 			state.builder.AddType(
-			    {OpVariable, state.ptr_output_vec4_float, binding.variable_id, StorageClassOutput});
+			    {OpVariable, pointer_type, binding.variable_id, StorageClassOutput});
 		}
 	}
 	if (state.depth_variable != 0) {
@@ -700,11 +729,11 @@ void EmitHeaderAndTypes(EmitterState& state) {
 	}
 	for (uint32_t i = 0; i < state.sampled_images.size(); i++) {
 		auto&      image     = state.sampled_images[i];
-		const auto view      = static_cast<ImageViewKind>(i % ImageViewKindCount);
-		const bool integer   = i >= ImageViewKindCount;
+		const auto view      = static_cast<ImageViewKind>(i % SampledImageViewKindCount);
+		const bool integer   = i >= SampledImageViewKindCount;
 		const auto component = integer ? state.uint_type : state.float_type;
-		state.builder.AddType({OpTypeImage, image.image_type, component,
-		                       ImageSpirvDimension(view), 0, ImageSpirvArrayed(view), 0, 1,
+		state.builder.AddType({OpTypeImage, image.image_type, component, ImageSpirvDimension(view),
+		                       0, ImageSpirvArrayed(view), ImageSpirvMultisampled(view), 1,
 		                       ImageFormatUnknown});
 		state.builder.AddType({OpTypeSampledImage, image.sampled_image_type, image.image_type});
 		state.builder.AddType(
@@ -733,13 +762,12 @@ void EmitHeaderAndTypes(EmitterState& state) {
 	}
 	for (uint32_t i = 0; i < state.storage_images.size(); i++) {
 		auto&      image     = state.storage_images[i];
-		const auto view      = static_cast<ImageViewKind>(i % ImageViewKindCount);
-		const bool integer   = i >= ImageViewKindCount;
+		const auto view      = static_cast<ImageViewKind>(i % StorageImageViewKindCount);
+		const bool integer   = i >= StorageImageViewKindCount;
 		const auto component = integer ? state.uint_type : state.float_type;
 		const auto format    = integer ? ImageFormatR32ui : ImageFormatUnknown;
-		state.builder.AddType({OpTypeImage, image.image_type, component,
-		                       ImageSpirvDimension(view), 0, ImageSpirvArrayed(view), 0, 2,
-		                       format});
+		state.builder.AddType({OpTypeImage, image.image_type, component, ImageSpirvDimension(view),
+		                       0, ImageSpirvArrayed(view), 0, 2, format});
 		state.builder.AddType(
 		    {OpTypePointer, image.pointer_type, StorageClassUniformConstant, image.image_type});
 		if (image.variable != 0) {
@@ -786,15 +814,15 @@ void AllocateDescriptorVariables(EmitterState& state) {
 		state.flattened_srt_variable = state.builder.AllocateId();
 	}
 	for (uint32_t i = 0; i < state.sampled_images.size(); i++) {
-		const auto view = static_cast<ImageViewKind>(i % ImageViewKindCount);
-		if (DescriptorBinding(state, SampledBindingKind(i >= ImageViewKindCount, view)) !=
+		const auto view = static_cast<ImageViewKind>(i % SampledImageViewKindCount);
+		if (DescriptorBinding(state, SampledBindingKind(i >= SampledImageViewKindCount, view)) !=
 		    nullptr) {
 			state.sampled_images[i].variable = state.builder.AllocateId();
 		}
 	}
 	for (uint32_t i = 0; i < state.storage_images.size(); i++) {
-		const auto view = static_cast<ImageViewKind>(i % ImageViewKindCount);
-		if (DescriptorBinding(state, StorageBindingKind(i >= ImageViewKindCount, view)) !=
+		const auto view = static_cast<ImageViewKind>(i % StorageImageViewKindCount);
+		if (DescriptorBinding(state, StorageBindingKind(i >= StorageImageViewKindCount, view)) !=
 		    nullptr) {
 			state.storage_images[i].variable = state.builder.AllocateId();
 		}

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

@@ -13,16 +13,30 @@ namespace {
 constexpr uint32_t MaxPushConstantBytes = 128;
 
 constexpr std::array ImageBindingKinds = {
-    DescriptorBindingKind::Sampled1D,          DescriptorBindingKind::Sampled1DArray,
-    DescriptorBindingKind::Sampled2D,          DescriptorBindingKind::Sampled2DArray,
-    DescriptorBindingKind::Sampled3D,          DescriptorBindingKind::SampledUint1D,
-    DescriptorBindingKind::SampledUint1DArray, DescriptorBindingKind::SampledUint2D,
-    DescriptorBindingKind::SampledUint2DArray, DescriptorBindingKind::SampledUint3D,
-    DescriptorBindingKind::Storage1D,          DescriptorBindingKind::Storage1DArray,
-    DescriptorBindingKind::Storage2D,          DescriptorBindingKind::Storage2DArray,
-    DescriptorBindingKind::Storage3D,          DescriptorBindingKind::StorageUint1D,
-    DescriptorBindingKind::StorageUint1DArray, DescriptorBindingKind::StorageUint2D,
-    DescriptorBindingKind::StorageUint2DArray, DescriptorBindingKind::StorageUint3D,
+    DescriptorBindingKind::Sampled1D,
+    DescriptorBindingKind::Sampled1DArray,
+    DescriptorBindingKind::Sampled2D,
+    DescriptorBindingKind::Sampled2DArray,
+    DescriptorBindingKind::Sampled2DMsaa,
+    DescriptorBindingKind::Sampled2DMsaaArray,
+    DescriptorBindingKind::Sampled3D,
+    DescriptorBindingKind::SampledUint1D,
+    DescriptorBindingKind::SampledUint1DArray,
+    DescriptorBindingKind::SampledUint2D,
+    DescriptorBindingKind::SampledUint2DArray,
+    DescriptorBindingKind::SampledUint2DMsaa,
+    DescriptorBindingKind::SampledUint2DMsaaArray,
+    DescriptorBindingKind::SampledUint3D,
+    DescriptorBindingKind::Storage1D,
+    DescriptorBindingKind::Storage1DArray,
+    DescriptorBindingKind::Storage2D,
+    DescriptorBindingKind::Storage2DArray,
+    DescriptorBindingKind::Storage3D,
+    DescriptorBindingKind::StorageUint1D,
+    DescriptorBindingKind::StorageUint1DArray,
+    DescriptorBindingKind::StorageUint2D,
+    DescriptorBindingKind::StorageUint2DArray,
+    DescriptorBindingKind::StorageUint3D,
 };
 
 bool ImageBinding(const ImageResource& image, DescriptorBindingKind& result) {
@@ -36,6 +50,8 @@ bool ImageBinding(const ImageResource& image, DescriptorBindingKind& result) {
 				case Dimension::Dim1DArray: result = Kind::Sampled1DArray; return true;
 				case Dimension::Dim2D: result = Kind::Sampled2D; return true;
 				case Dimension::Dim2DArray: result = Kind::Sampled2DArray; return true;
+				case Dimension::Dim2DMsaa: result = Kind::Sampled2DMsaa; return true;
+				case Dimension::Dim2DMsaaArray: result = Kind::Sampled2DMsaaArray; return true;
 				case Dimension::Dim3D: result = Kind::Sampled3D; return true;
 				default: return false;
 			}
@@ -45,6 +61,8 @@ bool ImageBinding(const ImageResource& image, DescriptorBindingKind& result) {
 				case Dimension::Dim1DArray: result = Kind::SampledUint1DArray; return true;
 				case Dimension::Dim2D: result = Kind::SampledUint2D; return true;
 				case Dimension::Dim2DArray: result = Kind::SampledUint2DArray; return true;
+				case Dimension::Dim2DMsaa: result = Kind::SampledUint2DMsaa; return true;
+				case Dimension::Dim2DMsaaArray: result = Kind::SampledUint2DMsaaArray; return true;
 				case Dimension::Dim3D: result = Kind::SampledUint3D; return true;
 				default: return false;
 			}
@@ -165,8 +183,7 @@ bool CollectUserData(const Program& program, std::vector<uint32_t>& result) {
 				return false;
 			}
 			for (uint32_t i = 0; i < inst.src_count; i++) {
-				if (!CollectValue(program.provenance, inst.scalar_sources[i], visited,
-				                  registers)) {
+				if (!CollectValue(program.provenance, inst.scalar_sources[i], visited, registers)) {
 					return false;
 				}
 			}

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

@@ -0,0 +1,323 @@
+#include "graphics/shader/recompiler/ir/ReadLaneElimination.h"
+
+#include "graphics/shader/recompiler/ir/SrtWalker.h"
+
+#include <algorithm>
+#include <iterator>
+#include <map>
+#include <set>
+#include <utility>
+
+namespace Libs::Graphics::ShaderRecompiler::IR {
+
+namespace {
+
+constexpr uint32_t FirstTemporaryScalarRegister = 128;
+
+struct LaneKey {
+	uint32_t reg  = 0;
+	uint32_t lane = 0;
+
+	auto operator<=>(const LaneKey&) const = default;
+};
+
+using LaneSet = std::set<LaneKey>;
+
+bool PairDwordOpcode(Opcode op) {
+	switch (op) {
+		case Opcode::MoveU64:
+		case Opcode::WqmB64:
+		case Opcode::SaveexecB64:
+		case Opcode::BitwiseAndU64:
+		case Opcode::BitwiseAndNotU64:
+		case Opcode::BitwiseOrU64:
+		case Opcode::BitwiseOrNotU64:
+		case Opcode::BitwiseXorU64:
+		case Opcode::BitwiseNandU64:
+		case Opcode::BitwiseNorU64:
+		case Opcode::BitwiseXnorU64:
+		case Opcode::BitwiseNotU64:
+		case Opcode::BitFieldMaskU64:
+		case Opcode::BitFieldExtractU64:
+		case Opcode::BitReplicateB64B32:
+		case Opcode::ShiftLeftLogicalU64:
+		case Opcode::ShiftRightLogicalU64:
+		case Opcode::SelectU64: return true;
+		default: return false;
+	}
+}
+
+bool ResolveLane(const Program& program, const Instruction& inst, uint32_t source_index,
+                 uint32_t& lane) {
+	if (source_index >= inst.src_count || (program.wave_size != 32 && program.wave_size != 64)) {
+		return false;
+	}
+	const auto& selector = inst.src[source_index];
+	if (selector.kind == OperandKind::ImmediateU32) {
+		lane = selector.imm % program.wave_size;
+		return true;
+	}
+	uint32_t folded = 0;
+	if (!FoldScalarConstant(program.provenance, inst.scalar_sources[source_index], folded)) {
+		return false;
+	}
+	lane = folded % program.wave_size;
+	return true;
+}
+
+bool UniformWriteSource(const Instruction& inst) {
+	if (inst.src_count == 0) {
+		return false;
+	}
+	const auto& source = inst.src[0];
+	if (source.kind == OperandKind::ImmediateU32 || source.kind == OperandKind::PcRelativeU32) {
+		return true;
+	}
+	return source.kind == OperandKind::Register &&
+	       (source.reg.file == RegisterFile::Scalar || source.reg.file == RegisterFile::Scc ||
+	        source.reg.file == RegisterFile::M0);
+}
+
+bool WriteLaneKey(const Program& program, const Instruction& inst, LaneKey& key) {
+	if (inst.op != Opcode::WriteLaneU32 || inst.dst.kind != OperandKind::Register ||
+	    inst.dst.reg.file != RegisterFile::Vector || !UniformWriteSource(inst)) {
+		return false;
+	}
+	uint32_t lane = 0;
+	if (!ResolveLane(program, inst, 1, lane)) {
+		return false;
+	}
+	key = {inst.dst.reg.index, lane};
+	return true;
+}
+
+bool ReadLaneKey(const Program& program, const Instruction& inst, LaneKey& key) {
+	if (inst.op != Opcode::ReadLaneU32 || inst.src_count < 2 ||
+	    inst.src[0].kind != OperandKind::Register || inst.src[0].reg.file != RegisterFile::Vector) {
+		return false;
+	}
+	uint32_t lane = 0;
+	if (!ResolveLane(program, inst, 1, lane)) {
+		return false;
+	}
+	key = {inst.src[0].reg.index, lane};
+	return true;
+}
+
+void InvalidateRegister(LaneSet& valid, uint32_t reg) {
+	const auto first = valid.lower_bound({reg, 0});
+	const auto last  = valid.lower_bound({reg + 1u, 0});
+	valid.erase(first, last);
+}
+
+void ApplyInstruction(const Program& program, const Instruction& inst, LaneSet& valid) {
+	if (inst.op == Opcode::WriteLaneU32 && inst.dst.kind == OperandKind::Register &&
+	    inst.dst.reg.file == RegisterFile::Vector) {
+		LaneKey key;
+		if (WriteLaneKey(program, inst, key)) {
+			valid.insert(key);
+			return;
+		}
+		uint32_t lane = 0;
+		if (ResolveLane(program, inst, 1, lane)) {
+			valid.erase({inst.dst.reg.index, lane});
+		} else {
+			InvalidateRegister(valid, inst.dst.reg.index);
+		}
+		return;
+	}
+
+	if (inst.op == Opcode::MoveRelDestU32 && inst.dst.kind == OperandKind::Register &&
+	    inst.dst.reg.file == RegisterFile::Vector) {
+		valid.clear();
+		return;
+	}
+
+	if (inst.dst.kind == OperandKind::Register && inst.dst.reg.file == RegisterFile::Vector) {
+		uint32_t dwords = std::max(inst.memory.data_dwords, 1u);
+		if (PairDwordOpcode(inst.op) || inst.op == Opcode::UMadU64U32) {
+			dwords = std::max(dwords, 2u);
+		}
+		for (uint32_t i = 0; i < dwords && inst.dst.reg.index <= UINT32_MAX - i; i++) {
+			InvalidateRegister(valid, inst.dst.reg.index + i);
+		}
+	}
+	if (inst.dst2.kind == OperandKind::Register && inst.dst2.reg.file == RegisterFile::Vector) {
+		InvalidateRegister(valid, inst.dst2.reg.index);
+	}
+}
+
+LaneSet TransferBlock(const Program& program, const BasicBlock& block, LaneSet state) {
+	for (const auto& inst: block.instructions) {
+		ApplyInstruction(program, inst, state);
+	}
+	return state;
+}
+
+LaneSet Intersect(const LaneSet& left, const LaneSet& right) {
+	LaneSet result;
+	std::set_intersection(left.begin(), left.end(), right.begin(), right.end(),
+	                      std::inserter(result, result.end()));
+	return result;
+}
+
+uint32_t NextTemporaryScalarRegister(const Program& program) {
+	uint32_t   next     = FirstTemporaryScalarRegister;
+	const auto consider = [&next](const Operand& operand) {
+		if (operand.kind == OperandKind::Register && operand.reg.file == RegisterFile::Scalar &&
+		    operand.reg.index >= next && operand.reg.index != UINT32_MAX) {
+			next = operand.reg.index + 1u;
+		}
+	};
+	for (const auto& block: program.blocks) {
+		for (const auto& inst: block.instructions) {
+			consider(inst.dst);
+			consider(inst.dst2);
+			for (uint32_t i = 0; i < inst.src_count; i++) {
+				consider(inst.src[i]);
+			}
+		}
+	}
+	return next;
+}
+
+Operand ScalarRegisterOperand(uint32_t reg) {
+	Operand operand;
+	operand.kind      = OperandKind::Register;
+	operand.reg.file  = RegisterFile::Scalar;
+	operand.reg.index = reg;
+	return operand;
+}
+
+Instruction ShadowWrite(const Instruction& write, uint32_t temporary) {
+	Instruction shadow;
+	shadow.pc        = write.pc;
+	shadow.op        = Opcode::MoveU32;
+	shadow.dst       = ScalarRegisterOperand(temporary);
+	shadow.src[0]    = write.src[0];
+	shadow.src_count = 1;
+	return shadow;
+}
+
+Instruction ShadowRead(const Instruction& read, uint32_t temporary) {
+	Instruction rewritten;
+	rewritten.pc        = read.pc;
+	rewritten.op        = Opcode::MoveU32;
+	rewritten.dst       = read.dst;
+	rewritten.src[0]    = ScalarRegisterOperand(temporary);
+	rewritten.src_count = 1;
+	return rewritten;
+}
+
+} // namespace
+
+ReadLaneEliminationStats EliminateReadLane(Program& program) {
+	ReadLaneEliminationStats stats;
+	if (program.blocks.empty() || (program.wave_size != 32 && program.wave_size != 64)) {
+		return stats;
+	}
+
+	LaneSet universe;
+	for (const auto& block: program.blocks) {
+		for (const auto& inst: block.instructions) {
+			LaneKey key;
+			if (WriteLaneKey(program, inst, key)) {
+				universe.insert(key);
+			}
+		}
+	}
+	if (universe.empty()) {
+		return stats;
+	}
+
+	const size_t         block_count = program.blocks.size();
+	std::vector<LaneSet> entry(block_count, universe);
+	std::vector<LaneSet> exit(block_count, universe);
+	entry[0].clear();
+	for (size_t block = 0; block < block_count; block++) {
+		exit[block] = TransferBlock(program, program.blocks[block], entry[block]);
+	}
+
+	bool changed = true;
+	while (changed) {
+		changed = false;
+		for (size_t block_index = 0; block_index < block_count; block_index++) {
+			LaneSet     next_entry;
+			const auto& block = program.blocks[block_index];
+			if (block_index != 0 && !block.predecessors.empty()) {
+				next_entry = universe;
+				for (const auto predecessor: block.predecessors) {
+					if (predecessor >= block_count) {
+						next_entry.clear();
+						break;
+					}
+					next_entry = Intersect(next_entry, exit[predecessor]);
+				}
+			}
+			auto next_exit = TransferBlock(program, block, next_entry);
+			if (next_entry != entry[block_index] || next_exit != exit[block_index]) {
+				entry[block_index] = std::move(next_entry);
+				exit[block_index]  = std::move(next_exit);
+				changed            = true;
+			}
+		}
+	}
+
+	LaneSet forwarded;
+	for (size_t block_index = 0; block_index < block_count; block_index++) {
+		auto state = entry[block_index];
+		for (const auto& inst: program.blocks[block_index].instructions) {
+			LaneKey key;
+			if (ReadLaneKey(program, inst, key) && state.contains(key)) {
+				forwarded.insert(key);
+			}
+			ApplyInstruction(program, inst, state);
+		}
+	}
+	if (forwarded.empty()) {
+		return stats;
+	}
+
+	std::map<LaneKey, uint32_t> temporaries;
+	auto                        next_temporary = NextTemporaryScalarRegister(program);
+	for (const auto& key: forwarded) {
+		if (next_temporary == UINT32_MAX) {
+			return {};
+		}
+		temporaries.emplace(key, next_temporary++);
+	}
+
+	for (size_t block_index = 0; block_index < block_count; block_index++) {
+		const auto original  = std::move(program.blocks[block_index].instructions);
+		auto&      rewritten = program.blocks[block_index].instructions;
+		rewritten.clear();
+		rewritten.reserve(original.size() + temporaries.size());
+		auto state = entry[block_index];
+		for (const auto& inst: original) {
+			LaneKey read_key;
+			if (ReadLaneKey(program, inst, read_key) && state.contains(read_key)) {
+				const auto temporary = temporaries.find(read_key);
+				if (temporary != temporaries.end()) {
+					rewritten.push_back(ShadowRead(inst, temporary->second));
+					stats.rewritten_reads++;
+					ApplyInstruction(program, inst, state);
+					continue;
+				}
+			}
+
+			rewritten.push_back(inst);
+			LaneKey write_key;
+			if (WriteLaneKey(program, inst, write_key)) {
+				const auto temporary = temporaries.find(write_key);
+				if (temporary != temporaries.end()) {
+					rewritten.push_back(ShadowWrite(inst, temporary->second));
+					stats.shadow_writes++;
+				}
+			}
+			ApplyInstruction(program, inst, state);
+		}
+	}
+	return stats;
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::IR

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

@@ -0,0 +1,20 @@
+#ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_READLANEELIMINATION_H_
+#define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_READLANEELIMINATION_H_
+
+#include "graphics/shader/recompiler/ir/ShaderIR.h"
+
+namespace Libs::Graphics::ShaderRecompiler::IR {
+
+struct ReadLaneEliminationStats {
+	uint32_t rewritten_reads = 0;
+	uint32_t shadow_writes   = 0;
+};
+
+// Replaces fixed-lane ReadLane operations that are reached by a matching WriteLane on every
+// control-flow path. A synthetic scalar register snapshots the value at WriteLane execution time,
+// so the rewrite remains valid when the source SGPR is subsequently overwritten.
+[[nodiscard]] ReadLaneEliminationStats EliminateReadLane(Program& program);
+
+} // namespace Libs::Graphics::ShaderRecompiler::IR
+
+#endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_READLANEELIMINATION_H_ */

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

@@ -15,7 +15,8 @@ constexpr uint64_t AddressMask = 0x0000ffffffffffffull;
 Decoder::ImageDimension DescriptorDimension(const DescriptorValue&  descriptor,
                                             Decoder::ImageDimension requested) {
 	const bool is_array = requested == Decoder::ImageDimension::Dim1DArray ||
-	                      requested == Decoder::ImageDimension::Dim2DArray;
+	                      requested == Decoder::ImageDimension::Dim2DArray ||
+	                      requested == Decoder::ImageDimension::Dim2DMsaaArray;
 	switch (static_cast<Prospero::ImageType>((descriptor.dwords[3] >> 28u) & 0xfu)) {
 		case Prospero::ImageType::kColor1D: return Decoder::ImageDimension::Dim1D;
 		case Prospero::ImageType::kColor1DArray:
@@ -26,13 +27,17 @@ Decoder::ImageDimension DescriptorDimension(const DescriptorValue&       descrip
 		case Prospero::ImageType::kColor3D: return Decoder::ImageDimension::Dim3D;
 		case Prospero::ImageType::kCube: return Decoder::ImageDimension::Dim2DArray;
 		case Prospero::ImageType::kColor2DArray:
-		case Prospero::ImageType::kColor2DMsaaArray:
 			if (is_array) {
 				return Decoder::ImageDimension::Dim2DArray;
 			}
 			return Decoder::ImageDimension::Dim2D;
-		case Prospero::ImageType::kColor2D:
-		case Prospero::ImageType::kColor2DMsaa: return Decoder::ImageDimension::Dim2D;
+		case Prospero::ImageType::kColor2DMsaaArray:
+			if (is_array) {
+				return Decoder::ImageDimension::Dim2DMsaaArray;
+			}
+			return Decoder::ImageDimension::Dim2DMsaa;
+		case Prospero::ImageType::kColor2D: return Decoder::ImageDimension::Dim2D;
+		case Prospero::ImageType::kColor2DMsaa: return Decoder::ImageDimension::Dim2DMsaa;
 		default: return Decoder::ImageDimension::Unknown;
 	}
 }
@@ -43,7 +48,8 @@ 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 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 ||
@@ -51,8 +57,7 @@ bool ValidImageDescriptor(const DescriptorValue& descriptor) {
 		const auto base_level = (descriptor.dwords[3] >> 12u) & 0xfu;
 		const auto fragments  = (descriptor.dwords[3] >> 16u) & 0xfu;
 		const auto max_mip    = (descriptor.dwords[5] >> 4u) & 0xfu;
-		return base_level == 0 && fragments >= 1 && fragments <= 3 &&
-		       max_mip == fragments;
+		return base_level == 0 && fragments >= 1 && fragments <= 3 && max_mip == fragments;
 	}
 	return true;
 }
@@ -61,6 +66,11 @@ uint32_t DescriptorImageSwizzle(const DescriptorValue& descriptor) {
 	return descriptor.dwords[3] & 0xfffu;
 }
 
+bool DescriptorIsCube(const DescriptorValue& descriptor) {
+	return static_cast<Prospero::ImageType>((descriptor.dwords[3] >> 28u) & 0xfu) ==
+	       Prospero::ImageType::kCube;
+}
+
 bool DecodeBufferDescriptor(const DescriptorValue& descriptor, ShaderBufferResource& result) {
 	if (descriptor.dword_count != std::size(result.fields)) {
 		return false;
@@ -171,12 +181,13 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 		const auto& image      = program.info.images[i];
 		const auto& descriptor = snapshot.images[i];
 		if (NullImageDescriptor(descriptor)) {
-			bool canonical_kind = image.kind == ResourceKind::Image ||
-			                      image.kind == ResourceKind::StorageImage;
+			bool canonical_kind =
+			    image.kind == ResourceKind::Image || image.kind == ResourceKind::StorageImage;
 			if (image.atomic) {
 				canonical_kind = image.kind == ResourceKind::StorageImageUint;
 			}
-			if (image.dimension != Decoder::ImageDimension::Dim2D || !canonical_kind) {
+			if (image.dimension != Decoder::ImageDimension::Dim2D || image.cube ||
+			    !canonical_kind) {
 				if (error != nullptr) {
 					*error = fmt::format(
 					    "image descriptor {} no longer matches canonical null specialization", i);
@@ -186,10 +197,15 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 			continue;
 		}
 		const auto dimension = DescriptorDimension(descriptor, image.dimension);
-		if (dimension == Decoder::ImageDimension::Unknown || dimension != image.dimension) {
+		if (dimension == Decoder::ImageDimension::Unknown || dimension != image.dimension ||
+		    DescriptorIsCube(descriptor) != image.cube) {
 			if (error != nullptr) {
 				*error =
-				    fmt::format("image descriptor {} no longer matches specialized dimension", 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;
 		}
@@ -204,14 +220,21 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 				}
 				return false;
 			}
-			const auto uint_descriptor =
-			    Prospero::IsUintTextureFormat((descriptor.dwords[1] >> 20u) & 0x1ffu);
+			const auto 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;
 			}
@@ -361,6 +384,7 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 		auto&       image      = next.images[i];
 		if (NullImageDescriptor(descriptor)) {
 			image.dimension = Decoder::ImageDimension::Dim2D;
+			image.cube      = false;
 			switch (image.kind) {
 				case ResourceKind::ImageUint: image.kind = ResourceKind::Image; break;
 				case ResourceKind::StorageImageUint:
@@ -386,11 +410,19 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 			return false;
 		}
 		image.dimension = descriptor_dimension;
+		image.cube      = DescriptorIsCube(descriptor);
 		if (image.kind == ResourceKind::StorageImage ||
 		    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;
@@ -402,6 +434,7 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 		std::reference_wrapper<Instruction> inst;
 		ResourceKind                        kind;
 		Decoder::ImageDimension             dimension;
+		bool                                cube;
 	};
 	std::vector<ImagePatch> patches;
 	for (auto& block: program.blocks) {
@@ -420,13 +453,14 @@ bool SpecializeResources(Program& program, const ResourceSnapshot& snapshot, std
 				return false;
 			}
 			const auto& image = next.images[inst.memory.resource];
-			patches.push_back({std::ref(inst), image.kind, image.dimension});
+			patches.push_back({std::ref(inst), image.kind, image.dimension, image.cube});
 		}
 	}
 	program.info = std::move(next);
 	for (const auto& patch: patches) {
 		patch.inst.get().memory.kind            = patch.kind;
 		patch.inst.get().memory.image_dimension = patch.dimension;
+		patch.inst.get().memory.image_cube      = patch.cube;
 	}
 	return true;
 }

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);
 			}
 		}
 	}