kernel: share directory streams across seeks and reads

Fix PS5 boot failure when an optional empty sound path opens /app0/ and the game seeks and reads that directory. Pack directory records once and share their bytes and position across raw reads, enumeration, seeking, and size reporting. Apply seek and read error semantics with checked offsets.

Add multiblock directory regression coverage using the existing production Vulkan resource lifecycle. Existing CMake build and kernel_file_system test pass. Independent correctness, fixture, runtime, and final compactness audits passed; launch.json run reached Jurassic Park gameplay with 4560 sampled flips, no Vulkan validation errors, and normal window-close exit.

.coderabbit.yaml

@@ -1,37 +0,0 @@
-# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
-
-# Keep inherited review profiles, automatic reviews, tools, and pre-merge checks.
-inheritance: true
-
-reviews:
-  # Trim supplemental output and keep the remaining walkthrough collapsed.
-  high_level_summary: false
-  review_status: false
-  review_details: false
-  collapse_walkthrough: true
-  changed_files_summary: false
-  sequence_diagrams: false
-  estimate_code_review_effort: false
-  related_issues: false
-  related_prs: false
-  in_progress_fortune: false
-  poem: false
-  enable_prompt_for_ai_agents: false
-
-  # Hide standard finishing-touch controls; this also disables their commands.
-  finishing_touches:
-    docstrings:
-      enabled: false
-    unit_tests:
-      enabled: false
-    simplify:
-      enabled: false
-    autofix:
-      enabled: false
-    fix_ci:
-      enabled: false
-    resolve_merge_conflict:
-      enabled: false
-
-chat:
-  art: false

.github/workflows/build.yml

@@ -6,12 +6,7 @@ on:
     branches: [ main, master ]
   pull_request:
 
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
-  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
-
 env:
-  SCCACHE_GHA_ENABLED: "true"
   KYTY_BUILD_ORIGIN: ${{ github.repository != 'KytyPS5/KytyPS5' && 'Fork' || github.event_name == 'push' && 'Official' || 'Source' }}
   KYTY_BUILD_REPOSITORY: ${{ github.repository }}
 
@@ -43,6 +38,9 @@ jobs:
       - name: Setup Visual Studio environment
         uses: ilammy/msvc-dev-cmd@v1
 
+      - name: Setup Ninja
+        uses: seanmiddleditch/gha-setup-ninja@v5
+
       - name: Locate vcpkg
         id: vcpkg
         shell: pwsh
@@ -80,9 +78,6 @@ jobs:
           arch: win64_msvc2022_64
           cache: true
 
-      - name: Setup sccache
-        uses: mozilla-actions/sccache-action@v0.0.11
-
       - name: Verify toolchain
         shell: cmd
         run: |
@@ -105,22 +100,17 @@ jobs:
             -DKYTY_RELEASE_TAG=${{ needs.metadata.outputs.release_tag }} ^
             -DCMAKE_C_COMPILER=clang-cl ^
             -DCMAKE_CXX_COMPILER=clang-cl ^
-            -DCMAKE_C_COMPILER_LAUNCHER=sccache ^
-            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache ^
             -DCMAKE_PREFIX_PATH="%Qt6_DIR%"
 
       - name: Build
         shell: cmd
         run: |
-          cmake --build _Build/windows --target launcher ^
-            audio_out2_port_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests ^
-            resource_tracking_tests scalar_provenance_tests ^
-            resource_materialization_tests shader_vertex_metadata_tests --parallel
+          cmake --build _Build/windows --target launcher audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests --parallel
 
       - name: Test
         shell: cmd
         run: |
-          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|ime_dialog|lru_cache|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|virtual_memory_allocation)$"
+          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|ime_dialog|virtual_memory_allocation)$"
 
       - name: Install
         shell: cmd
@@ -155,7 +145,7 @@ jobs:
 
       - name: Install build dependencies
         shell: bash
-        run: brew install glslang
+        run: brew install glslang ninja
 
       - name: Install Qt 6.10.3
         uses: jurplel/install-qt-action@v4
@@ -166,9 +156,6 @@ jobs:
           arch: clang_64
           cache: true
 
-      - name: Setup sccache
-        uses: mozilla-actions/sccache-action@v0.0.11
-
       - name: Verify toolchain
         shell: bash
         run: |
@@ -193,24 +180,20 @@ jobs:
             -DCMAKE_OSX_ARCHITECTURES=x86_64 \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
-            -DCMAKE_C_COMPILER_LAUNCHER=sccache \
-            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
             -DCMAKE_PREFIX_PATH="$QT_ROOT_DIR"
 
       - name: Build
         shell: bash
         run: |
           cmake --build _Build/macos \
-            --target launcher audio_out2_port_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests \
-              resource_tracking_tests scalar_provenance_tests \
-              resource_materialization_tests shader_vertex_metadata_tests \
+            --target launcher audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests \
             --parallel
 
       - name: Test
         shell: bash
         run: |
           ctest --test-dir _Build/macos --output-on-failure \
-            -R '^(audio_out2_port|ime_dialog|lru_cache|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|virtual_memory_allocation)$'
+            -R '^(audio_out2_port|ime_dialog|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -319,9 +302,8 @@ jobs:
             libxkbcommon-dev \
             libxrandr-dev \
             libxss-dev \
-            libxtst-dev \
             lld \
-            pkg-config \
+            ninja-build \
             wayland-protocols
 
       - name: Install Qt 6.10.3
@@ -333,9 +315,6 @@ jobs:
           arch: linux_gcc_64
           cache: true
 
-      - name: Setup sccache
-        uses: mozilla-actions/sccache-action@v0.0.11
-
       - name: Verify toolchain
         shell: bash
         run: |
@@ -360,12 +339,10 @@ jobs:
             -DKYTY_RELEASE_TAG="${{ needs.metadata.outputs.release_tag }}" \
             -DCMAKE_C_COMPILER=clang \
             -DCMAKE_CXX_COMPILER=clang++ \
-            -DCMAKE_C_COMPILER_LAUNCHER=sccache \
-            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
             -DCMAKE_PREFIX_PATH="$Qt6_DIR" 2>&1 | tee _Build/configure.log
           exit "${PIPESTATUS[0]}"
 
-      - name: Verify SDL3 backends
+      - name: Verify SDL2 backends
         shell: bash
         run: |
           status=0
@@ -384,16 +361,14 @@ jobs:
         run: |
           cmake --build _Build/linux \
             --target launcher page_manager_tests memory_tracker_tests \
-              audio_out2_port_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests \
-              resource_tracking_tests scalar_provenance_tests \
-              resource_materialization_tests shader_vertex_metadata_tests \
+              audio_out2_port_tests ime_dialog_tests virtual_memory_allocation_tests \
             --parallel
 
       - name: Test
         shell: bash
         run: |
           ctest --test-dir _Build/linux --output-on-failure \
-            -R '^(audio_out2_port|ime_dialog|lru_cache|memory_tracker|page_manager|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|virtual_memory_allocation)$'
+            -R '^(audio_out2_port|ime_dialog|page_manager|memory_tracker|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash

.gitmodules

@@ -1,7 +1,7 @@
-[submodule "3rdparty/SDL3"]
-	path = 3rdparty/SDL3
+[submodule "3rdparty/SDL2"]
+	path = 3rdparty/SDL2
 	url = https://github.com/libsdl-org/SDL.git
-	branch = release-3.4.x
+	branch = SDL2
 [submodule "3rdparty/magic_enum"]
 	path = 3rdparty/magic_enum
 	url = https://github.com/Neargye/magic_enum.git
@@ -22,7 +22,7 @@
 	url = https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator.git
 [submodule "3rdparty/ffmpeg-core"]
 	path = 3rdparty/ffmpeg-core
-	url = https://github.com/KytyPS5/ext-ffmpeg-core.git
+	url = https://github.com/shadps4-emu/ext-ffmpeg-core.git
 	shallow = true
 [submodule "3rdparty/fmt"]
 	path = 3rdparty/fmt

.vscode/settings.json

@@ -2,17 +2,15 @@
     "cmake.sourceDirectory": "${workspaceFolder}",
     "cmake.buildDirectory": "${workspaceFolder}/_Build/vscode-clang",
     "cmake.generator": "Ninja",
-    "cmake.useVsDeveloperEnvironment": "always",
     "cmake.environment": {
         "CC": "clang-cl",
-        "CXX": "clang-cl"
+        "CXX": "clang-cl",
+        "PATH": "${env:ProgramFiles}/Microsoft Visual Studio/18/Community/VC/Tools/Llvm/x64/bin;${env:ProgramFiles}/Microsoft Visual Studio/18/Community/Common7/IDE/CommonExtensions/Microsoft/CMake/Ninja;${env:PATH}"
     },
     "cmake.configureSettings": {
         "CMAKE_C_COMPILER": "clang-cl",
         "CMAKE_CXX_COMPILER": "clang-cl",
         "CMAKE_PREFIX_PATH": "C:/Qt/6.11.0/msvc2022_64"
-    },
-    "C_Cpp.default.intelliSenseMode": "windows-clang-x64",
-    "C_Cpp.default.compileCommands": "${workspaceFolder}/_Build/vscode-clang/compile_commands.json",
-    "C_Cpp.default.cppStandard": "c++20"
+    }
 }
+

3rdparty/CMakeLists.txt

@@ -29,7 +29,7 @@ FetchContent_Declare(
 FetchContent_MakeAvailable(xbyak zydis)
 
 set(SDL_SHARED OFF CACHE BOOL "" FORCE)
-add_subdirectory(SDL3)
+add_subdirectory(SDL2)
 
 add_library(Vulkan-Headers INTERFACE)
 add_library(Vulkan::Headers ALIAS Vulkan-Headers)
@@ -86,9 +86,6 @@ set(SPDLOG_BUILD_WARNINGS OFF CACHE BOOL "" FORCE)
 set(SPDLOG_INSTALL OFF CACHE BOOL "" FORCE)
 set(SPDLOG_FMT_EXTERNAL ON CACHE BOOL "" FORCE)
 set(SPDLOG_NO_EXCEPTIONS ON CACHE BOOL "" FORCE)
-if(WIN32)
-	set(SPDLOG_WCHAR_FILENAMES ON CACHE BOOL "" FORCE)
-endif()
 add_subdirectory(spdlog EXCLUDE_FROM_ALL)
 
 if (NOT TARGET FFmpeg::ffmpeg)

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.16...4.3)
+cmake_minimum_required(VERSION 3.12...4.3)
 
 if(CMAKE_VERSION VERSION_GREATER_EQUAL 4.0)
 	set(CMAKE_POLICY_VERSION_MINIMUM 3.5 CACHE STRING "Minimum policy version for bundled dependencies")
@@ -12,22 +12,6 @@ endif()
 
 project(Kyty)
 
-find_program(SCCACHE_PROGRAM sccache)
-find_program(CCACHE_PROGRAM ccache)
-
-foreach(lang IN ITEMS C CXX)
-    # An explicitly empty launcher disables caching for this language.
-    if(NOT DEFINED CMAKE_${lang}_COMPILER_LAUNCHER)
-        if(SCCACHE_PROGRAM)
-            set(CMAKE_${lang}_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}")
-            message(STATUS "sccache enabled for ${lang}: ${SCCACHE_PROGRAM}")
-        elseif(CCACHE_PROGRAM)
-            set(CMAKE_${lang}_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
-            message(STATUS "ccache enabled for ${lang}: ${CCACHE_PROGRAM}")
-        endif()
-    endif()
-endforeach()
-
 set(KYTY_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/src")
 set(KYTY_TESTS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/tests")
 set(KYTY_THIRD_PARTY_DIR "${CMAKE_CURRENT_SOURCE_DIR}/3rdparty")
@@ -301,7 +285,6 @@ set(gpu_blit_generated_dir "${PROJECT_BINARY_DIR}/gpu_blit_shaders")
 set(gpu_blit_shader_sources
 	"${gpu_tiler_shader_dir}/gpu_blit_fs_triangle.vert"
 	"${gpu_tiler_shader_dir}/gpu_blit_color_to_ms_depth.frag"
-	"${gpu_tiler_shader_dir}/gpu_video_out_overlay.frag"
 )
 foreach(shader_source IN LISTS gpu_blit_shader_sources)
 	get_filename_component(shader_name "${shader_source}" NAME_WE)
@@ -332,19 +315,33 @@ if (KYTY_CLANG_CL)
 	set_source_files_properties(${kyty_emulator_src} PROPERTIES COMPILE_FLAGS "-Wno-pragma-pack -Wno-deprecated-declarations -D_TIMESPEC_DEFINED")
 endif()
 
-set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL3::SDL3-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
+set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL2-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
 
-# Linux system libraries.
+# Linux system libraries required by the static FFmpeg archive.
 if(LINUX)
 	find_package(Threads REQUIRED)
 	list(APPEND kyty_emulator_link_libraries m ${CMAKE_DL_LIBS} Threads::Threads)
+
+	# Optional FFmpeg dependencies.
+	find_package(ZLIB)
+	if(ZLIB_FOUND)
+		list(APPEND kyty_emulator_link_libraries ZLIB::ZLIB)
+	endif()
+	find_library(KYTY_BZ2_LIBRARY bz2)
+	if(KYTY_BZ2_LIBRARY)
+		list(APPEND kyty_emulator_link_libraries ${KYTY_BZ2_LIBRARY})
+	endif()
+	find_library(KYTY_LZMA_LIBRARY lzma)
+	if(KYTY_LZMA_LIBRARY)
+		list(APPEND kyty_emulator_link_libraries ${KYTY_LZMA_LIBRARY})
+	endif()
 endif()
 
 set(inc_headers
 	${KYTY_SOURCE_DIR}
+	${KYTY_THIRD_PARTY_DIR}/SDL2/include
 	${KYTY_THIRD_PARTY_DIR}/VulkanMemoryAllocator/include
 	${KYTY_THIRD_PARTY_DIR}/SPIRV-Tools/include
-	${KYTY_THIRD_PARTY_DIR}/SPIRV-Headers/include
 	${KYTY_THIRD_PARTY_DIR}/fmt/include
 	${KYTY_THIRD_PARTY_DIR}/xxhash
 	${KYTY_THIRD_PARTY_DIR}/cpuinfo/include
@@ -412,11 +409,6 @@ add_executable(bit_array_tests EXCLUDE_FROM_ALL
 )
 target_include_directories(bit_array_tests PRIVATE ${inc_headers})
 
-add_executable(lru_cache_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/LruCacheTests.cpp"
-)
-target_include_directories(lru_cache_tests PRIVATE ${inc_headers})
-
 add_executable(ime_dialog_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ImeDialogTests.cpp"
 	"${KYTY_SOURCE_DIR}/libs/imeCommon.cpp"
@@ -471,25 +463,6 @@ add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
 target_link_libraries(audio_out2_port_tests common fmt::fmt)
 target_include_directories(audio_out2_port_tests PRIVATE ${inc_headers})
 
-add_executable(ngs2_sampler_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/Ngs2SamplerTests.cpp"
-	"${KYTY_SOURCE_DIR}/loader/timer.cpp"
-)
-target_link_libraries(ngs2_sampler_tests common fmt::fmt LibAtrac9)
-target_include_directories(ngs2_sampler_tests PRIVATE ${inc_headers})
-
-add_executable(pad_haptics_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/PadHapticsTests.cpp"
-)
-target_link_libraries(pad_haptics_tests common fmt::fmt SDL3::SDL3-static)
-target_include_directories(pad_haptics_tests PRIVATE ${inc_headers})
-
-add_executable(save_data_memory_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/SaveDataMemoryTests.cpp"
-)
-target_link_libraries(save_data_memory_tests common fmt::fmt)
-target_include_directories(save_data_memory_tests PRIVATE ${inc_headers})
-
 add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/EventQueueLifetimeTests.cpp"
 	"${KYTY_SOURCE_DIR}/kernel/eventQueue.cpp"
@@ -567,7 +540,6 @@ if(BUILD_TESTING)
 	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 lru_cache COMMAND $<TARGET_FILE:lru_cache_tests>)
 	add_test(NAME shader_vertex_metadata COMMAND $<TARGET_FILE:shader_vertex_metadata_tests>)
 	add_test(NAME resource_materialization COMMAND $<TARGET_FILE:resource_materialization_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
@@ -575,16 +547,9 @@ if(BUILD_TESTING)
 	add_test(NAME kernel_file_system COMMAND $<TARGET_FILE:kernel_file_system_tests>)
 	add_test(NAME sync_on_address COMMAND $<TARGET_FILE:sync_on_address_tests>)
 	add_test(NAME audio_out2_port COMMAND $<TARGET_FILE:audio_out2_port_tests>)
-	add_test(NAME ngs2_sampler COMMAND $<TARGET_FILE:ngs2_sampler_tests>)
-	add_test(NAME pad_haptics COMMAND $<TARGET_FILE:pad_haptics_tests>)
-	add_test(NAME save_data_memory COMMAND $<TARGET_FILE:save_data_memory_tests>)
 	add_test(NAME shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
 	add_test(NAME shader_recompiler_alignbyte
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --alignbyte-only)
-	add_test(NAME graphics_pipeline_rasterization
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --polygon-mode-only)
-	add_test(NAME graphics_draw_offsets
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --draw-offset-only)
 	add_test(NAME virtual_memory_allocation
 		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests>)
 	add_test(NAME guest_red_zone_patcher
@@ -633,7 +598,6 @@ if(BUILD_TESTING)
 		memory_tracker_tests
 		page_manager_tests
 		bit_array_tests
-		lru_cache_tests
 		shader_vertex_metadata_tests
 		resource_materialization_tests
 		resource_tracking_tests
@@ -643,27 +607,9 @@ if(BUILD_TESTING)
 		shader_recompiler_compute_tests
 		virtual_memory_allocation_tests
 		audio_out2_port_tests
-		ngs2_sampler_tests
-		pad_haptics_tests
-		save_data_memory_tests
 	)
 endif()
 
-# Test Darwin signal-context restoration for SSE4a instructions.
-if(APPLE AND (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR
-		(NOT CMAKE_OSX_ARCHITECTURES AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")))
-	add_executable(macos_sse4a_tests EXCLUDE_FROM_ALL
-		"${KYTY_TESTS_DIR}/macosSse4aTests.cpp"
-		"${KYTY_SOURCE_DIR}/loader/x64InstructionEmulator.cpp"
-	)
-	target_link_libraries(macos_sse4a_tests common fmt::fmt Zydis::Zydis)
-	target_include_directories(macos_sse4a_tests PRIVATE ${inc_headers})
-	if(BUILD_TESTING)
-		add_test(NAME macos_sse4a COMMAND $<TARGET_FILE:macos_sse4a_tests>)
-		set_tests_properties(macos_sse4a PROPERTIES TIMEOUT 15)
-		add_dependencies(kyty_tests macos_sse4a_tests)
-	endif()
-endif()
 
 add_executable(kyty_emulator "${KYTY_SOURCE_DIR}/main.cpp" ${kyty_emulator_src})
 configure_macos_guest_address_space(kyty_emulator)
@@ -703,16 +649,7 @@ if(KYTY_CLANG_CL)
 	)
 	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}")
-	# LINK_FLAGS is a space-delimited string: a -Map path containing spaces would
-	# be split into bogus arguments, so pass it as a discrete link option instead.
-	target_link_options(kyty_emulator PRIVATE "-Wl,-Map=${KYTY_EMULATOR_MAP_LINK_PATH}")
-endif()
-
-if(LINUX)
-    target_link_options(kyty_emulator PRIVATE
-        $<$<CONFIG:Release>:-Wl,--strip-all>
-    )
+	set_target_properties(kyty_emulator PROPERTIES LINK_FLAGS "${KYTY_LD_OPTIONS} -Wl,-Map=${KYTY_EMULATOR_MAP_LINK_PATH}")
 endif()
 
 if(APPLE)

README.md

@@ -4,13 +4,13 @@
 [![Build KytyPS5 (Linux)](https://img.shields.io/github/actions/workflow/status/KytyPS5/KytyPS5/build.yml?branch=main&event=push&label=Build%20KytyPS5%20%28Linux%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
 [![Build KytyPS5 (macOS)](https://img.shields.io/github/actions/workflow/status/KytyPS5/KytyPS5/build.yml?branch=main&event=push&label=Build%20KytyPS5%20%28macOS%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
 [![Platform](https://img.shields.io/badge/platform-Windows%20x64%20%7C%20Linux%20x64%20%7C%20macOS%20x86__64-0078D4.svg)](#system-requirements)
-[![Status](https://img.shields.io/badge/status-active%20development-orange.svg)](#current-status)
+[![Status](https://img.shields.io/badge/status-early%20development-orange.svg)](#current-status)
 [![License](https://img.shields.io/badge/license-GPL--2.0-blue.svg)](LICENSE)
 
 KytyPS5 is a free and open-source PlayStation 5 emulator written in C++ for Windows and Linux,
 with experimental macOS support. It is based on a heavily modified version of
-[Kyty](https://github.com/InoriRus/Kyty). The project is in active development, and behavior
-can change significantly between builds.
+[Kyty](https://github.com/InoriRus/Kyty). The project is in an early stage of development, so
+compatibility is limited and behavior may change significantly between builds.
 
 > [!IMPORTANT]
 > KytyPS5 is not affiliated with Sony Interactive Entertainment or PlayStation. The project does
@@ -20,12 +20,12 @@ can change significantly between builds.
 ## Current Status
 
 KytyPS5 can boot 2D games and a selection of 3D games, including titles built with Unreal Engine
-4/5, Unity, and custom engines. External low-level emulation modules are neither required nor
-planned.
+4/5, Unity, and custom engines. No external low-level emulation modules are currently required.
 
-Development is currently focused on expanding game compatibility and improving boot reliability.
+Development is focused on compatibility and boot reliability.
 
-Windows and Linux are the primary platforms and receive the most testing.
+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. 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
@@ -36,16 +36,16 @@ Community game test results are available in the
 
 ## Bugs and Issues
 
-Compatibility, stability, and performance can vary between versions. You may encounter crashes
-or graphical glitches, so please include the version you tested when reporting an issue.
+The project is in an early stage, so please be mindful when opening new issues. Expect crashes,
+graphical glitches, low compatibility, and poor performance.
 
 ## Screenshots
 
 <table align="center">
   <tr>
     <td align="center">
-      <strong>Astro Bot</strong><br>
-      <img src="docs/screenshots/ps5-01.png" width="300" alt="Astro Bot running in KytyPS5">
+      <strong>Disgaea 6</strong><br>
+      <img src="docs/screenshots/ps5-01.png" width="300" alt="Disgaea 6 running in KytyPS5">
     </td>
     <td align="center">
       <strong>Dreaming Sarah</strong><br>
@@ -64,12 +64,12 @@ or graphical glitches, so please include the version you tested when reporting a
   </tr>
   <tr>
     <td align="center">
-      <strong>Demon's Souls</strong><br>
-      <img src="docs/screenshots/ps5-02.png" width="300" alt="Demon's Souls running in KytyPS5">
+      <strong>Hellboy</strong><br>
+      <img src="docs/screenshots/ps5-02.png" width="300" alt="Hellboy running in KytyPS5">
     </td>
     <td align="center">
-      <strong>Hellboy</strong><br>
-      <img src="docs/screenshots/ps5-06.png" width="300" alt="Hellboy running in KytyPS5">
+      <strong>Paleo Pines</strong><br>
+      <img src="docs/screenshots/ps5-06.png" width="300" alt="Paleo Pines running in KytyPS5">
     </td>
   </tr>
 </table>
@@ -79,7 +79,7 @@ or graphical glitches, so please include the version you tested when reporting a
 ## Contributing
 
 Testing games and submitting detailed bug reports are useful ways to contribute. Search existing
-issues first, then use the **Game Emulation Status Report** template and attach the complete log file.
+issues first, then use the **Game Emulation Bug Report** template and attach the complete log file.
 
 Code contributions should be focused, build successfully on the platforms they touch, and include
 relevant tests where practical. Windows is the primary target, so a change that alters shared code
@@ -90,14 +90,8 @@ starting a large change.
 
 Set up the clang-format hook after cloning:
 
-Install `pre-commit` using the method appropriate for your platform:
-
-- **Arch Linux / CachyOS:** `sudo pacman -S pre-commit`
-- **Other Linux / macOS / Windows:** `python -m pip install pre-commit`
-
-Then install the Git hook:
-
-```bash
+```powershell
+python -m pip install pre-commit
 python -m pre_commit install --install-hooks
 ```
 
@@ -132,12 +126,11 @@ the Vulkan/SPIR-V validation rules.
 ### Build requirements (Windows)
 
 - Git
-- CMake 3.22.1 or newer
+- CMake 3.12 or newer
 - Ninja
 - Visual Studio 2022 or Build Tools 2022 with the **Desktop development with C++** workload and
   **C++ Clang tools for Windows** component
 - Qt 6 for MSVC 2022 64-bit, including Concurrent, Network, and Widgets
-- [glslang](https://github.com/KhronosGroup/glslang/releases) (`glslangValidator`) on `PATH`
 
 The Microsoft C++ compiler (`cl.exe`) is not supported; use `clang-cl`.
 
@@ -166,19 +159,19 @@ The finished application and its runtime dependencies will be placed in
 
 ### Building on Linux
 
-Install the toolchain and the libraries the bundled SDL3 needs. Without the audio, Wayland and
-udev development packages SDL3 quietly configures itself without those backends, and the resulting
+Install the toolchain and the libraries the bundled SDL2 needs. Without the audio, Wayland and
+udev development packages SDL2 quietly configures itself without those backends, and the resulting
 build has no working sound and no gamepad hotplug:
 
 ```bash
 sudo apt-get install --no-install-recommends \
-  clang lld ninja-build cmake git glslang-tools pkg-config \
+  clang lld ninja-build cmake git glslang-tools \
   libgl1-mesa-dev libx11-dev libxcursor-dev libxext-dev libxfixes-dev \
-  libxi-dev libxrandr-dev libxss-dev libxtst-dev libxkbcommon-dev \
+  libxi-dev libxrandr-dev libxss-dev libxkbcommon-dev \
   libasound2-dev libpulse-dev libudev-dev libdbus-1-dev libwayland-dev wayland-protocols
 ```
 
-Qt 6 (Concurrent, Network, Widgets) is required for the launcher — either the distribution packages
+Qt 6 (Concurrent, Network, Widgets) is also required — either the distribution packages
 (`qt6-base-dev`) or an official Qt installation.
 
 ```bash
@@ -193,49 +186,13 @@ cmake --install _Build/linux --prefix _Build/linux/install
 ```
 
 The install step copies the Qt libraries and plugins next to the binaries, so
-`_Build/linux/install` runs without a matching system Qt. FFmpeg is linked statically
-from the pinned [KytyPS5 FFmpeg core](https://github.com/KytyPS5/ext-ffmpeg-core)
-release, including VP9 and WebM support. System FFmpeg packages are not required.
-
-To build `kyty_emulator` and the `kyty_tests` target without Qt, use a separate build directory:
-
-```bash
-git submodule update --init --recursive
-
-cmake -S . -B _Build/linux-no-qt -G Ninja -DCMAKE_BUILD_TYPE=Release \
-  -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
-  -DKYTY_BUILD_LAUNCHER=OFF
-
-cmake --build _Build/linux-no-qt --target kyty_emulator kyty_tests --parallel
-```
+`_Build/linux/install` runs without a matching system Qt.
 
 As on Windows, the MSVC compiler is not used; Clang is required. `cl.exe` is rejected at configure
 time.
 
 The CMake source root is the repository root.
 
-### Building on NixOS
-
-A development shell provides Clang, CMake, Ninja, Qt 6, the Vulkan headers, and the SDL3 backend
-libraries. Enter it and configure exactly as on other Linux distributions; the shell exports
-`CMAKE_PREFIX_PATH` and `QT_PLUGIN_PATH`, so the `-DCMAKE_PREFIX_PATH="$Qt6_DIR"` argument is not
-needed:
-
-```bash
-nix-shell # or: nix develop
-git submodule update --init --recursive
-
-cmake -S . -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \
-  -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++
-
-cmake --build _Build/linux --target launcher --parallel
-cmake --install _Build/linux --prefix _Build/linux/install
-```
-
-The configure step downloads the FFmpeg prebuilts and the `xbyak`/`zydis` sources, so it needs
-network access; a fully sandboxed `nix build` would require vendoring those inputs. A Vulkan 1.3
-driver must be available at runtime (on NixOS, `hardware.graphics.enable = true`).
-
 ### Building on macOS
 
 macOS builds target x86-64 and run under Rosetta 2 on Apple Silicon, so the PS5's x86-64 game
@@ -382,6 +339,5 @@ licenses included with those components.
 
 - [InoriRus/Kyty](https://github.com/InoriRus/Kyty) — KytyPS5 is based on a heavily modified version
   of the original Kyty project.
-- [shadps4-emu/shadPS4](https://github.com/shadps4-emu/shadPS4) — reference for understanding PS4
-  memory behavior, GPU resource aliasing and cache coherency,
-  and the AVPlayer implementation.
+- [shadps4-emu/shadPS4](https://github.com/shadps4-emu/shadPS4) — reference for memory-model
+  understanding and the AVPlayer implementation.

flake.lock

@@ -1,27 +0,0 @@
-{
-  "nodes": {
-    "nixpkgs": {
-      "locked": {
-        "lastModified": 1789426278,
-        "narHash": "sha256-GnotcvKtbTMnoVx6G4E7ZdUX6487Dky74dqq3pU7iRk=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "efe6f071ede9d21c37462d2d6682d5e670099684",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "ref": "nixos-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
-    "root": {
-      "inputs": {
-        "nixpkgs": "nixpkgs"
-      }
-    }
-  },
-  "root": "root",
-  "version": 7
-}

flake.nix

@@ -1,20 +0,0 @@
-{
-  description = "KytyPS5 development shell";
-
-  inputs.nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
-
-  outputs = { self, nixpkgs }:
-    let
-      systems = [ "x86_64-linux" ];
-      forAllSystems = nixpkgs.lib.genAttrs systems;
-    in
-    {
-      devShells = forAllSystems (system:
-        let
-          pkgs = import nixpkgs { inherit system; };
-        in
-        {
-          default = import ./nix/devshell.nix { inherit pkgs; };
-        });
-    };
-}

nix/devshell.nix

@@ -1,85 +0,0 @@
-{ pkgs }:
-
-let
-  # Build-time dependencies for the bundled SDL3 audio, video and input backends.
-  buildDeps = with pkgs; [
-    qt6.qtbase # Concurrent, Network, Widgets (launcher)
-    vulkan-headers
-    vulkan-loader
-    vulkan-tools
-    vulkan-validation-layers
-    libglvnd
-    # X11 / input
-    libX11
-    libXcursor
-    libXext
-    libXfixes
-    libXi
-    libXrandr
-    libXScrnSaver
-    libXtst
-    libxkbcommon
-    # Wayland
-    wayland
-    wayland-protocols
-    libdecor
-    # audio + device/dbus
-    alsa-lib
-    libpulseaudio
-    systemd # libudev
-    dbus
-  ];
-
-  # Libraries the built binaries need at runtime.
-  runtimeLibs = with pkgs; [
-    stdenv.cc.cc.lib # libstdc++ / libgcc_s
-    libgcc.lib
-    qt6.qtbase # libQt6Concurrent/Network/Widgets
-    vulkan-loader
-    libglvnd
-    alsa-lib
-    libpulseaudio
-    libX11
-    libXcursor
-    libXext
-    libXfixes
-    libXi
-    libXrandr
-    libXScrnSaver
-    libXtst
-    libxkbcommon
-    wayland
-    libdecor
-    systemd
-    dbus
-  ];
-in
-pkgs.mkShell {
-  name = "kytyps5";
-
-  nativeBuildInputs = with pkgs; [
-    clang
-    lld
-    cmake
-    ninja
-    pkg-config
-    git
-    glslang # provides glslangValidator for the bundled shaders
-  ];
-
-  buildInputs = buildDeps;
-
-  shellHook = ''
-    export CMAKE_PREFIX_PATH="${pkgs.qt6.qtbase}:''${CMAKE_PREFIX_PATH:-}"
-    export QT_PLUGIN_PATH="${pkgs.qt6.qtbase}/lib/qt-6/plugins:${pkgs.qt6.qtbase}/lib/qt6/plugins:''${QT_PLUGIN_PATH:-}"
-    export LD_LIBRARY_PATH="${pkgs.lib.makeLibraryPath runtimeLibs}:''${LD_LIBRARY_PATH:-}"
-
-    if [ -z "''${KYTY_SHELL_QUIET:-}" ]; then
-      echo "KytyPS5 dev shell"
-      echo "  cmake -S . -B _Build/linux -G Ninja -DCMAKE_BUILD_TYPE=Release \\"
-      echo "    -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++"
-      echo "  cmake --build _Build/linux --target launcher --parallel"
-      echo "  cmake --install _Build/linux --prefix _Build/linux/install"
-    fi
-  '';
-}

shell.nix

@@ -1,4 +0,0 @@
-let
-  pkgs = import <nixpkgs> { };
-in
-import ./nix/devshell.nix { inherit pkgs; }

src/common/alignment.h

@@ -1,25 +0,0 @@
-// SPDX-FileCopyrightText: 2014 Jannik Vogel <email@jannikvogel.de>
-// SPDX-License-Identifier: CC0-1.0
-
-#pragma once
-
-#include <cstddef>
-#include <type_traits>
-
-namespace Common {
-
-template <typename T>
-[[nodiscard]] constexpr T AlignUp(T value, std::size_t size) {
-	static_assert(std::is_unsigned_v<T>, "T must be an unsigned value.");
-	auto mod{static_cast<T>(value % size)};
-	value -= mod;
-	return static_cast<T>(mod == T{0} ? value : value + size);
-}
-
-template <typename T>
-[[nodiscard]] constexpr T AlignDown(T value, std::size_t size) {
-	static_assert(std::is_unsigned_v<T>, "T must be an unsigned value.");
-	return static_cast<T>(value - value % size);
-}
-
-} // namespace Common

src/common/byteBuffer.h

@@ -0,0 +1,166 @@
+#ifndef KYTY_COMMON_BYTEBUFFER_H_
+#define KYTY_COMMON_BYTEBUFFER_H_
+
+#include "common/assert.h"
+#include "common/common.h"
+
+#include <algorithm>
+#include <cstddef> // IWYU pragma: export
+#include <cstring>
+#include <utility>
+#include <vector>
+
+namespace Common {
+
+using Byte = std::byte;
+
+class ByteBuffer {
+public:
+	using DataType = std::vector<Byte>;
+
+	static constexpr uint32_t INVALID_INDEX = static_cast<uint32_t>(-1);
+
+	using iterator       = DataType::iterator;
+	using const_iterator = DataType::const_iterator;
+
+	ByteBuffer()                  = default;
+	ByteBuffer(const ByteBuffer&) = default;
+	ByteBuffer(ByteBuffer&&)      = default;
+
+	ByteBuffer& operator=(const ByteBuffer&) = default;
+	ByteBuffer& operator=(ByteBuffer&&)      = default;
+
+	explicit ByteBuffer(uint32_t size): m_data(size) {}
+
+	ByteBuffer(std::initializer_list<uint8_t> list) {
+		for (const uint8_t& e: list) {
+			m_data.emplace_back(static_cast<Byte>(e));
+		}
+	}
+
+	explicit ByteBuffer(const void* buf, uint32_t size)
+	    : ByteBuffer(size) // @suppress("Ambiguous problem")
+	{
+		EXIT_IF(buf == nullptr && size != 0);
+
+		if (size != 0) {
+			std::memcpy(m_data.data(), buf, size);
+		}
+	}
+
+	[[nodiscard]] uint32_t Size() const { return static_cast<uint32_t>(m_data.size()); }
+	[[nodiscard]] bool     IsEmpty() const { return m_data.empty(); }
+
+	void Clear() { m_data.clear(); }
+
+	void Add(const Byte& val) { m_data.push_back(val); }
+	void Add(Byte&& val) { m_data.push_back(std::move(val)); }
+
+	void Add(const Byte* val, uint32_t num) {
+		EXIT_IF(val == nullptr && num != 0);
+		m_data.insert(m_data.end(), val, val + num);
+	}
+
+	void Add(const ByteBuffer& v) {
+		if (this == &v) {
+			const ByteBuffer copy(v);
+			Add(copy.GetDataConst(), copy.Size());
+			return;
+		}
+		Add(v.GetDataConst(), v.Size());
+	}
+
+	void InsertAt(uint32_t index, const Byte& val) {
+		EXIT_IF(index > Size());
+		m_data.insert(m_data.begin() + index, val);
+	}
+
+	void InsertAt(uint32_t index, Byte&& val) {
+		EXIT_IF(index > Size());
+		m_data.insert(m_data.begin() + index, std::move(val));
+	}
+
+	[[nodiscard]] uint32_t Find(const Byte& t) const {
+		const auto it = std::find(m_data.begin(), m_data.end(), t);
+		return (it == m_data.end() ? INVALID_INDEX
+		                           : static_cast<uint32_t>(std::distance(m_data.begin(), it)));
+	}
+
+	[[nodiscard]] bool Contains(const Byte& t) const { return Find(t) != INVALID_INDEX; }
+
+	bool Remove(const Byte& t) {
+		const auto it = std::find(m_data.begin(), m_data.end(), t);
+		if (it == m_data.end()) {
+			return false;
+		}
+		m_data.erase(it);
+		return true;
+	}
+
+	bool RemoveAt(uint32_t index, uint32_t count = 1) {
+		if (index >= Size() || count == 0) {
+			return false;
+		}
+
+		const uint32_t erase_count = std::min(count, Size() - index);
+		m_data.erase(m_data.begin() + index, m_data.begin() + index + erase_count);
+		return true;
+	}
+
+	[[nodiscard]] bool IndexValid(uint32_t index) const { return index < Size(); }
+
+	Byte& operator[](uint32_t index) {
+		EXIT_IF(index >= Size());
+		return m_data[index];
+	}
+
+	const Byte& operator[](uint32_t index) const {
+		EXIT_IF(index >= Size());
+		return m_data[index];
+	}
+
+	[[nodiscard]] const Byte& At(uint32_t index) const {
+		EXIT_IF(index >= Size());
+		return m_data[index];
+	}
+
+	Byte*                     GetData() { return m_data.data(); }
+	[[nodiscard]] const Byte* GetData() const { return m_data.data(); }
+	[[nodiscard]] const Byte* GetDataConst() const { return m_data.data(); }
+
+	void Sort() {
+		std::sort(m_data.begin(), m_data.end(), [](Byte a, Byte b) {
+			return std::to_integer<uint8_t>(a) < std::to_integer<uint8_t>(b);
+		});
+	}
+
+	template <typename OP>
+	void Sort(OP&& comp_func) {
+		std::sort(m_data.begin(), m_data.end(), std::forward<OP>(comp_func));
+	}
+
+	bool operator==(const ByteBuffer& s) const { return m_data == s.m_data; }
+	bool operator!=(const ByteBuffer& s) const { return !(*this == s); }
+
+	iterator begin() { return m_data.begin(); } // NOLINT(readability-identifier-naming)
+	iterator end() { return m_data.end(); }     // NOLINT(readability-identifier-naming)
+	[[nodiscard]] const_iterator begin() const {
+		return m_data.begin();
+	} // NOLINT(readability-identifier-naming)
+	[[nodiscard]] const_iterator end() const {
+		return m_data.end();
+	} // NOLINT(readability-identifier-naming)
+	[[nodiscard]] const_iterator cbegin() const {
+		return m_data.cbegin();
+	} // NOLINT(readability-identifier-naming)
+	[[nodiscard]] const_iterator cend() const {
+		return m_data.cend();
+	} // NOLINT(readability-identifier-naming)
+
+private:
+	DataType m_data;
+};
+
+} // namespace Common
+
+#endif /* KYTY_COMMON_BYTEBUFFER_H_ */

src/common/debug.cpp

@@ -24,4 +24,8 @@ std::string Debug::GetLinker() {
 #endif
 }
 
+std::string Debug::GetBitness() {
+	return "64";
+}
+
 } // namespace Common

src/common/debug.h

@@ -10,6 +10,7 @@ namespace Common {
 namespace Debug {
 std::string GetCompiler();
 std::string GetLinker();
+std::string GetBitness();
 } // namespace Debug
 
 } // namespace Common

src/common/emulatorConfig.cpp

@@ -43,10 +43,6 @@ int32_t GetUserId() {
 	return g_config->user_id;
 }
 
-const std::string& GetAudioInputDevice() {
-	return g_config->audio_input_device;
-}
-
 PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
@@ -59,14 +55,6 @@ bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }
 
-bool VrEnabled() {
-	return g_config->vr_enabled;
-}
-
-bool AmdCpuEnabled() {
-	return g_config->amd_cpu_enabled;
-}
-
 uint32_t GetVblankFrequency() {
 	return std::clamp(g_config->vblank_frequency, 30u, 360u);
 }
@@ -87,7 +75,7 @@ ShaderOptimizationType GetShaderOptimizationType() {
 	return g_config->shader_optimization_type;
 }
 
-LogDirection GetShaderLogDirection() {
+ShaderLogDirection GetShaderLogDirection() {
 	return g_config->shader_log_direction;
 }
 
@@ -107,7 +95,7 @@ bool GraphicsDebugDumpEnabled() {
 	return g_config->graphics_debug_dump_enabled;
 }
 
-LogDirection GetPrintfDirection() {
+OutputDirection GetPrintfDirection() {
 	return g_config->printf_direction;
 }
 
@@ -115,8 +103,8 @@ std::filesystem::path GetPrintfOutputFile() {
 	return g_config->printf_output_file;
 }
 
-bool ProfilerEnabled() {
-	return g_config->profiler_enabled;
+ProfilerDirection GetProfilerDirection() {
+	return g_config->profiler_direction;
 }
 
 bool SpirvDebugPrintfEnabled() {
@@ -135,10 +123,6 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
-bool TessellationEnabled() {
-	return g_config->tessellation_enabled;
-}
-
 bool PlayGoHackEnabled() {
 	return g_config->playgo_hack_enabled;
 }

src/common/emulatorConfig.h

@@ -21,7 +21,11 @@ struct Lifecycle {
 
 enum class ShaderOptimizationType { None, Size, Performance };
 
-enum class LogDirection { Silent, Console, File };
+enum class ShaderLogDirection { Silent, Console, File };
+
+enum class ProfilerDirection { None, Network };
+
+enum class OutputDirection { Silent, Console, File };
 
 enum class PresentMode { Fifo, Mailbox, Immediate };
 
@@ -43,30 +47,26 @@ struct ConfigOptions {
 	uint32_t               screen_height               = 720;
 	std::string            user_name                   = "Kyty";
 	int32_t                user_id                     = DEFAULT_USER_ID;
-	std::string            audio_input_device;
-	PresentMode            present_mode                = PresentMode::Mailbox;
+	PresentMode            present_mode                = PresentMode::Fifo;
 	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
-	bool                   vr_enabled                  = false;
-	bool                   amd_cpu_enabled             = false;
 	uint32_t               vblank_frequency            = 60;
 	uint32_t               console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = false;
 	bool                   shader_validation_enabled   = false;
 	ShaderOptimizationType shader_optimization_type    = ShaderOptimizationType::None;
-	LogDirection           shader_log_direction        = LogDirection::Silent;
+	ShaderLogDirection     shader_log_direction        = ShaderLogDirection::Silent;
 	std::filesystem::path  shader_log_folder           = "_Shaders";
 	bool                   command_buffer_dump_enabled = false;
 	std::filesystem::path  command_buffer_dump_folder  = "_Buffers";
 	bool                   graphics_debug_dump_enabled = false;
-	LogDirection           printf_direction            = LogDirection::Silent;
+	OutputDirection        printf_direction            = OutputDirection::Silent;
 	std::filesystem::path  printf_output_file          = "_kyty.txt";
-	bool                   profiler_enabled            = false;
+	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   spirv_debug_printf_enabled  = false;
 	bool                   gpu_assisted_validation_enabled = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
-	bool                   tessellation_enabled        = false;
 	bool                   playgo_hack_enabled         = false;
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	bool red_zone_protection_enabled = false;
@@ -80,19 +80,16 @@ uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
-const std::string& GetAudioInputDevice();
 PresentMode GetPresentMode();
 int32_t GetGpuIndex();
 bool     FullscreenEnabled();
-bool     VrEnabled();
-bool     AmdCpuEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();
 
 bool                   ShaderValidationEnabled();
 ShaderOptimizationType GetShaderOptimizationType();
-LogDirection           GetShaderLogDirection();
+ShaderLogDirection     GetShaderLogDirection();
 std::filesystem::path  GetShaderLogFolder();
 
 bool                  CommandBufferDumpEnabled();
@@ -100,10 +97,10 @@ std::filesystem::path GetCommandBufferDumpFolder();
 
 bool GraphicsDebugDumpEnabled();
 
-LogDirection          GetPrintfDirection();
+OutputDirection       GetPrintfDirection();
 std::filesystem::path GetPrintfOutputFile();
 
-bool ProfilerEnabled();
+ProfilerDirection GetProfilerDirection();
 
 bool SpirvDebugPrintfEnabled();
 
@@ -111,7 +108,6 @@ bool GpuAssistedValidationEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
-bool TessellationEnabled();
 bool PlayGoHackEnabled();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();

src/common/file.cpp

@@ -136,6 +136,10 @@ bool File::OpenInMem(void* buf, uint32_t buf_size) {
 	return true;
 }
 
+bool File::OpenInMem(ByteBuffer& buf) {
+	return OpenInMem(buf.GetData(), buf.Size());
+}
+
 bool File::CreateInMem() {
 	EXIT_IF(m_p->f != nullptr);
 
@@ -245,78 +249,88 @@ void File::Printf(const char* format, ...) {
 	Write(s.data(), static_cast<uint32_t>(s.size()));
 }
 
-static std::filesystem::path WithoutTrailingSeparator(
-	const std::filesystem::path& path) {
-		if (!path.empty() && !path.has_filename() &&
-			path != path.root_path()) {
-			return path.parent_path();
-		}
-
-		return path;
-}
-
-
 bool File::IsDirectoryExisting(const std::filesystem::path& path) {
-	return SysFileIsDirectoryExisting(WithoutTrailingSeparator(path));
+	auto path_str = PathToGenericString(path);
+	return SysFileIsDirectoryExisting(
+	    Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1) : path_str);
 }
 
 bool File::IsFileExisting(const std::filesystem::path& name) {
 	return SysFileIsFileExisting(name);
 }
 
-bool File::CreateDirectory(const std::filesystem::path& path) {
-    return SysFileCreateDirectory(
-        WithoutTrailingSeparator(path));
+bool File::CreateDirectory(
+    const std::filesystem::path& path) // @suppress("Member declaration not found")
+{
+	auto path_str = PathToGenericString(path);
+	return SysFileCreateDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
+	                                                              : path_str);
 }
 
 bool File::DeleteDirectory(const std::filesystem::path& path) {
-    return SysFileDeleteDirectory(
-        WithoutTrailingSeparator(path));
+	auto path_str = PathToGenericString(path);
+	return SysFileDeleteDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
+	                                                              : path_str);
 }
 
 bool File::CreateDirectories(const std::filesystem::path& path) {
-    const auto normalized_path =
-        WithoutTrailingSeparator(path);
+	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
 
-    auto current_path = normalized_path.root_path();
+	std::vector<std::string> list = Common::Split(real_path, "/");
 
-    for (const auto& component:
-         normalized_path.relative_path()) {
-        current_path /= component;
+	std::string p;
 
-        if (IsDirectoryExisting(current_path)) {
-            continue;
-        }
+	for (uint32_t si = 0; si < list.size(); si++) {
+		const std::string& s = list[si];
 
-        if (!CreateDirectory(current_path)) {
-            return false;
-        }
-    }
+		if (si != 0 || Common::StartsWith(real_path, "/")) {
+			p += "/";
+		}
 
-    return true;
+		p += s;
+
+		if (IsDirectoryExisting(p)) {
+			continue;
+		}
+
+		if (!CreateDirectory(p)) // @suppress("Invalid arguments")
+		{
+			return false;
+		}
+	}
+
+	return true;
 }
 
 bool File::DeleteDirectories(const std::filesystem::path& path) {
-    const auto normalized_path =
-        WithoutTrailingSeparator(path);
+	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
 
-    std::vector<std::filesystem::path> directories;
-    auto current_path = normalized_path.root_path();
+	std::vector<std::string> list = Common::Split(real_path, "/");
 
-    for (const auto& component:
-         normalized_path.relative_path()) {
-        current_path /= component;
-        directories.push_back(current_path);
-    }
+	std::string              p;
+	std::vector<std::string> list2;
 
-    for (auto it = directories.rbegin();
-         it != directories.rend(); ++it) {
-        if (!DeleteDirectory(*it)) {
-            return false;
-        }
-    }
+	for (uint32_t si = 0; si < list.size(); si++) {
+		const std::string& s = list[si];
 
-    return true;
+		if (si != 0 || Common::StartsWith(real_path, "/")) {
+			p += "/";
+		}
+
+		p += s;
+
+		list2.push_back(p);
+	}
+
+	uint32_t num = list2.size();
+
+	for (uint32_t si = num - 1; si < num; si--) {
+		if (!DeleteDirectory(list2[si])) {
+			return false;
+		}
+	}
+
+	return true;
 }
 
 bool File::DeleteFile(
@@ -331,7 +345,7 @@ bool File::Flush() {
 	return SysFileFlush(*m_p->f);
 }
 
-std::vector<std::byte> File::ReadWholeBuffer() {
+ByteBuffer File::ReadWholeBuffer() {
 	EXIT_IF(IsInvalid());
 	EXIT_IF(Tell() != 0);
 
@@ -339,14 +353,25 @@ std::vector<std::byte> File::ReadWholeBuffer() {
 
 	EXIT_IF((s >> 32u) != 0);
 
-	const auto            read_size = static_cast<uint32_t>(s);
-	std::vector<std::byte> buf(read_size);
+	ByteBuffer buf(s);
 
-	Read(buf.data(), read_size);
+	Read(buf.GetData(), s);
 
 	return buf;
 }
 
+ByteBuffer File::Read(uint32_t size) {
+	ByteBuffer buf(size);
+	uint32_t   b = 0;
+	Read(buf.GetData(), size, &b);
+	buf.RemoveAt(b, size - b);
+	return buf;
+}
+
+void File::Write(const ByteBuffer& buf, uint32_t* bytes_written) {
+	Write(buf.GetDataConst(), buf.Size(), bytes_written);
+}
+
 DateTime File::GetLastAccessTimeUTC(const std::filesystem::path& name) {
 	SysTimeStruct t {};
 	SysFileToSystemTimeUtc(SysFileGetLastAccessTimeUtc(name), t);
@@ -445,6 +470,51 @@ bool File::SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name, const
 	return SysFileSetLastAccessAndWriteTimeUtc(name, af, wf);
 }
 
+std::vector<File::FindInfo> File::FindFiles(const std::filesystem::path& path) {
+	std::vector<sys_file_find_t> files;
+
+	SysFileFindFiles(path, files);
+
+	auto     path_str = PathToGenericString(path);
+	uint32_t len      = static_cast<uint32_t>(path_str.size());
+
+	if (!Common::EndsWith(path_str, "/") && !Common::EndsWith(path_str, "\\")) {
+		len++;
+	}
+
+	std::vector<File::FindInfo> ret;
+	ret.reserve(files.size());
+
+	for (const auto& f: files) {
+		File::FindInfo r {};
+
+		r.path_with_name     = f.path_with_name;
+		r.rel_path_with_name = Common::Mid(PathToGenericString(f.path_with_name), len);
+		r.size               = f.size;
+
+		SysTimeStruct at {};
+		SysTimeStruct wt {};
+
+		SysFileToSystemTimeUtc(f.last_access_time, at);
+		SysFileToSystemTimeUtc(f.last_write_time, wt);
+
+		if (!at.is_invalid && !wt.is_invalid) {
+			r.last_access_time = DateTime(Date(at.Year, at.Month, at.Day),
+			                              Time(at.Hour, at.Minute, at.Second, at.Milliseconds));
+			r.last_write_time  = DateTime(Date(wt.Year, wt.Month, wt.Day),
+			                              Time(wt.Hour, wt.Minute, wt.Second, wt.Milliseconds));
+		}
+
+		// printf("%s, %s, %" PRIu64", %s, %s\n", r.path_with_name.c_str(),
+		// r.rel_path_with_name.c_str(), r.size, r.last_access_time.ToString().c_str(),
+		// r.last_write_time.ToString().c_str());
+
+		ret.push_back(std::move(r));
+	}
+
+	return ret;
+}
+
 std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& path) {
 	std::vector<sys_dir_entry_t> files;
 
@@ -453,8 +523,13 @@ std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& pat
 	std::vector<File::DirEntry> ret;
 	ret.reserve(files.size());
 
-	for (auto& f: files) {
-		ret.emplace_back(std::move(f.name), f.is_file);
+	for (const auto& f: files) {
+		File::DirEntry r {};
+
+		r.name    = f.name;
+		r.is_file = f.is_file;
+
+		ret.push_back(std::move(r));
 	}
 
 	return ret;

src/common/file.h

@@ -1,10 +1,10 @@
 #ifndef KYTY_COMMON_FILE_H_
 #define KYTY_COMMON_FILE_H_
 
+#include "common/byteBuffer.h"
 #include "common/common.h"
 #include "common/dateTime.h"
 
-#include <cstddef>
 #include <filesystem>
 #include <memory>
 #include <string>
@@ -30,6 +30,14 @@ class File {
 public:
 	enum class Mode { Read, Write, ReadWrite, WriteRead };
 
+	struct FindInfo {
+		std::filesystem::path path_with_name;
+		std::filesystem::path rel_path_with_name;
+		DateTime              last_access_time;
+		DateTime              last_write_time;
+		uint64_t              size;
+	};
+
 	struct DirEntry {
 		std::string name;
 		bool        is_file;
@@ -43,6 +51,7 @@ public:
 	bool Create(const std::filesystem::path& name);
 	bool Open(const std::filesystem::path& name, Mode mode);
 	bool OpenInMem(void* buf, uint32_t buf_size);
+	bool OpenInMem(ByteBuffer& buf); // NOLINT(google-runtime-references)
 	bool CreateInMem();
 
 	void Close();
@@ -64,12 +73,14 @@ public:
 
 	void GetLastAccessAndWriteTimeUTC(DateTime* access, DateTime* write);
 
-	void Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
-	void Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
+	void       Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
+	ByteBuffer Read(uint32_t size);
+	void       Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
+	void       Write(const ByteBuffer& buf, uint32_t* bytes_written = nullptr);
 
 	void Printf(const char* format, ...) KYTY_FORMAT_PRINTF(2, 3);
 
-	std::vector<std::byte> ReadWholeBuffer();
+	ByteBuffer ReadWholeBuffer();
 
 	static uint64_t Size(const std::filesystem::path& name);
 
@@ -90,6 +101,7 @@ public:
 	static bool SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name,
 	                                         const DateTime& access, const DateTime& write);
 
+	static std::vector<FindInfo> FindFiles(const std::filesystem::path& path);
 	static std::vector<DirEntry> GetDirEntries(const std::filesystem::path& path);
 	static bool CopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 	static bool RenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/hash.h

@@ -54,6 +54,103 @@ inline uint32_t hash(const void* key, uint32_t key_len) {
 	return hash;
 }
 
+inline uint32_t hash8(uint8_t key) {
+	uint32_t hash = 0;
+
+	uint8_t* ptr = &key;
+
+	hash += ptr[0];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+
+	hash += (hash << 3u);
+	hash ^= (hash >> 11u);
+	hash += (hash << 15u);
+
+	return hash;
+}
+
+inline uint32_t hash16(uint16_t key) {
+	uint32_t hash = 0;
+
+	auto* ptr = reinterpret_cast<uint8_t*>(&key);
+
+	hash += ptr[0];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[1];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+
+	hash += (hash << 3u);
+	hash ^= (hash >> 11u);
+	hash += (hash << 15u);
+
+	return hash;
+}
+
+inline uint32_t hash32(uint32_t key) {
+	uint32_t hash = 0;
+
+	auto* ptr = reinterpret_cast<uint8_t*>(&key);
+
+	hash += ptr[0];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[1];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[2];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[3];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+
+	hash += (hash << 3u);
+	hash ^= (hash >> 11u);
+	hash += (hash << 15u);
+
+	return hash;
+}
+
+inline uint32_t hash64(uint64_t key) {
+	uint32_t hash = 0;
+
+	auto* ptr = reinterpret_cast<uint8_t*>(&key);
+
+	hash += ptr[0];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[1];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[2];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[3];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[4];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[5];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[6];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+	hash += ptr[7];
+	hash += (hash << 10u);
+	hash ^= (hash >> 6u);
+
+	hash += (hash << 3u);
+	hash ^= (hash >> 11u);
+	hash += (hash << 15u);
+
+	return hash;
+}
+
 } // namespace Common
 
 #endif /* KYTY_COMMON_HASH_H_ */

src/common/logging/log.cpp

@@ -3,6 +3,7 @@
 
 #include "common/assert.h"
 #include "common/emulatorConfig.h"
+#include "common/stringUtils.h"
 
 #include <cstdio>
 #include <filesystem>
@@ -46,7 +47,7 @@ std::shared_ptr<spdlog::logger> MakeFileLogger(std::string                  name
 	}
 
 	auto sink =
-	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(path.native(), true);
+	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(Common::PathToString(path), true);
 	return MakeLogger(std::move(name), std::move(sink));
 }
 
@@ -128,14 +129,6 @@ static void WriteImpl(std::string_view text, fmt::text_style style = {}) {
 	}
 }
 
-void WriteToConsoleAndLog(std::string_view text) {
-	WriteImpl(text);
-	if (g_initialized && g_direction != Direction::Console) {
-		WriteStdout(text);
-	}
-	Flush();
-}
-
 void WriteFatal(std::string_view text) {
 	if (g_direction == Direction::Silent || !g_initialized) {
 		WriteStdout(text);
@@ -159,9 +152,9 @@ void WriteFatal(fmt::text_style style, std::string_view text) {
 void Initialize() {
 	g_initialized = true;
 	switch (Config::GetPrintfDirection()) {
-		case Config::LogDirection::Silent: g_direction = Direction::Silent; break;
-		case Config::LogDirection::Console: g_direction = Direction::Console; break;
-		case Config::LogDirection::File: g_direction = Direction::File; break;
+		case Config::OutputDirection::Silent: g_direction = Direction::Silent; break;
+		case Config::OutputDirection::Console: g_direction = Direction::Console; break;
+		case Config::OutputDirection::File: g_direction = Direction::File; break;
 	}
 	g_output_file =
 	    (g_direction == Direction::File ? Config::GetPrintfOutputFile() : std::filesystem::path {});

src/common/logging/log.h

@@ -25,7 +25,6 @@ Direction GetDirection();
 bool      IsSilent();
 void      Write(std::string_view text);
 void      Write(fmt::text_style style, std::string_view text);
-void      WriteToConsoleAndLog(std::string_view text);
 void      WriteFatal(std::string_view text);
 void      WriteFatal(fmt::text_style style, std::string_view text);
 void      Flush();

src/common/magicEnum.h

@@ -0,0 +1,42 @@
+#ifndef KYTY_COMMON_MAGICENUM_H_
+#define KYTY_COMMON_MAGICENUM_H_
+
+#include "common/common.h"
+#include "common/stringUtils.h"
+#include "magic_enum.hpp" // IWYU pragma: export
+
+namespace Common {
+
+template <typename E>
+inline std::string EnumName(E v) {
+	auto str = magic_enum::enum_name(v);
+	return std::string(str.data(), str.length());
+}
+
+template <typename E>
+inline std::string EnumName8(E v) {
+	auto str = magic_enum::enum_name(v);
+	return std::string(str.data(), str.length());
+}
+
+template <typename E>
+inline E EnumValue(const std::string& str, E default_value) {
+	auto v = magic_enum::enum_cast<E>(str.c_str());
+	if (v.has_value()) {
+		return v.value();
+	}
+	return default_value;
+}
+
+#define KYTY_ENUM_RANGE(e, mx, mn)                                                                 \
+	namespace magic_enum::customize {                                                              \
+	template <>                                                                                    \
+	struct enum_range<e> {                                                                         \
+		static constexpr int min = (mx);                                                           \
+		static constexpr int max = (mn);                                                           \
+	};                                                                                             \
+	}
+
+} // namespace Common
+
+#endif /* KYTY_COMMON_MAGICENUM_H_ */

src/common/platform/sysFileIO.h

@@ -18,6 +18,14 @@ enum sys_file_cache_type_t {
 	SYS_FILE_CACHE_SEQUENTIAL_SCAN = 2  // NOLINT(readability-identifier-naming)
 };
 
+// NOLINTNEXTLINE(readability-identifier-naming)
+struct sys_file_find_t {
+	std::filesystem::path path_with_name;
+	SysFileTimeStruct     last_access_time;
+	SysFileTimeStruct     last_write_time;
+	uint64_t              size;
+};
+
 // NOLINTNEXTLINE(readability-identifier-naming)
 struct sys_dir_entry_t {
 	std::string name;
@@ -44,6 +52,7 @@ uint64_t    SysFileTell(sys_file_t& f);                    // NOLINT(google-runt
 bool        SysFileTruncate(sys_file_t& f, uint64_t size); // NOLINT(google-runtime-references)
 bool        SysFileUnlink(sys_file_t&                  f,
                           const std::filesystem::path& name); // NOLINT(google-runtime-references)
+void        SysFileWrite(uint32_t n, sys_file_t& f);          // NOLINT(google-runtime-references)
 uint64_t    SysFileSize(const std::filesystem::path& file_name);
 bool        SysFileIsError(sys_file_t& f); // NOLINT(google-runtime-references)
 bool        SysFileIsDirectoryExisting(const std::filesystem::path& path);
@@ -67,6 +76,8 @@ bool SysFileSetLastWriteTimeUtc(const std::filesystem::path& name, SysFileTimeSt
 bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
                                          SysFileTimeStruct& access, SysFileTimeStruct& write);
 // NOLINTNEXTLINE(google-runtime-references)
+void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out);
+// NOLINTNEXTLINE(google-runtime-references)
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out);
 bool SysFileCopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 bool SysFileRenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/platform/sysLinuxFileIO.cpp

@@ -11,7 +11,6 @@
 
 #include <cerrno>
 #include <cstdlib>
-#include <cstring>
 #include <dirent.h>
 #include <fcntl.h>
 #include <filesystem>
@@ -152,6 +151,10 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
+void SysFileWrite(uint32_t n, sys_file_t& f) {
+	SysFileWrite(&n, 4, f);
+}
+
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 
@@ -610,6 +613,51 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	//	}
 }
 
+// Recursively collect regular files.
+void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
+	auto real_path = get_internal_name(path);
+
+	DIR* dir = opendir(real_path.string().c_str());
+	if (dir == nullptr) {
+		return;
+	}
+
+	for (const dirent* entry = readdir(dir); entry != nullptr; entry = readdir(dir)) {
+		const std::string file_name(entry->d_name);
+
+		if (file_name == "." || file_name == "..") {
+			continue;
+		}
+
+		auto        child = real_path / file_name;
+		struct stat s {};
+
+		// lstat, so a symlink is never followed into a cycle during the recursive walk.
+		if (0 != lstat(child.string().c_str(), &s)) {
+			continue;
+		}
+
+		if (S_ISDIR(s.st_mode)) {
+			SysFileFindFiles(child, out);
+		} else if (S_ISREG(s.st_mode)) {
+			sys_file_find_t r {};
+
+			r.path_with_name              = child;
+			r.size                        = static_cast<uint64_t>(s.st_size);
+			r.last_access_time.is_invalid = false;
+			r.last_access_time.time       = s.st_atime;
+			r.last_access_time.nanos      = KYTY_STAT_ATIME_NS(s);
+			r.last_write_time.is_invalid  = false;
+			r.last_write_time.time        = s.st_mtime;
+			r.last_write_time.nanos       = KYTY_STAT_MTIME_NS(s);
+
+			out.push_back(r);
+		}
+	}
+
+	closedir(dir);
+}
+
 // Keep "." and ".." to match FindFirstFileW.
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
 	auto real_path = get_internal_name(path);

src/common/platform/sysLinuxVirtual.cpp

@@ -5,6 +5,7 @@
 #else
 
 #include "common/assert.h"
+#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 #include <atomic>
@@ -27,12 +28,13 @@
 #define KYTY_FIXED_NOREPLACE
 #endif
 
-namespace Common::VirtualMemory {
+namespace Common {
 
 static pthread_mutex_t              g_virtual_mutex {};
-static std::map<uintptr_t, size_t>* g_allocs = nullptr;
+static std::map<uintptr_t, size_t>* g_allocs   = nullptr;
+static std::map<uintptr_t, int>*    g_protects = nullptr;
 
-void Init() {
+void SysVirtualInit() {
 	pthread_mutexattr_t attr {};
 
 	pthread_mutexattr_init(&attr);
@@ -44,27 +46,43 @@ void Init() {
 	pthread_mutex_init(&g_virtual_mutex, &attr);
 	pthread_mutexattr_destroy(&attr);
 
-	g_allocs = new std::map<uintptr_t, size_t>;
+	g_allocs   = new std::map<uintptr_t, size_t>;
+	g_protects = new std::map<uintptr_t, int>;
 }
 
-static int get_protection_flag(Mode mode) {
+static int get_protection_flag(VirtualMemory::Mode mode) {
 	int protect = PROT_NONE;
 	switch (mode) {
-		case Mode::Read: protect = PROT_READ; break;
-		case Mode::Write:
-		case Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
-		case Mode::Execute: protect = PROT_EXEC; break;
-		case Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
-		case Mode::ExecuteWrite:
-		case Mode::ExecuteReadWrite:
+		case VirtualMemory::Mode::Read: protect = PROT_READ; break;
+		case VirtualMemory::Mode::Write:
+		case VirtualMemory::Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
+		case VirtualMemory::Mode::Execute: protect = PROT_EXEC; break;
+		case VirtualMemory::Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
+		case VirtualMemory::Mode::ExecuteWrite:
+		case VirtualMemory::Mode::ExecuteReadWrite:
 			protect = PROT_EXEC | PROT_WRITE | PROT_READ;
 			break; // NOLINT
-		case Mode::NoAccess:
+		case VirtualMemory::Mode::NoAccess:
 		default: protect = PROT_NONE; break;
 	}
 	return protect;
 }
 
+static VirtualMemory::Mode get_protection_flag(int mode) {
+	switch (mode) {
+		case PROT_NONE: return VirtualMemory::Mode::NoAccess;
+		case PROT_READ: return VirtualMemory::Mode::Read;
+		case PROT_WRITE: return VirtualMemory::Mode::Write;
+		case PROT_READ | PROT_WRITE: return VirtualMemory::Mode::ReadWrite; // NOLINT
+		case PROT_EXEC: return VirtualMemory::Mode::Execute;
+		case PROT_EXEC | PROT_WRITE: return VirtualMemory::Mode::ExecuteWrite; // NOLINT
+		case PROT_EXEC | PROT_READ: return VirtualMemory::Mode::ExecuteRead;   // NOLINT
+		case PROT_EXEC | PROT_WRITE | PROT_READ:
+			return VirtualMemory::Mode::ExecuteReadWrite; // NOLINT
+		default: return VirtualMemory::Mode::NoAccess;
+	}
+}
+
 // Keep automatic mappings inside the guest and GPU-addressable low window.
 #ifdef KYTY_FIXED_NOREPLACE
 static constexpr uintptr_t LOW_ARENA_LIMIT = 0x000000FC00000000ULL; // libc mspace window ceiling
@@ -129,7 +147,7 @@ static void* map_anonymous(uintptr_t addr, size_t size, int protect, int flags)
 	return mmap(nullptr, size, protect, flags, -1, 0); // NOLINT
 }
 
-uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
+uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -143,6 +161,11 @@ uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
 	if (ptr != MAP_FAILED) {
 		pthread_mutex_lock(&g_virtual_mutex);
 		record_alloc(ret_addr, size);
+		uintptr_t page_start = ret_addr >> 12u;
+		uintptr_t page_end   = (ret_addr + size - 1) >> 12u;
+		for (uintptr_t page = page_start; page <= page_end; page++) {
+			(*g_protects)[page] = protect;
+		}
 		pthread_mutex_unlock(&g_virtual_mutex);
 	}
 
@@ -153,7 +176,8 @@ static uintptr_t align_up(uintptr_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
+uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                                uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -222,11 +246,16 @@ uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t align
 	}
 
 	if (ptr == MAP_FAILED) {
-		return AllocAligned(address, size, mode, alignment << 1u);
+		return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
 	record_alloc(ret_addr, size);
+	uintptr_t page_start = ret_addr >> 12u;
+	uintptr_t page_end   = (ret_addr + size - 1) >> 12u;
+	for (uintptr_t page = page_start; page <= page_end; page++) {
+		(*g_protects)[page] = protect;
+	}
 	pthread_mutex_unlock(&g_virtual_mutex);
 
 	return ret_addr;
@@ -279,7 +308,7 @@ static bool is_mapped(void* ptr, size_t length) {
 }
 #endif
 
-bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
+bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr    = static_cast<uintptr_t>(address);
@@ -308,6 +337,11 @@ bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	if (ptr != MAP_FAILED) {
 		pthread_mutex_lock(&g_virtual_mutex);
 		record_alloc(ret_addr, size);
+		uintptr_t page_start = ret_addr >> 12u;
+		uintptr_t page_end   = (ret_addr + size - 1) >> 12u;
+		for (uintptr_t page = page_start; page <= page_end; page++) {
+			(*g_protects)[page] = protect;
+		}
 		pthread_mutex_unlock(&g_virtual_mutex);
 
 		return true;
@@ -316,15 +350,15 @@ bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	return false;
 }
 
-bool Commit(uint64_t address, uint64_t size, Mode mode) {
-	return Protect(address, size, mode);
+bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+	return SysVirtualProtect(address, size, mode);
 }
 
-uint64_t Reserve(uint64_t address, uint64_t size) {
-	return ReserveAligned(address, size, 1);
+uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
+	return SysVirtualReserveAligned(address, size, 1);
 }
 
-uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -393,7 +427,7 @@ uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	}
 
 	if (ptr == MAP_FAILED) {
-		return ReserveAligned(address, size, alignment << 1u);
+		return SysVirtualReserveAligned(address, size, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
@@ -403,7 +437,7 @@ uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	return ret_addr;
 }
 
-bool ReserveFixed(uint64_t address, uint64_t size) {
+bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -439,9 +473,9 @@ bool ReserveFixed(uint64_t address, uint64_t size) {
 	return false;
 }
 
-bool Decommit(uint64_t address, uint64_t size) {
+bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	// Drop physical pages while preserving the reservation.
-	if (!Protect(address, size, Mode::NoAccess)) {
+	if (!SysVirtualProtect(address, size, VirtualMemory::Mode::NoAccess)) {
 		return false;
 	}
 
@@ -465,7 +499,7 @@ bool Decommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool Free(uint64_t address) {
+bool SysVirtualFree(uint64_t address) {
 	EXIT_IF(g_allocs == nullptr);
 	size_t size = 0;
 
@@ -482,10 +516,21 @@ bool Free(uint64_t address) {
 		return false;
 	}
 
-	return munmap(reinterpret_cast<void*>(addr), size) == 0;
+	if (munmap(reinterpret_cast<void*>(addr), size) == 0) {
+		uintptr_t page_start = addr >> 12u;
+		uintptr_t page_end   = (addr + size - 1) >> 12u;
+		pthread_mutex_lock(&g_virtual_mutex);
+		for (uintptr_t page = page_start; page <= page_end; page++) {
+			g_protects->erase(page);
+		}
+		pthread_mutex_unlock(&g_virtual_mutex);
+		return true;
+	}
+
+	return false;
 }
 
-bool FreeRange(uint64_t address, uint64_t size) {
+bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 	if (size == 0 || (address & 0xfffu) != 0 || (size & 0xfffu) != 0) {
 		return false;
@@ -537,22 +582,46 @@ bool FreeRange(uint64_t address, uint64_t size) {
 	if (end < alloc_end) {
 		(*g_allocs)[end] = alloc_end - end;
 	}
+	for (uintptr_t page = addr >> 12u; page <= (end - 1u) >> 12u; page++) {
+		g_protects->erase(page);
+	}
 	pthread_mutex_unlock(&g_virtual_mutex);
 	return true;
 }
 
-bool Protect(uint64_t address, uint64_t size, Mode mode) {
-	const auto addr       = static_cast<uintptr_t>(address);
-	const auto page_start = addr >> 12u;
-	const auto page_end   = (addr + size - 1) >> 12u;
-	return mprotect(reinterpret_cast<void*>(page_start << 12u), (page_end - page_start + 1) << 12u,
-	                get_protection_flag(mode)) == 0;
+bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                       VirtualMemory::Mode* old_mode) {
+	auto addr = static_cast<uintptr_t>(address);
+
+	pthread_mutex_lock(&g_virtual_mutex);
+	if (old_mode != nullptr) {
+		if (auto s = g_protects->find(addr >> 12u); s != g_protects->end()) {
+			*old_mode = get_protection_flag(s->second);
+		} else {
+			*old_mode = VirtualMemory::Mode::NoAccess;
+		}
+	}
+	pthread_mutex_unlock(&g_virtual_mutex);
+
+	uintptr_t page_start = addr >> 12u;
+	uintptr_t page_end   = (addr + size - 1) >> 12u;
+	if (mprotect(reinterpret_cast<void*>(page_start << 12u), (page_end - page_start + 1) << 12u,
+	             get_protection_flag(mode)) == 0) {
+		pthread_mutex_lock(&g_virtual_mutex);
+		for (uintptr_t page = page_start; page <= page_end; page++) {
+			(*g_protects)[page] = get_protection_flag(mode);
+		}
+		pthread_mutex_unlock(&g_virtual_mutex);
+		return true;
+	}
+
+	return false;
 }
 
-bool FlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
+bool SysVirtualFlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
 	return true;
 }
 
-} // namespace Common::VirtualMemory
+} // namespace Common
 
 #endif

src/common/platform/sysSwapByteOrder.h

@@ -0,0 +1,53 @@
+#ifndef KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
+#define KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
+
+#include "common/common.h"
+
+#include <type_traits>
+
+inline uint16_t SwapByteOrder16(uint16_t value) {
+	uint16_t hi = value << 8u;
+	uint16_t lo = value >> 8u;
+	return hi | lo;
+}
+
+inline uint32_t SwapByteOrder32(uint32_t value) {
+	return __builtin_bswap32(value);
+}
+
+inline uint64_t SwapByteOrder64(uint64_t value) {
+	return __builtin_bswap64(value);
+}
+
+template <typename T>
+inline void SwapByteOrder(T& x) {
+	if (sizeof(x) == 2) {
+		if (std::is_signed_v<T>) {
+			x = static_cast<std::make_signed_t<T>>(
+			    SwapByteOrder16(static_cast<std::make_unsigned_t<uint16_t>>(x)));
+		} else {
+			x = SwapByteOrder16(x);
+		}
+	}
+	if (sizeof(x) == 4) {
+		if (std::is_signed_v<T>) {
+			x = static_cast<std::make_signed_t<T>>(
+			    SwapByteOrder32(static_cast<std::make_unsigned_t<uint32_t>>(x)));
+		} else {
+			x = SwapByteOrder32(x);
+		}
+	}
+	if (sizeof(x) == 8) {
+		if (std::is_signed_v<T>) {
+			x = static_cast<std::make_signed_t<T>>(
+			    SwapByteOrder64(static_cast<std::make_unsigned_t<uint64_t>>(x)));
+		} else {
+			x = SwapByteOrder64(x);
+		}
+	}
+}
+
+template <typename T>
+inline void NoSwapByteOrder(T& x) {}
+
+#endif /* KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_ */

src/common/platform/sysTimer.h

@@ -151,6 +151,10 @@ inline void SysTimeTToSystem(time_t t, SysTimeStruct& s) {
 	SysFileToSystemTimeUtc(ft, s);
 }
 
+inline time_t SysTimegm(struct tm* tm) {
+	return timegm(tm);
+}
+
 inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t) {
 	struct tm i {};
 
@@ -161,7 +165,7 @@ inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t)
 	i.tm_min  = f.Minute;
 	i.tm_sec  = f.Second;
 
-	t.is_invalid = (f.is_invalid || (t.time = ::timegm(&i)) == static_cast<time_t>(-1));
+	t.is_invalid = (f.is_invalid || (t.time = SysTimegm(&i)) == static_cast<time_t>(-1));
 }
 
 // Retrieves the current local date and time.

src/common/platform/sysVirtual.h

@@ -0,0 +1,28 @@
+#ifndef KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
+#define KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
+
+#include "common/common.h"
+#include "common/virtualMemory.h"
+
+namespace Common {
+
+void SysVirtualInit();
+
+uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
+uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                                uint64_t alignment);
+bool     SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
+bool     SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
+uint64_t SysVirtualReserve(uint64_t address, uint64_t size);
+uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment);
+bool     SysVirtualReserveFixed(uint64_t address, uint64_t size);
+bool     SysVirtualDecommit(uint64_t address, uint64_t size);
+bool     SysVirtualFree(uint64_t address);
+bool     SysVirtualFreeRange(uint64_t address, uint64_t size);
+bool     SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                           VirtualMemory::Mode* old_mode = nullptr);
+bool     SysVirtualFlushInstructionCache(uint64_t address, uint64_t size);
+
+} // namespace Common
+
+#endif /* KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_ */

src/common/platform/sysWindowsFileIO.cpp

@@ -14,7 +14,6 @@
 #include "common/stringUtils.h"
 
 #include <cstdlib>
-#include <cstring>
 #include <vector>
 
 // NOLINTNEXTLINE(readability-identifier-naming)
@@ -57,10 +56,10 @@ static DWORD GetCacheAccessType(sys_file_cache_type_t t) {
 	}
 
 	if (t == SYS_FILE_CACHE_SEQUENTIAL_SCAN) {
-		return FILE_FLAG_SEQUENTIAL_SCAN;
+		return SYS_FILE_CACHE_SEQUENTIAL_SCAN;
 	}
 
-	return FILE_ATTRIBUTE_NORMAL;
+	return SYS_FILE_CACHE_AUTO;
 }
 
 void SysFileRead(void* data, uint32_t size, sys_file_t& f, uint32_t* bytes_read) {
@@ -136,6 +135,10 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
+void SysFileWrite(uint32_t n, sys_file_t& f) {
+	SysFileWrite(&n, 4, f);
+}
+
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 
@@ -476,11 +479,66 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	return ok;
 }
 
-void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
-	const auto pattern = path / L"*";
+void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
+	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
+	if (!Common::EndsWith(real_path, "/")) {
+		real_path += "/";
+	}
 
-	WIN32_FIND_DATAW data {};
-	HANDLE h = FindFirstFileW(pattern.c_str(), &data);
+	std::string pattern = real_path + "*";
+
+	HANDLE           h = nullptr;
+	WIN32_FIND_DATAW data;
+
+	auto wide_pattern = std::filesystem::path(pattern).wstring();
+	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
+
+	if (h == INVALID_HANDLE_VALUE) {
+		return;
+	}
+
+	do {
+		std::filesystem::path file_name(data.cFileName);
+		auto                  file_name_str = Common::PathToString(file_name);
+
+		if (file_name_str == "." || file_name_str == "..") {
+			continue;
+		}
+
+		if ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) != 0u) {
+			SysFileFindFiles(std::filesystem::path(real_path) / file_name, out);
+		} else {
+			sys_file_find_t r {};
+
+			r.path_with_name              = std::filesystem::path(real_path) / file_name;
+			r.size                        = (static_cast<uint64_t>(data.nFileSizeHigh) << 32u) +
+			                                static_cast<uint64_t>(data.nFileSizeLow);
+			r.last_access_time.is_invalid = false;
+			r.last_access_time.time       = data.ftLastAccessTime;
+			r.last_write_time.is_invalid  = false;
+			r.last_write_time.time        = data.ftLastWriteTime;
+
+			out.push_back(r);
+		}
+
+	} while (FindNextFileW(h, &data) != 0);
+
+	FindClose(h);
+}
+
+void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
+	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
+	if (!Common::EndsWith(real_path, "/")) {
+		real_path += "/";
+	}
+
+	std::string pattern = real_path + "*";
+
+	HANDLE           h = nullptr;
+	WIN32_FIND_DATAW data;
+
+	auto wide_pattern = std::filesystem::path(pattern).wstring();
+	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
 
 	if (h == INVALID_HANDLE_VALUE) {
 		return;
@@ -494,7 +552,7 @@ void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entr
 		r.is_file = ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) == 0u);
 		r.name    = Common::PathToString(file_name);
 
-		out.push_back(std::move(r));
+		out.push_back(r);
 
 	} while (FindNextFileW(h, &data) != 0);
 

src/common/platform/sysWindowsVirtual.cpp

@@ -10,6 +10,7 @@
 #include <windows.h> // IWYU pragma: keep
 
 #include "common/assert.h"
+#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 // IWYU pragma: no_include <basetsd.h>
@@ -21,33 +22,45 @@
 // IWYU pragma: no_include <winerror.h>
 // IWYU pragma: no_include <wtypes.h>
 
-namespace Common::VirtualMemory {
+namespace Common {
 
-static DWORD GetProtectionFlag(Mode mode) {
+static DWORD GetProtectionFlag(VirtualMemory::Mode mode) {
 	DWORD protect = PAGE_NOACCESS;
 	switch (mode) {
-		case Mode::Read: protect = PAGE_READONLY; break;
+		case VirtualMemory::Mode::Read: protect = PAGE_READONLY; break;
 
-		case Mode::Write:
-		case Mode::ReadWrite: protect = PAGE_READWRITE; break;
+		case VirtualMemory::Mode::Write:
+		case VirtualMemory::Mode::ReadWrite: protect = PAGE_READWRITE; break;
 
-		case Mode::Execute: protect = PAGE_EXECUTE; break;
+		case VirtualMemory::Mode::Execute: protect = PAGE_EXECUTE; break;
 
-		case Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
+		case VirtualMemory::Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
 
-		case Mode::ExecuteWrite:
-		case Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
+		case VirtualMemory::Mode::ExecuteWrite:
+		case VirtualMemory::Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
 
-		case Mode::NoAccess:
+		case VirtualMemory::Mode::NoAccess:
 		default: protect = PAGE_NOACCESS; break;
 	}
 	return protect;
 }
 
-void Init() {}
+static VirtualMemory::Mode GetProtectionFlag(DWORD mode) {
+	switch (mode) {
+		case PAGE_NOACCESS: return VirtualMemory::Mode::NoAccess;
+		case PAGE_READONLY: return VirtualMemory::Mode::Read;
+		case PAGE_READWRITE: return VirtualMemory::Mode::ReadWrite;
+		case PAGE_EXECUTE: return VirtualMemory::Mode::Execute;
+		case PAGE_EXECUTE_READ: return VirtualMemory::Mode::ExecuteRead;
+		case PAGE_EXECUTE_READWRITE: return VirtualMemory::Mode::ExecuteReadWrite;
+		default: return VirtualMemory::Mode::NoAccess;
+	}
+}
 
-uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
-	auto ptr = (address == 0 ? AllocAligned(address, size, mode, 1)
+void SysVirtualInit() {}
+
+uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+	auto ptr = (address == 0 ? SysVirtualAllocAligned(address, size, mode, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE),
@@ -58,7 +71,7 @@ uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc() failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return AllocAligned(address, size, mode, 1);
+			return SysVirtualAllocAligned(address, size, mode, 1);
 		}
 	}
 	return ptr;
@@ -79,7 +92,8 @@ static uint64_t AlignUp(uint64_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
+uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                                uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2 failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
 		return 0;
@@ -133,13 +147,13 @@ uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t align
 			printf("VirtualAlloc2(alignment = 0x%016" PRIx64 ") failed: 0x%08" PRIx32 "\n",
 			       alignment, err);
 		} else {
-			return AllocAligned(address, size, mode, alignment << 1u);
+			return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
+bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(VirtualAlloc(
 	    reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	    static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE), GetProtectionFlag(mode)));
@@ -159,7 +173,7 @@ bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	return true;
 }
 
-bool Commit(uint64_t address, uint64_t size, Mode mode) {
+bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_COMMIT), GetProtectionFlag(mode)));
@@ -177,8 +191,8 @@ bool Commit(uint64_t address, uint64_t size, Mode mode) {
 	return true;
 }
 
-uint64_t Reserve(uint64_t address, uint64_t size) {
-	auto ptr = (address == 0 ? ReserveAligned(address, size, 1)
+uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
+	auto ptr = (address == 0 ? SysVirtualReserveAligned(address, size, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS)));
@@ -188,13 +202,13 @@ uint64_t Reserve(uint64_t address, uint64_t size) {
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc(MEM_RESERVE) failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return ReserveAligned(address, size, 1);
+			return SysVirtualReserveAligned(address, size, 1);
 		}
 	}
 	return ptr;
 }
 
-uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2(MEM_RESERVE) failed: 0x%08" PRIx32 "\n",
 		       static_cast<uint32_t>(GetLastError()));
@@ -248,13 +262,13 @@ uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 			       "\n",
 			       alignment, err);
 		} else {
-			return ReserveAligned(address, size, alignment << 1u);
+			return SysVirtualReserveAligned(address, size, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool ReserveFixed(uint64_t address, uint64_t size) {
+bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS));
@@ -274,7 +288,7 @@ bool ReserveFixed(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool Decommit(uint64_t address, uint64_t size) {
+bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                MEM_DECOMMIT) == 0) {
 		printf("VirtualFree(MEM_DECOMMIT) failed: 0x%08" PRIx32 "\n",
@@ -284,7 +298,7 @@ bool Decommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool Free(uint64_t address) {
+bool SysVirtualFree(uint64_t address) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), 0, MEM_RELEASE) ==
 	    0) {
 		printf("VirtualFree() failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
@@ -293,7 +307,7 @@ bool Free(uint64_t address) {
 	return true;
 }
 
-bool FreeRange(uint64_t address, uint64_t size) {
+bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 	if (address == 0 || size == 0) {
 		return false;
 	}
@@ -311,20 +325,24 @@ bool FreeRange(uint64_t address, uint64_t size) {
 		}
 		current = next;
 	}
-	return current - address == size && Free(address);
+	return current - address == size && SysVirtualFree(address);
 }
 
-bool Protect(uint64_t address, uint64_t size, Mode mode) {
+bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
+                       VirtualMemory::Mode* old_mode) {
 	DWORD old_protect = 0;
 	if (VirtualProtect(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                   GetProtectionFlag(mode), &old_protect) == 0) {
 		printf("VirtualProtect() failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
 		return false;
 	}
+	if (old_mode != nullptr) {
+		*old_mode = GetProtectionFlag(old_protect);
+	}
 	return true;
 }
 
-bool FlushInstructionCache(uint64_t address, uint64_t size) {
+bool SysVirtualFlushInstructionCache(uint64_t address, uint64_t size) {
 	if (::FlushInstructionCache(GetCurrentProcess(),
 	                            reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)),
 	                            size) == 0) {
@@ -335,6 +353,6 @@ bool FlushInstructionCache(uint64_t address, uint64_t size) {
 	return true;
 }
 
-} // namespace Common::VirtualMemory
+} // namespace Common
 
 #endif

src/common/profiler.cpp

@@ -55,13 +55,19 @@ void SetThreadName(const char* name) {
 }
 
 void Initialize() {
-	if (Config::ProfilerEnabled() && !tracy::ProfilerAvailable()) {
-		tracy::StartupProfiler();
-		TracySetProgramName("KytyPS5");
-		::printf("Tracy profiler enabled: client %d.%d.%d, protocol %u, "
-		         "broadcast %u, connect to 127.0.0.1:8086\n",
-		         tracy::Version::Major, tracy::Version::Minor, tracy::Version::Patch,
-		         tracy::ProtocolVersion, tracy::BroadcastVersion);
+	switch (Config::GetProfilerDirection()) {
+		case Config::ProfilerDirection::Network:
+			if (!tracy::ProfilerAvailable()) {
+				tracy::StartupProfiler();
+				TracySetProgramName("KytyPS5");
+				::printf("Tracy profiler enabled: client %d.%d.%d, protocol %u, "
+				         "broadcast %u, connect to 127.0.0.1:8086\n",
+				         tracy::Version::Major, tracy::Version::Minor, tracy::Version::Patch,
+				         tracy::ProtocolVersion, tracy::BroadcastVersion);
+			}
+			break;
+		case Config::ProfilerDirection::None:
+		default: break;
 	}
 }
 

src/common/stringUtils.h

@@ -1,6 +1,7 @@
 #ifndef KYTY_COMMON_STRING_UTILS_H_
 #define KYTY_COMMON_STRING_UTILS_H_
 
+#include "common/byteBuffer.h"
 #include "common/common.h"
 
 #include <algorithm>
@@ -8,6 +9,7 @@
 #include <charconv>
 #include <codecvt>
 #include <filesystem>
+#include <fmt/format.h>
 #include <locale>
 #include <string>
 #include <string_view>
@@ -16,6 +18,8 @@
 
 namespace Common {
 
+constexpr uint32_t FIND_INVALID_INDEX = static_cast<uint32_t>(-1);
+
 inline std::string PathToString(const std::filesystem::path& path) {
 #if defined(__cpp_char8_t)
 	auto u8 = path.u8string();
@@ -34,12 +38,8 @@ inline std::string PathToGenericString(const std::filesystem::path& path) {
 #endif
 }
 
-inline std::filesystem::path PathFromUtf8(std::string_view text) {
-#if defined(__cpp_char8_t)
-	return std::filesystem::path(std::u8string(text.begin(), text.end()));
-#else
-	return std::filesystem::u8path(text.begin(), text.end());
-#endif
+inline bool IsSpace(char c) {
+	return std::isspace(static_cast<unsigned char>(c)) != 0;
 }
 
 inline bool EqualNoCase(std::string_view lhs, std::string_view rhs) {
@@ -63,6 +63,65 @@ inline std::string ToLower(std::string_view text) {
 	return ret;
 }
 
+inline bool ContainsStr(std::string_view text, std::string_view needle) {
+	return text.find(needle) != std::string_view::npos;
+}
+
+inline bool StartsWith(std::string_view text, std::string_view prefix) {
+	return text.starts_with(prefix);
+}
+
+inline bool StartsWith(std::string_view text, char prefix) {
+	return !text.empty() && text.front() == prefix;
+}
+
+inline bool EndsWith(std::string_view text, std::string_view suffix) {
+	return text.ends_with(suffix);
+}
+
+inline bool EndsWith(std::string_view text, char suffix) {
+	return !text.empty() && text.back() == suffix;
+}
+
+inline bool ContainsChar(std::string_view text, char needle) {
+	return text.find(needle) != std::string_view::npos;
+}
+
+inline bool IndexValid(std::string_view text, uint32_t index) {
+	return index < text.size();
+}
+
+inline uint32_t FindIndex(std::string_view text, std::string_view needle, size_t from = 0) {
+	const auto pos = text.find(needle, from);
+	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
+}
+
+inline uint32_t FindIndex(std::string_view text, char needle, size_t from = 0) {
+	const auto pos = text.find(needle, from);
+	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
+}
+
+inline uint32_t FindLastIndex(std::string_view text, std::string_view needle,
+                              size_t from = FIND_INVALID_INDEX) {
+	if (needle.empty()) {
+		return 0;
+	}
+	if (from == FIND_INVALID_INDEX || from > text.size()) {
+		from = text.size();
+	}
+	const auto pos = text.rfind(needle, from);
+	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
+}
+
+inline uint32_t FindLastIndex(std::string_view text, char needle,
+                              size_t from = FIND_INVALID_INDEX) {
+	if (from == FIND_INVALID_INDEX || from > text.size()) {
+		from = text.size();
+	}
+	const auto pos = text.rfind(needle, from);
+	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
+}
+
 inline std::string Mid(std::string_view text, size_t first, size_t count = std::string_view::npos) {
 	if (first >= text.size()) {
 		return {};
@@ -70,18 +129,49 @@ inline std::string Mid(std::string_view text, size_t first, size_t count = std::
 	return std::string(text.substr(first, count));
 }
 
+inline std::string Left(std::string_view text, size_t count) {
+	return Mid(text, 0, count);
+}
+
+inline std::string Right(std::string_view text, size_t count) {
+	if (count >= text.size()) {
+		return std::string(text);
+	}
+	return Mid(text, text.size() - count, count);
+}
+
 inline std::string RemoveLast(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return std::string(text.substr(0, text.size() - count));
+	return Left(text, text.size() - count);
 }
 
 inline std::string RemoveFirst(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return std::string(text.substr(count));
+	return Right(text, text.size() - count);
+}
+
+inline std::string TrimRight(std::string_view text) {
+	size_t end = text.size();
+	while (end > 0 && IsSpace(text[end - 1])) {
+		end--;
+	}
+	return std::string(text.substr(0, end));
+}
+
+inline std::string TrimLeft(std::string_view text) {
+	size_t first = 0;
+	while (first < text.size() && IsSpace(text[first])) {
+		first++;
+	}
+	return std::string(text.substr(first));
+}
+
+inline std::string Trim(std::string_view text) {
+	return TrimRight(TrimLeft(text));
 }
 
 inline std::string ReplaceChar(std::string_view text, char old_char, char new_char) {
@@ -114,6 +204,16 @@ inline std::string FilenameWithoutDirectory(std::string_view text) {
 	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(pos + 1));
 }
 
+inline std::string FilenameWithoutExtension(std::string_view text) {
+	const auto pos = text.find_last_of('.');
+	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(0, pos));
+}
+
+inline std::string ExtensionWithoutFilename(std::string_view text) {
+	const auto pos = text.find_last_of('.');
+	return pos == std::string_view::npos ? std::string() : std::string(text.substr(pos));
+}
+
 inline std::string FixFilenameSlash(std::string_view text) {
 	return ReplaceChar(text, '\\', '/');
 }
@@ -157,12 +257,65 @@ inline std::vector<std::string> Split(std::string_view text, char sep, bool keep
 	return Split(text, std::string_view(&sep, 1), keep_empty);
 }
 
+inline std::string Concat(const std::vector<std::string>& list, std::string_view sep) {
+	std::string ret;
+	for (size_t i = 0; i < list.size(); i++) {
+		if (i != 0) {
+			ret += sep;
+		}
+		ret += list[i];
+	}
+	return ret;
+}
+
+inline std::string Concat(const std::vector<std::string>& list, char sep) {
+	return Concat(list, std::string_view(&sep, 1));
+}
+
+inline uint32_t ToUint32(std::string_view text, int base = 10) {
+	uint32_t value = 0;
+	std::from_chars(text.data(), text.data() + text.size(), value, base);
+	return value;
+}
+
+inline uint64_t ToUint64(std::string_view text, int base = 10) {
+	uint64_t value = 0;
+	std::from_chars(text.data(), text.data() + text.size(), value, base);
+	return value;
+}
+
 inline int32_t ToInt32(std::string_view text, int base = 10) {
 	int32_t value = 0;
 	std::from_chars(text.data(), text.data() + text.size(), value, base);
 	return value;
 }
 
+inline int64_t ToInt64(std::string_view text, int base = 10) {
+	int64_t value = 0;
+	std::from_chars(text.data(), text.data() + text.size(), value, base);
+	return value;
+}
+
+inline double ToDouble(std::string_view text) {
+	return std::stod(std::string(text));
+}
+
+inline float ToFloat(std::string_view text) {
+	return std::stof(std::string(text));
+}
+
+inline std::string SafeCsv(std::string_view text) {
+	const bool add_space =
+	    !text.empty() && (text.front() == '+' || text.front() == '=' || text.front() == '-');
+	if (text.find('"') != std::string_view::npos || text.find(';') != std::string_view::npos ||
+	    text.find('+') != std::string_view::npos || text.find('=') != std::string_view::npos ||
+	    text.find('-') != std::string_view::npos) {
+		return "\"" + std::string(add_space ? " " : "") +
+		       ReplaceStr(std::string(text), "\"", "\"\"") + "\"";
+	}
+	return std::string(text);
+}
+
 inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	if (utf16.empty()) {
 		return {};
@@ -171,6 +324,23 @@ inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
 }
 
+inline ByteBuffer HexToBin(std::string_view text) {
+	ByteBuffer ret;
+	for (size_t i = 0; i + 1 < text.size(); i += 2) {
+		ret.Add(static_cast<Byte>((ToUint32(text.substr(i, 2), 16) & 0xffu)));
+	}
+	return ret;
+}
+
+inline std::string HexFromBin(const ByteBuffer& bin) {
+	std::string ret;
+	ret.reserve(bin.Size() * 2);
+	for (uint32_t i = 0; i < bin.Size(); i++) {
+		ret += fmt::format("{:02X}", std::to_integer<uint8_t>(bin.At(i)));
+	}
+	return ret;
+}
+
 } // namespace Common
 
 #endif /* KYTY_COMMON_STRING_UTILS_H_ */

src/common/threads.cpp

@@ -8,6 +8,7 @@
 #include <chrono>             // IWYU pragma: keep
 #include <condition_variable> // IWYU pragma: keep
 #include <mutex>
+#include <vector>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS && KYTY_COMPILER == KYTY_COMPILER_CLANG
 #define KYTY_WIN_CS
@@ -196,7 +197,37 @@ struct CondVarPrivate {
 #endif
 };
 
-static wait_poll_func_t g_cond_wait_poll_callback = nullptr;
+static std::recursive_mutex                         g_cond_waiters_mutex;
+static std::vector<std::pair<int, CondVarPrivate*>> g_cond_waiters;
+static wait_poll_func_t                             g_cond_wait_poll_callback = nullptr;
+
+static void WakeCondVar(CondVarPrivate* cond_var) {
+#ifdef KYTY_WIN_CS
+	static auto func = ResolveWakeAllConditionVariable();
+	EXIT_NOT_IMPLEMENTED(func == nullptr);
+	func(&cond_var->m_cv);
+#else
+	cond_var->m_cv.notify_all();
+#endif
+}
+
+static void RegisterCondWaiter(CondVarPrivate* cond_var) {
+	std::lock_guard lock(g_cond_waiters_mutex);
+	g_cond_waiters.emplace_back(Thread::GetThreadIdUnique(), cond_var);
+}
+
+static void UnregisterCondWaiter(CondVarPrivate* cond_var) {
+	const auto      thread_id = Thread::GetThreadIdUnique();
+	std::lock_guard lock(g_cond_waiters_mutex);
+
+	const auto it = std::find_if(g_cond_waiters.begin(), g_cond_waiters.end(),
+	                             [thread_id, cond_var](const auto& waiter) {
+		                             return waiter.first == thread_id && waiter.second == cond_var;
+	                             });
+	if (it != g_cond_waiters.end()) {
+		g_cond_waiters.erase(it);
+	}
+}
 
 struct ThreadPrivate {
 	ThreadPrivate(thread_func_t f, void* a): func(f), arg(a), m_thread(&Run, this) {}
@@ -253,6 +284,10 @@ void Thread::Detach() {
 	m_thread->m_thread.detach();
 }
 
+void Thread::Sleep(uint32_t millis) {
+	std::this_thread::sleep_for(std::chrono::milliseconds(millis));
+}
+
 void Thread::SleepMicro(uint32_t micros) {
 #ifdef KYTY_WIN_CS
 	SleepHighResolution100ns(static_cast<uint64_t>(micros) * 10);
@@ -330,6 +365,7 @@ CondVar::~CondVar() {
 }
 
 void CondVar::Wait(Mutex* mutex) {
+	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -370,6 +406,7 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
+	UnregisterCondWaiter(m_cond_var.get());
 }
 
 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -378,6 +415,7 @@ void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
 
 bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	bool ok = false;
+	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -392,6 +430,7 @@ bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	      std::cv_status::no_timeout);
 	cpp_lock.release();
 #endif
+	UnregisterCondWaiter(m_cond_var.get());
 	return ok;
 }
 
@@ -406,13 +445,16 @@ void CondVar::Signal() {
 }
 
 void CondVar::SignalAll() {
-#ifdef KYTY_WIN_CS
-	static auto func = ResolveWakeAllConditionVariable();
-	EXIT_NOT_IMPLEMENTED(func == nullptr);
-	func(&m_cond_var->m_cv);
-#else
-	m_cond_var->m_cv.notify_all();
-#endif
+	WakeCondVar(m_cond_var.get());
+}
+
+void CondVar::SignalThread(int thread_id) {
+	std::lock_guard lock(g_cond_waiters_mutex);
+	for (const auto& waiter: g_cond_waiters) {
+		if (waiter.first == thread_id) {
+			WakeCondVar(waiter.second);
+		}
+	}
 }
 
 int Thread::GetThreadIdUnique() {

src/common/threads.h

@@ -31,6 +31,7 @@ public:
 	// The id is unique and can't be reused by another thread.
 	[[nodiscard]] int GetUniqueId() const;
 
+	static void Sleep(uint32_t millis);
 	static void SleepMicro(uint32_t micros);
 	static void SleepNano(uint64_t nanos);
 	static bool IsMainThread();
@@ -76,6 +77,7 @@ public:
 	void Signal();
 	void SignalAll();
 
+	static void SignalThread(int thread_id);
 	static void SetWaitPollCallback(wait_poll_func_t callback);
 
 	KYTY_CLASS_NO_COPY(CondVar);

src/common/timer.cpp

@@ -38,7 +38,16 @@ bool Timer::IsPaused() const {
 
 // return time in milliseconds
 double Timer::GetTimeMs() const {
-	return GetTimeS() * 1000.0;
+	if (m_is_paused) {
+		return 1000.0 * (static_cast<double>(m_PauseTime - m_StartTime)) /
+		       static_cast<double>(m_Frequency);
+	}
+
+	uint64_t current_time = 0;
+	SysQueryPerformanceCounter(&current_time);
+
+	return 1000.0 * (static_cast<double>(current_time - m_StartTime)) /
+	       static_cast<double>(m_Frequency);
 }
 
 // return time in seconds
@@ -53,6 +62,23 @@ double Timer::GetTimeS() const {
 	return (static_cast<double>(current_time - m_StartTime)) / static_cast<double>(m_Frequency);
 }
 
+// return time in ticks
+uint64_t Timer::GetTicks() const {
+	if (m_is_paused) {
+		return (m_PauseTime - m_StartTime);
+	}
+
+	uint64_t current_time = 0;
+	SysQueryPerformanceCounter(&current_time);
+
+	return (current_time - m_StartTime);
+}
+
+// return ticks frequency
+uint64_t Timer::GetFrequency() const {
+	return m_Frequency;
+}
+
 uint64_t Timer::QueryPerformanceFrequency() {
 	uint64_t ret = 0;
 	SysQueryPerformanceFrequency(&ret);

src/common/timer.h

@@ -24,6 +24,12 @@ public:
 	// return time in seconds
 	[[nodiscard]] double GetTimeS() const;
 
+	// return time in ticks
+	[[nodiscard]] uint64_t GetTicks() const;
+
+	// return ticks frequency
+	[[nodiscard]] uint64_t GetFrequency() const;
+
 	KYTY_CLASS_NO_COPY(Timer);
 
 	[[nodiscard]] static uint64_t QueryPerformanceFrequency();

src/common/virtualMemory.cpp

@@ -0,0 +1,63 @@
+#include "common/virtualMemory.h"
+
+#include "common/platform/sysVirtual.h"
+
+namespace Common {
+
+namespace VirtualMemory {
+
+void Init() {
+	SysVirtualInit();
+}
+
+uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
+	return SysVirtualAlloc(address, size, mode);
+}
+
+uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
+	return SysVirtualAllocAligned(address, size, mode, alignment);
+}
+
+bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
+	return SysVirtualAllocFixed(address, size, mode);
+}
+
+bool Commit(uint64_t address, uint64_t size, Mode mode) {
+	return SysVirtualCommit(address, size, mode);
+}
+
+uint64_t Reserve(uint64_t address, uint64_t size) {
+	return SysVirtualReserve(address, size);
+}
+
+uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+	return SysVirtualReserveAligned(address, size, alignment);
+}
+
+bool ReserveFixed(uint64_t address, uint64_t size) {
+	return SysVirtualReserveFixed(address, size);
+}
+
+bool Decommit(uint64_t address, uint64_t size) {
+	return SysVirtualDecommit(address, size);
+}
+
+bool Free(uint64_t address) {
+	return SysVirtualFree(address);
+}
+
+bool FreeRange(uint64_t address, uint64_t size) {
+	return SysVirtualFreeRange(address, size);
+}
+
+bool Protect(uint64_t address, uint64_t size, Mode mode, Mode* old_mode) {
+	return SysVirtualProtect(address, size, mode, old_mode);
+}
+
+bool FlushInstructionCache(uint64_t address, uint64_t size) {
+	return SysVirtualFlushInstructionCache(address, size);
+}
+
+} // namespace VirtualMemory
+
+} // namespace Common

src/common/virtualMemory.h

@@ -35,7 +35,7 @@ bool     ReserveFixed(uint64_t address, uint64_t size);
 bool     Decommit(uint64_t address, uint64_t size);
 bool     Free(uint64_t address);
 bool     FreeRange(uint64_t address, uint64_t size);
-bool     Protect(uint64_t address, uint64_t size, Mode mode);
+bool     Protect(uint64_t address, uint64_t size, Mode mode, Mode* old_mode = nullptr);
 bool     FlushInstructionCache(uint64_t address, uint64_t size);
 
 } // namespace VirtualMemory

src/emulator.cpp

@@ -196,10 +196,6 @@ void Run(const RunOptions& options) {
 	ClearDebugTextureFolder();
 
 	PrintSystemInfo();
-	std::string title_id;
-	if (Loader::SystemContentParamSfoGetString("TITLE_ID", &title_id) && !title_id.empty()) {
-		Log::WriteToConsoleAndLog(fmt::format("Title ID: {}\n", title_id));
-	}
 
 	int ok = atexit(KytyClose);
 	EXIT_NOT_IMPLEMENTED(ok != 0);

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -32,12 +32,11 @@ private:
 
 	struct BufferCursor {
 		std::span<const uint32_t> commands;
-		uint32_t                  offset_dw = 0;
+		uint32_t                  offset_dw           = 0;
+		uint32_t                  deferred_advance_dw = 0;
 	};
 
 	std::vector<BufferCursor> m_buffer_stack;
-	std::span<const uint32_t> m_next_buffer;
-	bool                      m_chain         = false;
 	bool                      m_suspended     = false;
 	bool                      m_made_progress = false;
 };
@@ -73,9 +72,6 @@ public:
 	void SetIndexBufferSize(uint32_t index_buffer_size);
 	void SetDrawIndirectArgsBaseAddress(uint64_t draw_indirect_args_base_addr);
 	void SetDispatchIndirectArgsBaseAddress(uint64_t dispatch_indirect_args_base_addr);
-	[[nodiscard]] uint64_t GetDispatchIndirectArgsBaseAddress() const {
-		return m_dispatch_indirect_args_base_addr;
-	}
 	void SetNumInstances(uint32_t num_instances);
 	void DrawIndex(DrawIndexArgs args);
 	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count);
@@ -102,7 +98,7 @@ public:
 	void TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id);
 	void DispatchDirect(uint32_t thread_group_x, uint32_t thread_group_y, uint32_t thread_group_z,
 	                    uint32_t mode);
-	void DispatchIndirect(uint64_t args_addr, uint32_t mode);
+	void DispatchIndirect(uint32_t data_offset, uint32_t mode);
 	void WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index);
 	void TriggerEvent(uint32_t event_type, uint32_t event_index, uint64_t event_address = 0);
 
@@ -133,7 +129,7 @@ public:
 	[[nodiscard]] bool ShouldSkipPredicatedPackets() const { return m_predicate_skip; }
 
 	Pm4ProcessResult Process(Pm4Execution& execution, std::span<const uint32_t> commands);
-	void             ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain);
+	void             ProcessIndirectBuffer(std::span<const uint32_t> commands);
 
 	void SetFlip(const FlipInfo& flip) { m_flip = flip; }
 
@@ -147,10 +143,12 @@ private:
 	                      uint32_t cache_action, uint32_t event_index, uint32_t event_write_source,
 	                      void* dst_gpu_addr, T value, uint32_t interrupt_selector,
 	                      uint32_t interrupt_context_id);
-	void ProcessPm4(Pm4Execution& execution);
+	void ProcessPm4(Pm4Execution& execution, size_t stop_depth);
 	void SuspendPm4();
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }
+	void                CheckBuffer() const { GetScheduler().CheckActive(); }
+	GpuResourceManager& GetGpuResources() const { return m_renderer.GetGpuResources(); }
 
 	RenderContext&   m_renderer;
 	HW::Context      m_ctx;

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -1331,13 +1331,26 @@ KYTY_CP_OP_PARSER(CpOpDispatchIndirect) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0011600 && cmd_id != 0xc0021600);
 
 	if (cmd_id == 0xc0021600) {
-		cp.DispatchIndirect(buffer[0] | (static_cast<uint64_t>(buffer[1]) << 32u), buffer[2]);
+		struct DispatchIndirectArgs {
+			uint32_t thread_group_x;
+			uint32_t thread_group_y;
+			uint32_t thread_group_z;
+		};
+
+		auto* args = reinterpret_cast<const DispatchIndirectArgs*>(
+		    buffer[0] | (static_cast<uint64_t>(buffer[1]) << 32u));
+		uint32_t mode = buffer[2];
+
+		EXIT_NOT_IMPLEMENTED(args == nullptr);
+		cp.DispatchDirect(args->thread_group_x, args->thread_group_y, args->thread_group_z, mode);
+
 		return 3;
 	}
 
-	const auto base_addr = cp.GetDispatchIndirectArgsBaseAddress();
-	EXIT_NOT_IMPLEMENTED(base_addr == 0);
-	cp.DispatchIndirect(base_addr + buffer[0], buffer[1]);
+	uint32_t data_offset = buffer[0];
+	uint32_t mode        = buffer[1];
+
+	cp.DispatchIndirect(data_offset, mode);
 
 	return 2;
 }
@@ -1482,10 +1495,10 @@ KYTY_CP_OP_PARSER(CpOpBranch) {
 	     reinterpret_cast<uint64_t>(else_buffer), else_num_dw);
 
 	if (take_then) {
-		cp.ProcessIndirectBuffer({then_buffer, then_num_dw}, true);
+		cp.ProcessIndirectBuffer({then_buffer, then_num_dw});
 	} else if (mode == 2 && else_num_dw != 0) {
 		EXIT_NOT_IMPLEMENTED(else_buffer == nullptr);
-		cp.ProcessIndirectBuffer({else_buffer, else_num_dw}, true);
+		cp.ProcessIndirectBuffer({else_buffer, else_num_dw});
 	}
 
 	return payload_dw;
@@ -1946,7 +1959,7 @@ KYTY_CP_OP_PARSER(CpOpIndirectBuffer) {
 
 	GraphicsDbgDumpDcb("ci", indirect_num_dw, indirect_buffer);
 
-	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw}, (control & (1u << 20u)) != 0);
+	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw});
 
 	return 3;
 }
@@ -2004,14 +2017,6 @@ KYTY_CP_OP_PARSER(CpOpIndirectCxRegs) {
 		auto pfunc = g_hw_ctx_indirect_func[cmd_offset & (Pm4::CX_NUM - 1)];
 
 		if (pfunc == nullptr) {
-			if (raw_cmd_offset == 0x24au && value == 0u) {
-				static std::atomic_flag logged = ATOMIC_FLAG_INIT;
-				if (!logged.test_and_set(std::memory_order_relaxed)) {
-					Log::WriteToConsoleAndLog(
-					    "\t diagnostic: ignoring indirect CX {0x24a, 0}; hardware effect unresolved\n");
-				}
-				continue;
-			}
 			EXIT("unknown cx reg at %05" PRIx32 ": 0x%" PRIx32 "\n", num_dw - dw, cmd_offset);
 		}
 
@@ -3630,11 +3635,6 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC4_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	// Toolkit state restoration emits the compiler's GS-front allocation
-	// metadata here. Native GS resource registers already carry that allocation.
-	g_hw_sh_indirect_func[0x0ca] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
@@ -3654,13 +3654,12 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS;
-	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS; offset++) {
-		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-			cp.GetShCtx().SetHsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS,
-			                                   value);
-		};
-	}
+	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
+	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		auto base = cp.GetShCtx().GetVs().hs_regs.data_addr;
 		base &= 0xFFFFFF00000000FFull;
@@ -3819,25 +3818,6 @@ void GraphicsInitJmpTablesUcIndirect() {
 	for (auto& func: g_hw_uc_indirect_func) {
 		func = nullptr;
 	}
-	for (uint32_t i = 0; i < 4; ++i) {
-		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-			cp.GetUcfg().SetFsrControlPoint(0, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_X, value);
-		};
-		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-			cp.GetUcfg().SetFsrControlPoint(1, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_Y, value);
-		};
-	}
-	for (uint32_t i = 0; i < 2; ++i) {
-		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-			cp.GetUcfg().SetFsrAlpha(0, cmd_offset - Pm4::FSR_ALPHA_LEFT_X, value);
-		};
-		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-			cp.GetUcfg().SetFsrAlpha(1, cmd_offset - Pm4::FSR_ALPHA_LEFT_Y, value);
-		};
-		g_hw_uc_indirect_func[Pm4::FSR_WINDOW_LEFT + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-			cp.GetUcfg().SetFsrWindow(cmd_offset - Pm4::FSR_WINDOW_LEFT, value);
-		};
-	}
 
 	g_hw_uc_indirect_func[Pm4::GE_CNTL] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeControl r;

src/graphics/guest_gpu/gpu_defs.h

@@ -26,10 +26,6 @@ enum class PrimitiveType : uint32_t {
 	kPolygon            = 21,
 };
 
-constexpr bool IsRectList(PrimitiveType type) {
-	return type == PrimitiveType::kRectList || type == PrimitiveType::kRectListLegacy;
-}
-
 enum class IndexType : uint32_t {
 	kIndex16 = 0,
 	kIndex32 = 1,

src/graphics/guest_gpu/graphicsRun.cpp

@@ -77,7 +77,7 @@ private:
 
 static bool GraphicsRunDebugDumpEnabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
-	       Config::GetPrintfDirection() != Config::LogDirection::Silent;
+	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
 }
 
 GuestGpu::GuestGpu(RenderContext& renderer): m_renderer(renderer) {
@@ -386,12 +386,8 @@ void CommandProcessor::WriteReferenceClock(uint64_t dst_address, uint32_t num_by
 	}
 	const auto value = Sync::ReadReferenceClock();
 	std::memcpy(reinterpret_cast<void*>(dst_address), &value, num_bytes);
-	static std::atomic<uint32_t> clock_log_count {0};
-	if (clock_log_count.fetch_add(1) < 64) {
-		LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64
-		     " size=%u\n",
-		     dst_address, value, num_bytes);
-	}
+	LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64 " size=%u\n",
+	     dst_address, value, num_bytes);
 }
 
 void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cache_policy,
@@ -431,7 +427,7 @@ void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cach
 	if (!decode_gds(dst_sel, dst_gds)) {
 		EXIT("unsupported dmaData destination selector 0x%02" PRIx8 "\n", dst_sel);
 	}
-	auto& buffer_cache = m_renderer.GetBufferCache();
+	auto& buffer_cache = GetGpuResources().GetBufferCache();
 	if (src_sel == 2) {
 		buffer_cache.FillBuffer(
 		    dst_address_or_offset, num_bytes,
@@ -607,11 +603,11 @@ bool GuestGpu::Process(Submission& submission) {
 			}
 			if (progressed) {
 				if (complete) {
-					m_renderer.RunGarbageCollector();
+					m_renderer.GetGpuResources().RunGarbageCollector();
 				}
 				cp.BufferFlush();
 			} else if (complete) {
-				m_renderer.RunGarbageCollector();
+				m_renderer.GetGpuResources().RunGarbageCollector();
 			}
 			break;
 		}
@@ -633,16 +629,16 @@ bool GuestGpu::Process(Submission& submission) {
 			           Pm4ProcessResult::Complete;
 			if (submission.command_execution.MadeProgress()) {
 				if (complete) {
-					m_renderer.RunGarbageCollector();
+					m_renderer.GetGpuResources().RunGarbageCollector();
 				}
 				cp.BufferFlush();
 			} else if (complete) {
-				m_renderer.RunGarbageCollector();
+				m_renderer.GetGpuResources().RunGarbageCollector();
 			}
 			break;
 		}
 		case SubmissionType::FlipPreparation:
-			m_renderer.RunGarbageCollector();
+			m_renderer.GetGpuResources().RunGarbageCollector();
 			cp.PrepareCpuFlip(submission.flip_request_id);
 			break;
 	}
@@ -676,16 +672,20 @@ Pm4ProcessResult CommandProcessor::Process(Pm4Execution&             execution,
 		Pm4Execution*     previous_execution;
 	} execution_scope(*this, execution);
 
-	ProcessPm4(execution);
+	ProcessPm4(execution, 0);
 	return execution.m_buffer_stack.empty() ? Pm4ProcessResult::Complete
 	                                        : Pm4ProcessResult::Blocked;
 }
 
-void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain) {
+void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands) {
 	EXIT_IF(g_current_execution == nullptr);
-	EXIT_IF(!g_current_execution->m_next_buffer.empty());
-	g_current_execution->m_next_buffer = commands;
-	g_current_execution->m_chain       = chain;
+	if (commands.empty()) {
+		return;
+	}
+	auto&      execution  = *g_current_execution;
+	const auto stop_depth = execution.m_buffer_stack.size();
+	execution.m_buffer_stack.push_back({commands});
+	ProcessPm4(execution, stop_depth);
 }
 
 void CommandProcessor::SuspendPm4() {
@@ -693,13 +693,21 @@ void CommandProcessor::SuspendPm4() {
 	g_current_execution->m_suspended = true;
 }
 
-void CommandProcessor::ProcessPm4(Pm4Execution& execution) {
-	while (!execution.m_buffer_stack.empty()) {
+void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
+	while (execution.m_buffer_stack.size() > stop_depth) {
 		if (g_gpu_state != nullptr) {
 			g_gpu_state->ProcessCommands();
 		}
-		auto& cursor = execution.m_buffer_stack.back();
+		const auto buffer_index = execution.m_buffer_stack.size() - 1;
+		auto&      cursor       = execution.m_buffer_stack[buffer_index];
 		EXIT_IF(cursor.offset_dw > cursor.commands.size());
+		if (cursor.deferred_advance_dw != 0) {
+			EXIT_IF(cursor.deferred_advance_dw > cursor.commands.size() - cursor.offset_dw);
+			cursor.offset_dw += cursor.deferred_advance_dw;
+			cursor.deferred_advance_dw = 0;
+			execution.m_made_progress  = true;
+			continue;
+		}
 		if (cursor.offset_dw == cursor.commands.size()) {
 			execution.m_buffer_stack.pop_back();
 			continue;
@@ -775,19 +783,14 @@ void CommandProcessor::ProcessPm4(Pm4Execution& execution) {
 		    handler(*this, packet_header & ~1u, packet + 1, remaining_dw, total_dw) + 1;
 		EXIT_IF(packet_dw > remaining_dw);
 		if (execution.m_suspended) {
+			if (execution.m_buffer_stack.size() > buffer_index + 1) {
+				execution.m_buffer_stack[buffer_index].deferred_advance_dw = packet_dw;
+			}
 			return;
 		}
-		cursor.offset_dw += packet_dw;
+		EXIT_IF(execution.m_buffer_stack.size() != buffer_index + 1);
+		execution.m_buffer_stack[buffer_index].offset_dw += packet_dw;
 		execution.m_made_progress = true;
-		if (!execution.m_next_buffer.empty()) {
-			// Chains and taken branches reuse the fetcher; only calls retain a return cursor.
-			if (execution.m_chain) {
-				cursor = {execution.m_next_buffer};
-			} else {
-				execution.m_buffer_stack.push_back({execution.m_next_buffer});
-			}
-			execution.m_next_buffer = {};
-		}
 	}
 }
 
@@ -822,63 +825,46 @@ void CommandProcessor::SetNumInstances(uint32_t num_instances) {
 
 void CommandProcessor::SetPredication(uint32_t condition, uint32_t op, uint32_t wait_op,
                                       const volatile void* address, uint32_t count_in_dwords) {
+	if (wait_op != 0) {
+		BufferFlushAndWait();
+	}
+
 	(void)count_in_dwords;
-	uint64_t value = 0;
 
 	switch (op) {
-		case 0x00:
+		case 0x00: {
 			m_predicate_skip = false;
-			return;
-		case 0x01: {
+		} break;
+		case 0x03: {
 			EXIT_NOT_IMPLEMENTED(address == nullptr);
-			// One begin/end pair per DB; bit 63 marks each counter ready.
-			constexpr uint64_t ready_bit = 1ull << 63u;
-			const auto* results = reinterpret_cast<const volatile uint64_t*>(address);
-			for (uint32_t db = 0; db < 16u; db++) {
-				const auto begin = results[db * 2u];
-				const auto end   = results[db * 2u + 1u];
-				if ((begin & end & ready_bit) == 0) {
-					if (wait_op == 0) {
-						SuspendPm4();
-					} else {
-						m_predicate_skip = false;
-					}
-					return;
-				}
-				value += end - begin;
+
+			auto value = *reinterpret_cast<const volatile uint64_t*>(address);
+
+			switch (condition) {
+				case 0x00: m_predicate_skip = (value != 0); break;
+				case 0x01: m_predicate_skip = (value == 0); break;
+				default: EXIT("unknown predication condition: 0x%08" PRIx32 "\n", condition);
+			}
+			static std::atomic<uint32_t> log_count {0};
+			if (log_count.fetch_add(1) < 128) {
+				LOGF("\t bool predication: addr=0x%016" PRIx64 ", value=0x%016" PRIx64
+				     ", condition=%" PRIu32 ", skip=%u, wait_op=%" PRIu32 "\n",
+				     reinterpret_cast<uint64_t>(address), value, condition,
+				     m_predicate_skip ? 1u : 0u, wait_op);
 			}
 		} break;
-		case 0x03:
-			if (wait_op != 0) {
-				BufferFlushAndWait();
-			}
-			EXIT_NOT_IMPLEMENTED(address == nullptr);
-			value = *reinterpret_cast<const volatile uint64_t*>(address);
-			break;
 		default: EXIT("unknown predication op: 0x%08" PRIx32 "\n", op);
 	}
-	switch (condition) {
-		case 0x00: m_predicate_skip = (value != 0); break;
-		case 0x01: m_predicate_skip = (value == 0); break;
-		default: EXIT("unknown predication condition: 0x%08" PRIx32 "\n", condition);
-	}
-	if (op == 0x03) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1) < 128) {
-			LOGF("\t bool predication: addr=0x%016" PRIx64 ", value=0x%016" PRIx64
-			     ", condition=%" PRIu32 ", skip=%u, wait_op=%" PRIu32 "\n",
-			     reinterpret_cast<uint64_t>(address), value, condition,
-			     m_predicate_skip ? 1u : 0u, wait_op);
-		}
-	}
 }
 
 void CommandProcessor::DrawIndex(DrawIndexArgs args) {
+	CheckBuffer();
+
 	args.index_type_and_size = m_index_type_and_size;
 	if (args.instance_count == 0) {
 		args.instance_count = m_num_instances;
 	}
-	if (GraphicsRunDebugDumpEnabled() && (args.base_vertex != 0 || args.first_instance != 0)) {
+	if (args.base_vertex != 0 || args.first_instance != 0) {
 		LOGF("\t draw indexed offsets: base_vertex = %" PRId32 ", first_instance = %" PRIu32 "\n",
 		     args.base_vertex, args.first_instance);
 	}
@@ -910,6 +896,17 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 	if (!indexed) {
 		DrawIndirectArgs args {};
 		std::memcpy(&args, args_addr, sizeof(args));
+		if (args.instance_count != 1u || args.start_vertex_location != 0u ||
+		    args.start_instance_location != 0u) {
+			static std::atomic<uint32_t> log_count {0};
+			if (log_count.fetch_add(1) < 64) {
+				LOGF("\t warning: partial DrawIndirect args: vertex_count=%" PRIu32
+				     ", instance_count=%" PRIu32 ", start_vertex=%" PRIu32
+				     ", start_instance=%" PRIu32 "\n",
+				     args.vertex_count_per_instance, args.instance_count,
+				     args.start_vertex_location, args.start_instance_location);
+			}
+		}
 		m_num_instances = args.instance_count;
 		DrawIndexAuto({.vertex_count   = args.vertex_count_per_instance,
 		               .instance_count = args.instance_count,
@@ -921,6 +918,16 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 
 	DrawIndexedIndirectArgs args {};
 	std::memcpy(&args, args_addr, sizeof(args));
+	if (args.base_vertex_location != 0u || args.start_instance_location != 0u) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1) < 64) {
+			LOGF("\t warning: partial DrawIndexIndirect args: index_count=%" PRIu32
+			     ", instance_count=%" PRIu32 ", start_index=%" PRIu32 ", base_vertex=%" PRIu32
+			     ", start_instance=%" PRIu32 "\n",
+			     args.index_count_per_instance, args.instance_count, args.start_index_location,
+			     args.base_vertex_location, args.start_instance_location);
+		}
+	}
 
 	uint64_t index_size = 0;
 	switch (m_index_type_and_size) {
@@ -976,22 +983,23 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 	const auto args_size = indexed ? sizeof(DrawIndexedIndirectArgs) : sizeof(DrawIndirectArgs);
 	EXIT_NOT_IMPLEMENTED(stride_in_bytes < args_size);
 
-	uint64_t index_size = 0;
-	if (indexed) {
-		switch (m_index_type_and_size) {
-			case 0: index_size = 2; break;
-			case 1: index_size = 4; break;
-			case 2: index_size = 1; break;
-			default: EXIT("unknown index_type_and_size: %u\n", m_index_type_and_size);
-		}
-	}
-
 	for (uint32_t i = 0; i < draw_count; i++) {
 		const auto args_addr = m_draw_indirect_args_base_addr + data_offset +
 		                       static_cast<uint64_t>(i) * stride_in_bytes;
 
 		if (!indexed) {
 			auto* args = reinterpret_cast<const DrawIndirectArgs*>(args_addr);
+			if (args->instance_count != 1u || args->start_vertex_location != 0u ||
+			    args->start_instance_location != 0u) {
+				static std::atomic<uint32_t> log_count {0};
+				if (log_count.fetch_add(1) < 64) {
+					LOGF("\t warning: partial DrawIndirectMulti args[%u]: vertex_count=%" PRIu32
+					     ", instance_count=%" PRIu32 ", start_vertex=%" PRIu32
+					     ", start_instance=%" PRIu32 "\n",
+					     i, args->vertex_count_per_instance, args->instance_count,
+					     args->start_vertex_location, args->start_instance_location);
+				}
+			}
 			m_num_instances = args->instance_count;
 			DrawIndexAuto({.vertex_count   = args->vertex_count_per_instance,
 			               .instance_count = args->instance_count,
@@ -1002,6 +1010,25 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 		}
 
 		auto* args = reinterpret_cast<const DrawIndexedIndirectArgs*>(args_addr);
+		if (args->base_vertex_location != 0u || args->start_instance_location != 0u) {
+			static std::atomic<uint32_t> log_count {0};
+			if (log_count.fetch_add(1) < 64) {
+				LOGF("\t warning: partial DrawIndexIndirectMulti args[%u]: index_count=%" PRIu32
+				     ", instance_count=%" PRIu32 ", start_index=%" PRIu32 ", base_vertex=%" PRIu32
+				     ", start_instance=%" PRIu32 "\n",
+				     i, args->index_count_per_instance, args->instance_count,
+				     args->start_index_location, args->base_vertex_location,
+				     args->start_instance_location);
+			}
+		}
+
+		uint64_t index_size = 0;
+		switch (m_index_type_and_size) {
+			case 0: index_size = 2; break;
+			case 1: index_size = 4; break;
+			case 2: index_size = 1; break;
+			default: EXIT("unknown index_type_and_size: %u\n", m_index_type_and_size);
+		}
 
 		auto* index_addr = reinterpret_cast<const void*>(
 		    m_index_base_addr + static_cast<uint64_t>(args->start_index_location) * index_size);
@@ -1039,6 +1066,7 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 	// uint32_t local_z   = 1;
 
 	{
+		CheckBuffer();
 		frame_num = m_renderer.GetGpu().GetFrameNum();
 		if (GraphicsRunDebugDumpEnabled()) {
 			static std::atomic<uint32_t> log_count {0};
@@ -1062,6 +1090,17 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		// local_x        = std::max(cs.num_thread_x, 1u);
 		// local_y        = std::max(cs.num_thread_y, 1u);
 		// local_z        = std::max(cs.num_thread_z, 1u);
+		if (cs.wave_size == 64u) {
+			static std::atomic_bool logged_wave64_shader {false};
+			if (!logged_wave64_shader.exchange(true, std::memory_order_relaxed)) {
+				LOGF("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
+				     cs.data_addr);
+				std::printf("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
+				            cs.data_addr);
+				std::fflush(stdout);
+			}
+		}
+
 		m_renderer.GetRenderExecutor().DispatchDirect(m_submit_id, CurrentBuffer(), thread_group_x,
 		                                              thread_group_y, thread_group_z, mode);
 	}
@@ -1089,18 +1128,24 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 	}*/
 }
 
-void CommandProcessor::DispatchIndirect(uint64_t args_addr, uint32_t mode) {
-	EXIT_NOT_IMPLEMENTED(args_addr == 0 || (args_addr & 3u) != 0);
-	if ((mode & Pm4::COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS) != 0) {
-		const auto* args = reinterpret_cast<const vk::DispatchIndirectCommand*>(args_addr);
-		DispatchDirect(args->x, args->y, args->z, mode);
-		return;
-	}
-	m_sh_ctx.SetCsWaveSize(Pm4::ComputeWaveSize(mode));
-	m_renderer.GetRenderExecutor().DispatchIndirect(m_submit_id, CurrentBuffer(), args_addr, mode);
+void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
+	struct DispatchIndirectArgs {
+		uint32_t thread_group_x;
+		uint32_t thread_group_y;
+		uint32_t thread_group_z;
+	};
+
+	EXIT_NOT_IMPLEMENTED(m_dispatch_indirect_args_base_addr == 0);
+
+	const auto args_addr = m_dispatch_indirect_args_base_addr + data_offset;
+	auto*      args      = reinterpret_cast<const DispatchIndirectArgs*>(args_addr);
+
+	DispatchDirect(args->thread_group_x, args->thread_group_y, args->thread_group_z, mode);
 }
 
 void CommandProcessor::DrawIndexAuto(DrawAutoArgs args) {
+	CheckBuffer();
+
 	if (args.instance_count == 0) {
 		args.instance_count = m_num_instances;
 	}
@@ -1122,7 +1167,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
                                         uint32_t interrupt_context_id) {
 	static_assert(sizeof(T) == sizeof(uint32_t) || sizeof(T) == sizeof(uint64_t));
 
-	auto& command = CurrentBuffer();
+	CheckBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		const auto bits      = static_cast<unsigned>(sizeof(T) * 8u);
@@ -1153,7 +1198,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 		case 0x03: with_interrupt = false; break;
 		case 0x01:
 			if (!IsAsyncComputeQueue()) {
-				Sync::TriggerEopEventAtEndOfPipe(command, m_interrupt_event_id,
+				Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id,
 				                                 interrupt_context_id);
 				return;
 			}
@@ -1170,17 +1215,17 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 
 		if (with_interrupt) {
 			if (with_writeback) {
-				Sync::WriteAtEndOfPipeWithInterruptWriteBack32(m_submit_id, command, dst,
+				Sync::WriteAtEndOfPipeWithInterruptWriteBack32(m_submit_id, CurrentBuffer(), dst,
 				                                               data, m_interrupt_event_id,
 				                                               interrupt_context_id);
 			} else {
-				Sync::WriteAtEndOfPipeWithInterrupt32(m_submit_id, command, dst, data,
+				Sync::WriteAtEndOfPipeWithInterrupt32(m_submit_id, CurrentBuffer(), dst, data,
 				                                      m_interrupt_event_id, interrupt_context_id);
 			}
 		} else if (with_writeback) {
-			Sync::WriteAtEndOfPipeWithWriteBack32(m_submit_id, command, dst, data);
+			Sync::WriteAtEndOfPipeWithWriteBack32(m_submit_id, CurrentBuffer(), dst, data);
 		} else {
-			Sync::WriteAtEndOfPipe32(m_submit_id, command, dst, data);
+			Sync::WriteAtEndOfPipe32(m_submit_id, CurrentBuffer(), dst, data);
 		}
 	};
 
@@ -1192,7 +1237,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					SynchronizeGpu();
 					Sync::ReadGds(*m_renderer.GetBufferCache().GetGdsBuffer(), dst, value & 0xffffu,
 					              value >> 16u);
-					Sync::WriteAtEndOfPipeGds32(m_submit_id, command, dst, value & 0xffffu,
+					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
 					                            value >> 16u);
 					if (with_interrupt) {
 						m_renderer.TriggerInterrupt(m_interrupt_event_id, interrupt_context_id);
@@ -1205,9 +1250,8 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 			}
 			break;
 		case 0x02:
-		case 0x04:
 			if constexpr (sizeof(T) == sizeof(uint32_t)) {
-				if (event_write_source == 0x02 && eop_event_type == 0x2f && event_index == 0x06) {
+				if (eop_event_type == 0x2f && event_index == 0x06) {
 					switch (cache_action) {
 						case 0x00: write32(false); return;
 						case 0x38: write32(true); return;
@@ -1215,9 +1259,6 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					}
 				}
 			} else {
-				if (event_write_source == 0x04) {
-					value = Sync::ReadReferenceClock();
-				}
 				auto write64 = [&](bool with_writeback) {
 					auto* dst = static_cast<uint64_t*>(dst_gpu_addr);
 					std::memcpy(dst, &value, sizeof(value));
@@ -1225,18 +1266,18 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					if (with_interrupt) {
 						if (with_writeback) {
 							Sync::WriteAtEndOfPipeWithInterruptWriteBack64(
-							    m_submit_id, command, dst, value, m_interrupt_event_id,
+							    m_submit_id, CurrentBuffer(), dst, value, m_interrupt_event_id,
 							    interrupt_context_id);
 						} else {
-							Sync::WriteAtEndOfPipeWithInterrupt64(m_submit_id, command, dst,
+							Sync::WriteAtEndOfPipeWithInterrupt64(m_submit_id, CurrentBuffer(), dst,
 							                                      value, m_interrupt_event_id,
 							                                      interrupt_context_id);
 						}
 					} else if (with_writeback) {
-						Sync::WriteAtEndOfPipeWithWriteBack64(m_submit_id, command, dst,
+						Sync::WriteAtEndOfPipeWithWriteBack64(m_submit_id, CurrentBuffer(), dst,
 						                                      value);
 					} else {
-						Sync::WriteAtEndOfPipe64(m_submit_id, command, dst, value);
+						Sync::WriteAtEndOfPipe64(m_submit_id, CurrentBuffer(), dst, value);
 					}
 				};
 
@@ -1251,7 +1292,6 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x14:
 							case 0x28:
-							case 0x2f:
 								if (event_index == 0x00) {
 									write64(false);
 									return;
@@ -1259,6 +1299,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x2b:
 							case 0x2d:
+							case 0x2f:
 							case 0x30:
 								if (event_index == 0x00 && !with_interrupt) {
 									write64(false);
@@ -1274,7 +1315,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 							case 0x14:
 							case 0x28:
 								if (((eop_event_type == 0x04 || eop_event_type == 0x28) &&
-								     event_index == 0x05) ||
+								     event_index == 0x05 && !with_interrupt) ||
 								    (event_index == 0x00)) {
 									write64(true);
 									return;
@@ -1306,6 +1347,45 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 				}
 			}
 			break;
+		case 0x04:
+			if constexpr (sizeof(T) == sizeof(uint64_t)) {
+				const auto clock = Sync::ReadReferenceClock();
+				auto*      dst   = static_cast<uint64_t*>(dst_gpu_addr);
+				std::memcpy(dst, &clock, sizeof(clock));
+				switch (cache_action) {
+					case 0x00:
+						if ((eop_event_type == 0x04 && event_index == 0x05) ||
+						    (eop_event_type == 0x28 && event_index == 0x00)) {
+							if (with_interrupt) {
+								Sync::WriteAtEndOfPipeWithInterrupt64(
+								    m_submit_id, CurrentBuffer(), dst, clock, m_interrupt_event_id,
+								    interrupt_context_id);
+							} else {
+								Sync::WriteAtEndOfPipeClockCounter(m_submit_id, CurrentBuffer(),
+								                                   dst, clock);
+							}
+							return;
+						}
+						break;
+					case 0x38:
+						if ((eop_event_type == 0x04 &&
+						     (event_index == 0x00 || event_index == 0x05)) ||
+						    (eop_event_type == 0x28 && event_index == 0x00)) {
+							if (with_interrupt) {
+								Sync::WriteAtEndOfPipeWithInterruptWriteBack64(
+								    m_submit_id, CurrentBuffer(), dst, clock, m_interrupt_event_id,
+								    interrupt_context_id);
+							} else {
+								Sync::WriteAtEndOfPipeClockCounterWithWriteBack(
+								    m_submit_id, CurrentBuffer(), dst, clock);
+							}
+							return;
+						}
+						break;
+					default: break;
+				}
+			}
+			break;
 		default: break;
 	}
 
@@ -1335,6 +1415,8 @@ void CommandProcessor::WriteAtEndOfPipe64(uint32_t cache_policy, uint32_t event_
 }
 
 void CommandProcessor::EmitGlobalBarrier() {
+	CheckBuffer();
+
 	Common::LockGuard lock(m_renderer.GetMutex());
 
 	vk::MemoryBarrier2 barrier {};
@@ -1351,6 +1433,8 @@ void CommandProcessor::EmitGlobalBarrier() {
 }
 
 void CommandProcessor::TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id) {
+	CheckBuffer();
+
 	Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id, interrupt_context_id);
 }
 
@@ -1434,6 +1518,8 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index,
 }
 
 void CommandProcessor::Flip() {
+	CheckBuffer();
+
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n");
 	}
@@ -1447,7 +1533,7 @@ void CommandProcessor::Flip() {
 }
 
 void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
-	auto& command = CurrentBuffer();
+	CheckBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n"
@@ -1457,6 +1543,7 @@ void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
 	}
 
 	std::memcpy(dst_gpu_addr, &value, sizeof(value));
+	auto& command = CurrentBuffer();
 	auto request = Sync::PrepareVideoOutFlip(command, m_flip.handle, m_flip.index, m_flip.flip_mode,
 	                                         m_flip.flip_arg);
 	Sync::WriteAtEndOfPipeWithFlip32(m_submit_id, command, static_cast<uint32_t*>(dst_gpu_addr),
@@ -1467,7 +1554,7 @@ void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
 
 void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache_action,
                                          void* dst_gpu_addr, uint32_t value) {
-	auto& command = CurrentBuffer();
+	CheckBuffer();
 
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::FlipWithInterrupt()\n"
@@ -1482,6 +1569,7 @@ void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache
 		EXIT("unknown event type\n");
 	}
 	std::memcpy(dst_gpu_addr, &value, sizeof(value));
+	auto& command = CurrentBuffer();
 	auto request = Sync::PrepareVideoOutFlip(command, m_flip.handle, m_flip.index, m_flip.flip_mode,
 	                                         m_flip.flip_arg);
 	Sync::WriteAtEndOfPipeWithInterruptWriteBackFlip32(
@@ -1491,7 +1579,7 @@ void CommandProcessor::FlipWithInterrupt(uint32_t eop_event_type, uint32_t cache
 }
 
 void CommandProcessor::PrepareCpuFlip(uint64_t request_id) {
-	auto& command = CurrentBuffer();
+	CheckBuffer();
 	if (g_current_processor != nullptr) {
 		EXIT("invalid graphics-thread CPU flip preparation\n");
 	}
@@ -1501,10 +1589,9 @@ void CommandProcessor::PrepareCpuFlip(uint64_t request_id) {
 	};
 	ProcessorScope processor_scope(*this);
 
-	m_renderer.GetVideoOut().PrepareFlip(request_id, command);
-	GetScheduler().DeferPriorityOperation(
-	    [this, request_id] { m_renderer.GetVideoOut().CompleteFlip(request_id); });
+	m_renderer.GetVideoOut().PrepareFlip(request_id, CurrentBuffer());
 	GetScheduler().Flush();
+	m_renderer.GetVideoOut().CompleteFlip(request_id);
 }
 
 void CommandProcessor::SynchronizeGpu() {

src/graphics/guest_gpu/hardwareContext.h

@@ -498,8 +498,7 @@ struct LsStageRegisters {
 };
 
 struct HsStageRegisters {
-	uint64_t          data_addr      = 0;
-	uint64_t          user_data_addr = 0;
+	uint64_t          data_addr = 0;
 	HsShaderResource1 rsrc1;
 	HsShaderResource2 rsrc2;
 };
@@ -936,12 +935,6 @@ private:
 	ShaderRegisters m_sh_regs;
 };
 
-struct FsrView {
-	uint32_t control_points[2][4] {};
-	uint32_t alphas[2][2] {};
-	uint32_t window[2] {};
-};
-
 class UserConfig {
 public:
 	UserConfig()  = default;
@@ -971,17 +964,8 @@ public:
 	void SetGdsOaCntl(uint32_t value) { m_gds_oa.cntl = value; }
 	void SetGdsOaCounter(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].counter = value; }
 	void SetGdsOaAddress(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].address = value; }
-	void SetFsrControlPoint(uint32_t axis, uint32_t index, uint32_t value) {
-		m_fsr_view.control_points[axis][index] = value;
-	}
-	void SetFsrAlpha(uint32_t axis, uint32_t index, uint32_t value) {
-		m_fsr_view.alphas[axis][index] = value;
-	}
-	void SetFsrWindow(uint32_t index, uint32_t value) { m_fsr_view.window[index] = value; }
-	[[nodiscard]] const FsrView& GetFsrView() const { return m_fsr_view; }
 
 private:
-	FsrView m_fsr_view;
 	Prospero::PrimitiveType m_prim_type               = Prospero::PrimitiveType::kNone;
 	uint32_t                m_index_offset            = 0;
 	uint32_t                m_object_id               = 0;
@@ -1004,14 +988,14 @@ public:
 	void SetEsShaderBase(uint64_t addr) { m_vs.es_regs.data_addr = addr; }
 	void SetLsShaderBase(uint64_t addr) { m_vs.ls_regs.data_addr = addr; }
 	void SetHsShaderBase(uint64_t addr) { m_vs.hs_regs.data_addr = addr; }
-	void SetHsUserDataAddress(uint32_t word, uint32_t value) {
-		SetUserDataAddressWord(m_vs.hs_regs.user_data_addr, word, value);
-	}
 	void SetHsShaderResource1(const HsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
 	void SetHsShaderResource2(const HsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
 	void SetGsShaderBase(uint64_t addr) { m_vs.gs_regs.data_addr = addr; }
 	void SetGsUserDataAddress(uint32_t word, uint32_t value) {
-		SetUserDataAddressWord(m_vs.gs_regs.user_data_addr, word, value);
+		const auto shift   = word * 32u;
+		auto&      address = m_vs.gs_regs.user_data_addr;
+		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
+		          (static_cast<uint64_t>(value) << shift);
 	}
 	void SetGsShaderResource1(const GsShaderResource1& rsrc1) { m_vs.gs_regs.rsrc1 = rsrc1; }
 	void SetGsShaderResource2(const GsShaderResource2& rsrc2) { m_vs.gs_regs.rsrc2 = rsrc2; }
@@ -1053,12 +1037,6 @@ public:
 	[[nodiscard]] const ComputeShaderInfo& GetCs() const { return m_cs; }
 
 private:
-	static void SetUserDataAddressWord(uint64_t& address, uint32_t word, uint32_t value) {
-		const auto shift = word * 32u;
-		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
-		          (static_cast<uint64_t>(value) << shift);
-	}
-
 	VertexShaderInfo  m_vs;
 	PixelShaderInfo   m_ps;
 	ComputeShaderInfo m_cs;

src/graphics/guest_gpu/pm4.h

@@ -910,7 +910,6 @@ constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_MASK    = 0x1;
 constexpr uint32_t COMPUTE_PGM_RSRC1_FWD_PROGRESS_SHIFT = 31;
 constexpr uint32_t COMPUTE_PGM_RSRC1_FWD_PROGRESS_MASK  = 0x1;
 
-constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS = 1u << 5u;
 constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_SHIFT = 15;
 constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_MASK  = 0x1;
 
@@ -988,12 +987,6 @@ constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR2_MASK  = 0x1;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_SHIFT = 2;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_MASK  = 0x1;
 
-constexpr uint32_t FSR_WINDOW_LEFT           = 0x2BC;
-constexpr uint32_t FSR_CONTROL_POINTS_LEFT_X = 0x2C4;
-constexpr uint32_t FSR_CONTROL_POINTS_LEFT_Y = 0x2CC;
-constexpr uint32_t FSR_ALPHA_LEFT_X          = 0x2E4;
-constexpr uint32_t FSR_ALPHA_LEFT_Y          = 0x2E8;
-
 constexpr uint32_t TA_CS_BC_BASE_ADDR       = 0x380;
 constexpr uint32_t TA_CS_BC_BASE_ADDR_HI    = 0x381;
 constexpr uint32_t TEXTURE_GRADIENT_FACTORS = 0x383;

src/graphics/guest_gpu/tile.cpp

@@ -1,18 +1,24 @@
 #include "graphics/guest_gpu/tile.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
+#include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/gpu_format.h"
 
 #include <algorithm>
 #include <array>
 #include <bit>
+#include <fmt/format.h>
+#include <vector>
 
 namespace Libs::Graphics {
 
+static uint32_t AlignUp(uint32_t value, uint32_t alignment) {
+	return (value + alignment - 1u) & ~(alignment - 1u);
+}
+
 static uint32_t ShiftCeil(uint32_t value, uint32_t shift) {
 	return static_cast<uint32_t>((static_cast<uint64_t>(value) + (1ull << shift) - 1ull) >> shift);
 }
@@ -27,7 +33,7 @@ static uint32_t CalcLinearAlignedLevelPitch(uint32_t base_width, uint32_t base_h
 	const uint32_t level_width  = ShiftCeil(base_width, level);
 	const uint32_t level_height = ShiftCeil(base_height, level);
 	const uint32_t padded_width =
-	    Common::AlignUp(std::max(level_width, 1u), CalcLinearBlockWidth(bytes_per_element));
+	    AlignUp(std::max(level_width, 1u), CalcLinearBlockWidth(bytes_per_element));
 	const uint64_t size =
 	    static_cast<uint64_t>(padded_width) * std::max(level_height, 1u) * bytes_per_element;
 	EXIT_NOT_IMPLEMENTED(size > 0xffffffffull);
@@ -55,7 +61,7 @@ static uint32_t SetLinearMipChainLayout(uint32_t levels, const uint32_t* mip_pit
 		offset += mip_size[level];
 	}
 
-	return Common::AlignUp(offset, 256u);
+	return AlignUp(offset, 256u);
 }
 
 struct Log2BlockDimensions {
@@ -360,8 +366,8 @@ bool TileGetTiledTextureLayout(const TileSurfaceDescription& description, TileSu
 		    ((description.width >> level) + texture.texel_width - 1u) / texture.texel_width, 1u);
 		mip.height = std::max(
 		    ((description.height >> level) + texture.texel_height - 1u) / texture.texel_height, 1u);
-		mip.padded_width  = Common::AlignUp(std::max(ShiftCeil(width0, level), 1u), block.block_width);
-		mip.padded_height = Common::AlignUp(std::max(ShiftCeil(height0, level), 1u), block.block_height);
+		mip.padded_width  = AlignUp(std::max(ShiftCeil(width0, level), 1u), block.block_width);
+		mip.padded_height = AlignUp(std::max(ShiftCeil(height0, level), 1u), block.block_height);
 		mip.size = static_cast<uint64_t>(block.block_depth) * mip.padded_width * mip.padded_height *
 		           block.bytes_per_element;
 		result.block_slice_size += mip.size;
@@ -1115,8 +1121,8 @@ static bool TileGetHtileSize(uint32_t width, uint32_t height, TileSizeAlign& hti
 	}
 	// Prospero HTile stores one DWORD per depth tile. Its 32 KiB allocation blocks cover
 	// 1024x512 pixels, independently of the attachment's fragment count.
-	const uint64_t size = static_cast<uint64_t>(Common::AlignUp(width, 1024u) / 1024u) *
-	                      (Common::AlignUp(height, 512u) / 512u) * 32768u;
+	const uint64_t size = static_cast<uint64_t>(AlignUp(width, 1024u) / 1024u) *
+	                      (AlignUp(height, 512u) / 512u) * 32768u;
 	if (size == 0 || size > UINT32_MAX) {
 		return false;
 	}
@@ -1148,13 +1154,13 @@ bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch,
 		                                           &stencil_block_height));
 		const uint32_t fragments = 1u << num_fragments_log2;
 		const uint64_t depth_bytes_total =
-		    valid_blocks ? static_cast<uint64_t>(Common::AlignUp(width, depth_block_width)) *
-		                       Common::AlignUp(height, depth_block_height) * depth_bytes * fragments
+		    valid_blocks ? static_cast<uint64_t>(AlignUp(width, depth_block_width)) *
+		                       AlignUp(height, depth_block_height) * depth_bytes * fragments
 		                 : 0;
 		const uint64_t stencil_bytes_total =
 		    stencil_format == Prospero::StencilFormat::k8UInt && valid_blocks
-		        ? static_cast<uint64_t>(Common::AlignUp(width, stencil_block_width)) *
-		              Common::AlignUp(height, stencil_block_height) * fragments
+		        ? static_cast<uint64_t>(AlignUp(width, stencil_block_width)) *
+		              AlignUp(height, stencil_block_height) * fragments
 		        : 0;
 		TileSizeAlign calculated_htile {};
 		const bool    htile_valid = !htile || TileGetHtileSize(width, height, calculated_htile);
@@ -1181,7 +1187,8 @@ uint32_t TileGetRenderTargetPitch(uint32_t width, uint32_t bytes_per_element,
 	                                                         &block_width, &block_height)) {
 		return 0;
 	}
-	const uint64_t pitch = Common::AlignUp(static_cast<uint64_t>(width), block_width);
+	const uint64_t pitch = (static_cast<uint64_t>(width) + block_width - 1u) &
+	                       ~static_cast<uint64_t>(block_width - 1u);
 	return pitch <= UINT32_MAX ? static_cast<uint32_t>(pitch) : 0;
 }
 
@@ -1202,7 +1209,8 @@ bool TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch,
 	    pitch != TileGetRenderTargetPitch(width, bytes_per_element, num_fragments_log2)) {
 		return false;
 	}
-	const uint64_t padded_height = Common::AlignUp(static_cast<uint64_t>(height), block_height);
+	const uint64_t padded_height = (static_cast<uint64_t>(height) + block_height - 1u) &
+	                               ~static_cast<uint64_t>(block_height - 1u);
 	const uint64_t size = static_cast<uint64_t>(pitch) * padded_height * bytes_per_element *
 	                      (1u << num_fragments_log2);
 	if (size == 0 || size > UINT32_MAX) {
@@ -1213,12 +1221,20 @@ bool TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch,
 	return true;
 }
 
-static bool TileGetColorMetadataSize(uint32_t width, uint32_t height, uint32_t slices,
-                                      uint32_t block_width, uint32_t block_height,
-                                      TileSizeAlign& total_size) {
-	if (width == 0 || height == 0 || slices == 0) {
+bool TileGetDccSize(uint32_t width, uint32_t height, uint32_t slices,
+                    uint32_t bytes_per_element, uint32_t levels, Prospero::TileMode tile,
+                    TileSizeAlign& total_size, uint32_t num_fragments_log2) {
+	total_size = {};
+	if (width == 0 || height == 0 || slices == 0 || levels != 1 || num_fragments_log2 != 0 ||
+	    !std::has_single_bit(bytes_per_element) || bytes_per_element > 16 ||
+	    (tile != Prospero::TileMode::kRenderTarget && tile != Prospero::TileMode::kDepth)) {
 		return false;
 	}
+	// Gen5 color metadata uses 4 KiB blocks, with one byte per 256 bytes of color data.
+	// Metadata stays thin for volume surfaces, so every depth slice has its own block raster.
+	const uint32_t coverage_bits = 20u - std::countr_zero(bytes_per_element);
+	const uint32_t block_width   = 1u << ((coverage_bits + 1u) / 2u);
+	const uint32_t block_height  = 1u << (coverage_bits / 2u);
 	const uint64_t blocks_x = (static_cast<uint64_t>(width) + block_width - 1u) / block_width;
 	const uint64_t blocks_y = (static_cast<uint64_t>(height) + block_height - 1u) / block_height;
 	const uint64_t blocks   = blocks_x * blocks_y;
@@ -1230,27 +1246,36 @@ static bool TileGetColorMetadataSize(uint32_t width, uint32_t height, uint32_t s
 	return true;
 }
 
-bool TileGetDccSize(uint32_t width, uint32_t height, uint32_t slices,
-                    uint32_t bytes_per_element, uint32_t levels, Prospero::TileMode tile,
-                    TileSizeAlign& total_size, uint32_t num_fragments_log2) {
+bool TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t pitch,
+                                  uint32_t bytes_per_element, uint32_t levels,
+                                  TileSizeAlign& total_size, TileSizeOffset* level_sizes,
+                                  TilePaddedSize* padded_size) {
 	total_size = {};
-	if (levels != 1 || num_fragments_log2 != 0 ||
-	    !std::has_single_bit(bytes_per_element) || bytes_per_element > 16 ||
-	    (tile != Prospero::TileMode::kRenderTarget && tile != Prospero::TileMode::kDepth)) {
+	if (width == 0 || height == 0 || levels == 0 || levels > 16 ||
+	    pitch != TileGetRenderTargetPitch(width, bytes_per_element)) {
 		return false;
 	}
-	// Gen5 DCC uses 4 KiB blocks, with one byte per 256 bytes of color data.
-	// Metadata stays thin for volume surfaces, so every depth slice has its own block raster.
-	const uint32_t coverage_bits = 20u - std::countr_zero(bytes_per_element);
-	return TileGetColorMetadataSize(width, height, slices, 1u << ((coverage_bits + 1u) / 2u),
-	                                1u << (coverage_bits / 2u), total_size);
-}
-
-bool TileGetCmaskSize(uint32_t width, uint32_t height, uint32_t slices, uint32_t levels,
-                      TileSizeAlign& total_size) {
-	total_size = {};
-	// Gen5 CMASK stores four bits per 8x8 tile in 4 KiB blocks covering 1024x512 pixels.
-	return levels == 1 && TileGetColorMetadataSize(width, height, slices, 1024, 512, total_size);
+	uint32_t max_levels    = 1;
+	uint32_t max_dimension = std::max(width, height);
+	while (max_dimension > 1) {
+		max_dimension >>= 1u;
+		max_levels++;
+	}
+	if (levels > max_levels) {
+		return false;
+	}
+	auto format = Prospero::BufferFormat::kInvalid;
+	switch (bytes_per_element) {
+		case 1: format = Prospero::BufferFormat::k8UNorm; break;
+		case 2: format = Prospero::BufferFormat::k16UNorm; break;
+		case 4: format = Prospero::BufferFormat::k32Float; break;
+		case 8: format = Prospero::BufferFormat::k16_16_16_16Float; break;
+		case 16: format = Prospero::BufferFormat::k32_32_32_32Float; break;
+		default: return false;
+	}
+	TileGetTextureSize(format, width, height, levels, Prospero::TileMode::kRenderTarget,
+	                   &total_size, level_sizes, padded_size);
+	return total_size.size != 0 && total_size.align == 65536;
 }
 
 void TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
@@ -1302,8 +1327,13 @@ void TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t
 		return;
 	}
 	if (total_size != nullptr && total_size->size == 0) {
-		EXIT("unknown format:\nformat = %u\nwidth  = %u\nheight = %u\nlevels = %u\ntile   = %u\n",
-		     static_cast<uint32_t>(format), width, height, levels, static_cast<uint32_t>(tile));
+		std::vector<std::string> list;
+		list.push_back(fmt::format("format = {}", static_cast<uint32_t>(format)));
+		list.push_back(fmt::format("width  = {}", width));
+		list.push_back(fmt::format("height = {}", height));
+		list.push_back(fmt::format("levels = {}", levels));
+		list.push_back(fmt::format("tile   = {}", static_cast<uint32_t>(tile)));
+		EXIT("unknown format:\n%s\n", Common::Concat(list, '\n').c_str());
 	}
 }
 
@@ -1342,7 +1372,7 @@ uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
 			TileTextureElementLayout element {};
 			if (TileGetTextureElementLayout(format, element) && element.texel_width == 1 &&
 			    element.texel_height == 1) {
-				pitch = Common::AlignUp(pitch, CalcLinearBlockWidth(element.bytes));
+				pitch = AlignUp(pitch, CalcLinearBlockWidth(element.bytes));
 			}
 			break;
 		}
@@ -1353,7 +1383,7 @@ uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
 		case Prospero::TileMode::kRenderTarget: {
 			TileTextureBlockLayout layout {};
 			if (TileGetTextureBlockLayout(format, tile, false, layout)) {
-				pitch = Common::AlignUp(pitch, layout.block.block_width * layout.texel_width);
+				pitch = AlignUp(pitch, layout.block.block_width * layout.texel_width);
 			}
 			break;
 		}

src/graphics/guest_gpu/tile.h

@@ -86,7 +86,6 @@ struct TileMipLayout {
 	uint32_t padded_height = 0;
 	uint32_t tail_x        = 0;
 	uint32_t tail_y        = 0;
-	bool     operator==(const TileMipLayout&) const = default;
 };
 
 struct TileSurfaceLayout {
@@ -126,8 +125,10 @@ bool     TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch
 bool     TileGetDccSize(uint32_t width, uint32_t height, uint32_t slices,
                         uint32_t bytes_per_element, uint32_t levels, Prospero::TileMode tile,
                         TileSizeAlign& total_size, uint32_t num_fragments_log2 = 0);
-bool     TileGetCmaskSize(uint32_t width, uint32_t height, uint32_t slices, uint32_t levels,
-                          TileSizeAlign& total_size);
+bool     TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t pitch,
+                                      uint32_t bytes_per_element, uint32_t levels,
+                                      TileSizeAlign& total_size, TileSizeOffset* level_sizes,
+                                      TilePaddedSize* padded_size);
 void     TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
                             uint32_t levels, Prospero::TileMode tile, TileSizeAlign* total_size,
                             TileSizeOffset* level_sizes, TilePaddedSize* padded_size);

src/graphics/host_gpu/graphicContext.h

@@ -27,9 +27,9 @@ struct GraphicContext {
 	vk::Device                         device                                = nullptr;
 	VmaAllocator                       allocator                             = nullptr;
 	bool                               memory_budget_ext_enabled             = false;
+	bool                               rt_extensions_enabled                 = false;
 	bool                               compute_subgroup_size_control_enabled = false;
 	bool                               sample_rate_shading_enabled           = false;
-	// bool fp64_denorm_preserve = false; // Temporarily disabled.
 	bool                               attachment_feedback_loop_enabled      = false;
 	bool                               provoking_vertex_last_enabled         = false;
 	bool                               supports_block_texel_view              = false;
@@ -99,6 +99,10 @@ struct GraphicContext {
 	[[nodiscard]] uint64_t GetTotalMemoryBudget() const;
 	[[nodiscard]] bool     CreateImage(const vk::ImageCreateInfo& info, VulkanImage& image);
 	void                   DeleteImage(VulkanImage& image);
+	void                   AppendHardwareRayTracingDeviceExtensions(
+	    const std::vector<vk::ExtensionProperties>& available_extensions,
+	    std::vector<const char*>&                   device_extensions);
+	void LoadHardwareRayTracingFunctions() const;
 
 	uint32_t screen_width  = 0;
 	uint32_t screen_height = 0;

src/graphics/host_gpu/memoryTracker.cpp

@@ -1,6 +1,5 @@
 #include "graphics/host_gpu/memoryTracker.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 
 namespace Libs::Graphics {
@@ -9,6 +8,9 @@ static_assert(std::atomic<void*>::is_always_lock_free);
 
 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);
+	}
 }
 
 MemoryTracker::~MemoryTracker() = default;
@@ -32,8 +34,8 @@ void MemoryTracker::ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr,
 void MemoryTracker::ValidateGpuDirtyOwnership(const RangeSet& dirty, uint64_t vaddr, uint64_t size,
                                               const char* operation) {
 	ValidateRange(vaddr, size);
-	const auto begin = Common::AlignDown(vaddr, TRACKER_PAGE_SIZE);
-	const auto end   = Common::AlignUp(vaddr + size, TRACKER_PAGE_SIZE);
+	const auto begin = vaddr & ~(TRACKER_PAGE_SIZE - 1);
+	const auto end   = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
 	for (auto page = begin; page < end; page += TRACKER_PAGE_SIZE) {
 		const bool has_dirty_bytes = dirty.Intersects(page, TRACKER_PAGE_SIZE);
 		if (IsRegionGpuModified(page, TRACKER_PAGE_SIZE) != has_dirty_bytes) {

src/graphics/host_gpu/memoryTracker.h

@@ -62,18 +62,43 @@ public:
 	void ValidateGpuDirtyOwnership(const RangeSet&, uint64_t, uint64_t, const char*) {}
 #endif
 
-	template <bool clear, typename Func>
-	void ForEachDownloadRange(uint64_t vaddr, uint64_t size, Func&& func) {
+	template <bool clear, typename Preflight, typename Func>
+	void ForEachDownloadRange(uint64_t vaddr, uint64_t size, Preflight&& preflight, Func&& func) {
+		static_assert(std::is_nothrow_invocable_v<Preflight&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<Func&, uint64_t, uint64_t>);
 		CheckNotInUploadCallback();
-		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
-			std::scoped_lock lock(manager->lock);
-			const auto       address = manager->GetCpuAddr() + offset;
-			manager->template ForEachModifiedRange<DirtySource::Gpu, false>(address, bytes, func);
-			if constexpr (clear) {
-				manager->template ChangeState<DirtySource::Gpu, false>(address, bytes);
-			}
+		std::vector<RegionManager*> managers;
+		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t, uint64_t) {
+			managers.push_back(manager);
 		});
+		std::vector<std::unique_lock<TrackingSpinLock>> locks;
+		locks.reserve(managers.size());
+		for (auto* manager: managers) {
+			locks.emplace_back(manager->lock);
+		}
+		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+			const auto address = manager->GetCpuAddr() + offset;
+			manager->template ForEachModifiedRange<DirtySource::Gpu, false>(address, bytes,
+			                                                                preflight);
+		});
+		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+			manager->template ForEachModifiedRange<DirtySource::Gpu, false>(
+			    manager->GetCpuAddr() + offset, bytes, func);
+		});
+		if constexpr (clear) {
+			Iterate<false>(vaddr, size,
+			               [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+				               const auto address = manager->GetCpuAddr() + offset;
+				               manager->template ForEachModifiedRange<DirtySource::Gpu, true>(
+				                   address, bytes, [](uint64_t, uint64_t) noexcept {});
+			               });
+		}
+	}
+
+	template <bool clear, typename Func>
+	void ForEachDownloadRange(uint64_t vaddr, uint64_t size, Func&& func) {
+		ForEachDownloadRange<clear>(
+		    vaddr, size, [](uint64_t, uint64_t) noexcept {}, std::forward<Func>(func));
 	}
 
 	template <typename RangeFunc, typename UploadFunc>

src/graphics/host_gpu/pageManager.cpp

@@ -1,7 +1,5 @@
 #include "graphics/host_gpu/pageManager.h"
 
-#include "common/alignment.h"
-#include "common/virtualMemory.h"
 #include "graphics/host_gpu/regionDefinitions.h"
 #include "kernel/memory.h"
 
@@ -34,8 +32,19 @@ constexpr uint64_t REGION_SIZE  = TRACKER_REGION_SIZE;
 constexpr uint64_t ADDRESS_SIZE = TRACKER_ADDRESS_SIZE;
 constexpr uint64_t REGION_COUNT = ADDRESS_SIZE / REGION_SIZE;
 
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
+// The tracker reuses Win32 memory-protection tags as internal page-state values.
+// Mirror their canonical numeric values so the shared state-machine logic is identical.
+constexpr uint32_t PAGE_NOACCESS  = 0x01;
+constexpr uint32_t PAGE_READONLY  = 0x02;
+constexpr uint32_t PAGE_READWRITE = 0x04;
+#endif
 constexpr uint64_t REGION_PAGES = REGION_SIZE / PAGE_SIZE;
 
+constexpr uint32_t NO_ACCESS_PROTECTION  = PAGE_NOACCESS;
+constexpr uint32_t READ_ONLY_PROTECTION  = PAGE_READONLY;
+constexpr uint32_t READ_WRITE_PROTECTION = PAGE_READWRITE;
+
 [[noreturn]] void FailFast(const char* reason = nullptr) noexcept {
 	std::fputs("PageManager fail-fast: ", stderr);
 	std::fputs(reason != nullptr ? reason : "invalid page state", stderr);
@@ -58,6 +67,15 @@ constexpr uint64_t REGION_PAGES = REGION_SIZE / PAGE_SIZE;
 	std::_Exit(322);
 }
 
+Common::VirtualMemory::Mode ToMemoryMode(uint32_t protection) {
+	switch (protection) {
+		case NO_ACCESS_PROTECTION: return Common::VirtualMemory::Mode::NoAccess;
+		case READ_ONLY_PROTECTION: return Common::VirtualMemory::Mode::Read;
+		case READ_WRITE_PROTECTION: return Common::VirtualMemory::Mode::ReadWrite;
+		default: Fatal("unmappable protection 0x%08" PRIx32, protection);
+	}
+}
+
 class SpinGuard final {
 public:
 	explicit SpinGuard(std::atomic_flag& lock): m_lock(lock) {
@@ -78,6 +96,15 @@ void ValidateRange(uint64_t vaddr, uint64_t size) {
 	}
 }
 
+uint64_t PageStart(uint64_t vaddr) {
+	return vaddr & ~(PAGE_SIZE - 1);
+}
+
+uint64_t PageEnd(uint64_t vaddr, uint64_t size) {
+	ValidateRange(vaddr, size);
+	return PageStart(vaddr + size - 1) + PAGE_SIZE;
+}
+
 } // namespace
 
 struct PageManager::Impl {
@@ -85,14 +112,14 @@ struct PageManager::Impl {
 		uint8_t write_watchers  : 7 = 0;
 		uint8_t access_watchers : 1 = 0;
 
-		[[nodiscard]] Common::VirtualMemory::Mode Perms() const noexcept {
+		[[nodiscard]] uint32_t Perms() const noexcept {
 			if (access_watchers != 0) {
-				return Common::VirtualMemory::Mode::NoAccess;
+				return NO_ACCESS_PROTECTION;
 			}
 			if (write_watchers != 0) {
-				return Common::VirtualMemory::Mode::Read;
+				return READ_ONLY_PROTECTION;
 			}
-			return Common::VirtualMemory::Mode::ReadWrite;
+			return READ_WRITE_PROTECTION;
 		}
 
 		template <int delta, bool is_read>
@@ -156,6 +183,9 @@ struct PageManager::Impl {
 		}
 #endif
 		regions = std::make_unique<std::atomic<Region*>[]>(REGION_COUNT);
+		for (uint64_t i = 0; i < REGION_COUNT; i++) {
+			regions[i].store(nullptr, std::memory_order_relaxed);
+		}
 	}
 
 	~Impl() {
@@ -190,10 +220,11 @@ struct PageManager::Impl {
 		return ptr;
 	}
 
-	void Protect(uint64_t vaddr, uint64_t size, Common::VirtualMemory::Mode mode) noexcept {
-		if (!Libs::LibKernel::Memory::ProtectGuestHostMemory(vaddr, size, mode)) {
-			Fatal("address-space protection failed at 0x%016" PRIx64 ", mode=0x%08" PRIx32, vaddr,
-			      static_cast<uint32_t>(mode));
+	void Protect(uint64_t vaddr, uint64_t size, uint32_t protection) noexcept {
+		if (!Libs::LibKernel::Memory::ProtectGuestHostMemory(vaddr, size,
+		                                                     ToMemoryMode(protection))) {
+			Fatal("address-space protection failed at 0x%016" PRIx64 ", new=0x%08" PRIx32, vaddr,
+			      protection);
 		}
 	}
 
@@ -250,12 +281,11 @@ struct PageManager::Impl {
 
 	template <bool track, bool is_read>
 	void UpdatePageWatchers(uint64_t vaddr, uint64_t size) {
-		ValidateRange(vaddr, size);
-		const auto begin = Common::AlignDown(vaddr, PAGE_SIZE);
-		const auto end   = Common::AlignUp(vaddr + size, PAGE_SIZE);
+		const auto begin = PageStart(vaddr);
+		const auto end   = PageEnd(vaddr, size);
 		for (auto chunk_begin = begin; chunk_begin < end;) {
-			const auto chunk_end = std::min(end, Common::AlignUp(chunk_begin + 1, REGION_SIZE));
-			const auto region_base = Common::AlignDown(chunk_begin, REGION_SIZE);
+			const auto chunk_end   = std::min(end, (chunk_begin / REGION_SIZE + 1) * REGION_SIZE);
+			const auto region_base = chunk_begin / REGION_SIZE * REGION_SIZE;
 			auto*      region = track ? GetOrCreateRegion(chunk_begin) : FindRegion(chunk_begin);
 			if (region == nullptr) {
 				Fatal("untracking unknown page 0x%016" PRIx64, chunk_begin);

src/graphics/host_gpu/rangeSet.h

@@ -6,11 +6,17 @@
 #include <algorithm>
 #include <cstdint>
 #include <map>
+#include <vector>
 
 namespace Libs::Graphics {
 
 class RangeSet final {
 public:
+	struct Range {
+		uint64_t address = 0;
+		uint64_t size    = 0;
+	};
+
 	void Add(uint64_t address, uint64_t size) {
 		const auto end = End(address, size);
 		auto       it  = m_ranges.lower_bound(address);
@@ -56,6 +62,12 @@ public:
 		}
 	}
 
+	[[nodiscard]] std::vector<Range> Intersections(uint64_t address, uint64_t size) const {
+		std::vector<Range> result;
+		ForEachIntersection(address, size, [&result](Range range) { result.push_back(range); });
+		return result;
+	}
+
 	[[nodiscard]] bool Intersects(uint64_t address, uint64_t size) const {
 		const auto end = End(address, size);
 		auto       it  = m_ranges.lower_bound(address);
@@ -76,7 +88,7 @@ public:
 	}
 
 	template <typename Func>
-	void ForEachInRange(uint64_t address, uint64_t size, Func&& func) const {
+	void ForEachIntersection(uint64_t address, uint64_t size, Func&& func) const {
 		const auto end = End(address, size);
 		auto       it  = m_ranges.upper_bound(address);
 		if (it != m_ranges.begin()) {
@@ -86,7 +98,7 @@ public:
 			const auto begin = std::max(address, it->first);
 			const auto last  = std::min(end, it->second);
 			if (begin < last) {
-				func(begin, last);
+				func(Range {begin, last - begin});
 			}
 		}
 	}

src/graphics/host_gpu/regionManager.h

@@ -16,7 +16,6 @@
 #include <windows.h>
 #undef min
 #undef max
-#undef MemoryBarrier
 #elif defined(__APPLE__)
 #include <pthread.h>
 #elif defined(__linux__)
@@ -112,43 +111,56 @@ public:
 			bits.UnsetRange(start, end);
 		}
 		if constexpr (source == DirtySource::Cpu) {
-			UpdateProtection<!enable, false>();
+			UpdateCpuProtection<!enable>();
 		} else {
-			UpdateProtection<enable, true>();
+			UpdateGpuProtection<enable>();
 		}
 	}
 
 	template <DirtySource source, bool clear, typename Func>
 	void ForEachModifiedRange(uint64_t vaddr, uint64_t size, Func&& func) {
 		const auto [start, end] = GetPageRange(vaddr, size);
-		auto&      bits         = GetBits<source>();
-		RegionBits mask(bits, start, end);
+		RegionBits mask(GetBits<source>(), start, end);
 		if constexpr (clear) {
-			bits.UnsetRange(start, end);
-			if constexpr (source == DirtySource::Cpu) {
-				UpdateProtection<true, false>();
-			} else {
-				UpdateProtection<false, true>();
-			}
+			GetBits<source>().UnsetRange(start, end);
 		}
-		for (const auto [first, last]: mask) {
-			func(m_cpu_addr + first * TRACKER_PAGE_SIZE, (last - first) * TRACKER_PAGE_SIZE);
+		if constexpr (source == DirtySource::Cpu && clear) {
+			UpdateCpuProtection<true>();
+			ForEachRange(mask, std::forward<Func>(func));
+			return;
 		}
+		if constexpr (source == DirtySource::Gpu && clear) {
+			UpdateGpuProtection<false>();
+		}
+		ForEachRange(mask, std::forward<Func>(func));
 	}
 
 	TrackingSpinLock lock;
 
 private:
-	template <bool track, bool is_read>
-	void UpdateProtection() {
-		const auto protection = is_read ? ~m_gpu_dirty : m_cpu_dirty;
-		auto&      previous   = is_read ? m_readable : m_writable;
-		auto       mask       = protection ^ previous;
+	template <bool track>
+	void UpdateCpuProtection() {
+		auto mask  = m_cpu_dirty ^ m_writable;
+		m_writable = m_cpu_dirty;
 		if (mask.None()) {
 			return;
 		}
-		previous = protection;
-		m_page_manager.UpdatePageWatchersForRegion<track, is_read>(m_cpu_addr, mask);
+		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>
@@ -179,6 +191,13 @@ private:
 		        static_cast<size_t>((offset + size + TRACKER_PAGE_SIZE - 1) / TRACKER_PAGE_SIZE)};
 	}
 
+	template <typename Func>
+	void ForEachRange(const RegionBits& bits, Func&& func) const {
+		for (const auto [start, end]: bits) {
+			func(m_cpu_addr + start * TRACKER_PAGE_SIZE, (end - start) * TRACKER_PAGE_SIZE);
+		}
+	}
+
 	PageManager& m_page_manager;
 	uint64_t     m_cpu_addr = 0;
 	RegionBits   m_cpu_dirty;

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

@@ -1,6 +1,5 @@
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -43,6 +42,13 @@ void BufferCache::WriteDataBuffer(Buffer& buffer, uint64_t address, const void*
 	}
 }
 
+struct BufferCache::DownloadCopy {
+	Buffer*  buffer        = nullptr;
+	uint64_t source_offset = 0;
+	uint64_t address       = 0;
+	uint64_t size          = 0;
+};
+
 void BufferCache::Register(BufferId id) {
 	ChangeRegister<true>(id);
 }
@@ -97,7 +103,8 @@ void BufferCache::TouchBuffer(const Buffer& buffer) {
 }
 
 void BufferCache::DeleteBuffer(BufferId id) {
-	if (IsBufferInvalid(id)) {
+	auto* buffer = m_slot_buffers.try_get(id);
+	if (buffer == nullptr || buffer->is_deleted) {
 		return;
 	}
 	Unregister(id);
@@ -108,70 +115,79 @@ void BufferCache::DeleteBuffer(BufferId id) {
 	}
 }
 
-bool BufferCache::DownloadBufferMemory(Buffer& buffer, uint64_t vaddr, uint64_t size) {
-	std::vector<vk::BufferCopy> copies;
-	uint64_t                    total_size     = 0;
-	const auto                  buffer_address = buffer.CpuAddress();
-	m_memory_tracker.ForEachDownloadRange<false>(
-	    vaddr, size, [&](uint64_t address, uint64_t bytes) noexcept {
-		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-		                                           "buffer download");
-		    m_gpu_modified_ranges.ForEachInRange(address, bytes, [&](uint64_t start, uint64_t end) {
-			    copies.emplace_back(start - buffer_address, total_size, end - start);
-			    // Keep packed ranges on separate cache lines, as in shadPS4.
-			    total_size += Common::AlignUp(end - start, 64);
-		    });
-		    m_gpu_modified_ranges.Subtract(address, bytes);
-	    });
-	if (copies.empty()) {
-		return false;
+std::pair<uint64_t, uint64_t> BufferCache::DownloadEnvelope(const DownloadCopy& copy) {
+	if (copy.buffer == nullptr || copy.size == 0 || copy.source_offset > copy.buffer->Size() ||
+	    copy.size > copy.buffer->Size() - copy.source_offset) {
+		EXIT("BufferCache: invalid download copy\n");
 	}
-
-	const auto [mapped, offset] = m_download_buffer.Map(total_size, 64);
-	if (mapped == nullptr) {
-		EXIT("BufferCache: download exceeds 64 MiB staging buffer capacity\n");
+	const auto begin = copy.source_offset & ~uint64_t {3};
+	if (copy.source_offset > UINT64_MAX - copy.size ||
+	    copy.source_offset + copy.size > UINT64_MAX - 3) {
+		EXIT("BufferCache: download copy alignment overflow\n");
 	}
-	m_download_buffer.Commit();
-	for (auto& copy: copies) {
-		copy.dstOffset += offset;
+	const auto end = (copy.source_offset + copy.size + 3) & ~uint64_t {3};
+	if (end > copy.buffer->Size()) {
+		EXIT("BufferCache: aligned download copy exceeds its owner\n");
 	}
+	return {begin, end - begin};
+}
 
-	auto& command = m_scheduler.Current();
-	command.EndRendering();
-	const auto              native = command.Handle();
-	vk::BufferMemoryBarrier before {};
-	before.srcAccessMask       = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
-	before.dstAccessMask       = vk::AccessFlagBits::eTransferRead;
-	before.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-	before.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-	before.buffer              = buffer.Handle();
-	before.offset              = 0;
-	before.size                = buffer.Size();
-	native.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
-	                       vk::PipelineStageFlagBits::eTransfer, {}, 0, nullptr, 1, &before, 0,
-	                       nullptr);
-	native.copyBuffer(buffer.Handle(), m_download_buffer.Handle(),
-	                  static_cast<uint32_t>(copies.size()), copies.data());
-
-	auto after          = before;
-	after.srcAccessMask = vk::AccessFlagBits::eTransferWrite;
-	after.dstAccessMask = vk::AccessFlagBits::eHostRead;
-	after.buffer        = m_download_buffer.Handle();
-	after.offset        = offset;
-	after.size          = total_size;
-	native.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
-	                       vk::PipelineStageFlagBits::eAllCommands |
-	                           vk::PipelineStageFlagBits::eHost,
-	                       {}, 0, nullptr, 1, &after, 0, nullptr);
-	m_scheduler.DeferPriorityOperation([this, mapped, offset, total_size, buffer_address,
-	                                    copies = std::move(copies)] {
-		m_download_buffer.Invalidate(offset, total_size);
-		for (const auto& copy: copies) {
-			Libs::LibKernel::Memory::WriteBacking(buffer_address + copy.srcOffset,
-			                                      mapped + (copy.dstOffset - offset), copy.size);
+void BufferCache::DownloadBufferMemory(std::span<const DownloadCopy> copies) {
+	std::vector<DownloadCopy> batch;
+	batch.reserve(copies.size());
+	uint64_t                  packed_size = 0;
+	auto&                     download    = m_download_buffer;
+	const auto flush = [&] {
+		const auto [mapped, base_offset] = download.Map(packed_size, DOWNLOAD_ALIGNMENT);
+		EXIT_IF(mapped == nullptr);
+		uint64_t cursor = 0;
+		for (const auto& copy: batch) {
+			const auto [source_begin, envelope_size] = DownloadEnvelope(copy);
+			download.CopyFrom(m_scheduler.Current(), *copy.buffer, source_begin, base_offset + cursor,
+			                  envelope_size, vk::AccessFlagBits::eMemoryWrite, vk::AccessFlags {},
+			                  vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
+			                  vk::AccessFlagBits::eHostRead);
+			cursor += AlignDownload(envelope_size);
 		}
-	});
-	return true;
+		download.Commit();
+		const auto completion_tick = m_scheduler.CurrentTick();
+		m_scheduler.Finish();
+		m_scheduler.WaitPriorityOperations(completion_tick);
+		cursor = 0;
+		for (const auto& copy: batch) {
+			const auto [source_begin, envelope_size] = DownloadEnvelope(copy);
+			const auto offset = cursor + copy.source_offset - source_begin;
+			download.Invalidate(base_offset + offset, copy.size);
+			Libs::LibKernel::Memory::WriteBacking(copy.address, mapped + offset, copy.size);
+			cursor += AlignDownload(envelope_size);
+		}
+		batch.clear();
+		packed_size = 0;
+	};
+	for (auto copy: copies) {
+		while (copy.size != 0) {
+			const auto available = download.Size() - packed_size;
+			const auto prefix    = copy.source_offset & 3u;
+			const auto bytes     = std::min(copy.size, available - prefix);
+			DownloadCopy part {copy.buffer, copy.source_offset, copy.address, bytes};
+			const auto [source_begin, envelope_size] = DownloadEnvelope(part);
+			(void)source_begin;
+			packed_size += AlignDownload(envelope_size);
+			batch.push_back(part);
+			copy.source_offset += bytes;
+			copy.address += bytes;
+			copy.size -= bytes;
+			if (packed_size == download.Size()) {
+				flush();
+			}
+		}
+	}
+	if (!batch.empty()) {
+		flush();
+	}
+	for (const auto& copy: copies) {
+		m_gpu_modified_ranges.Subtract(copy.address, copy.size);
+	}
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
@@ -183,7 +199,7 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
       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, 64 * MiB),
+      m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
       m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
       m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
@@ -236,29 +252,44 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size, bool is_write) {
 		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 		     vaddr, size);
 	}
-	m_scheduler.Context().GetGpu().SendCommandSync([this, vaddr, size, is_write] {
-		if (is_write && !IsRegionRegistered(vaddr, size)) {
-			return;
-		}
-		auto& buffer = m_slot_buffers[FindBuffer(vaddr, size)];
+	m_scheduler.Context().GetGpu().SendCommandSync(
+	    [this, vaddr, size, is_write] { ReadMemoryOnGpu(vaddr, size, is_write); });
+}
 
-		// Widen nearby CPU reads so they share one GPU drain.
-		constexpr uint64_t WindowSize   = 512 * 1024;
-		const auto         buffer_begin = buffer.CpuAddress();
-		const auto         buffer_end   = buffer_begin + buffer.Size();
-		const auto window_begin = std::max(Common::AlignDown(vaddr, WindowSize), buffer_begin);
-		const auto window_end = std::min(std::max(window_begin + WindowSize, vaddr + size), buffer_end);
+void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
+	if (is_write && !IsRegionRegistered(vaddr, size)) {
+		return;
+	}
+	auto& buffer = m_slot_buffers[FindBuffer(vaddr, size)];
 
-		if (DownloadBufferMemory(buffer, window_begin, window_end - window_begin)) {
-			const auto tick = m_scheduler.CurrentTick();
-			m_scheduler.Wait(tick);
-			m_scheduler.WaitPriorityOperations(tick);
-			m_memory_tracker.UnmarkRegionAsGpuModified(window_begin, window_end - window_begin);
-		}
-		if (is_write) {
-			m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
-		}
-	});
+	// Widen nearby CPU reads so they share one GPU drain.
+	constexpr uint64_t WindowSize   = 512 * 1024;
+	const auto         buffer_begin = buffer.CpuAddress();
+	const auto         buffer_end   = buffer_begin + buffer.Size();
+	const auto         window_begin = std::max(vaddr & ~(WindowSize - 1), buffer_begin);
+	const auto window_end = std::min(std::max(window_begin + WindowSize, vaddr + size), buffer_end);
+
+	std::vector<DownloadCopy> copies;
+	m_memory_tracker.ForEachDownloadRange<false>(
+	    window_begin, window_end - window_begin,
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+		                                           "memory invalidation");
+	    },
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_gpu_modified_ranges.ForEachIntersection(address, bytes, [&](RangeSet::Range range) {
+			    copies.push_back(
+			        {&buffer, buffer.Offset(range.address), range.address, range.size});
+		    });
+	    });
+	if (!copies.empty()) {
+		DownloadBufferMemory(copies);
+		// The enumeration covered whole dirty pages and every exact interval on them.
+		m_memory_tracker.UnmarkRegionAsGpuModified(window_begin, window_end - window_begin);
+	}
+	if (is_write) {
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+	}
 }
 
 BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
@@ -309,7 +340,7 @@ BufferCache::OverlapResult BufferCache::ResolveOverlaps(uint64_t vaddr, uint64_t
 		end                       = std::max(end, buffer_end);
 		if (!has_stream_leap && (stream_score += buffer.StreamScore()) > StreamLeapThreshold) {
 			has_stream_leap = true;
-			// Reserve space in the incoming stream's direction of growth.
+			// Fix the shadPS4 bug that reserves space opposite to the incoming stream's growth.
 			// The old buffer extending left of the request predicts growth to the right, and vice versa.
 			if (expands_left) {
 				end += std::min(StreamLeapSize, PageTable::kAddressSpaceSize - end);
@@ -340,8 +371,8 @@ void BufferCache::JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumul
 
 BufferId BufferCache::CreateBuffer(uint64_t vaddr, uint64_t size) {
 	EXIT_IF(m_scheduler.Current().IsInvalid());
-	const auto end = Common::AlignUp(vaddr + size, CACHING_PAGESIZE);
-	vaddr = Common::AlignDown(vaddr, CACHING_PAGESIZE);
+	const auto end = (vaddr + size + CACHING_PAGESIZE - 1) & ~(CACHING_PAGESIZE - 1);
+	vaddr &= ~(CACHING_PAGESIZE - 1);
 	size               = end - vaddr;
 	const auto overlap = ResolveOverlaps(vaddr, size);
 
@@ -454,16 +485,17 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBuffer(uint64_t vaddr, uint64_t
 		}
 	}
 
-	if (IsBufferInvalid(id) || !m_slot_buffers[id].IsInBounds(vaddr, size)) {
-		id = FindBuffer(vaddr, size);
+	auto* buffer = m_slot_buffers.try_get(id);
+	if (buffer == nullptr || buffer->is_deleted || !buffer->IsInBounds(vaddr, size)) {
+		id     = FindBuffer(vaddr, size);
+		buffer = &m_slot_buffers[id];
 	}
-	auto& buffer = m_slot_buffers[id];
-	TouchBuffer(buffer);
-	(void)SynchronizeBuffer(buffer, vaddr, size, is_written, is_texel_buffer);
+	TouchBuffer(*buffer);
+	(void)SynchronizeBuffer(*buffer, vaddr, size, is_written, is_texel_buffer);
 	if (is_written) {
 		m_gpu_modified_ranges.Add(vaddr, size);
 	}
-	return {&buffer, buffer.Offset(vaddr)};
+	return {buffer, buffer->Offset(vaddr)};
 }
 
 std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t size) {
@@ -515,7 +547,9 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 	}
 
 	m_texture_cache.InvalidateMemoryFromGPU(vaddr, size);
-	auto [dst, dst_offset] = ObtainBuffer(vaddr, size, true, true);
+	const auto id          = FindBuffer(vaddr, size);
+	auto [dst, dst_offset] = ObtainBuffer(vaddr, size, true, true, id);
+	EXIT_IF(dst == nullptr);
 	dst->Fill(dst_offset, size, value);
 }
 
@@ -549,6 +583,10 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	                                    : std::pair {&m_gds_buffer, src_vaddr};
 	auto [dst, dst_offset] = dst_memory ? ObtainBuffer(dst_vaddr, size, true, true, dst_id)
 	                                    : std::pair {&m_gds_buffer, dst_vaddr};
+	EXIT_IF(src == nullptr || dst == nullptr);
+	if (src == dst && src_offset < dst_offset + size && dst_offset < src_offset + size) {
+		EXIT("BufferCache: resolved Vulkan copy ranges overlap\n");
+	}
 	dst->CopyFrom(command, *src, src_offset, dst_offset, size);
 }
 
@@ -592,7 +630,8 @@ void BufferCache::RunGarbageCollector() {
 	const size_t   limit      = aggressive ? 64 : 32;
 
 	std::vector<BufferId> dirty_buffers;
-	size_t                retire_count = 0;
+	std::vector<DownloadCopy> copies;
+	size_t                    retire_count = 0;
 	m_lru_cache.ForEachItemBelow(tick - age, [&](BufferId id) {
 		auto& buffer = m_slot_buffers[id];
 		EXIT_IF(buffer.is_deleted);
@@ -603,7 +642,19 @@ void BufferCache::RunGarbageCollector() {
 			return false;
 		}
 		if (dirty) {
-			EXIT_IF(!DownloadBufferMemory(buffer, buffer.CpuAddress(), buffer.Size()));
+			m_memory_tracker.ForEachDownloadRange<false>(
+			    buffer.CpuAddress(), buffer.Size(),
+			    [&](uint64_t dirty_address, uint64_t dirty_size) noexcept {
+				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, dirty_address,
+				                                           dirty_size, "garbage collection");
+			    },
+			    [&](uint64_t dirty_address, uint64_t dirty_size) noexcept {
+				    m_gpu_modified_ranges.ForEachIntersection(
+				        dirty_address, dirty_size, [&](RangeSet::Range range) {
+					    copies.push_back({&buffer, range.address - buffer.CpuAddress(),
+					                      range.address, range.size});
+				        });
+				});
 			dirty_buffers.push_back(id);
 		} else {
 			m_memory_tracker.UntrackMemory(buffer.CpuAddress(), buffer.Size());
@@ -615,10 +666,8 @@ void BufferCache::RunGarbageCollector() {
 		return;
 	}
 
-	// Publish all queued downloads before releasing their tracked pages and owners.
-	const auto completion_tick = m_scheduler.CurrentTick();
-	m_scheduler.Wait(completion_tick);
-	m_scheduler.WaitPriorityOperations(completion_tick);
+	EXIT_IF(copies.empty());
+	DownloadBufferMemory(copies);
 	for (const auto id: dirty_buffers) {
 		auto& buffer = m_slot_buffers[id];
 		m_memory_tracker.UnmarkRegionAsGpuModified(buffer.CpuAddress(), buffer.Size());

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

@@ -74,11 +74,6 @@ public:
 private:
 	friend struct BufferCacheTestAccess;
 
-	bool IsBufferInvalid(BufferId id) const {
-		const auto* buffer = m_slot_buffers.try_get(id);
-		return buffer == nullptr || buffer->is_deleted;
-	}
-
 	using BufferMap = std::map<uint64_t, BufferId>;
 	struct OverlapResult {
 		BufferMap::iterator first;
@@ -88,8 +83,14 @@ private:
 		bool                has_stream_leap;
 	};
 
+	struct DownloadCopy;
 	using PageTable = MultiLevelPageTable<BufferId, CACHING_PAGEBITS, 40, 16>;
 	static_assert(CACHING_PAGESIZE == (uint64_t {1} << PageTable::kPageBits));
+	static constexpr uint64_t               DOWNLOAD_ALIGNMENT = 64;
+	[[nodiscard]] static constexpr uint64_t AlignDownload(uint64_t size) noexcept {
+		return (size + DOWNLOAD_ALIGNMENT - 1) & ~(DOWNLOAD_ALIGNMENT - 1);
+	}
+	[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
 	void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size);
 	void TouchBuffer(const Buffer& buffer);
 	[[nodiscard]] OverlapResult ResolveOverlaps(uint64_t vaddr, uint64_t size);
@@ -105,8 +106,8 @@ private:
 	[[nodiscard]] vk::Buffer UploadCopies(Buffer& buffer, std::span<vk::BufferCopy> copies,
 	                                      uint64_t total_size);
 	[[nodiscard]] bool SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uint64_t size);
-	// Queues backing publication; callers wait before clearing dirty pages or reusing their data.
-	[[nodiscard]] bool DownloadBufferMemory(Buffer& buffer, uint64_t vaddr, uint64_t size);
+	void DownloadBufferMemory(std::span<const DownloadCopy> copies);
+	void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);
 
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;

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

@@ -44,7 +44,12 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	                                                &m_fault_process_desc_layout),
 	    "create fault-buffer descriptor layout");
 
-	const auto module = CompileSPV(FAULT_BUFFER_PROCESS_SPV, m_graphics.device);
+	vk::ShaderModuleCreateInfo module_info {};
+	module_info.codeSize = std::size(FAULT_BUFFER_PROCESS_SPV) * sizeof(uint32_t);
+	module_info.pCode    = FAULT_BUFFER_PROCESS_SPV;
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
+	                     "create fault-buffer shader module");
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
 	pipeline_layout_info.setLayoutCount = 1;

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

@@ -0,0 +1,96 @@
+#include "graphics/host_gpu/renderer/cache/gpuResourceManager.h"
+
+#include "common/assert.h"
+#include "graphics/guest_gpu/graphicsRun.h"
+#include "graphics/host_gpu/renderer/commandScheduler.h"
+namespace Libs::Graphics {
+
+GpuResourceManager::GpuResourceManager(GraphicContext& graphics, CommandScheduler& scheduler)
+    : m_scheduler(scheduler), m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache),
+      m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache) {}
+
+GpuResourceManager::~GpuResourceManager() = default;
+
+bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
+	// The host reports the faulting byte, not the instruction's access width. Both caches
+	// resolve its page; guessing a width can cross the end of a valid guest mapping.
+	constexpr uint64_t fault_size = 1;
+	if (!IsMapped(fault_vaddr, fault_size)) {
+		return false;
+	}
+	if (access == PageFaultAccess::Write) {
+		m_buffer_cache.InvalidateMemory(fault_vaddr, fault_size);
+		m_texture_cache.InvalidateMemory(fault_vaddr, fault_size);
+	} else {
+		m_buffer_cache.ReadMemory(fault_vaddr, fault_size);
+	}
+	return true;
+}
+
+bool GpuResourceManager::InvalidateMemory(uint64_t vaddr, uint64_t size) {
+	if (!IsMapped(vaddr, size)) {
+		return false;
+	}
+	m_buffer_cache.InvalidateMemory(vaddr, size);
+	m_texture_cache.InvalidateMemory(vaddr, size);
+	return true;
+}
+
+bool GpuResourceManager::IsMapped(uint64_t vaddr, uint64_t size) const noexcept {
+	if (!GuestRange {vaddr, size}.Valid()) {
+		return false;
+	}
+	std::shared_lock lock(m_mapped_ranges_mutex);
+	return m_mapped_ranges.Contains(vaddr, size);
+}
+
+void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size) {
+	{
+		std::lock_guard lock(m_mapped_ranges_mutex);
+		m_mapped_ranges.Add(vaddr, size);
+	}
+}
+
+void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
+	if (CommandScheduler::InDeferredOperation()) {
+		EXIT("unsupported memory unmap from an asynchronous GPU completion, "
+		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
+		     vaddr, size);
+	}
+	const auto unmap = [this, vaddr, size] {
+		if (m_scheduler.Active()) {
+			const auto tick = m_scheduler.CurrentTick();
+			m_scheduler.Finish();
+			m_scheduler.WaitPriorityOperations(tick);
+		}
+		m_buffer_cache.InvalidateMemory(vaddr, size);
+		m_texture_cache.UnmapMemory(vaddr, size);
+		std::lock_guard lock(m_mapped_ranges_mutex);
+		m_mapped_ranges.Subtract(vaddr, size);
+	};
+	if (m_gpu == nullptr) {
+		unmap();
+		return;
+	}
+	m_gpu->SendCommandSync(unmap);
+}
+
+void GpuResourceManager::PrepareBda() {
+	std::shared_lock lock(m_mapped_ranges_mutex);
+	m_mapped_ranges.ForEach([this](uint64_t start, uint64_t end) {
+		m_buffer_cache.SynchronizeBuffersInRange(start, end - start);
+	});
+	m_fault_process_pending = true;
+}
+
+void GpuResourceManager::RunGarbageCollector() {
+	if (m_fault_process_pending) {
+		m_fault_process_pending = false;
+		m_buffer_cache.ProcessFaultBuffer();
+	}
+	m_texture_cache.ProcessDownloadImages();
+	m_texture_cache.RunGarbageCollector();
+	m_buffer_cache.RunGarbageCollector();
+}
+
+} // namespace Libs::Graphics

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

@@ -0,0 +1,49 @@
+#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_GPURESOURCEMANAGER_H_
+#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_GPURESOURCEMANAGER_H_
+
+#include "common/abi.h"
+#include "common/common.h"
+#include "graphics/host_gpu/pageManager.h"
+#include "graphics/host_gpu/renderer/cache/bufferCache.h"
+#include "graphics/host_gpu/renderer/cache/textureCache.h"
+
+#include <cstdint>
+#include <shared_mutex>
+
+namespace Libs::Graphics {
+
+class CommandScheduler;
+class GuestGpu;
+
+class GpuResourceManager {
+public:
+	GpuResourceManager(GraphicContext& graphics, CommandScheduler& scheduler);
+	~GpuResourceManager();
+	KYTY_CLASS_NO_COPY(GpuResourceManager);
+
+	[[nodiscard]] BufferCache&  GetBufferCache() { return m_buffer_cache; }
+	[[nodiscard]] TextureCache& GetTextureCache() { return m_texture_cache; }
+	void                        SetGpu(GuestGpu* gpu) noexcept { m_gpu = gpu; }
+
+	[[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);
+	void               UnmapMemory(uint64_t vaddr, uint64_t size);
+	void               PrepareBda();
+	void               RunGarbageCollector();
+
+private:
+	PageManager               m_page_manager;
+	CommandScheduler&         m_scheduler;
+	BufferCache               m_buffer_cache;
+	TextureCache              m_texture_cache;
+	mutable std::shared_mutex m_mapped_ranges_mutex;
+	RangeSet                  m_mapped_ranges;
+	GuestGpu*                 m_gpu = nullptr;
+	bool                      m_fault_process_pending = false;
+};
+
+} // namespace Libs::Graphics
+
+#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_GPURESOURCEMANAGER_H_

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

@@ -1,12 +1,12 @@
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/profiler.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 
 #include <cstring>
+#include <limits>
 #include <numeric>
 #include <vk_mem_alloc.h>
 
@@ -45,11 +45,16 @@ constexpr size_t WATCHES_RESERVE_CHUNK   = 0x1000;
 		result = value;
 		return true;
 	}
-	const auto aligned = Common::AlignUp(value, alignment);
-	if (aligned < value) {
+	const auto remainder = value % alignment;
+	if (remainder == 0) {
+		result = value;
+		return true;
+	}
+	const auto increment = alignment - remainder;
+	if (value > std::numeric_limits<uint64_t>::max() - increment) {
 		return false;
 	}
-	result = aligned;
+	result = value + increment;
 	return true;
 }
 
@@ -119,6 +124,12 @@ bool Buffer::IsInBounds(uint64_t address, uint64_t size) const noexcept {
 	return address >= m_cpu_address && size <= Size() && address - m_cpu_address <= Size() - size;
 }
 
+void Buffer::Write(uint64_t offset, const void* source, uint64_t size) {
+	EXIT_IF(source == nullptr || m_mapped.empty() || offset > Size() || size > Size() - offset);
+	std::memcpy(m_mapped.data() + offset, source, static_cast<size_t>(size));
+	Flush(offset, size);
+}
+
 void Buffer::Flush(uint64_t offset, uint64_t size) {
 	EXIT_IF(m_mapped.empty() || offset > Size() || size > Size() - offset);
 	if (!IsCoherent() && size != 0) {

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

@@ -55,6 +55,7 @@ public:
 	[[nodiscard]] bool IsInBounds(uint64_t address, uint64_t size) const noexcept;
 	void               IncreaseStreamScore(int score) noexcept { stream_score += score; }
 	[[nodiscard]] int  StreamScore() const noexcept { return stream_score; }
+	void               Write(uint64_t offset, const void* source, uint64_t size);
 	void               Flush(uint64_t offset, uint64_t size);
 	void               Invalidate(uint64_t offset, uint64_t size);
 	void CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t source_offset,

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

@@ -13,7 +13,6 @@
 #include "graphics/host_gpu/renderer/image/tiler.h"
 
 #include <map>
-#include <type_traits>
 #include <unordered_map>
 #include <unordered_set>
 #include <vector>
@@ -64,8 +63,11 @@ public:
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t address, uint64_t size);
 
 	[[nodiscard]] bool IsMeta(uint64_t address);
-	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice);
+	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice,
+	                                 uint32_t* fill_value = nullptr);
 	[[nodiscard]] bool ClearMeta(uint64_t address);
+	// Record deferred DCC state while the original guest dispatch writes the metadata.
+	void               TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
 	void UnmapMemory(uint64_t address, uint64_t size);
@@ -74,14 +76,21 @@ public:
 
 private:
 	enum class TransferDirection { Upload, Download };
-	struct TextureTransfer;
-	struct ImageDownload;
+	struct TextureTransferPlan;
+	struct DownloadPlan;
 
 	struct MetaDataInfo {
-		enum class Type : uint8_t { CMask, FMask, HTile };
+		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
+		// PendingDcc retains that exact fill until an image binding classifies the address,
+		// without exposing an unconfirmed buffer address to the normal metadata heuristics.
+		// Keep all surface metadata in one entry so CMask/FMask can be
+		// registered beside HTile and DCC without introducing parallel tracking paths.
+		enum class Type : uint8_t { PendingDcc, CMask, FMask, HTile, Dcc };
 
-		Type     type;
-		uint32_t clear_mask = UINT32_MAX;
+		Type     type       = Type::PendingDcc;
+		uint32_t clear_mask = 0;
+		uint32_t fill_value = 0xffffffffu;
+		uint64_t fill_size  = 0;
 	};
 
 	struct OverlapResult {
@@ -93,23 +102,6 @@ private:
 	using ImageIds       = InlinePageOwnerList<ImageId, 16>;
 	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
 
-	// Callers have validated the nonempty 40-bit range with TryGetPageRange.
-	template <typename Func>
-	static void ForEachPage(uint64_t address, size_t size, Func&& func) {
-		using FuncReturn = typename std::invoke_result<Func, uint64_t>::type;
-		static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>;
-		const uint64_t page_end = (address + size - 1) >> ImagePageTable::kPageBits;
-		for (uint64_t page = address >> ImagePageTable::kPageBits; page <= page_end; ++page) {
-			if constexpr (RETURNS_BOOL) {
-				if (func(page)) {
-					break;
-				}
-			} else {
-				func(page);
-			}
-		}
-	}
-
 	[[nodiscard]] ImageId     InsertImage(const ImageInfo& info);
 	[[nodiscard]] ImageId     GetNullImage(const ImageDesc& desc);
 	void                      RegisterImage(ImageId id);
@@ -139,30 +131,29 @@ private:
 	                                                ImageId cached);
 	[[nodiscard]] ImageId       ExpandImage(const ImageInfo& info, ImageId source);
 	void                        RefreshImage(ImageId id);
-	void                        MaterializeColorClear(ImageId id, const ImageDesc& desc,
-	                                                uint32_t metadata_base_layer);
+	void                        PrepareDccClear(ImageId id, const ImageDesc& desc);
 	void                        InitializeImage(ImageId id);
-	[[nodiscard]] TextureTransfer
+	[[nodiscard]] TextureTransferPlan
 	BuildTextureTransfer(const Image& image, BindingType binding, TransferDirection direction) const;
-	[[nodiscard]] ImageDownload BuildDownload(const Image& image) const;
+	[[nodiscard]] DownloadPlan BuildDownload(const Image& image) const;
 	void UploadImage(Image& image, Buffer& source, uint64_t source_offset);
-	void DownloadImage(Image& image, Buffer& destination, uint64_t destination_offset,
-	                       uint64_t destination_size, ImageDownload transfer);
+	void DownloadImageData(Image& image, Buffer& destination, uint64_t destination_offset,
+	                       uint64_t destination_size, DownloadPlan plan);
 	void DownloadDepth(Image& image, Buffer& destination, uint64_t destination_offset);
 	void CommitGpuWrite(Image& image);
 	// Caller holds m_lock. Volume layer ranges select depth slices.
-	void ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
-	                const vk::ImageSubresourceRange& range, const vk::ClearValue& clear);
+	void ClearImage(CommandBuffer& command, ImageId id, const vk::ImageSubresourceRange& range,
+	                const vk::ClearValue& clear);
 	void PrepareImageCopy(Image& image);
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);
 	void               CopyImage(ImageId destination, ImageId source);
-	[[nodiscard]] ImageId AssociateStencil(ImageId depth, GuestRange stencil);
+	void               AssociateStencil(ImageId depth, GuestRange stencil);
 	void CopyImageMip(ImageId destination, ImageId source, uint32_t mip, uint32_t layer);
 	void ValidateImageDesc(const ImageDesc& desc) const;
 
 	void               InvalidateCpuAliases(uint64_t address, uint64_t size);
-	[[nodiscard]] bool DownloadImageMemory(ImageId id);
+	[[nodiscard]] bool TryDownloadImage(ImageId id);
 
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -17,7 +17,6 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
-#include <bit>
 
 namespace Libs::Graphics {
 
@@ -139,7 +138,8 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 	// Color-control state selects the color-buffer operation and logical blend operation.
 	// The normal copy operation is a regular color write, not an attachment clear.
 	// Nonlinear clear values are still stored as normalized components.
-	// Metadata clears are materialized during image discovery; render-pass loads preserve contents.
+	// Fast color clears are metadata driven and must be handled explicitly when
+	// that metadata path is implemented; render-pass load must preserve contents.
 	uint32_t   width  = 0;
 	uint32_t   height = 0;
 	uint32_t   pitch  = 0;
@@ -160,6 +160,9 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 			break;
 		default: EXIT("unknown tile mode: %u\n", static_cast<uint32_t>(rt.attrib3.tile_mode));
 	}
+	if (!tile && levels > 1) {
+		EXIT("linear mipmapped render targets are unsupported\n");
+	}
 	if (samples > 1 && (!tile || levels != 1)) {
 		EXIT("multisampled render targets require a single-mip tiled surface\n");
 	}
@@ -200,7 +203,8 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 		     " layer=%u/%u\n",
 		     rt.attrib3.dimension, rt.attrib3.depth, view.base_layer, view.image_layers);
 	}
-	if (samples == 1) {
+	// PPSA28068: the linear EULA target and its sampled view must share the padded pitch.
+	if (!tile || volume || texture_tile) {
 		pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
 	} else {
 		pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
@@ -228,26 +232,37 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 		}
 		size         = volume_layout.block_slice_size;
 		backing_size = volume_layout.total_size;
-	} else {
+	} else if (tile) {
 		TileSizeAlign layout {};
 		bool          valid_layout = false;
-		if (samples == 1) {
+		if (texture_tile) {
 			TileGetTextureSize(transfer_format, width, height, levels, rt.attrib3.tile_mode,
 			                   &layout, mip_sizes, mip_padded);
-			valid_layout = layout.size != 0 && layout.align != 0 &&
-			               (rt.attrib3.tile_mode != Prospero::TileMode::kRenderTarget ||
-			                levels <= std::bit_width(std::max(width, height)));
+			valid_layout = layout.size != 0 && layout.align == texture_tile_layout.block.block_size;
 		} else {
-			valid_layout = TileGetRenderTargetSize(width, height, pitch, bytes_per_element,
-			                                       layout, rt.attrib.num_fragments);
-			mip_sizes[0]  = {layout.size, 0, 0, 0, 0, 0};
-			mip_padded[0] = {pitch, height};
+			valid_layout =
+			    levels == 1 ? TileGetRenderTargetSize(width, height, pitch, bytes_per_element,
+			                                          layout, rt.attrib.num_fragments)
+			                : TileGetRenderTargetMipLayout(width, height, pitch, bytes_per_element,
+			                                               levels, layout, mip_sizes, mip_padded);
 		}
 		if (!valid_layout) {
 			EXIT("unsupported render-target layout: %ux%u pitch=%u bytes=%u levels=%u\n", width,
 			     height, pitch, bytes_per_element, levels);
 		}
 		size = layout.size;
+		EXIT_IF(size > UINT32_MAX);
+		if (levels == 1) {
+			mip_sizes[0]  = {static_cast<uint32_t>(size), 0, 0, 0, 0, 0};
+			mip_padded[0] = {pitch, height};
+		}
+	} else {
+		size = static_cast<uint64_t>(pitch) * height * bytes_per_element * samples;
+		if (size > UINT32_MAX) {
+			EXIT("linear render-target slice exceeds the supported layout size\n");
+		}
+		mip_sizes[0]  = {static_cast<uint32_t>(size), 0, 0, 0, 0, 0};
+		mip_padded[0] = {pitch, height};
 	}
 	if (size == 0 || (!volume && size > UINT64_MAX / view.image_layers)) {
 		EXIT("render-target memory footprint is invalid\n");
@@ -289,25 +304,15 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 	desc.info.samples         = samples;
 	desc.info.tile_mode       = rt.attrib3.tile_mode;
 	const bool has_dcc        = rt.info.dcc_compression_enable && rt.dcc_addr.addr != 0;
-	const bool has_cmask      = !rt.info.dcc_compression_enable && rt.info.cmask_fast_clear_enable &&
-	                            rt.cmask.addr != 0 && samples == 1 &&
-	                            !rt.info.fmask_compression_enable &&
-	                            !rt.attrib3.write_vrs_rate_hint_to_cmask;
-	if (has_dcc || has_cmask) {
+	if (has_dcc) {
 		TileSizeAlign metadata_size {};
-		const auto layers = volume ? depth : view.image_layers;
-		if (has_dcc) {
-			(void)TileGetDccSize(width, height, layers, bytes_per_element, levels,
-			                     rt.attrib3.tile_mode, metadata_size, rt.attrib.num_fragments);
-			desc.info.metadata.dcc_alpha_msb = DccAlphaOnMsb(rt.info);
-		} else {
-			(void)TileGetCmaskSize(width, height, layers, levels, metadata_size);
-		}
-		// DCC owns the clear when both planes are enabled; single-sample CMASK stays expanded.
-		desc.info.metadata.kind = has_dcc ? ImageMetadataKind::Dcc : ImageMetadataKind::Cmask;
-		desc.info.metadata.range = {has_dcc ? rt.dcc_addr.addr : rt.cmask.addr, metadata_size.size};
-		desc.info.metadata.clear_word           = rt.clear_word0.word0;
-		desc.info.metadata.clear_register_valid = true;
+		(void)TileGetDccSize(width, height, volume ? depth : view.image_layers, bytes_per_element,
+		                     levels, rt.attrib3.tile_mode, metadata_size, rt.attrib.num_fragments);
+		desc.info.metadata.kind                     = ImageMetadataKind::Dcc;
+		desc.info.metadata.range                    = {rt.dcc_addr.addr, metadata_size.size};
+		desc.info.metadata.dcc_clear_word           = rt.clear_word0.word0;
+		desc.info.metadata.dcc_clear_register_valid = true;
+		desc.info.metadata.dcc_alpha_msb            = DccAlphaOnMsb(rt.info);
 	}
 	for (uint32_t level = 0; level < levels; level++) {
 		if (volume) {

src/graphics/host_gpu/renderer/context.cpp

@@ -61,11 +61,11 @@ void CommandBuffer::SetDebugInfo(uint32_t op, uint64_t submit_id, uint32_t arg0,
 }
 
 void CommandBuffer::BeginRendering(const RenderState& state) const {
+	EXIT_IF(state.width == 0 || state.height == 0 || state.num_layers == 0 ||
+	        state.num_color_attachments > RENDER_COLOR_ATTACHMENTS_MAX);
 	if (m_rendering && m_render_state == state) {
 		return;
 	}
-	EXIT_IF(state.width == 0 || state.height == 0 || state.num_layers == 0 ||
-	        state.num_color_attachments > RENDER_COLOR_ATTACHMENTS_MAX);
 	EndRendering();
 
 	std::array<vk::RenderingAttachmentInfo, RENDER_COLOR_ATTACHMENTS_MAX> colors {};

src/graphics/host_gpu/renderer/debug.cpp

@@ -4,6 +4,7 @@
 #include "common/common.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
+#include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/host_gpu/renderer/render.h"
@@ -56,7 +57,7 @@ uint32_t render_target_first_bound_slot(const CommandBuffer& buffer) {
 
 bool graphics_debug_dump_enabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
-	       Config::GetPrintfDirection() != Config::LogDirection::Silent;
+	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
 }
 
 void uc_print(const char* func, const HW::UserConfig& uc) {
@@ -88,75 +89,91 @@ void uc_check(const HW::UserConfig& uc) {
 	EXIT_NOT_IMPLEMENTED(user_en.vgpr3 != false);
 }
 
-std::string rt_print(const char* func, const HW::RenderTarget& rt) {
-	std::string dst;
-	dst.reserve(4096);
+void sh_print(const char* func, const HW::Shader& /*uc*/) {
+	LOGF("%s\n", func);
+}
 
-	dst += fmt::format("{}\n", func);
+std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt) {
+	std::vector<std::string> dst;
+	dst.reserve(40);
 
-	dst += fmt::format("\t base.addr                       = 0x{:016x}\n", rt.base.addr);
-	dst += fmt::format("\t view.base_array_slice_index     = 0x{:08x}\n",
-	                  rt.view.base_array_slice_index);
-	dst += fmt::format("\t view.last_array_slice_index     = 0x{:08x}\n",
-	                  rt.view.last_array_slice_index);
-	dst += fmt::format("\t view.current_mip_level          = 0x{:08x}\n", rt.view.current_mip_level);
-	dst += fmt::format("\t info.fmask_compression_enable   = {}\n",
-	                  rt.info.fmask_compression_enable ? "true" : "false");
+	dst.push_back(fmt::format("{}\n", func));
 
-	dst += fmt::format("\t info.fmask_data_compression_disable = {}\n",
-	                  rt.info.fmask_data_compression_disable ? "true" : "false");
-	dst += fmt::format("\t info.fmask_one_frag_mode        = {}\n",
-	                  rt.info.fmask_one_frag_mode ? "true" : "false");
+	dst.push_back(fmt::format("\t base.addr                       = 0x{:016x}\n", rt.base.addr));
+	dst.push_back(fmt::format("\t view.base_array_slice_index     = 0x{:08x}\n",
+	                          rt.view.base_array_slice_index));
+	dst.push_back(fmt::format("\t view.last_array_slice_index     = 0x{:08x}\n",
+	                          rt.view.last_array_slice_index));
+	dst.push_back(
+	    fmt::format("\t view.current_mip_level          = 0x{:08x}\n", rt.view.current_mip_level));
+	dst.push_back(fmt::format("\t info.fmask_compression_enable   = {}\n",
+	                          rt.info.fmask_compression_enable ? "true" : "false"));
 
-	dst += fmt::format("\t info.cmask_fast_clear_enable    = {}\n",
-	                  rt.info.cmask_fast_clear_enable ? "true" : "false");
-	dst += fmt::format("\t info.dcc_compression_enable     = {}\n",
-	                  rt.info.dcc_compression_enable ? "true" : "false");
-	dst += fmt::format("\t info.format                     = 0x{:08x}\n",
-	                  static_cast<uint32_t>(rt.info.format));
-	dst += fmt::format("\t info.channel_type               = 0x{:08x}\n",
-	                  static_cast<uint32_t>(rt.info.channel_type));
-	dst += fmt::format("\t info.channel_order              = 0x{:08x}\n",
-	                  static_cast<uint32_t>(rt.info.channel_order));
-	dst += fmt::format("\t info.blend_bypa                 = {}\n",
-	                  rt.info.blend_bypass ? "true" : "false");
-	dst += fmt::format("\t info.blend_clamp                = {}\n",
-	                  rt.info.blend_clamp ? "true" : "false");
-	dst += fmt::format("\t info.round_mode                 = {}\n",
-	                  rt.info.round_mode ? "true" : "false");
-	dst += fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
-	                  rt.attrib.force_dest_alpha_to_one ? "true" : "false");
-	dst += fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples);
-	dst += fmt::format("\t attrib.num_fragments            = 0x{:08x}\n", rt.attrib.num_fragments);
-	dst += fmt::format("\t attrib2.width                   = 0x{:08x}\n", rt.attrib2.width);
-	dst += fmt::format("\t attrib2.height                  = 0x{:08x}\n", rt.attrib2.height);
-	dst += fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels);
-	dst += fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth);
-	dst += fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
-	                  static_cast<uint32_t>(rt.attrib3.tile_mode));
-	dst += fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension);
-	dst += fmt::format("\t attrib3.metadata_pipe_aligned   = {}\n",
-	                  rt.attrib3.metadata_pipe_aligned ? "true" : "false");
-	dst += fmt::format("\t attrib3.write_vrs_rate_hint_to_cmask = {}\n",
-	                  rt.attrib3.write_vrs_rate_hint_to_cmask ? "true" : "false");
-	dst += fmt::format("\t dcc.max_uncompressed_block_size = 0x{:08x}\n",
-	                  rt.dcc.max_uncompressed_block_size);
-	dst += fmt::format("\t dcc.max_compressed_block_size   = 0x{:08x}\n",
-	                  rt.dcc.max_compressed_block_size);
-	dst += fmt::format("\t dcc.color_transform             = 0x{:08x}\n", rt.dcc.color_transform);
-	dst += fmt::format("\t dcc.overwrite_combiner_disable  = {}\n",
-	                  rt.dcc.overwrite_combiner_disable ? "true" : "false");
-	dst += fmt::format("\t dcc.independent_block_size      = 0x{:02x}\n",
-	                  static_cast<uint8_t>(rt.dcc.independent_block_size));
-	dst += fmt::format("\t data_write_on_dcc_clear_to_reg  = {}\n",
-	                  rt.dcc.data_write_on_dcc_clear_to_reg ? "true" : "false");
-	dst += fmt::format("\t dcc.dcc_clear_key_enable        = {}\n",
-	                  rt.dcc.dcc_clear_key_enable ? "true" : "false");
-	dst += fmt::format("\t cmask.addr                      = 0x{:016x}\n", rt.cmask.addr);
-	dst += fmt::format("\t fmask.addr                      = 0x{:016x}\n", rt.fmask.addr);
-	dst += fmt::format("\t clear_word0.word0               = 0x{:08x}\n", rt.clear_word0.word0);
-	dst += fmt::format("\t clear_word1.word1               = 0x{:08x}\n", rt.clear_word1.word1);
-	dst += fmt::format("\t dcc_addr.addr                   = 0x{:016x}\n", rt.dcc_addr.addr);
+	// dst.push_back(fmt::format("\t info.fmask_compression_mode     = 0x{:08x}\n", //
+	// rt.info.fmask_compression_mode));
+	dst.push_back(fmt::format("\t info.fmask_data_compression_disable = {}\n",
+	                          rt.info.fmask_data_compression_disable ? "true" : "false"));
+	dst.push_back(fmt::format("\t info.fmask_one_frag_mode        = {}\n",
+	                          rt.info.fmask_one_frag_mode ? "true" : "false"));
+
+	dst.push_back(fmt::format("\t info.cmask_fast_clear_enable    = {}\n",
+	                          rt.info.cmask_fast_clear_enable ? "true" : "false"));
+	dst.push_back(fmt::format("\t info.dcc_compression_enable     = {}\n",
+	                          rt.info.dcc_compression_enable ? "true" : "false"));
+	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.format)));
+	dst.push_back(fmt::format("\t info.channel_type               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_type)));
+	dst.push_back(fmt::format("\t info.channel_order              = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_order)));
+	dst.push_back(fmt::format("\t info.blend_bypa                 = {}\n",
+	                          rt.info.blend_bypass ? "true" : "false"));
+	dst.push_back(fmt::format("\t info.blend_clamp                = {}\n",
+	                          rt.info.blend_clamp ? "true" : "false"));
+	dst.push_back(fmt::format("\t info.round_mode                 = {}\n",
+	                          rt.info.round_mode ? "true" : "false"));
+	dst.push_back(fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
+	                          rt.attrib.force_dest_alpha_to_one ? "true" : "false"));
+	dst.push_back(
+	    fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples));
+	dst.push_back(
+	    fmt::format("\t attrib.num_fragments            = 0x{:08x}\n", rt.attrib.num_fragments));
+	dst.push_back(fmt::format("\t attrib2.width                   = 0x{:08x}\n", rt.attrib2.width));
+	dst.push_back(
+	    fmt::format("\t attrib2.height                  = 0x{:08x}\n", rt.attrib2.height));
+	dst.push_back(
+	    fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels));
+	dst.push_back(fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth));
+	dst.push_back(fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib3.tile_mode)));
+	dst.push_back(
+	    fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension));
+	dst.push_back(fmt::format("\t attrib3.metadata_pipe_aligned   = {}\n",
+	                          rt.attrib3.metadata_pipe_aligned ? "true" : "false"));
+	dst.push_back(fmt::format("\t attrib3.write_vrs_rate_hint_to_cmask = {}\n",
+	                          rt.attrib3.write_vrs_rate_hint_to_cmask ? "true" : "false"));
+	dst.push_back(fmt::format("\t dcc.max_uncompressed_block_size = 0x{:08x}\n",
+	                          rt.dcc.max_uncompressed_block_size));
+	dst.push_back(fmt::format("\t dcc.max_compressed_block_size   = 0x{:08x}\n",
+	                          rt.dcc.max_compressed_block_size));
+	dst.push_back(
+	    fmt::format("\t dcc.color_transform             = 0x{:08x}\n", rt.dcc.color_transform));
+	dst.push_back(fmt::format("\t dcc.overwrite_combiner_disable  = {}\n",
+	                          rt.dcc.overwrite_combiner_disable ? "true" : "false"));
+	dst.push_back(fmt::format("\t dcc.independent_block_size      = 0x{:02x}\n",
+	                          static_cast<uint8_t>(rt.dcc.independent_block_size)));
+	dst.push_back(fmt::format("\t data_write_on_dcc_clear_to_reg  = {}\n",
+	                          rt.dcc.data_write_on_dcc_clear_to_reg ? "true" : "false"));
+	dst.push_back(fmt::format("\t dcc.dcc_clear_key_enable        = {}\n",
+	                          rt.dcc.dcc_clear_key_enable ? "true" : "false"));
+	dst.push_back(fmt::format("\t cmask.addr                      = 0x{:016x}\n", rt.cmask.addr));
+	dst.push_back(fmt::format("\t fmask.addr                      = 0x{:016x}\n", rt.fmask.addr));
+	dst.push_back(
+	    fmt::format("\t clear_word0.word0               = 0x{:08x}\n", rt.clear_word0.word0));
+	dst.push_back(
+	    fmt::format("\t clear_word1.word1               = 0x{:08x}\n", rt.clear_word1.word1));
+	dst.push_back(
+	    fmt::format("\t dcc_addr.addr                   = 0x{:016x}\n", rt.dcc_addr.addr));
 
 	return dst;
 }
@@ -209,16 +226,28 @@ static void RtCheck(const HW::RenderTarget& rt) {
 		}
 		if (rt.info.fmask_compression_enable) {
 			EXIT_NOT_IMPLEMENTED(rt.attrib.num_samples == 0 && rt.attrib.num_fragments == 0);
-			// Native MSAA stores expanded samples, independent of FMASK metadata compression.
 			static bool logged = false;
 			if (!logged) {
-				LOGF("RenderTarget: using expanded native Vulkan MSAA samples, "
+				LOGF("RenderTarget: using native Vulkan MSAA without guest FMASK metadata, "
 				     "fmask=0x%016" PRIx64 "\n",
 				     rt.fmask.addr);
 				logged = true;
 			}
 		}
 
+		EXIT_NOT_IMPLEMENTED(rt.info.fmask_data_compression_disable != false);
+
+		if (rt.info.cmask_fast_clear_enable || rt.info.dcc_compression_enable) {
+			static bool logged = false;
+			if (!logged) {
+				LOGF("RenderTarget: temporary: ignoring PS5 color metadata fast_clear=%s dcc=%s "
+				     "cmask=0x%016" PRIx64 " dcc_addr=0x%016" PRIx64 "\n",
+				     rt.info.cmask_fast_clear_enable ? "true" : "false",
+				     rt.info.dcc_compression_enable ? "true" : "false", rt.cmask.addr,
+				     rt.dcc_addr.addr);
+				logged = true;
+			}
+		}
 		if (rt.info.blend_bypass) {
 			static bool logged = false;
 			if (!logged) {
@@ -475,7 +504,16 @@ static void BcPrint(const char* func, const HW::BlendControl& c, const HW::Blend
 	     color.red, color.green, color.blue, color.alpha, cc.mode, cc.op);
 }
 
-static void BcCheck(const HW::BlendColor& color, const HW::ColorControl& cc) {
+static void BcCheck(const HW::BlendControl& /*c*/, const HW::BlendColor& color,
+                    const HW::ColorControl& cc) {
+	// EXIT_NOT_IMPLEMENTED(c.color_srcblend != 0);
+	// EXIT_NOT_IMPLEMENTED(c.color_comb_fcn != 0);
+	// EXIT_NOT_IMPLEMENTED(c.color_destblend != 0);
+	// EXIT_NOT_IMPLEMENTED(c.alpha_srcblend != 0);
+	// EXIT_NOT_IMPLEMENTED(c.alpha_comb_fcn != 0);
+	// EXIT_NOT_IMPLEMENTED(c.alpha_destblend != 0);
+	// EXIT_NOT_IMPLEMENTED(c.separate_alpha_blend != false);
+	// EXIT_NOT_IMPLEMENTED(c.enable != false);
 	if (color.red != 0.0f || color.green != 0.0f || color.blue != 0.0f || color.alpha != 0.0f) {
 		static bool logged = false;
 		if (!logged) {
@@ -538,6 +576,29 @@ static void DPrint(const char* func, const HW::DepthControl& c, const HW::Stenci
 	     sm.stencil_opval_bf);
 }
 
+static void DCheck(const HW::DepthControl& c, const HW::StencilControl& s,
+                   const HW::StencilMask& /*sm*/) {
+	// EXIT_NOT_IMPLEMENTED(c.stencil_enable != false);
+	// EXIT_NOT_IMPLEMENTED(c.z_enable != false);
+	// EXIT_NOT_IMPLEMENTED(c.z_write_enable != false);
+	// EXIT_NOT_IMPLEMENTED(c.zfunc != 0);
+	// Back-face stencil state is handled separately when enabled.
+	// EXIT_NOT_IMPLEMENTED(c.stencilfunc != 0);
+	// EXIT_NOT_IMPLEMENTED(c.stencilfunc_bf != 0);
+	// EXIT_NOT_IMPLEMENTED(s.stencil_fail != 0);
+	// EXIT_NOT_IMPLEMENTED(s.stencil_zpass != 0);
+	// EXIT_NOT_IMPLEMENTED(s.stencil_zfail != 0);
+	// Back-face stencil ops may legitimately differ from the front-face ops.
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_testval != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_mask != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_writemask != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_opval != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_testval_bf != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_mask_bf != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_writemask_bf != 0);
+	// EXIT_NOT_IMPLEMENTED(sm.stencil_opval_bf != 0);
+}
+
 static void EqaaPrint(const char* func, const HW::EqaaControl& c) {
 	LOGF("%s\n", func);
 
@@ -843,10 +904,14 @@ void hw_check(const CommandBuffer& buffer) {
 	const auto& hw      = buffer.GetRegisters();
 	const auto  rt_slot = render_target_first_bound_slot(buffer);
 	const auto& rt      = hw.GetRenderTarget(rt_slot);
+	const auto& bc      = hw.GetBlendControl(rt_slot);
 	const auto& bclr    = hw.GetBlendColor();
 	const auto& vp      = hw.GetScreenViewport();
 	const auto& c       = hw.GetClipControl();
 	const auto& rc      = hw.GetRenderControl();
+	const auto& d       = hw.GetDepthControl();
+	const auto& s       = hw.GetStencilControl();
+	const auto& sm      = hw.GetStencilMask();
 	const auto& mc      = hw.GetModeControl();
 	const auto& eqaa    = hw.GetEqaaControl();
 	const auto& cc      = hw.GetColorControl();
@@ -872,10 +937,11 @@ void hw_check(const CommandBuffer& buffer) {
 	log_phase("rc");
 	RcCheck(rc);
 	log_phase("depth");
+	DCheck(d, s, sm);
 	log_phase("mode");
 	McCheck(mc);
 	log_phase("blend");
-	BcCheck(bclr, cc);
+	BcCheck(bc, bclr, cc);
 	log_phase("eqaa");
 	EqaaCheck(eqaa, ac);
 	log_phase("aa");
@@ -934,7 +1000,7 @@ void hw_print(const CommandBuffer& buffer) {
 		     hw.GetRenderTargetMask(), hw.GetDepthClearValue(), hw.GetStencilClearValue(),
 		     hw.GetLineWidth(), hw.GetPrimitiveResetIndex());
 
-		LOGF("%s", rt_print("RenderTraget:", rt).c_str());
+		LOGF("%s", Common::Concat(rt_print("RenderTraget:", rt), "").c_str());
 
 		ZPrint("DepthRenderTraget:", z);
 		VpPrint("ScreenViewport:", vp, smc);

src/graphics/host_gpu/renderer/debug.h

@@ -5,6 +5,7 @@
 
 #include <cstdint>
 #include <string>
+#include <vector>
 
 namespace Libs::Graphics {
 
@@ -12,6 +13,7 @@ class CommandBuffer;
 
 namespace HW {
 class Context;
+class Shader;
 class UserConfig;
 struct RenderTarget;
 struct ScanModeControl;
@@ -30,7 +32,8 @@ uint32_t                 render_target_first_bound_slot(const CommandBuffer& buf
 bool                     graphics_debug_dump_enabled();
 void                     uc_print(const char* func, const HW::UserConfig& uc);
 void                     uc_check(const HW::UserConfig& uc);
-std::string              rt_print(const char* func, const HW::RenderTarget& rt);
+void                     sh_print(const char* func, const HW::Shader& uc);
+std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt);
 bool                     RenderIsColorTileModeLinear(Prospero::TileMode tile_mode);
 void                     hw_print(const CommandBuffer& buffer);
 void                     hw_check(const CommandBuffer& buffer);

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -19,7 +19,6 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <algorithm>
-#include <array>
 #include <atomic>
 #include <cmath>
 #include <cstdarg>
@@ -46,12 +45,8 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
-		case Prospero::StencilOp::kReplaceTest: return vk::StencilOp::eReplace;
-		case Prospero::StencilOp::kReplaceOp:
-			if ((op_value & write_mask) == 0) {
-				return vk::StencilOp::eZero;
-			}
-			return vk::StencilOp::eReplace;
+		case Prospero::StencilOp::kReplaceTest:
+		case Prospero::StencilOp::kReplaceOp: return vk::StencilOp::eReplace;
 		case Prospero::StencilOp::kAddClamp: return vk::StencilOp::eIncrementAndClamp;
 		case Prospero::StencilOp::kSubClamp: return vk::StencilOp::eDecrementAndClamp;
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
@@ -71,38 +66,9 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	}
 }
 
-static vk::StencilOpState ConvertStencilState(
-    uint8_t compare, const std::array<uint8_t, 3>& operations, uint8_t op_value,
-    const vk::StencilOpState& state) {
-	const auto test_value = state.reference;
-	auto reference       = test_value;
-	auto required_bits   = state.compareMask;
-	if (compare == static_cast<uint8_t>(vk::CompareOp::eAlways) ||
-	    compare == static_cast<uint8_t>(vk::CompareOp::eNever)) {
-		required_bits = 0;
-	}
-	std::array<vk::StencilOp, 3> converted {};
-	for (size_t i = 0; i < operations.size(); i++) {
-		converted[i] = ConvertStencilOp(operations[i], state.writeMask, op_value);
-		if (converted[i] != vk::StencilOp::eReplace) {
-			continue;
-		}
-		auto replacement = test_value;
-		if (static_cast<Prospero::StencilOp>(operations[i]) == Prospero::StencilOp::kReplaceOp) {
-			replacement = op_value;
-		}
-		if (((reference ^ replacement) & required_bits & state.writeMask) != 0) {
-			DepthFatal("unsupported stencil replacement: compare=%u, compare mask=0x%02" PRIx32
-			           ", write mask=0x%02" PRIx32 ", operation value=0x%02" PRIx8
-			           ", test value=0x%02" PRIx32,
-			           compare, state.compareMask, state.writeMask, op_value, test_value);
-		}
-		// Vulkan shares one reference between comparison and every replacement on this face.
-		reference = (reference & ~state.writeMask) | (replacement & state.writeMask);
-		required_bits |= state.writeMask;
-	}
-	return {converted[0], converted[1], converted[2], static_cast<vk::CompareOp>(compare),
-	        state.compareMask, state.writeMask, reference};
+static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
+	constexpr auto replace_op = static_cast<uint8_t>(Prospero::StencilOp::kReplaceOp);
+	return fail == replace_op || pass == replace_op || depth_fail == replace_op;
 }
 
 [[nodiscard]] static vk::Format ResolveHostDepthAttachmentFormat(const CommandBuffer&     buffer,
@@ -159,8 +125,8 @@ static TextureCache::ImageDesc MakeDepthTargetDesc(const CommandBuffer& buffer,
 			DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
 			           z.depth_view.slice_max);
 	}
-	// EXPCLEAR permits an HTile acceleration state; the host attachment is already expanded.
-	if (z.z_info.partially_resident ||
+	if (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 || unsupported_shading_rate_encoding ||
 	    depth_address == 0 || (depth_address & 0xffffu) != 0) {
@@ -330,20 +296,31 @@ void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepth
 		const uint8_t back_write_mask  = stencil_ops_disabled ? 0 : sm.stencil_writemask_bf;
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
-		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways))) {
-			DepthFatal("unsupported stencil compare state");
+		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
+		    (front_write_mask != 0 &&
+		     UsesStencilOpValue(sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail) &&
+		     sm.stencil_opval != sm.stencil_testval) ||
+		    (dc.backface_enable && back_write_mask != 0 &&
+		     UsesStencilOpValue(sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf) &&
+		     sm.stencil_opval_bf != sm.stencil_testval_bf)) {
+			DepthFatal("unsupported stencil compare or replacement state");
 		}
-		r.stencil_front = ConvertStencilState(
-		    dc.stencilfunc, {sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail},
-		    sm.stencil_opval, {.compareMask = sm.stencil_mask, .writeMask = front_write_mask,
-		                       .reference = sm.stencil_testval});
+		r.stencil_static_front = {
+		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
+		    static_cast<vk::CompareOp>(dc.stencilfunc)};
+		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
 		if (dc.backface_enable) {
-			r.stencil_back = ConvertStencilState(
-			    dc.stencilfunc_bf, {sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf},
-			    sm.stencil_opval_bf, {.compareMask = sm.stencil_mask_bf, .writeMask = back_write_mask,
-			                          .reference = sm.stencil_testval_bf});
+			r.stencil_static_back = {
+			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
+			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
+			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
 		} else {
-			r.stencil_back = r.stencil_front;
+			r.stencil_static_back  = r.stencil_static_front;
+			r.stencil_dynamic_back = r.stencil_dynamic_front;
 		}
 	}
 	auto& cache = m_context.GetTextureCache();
@@ -429,13 +406,14 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		return writes;
 	}
 
-	const auto face_writes = [&](const vk::StencilOpState& state) {
-		if (state.writeMask == 0) {
+	const auto face_writes = [&](const PipelineStencilStaticState&  state,
+	                             const PipelineStencilDynamicState& dynamic) {
+		if (dynamic.writeMask == 0) {
 			return false;
 		}
 		bool can_pass = state.compareOp != vk::CompareOp::eNever;
 		bool can_fail = state.compareOp != vk::CompareOp::eAlways;
-		if (state.compareMask == 0) {
+		if (dynamic.compareMask == 0) {
 			switch (state.compareOp) {
 				case vk::CompareOp::eEqual:
 				case vk::CompareOp::eLessOrEqual:
@@ -461,7 +439,8 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		       (can_pass && depth_fail && state.depthFailOp != vk::StencilOp::eKeep);
 	};
 	if (stencil_clear_enable ||
-	    (stencil_test_enable && (face_writes(stencil_front) || face_writes(stencil_back)))) {
+	    (stencil_test_enable && (face_writes(stencil_static_front, stencil_dynamic_front) ||
+	                             face_writes(stencil_static_back, stencil_dynamic_back)))) {
 		writes |= vk::ImageAspectFlagBits::eStencil;
 	}
 	return writes;

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -34,49 +34,22 @@ struct RenderDepthInfo {
 	bool                        stencil_clear_enable     = false;
 	uint8_t                     stencil_clear_value      = 0;
 	bool                        stencil_test_enable      = false;
-	vk::StencilOpState          stencil_front;
-	vk::StencilOpState          stencil_back;
+	PipelineStencilStaticState  stencil_static_front;
+	PipelineStencilStaticState  stencil_static_back;
+	PipelineStencilDynamicState stencil_dynamic_front;
+	PipelineStencilDynamicState stencil_dynamic_back;
 	ImageId                     image_id;
 
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;
 };
 
-inline vk::ImageAspectFlags DepthFeedbackAspects(vk::ImageAspectFlags draw_writes,
-                                                 const ImageViewInfo& target,
-                                                 const ImageViewInfo& sampled) {
-	if (!ImageRangeOverlaps(target.base_level, target.level_count, sampled.base_level,
-	                        sampled.level_count) ||
-	    !ImageRangeOverlaps(target.base_layer, target.layer_count, sampled.base_layer,
-	                        sampled.layer_count)) {
-		return {};
-	}
-	return draw_writes & sampled.aspect;
-}
-
-inline vk::ImageAspectFlags DepthReadableAspects(vk::ImageLayout layout) {
-	switch (layout) {
-		case vk::ImageLayout::eDepthReadOnlyOptimal:
-		case vk::ImageLayout::eDepthReadOnlyStencilAttachmentOptimal:
-			return vk::ImageAspectFlagBits::eDepth;
-		case vk::ImageLayout::eStencilReadOnlyOptimal:
-		case vk::ImageLayout::eDepthAttachmentStencilReadOnlyOptimal:
-			return vk::ImageAspectFlagBits::eStencil;
-		case vk::ImageLayout::eDepthStencilReadOnlyOptimal:
-		case vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT:
-		case vk::ImageLayout::eGeneral:
-			return vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil;
-		default:
-			return {};
-	}
-}
-
 inline vk::ImageLayout depth_attachment_layout(const RenderDepthInfo& depth) {
 	const auto available     = ImageViewOps::DepthAspectMask(depth.desc.view_info.format);
 	const auto writes        = depth.AttachmentWriteAspects();
 	const bool has_depth     = static_cast<bool>(available & vk::ImageAspectFlagBits::eDepth);
 	const bool has_stencil   = static_cast<bool>(available & vk::ImageAspectFlagBits::eStencil);
-	// The attachment layout must permit load clears as well as draw writes.
-	const bool depth_write   = static_cast<bool>(writes & vk::ImageAspectFlagBits::eDepth);
+	// LoadOp clears are separate from guest depth writes.
+	const bool depth_write   = depth.depth_write_enable;
 	const bool stencil_write = static_cast<bool>(writes & vk::ImageAspectFlagBits::eStencil);
 	if (!has_stencil) {
 		return depth_write ? vk::ImageLayout::eDepthAttachmentOptimal

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

@@ -10,6 +10,7 @@
 
 #include <algorithm>
 #include <array>
+#include <iterator>
 
 namespace Libs::Graphics {
 
@@ -36,8 +37,9 @@ BlitHelper::BlitHelper(GraphicContext& graphics, CommandScheduler& scheduler)
 	    m_graphics.device.createPipelineLayout(&layout_info, nullptr, &m_pipeline_layout),
 	    "create BlitHelper pipeline layout");
 
-	m_vertex_shader   = CompileSPV(GPU_BLIT_FS_TRIANGLE_SPV, m_graphics.device);
-	m_fragment_shader = CompileSPV(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV, m_graphics.device);
+	m_vertex_shader = CreateShader(GPU_BLIT_FS_TRIANGLE_SPV, std::size(GPU_BLIT_FS_TRIANGLE_SPV));
+	m_fragment_shader =
+	    CreateShader(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV, std::size(GPU_BLIT_COLOR_TO_MS_DEPTH_SPV));
 }
 
 BlitHelper::~BlitHelper() {
@@ -58,6 +60,17 @@ BlitHelper::~BlitHelper() {
 	}
 }
 
+vk::ShaderModule BlitHelper::CreateShader(const uint32_t* code, size_t words) const {
+	EXIT_IF(code == nullptr || words == 0);
+	vk::ShaderModuleCreateInfo create {};
+	create.codeSize         = words * sizeof(uint32_t);
+	create.pCode            = code;
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(m_graphics.device.createShaderModule(&create, nullptr, &module),
+	                     "create BlitHelper shader module");
+	return module;
+}
+
 vk::Pipeline BlitHelper::GetPipeline(PipelineKey key) {
 	const auto cached = std::ranges::find(m_pipelines, key, &Pipeline::key);
 	if (cached != m_pipelines.end()) {

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

@@ -37,7 +37,8 @@ private:
 		vk::Pipeline handle = nullptr;
 	};
 
-	[[nodiscard]] vk::Pipeline GetPipeline(PipelineKey key);
+	[[nodiscard]] vk::ShaderModule CreateShader(const uint32_t* code, size_t words) const;
+	[[nodiscard]] vk::Pipeline     GetPipeline(PipelineKey key);
 
 	GraphicContext&         m_graphics;
 	CommandScheduler&       m_scheduler;

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

@@ -627,13 +627,12 @@ void Validate(const ImageInfo& info) {
 			}
 			break;
 		case ImageMetadataKind::Dcc:
-		case ImageMetadataKind::Cmask:
 			if (info.metadata.range.address == 0 ||
 			    info.metadata.range.address >= TRACKER_ADDRESS_SIZE ||
 			    (info.metadata.range.size != 0 &&
 			     info.metadata.range.size > TRACKER_ADDRESS_SIZE - info.metadata.range.address) ||
 			    info.metadata.compression == VideoOutCompression::Unsupported) {
-				EXIT("invalid color metadata\n");
+				EXIT("invalid DCC metadata\n");
 			}
 			break;
 	}
@@ -691,21 +690,14 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
 		     create.extent.width, create.extent.height, create.extent.depth,
 		     static_cast<int>(create.format), create.arrayLayers, create.mipLevels);
 	}
-	SetVulkanObjectNameF(
-	    graphics.device, backing.image,
-	    "Kyty.Image[guest=0x{:016x} size=0x{:x} extent={}x{}x{} format={} mips={} layers={} samples={}]",
-	    info.data.address, info.data.size, info.extent.width, info.extent.height, info.extent.depth,
-	    static_cast<uint32_t>(info.pixel_format), info.resources.levels, info.resources.layers,
-	    info.samples);
 }
 
 uint64_t Image::HashGuestEdges() const {
+	constexpr uint64_t                         page_mask = TRACKER_PAGE_SIZE - 1;
 	std::array<uint8_t, TRACKER_PAGE_SIZE * 2> bytes {};
 	const auto                                 range = info.data;
-	const uint64_t head_end =
-	    std::min(range.End(), Common::AlignUp(range.address, TRACKER_PAGE_SIZE));
-	const uint64_t tail_begin =
-	    std::max(range.address, Common::AlignDown(range.End(), TRACKER_PAGE_SIZE));
+	const uint64_t head_end     = std::min(range.End(), (range.address + page_mask) & ~page_mask);
+	const uint64_t tail_begin   = std::max(range.address, range.End() & ~page_mask);
 	const uint64_t head_size    = head_end - range.address;
 	const uint64_t tail_address = tail_begin < head_end ? head_end : tail_begin;
 	const uint64_t tail_size    = range.End() - tail_address;

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

@@ -1,7 +1,6 @@
 #ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_IMAGE_H_
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_IMAGE_H_
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/slotVector.h"
 #include "graphics/host_gpu/graphicContext.h"
@@ -134,7 +133,7 @@ public:
 	}
 	[[nodiscard]] bool IsTracked() const noexcept { return track_addr != 0 && track_addr_end != 0; }
 	[[nodiscard]] uint64_t AccountedSize() const noexcept {
-		return backing.image == nullptr ? 0 : Common::AlignUp(info.data.size, 1024);
+		return backing.image == nullptr ? 0 : (info.data.size + 1023) & ~uint64_t {1023};
 	}
 	[[nodiscard]] uint64_t HashGuestEdges() const;
 

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

@@ -17,17 +17,17 @@ namespace Libs::Graphics {
 
 enum class VideoOutCompression : uint8_t { Uncompressed, Dcc256_256_0, Dcc256_64_64, Unsupported };
 
-enum class ImageMetadataKind : uint8_t { None, Htile, Dcc, Cmask };
+enum class ImageMetadataKind : uint8_t { None, Htile, Dcc };
 
 struct ImageMetadataInfo {
 	GuestRange          range;
-	ImageMetadataKind   kind                 = ImageMetadataKind::None;
-	uint32_t            control              = 0;
-	uint32_t            clear_word           = 0;
-	VideoOutCompression compression          = VideoOutCompression::Uncompressed;
-	bool                stencil_compressed   = false;
-	bool                clear_register_valid = false;
-	bool                dcc_alpha_msb        = true;
+	ImageMetadataKind   kind               = ImageMetadataKind::None;
+	uint32_t            control            = 0;
+	uint32_t            dcc_clear_word           = 0;
+	VideoOutCompression compression        = VideoOutCompression::Uncompressed;
+	bool                stencil_compressed = false;
+	bool                dcc_clear_register_valid = false;
+	bool                dcc_alpha_msb            = true;
 };
 
 struct ImageSubresources {
@@ -82,6 +82,9 @@ struct ImageInfo {
 	[[nodiscard]] constexpr bool IsVolume() const noexcept {
 		return type == Prospero::ImageType::kColor3D;
 	}
+	[[nodiscard]] constexpr bool IsLayered() const noexcept {
+		return !IsVolume() && resources.layers > 1;
+	}
 	[[nodiscard]] constexpr uint32_t TransferLayers() const noexcept {
 		return IsVolume() ? extent.depth : resources.layers;
 	}
@@ -94,24 +97,9 @@ struct ImageInfo {
 		       bytes_per_block == other.bytes_per_block;
 	}
 	[[nodiscard]] int32_t MipOf(const ImageInfo& container) const noexcept {
-		if (container.resources.levels > container.mip_layout.size()) {
-			return -1;
-		}
-		for (uint32_t level = 0; level < container.resources.levels; level++) {
-			if (SliceOf(container, static_cast<int32_t>(level)) >= 0) {
-				return static_cast<int32_t>(level);
-			}
-		}
-		return -1;
-	}
-	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
-		if (!IsCompatible(container) || tile_mode != container.tile_mode || type != container.type ||
-		    IsVolume() || resources.levels != 1 || resources.layers == 0 ||
-		    container.resources.layers == 0 || mip < 0 ||
-		    static_cast<uint32_t>(mip) >= container.resources.levels ||
-		    container.resources.levels > container.mip_layout.size() ||
-		    !data.Valid() || !container.data.Valid() || data.address < container.data.address ||
-		    data.End() > container.data.End()) {
+		if (!IsCompatible(container) || tile_mode != container.tile_mode || resources.levels != 1 ||
+		    container.resources.layers == 0 ||
+		    container.resources.levels > container.mip_layout.size()) {
 			return -1;
 		}
 		if (HasStencil() != container.HasStencil() ||
@@ -120,36 +108,75 @@ struct ImageInfo {
 			return -1;
 		}
 
-		const auto  level  = static_cast<uint32_t>(mip);
-		const auto& layout = container.mip_layout[level];
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u) ||
-		    extent.depth != container.extent.depth || mip_layout[0].pitch != layout.pitch ||
-		    mip_layout[0].height != layout.height || layout.size == 0 ||
-		    layout.size % container.resources.layers != 0 ||
-		    container.data.size % container.resources.layers != 0 ||
-		    data.size % resources.layers != 0) {
+		int32_t mip = -1;
+		for (uint32_t level = 0; level < container.resources.levels; level++) {
+			const auto& layout = container.mip_layout[level];
+			if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
+			    container.data.address > UINT64_MAX - layout.offset) {
+				continue;
+			}
+			const auto mip_base   = container.data.address + layout.offset;
+			const auto slice_size = layout.size / container.resources.layers;
+			if (slice_size == 0 || mip_base > UINT64_MAX - layout.size) {
+				continue;
+			}
+			const auto mip_end = mip_base + layout.size;
+			if (data.address >= mip_base && data.address < mip_end &&
+			    (data.address - mip_base) % slice_size == 0) {
+				mip = static_cast<int32_t>(level);
+				break;
+			}
+		}
+		if (mip < 0) {
 			return -1;
 		}
 
-		// PS5 block slices contain the entire reversed mip chain.
-		const auto slice_stride = container.data.size / container.resources.layers;
-		const auto child_stride = data.size / resources.layers;
-		if (child_stride != layout.size / container.resources.layers ||
-		    layout.offset >= slice_stride || child_stride > slice_stride - layout.offset ||
-		    (resources.layers > 1 && child_stride != slice_stride)) {
+		const auto level = static_cast<uint32_t>(mip);
+		if (extent.width != std::max(container.extent.width >> level, 1u) ||
+		    extent.height != std::max(container.extent.height >> level, 1u)) {
 			return -1;
 		}
-		const auto address_delta = data.address - container.data.address;
-		if (address_delta < layout.offset || (address_delta - layout.offset) % slice_stride != 0) {
+		const auto mip_depth = std::max(container.extent.depth >> level, 1u);
+		if (container.type == Prospero::ImageType::kColor3D &&
+		    type == Prospero::ImageType::kColor2D) {
+			if (resources.layers != mip_depth) {
+				return -1;
+			}
+		} else if (type != container.type) {
 			return -1;
 		}
-		const auto layer = (address_delta - layout.offset) / slice_stride;
-		if (layer > INT32_MAX || layer >= container.resources.layers ||
-		    resources.layers > container.resources.layers - layer) {
+		return mip;
+	}
+	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
+		if (!IsCompatible(container) || type != container.type || mip < 0 ||
+		    static_cast<uint32_t>(mip) >= container.resources.levels ||
+		    container.resources.levels > container.mip_layout.size() ||
+		    container.resources.layers == 0 || data.size == 0) {
 			return -1;
 		}
-		return static_cast<int32_t>(layer);
+		const auto level = static_cast<uint32_t>(mip);
+		if (extent.width != std::max(container.extent.width >> level, 1u) ||
+		    extent.height != std::max(container.extent.height >> level, 1u)) {
+			return -1;
+		}
+		const auto& layout = container.mip_layout[level];
+		if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
+		    container.data.address > UINT64_MAX - layout.offset) {
+			return -1;
+		}
+		const auto slice_size = layout.size / container.resources.layers;
+		if (slice_size == 0 || data.size % slice_size != 0) {
+			return -1;
+		}
+		const auto mip_base = container.data.address + layout.offset;
+		if (data.address < mip_base) {
+			return -1;
+		}
+		const auto address_delta = data.address - mip_base;
+		if (address_delta % data.size != 0 || address_delta / data.size > INT32_MAX) {
+			return -1;
+		}
+		return static_cast<int32_t>(address_delta / data.size);
 	}
 };
 
@@ -159,7 +186,6 @@ struct ImageViewInfo {
 	vk::ImageAspectFlags aspect      = vk::ImageAspectFlagBits::eColor;
 	uint32_t             base_level  = 0;
 	uint32_t             level_count = 1;
-	uint32_t             min_lod     = 0; // U4.8 clamp relative to base_level.
 	uint32_t             base_layer  = 0;
 	uint32_t             layer_count = 1;
 	vk::ComponentMapping mapping     = {};
@@ -168,10 +194,9 @@ struct ImageViewInfo {
 	[[nodiscard]] bool operator==(const ImageViewInfo& rhs) const noexcept {
 		return format == rhs.format && type == rhs.type && aspect == rhs.aspect &&
 		       base_level == rhs.base_level && level_count == rhs.level_count &&
-		       min_lod == rhs.min_lod && base_layer == rhs.base_layer &&
-		       layer_count == rhs.layer_count && mapping.r == rhs.mapping.r &&
-		       mapping.g == rhs.mapping.g && mapping.b == rhs.mapping.b &&
-		       mapping.a == rhs.mapping.a && usage == rhs.usage;
+		       base_layer == rhs.base_layer && layer_count == rhs.layer_count &&
+		       mapping.r == rhs.mapping.r && mapping.g == rhs.mapping.g &&
+		       mapping.b == rhs.mapping.b && mapping.a == rhs.mapping.a && usage == rhs.usage;
 	}
 };
 
@@ -317,6 +342,14 @@ ClassifyVideoOutCompression(bool compressed, uint64_t metadata_address, uint32_t
 	}
 }
 
+[[nodiscard]] inline constexpr bool
+CanUseVideoOutNativeWithoutUpload(VideoOutCompression compression, bool render_target,
+                                  bool gpu_modified, bool guest_modified) noexcept {
+	return compression != VideoOutCompression::Uncompressed &&
+	       compression != VideoOutCompression::Unsupported && !guest_modified &&
+	       (render_target || gpu_modified);
+}
+
 struct VideoOutPixelFormatInfo {
 	vk::Format             format            = vk::Format::eUndefined;
 	Prospero::BufferFormat guest_format      = Prospero::BufferFormat::kInvalid;
@@ -329,9 +362,7 @@ struct VideoOutFormatPolicy {
 	VideoOutPixelFormatInfo info;
 };
 
-inline constexpr uint64_t VIDEO_OUT_PIXEL_FORMAT_R10_G10_B10_A2_BT2100_PQ = 0x8100070422000000ull;
-
-inline constexpr std::array<VideoOutFormatPolicy, 7> VIDEO_OUT_FORMAT_POLICIES {{
+inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {{
     {0x8000000022000000ull,
      {vk::Format::eR8G8B8A8Srgb, Prospero::BufferFormat::k8_8_8_8Srgb, 4, false}},
     {0x8000000000000000ull,
@@ -340,8 +371,6 @@ inline constexpr std::array<VideoOutFormatPolicy, 7> VIDEO_OUT_FORMAT_POLICIES {
      {vk::Format::eA2B10G10R10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0x8100000000000000ull,
      {vk::Format::eA2R10G10B10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
-    {VIDEO_OUT_PIXEL_FORMAT_R10_G10_B10_A2_BT2100_PQ,
-     {vk::Format::eA2B10G10R10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0xc001000622000000ull,
      {vk::Format::eR16G16B16A16Sfloat, Prospero::BufferFormat::k16_16_16_16Float, 8, false}},
     {0xc001000600000000ull,
@@ -371,6 +400,32 @@ inline constexpr std::array<VideoOutFormatPolicy, 7> VIDEO_OUT_FORMAT_POLICIES {
 	return false;
 }
 
+[[nodiscard]] inline constexpr bool
+IsSupportedDisplayRenderTargetTileMode(Prospero::TileMode tile_mode) noexcept {
+	return tile_mode == Prospero::TileMode::kRenderTarget;
+}
+
+[[nodiscard]] inline constexpr bool IsSupportedStandard64RenderTarget(const ImageInfo& info) {
+	if (info.tile_mode != Prospero::TileMode::kStandard64KB || info.data.address == 0 ||
+	    (info.data.address & 0xffffu) != 0 || info.extent.width == 0 || info.extent.height == 0 ||
+	    info.bytes_per_block != 4 || info.resources.levels != 1 || info.resources.layers != 1 ||
+	    info.samples != 1) {
+		return false;
+	}
+	const auto expected_pitch =
+	    (static_cast<uint64_t>(info.extent.width) + 127u) & ~uint64_t {127u};
+	const auto padded_height =
+	    (static_cast<uint64_t>(info.extent.height) + 127u) & ~uint64_t {127u};
+	return expected_pitch <= UINT32_MAX && info.pitch == expected_pitch &&
+	       expected_pitch <= UINT64_MAX / padded_height / info.bytes_per_block &&
+	       info.data.size == expected_pitch * padded_height * info.bytes_per_block;
+}
+
+[[nodiscard]] inline constexpr bool IsTiledRenderTarget(const ImageInfo& info) noexcept {
+	return info.tile_mode == Prospero::TileMode::kRenderTarget ||
+	       IsSupportedStandard64RenderTarget(info);
+}
+
 [[nodiscard]] inline bool DecodePackedColorClear(vk::Format format, uint32_t packed,
                                                  vk::ClearColorValue& clear) {
 	vk::ClearColorValue next {};

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

@@ -322,12 +322,6 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		normalized.aspect = vk::ImageAspectFlagBits::eStencil;
 	}
 	normalized.usage = is_storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlags {};
-	for (const auto& cached: views) {
-		if (cached.info == normalized) {
-			return cached.view;
-		}
-	}
-
 	const bool format_compatible = normalized.format != vk::Format::eUndefined &&
 	                               ImageViewOps::FormatsCompatible(image.format, normalized.format);
 	const bool slice_view =
@@ -357,17 +351,17 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		     image.layers);
 	}
 
+	for (const auto& cached: views) {
+		if (cached.info == normalized) {
+			return cached.view;
+		}
+	}
+
 	vk::ImageViewUsageCreateInfo usage {};
 	usage.usage = image.usage;
 	if (!is_storage) {
 		usage.usage &= ~vk::ImageUsageFlagBits::eStorage;
 	}
-	vk::ImageViewMinLodCreateInfoEXT min_lod {};
-	if (normalized.min_lod != 0) {
-		min_lod.minLod = static_cast<float>(normalized.base_level) +
-		                 static_cast<float>(normalized.min_lod) / 256.0f;
-		usage.pNext    = &min_lod;
-	}
 	vk::ImageViewCreateInfo create {};
 	create.pNext                           = &usage;
 	create.image                           = image.image;
@@ -391,12 +385,6 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		     view_info.level_count, view_info.base_layer, view_info.layer_count,
 		     static_cast<vk::ImageUsageFlags::MaskType>(view_info.usage));
 	}
-	SetVulkanObjectNameF(
-	    m_graphics.device, view,
-	    "Kyty.ImageView[guest=0x{:016x} format={} aspect=0x{:x} mip={}+{} layer={}+{}]",
-	    info.data.address, static_cast<uint32_t>(normalized.format),
-	    static_cast<vk::ImageAspectFlags::MaskType>(normalized.aspect), normalized.base_level,
-	    normalized.level_count, normalized.base_layer, normalized.layer_count);
 	views.push_back({normalized, view});
 	return view;
 }

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

@@ -1,6 +1,5 @@
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/gpu_format.h"
@@ -261,10 +260,12 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 			region.bufferRowLength   = layout.mips[mip_level].row_length;
 			region.bufferImageHeight = layout.mips[mip_level].image_height;
 		} else {
-			const auto texel_width   = layout.surface.texture.texel_width;
-			const auto aligned_pitch = Common::AlignUp(mip_pitch, texel_width);
-			region.bufferRowLength =
-			    aligned_pitch > Common::AlignUp(mip_width, texel_width) ? aligned_pitch : 0;
+			const auto texel_width    = layout.surface.texture.texel_width;
+			const auto align_to_texel = [texel_width](uint32_t dimension) {
+				return ((dimension + texel_width - 1u) / texel_width) * texel_width;
+			};
+			const auto aligned_pitch = align_to_texel(mip_pitch);
+			region.bufferRowLength = aligned_pitch > align_to_texel(mip_width) ? aligned_pitch : 0;
 		}
 		for (uint32_t slice_index = 0; slice_index < mip_depth; ++slice_index) {
 			region.bufferOffset = layout.mips[mip_level].offset + slice_index * layout.slice_stride;

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

@@ -1,6 +1,5 @@
 #include "graphics/host_gpu/renderer/image/tiler.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "gpu_tiler_shaders/gpu_tiler_demote_d16_spv.h"
 #include "gpu_tiler_shaders/gpu_tiler_depth_spv.h"
@@ -222,7 +221,7 @@ void TileManager::Prepare(bool tile, uint64_t tiled_capacity, uint64_t linear_ca
 
 	const uint64_t uniform_alignment =
 	    std::max<uint64_t>(limits.minUniformBufferOffsetAlignment, 1);
-	const uint64_t stride = Common::AlignUp<uint64_t>(sizeof(Push), uniform_alignment);
+	const uint64_t stride = (sizeof(Push) + uniform_alignment - 1) & ~(uniform_alignment - 1);
 	EXIT_NOT_IMPLEMENTED(dispatches.size() > UINT64_MAX / stride);
 	const uint64_t bytes  = dispatches.size() * stride;
 	auto [mapped, offset] = m_stream_buffer.Map(bytes, uniform_alignment);
@@ -260,8 +259,12 @@ vk::Pipeline TileManager::GetPipeline(uint32_t slot) {
 	const uint32_t                   values[] {1u << element_index, direction_index};
 	const vk::SpecializationMapEntry entries[] {{0, 0, 4}, {1, 4, 4}};
 	const vk::SpecializationInfo     specialization {2, entries, sizeof(values), values};
-	const auto module =
-	    CompileSPV({shaders[family_index].code, shaders[family_index].words}, m_graphics.device);
+	vk::ShaderModuleCreateInfo       module_info {};
+	module_info.codeSize    = shaders[family_index].words * sizeof(uint32_t);
+	module_info.pCode       = shaders[family_index].code;
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
+	                     "create TileManager shader module");
 	vk::PipelineShaderStageCreateInfo stage {};
 	stage.stage               = vk::ShaderStageFlagBits::eCompute;
 	stage.module              = module;
@@ -284,12 +287,12 @@ void TileManager::Record(vk::Buffer source, uint64_t source_offset,
 	const auto&    limits = m_graphics.GetPhysicalDeviceProperties().limits;
 	const uint64_t descriptor_alignment =
 	    std::max<uint64_t>(limits.minStorageBufferOffsetAlignment, 4);
-	const uint64_t source_descriptor_offset = Common::AlignDown(source_offset, descriptor_alignment);
-	const uint64_t target_descriptor_offset = Common::AlignDown(target_offset, descriptor_alignment);
+	const uint64_t source_descriptor_offset = source_offset & ~(descriptor_alignment - 1);
+	const uint64_t target_descriptor_offset = target_offset & ~(descriptor_alignment - 1);
 	const uint64_t source_base              = source_offset - source_descriptor_offset;
 	const uint64_t target_base              = target_offset - target_descriptor_offset;
-	const uint64_t source_range             = Common::AlignUp(source_base + source_capacity, 4);
-	const uint64_t target_range             = Common::AlignUp(target_base + target_capacity, 4);
+	const uint64_t source_range             = (source_base + source_capacity + 3u) & ~uint64_t {3};
+	const uint64_t target_range             = (target_base + target_capacity + 3u) & ~uint64_t {3};
 	EXIT_NOT_IMPLEMENTED(source_range > limits.maxStorageBufferRange ||
 	                     target_range > limits.maxStorageBufferRange || target_offset % 4 != 0 ||
 	                     target_capacity % 4 != 0);
@@ -369,7 +372,7 @@ TileManager::Result TileManager::Detile(vk::Buffer tiled, uint64_t tiled_offset,
 	const uint64_t        source_base = tiled_offset & (descriptor_alignment - 1);
 	std::vector<Dispatch> dispatches;
 	Prepare(false, tiled_capacity, linear_capacity, infos, source_base, 0, dispatches);
-	auto scratch = AllocateScratch(Common::AlignUp(linear_capacity, 4));
+	auto scratch = AllocateScratch((linear_capacity + 3u) & ~uint64_t {3});
 	DeferDestroy(scratch);
 	Record(tiled, tiled_offset, tiled_capacity, scratch.buffer, 0, scratch.size, dispatches,
 	       true);
@@ -403,7 +406,7 @@ void TileManager::TileImage(Image& image, std::span<const vk::BufferImageCopy> r
 	// Reserve all stream parameters before creating a scheduler-lived scratch dependency:
 	// StreamBuffer::Map is allowed to submit the current tick when it wraps.
 	Prepare(true, tiled_capacity, linear_capacity, infos, 0, target_base, dispatches);
-	auto linear = AllocateScratch(Common::AlignUp(linear_capacity, 4));
+	auto linear = AllocateScratch((linear_capacity + 3u) & ~uint64_t {3});
 	DeferDestroy(linear);
 	image.Download(regions, linear.buffer, 0, linear.size);
 	Result source {linear.buffer, 0, linear.size};
@@ -415,7 +418,7 @@ void TileManager::TileImage(Image& image, std::span<const vk::BufferImageCopy> r
 }
 
 TileManager::Result TileManager::GetScratchBuffer(uint64_t size) {
-	auto scratch = AllocateScratch(Common::AlignUp(size, 4));
+	auto scratch = AllocateScratch((size + 3u) & ~uint64_t {3});
 	DeferDestroy(scratch);
 	return {scratch.buffer, 0, scratch.size};
 }
@@ -423,11 +426,11 @@ TileManager::Result TileManager::GetScratchBuffer(uint64_t size) {
 TileManager::StorageBinding TileManager::BindStorage(Result buffer, uint64_t size) const {
 	const auto& limits            = m_graphics.GetPhysicalDeviceProperties().limits;
 	const auto  alignment         = std::max<uint64_t>(limits.minStorageBufferOffsetAlignment, 4);
-	const auto  descriptor_offset = Common::AlignDown(buffer.offset, alignment);
+	const auto  descriptor_offset = buffer.offset - buffer.offset % alignment;
 	const auto  base              = buffer.offset - descriptor_offset;
 	EXIT_IF(buffer.buffer == nullptr || size == 0 || buffer.size < size || base > UINT32_MAX ||
 	        size > UINT64_MAX - base || base + size > UINT64_MAX - 3);
-	const auto range = Common::AlignUp(base + size, 4);
+	const auto range = (base + size + 3) & ~uint64_t {3};
 	EXIT_IF(range > limits.maxStorageBufferRange || range > UINT32_MAX);
 	return {{buffer.buffer, descriptor_offset, range}, static_cast<uint32_t>(base)};
 }
@@ -468,7 +471,12 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 			code  = GPU_TILER_DEMOTE_D16_SPV;
 			words = std::size(GPU_TILER_DEMOTE_D16_SPV);
 		}
-		const auto module = CompileSPV({code, words}, m_graphics.device);
+		vk::ShaderModuleCreateInfo module_info {};
+		module_info.codeSize    = words * sizeof(uint32_t);
+		module_info.pCode       = code;
+		vk::ShaderModule module = nullptr;
+		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
+		                     "create D16 conversion shader module");
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage               = vk::ShaderStageFlagBits::eCompute;
 		stage.module              = module;
@@ -504,8 +512,8 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 	const auto target_required = required(layout.height, layout.layers, layout.target_row_stride,
 	                                      layout.target_slice_stride, target_active);
 	EXIT_IF(source_required > UINT64_MAX - 3 || target_required > UINT64_MAX - 3);
-	const auto source_barrier_size = Common::AlignUp(source_required, 4);
-	const auto target_barrier_size = Common::AlignUp(target_required, 4);
+	const auto source_barrier_size = (source_required + 3) & ~uint64_t {3};
+	const auto target_barrier_size = (target_required + 3) & ~uint64_t {3};
 	EXIT_IF(source.size < source_barrier_size || target.size < target_barrier_size);
 
 	m_scheduler.EndRendering();
@@ -602,7 +610,12 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 
 void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 	if (m_swap_bgra16 == nullptr) {
-		const auto module = CompileSPV(GPU_TILER_SWAP_BGRA16_SPV, m_graphics.device);
+		vk::ShaderModuleCreateInfo module_info {};
+		module_info.codeSize    = std::size(GPU_TILER_SWAP_BGRA16_SPV) * sizeof(uint32_t);
+		module_info.pCode       = GPU_TILER_SWAP_BGRA16_SPV;
+		vk::ShaderModule module = nullptr;
+		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
+		                     "create BGRA16 swap shader module");
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage  = vk::ShaderStageFlagBits::eCompute;
 		stage.module = module;

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

@@ -20,7 +20,7 @@ constexpr std::array DescriptorPoolSizes = {
 
 DescriptorHeap::DescriptorHeap(GraphicContext& graphics, MasterSemaphore& master_semaphore)
     : m_graphics(graphics), m_master_semaphore(master_semaphore) {
-	CreateDescriptorPool();
+	CreatePool();
 }
 
 DescriptorHeap::~DescriptorHeap() {
@@ -49,7 +49,7 @@ vk::DescriptorSet DescriptorHeap::Commit(vk::DescriptorSetLayout layout) {
 		m_pending_pools.pop_front();
 		EXIT_IF(m_graphics.device.resetDescriptorPool(m_current_pool, {}) != vk::Result::eSuccess);
 	} else {
-		CreateDescriptorPool();
+		CreatePool();
 	}
 
 	m_sets.clear();
@@ -82,7 +82,7 @@ bool DescriptorHeap::Allocate(vk::DescriptorSetLayout layout, Batch& batch) {
 	}
 }
 
-void DescriptorHeap::CreateDescriptorPool() {
+void DescriptorHeap::CreatePool() {
 	vk::DescriptorPoolCreateInfo create {};
 	create.maxSets       = DescriptorHeapCount;
 	create.poolSizeCount = static_cast<uint32_t>(DescriptorPoolSizes.size());

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

@@ -31,7 +31,7 @@ private:
 	};
 
 	[[nodiscard]] bool Allocate(vk::DescriptorSetLayout layout, Batch& batch);
-	void               CreateDescriptorPool();
+	void               CreatePool();
 
 	GraphicContext&                                     m_graphics;
 	MasterSemaphore&                                    m_master_semaphore;

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

@@ -1,6 +1,5 @@
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 
-#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/file.h"
@@ -15,9 +14,7 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/hostMemory.h"
-#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/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
@@ -95,15 +92,22 @@ static const char* ShaderStageResourceName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Vertex: return "Vertex";
 		case ShaderType::Mesh: return "Mesh";
-		case ShaderType::Local: return "Local";
-		case ShaderType::TessellationControl: return "Hull";
-		case ShaderType::TessellationEvaluation: return "Domain";
 		case ShaderType::Pixel: return "Pixel";
 		case ShaderType::Compute: return "Compute";
 		default: return "Unknown";
 	}
 }
 
+static vk::ShaderStageFlags NativeShaderStage(ShaderType stage) {
+	switch (stage) {
+		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
+		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
+		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
+		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
+		default: EXIT("unknown native shader stage\n");
+	}
+}
+
 static Prospero::ImageType TextureType(const ShaderTextureResource& descriptor) {
 	const auto type = descriptor.Type();
 	return type == Prospero::ImageType::kCube ? Prospero::ImageType::kColor2DArray : type;
@@ -141,7 +145,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	auto [buffer, offset] = context.GetBufferCache().ObtainBuffer(address, size, resource.written,
 	                                                              resource.formatted, id);
-	const auto aligned_offset = Common::AlignDown(offset, alignment);
+	const auto aligned_offset = offset - offset % alignment;
 	const auto adjustment     = offset - aligned_offset;
 	const auto max_range      = graphics.GetPhysicalDeviceProperties().limits.maxStorageBufferRange;
 	if (adjustment % sizeof(uint32_t) != 0 || adjustment >= 256 || size > max_range - adjustment) {
@@ -149,7 +153,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	buffer_offset = static_cast<uint32_t>(adjustment);
 	const vk::DescriptorBufferInfo result {buffer->Handle(), aligned_offset, size + adjustment};
-	if (resource.written) {
+	if (resource.formatted && resource.written) {
 		context.GetTextureCache().InvalidateMemoryFromGPU(address, size);
 	}
 	const char* access = "Read";
@@ -165,6 +169,34 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	return result;
 }
 
+static bool IsSupportedSampledColorResource(const ShaderRecompiler::IR::ImageResource& resource) {
+	bool supported_dimension = false;
+	switch (resource.dimension) {
+		case ShaderRecompiler::Decoder::ImageDimension::Dim1D:
+		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;
+	}
+	return resource.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled &&
+	       resource.numeric_class != Prospero::TextureNumericClass::Unsupported &&
+	       supported_dimension && resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None &&
+	       resource.read && !resource.written && !resource.atomic && !resource.depth_compare;
+}
+
+bool IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
+                                    const ShaderTextureResource& descriptor, const Image& image) {
+	return image.usage.video_out && image.info.resources.layers == 1 &&
+	       IsSupportedSampledColorResource(resource) &&
+	       resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D &&
+	       descriptor.Type() == Prospero::ImageType::kColor2D && descriptor.Depth() == 0 &&
+	       descriptor.BaseArray5() == 0;
+}
+
 bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, bool r128) {
 	constexpr uint32_t field1_reserved_mask = 0x200fff00u;
 	constexpr uint32_t field2_reserved_mask = 0xf0003000u;
@@ -185,18 +217,17 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, bo
 	}
 	const bool full = common && descriptor.fields[4] == field4_expected &&
 	                  descriptor.fields[5] == field5_expected;
-	if (!full ||
+	if (!full || (descriptor.fields[6] == 0 && descriptor.fields[7] != 0) ||
 	    (descriptor.MsaaDepth() && !IsMultisampledTexture(descriptor.Type()))) {
 		return false;
 	}
-	const auto metadata_control = descriptor.fields[6] & 0x00ffffffu;
-	if (metadata_control == 0) {
+	if (descriptor.fields[6] == 0) {
 		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 metadata_control == expected_control && metadata_addr != 0 &&
+	return (descriptor.fields[6] & 0x00ffffffu) == expected_control && metadata_addr != 0 &&
 	       metadata_addr < TRACKER_ADDRESS_SIZE && (metadata_addr & 0x7fffu) == 0 &&
 	       descriptor.TileMode() == Prospero::TileMode::kDepth;
 }
@@ -254,10 +285,10 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	    (is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth());
 	const bool is_2d =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D && valid_2d_slice;
-	// Storage cube coordinates address individual faces, including partial cube views.
 	const bool is_cube = resource.cube && descriptor.Type() == Prospero::ImageType::kCube &&
 	                     descriptor.Width5() == descriptor.Height5() &&
-	                     descriptor.BaseArray5() <= descriptor.Depth();
+	                     descriptor.BaseArray5() <= descriptor.Depth() &&
+	                     (descriptor.Depth() - descriptor.BaseArray5() + 1u) % 6u == 0;
 	const bool is_2d_array =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray &&
 	    ((!resource.cube && is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth()) ||
@@ -295,8 +326,13 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
+	const auto max_mip = resource.r128 ? descriptor.LastLevel() : descriptor.MaxMip();
+	const auto view_last_level =
+	    resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::DynamicStorage
+	        ? descriptor.LastLevel()
+	        : std::min(descriptor.LastLevel(), max_mip);
 	return (is_1d || is_1d_array || is_2d || is_2d_array || is_3d) && supported_tile &&
-	       descriptor.BaseLevel() <= descriptor.LastLevel() &&
+	       descriptor.BaseLevel() <= view_last_level && view_last_level <= max_mip &&
 	       descriptor.MinLod() == 0 && supported_swizzle && descriptor.BCSwizzle() == 0 &&
 	       !descriptor.MsaaDepth();
 }
@@ -335,10 +371,8 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	const bool raw_sint_storage = format == Prospero::BufferFormat::k32SInt && uint_resource &&
 	                              resource.written && !resource.read && !resource.atomic;
 	const auto numeric_class = Prospero::SampledTextureNumericClass(format);
-	const bool raw_float_atomic = format == Prospero::BufferFormat::k32Float && uint_resource &&
-	                              resource.atomic;
 	const bool format_ok =
-	    raw_sint_storage || raw_float_atomic ||
+	    raw_sint_storage ||
 	    (numeric_class != Prospero::TextureNumericClass::Unsupported &&
 	     numeric_class != Prospero::TextureNumericClass::Sint &&
 	     uint_resource == (numeric_class == Prospero::TextureNumericClass::Uint) &&
@@ -456,14 +490,6 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 	view.aspect      = vk::ImageAspectFlagBits::eColor;
 	view.base_level  = descriptor.BaseLevel();
 	view.level_count = view_levels;
-	if (descriptor.MinLod() > descriptor.LastLevel() * 256u) {
-		EXIT("texture minimum LOD exceeds last mip level: min_lod=%u last_level=%u\n",
-		     descriptor.MinLod(), descriptor.LastLevel());
-	}
-	const auto base_lod = view.base_level * 256u;
-	if (descriptor.MinLod() > base_lod) {
-		view.min_lod = descriptor.MinLod() - base_lod;
-	}
 	view.usage = storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlagBits::eSampled;
 	view.mapping =
 	    storage || surface_format.conversion_format != Prospero::BufferFormat::kInvalid
@@ -514,52 +540,6 @@ static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource&
 	return view;
 }
 
-static bool ResolveTextureMipView(const TileSurfaceDescription& description, bool metadata,
-                                   uint32_t view_levels, uint32_t& levels, uint32_t& base_level) {
-	TileSurfaceLayout physical {};
-	TileSurfaceLayout view {};
-	auto              view_description = description;
-	view_description.levels            = levels;
-	if (!TileGetTiledTextureLayout(description, physical) ||
-	    !TileGetTiledTextureLayout(view_description, view)) {
-		return false;
-	}
-	if (physical.first_tail_level == view.first_tail_level &&
-	    physical.block_slice_size == view.block_slice_size &&
-	    physical.total_size == view.total_size &&
-	    std::equal(std::begin(physical.mips), std::begin(physical.mips) + description.levels,
-	               std::begin(view.mips))) {
-		return true;
-	}
-	if (metadata || ((description.layers > 1 || description.depth > 1) &&
-	                 physical.block_slice_size != view.block_slice_size)) {
-		return false;
-	}
-	// T# addresses the last mip. A view can select the same stored subresources
-	// with different mip indices; inaccessible mips need no host representation.
-	for (uint32_t base = 0; base + view_levels <= description.levels; ++base) {
-		bool matches = true;
-		for (uint32_t i = 0; i < view_levels; ++i) {
-			const auto source = base_level + i;
-			const auto target = base + i;
-			if (physical.mips[target] != view.mips[source] ||
-			    (target >= physical.first_tail_level) != (source >= view.first_tail_level) ||
-			    std::max(description.width >> target, 1u) != std::max(description.width >> source, 1u) ||
-			    std::max(description.height >> target, 1u) != std::max(description.height >> source, 1u) ||
-			    std::max(description.depth >> target, 1u) != std::max(description.depth >> source, 1u)) {
-				matches = false;
-				break;
-			}
-		}
-		if (matches) {
-			levels     = description.levels;
-			base_level = base;
-			return true;
-		}
-	}
-	return false;
-}
-
 TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   resource,
                                               const ShaderRecompiler::IR::DescriptorValue& value) {
 	auto descriptor = DecodeNativeDescriptor<ShaderTextureResource>(value);
@@ -576,28 +556,24 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		return {id, nullptr, std::move(desc)};
 	}
 
-	const auto address         = descriptor.Base40();
-	const auto width           = static_cast<uint32_t>(descriptor.Width5()) + 1u;
-	const auto height          = static_cast<uint32_t>(descriptor.Height5()) + 1u;
-	const auto base_level      = descriptor.BaseLevel();
-	const auto last_level      = descriptor.LastLevel();
-	const auto type            = TextureType(descriptor);
-	const bool multisampled    = IsMultisampledTexture(type);
-	const auto max_mip         = resource.r128 ? last_level : descriptor.MaxMip();
-	const auto physical_levels = multisampled ? 1u : static_cast<uint32_t>(max_mip) + 1u;
-	// IMAGE_STORE addresses BASE_LEVEL; only IMAGE_STORE_MIP selects other view mips.
-	const bool single_storage_mip =
-	    storage && resource.mip_mode != ShaderRecompiler::IR::ImageMipMode::DynamicStorage;
-	const auto view_levels = multisampled || single_storage_mip
-	                             ? 1u
-	                             : static_cast<uint32_t>(last_level - base_level) + 1u;
-	auto levels =
-	    multisampled ? 1u : std::max(physical_levels, base_level + view_levels);
+	const auto address      = descriptor.Base40();
+	const auto width        = static_cast<uint32_t>(descriptor.Width5()) + 1u;
+	const auto height       = static_cast<uint32_t>(descriptor.Height5()) + 1u;
+	const auto base_level   = descriptor.BaseLevel();
+	const auto last_level   = descriptor.LastLevel();
+	const auto type         = TextureType(descriptor);
+	const bool multisampled = IsMultisampledTexture(type);
+	const auto max_mip      = resource.r128 ? last_level : descriptor.MaxMip();
+	const auto levels       = multisampled ? 1u : static_cast<uint32_t>(max_mip) + 1u;
+	const bool dynamic_storage =
+	    storage && resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::DynamicStorage;
+	const auto view_last_level =
+	    !multisampled && !dynamic_storage ? std::min(last_level, max_mip) : last_level;
 	const auto tile       = descriptor.TileMode();
 	const bool depth_tile = tile == Prospero::TileMode::kDepth;
 	const bool msaa_tile  = depth_tile || tile == Prospero::TileMode::kRenderTarget;
 	const bool msaa_array = type == Prospero::ImageType::kColor2DMsaaArray;
-	if ((!multisampled && base_level > last_level) ||
+	if ((!multisampled && (base_level > view_last_level || view_last_level >= levels)) ||
 	    (multisampled &&
 	     (base_level != 0 || last_level == 0 || last_level > 3 || max_mip != last_level ||
 	      !msaa_tile || (descriptor.MsaaDepth() && !depth_tile) ||
@@ -615,6 +591,8 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		     descriptor.fields[6], descriptor.fields[7]);
 	}
 	const auto samples = multisampled ? 1u << last_level : 1u;
+	const auto view_levels =
+	    multisampled ? 1u : static_cast<uint32_t>(view_last_level - base_level) + 1u;
 	const auto depth          = static_cast<uint32_t>(descriptor.Depth()) + 1u;
 	const auto format         = descriptor.Format();
 	const auto surface_format = TextureGetSurfaceFormatInfo(format);
@@ -633,20 +611,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	                             type == Prospero::ImageType::kColor2DArray ||
 	                             type == Prospero::ImageType::kColor2DMsaaArray;
 	const auto    image_layers = layered ? depth : 1u;
-	auto          view_base    = static_cast<uint32_t>(base_level);
-	if (levels > physical_levels) {
-		const TileSurfaceDescription physical {
-		    format, tile, volume ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
-		    width, height, volume ? depth : 1u, physical_levels, image_layers};
-		if (!ResolveTextureMipView(physical, !resource.r128 && descriptor.MetaCompress(),
-		                           view_levels, levels, view_base)) {
-			EXIT("unsupported texture mip view changes physical layout: base=%u last=%u max=%u "
-			     "extent=%ux%ux%u tile=%u\n",
-			     base_level, last_level, max_mip, width, height, depth,
-			     static_cast<uint32_t>(tile));
-		}
-	}
-	uint32_t      pitch = 0;
+	uint32_t      pitch        = 0;
 	TileSizeAlign size {};
 	if (multisampled) {
 		const auto bytes = Prospero::NumBytesPerElement(format);
@@ -659,8 +624,8 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 		size.size *= image_layers;
 	} else {
 		pitch = TileGetTexturePitch(format, width, tile);
-		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers,
-		                        physical_levels, tile, volume, size);
+		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, levels, tile,
+		                        volume, size);
 	}
 	EXIT_NOT_IMPLEMENTED(size.size == 0 || size.align == 0 ||
 	                     (address & (static_cast<uint64_t>(size.align) - 1u)) != 0);
@@ -674,8 +639,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 			pixel_format = depth_format->depth_attachment_format;
 		}
 	}
-	const auto storage_view_format = storage && (resource.atomic ||
-	                                            format == Prospero::BufferFormat::k32SInt)
+	const auto storage_view_format = storage && format == Prospero::BufferFormat::k32SInt
 	                                     ? vk::Format::eR32Uint
 	                                     : SrgbStorageViewFormat(pixel_format);
 	const auto view_format         = storage && storage_view_format != vk::Format::eUndefined
@@ -698,7 +662,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	    !desc.info.IsDepth()) {
 		TileSizeAlign metadata_size {};
 		(void)TileGetDccSize(width, height, volume ? depth : image_layers,
-		                     desc.info.bytes_per_block, physical_levels, tile, metadata_size,
+		                     desc.info.bytes_per_block, levels, tile, metadata_size,
 		                     std::countr_zero(samples));
 		desc.info.metadata.kind          = ImageMetadataKind::Dcc;
 		desc.info.metadata.range         = {descriptor.MetaAddr() << 8u, metadata_size.size};
@@ -711,7 +675,6 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	}
 	desc.view_info = TextureViewInfo(resource, descriptor, view_format, surface_format, storage,
 	                                 view_levels, desc.info.resources.layers);
-	desc.view_info.base_level = view_base;
 	desc.type = storage ? TextureCache::BindingType::Storage : TextureCache::BindingType::Texture;
 
 	auto       id                  = texture_cache.FindImage(desc, shader_conversion);
@@ -758,6 +721,13 @@ static vk::DescriptorBufferInfo NativeUpload(RenderContext&            context,
 	return {buffer.Handle(), offset, data.size_bytes()};
 }
 
+void RenderExecutor::TrackImageBinding(ImageId id) {
+	EXIT_IF(m_context.GetTextureCache().m_slot_images.try_get(id) == nullptr);
+	if (std::ranges::find(m_bound_images, id) == m_bound_images.end()) {
+		m_bound_images.push_back(id);
+	}
+}
+
 void RenderExecutor::BindImage(ImageId id, bool storage) {
 	auto& image = m_context.GetTextureCache().GetImage(id);
 	if (image.info.data.Empty()) {
@@ -768,13 +738,13 @@ void RenderExecutor::BindImage(ImageId id, bool storage) {
 	}
 	image.binding.is_bound = true;
 	image.binding.shader_write |= storage;
-	m_bound_images.push_back(id);
+	TrackImageBinding(id);
 }
 
 void RenderExecutor::BindRenderTarget(ImageId id) {
 	auto& image             = m_context.GetTextureCache().GetImage(id);
 	image.binding.is_target = true;
-	m_bound_images.push_back(id);
+	TrackImageBinding(id);
 }
 
 void RenderExecutor::ResetBindings() {
@@ -786,27 +756,18 @@ void RenderExecutor::ResetBindings() {
 	m_bound_images.clear();
 }
 
-void RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime,
-                                     PreparedBindings& prepared) {
+PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(!runtime);
 	const auto& program  = *runtime.program;
-	const auto& snapshot = *runtime.resources;
+	const auto& snapshot = runtime.resources;
+	PreparedBindings prepared;
 	prepared.runtime = &runtime;
-	prepared.gds = {nullptr, 0, VK_WHOLE_SIZE};
-	prepared.flattened_srt = {};
-	prepared.shader_data_buffer = {};
-	prepared.buffer_sources.clear();
-	prepared.buffers.clear();
-	prepared.images.resize(program.info.images.size());
-	prepared.samplers.clear();
-	prepared.shader_data.clear();
+	prepared.images.reserve(program.info.images.size());
 	for (uint32_t i = 0; i < program.info.images.size(); i++) {
 		auto binding = ResolveTexture(program.info.images[i], snapshot.images[i]);
 		BindImage(binding.image_id, binding.desc.type == TextureCache::BindingType::Storage);
-		binding.mip_views.swap(prepared.images[i].mip_views);
-		binding.mip_views.clear();
-		prepared.images[i] = std::move(binding);
+		prepared.images.push_back(std::move(binding));
 	}
 	prepared.samplers.reserve(program.info.samplers.size());
 	for (uint32_t i = 0; i < program.info.samplers.size(); i++) {
@@ -821,26 +782,25 @@ void RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime,
 	        program.bindings, ShaderRecompiler::IR::DescriptorBindingKind::Gds) != nullptr) {
 		prepared.gds.buffer = m_context.GetBufferCache().GetGdsBuffer()->Handle();
 	}
+	return prepared;
 }
 
 void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program  = *prepared.runtime->program;
-	const auto& snapshot = *prepared.runtime->resources;
+	const auto& snapshot = prepared.runtime->resources;
 	auto&       cache    = m_context.GetBufferCache();
 
 	prepared.buffer_sources.clear();
-	const auto& layout = program.bindings;
-	if (layout.memory_offset_count == 0) {
-		return;
-	}
-	const auto& resources = layout.descriptors.front().resources;
-	prepared.buffer_sources.reserve(resources.size());
-	for (const auto resource: resources) {
-		auto descriptor = DecodeNativeDescriptor<ShaderBufferResource>(snapshot.buffers[resource]);
+	prepared.buffer_sources.reserve(program.info.buffers.size());
+	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
+		auto descriptor = DecodeNativeDescriptor<ShaderBufferResource>(snapshot.buffers[i]);
 		const auto address = descriptor.Base48();
-		const auto requested_size = descriptor.GetSize();
+		const auto stride  = descriptor.Stride();
+		const auto records = descriptor.NumRecords();
+		// The descriptor has a 14-bit stride and 32-bit record count, so the product fits u64.
+		const auto requested_size = stride != 0 ? static_cast<uint64_t>(stride) * records : records;
 		if (address == 0 || requested_size == 0) {
 			prepared.buffer_sources.push_back({});
 			continue;
@@ -854,12 +814,12 @@ void RenderExecutor::RebindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program   = *prepared.runtime->program;
-	const auto& snapshot  = *prepared.runtime->resources;
+	const auto& snapshot  = prepared.runtime->resources;
 	const auto& layout    = program.bindings;
-	EXIT_IF(prepared.buffer_sources.size() != layout.memory_offset_count);
+	EXIT_IF(prepared.buffer_sources.size() != program.info.buffers.size());
 
 	prepared.buffers.clear();
-	prepared.buffers.reserve(layout.memory_offset_count);
+	prepared.buffers.reserve(program.info.buffers.size());
 	EXIT_IF(prepared.shader_data.size() != layout.ShaderDataDwords());
 	std::fill(prepared.shader_data.begin() + layout.memory_offset_dword,
 	          prepared.shader_data.end(), 0);
@@ -868,12 +828,11 @@ void RenderExecutor::RebindBuffers(PreparedBindings& prepared) {
 		const auto shift = (index % 4u) * 8u;
 		prepared.shader_data[dword] |= offset << shift;
 	};
-	for (uint32_t i = 0; i < layout.memory_offset_count; i++) {
-		const auto resource = layout.descriptors.front().resources[i];
+	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 		uint32_t buffer_offset = 0;
 		prepared.buffers.push_back(NativeStorageBuffer(m_context, prepared.buffer_sources[i],
-		                                               program.info.buffers[resource],
-		                                               program.stage, resource, buffer_offset));
+		                                               program.info.buffers[i], program.stage, i,
+		                                               buffer_offset));
 		pack_memory_offset(i, buffer_offset);
 	}
 	if (ShaderRecompiler::IR::FindBinding(
@@ -890,7 +849,7 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
 	const auto& program  = *prepared.runtime->program;
-	const auto& snapshot = *prepared.runtime->resources;
+	const auto& snapshot = prepared.runtime->resources;
 	auto&       images   = prepared.images;
 	EXIT_IF(images.size() != program.info.images.size());
 	auto& texture_cache = m_context.GetTextureCache();
@@ -922,7 +881,11 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 			}
 			binding.image_view = binding.mip_views.front();
 		} else {
-			binding.image_view = texture_cache.FindTexture(binding.image_id, binding.desc);
+			auto desc = binding.desc;
+			if (desc.type == TextureCache::BindingType::Storage) {
+				desc.view_info.level_count = 1;
+			}
+			binding.image_view = texture_cache.FindTexture(binding.image_id, desc);
 		}
 		auto&      image   = texture_cache.GetImage(binding.image_id);
 		const bool storage = binding.desc.type == TextureCache::BindingType::Storage;
@@ -931,39 +894,32 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	}
 }
 
-void RenderExecutor::PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
-                                             std::span<RenderColorInfo> colors) {
-	bool uses_dma = false;
-	for (auto* stage: stages) {
-		FindBuffers(*stage);
-		uses_dma |= stage->runtime->program->info.uses_dma;
+RenderExecutor::GraphicsBindings
+RenderExecutor::PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
+                                        const ShaderStageRuntime& pixel, bool pixel_active) {
+	GraphicsBindings bindings {
+	    .vertex = PrepareBindings(vertex),
+	};
+	if (pixel_active) {
+		bindings.pixel.emplace(PrepareBindings(pixel));
 	}
-	if (uses_dma) {
-		m_context.PrepareBda();
+	FindBuffers(bindings.vertex);
+	if (bindings.pixel) {
+		FindBuffers(*bindings.pixel);
 	}
-	for (auto* stage: stages) {
-		RebindImages(*stage);
+	if (bindings.vertex.runtime->program->info.uses_dma ||
+	    (bindings.pixel && bindings.pixel->runtime->program->info.uses_dma)) {
+		m_context.GetGpuResources().PrepareBda();
 	}
-	auto& cache = m_context.GetTextureCache();
-	for (auto& target: colors) {
-		EXIT_IF(!target.image_id);
-		const auto old_image = cache.m_slot_images.try_get(target.image_id);
-		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
-		    old_image->binding.needs_rebind) {
-			if (old_image != nullptr) {
-				old_image->binding = {};
-			}
-			target.desc.view_info.base_level = target.guest_mip_level;
-			target.desc.view_info.base_layer = target.guest_array_layer;
-			target.image_id = cache.FindImage(target.desc);
-			BindRenderTarget(target.image_id);
-		}
+	RebindBuffers(bindings.vertex);
+	if (bindings.pixel) {
+		RebindBuffers(*bindings.pixel);
 	}
-	// Discovery can read back PS5 metadata and submit the scheduler. Reserve draw buffers only
-	// after image identities are final; attachment layout transitions follow buffer alias copies.
-	for (auto* stage: stages) {
-		RebindBuffers(*stage);
+	RebindImages(bindings.vertex);
+	if (bindings.pixel) {
+		RebindImages(*bindings.pixel);
 	}
+	return bindings;
 }
 
 void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
@@ -976,10 +932,9 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 	size_t write_count      = 0;
 	ShaderRecompiler::IR::PushData push_data;
 	bool                           has_push_data = false;
-	constexpr auto                 GraphicsStages =
-	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eMeshEXT |
-	    vk::ShaderStageFlagBits::eTessellationControl |
-	    vk::ShaderStageFlagBits::eTessellationEvaluation | vk::ShaderStageFlagBits::eFragment;
+	constexpr auto                 GraphicsStages = vk::ShaderStageFlagBits::eVertex |
+	                                                vk::ShaderStageFlagBits::eMeshEXT |
+	                                                vk::ShaderStageFlagBits::eFragment;
 	vk::ShaderStageFlags push_stages = pipeline_bind_point == vk::PipelineBindPoint::eGraphics
 	                                       ? vk::ShaderStageFlagBits::eFragment
 	                                       : vk::ShaderStageFlags {};
@@ -1040,7 +995,19 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 					    std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
 					EXIT_IF(storage || host_view == image.views.end());
 					const auto aspect = host_view->info.aspect;
-					if (aspect & ~DepthReadableAspects(layout)) {
+					const bool depth_feedback =
+					    layout == vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT &&
+					    program.stage == ShaderType::Pixel;
+					const bool depth_read =
+					    depth_feedback || layout == vk::ImageLayout::eDepthReadOnlyOptimal ||
+					    layout == vk::ImageLayout::eDepthStencilReadOnlyOptimal ||
+					    layout == vk::ImageLayout::eDepthReadOnlyStencilAttachmentOptimal;
+					const bool stencil_read =
+					    layout == vk::ImageLayout::eStencilReadOnlyOptimal ||
+					    layout == vk::ImageLayout::eDepthStencilReadOnlyOptimal ||
+					    layout == vk::ImageLayout::eDepthAttachmentStencilReadOnlyOptimal;
+					if ((aspect & vk::ImageAspectFlagBits::eDepth && !depth_read) ||
+					    (aspect & vk::ImageAspectFlagBits::eStencil && !stencil_read)) {
 						EXIT("sampling a writable depth/stencil attachment aspect\n");
 					}
 				}
@@ -1084,8 +1051,8 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 			} else {
 				switch (binding.kind) {
 					case BindingKind::Buffers:
-						EXIT_IF(descriptors.buffers.size() != binding.resources.size());
-						for (const auto& view: descriptors.buffers) {
+						for (const auto resource: binding.resources) {
+							const auto& view = descriptors.buffers.at(resource);
 							EXIT_IF(view.buffer == nullptr);
 							m_descriptor_buffers.push_back(view);
 						}

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

@@ -66,6 +66,9 @@ template <typename T>
 
 [[nodiscard]] bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor,
                                                    bool r128 = false);
+[[nodiscard]] bool
+IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
+                               const ShaderTextureResource& descriptor, const Image& image);
 void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
                             const ShaderTextureResource& descriptor, uint64_t size);
 

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

@@ -89,12 +89,17 @@ template <typename... Args>
 void PipelineCacheLog(fmt::format_string<Args...> format, Args&&... args) {
 	auto message = fmt::format(format, std::forward<Args>(args)...);
 	message += '\n';
-	Log::WriteToConsoleAndLog(message);
+	if (Log::GetDirection() != Log::Direction::Console) {
+		std::fwrite(message.data(), 1, message.size(), stdout);
+		std::fflush(stdout);
+	}
+	Log::Write(message);
+	Log::Flush();
 }
 
-bool ReadShaderGuestMemory(void*, uint64_t address, std::span<uint32_t> values) {
-	return !values.empty() &&
-	       Libs::LibKernel::Memory::TryReadGpuCleanBacking(address, values.data(), values.size_bytes());
+bool ReadShaderGuestMemory(void*, uint64_t address, uint32_t* value) {
+	return value != nullptr &&
+	       Libs::LibKernel::Memory::TryReadGpuCleanBacking(address, value, sizeof(*value));
 }
 
 void DumpShaderSpirv(const char* stage_name, uint64_t shader_hash,
@@ -200,11 +205,8 @@ struct PipelineCache::ProgramCache {
 			permutations.reserve(8);
 		}
 
-		ShaderRecompiler::IR::ResourcePlan           resource_plan;
-		ShaderRecompiler::IR::ResourceSnapshot       resources;
-		ShaderRecompiler::IR::ResourceSpecialization specialization;
-		std::vector<Permutation>                    permutations;
-		bool                                        skip_dispatch = false;
+		ShaderRecompiler::IR::ResourcePlan resource_plan;
+		std::vector<Permutation>           permutations;
 	};
 
 	struct ProgramKeyHash {
@@ -234,9 +236,6 @@ struct PipelineCache::ProgramCache {
 		switch (options.stage) {
 			case ShaderType::Vertex: stage_name = "vs"; break;
 			case ShaderType::Mesh: stage_name = "ms"; break;
-			case ShaderType::Local: stage_name = "ls"; break;
-			case ShaderType::TessellationControl: stage_name = "hs"; break;
-			case ShaderType::TessellationEvaluation: stage_name = "ds"; break;
 			case ShaderType::Pixel: stage_name = "ps"; break;
 			case ShaderType::Compute: stage_name = "cs"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -251,9 +250,18 @@ struct PipelineCache::ProgramCache {
 		}
 		DumpShaderSpirv(stage_name, options.shader_hash, result.spirv);
 
-		const auto module = CompileSPV(result.spirv, device);
+		vk::ShaderModuleCreateInfo create_info {};
+		create_info.codeSize    = result.spirv.size() * sizeof(uint32_t);
+		create_info.pCode       = result.spirv.data();
+		vk::ShaderModule module = nullptr;
+		RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
+		                     "create recompiled shader module");
 		EXIT_IF(module == nullptr);
 		if (options.dump_ir) {
+			if (!options.early_dump) {
+				LOGF("%s decoded RDNA2:\n%s", options.dump_label, result.decoded_dump.c_str());
+				LOGF("%s IR:\n%s", options.dump_label, result.ir_dump.c_str());
+			}
 			LOGF("%s SPIR-V words=%" PRIu64 " wave_size=%u\n", options.dump_label,
 			     static_cast<uint64_t>(result.spirv.size()), options.wave_size);
 		}
@@ -269,7 +277,7 @@ struct PipelineCache::ProgramCache {
 	                  uint32_t& push_data_cursor) {
 		ShaderType stage;
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
-			stage = input_info.logical_stage;
+			stage = input_info.mesh.threads_num[0] != 0 ? ShaderType::Mesh : ShaderType::Vertex;
 		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
 			stage = ShaderType::Pixel;
 		} else {
@@ -277,36 +285,33 @@ struct PipelineCache::ProgramCache {
 			stage = ShaderType::Compute;
 		}
 
-		const auto user_data = std::span(params.user_data).first(params.user_data_count);
 		lookup_key.stage           = stage;
 		lookup_key.hash            = params.hash;
-		lookup_key.user_data_count = params.user_data_count;
+		lookup_key.user_data_count = static_cast<uint32_t>(params.user_data.size());
 		lookup_key.code_size       = static_cast<uint32_t>(params.code.size());
 		BuildStageStaticKey(input_info, lookup_key.static_state);
 		auto                                         entry = programs.find(lookup_key);
-		if (entry != programs.end() && entry->second.skip_dispatch) {
-			return {};
-		}
+		ShaderRecompiler::IR::ResourceSnapshot       resources;
+		ShaderRecompiler::IR::ResourceSpecialization specialization;
 		const ShaderRecompiler::IR::SrtRuntime       runtime {
-		    .user_data                  = user_data,
+		    .user_data                  = params.user_data,
 		    .shader_base                = params.Base(),
 		    .read_specialization_memory = ReadShaderGuestMemory,
 		};
 		if (entry != programs.end()) {
 			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(
-			    entry->second.resource_plan, runtime, entry->second.resources,
-			    entry->second.specialization));
+			    entry->second.resource_plan, runtime, resources, specialization));
 			if (const auto permutation = std::ranges::find_if(
 			        entry->second.permutations, [&](const Permutation& candidate) {
 				        const auto& layout = candidate.program.bindings;
 				        return layout.push_data_start_dword ==
 				                   ShaderRecompiler::IR::PushData::StartFor(
 				                       push_data_cursor, layout.ShaderDataDwords()) &&
-				               candidate.specialization == entry->second.specialization;
+				               candidate.specialization == specialization;
 			        });
 			    permutation != entry->second.permutations.end()) {
 				input_info.stage = {.program   = &permutation->program,
-				                    .resources = &entry->second.resources};
+				                    .resources = std::move(resources)};
 				permutation->program.bindings.AdvancePushData(push_data_cursor);
 				return permutation->handle;
 			}
@@ -324,9 +329,6 @@ struct PipelineCache::ProgramCache {
 		switch (stage) {
 			case ShaderType::Vertex: label = "ShaderRecompiler VS"; break;
 			case ShaderType::Mesh: label = "ShaderRecompiler MS"; break;
-			case ShaderType::Local: label = "ShaderRecompiler LS"; break;
-			case ShaderType::TessellationControl: label = "ShaderRecompiler HS"; break;
-			case ShaderType::TessellationEvaluation: label = "ShaderRecompiler DS"; break;
 			case ShaderType::Pixel: label = "ShaderRecompiler PS"; break;
 			case ShaderType::Compute: label = "ShaderRecompiler CS"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -334,55 +336,46 @@ struct PipelineCache::ProgramCache {
 		ShaderRecompiler::CompileOptions options;
 		options.stage       = stage;
 		options.shader_hash = params.hash;
-		options.user_data   = user_data;
+		options.user_data   = params.user_data;
 		options.back_code      = params.back_code;
-		options.dump_ir     = Config::GetShaderLogDirection() != Config::LogDirection::Silent;
+		options.dump_ir     = Config::GetShaderLogDirection() != Config::ShaderLogDirection::Silent;
 		options.early_dump  = options.dump_ir;
 		options.dump_label  = label;
 		options.input_info  = stage_input;
-
+		options.scratch_dwords = input_info.scratch_size_dwords;
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
 			options.user_data_base = 8;
-			options.wave_size = input_info.wave_size;
-			if (stage == ShaderType::Mesh || stage == ShaderType::TessellationControl) {
+			if (stage == ShaderType::Mesh) {
 				options.user_data_base = 0;
-				options.wave_size = stage == ShaderType::Mesh ? input_info.mesh.wave_size : 64u;
+				options.wave_size      = input_info.mesh.wave_size;
+				options.scratch_dwords = input_info.mesh.scratch_size_dwords;
 			}
-		} else {
+		} else if constexpr (std::is_same_v<InputInfo, ShaderComputeInputInfo>) {
 			options.wave_size = input_info.wave_size;
 		}
 		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
-		if (translated.skip_dispatch) {
-			entry = programs.try_emplace(lookup_key, ShaderRecompiler::IR::ResourcePlan {}).first;
-			entry->second.skip_dispatch = true;
-			return {};
-		}
 		if (entry == programs.end()) {
-			entry = programs.try_emplace(lookup_key,
-			    ShaderRecompiler::IR::ExtractResourcePlan(translated.program)).first;
-			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(
-			    entry->second.resource_plan, runtime, entry->second.resources,
-			    entry->second.specialization));
+			auto resource_plan = ShaderRecompiler::IR::ExtractResourcePlan(translated.program);
+			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(resource_plan, runtime, resources,
+			                                                    specialization));
+			entry = programs.try_emplace(lookup_key, std::move(resource_plan)).first;
 		}
 		entry->second.permutations.push_back(CompilePermutation(
-		    params, options, std::move(translated), entry->second.specialization, push_data_cursor));
+		    params, options, std::move(translated), std::move(specialization), push_data_cursor));
 		const auto& permutation = entry->second.permutations.back();
-		input_info.stage = {.program = &permutation.program, .resources = &entry->second.resources};
+		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
 		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
-		std::array<size_t, static_cast<size_t>(ShaderType::TessellationEvaluation) + 1> counts {};
+		std::array<size_t, static_cast<size_t>(ShaderType::Mesh) + 1> counts {};
 		for (const auto& [key, source]: programs) {
 			counts[static_cast<size_t>(key.stage)] += source.permutations.size();
 		}
 		// Guest geometry shaders are compiled through the host mesh stage.
-		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu | LS %zu | HS %zu | TES %zu\n",
+		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu\n",
 		            counts[static_cast<size_t>(ShaderType::Vertex)],
 		            counts[static_cast<size_t>(ShaderType::Pixel)],
 		            counts[static_cast<size_t>(ShaderType::Compute)],
-		            counts[static_cast<size_t>(ShaderType::Mesh)],
-		            counts[static_cast<size_t>(ShaderType::Local)],
-		            counts[static_cast<size_t>(ShaderType::TessellationControl)],
-		            counts[static_cast<size_t>(ShaderType::TessellationEvaluation)]);
+		            counts[static_cast<size_t>(ShaderType::Mesh)]);
 		return permutation.handle;
 	}
 
@@ -576,18 +569,12 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
     const HW::VertexShaderInfo& vertex_regs, const HW::PixelShaderInfo& pixel_regs,
     const HW::ShaderRegisters& sh, const HW::Context& context, const HW::UserConfig& user_config,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping, bool pixel_active,
-    std::array<ShaderVertexInputInfo, 3>& vertex_info, ShaderPixelInputInfo& pixel_info) {
-	const bool tess_active = user_config.GetPrimType() == Prospero::PrimitiveType::kPatch;
-	std::array<ShaderParams, 3> vertex_params;
-	if (tess_active) {
-		vertex_params = PrepareTessellationPrograms(vertex_regs, context, vertex_info);
-	} else {
-		vertex_params[0] = PrepareProgram(vertex_regs, context, user_config, vertex_info[0]);
-	}
-	const bool mesh_active = vertex_info[0].logical_stage == ShaderType::Mesh;
+    ShaderVertexInputInfo& vertex_info, ShaderPixelInputInfo& pixel_info) {
+	const auto vertex_params = PrepareProgram(vertex_regs, context, user_config, vertex_info);
+	const bool mesh_active   = vertex_info.mesh.threads_num[0] != 0;
 	if (mesh_active) {
 		EXIT_NOT_IMPLEMENTED(!m_graphics.mesh_shader_enabled);
-		auto& mesh              = vertex_info[0].mesh;
+		auto& mesh              = vertex_info.mesh;
 		mesh.host_subgroup_size = m_graphics.subgroup_size;
 		const auto& limits      = m_graphics.mesh_shader_properties;
 		const auto  logical_threads =
@@ -606,26 +593,11 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	ShaderParams pixel_params;
 	if (pixel_active) {
 		pixel_params = PrepareProgram(pixel_regs, sh, target_export_mapping, pixel_info);
-		const auto& blend          = context.GetBlendControl(0);
-		const auto  is_dual_source = [](uint8_t factor) {
-			return factor >= static_cast<uint8_t>(Prospero::BlendFactor::kSrc1Color) &&
-			       factor <= static_cast<uint8_t>(Prospero::BlendFactor::kOneMinusSrc1Alpha);
-		};
-		pixel_info.dual_source_blending =
-		    blend.enable && !context.GetRenderTarget(0).info.blend_bypass &&
-		    (is_dual_source(blend.color_srcblend) || is_dual_source(blend.color_destblend) ||
-		     (blend.separate_alpha_blend &&
-		      (is_dual_source(blend.alpha_srcblend) || is_dual_source(blend.alpha_destblend))));
-		if (pixel_info.dual_source_blending) {
-			// MRT1 supplies a second blend source for the same render target as MRT0.
-			pixel_info.target_output_mode[1]    = pixel_info.target_output_mode[0];
-			pixel_info.target_export_mapping[1] = pixel_info.target_export_mapping[0];
-		}
 	}
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];
 		const auto& limits   = m_graphics.GetPhysicalDeviceProperties().limits;
-		auto&       clip     = vertex_info[tess_active ? 2u : 0u].clip_space;
+		auto&       clip     = vertex_info.clip_space;
 		clip.scale[0]        = viewport.xscale;
 		clip.scale[1]        = viewport.yscale;
 		clip.offset[0]       = viewport.xoffset;
@@ -643,9 +615,7 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (pixel_active) {
 		result.pixel = m_program_cache->Get(pixel_params, pixel_info, push_data_cursor);
 	}
-	for (uint32_t i = 0; i < (tess_active ? 3u : 1u); i++) {
-		result.vertex[i] = m_program_cache->Get(vertex_params[i], vertex_info[i], push_data_cursor);
-	}
+	result.vertex = m_program_cache->Get(vertex_params, vertex_info, push_data_cursor);
 	return result;
 }
 
@@ -663,14 +633,12 @@ bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other)
 	return std::memcmp(this, &other, sizeof(*this)) == 0;
 }
 
-PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
+PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
     std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-    std::span<const ShaderVertexInputInfo> vertex_info, CommandBuffer& command,
+    const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
     const ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
-    bool primitive_restart_enable, const GraphicsPrograms& programs) {
-	const auto& vs_input_info  = vertex_info.front();
-	const auto& vertex_program = programs.vertex[0];
-	const auto& pixel_program  = programs.pixel;
+    bool primitive_restart_enable, const ShaderProgram& vertex_program,
+    const ShaderProgram& pixel_program) {
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Gfx)", profiler::colors::DeepOrangeA200);
 
 	EXIT_IF(colors.size() > RENDER_COLOR_ATTACHMENTS_MAX);
@@ -688,38 +656,23 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	const auto ps_id = ps_active ? pixel_program.id : 0;
 
 	GraphicsPipelineKey key {};
-	for (uint32_t i = 0; i < programs.vertex.size(); i++) {
-		key.vertex_shader_ids[i] = programs.vertex[i].id;
-	}
+	key.vs_shader_id            = vs_id;
 	key.ps_shader_id            = ps_id;
 	auto& static_params         = key.static_params;
 	auto& rendering             = key.rendering;
-	rendering.color_count       = 0;
+	rendering.color_count       = color_count;
 	uint32_t attachment_samples = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
-		const auto slot = colors[i].target_slot;
-		EXIT_IF(slot >= RENDER_COLOR_ATTACHMENTS_MAX);
-		rendering.color_count = std::max(rendering.color_count, slot + 1);
 		EXIT_IF(!colors[i].image_id || colors[i].desc.view_info.format == vk::Format::eUndefined);
-		static_params.color_mask[slot] = colors[i].export_mapping.ApplyMask(
+		static_params.color_mask[i] = colors[i].export_mapping.ApplyMask(
 		    render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot));
-		rendering.color_formats[slot] = colors[i].desc.view_info.format;
+		rendering.color_formats[i] = colors[i].desc.view_info.format;
 		if (attachment_samples == 0) {
 			attachment_samples = colors[i].desc.info.samples;
 		} else if (attachment_samples != colors[i].desc.info.samples) {
 			EXIT("mixed color attachment sample counts are unsupported: %u and %u\n",
 			     attachment_samples, colors[i].desc.info.samples);
 		}
-		const auto& rt                        = ctx.GetRenderTarget(colors[i].target_slot);
-		const auto& bc                        = ctx.GetBlendControl(colors[i].target_slot);
-		static_params.color_srcblend[slot]       = bc.color_srcblend;
-		static_params.color_comb_fcn[slot]       = bc.color_comb_fcn;
-		static_params.color_destblend[slot]      = bc.color_destblend;
-		static_params.alpha_srcblend[slot]       = bc.alpha_srcblend;
-		static_params.alpha_comb_fcn[slot]       = bc.alpha_comb_fcn;
-		static_params.alpha_destblend[slot]      = bc.alpha_destblend;
-		static_params.separate_alpha_blend[slot] = bc.separate_alpha_blend;
-		static_params.blend_enable[slot]         = bc.enable && !rt.info.blend_bypass;
 	}
 	const bool with_depth =
 	    depth.desc.view_info.format != vk::Format::eUndefined && static_cast<bool>(depth.image_id);
@@ -768,7 +721,10 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	static_params.depth_bounds_test_enable = depth.depth_bounds_test_enable;
 	static_params.depth_min_bounds         = depth.depth_min_bounds;
 	static_params.depth_max_bounds         = depth.depth_max_bounds;
-	const bool rect_list = Prospero::IsRectList(command.GetUserConfig().GetPrimType());
+	static_params.stencil_test_enable      = depth.stencil_test_enable;
+	static_params.stencil_front            = depth.stencil_static_front;
+	static_params.stencil_back             = depth.stencil_static_back;
+	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;
@@ -776,6 +732,19 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	static_params.polygon_mode =
 	    ResolvePolygonMode(mc, static_params.cull_front, static_params.cull_back);
 
+	for (uint32_t i = 0; i < color_count; i++) {
+		const auto& rt                        = ctx.GetRenderTarget(colors[i].target_slot);
+		const auto& bc                        = ctx.GetBlendControl(colors[i].target_slot);
+		static_params.color_srcblend[i]       = bc.color_srcblend;
+		static_params.color_comb_fcn[i]       = bc.color_comb_fcn;
+		static_params.color_destblend[i]      = bc.color_destblend;
+		static_params.alpha_srcblend[i]       = bc.alpha_srcblend;
+		static_params.alpha_comb_fcn[i]       = bc.alpha_comb_fcn;
+		static_params.alpha_destblend[i]      = bc.alpha_destblend;
+		static_params.separate_alpha_blend[i] = bc.separate_alpha_blend;
+		static_params.blend_enable[i]         = bc.enable;
+		static_params.blend_bypass[i]         = rt.info.blend_bypass;
+	}
 	if (vs_input_info.stage.program->stage != ShaderType::Mesh) {
 		EXIT_IF(vs_input_info.buffers_num < 0 ||
 		        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX ||
@@ -819,8 +788,9 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 
 	auto cached = std::make_unique<Pipeline>();
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id, ps_id);
-	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vertex_info,
-	                       ps_input_info, programs, static_params, m_driver_cache);
+	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vs_input_info,
+	                       vertex_program, ps_input_info, pixel_program, static_params,
+	                       m_driver_cache);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id, ps_id);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);
@@ -833,8 +803,8 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 }
 
 PipelineCache::Pipeline&
-PipelineCache::GetComputePipeline(const ShaderComputeInputInfo& input_info,
-                                  const ShaderProgram&          compute_program) {
+PipelineCache::CreateComputePipeline(const ShaderComputeInputInfo& input_info,
+                                     const ShaderProgram&          compute_program) {
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Compute)", profiler::colors::RedA100);
 
 	EXIT_IF(!compute_program);

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

@@ -42,6 +42,9 @@ struct PipelineStaticParameters {
 	bool                       depth_bounds_test_enable = false;
 	float                      depth_min_bounds         = 0.0f;
 	float                      depth_max_bounds         = 0.0f;
+	bool                       stencil_test_enable      = false;
+	PipelineStencilStaticState stencil_front;
+	PipelineStencilStaticState stencil_back;
 	uint32_t                   color_mask[RENDER_COLOR_ATTACHMENTS_MAX]           = {};
 	bool                       cull_front                                         = false;
 	bool                       cull_back                                          = false;
@@ -56,6 +59,7 @@ struct PipelineStaticParameters {
 	uint8_t                    alpha_destblend[RENDER_COLOR_ATTACHMENTS_MAX]      = {};
 	bool                       separate_alpha_blend[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	bool                       blend_enable[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
+	bool                       blend_bypass[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
 
 	bool operator==(const PipelineStaticParameters& other) const noexcept;
 };
@@ -65,7 +69,7 @@ struct PipelineStaticParameters {
 static_assert(std::is_trivially_copyable_v<PipelineStaticParameters>);
 static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
-static_assert(sizeof(PipelineStaticParameters) == 125);
+static_assert(sizeof(PipelineStaticParameters) == 166);
 
 struct PipelineRenderingState {
 	std::array<vk::Format, RENDER_COLOR_ATTACHMENTS_MAX> color_formats {};
@@ -118,10 +122,8 @@ public:
 	};
 
 	struct GraphicsPrograms {
-		std::array<ShaderProgram, 3> vertex;
+		ShaderProgram vertex;
 		ShaderProgram pixel;
-
-		[[nodiscard]] uint32_t VertexStageCount() const { return vertex[1] ? 3u : 1u; }
 	};
 
 	GraphicsPrograms
@@ -129,33 +131,33 @@ public:
 	                    const HW::PixelShaderInfo& pixel_regs, const HW::ShaderRegisters& sh,
 	                    const HW::Context& context, const HW::UserConfig& user_config,
 	                    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-	                    bool pixel_active, std::array<ShaderVertexInputInfo, 3>& vertex_info,
+	                    bool pixel_active, ShaderVertexInputInfo& vertex_info,
 	                    ShaderPixelInputInfo& pixel_info);
 	ShaderProgram GetComputeProgram(const HW::ComputeShaderInfo& regs,
 	                                const HW::ShaderRegisters&   sh,
 	                                ShaderComputeInputInfo&      input_info);
 
-	Pipeline& GetGraphicsPipeline(std::span<const RenderColorInfo>       colors,
-	                              const RenderDepthInfo&                 depth,
-	                              std::span<const ShaderVertexInputInfo> vertex_info,
-	                              CommandBuffer& command, const ShaderPixelInputInfo* ps_input_info,
-	                              vk::PrimitiveTopology topology, bool primitive_restart_enable,
-	                              const GraphicsPrograms& programs);
-	Pipeline& GetComputePipeline(const ShaderComputeInputInfo& input_info,
-	                             const ShaderProgram&          compute_program);
+	Pipeline&
+	CreateGraphicsPipeline(std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
+	                       const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
+	                       const ShaderPixelInputInfo* ps_input_info,
+	                       vk::PrimitiveTopology topology, bool primitive_restart_enable,
+	                       const ShaderProgram& vertex_program, const ShaderProgram& pixel_program);
+	Pipeline& CreateComputePipeline(const ShaderComputeInputInfo& input_info,
+	                                const ShaderProgram&          compute_program);
 
 private:
 	struct ProgramCache;
 
 	struct GraphicsPipelineKey {
 		PipelineRenderingState   rendering;
-		std::array<uint64_t, 3>  vertex_shader_ids {};
+		uint64_t                 vs_shader_id = 0;
 		uint64_t                 ps_shader_id = 0;
 		PipelineVertexInputState vertex_input;
 		PipelineStaticParameters static_params;
 
 		bool operator==(const GraphicsPipelineKey& other) const {
-			return rendering == other.rendering && vertex_shader_ids == other.vertex_shader_ids &&
+			return rendering == other.rendering && vs_shader_id == other.vs_shader_id &&
 			       ps_shader_id == other.ps_shader_id && vertex_input == other.vertex_input &&
 			       static_params == other.static_params;
 		}
@@ -188,9 +190,7 @@ private:
 		std::size_t operator()(const GraphicsPipelineKey& key) const {
 			std::size_t hash = 0;
 			PipelineKeyHash::MixRendering(hash, key.rendering);
-			for (const auto id: key.vertex_shader_ids) {
-				PipelineKeyHash::Mix(hash, id);
-			}
+			PipelineKeyHash::Mix(hash, key.vs_shader_id);
 			PipelineKeyHash::Mix(hash, key.ps_shader_id);
 			PipelineKeyHash::Mix(hash, key.vertex_input.binding_count);
 			for (uint32_t i = 0; i < key.vertex_input.binding_count; i++) {
@@ -220,14 +220,12 @@ private:
 };
 
 void LogPipelineTrace(const char* phase, uint64_t vertex_program_id, uint64_t pixel_program_id);
-void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-                            const PipelineRenderingState&          rendering,
-                            const PipelineVertexInputState&        vertex_input,
-                            std::span<const ShaderVertexInputInfo> vertex_info,
-                            const ShaderPixelInputInfo*            ps_input_info,
-                            const PipelineCache::GraphicsPrograms& programs,
-                            const PipelineStaticParameters&        static_params,
-                            vk::PipelineCache                      driver_cache);
+void CreatePipelineInternal(
+    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
+    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
+    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
+    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache);
 void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
                             vk::ShaderModule compute_module, vk::PipelineCache driver_cache);

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

@@ -8,20 +8,6 @@
 
 namespace Libs::Graphics {
 
-vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage) {
-	switch (stage) {
-		case ShaderType::Local:
-		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
-		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
-		case ShaderType::TessellationControl: return vk::ShaderStageFlagBits::eTessellationControl;
-		case ShaderType::TessellationEvaluation:
-			return vk::ShaderStageFlagBits::eTessellationEvaluation;
-		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
-		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
-		default: EXIT("unknown native shader stage\n");
-	}
-}
-
 vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	vk::PipelineStageFlags result = {};
 	if (stages & vk::ShaderStageFlagBits::eVertex) {
@@ -30,12 +16,6 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	if (stages & vk::ShaderStageFlagBits::eMeshEXT) {
 		result |= vk::PipelineStageFlagBits::eMeshShaderEXT;
 	}
-	if (stages & vk::ShaderStageFlagBits::eTessellationControl) {
-		result |= vk::PipelineStageFlagBits::eTessellationControlShader;
-	}
-	if (stages & vk::ShaderStageFlagBits::eTessellationEvaluation) {
-		result |= vk::PipelineStageFlagBits::eTessellationEvaluationShader;
-	}
 	if (stages & vk::ShaderStageFlagBits::eFragment) {
 		result |= vk::PipelineStageFlagBits::eFragmentShader;
 	}
@@ -46,8 +26,8 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	return result;
 }
 
-vk::MemoryBarrier MakeShaderWriteDependency() {
-	vk::MemoryBarrier barrier {};
+VulkanMemoryBarrier MakeShaderWriteDependency() {
+	VulkanMemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite |
 	                        vk::AccessFlagBits::eVertexAttributeRead |
@@ -58,15 +38,15 @@ vk::MemoryBarrier MakeShaderWriteDependency() {
 	return barrier;
 }
 
-vk::MemoryBarrier MakeShaderAccessDependency() {
-	vk::MemoryBarrier barrier {};
+VulkanMemoryBarrier MakeShaderAccessDependency() {
+	VulkanMemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	return barrier;
 }
 
-vk::MemoryBarrier MakeShaderWriteHazardDependency() {
-	vk::MemoryBarrier barrier {};
+VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
+	VulkanMemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	return barrier;
@@ -90,7 +70,7 @@ vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
 	EXIT_IF(!runtime);
 	const auto& program   = *runtime.program;
-	const auto& resources = *runtime.resources;
+	const auto& resources = runtime.resources;
 	EXIT_IF(resources.buffers.size() != program.info.buffers.size());
 	bool has_writes = false;
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {

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

@@ -8,11 +8,10 @@ namespace Libs::Graphics {
 
 struct ShaderStageRuntime;
 
-vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage);
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
-vk::MemoryBarrier       MakeShaderAccessDependency();
-vk::MemoryBarrier       MakeShaderWriteHazardDependency();
-vk::MemoryBarrier       MakeShaderWriteDependency();
+VulkanMemoryBarrier     MakeShaderAccessDependency();
+VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
+VulkanMemoryBarrier     MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
 void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);

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

@@ -8,7 +8,6 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
@@ -117,6 +116,10 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		}
 		format = vk::Format::eR16G16Sfloat;
 		size   = 2;
+		if (NarrowInputFormat(format, size, used_components)) {
+			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
+			     vk::to_string(format).c_str(), used_components);
+		}
 		return;
 	}
 
@@ -206,18 +209,12 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-                            const PipelineRenderingState&          rendering,
-                            const PipelineVertexInputState&        vertex_input,
-                            std::span<const ShaderVertexInputInfo> vertex_info,
-                            const ShaderPixelInputInfo*            ps_input_info,
-                            const PipelineCache::GraphicsPrograms& programs,
-                            const PipelineStaticParameters&        static_params,
-                            vk::PipelineCache                      driver_cache) {
-	const auto& vs_input_info  = vertex_info.front();
-	const auto& vertex_program = programs.vertex[0];
-	const auto& pixel_program  = programs.pixel;
-	const bool  tessellation   = vertex_info.size() == 3;
+void CreatePipelineInternal(
+    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
+    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
+    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
+    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache) {
 	const bool ps_active = ps_input_info != nullptr;
 	EXIT_IF(!vertex_program || (ps_active && !pixel_program));
 	const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
@@ -225,53 +222,72 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 	EXIT_IF(!vs_input_info.stage);
 	const bool mesh = vs_input_info.stage.program->stage == ShaderType::Mesh;
 	EXIT_NOT_IMPLEMENTED(mesh && !graphics.mesh_shader_enabled);
-	const bool rect_list =
-	    !mesh && !tessellation && static_params.topology == vk::PrimitiveTopology::ePatchList;
+	const auto vertex_stage =
+	    mesh ? vk::ShaderStageFlagBits::eMeshEXT : vk::ShaderStageFlagBits::eVertex;
+
+	const bool rect_list = !mesh && static_params.topology == vk::PrimitiveTopology::ePatchList;
 
 	vk::ShaderModule tess_control_shader_module = nullptr;
 	vk::ShaderModule tess_eval_shader_module    = nullptr;
 
+	vk::ShaderModuleCreateInfo create_info {};
+	vk::Result result {};
 	if (rect_list) {
 		const auto shaders =
 		    BuildRectListShaders(vs_input_info, ps_active ? ps_input_info : nullptr);
-		tess_control_shader_module = CompileSPV(shaders.control, graphics.device);
+		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 module=%p\n",
-			     static_cast<void*>(tess_control_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
+			     vk::to_string(result).c_str(), static_cast<void*>(tess_control_shader_module));
 		}
+		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
-		tess_eval_shader_module = CompileSPV(shaders.evaluation, graphics.device);
+		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 module=%p\n",
-			     static_cast<void*>(tess_eval_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
+			     vk::to_string(result).c_str(), static_cast<void*>(tess_eval_shader_module));
 		}
+		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
 
 	EXIT_NOT_IMPLEMENTED(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 
-	vk::PipelineShaderStageCreateInfo shader_stages[4] {};
+	vk::PipelineShaderStageCreateInfo vert_shader_stage_info {};
+	vert_shader_stage_info.stage  = vertex_stage;
+	vert_shader_stage_info.module = vertex_program.module;
+	vert_shader_stage_info.pName  = "main";
+
+	vk::PipelineShaderStageCreateInfo frag_shader_stage_info {};
+	frag_shader_stage_info.stage  = vk::ShaderStageFlagBits::eFragment;
+	frag_shader_stage_info.module = pixel_program.module;
+	frag_shader_stage_info.pName  = "main";
+
+	vk::PipelineShaderStageCreateInfo tess_control_shader_stage_info {};
+	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.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;
-	for (uint32_t i = 0; i < vertex_info.size(); i++) {
-		shader_stages[shader_stage_count++] = {.stage =
-		                                           NativeShaderStage(vertex_info[i].logical_stage),
-		                                       .module = programs.vertex[i].module,
-		                                       .pName  = "main"};
-	}
+	shader_stages[shader_stage_count++]                  = vert_shader_stage_info;
 	if (rect_list) {
-		shader_stages[shader_stage_count++] = {.stage =
-		                                           vk::ShaderStageFlagBits::eTessellationControl,
-		                                       .module = tess_control_shader_module,
-		                                       .pName  = "main"};
-		shader_stages[shader_stage_count++] = {.stage =
-		                                           vk::ShaderStageFlagBits::eTessellationEvaluation,
-		                                       .module = tess_eval_shader_module,
-		                                       .pName  = "main"};
+		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++] = {.stage  = vk::ShaderStageFlagBits::eFragment,
-		                                       .module = pixel_program.module,
-		                                       .pName  = "main"};
+		shader_stages[shader_stage_count++] = frag_shader_stage_info;
 	}
 
 	vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX] {};
@@ -368,8 +384,6 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 	// MoltenVK lacks VK_EXT_depth_clip_enable; omit the depth-clip struct on macOS and accept
 	// Vulkan's default depth clipping (enabled) instead of the PS5's clamp behavior.
 #if !defined(__APPLE__)
-	// The DB clamps depth to the viewport range after polygon offset is applied.
-	rasterizer.depthClampEnable = VK_TRUE;
 	rasterizer.pNext = &clip_ext;
 #endif
 	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provoking_vertex {};
@@ -396,7 +410,8 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 		EXIT_NOT_IMPLEMENTED((static_params.color_mask[i] & ~0x0fu) != 0);
 		color_blend_attachment[i].colorWriteMask =
 		    vk::ColorComponentFlags {static_params.color_mask[i]};
-		color_blend_attachment[i].blendEnable = static_params.blend_enable[i] ? VK_TRUE : VK_FALSE;
+		color_blend_attachment[i].blendEnable =
+		    (static_params.blend_enable[i] && !static_params.blend_bypass[i]) ? VK_TRUE : VK_FALSE;
 		color_blend_attachment[i].srcColorBlendFactor =
 		    GetBlendFactor(static_params.color_srcblend[i]);
 		color_blend_attachment[i].dstColorBlendFactor =
@@ -434,19 +449,15 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 	color_blending.pAttachments    = color_blend_attachment;
 
 	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
-	vk::ShaderStageFlags graphics_stages = vk::ShaderStageFlagBits::eFragment;
-	for (const auto& stage: vertex_info) {
-		const auto native_stage = NativeShaderStage(stage.logical_stage);
-		AddLayoutBindings(descriptor_bindings, *stage.stage.program, native_stage);
-		graphics_stages |= native_stage;
-	}
+	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program, vertex_stage);
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
 		                  vk::ShaderStageFlagBits::eFragment);
 	}
 	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
-	const vk::PushConstantRange push_constants {graphics_stages, 0,
+	const auto                  GraphicsStages = vertex_stage | vk::ShaderStageFlagBits::eFragment;
+	const vk::PushConstantRange push_constants {GraphicsStages, 0,
 	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
@@ -462,8 +473,8 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 		     " set_layouts=1 push_constants=%" PRIu32 "\n",
 		     vertex_program.id, ps_active ? pixel_program.id : 0, 1u);
 	}
-	auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
-	                                                   &pipeline.pipeline_layout);
+	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
+	                                              &pipeline.pipeline_layout);
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout done result=%s layout=%p\n",
 		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
@@ -479,6 +490,15 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 #else
 	    (static_params.depth_bounds_test_enable ? VK_TRUE : VK_FALSE);
 #endif
+	depth_stencil_info.stencilTestEnable = (static_params.stencil_test_enable ? VK_TRUE : VK_FALSE);
+	depth_stencil_info.front.failOp      = static_params.stencil_front.failOp;
+	depth_stencil_info.front.passOp      = static_params.stencil_front.passOp;
+	depth_stencil_info.front.depthFailOp = static_params.stencil_front.depthFailOp;
+	depth_stencil_info.front.compareOp   = static_params.stencil_front.compareOp;
+	depth_stencil_info.back.failOp       = static_params.stencil_back.failOp;
+	depth_stencil_info.back.passOp       = static_params.stencil_back.passOp;
+	depth_stencil_info.back.depthFailOp  = static_params.stencil_back.depthFailOp;
+	depth_stencil_info.back.compareOp    = static_params.stencil_back.compareOp;
 	depth_stencil_info.minDepthBounds    = static_params.depth_min_bounds;
 	depth_stencil_info.maxDepthBounds    = static_params.depth_max_bounds;
 
@@ -491,8 +511,6 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 	    vk::DynamicState::eDepthCompareOp,
 	    vk::DynamicState::eDepthBiasEnable,
 	    vk::DynamicState::eDepthBias,
-	    vk::DynamicState::eStencilTestEnable,
-	    vk::DynamicState::eStencilOp,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
@@ -523,9 +541,8 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
 	pipeline_info.pVertexInputState        = mesh ? nullptr : &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = mesh ? nullptr : &input_assembly;
 	vk::PipelineTessellationStateCreateInfo tessellation_state {};
-	tessellation_state.patchControlPoints =
-	    tessellation ? vs_input_info.tess.input_control_points : 3u;
-	pipeline_info.pTessellationState = (rect_list || tessellation) ? &tessellation_state : nullptr;
+	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;

src/graphics/host_gpu/renderer/render.h

@@ -45,7 +45,6 @@ enum class CommandBufferDebugOp : uint32_t {
 	EopFlip,
 	EopWriteBackFlip,
 	EopOnlyFlip,
-	DispatchIndirect,
 	Unknown,
 };
 
@@ -157,10 +156,8 @@ public:
 
 	void DispatchDirect(uint64_t submit_id, CommandBuffer& buffer, uint32_t thread_group_x,
 	                    uint32_t thread_group_y, uint32_t thread_group_z, uint32_t mode);
-	void DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer, uint64_t args_addr,
-	                      uint32_t mode);
 
-	void PrepareBindings(const ShaderStageRuntime& runtime, PreparedBindings& prepared);
+	[[nodiscard]] PreparedBindings PrepareBindings(const ShaderStageRuntime& runtime);
 	void                           FindBuffers(PreparedBindings& bindings);
 	void                           RebindBuffers(PreparedBindings& bindings);
 	void                           RebindImages(PreparedBindings& bindings);
@@ -173,14 +170,15 @@ private:
 	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const DrawAutoArgs& args);
 
 	struct GraphicsBindings {
-		std::array<PreparedBindings, 3> vertex;
+		PreparedBindings                vertex;
 		std::optional<PreparedBindings> pixel;
 	};
 
 	[[nodiscard]] TextureBinding ResolveTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	                                            const ShaderRecompiler::IR::DescriptorValue& value);
-	void PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
-	                             std::span<RenderColorInfo> colors);
+	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
+	                                                       const ShaderStageRuntime& pixel,
+	                                                       bool                      pixel_active);
 	void ResolveRenderColorTarget(CommandBuffer& buffer, RenderColorInfo& target,
 	                              uint32_t render_target_slice_offset, uint32_t render_target_slot,
 	                              bool ignore_target_mask = false, bool exact_format = false);
@@ -189,19 +187,20 @@ private:
 	[[nodiscard]] bool PrepareDrawRenderState(CommandBuffer& buffer,
 	                                          const DrawCallInfo& draw,
 	                                          uint32_t            render_target_slice_offset,
-	                                          DrawRenderState& state);
+	                                          bool log_setup_phases, DrawRenderState& state);
 	void ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buffer, const DrawCallInfo& draw,
 	                         DrawRenderState& state, vk::PrimitiveTopology topology,
 	                         const DrawEmitInfo& emit, const DrawIndexBufferSource& index_source,
-	                         bool primitive_restart_enable);
+	                         bool primitive_restart_enable, bool log_pipeline_phase,
+	                         bool set_bind_debug, bool set_auto_debug);
 	[[nodiscard]] RenderState AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
 	                                               uint32_t color_count, RenderDepthInfo& depth,
-	                                               vk::ImageAspectFlags& feedback_aspects,
-	                                               std::span<PreparedBindings* const> stages = {});
+	                                               const std::optional<PreparedBindings>& pixel = std::nullopt);
 	[[nodiscard]] bool        ResolveColorTargets(CommandBuffer& buffer,
 	                                              uint32_t render_target_slice_offset);
 	void                      BindImage(ImageId id, bool storage);
 	void                      BindRenderTarget(ImageId id);
+	void                      TrackImageBinding(ImageId id);
 	void                      ResetBindings();
 	[[nodiscard]] bool        TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
 	                                                     const CommandBuffer&          buffer);
@@ -210,8 +209,6 @@ private:
 	                                              uint32_t group_y, uint32_t group_z, uint32_t mode);
 
 	RenderContext&                        m_context;
-	GraphicsBindings                     m_graphics_bindings;
-	PreparedBindings                     m_compute_bindings;
 	std::vector<ImageId>                  m_bound_images;
 	std::vector<vk::DescriptorBufferInfo> m_descriptor_buffers;
 	std::vector<vk::DescriptorImageInfo>  m_descriptor_images;

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -9,7 +9,6 @@
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/guest_gpu/hardwareContext.h"
-#include "graphics/guest_gpu/pm4.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/image/imageInfo.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
@@ -37,12 +36,18 @@
 #include <vector>
 
 namespace Libs::Graphics {
+static uint64_t BufferDescriptorSize(const ShaderBufferResource& descriptor) {
+	const uint64_t records = descriptor.NumRecords();
+	const uint64_t stride  = descriptor.Stride();
+	return stride == 0 ? records : records * stride;
+}
+
 static bool FillSourcesDisjoint(std::span<const ShaderRecompiler::IR::DescriptorValue> sources,
                                  GuestRange destination, uint32_t output_buffer = UINT32_MAX) {
 	for (uint32_t i = 0; i < sources.size(); ++i) {
 		if (i == output_buffer) continue;
 		const auto source = DecodeNativeDescriptor<ShaderBufferResource>(sources[i]);
-		const auto bytes  = source.GetSize();
+		const auto bytes  = BufferDescriptorSize(source);
 		if (source.Base48() < destination.End() && destination.address < source.Base48() + bytes)
 			return false;
 	}
@@ -52,7 +57,7 @@ static bool FillSourcesDisjoint(std::span<const ShaderRecompiler::IR::Descriptor
 bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
                                                 const CommandBuffer&          buffer) {
 	const auto& program   = *input.stage.program;
-	const auto& resources = *input.stage.resources;
+	const auto& resources = input.stage.resources;
 	if (resources.buffers.size() != program.info.buffers.size()) {
 		EXIT("compute runtime buffer count does not match shader metadata\n");
 	}
@@ -62,7 +67,7 @@ bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& in
 		const auto  descriptor = DecodeNativeDescriptor<ShaderBufferResource>(resources.buffers[i]);
 		// A metadata resource that is also read is not proven to be a full overwrite. Execute it
 		// conservatively instead of replacing the dispatch with a coarse full-surface clear.
-		if ((!resource.written || resource.read) && cache.IsMeta(descriptor.Base48())) {
+		if (cache.IsMeta(descriptor.Base48()) && (!resource.written || resource.read)) {
 			return false;
 		}
 	}
@@ -86,7 +91,7 @@ bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t grou
                               uint32_t group_y, uint32_t group_z, uint32_t mode,
                               ShaderBufferResource& resolved_descriptor, uint32_t& resolved_clear,
                               uint64_t& resolved_size) {
-	const auto& resources = *input.stage.resources;
+	const auto& resources = input.stage.resources;
 	const auto& fill      = resources.uniform_fill;
 	if (fill.kind != ShaderRecompiler::IR::UniformFillKind::Buffer) {
 		return false;
@@ -110,7 +115,7 @@ bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t grou
 	const uint64_t invocations = input.dispatch_thread_dimensions
 	                                 ? group_x
 	                                 : static_cast<uint64_t>(group_x) * input.threads_num[0];
-	const auto     size        = descriptor.GetSize();
+	const auto     size        = BufferDescriptorSize(descriptor);
 	if (invocations != descriptor.NumRecords() || size == 0 || size > UINT32_MAX ||
 	    (input.dispatch_thread_dimensions &&
 	     (group_x % input.threads_num[0] != 0 || input.dispatch_threads_num[0] != group_x ||
@@ -129,7 +134,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
                                                 CommandBuffer& command, uint32_t group_x,
                                                 uint32_t group_y, uint32_t group_z, uint32_t mode) {
 	const auto& program   = *input.stage.program;
-	const auto& resources = *input.stage.resources;
+	const auto& resources = input.stage.resources;
 	const auto& fill      = resources.uniform_fill;
 	auto&       cache     = command.GetContext().GetTextureCache();
 	if (fill.kind == ShaderRecompiler::IR::UniformFillKind::Image) {
@@ -173,7 +178,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		                                       1, view.base_layer, view.layer_count};
 		vk::ClearValue clear {};
 		clear.depthStencil = vk::ClearDepthStencilValue {0.0f, fill.value};
-		cache.ClearImage(command, binding.image_id, image.backing.format, range, clear);
+		cache.ClearImage(command, binding.image_id, range, clear);
 		return true;
 	}
 	ShaderBufferResource descriptor;
@@ -184,6 +189,14 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		return false;
 	}
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
+		// Track deferred DCC state while the original dispatch writes the metadata allocation.
+		cache.TrackDccFill(descriptor.Base48(), size, packed_clear);
+		static std::atomic<uint32_t> logged_metadata_clears {0};
+		if (logged_metadata_clears.fetch_add(1, std::memory_order_relaxed) < 32) {
+			LOGF("GraphicsRenderDispatchDirect: metadata fill shader=0x%016" PRIx64
+			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
+			     input.stage.program->shader_hash, descriptor.Base48(), size, packed_clear);
+		}
 		return false;
 	}
 	static std::atomic<uint32_t> logged_clears {0};
@@ -207,17 +220,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	                    thread_group_x, thread_group_y, thread_group_z, mode,
 	                    sh_ctx.GetCs().cs_regs.data_addr);
 
-	if (thread_group_x == 0 || thread_group_y == 0 || thread_group_z == 0) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1, std::memory_order_relaxed) < 32) {
-			LOGF("GraphicsRenderDispatchDirect: skipping zero-sized dispatch groups=%ux%ux%u "
-			     "mode=0x%08" PRIx32 " shader=0x%016" PRIx64 "\n",
-			     thread_group_x, thread_group_y, thread_group_z, mode,
-			     sh_ctx.GetCs().cs_regs.data_addr);
-		}
-		return;
-	}
-
 	Common::LockGuard lock(m_context.GetMutex());
 	if (sh_ctx.GetCs().cs_regs.data_addr == 0) {
 		LOGF("GraphicsRenderDispatchDirect: temporary: ignoring dispatch with null CS shader, "
@@ -226,7 +228,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (sh_ctx.GetCs().cs_regs.data_addr == 0) {
+	if (!ShaderAddressValid(sh_ctx.GetCs().cs_regs.data_addr)) {
 		return;
 	}
 
@@ -256,18 +258,17 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	input_info.dispatch_thread_dimensions = use_thread_dimensions;
 	const auto compute_program =
 	    m_context.GetPipelineCache().GetComputeProgram(cs_regs, sh_regs, input_info);
-	if (!compute_program) {
-		// Temporary until RT is implemented.
-		return;
-	}
 	if (use_thread_dimensions) {
 		input_info.dispatch_threads_num[0]    = thread_group_x;
 		input_info.dispatch_threads_num[1]    = thread_group_y;
 		input_info.dispatch_threads_num[2]    = thread_group_z;
 	}
 
+	const uint32_t frame_num = static_cast<uint32_t>(m_context.GetGpu().GetFrameNum());
+	const bool     large_workgroup =
+	    (input_info.threads_num[0] * input_info.threads_num[1] * input_info.threads_num[2] >= 512);
 	const auto& program   = *input_info.stage.program;
-	const auto& resources = *input_info.stage.resources;
+	const auto& resources = input_info.stage.resources;
 	if (TryConsumeComputeMetaClear(input_info, buffer)) {
 		ResetBindings();
 		return;
@@ -277,17 +278,14 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		ResetBindings();
 		return;
 	}
-	const bool large_workgroup =
-	    (input_info.threads_num[0] * input_info.threads_num[1] * input_info.threads_num[2] >= 512);
+	const auto sampled_images = std::count_if(
+	    program.info.images.begin(), program.info.images.end(), [](const auto& image) {
+		    return image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled;
+	    });
 	const bool                   has_sampler = !program.info.samplers.empty();
 	static std::atomic<uint32_t> dispatch_log_count {0};
 	if ((large_workgroup || has_sampler) &&
 	    dispatch_log_count.fetch_add(1, std::memory_order_relaxed) < 512) {
-		const auto sampled_images = std::count_if(
-		    program.info.images.begin(), program.info.images.end(), [](const auto& image) {
-			    return image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled;
-		    });
-		const uint32_t frame_num = static_cast<uint32_t>(m_context.GetGpu().GetFrameNum());
 		LOGF("GraphicsRenderDispatchDirect: frame=%u shader=0x%016" PRIx64
 		     " groups=%ux%ux%u mode=0x%08" PRIx32 " local=%ux%ux%u "
 		     "buffers=%zu textures=%zu sampled=%zu storage=%zu samplers=%zu push=%u\n",
@@ -363,17 +361,27 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		}
 	}
 
+	if (thread_group_x == 0 || thread_group_y == 0 || thread_group_z == 0) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1, std::memory_order_relaxed) < 32) {
+			LOGF("GraphicsRenderDispatchDirect: skipping zero-sized dispatch groups=%ux%ux%u "
+			     "mode=0x%08" PRIx32 " shader=0x%016" PRIx64 "\n",
+			     thread_group_x, thread_group_y, thread_group_z, mode,
+			     sh_ctx.GetCs().cs_regs.data_addr);
+		}
+		return;
+	}
+
 	buffer.EndRendering();
 	auto& pipeline =
-	    m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
-	auto& bindings = m_compute_bindings;
-	PrepareBindings(input_info.stage, bindings);
+	    m_context.GetPipelineCache().CreateComputePipeline(input_info, compute_program);
+	auto bindings = PrepareBindings(input_info.stage);
 	FindBuffers(bindings);
 	if (program.info.uses_dma) {
-		m_context.PrepareBda();
+		m_context.GetGpuResources().PrepareBda();
 	}
-	RebindImages(bindings);
 	RebindBuffers(bindings);
+	RebindImages(bindings);
 
 	auto              vk_buffer        = buffer.Handle();
 	PreparedBindings* descriptor_stage = &bindings;
@@ -401,65 +409,4 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	ResetBindings();
 }
 
-void RenderExecutor::DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer,
-                                      uint64_t args_addr, uint32_t mode) {
-	EXIT_IF(buffer.IsInvalid() || args_addr == 0 || (args_addr & 3u) != 0 ||
-	        (mode & Pm4::COMPUTE_DISPATCH_INITIATOR_USE_THREAD_DIMENSIONS) != 0);
-	m_context.GetCommandScheduler().PopPendingOperations();
-	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::DispatchIndirect), submit_id,
-	                    static_cast<uint32_t>(args_addr), static_cast<uint32_t>(args_addr >> 32u),
-	                    0, mode, buffer.GetShaders().GetCs().cs_regs.data_addr);
-	Common::LockGuard lock(m_context.GetMutex());
-	const auto& cs_regs = buffer.GetShaders().GetCs();
-	if (cs_regs.cs_regs.data_addr == 0) {
-		return;
-	}
-	ShaderComputeInputInfo input_info {};
-	const auto compute_program = m_context.GetPipelineCache().GetComputeProgram(
-	    cs_regs, buffer.GetRegisters().GetShaderRegisters(), input_info);
-	if (!compute_program) {
-		// Temporary until RT is implemented.
-		return;
-	}
-	buffer.EndRendering();
-	auto& pipeline = m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
-	auto& bindings = m_compute_bindings;
-	PrepareBindings(input_info.stage, bindings);
-	FindBuffers(bindings);
-	const auto& program = *input_info.stage.program;
-	if (program.info.uses_dma) {
-		m_context.PrepareBda();
-	}
-	RebindImages(bindings);
-	// Acquiring arguments can merge cache buffers; finalize shader bindings afterward.
-	const auto [args_buffer, args_offset] = m_context.GetBufferCache().ObtainBuffer(
-	    args_addr, sizeof(vk::DispatchIndirectCommand), false);
-	EXIT_IF(args_buffer == nullptr || (args_offset & 3u) != 0);
-	RebindBuffers(bindings);
-	PreparedBindings* descriptor_stage = &bindings;
-	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline,
-	               std::span {&descriptor_stage, 1u});
-	const auto vk_buffer = buffer.Handle();
-	const bool has_storage_writes = HasShaderBufferWrites(input_info.stage) ||
-	    std::any_of(program.info.images.begin(), program.info.images.end(), [](const auto& image) {
-		    return image.written && image.resource_class ==
-		                                ShaderRecompiler::IR::ImageResourceClass::Storage;
-	    });
-	if (has_storage_writes) {
-		ShaderWriteHazardBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
-	}
-	vk::MemoryBarrier barrier {};
-	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite | vk::AccessFlagBits::eTransferWrite;
-	barrier.dstAccessMask = vk::AccessFlagBits::eIndirectCommandRead;
-	vk_buffer.pipelineBarrier(vk::PipelineStageFlagBits::eAllGraphics |
-	                              vk::PipelineStageFlagBits::eComputeShader |
-	                              vk::PipelineStageFlagBits::eTransfer,
-	                          vk::PipelineStageFlagBits::eDrawIndirect, {},
-	                          1, &barrier, 0, nullptr, 0, nullptr);
-	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
-	vk_buffer.dispatchIndirect(args_buffer->Handle(), args_offset);
-	ShaderAccessBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
-	ResetBindings();
-}
-
 } // namespace Libs::Graphics

src/graphics/host_gpu/renderer/renderContext.cpp

@@ -14,8 +14,7 @@ RenderContext::RenderContext(GraphicContext& graphics)
     : m_graphics(graphics), m_render_executor(*this), m_command_scheduler(*this, graphics),
       m_descriptor_heap(graphics, m_command_scheduler.GetMasterSemaphore()),
       m_pipeline_cache(graphics), m_sampler_cache(graphics),
-      m_buffer_cache(graphics, m_command_scheduler, m_page_manager, m_texture_cache),
-      m_texture_cache(graphics, m_command_scheduler, m_page_manager, m_buffer_cache) {
+      m_gpu_resources(graphics, m_command_scheduler) {
 	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
 }
 
@@ -28,10 +27,12 @@ void RenderContext::InitializeGpu(VideoOut::VideoOutDriver* video_out) {
 	EXIT_IF(m_gpu != nullptr);
 	m_video_out = video_out;
 	m_gpu       = std::make_unique<GuestGpu>(*this);
+	m_gpu_resources.SetGpu(m_gpu.get());
 }
 
 void RenderContext::ShutdownGpu() {
 	if (m_gpu != nullptr) {
+		m_gpu_resources.SetGpu(nullptr);
 		m_gpu->Shutdown();
 		m_gpu.reset();
 	}
@@ -54,88 +55,6 @@ VideoOut::VideoOutDriver& RenderContext::GetVideoOut() const {
 	return *m_video_out;
 }
 
-bool RenderContext::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
-	// The host reports the faulting byte, not the instruction's access width. Both caches
-	// resolve its page; guessing a width can cross the end of a valid guest mapping.
-	constexpr uint64_t fault_size = 1;
-	if (!IsMapped(fault_vaddr, fault_size)) {
-		return false;
-	}
-	if (access == PageFaultAccess::Write) {
-		m_buffer_cache.InvalidateMemory(fault_vaddr, fault_size);
-		m_texture_cache.InvalidateMemory(fault_vaddr, fault_size);
-	} else {
-		m_buffer_cache.ReadMemory(fault_vaddr, fault_size);
-	}
-	return true;
-}
-
-bool RenderContext::InvalidateMemory(uint64_t vaddr, uint64_t size) {
-	if (!IsMapped(vaddr, size)) {
-		return false;
-	}
-	m_buffer_cache.InvalidateMemory(vaddr, size);
-	m_texture_cache.InvalidateMemory(vaddr, size);
-	return true;
-}
-
-bool RenderContext::IsMapped(uint64_t vaddr, uint64_t size) const noexcept {
-	if (!GuestRange {vaddr, size}.Valid()) {
-		return false;
-	}
-	std::shared_lock lock(m_mapped_ranges_mutex);
-	return m_mapped_ranges.Contains(vaddr, size);
-}
-
-void RenderContext::MapMemory(uint64_t vaddr, uint64_t size) {
-	std::lock_guard lock(m_mapped_ranges_mutex);
-	m_mapped_ranges.Add(vaddr, size);
-}
-
-void RenderContext::UnmapMemory(uint64_t vaddr, uint64_t size) {
-	if (CommandScheduler::InDeferredOperation()) {
-		EXIT("unsupported memory unmap from an asynchronous GPU completion, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
-	const auto unmap = [this, vaddr, size] {
-		if (m_command_scheduler.Active()) {
-			const auto tick = m_command_scheduler.CurrentTick();
-			m_command_scheduler.Finish();
-			m_command_scheduler.WaitPriorityOperations(tick);
-		}
-		m_buffer_cache.InvalidateMemory(vaddr, size);
-		m_texture_cache.UnmapMemory(vaddr, size);
-		std::lock_guard lock(m_mapped_ranges_mutex);
-		m_mapped_ranges.Subtract(vaddr, size);
-	};
-	// Shutdown still owns the GPU while queued rendering drains, but its command lane no
-	// longer accepts external work. Use the guest GPU's state for the teardown route.
-	if (m_gpu == nullptr || m_gpu->IsStopping()) {
-		unmap();
-		return;
-	}
-	m_gpu->SendCommandSync(unmap);
-}
-
-void RenderContext::PrepareBda() {
-	std::shared_lock lock(m_mapped_ranges_mutex);
-	m_mapped_ranges.ForEach([this](uint64_t start, uint64_t end) {
-		m_buffer_cache.SynchronizeBuffersInRange(start, end - start);
-	});
-	m_fault_process_pending = true;
-}
-
-void RenderContext::RunGarbageCollector() {
-	if (m_fault_process_pending) {
-		m_fault_process_pending = false;
-		m_buffer_cache.ProcessFaultBuffer();
-	}
-	m_texture_cache.ProcessDownloadImages();
-	m_texture_cache.RunGarbageCollector();
-	m_buffer_cache.RunGarbageCollector();
-}
-
 void RenderContext::AddInterruptEq(LibKernel::EventQueue::KernelEqueue eq, int event_id) {
 	Common::LockGuard lock(m_interrupt_mutex);
 

src/graphics/host_gpu/renderer/renderContext.h

@@ -5,9 +5,8 @@
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/threads.h"
-#include "graphics/host_gpu/pageManager.h"
-#include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
+#include "graphics/host_gpu/renderer/cache/gpuResourceManager.h"
 #include "graphics/host_gpu/renderer/cache/samplerCache.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
@@ -16,7 +15,6 @@
 #include "kernel/eventQueue.h"
 
 #include <memory>
-#include <shared_mutex>
 #include <vector>
 
 namespace Libs::VideoOut {
@@ -44,18 +42,11 @@ public:
 	PipelineCache&      GetPipelineCache() { return m_pipeline_cache; }
 	DescriptorHeap&     GetDescriptorHeap() { return m_descriptor_heap; }
 	SamplerCache&       GetSamplerCache() { return m_sampler_cache; }
-	BufferCache&        GetBufferCache() { return m_buffer_cache; }
-	TextureCache&       GetTextureCache() { return m_texture_cache; }
+	GpuResourceManager& GetGpuResources() { return m_gpu_resources; }
+	BufferCache&        GetBufferCache() { return m_gpu_resources.GetBufferCache(); }
+	TextureCache&       GetTextureCache() { return m_gpu_resources.GetTextureCache(); }
 	RenderExecutor&     GetRenderExecutor() { return m_render_executor; }
 
-	[[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);
-	void               UnmapMemory(uint64_t vaddr, uint64_t size);
-	void               PrepareBda();
-	void               RunGarbageCollector();
-
 	void AddInterruptEq(LibKernel::EventQueue::KernelEqueue eq, int event_id);
 	void DeleteInterruptEq(LibKernel::EventQueue::KernelEqueue eq, int event_id);
 	void TriggerInterrupt(int event_id, uint32_t context_id);
@@ -73,14 +64,9 @@ private:
 	DescriptorHeap            m_descriptor_heap;
 	PipelineCache             m_pipeline_cache;
 	SamplerCache              m_sampler_cache;
-	PageManager               m_page_manager;
-	BufferCache               m_buffer_cache;
-	TextureCache              m_texture_cache;
-	mutable std::shared_mutex m_mapped_ranges_mutex;
-	RangeSet                  m_mapped_ranges;
+	GpuResourceManager        m_gpu_resources;
 	std::unique_ptr<GuestGpu> m_gpu;
 	VideoOut::VideoOutDriver* m_video_out = nullptr;
-	bool                      m_fault_process_pending = false;
 
 	Common::Mutex                        m_interrupt_mutex;
 	std::vector<InterruptEqRegistration> m_interrupt_eqs;

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"
@@ -16,7 +15,6 @@
 #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/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
@@ -48,34 +46,42 @@
 
 namespace Libs::Graphics {
 
-std::pair<int32_t, uint32_t> ResolveDrawOffsets(uint32_t index_offset,
-	                                           const ShaderVertexInputInfo& vs_input_info) {
-	auto     vertex_offset   = static_cast<int32_t>(index_offset);
-	uint32_t instance_offset = 0;
-	if (!vs_input_info.fetch_embedded) {
-		return {vertex_offset, instance_offset};
+int32_t ResolveVertexOffset(uint32_t index_offset, const ShaderVertexInputInfo& vs_input_info) {
+	if (index_offset != 0 || !vs_input_info.fetch_embedded) {
+		return static_cast<int32_t>(index_offset);
 	}
 
 	EXIT_IF(!vs_input_info.stage);
 	const auto& program   = *vs_input_info.stage.program;
-	const auto& resources = *vs_input_info.stage.resources;
-	if (index_offset == 0 &&
-	    program.info.vertex_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
+	const auto& resources = vs_input_info.stage.resources;
+	if (program.info.vertex_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
 		const auto index =
 		    static_cast<uint32_t>(program.info.vertex_offset_sgpr) - program.user_data_base;
 		if (index < resources.user_data.size()) {
-			vertex_offset = static_cast<int32_t>(resources.user_data[index]);
+			return static_cast<int32_t>(resources.user_data[index]);
 		}
 	}
+
+	return 0;
+}
+
+uint32_t ResolveInstanceOffset(const ShaderVertexInputInfo& vs_input_info) {
+	if (!vs_input_info.fetch_embedded) {
+		return 0;
+	}
+
+	EXIT_IF(!vs_input_info.stage);
+	const auto& program   = *vs_input_info.stage.program;
+	const auto& resources = vs_input_info.stage.resources;
 	if (program.info.instance_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
 		const auto index =
 		    static_cast<uint32_t>(program.info.instance_offset_sgpr) - program.user_data_base;
 		if (index < resources.user_data.size()) {
-			instance_offset = resources.user_data[index];
+			return resources.user_data[index];
 		}
 	}
 
-	return {vertex_offset, instance_offset};
+	return 0;
 }
 
 static std::atomic<uint32_t> g_draw_state_log_count   = 0;
@@ -310,12 +316,24 @@ static void LogDrawInputState(const CommandBuffer& buffer, const RenderColorInfo
 
 static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuffer vk_buffer,
                                      const ShaderVertexInputInfo& vs_input_info,
-                                     const RenderDepthInfo& depth, const RenderState& rendering) {
+                                     const RenderColorInfo* colors, uint32_t color_count,
+                                     const RenderDepthInfo& depth) {
 	KYTY_PROFILER_FUNCTION();
 
+	EXIT_IF(colors == nullptr);
 	const auto& ctx = buffer.GetRegisters();
-	const auto&        vp  = ctx.GetScreenViewport();
-	const vk::Extent2D framebuffer_extent {rendering.width, rendering.height};
+
+	const auto&  vp = ctx.GetScreenViewport();
+	vk::Extent2D framebuffer_extent {};
+	if (color_count > 0 && colors[0].image_id) {
+		framebuffer_extent = colors[0].Extent();
+	} else if (depth.image_id) {
+		framebuffer_extent = {depth.desc.info.extent.width, depth.desc.info.extent.height};
+	} else {
+		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
+		framebuffer_extent = {limits.maxFramebufferWidth, limits.maxFramebufferHeight};
+	}
+
 	const auto& outputs = vs_input_info.stage.program->info.outputs;
 	const bool  indexed_viewports =
 	    std::any_of(outputs.begin(), outputs.end(), [](const auto& output) {
@@ -393,16 +411,19 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 		vk_buffer.setDepthBias(constant_factor, poly_offset.clamp, slope_factor);
 	}
 
-	vk_buffer.setStencilTestEnable(depth.stencil_test_enable ? VK_TRUE : VK_FALSE);
 	if (depth.stencil_test_enable) {
-		const auto set_stencil = [&](vk::StencilFaceFlagBits face, const vk::StencilOpState& state) {
-			vk_buffer.setStencilOp(face, state.failOp, state.passOp, state.depthFailOp, state.compareOp);
-			vk_buffer.setStencilCompareMask(face, state.compareMask);
-			vk_buffer.setStencilWriteMask(face, state.writeMask);
-			vk_buffer.setStencilReference(face, state.reference);
-		};
-		set_stencil(vk::StencilFaceFlagBits::eFront, depth.stencil_front);
-		set_stencil(vk::StencilFaceFlagBits::eBack, depth.stencil_back);
+		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
+		                                depth.stencil_dynamic_front.compareMask);
+		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eBack,
+		                                depth.stencil_dynamic_back.compareMask);
+		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eFront,
+		                              depth.stencil_dynamic_front.writeMask);
+		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eBack,
+		                              depth.stencil_dynamic_back.writeMask);
+		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eFront,
+		                              depth.stencil_dynamic_front.reference);
+		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eBack,
+		                              depth.stencil_dynamic_back.reference);
 	}
 
 #if defined(__APPLE__)
@@ -410,11 +431,15 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 	// eColorWriteEnableEXT dynamic state and relies on the static colorWriteMask instead.
 #else
 	vk::Bool32 enable[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	for (uint32_t slot = 0; slot < rendering.num_color_attachments; slot++) {
-		enable[slot] = rendering.color_attachments[slot].image_view != nullptr;
+	// Color-control operation selects special color-buffer paths, not the normal component write
+	// mask. Attachment availability therefore follows the target write mask.
+	for (uint32_t i = 0; i < color_count; i++) {
+		enable[i] = render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot) != 0
+		                ? VK_TRUE
+		                : VK_FALSE;
 	}
-	if (rendering.num_color_attachments != 0) {
-		vk_buffer.setColorWriteEnableEXT(rendering.num_color_attachments, enable);
+	if (color_count != 0) {
+		vk_buffer.setColorWriteEnableEXT(color_count, enable);
 	}
 #endif
 }
@@ -422,7 +447,7 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 static bool DrawHasValidVertexShader(const HW::Shader& sh_ctx) {
 
 	const auto& vs = sh_ctx.GetVs();
-	return vs.es_regs.data_addr != 0;
+	return ShaderAddressValid(vs.es_regs.data_addr);
 }
 
 static bool PixelShaderHasDepthOrCoverageSideEffects(const HW::ShaderRegisters& sh_regs) {
@@ -436,44 +461,60 @@ struct DrawRenderState {
 	RenderColorInfo       color_info[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	uint32_t              color_count                              = 0;
 	bool                  ps_active                                = true;
-	std::array<ShaderVertexInputInfo, 3> vertex_info;
+	ShaderVertexInputInfo vs_input_info;
 	ShaderPixelInputInfo  ps_input_info;
 	PipelineCache::GraphicsPrograms programs;
 };
 
 struct DrawCallInfo {
+	const char*          name           = nullptr;
 	CommandBufferDebugOp debug_op       = CommandBufferDebugOp::DrawIndex;
 	uint32_t             index_count    = 0;
 	uint32_t             instance_count = 0;
 	uint32_t             first_instance = 0;
-
-	[[nodiscard]] bool IsIndexed() const { return debug_op == CommandBufferDebugOp::DrawIndex; }
-	[[nodiscard]] const char* Name() const { return IsIndexed() ? "DrawIndex" : "DrawIndexAuto"; }
 };
 
 RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
                                                  uint32_t color_count, RenderDepthInfo& depth,
-                                                 vk::ImageAspectFlags& feedback_aspects,
-                                                 std::span<PreparedBindings* const> stages) {
+                                                 const std::optional<PreparedBindings>& pixel) {
 	EXIT_IF(colors == nullptr || color_count > RENDER_COLOR_ATTACHMENTS_MAX);
-	feedback_aspects = {};
 	auto&       cache = m_context.GetTextureCache();
 	RenderState state {};
 	state.width                 = std::numeric_limits<uint32_t>::max();
 	state.height                = std::numeric_limits<uint32_t>::max();
 	state.num_layers            = std::numeric_limits<uint32_t>::max();
-	state.num_color_attachments = 0;
+	state.num_color_attachments = color_count;
+	uint32_t attachment_samples = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
-		const auto owner = cache.m_slot_images.try_get(target.image_id);
-		if (owner == nullptr || (!owner->registered && !owner->info.data.Empty()) ||
-		    owner->binding.needs_rebind) {
-			EXIT("color target changed after render-state discovery\n");
+		const auto old_image = cache.m_slot_images.try_get(target.image_id);
+		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
+		    old_image->binding.needs_rebind) {
+			if (old_image != nullptr) {
+				old_image->binding = {};
+			}
+			target.image_id = cache.FindImage(target.desc);
+			BindRenderTarget(target.image_id);
 		}
 		const auto image_view = cache.FindRenderTarget(target.image_id, target.desc);
 		auto&      image      = cache.GetImage(target.image_id);
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image.backing.image,
+		                     "Kyty.MRT{}.Image[guest=0x{:016x} size=0x{:x} format={}]",
+		                     target.target_slot, image.info.data.address, image.info.data.size,
+		                     static_cast<uint32_t>(image.info.pixel_format));
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image_view,
+		                     "Kyty.MRT{}.View[guest=0x{:016x} mip={} layer={}+{}]",
+		                     target.target_slot, image.info.data.address,
+		                     target.desc.view_info.base_level, target.desc.view_info.base_layer,
+		                     target.desc.view_info.layer_count);
 		EXIT_IF(image.backing.samples != target.desc.info.samples || image_view == nullptr);
+		if (attachment_samples == 0) {
+			attachment_samples = target.desc.info.samples;
+		} else if (attachment_samples != target.desc.info.samples) {
+			EXIT("mixed color attachment sample counts are unsupported: %u and %u\n",
+			     attachment_samples, target.desc.info.samples);
+		}
 		const auto& view   = target.desc.view_info;
 		const auto  layout = image.binding.is_bound ? vk::ImageLayout::eGeneral
 		                                            : vk::ImageLayout::eColorAttachmentOptimal;
@@ -488,8 +529,7 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		state.width             = std::min(state.width, extent.width);
 		state.height            = std::min(state.height, extent.height);
 		state.num_layers        = std::min(state.num_layers, view.layer_count);
-		state.num_color_attachments = std::max(state.num_color_attachments, target.target_slot + 1);
-		auto& attachment            = state.color_attachments[target.target_slot];
+		auto& attachment        = state.color_attachments[i];
 		attachment.image_view   = image_view;
 		attachment.image_layout = layout;
 	}
@@ -513,30 +553,43 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 			EXIT("failed to consume HTile clear state\n");
 		}
 		auto& image = cache.GetImage(depth.image_id);
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image.backing.image,
+		                     "Kyty.DepthTarget.Image[guest=0x{:016x} size=0x{:x} format={}]",
+		                     image.info.data.address, image.info.data.size,
+		                     static_cast<uint32_t>(image.info.pixel_format));
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image_view,
+		                     "Kyty.DepthTarget.View[guest=0x{:016x} layer={}+{}]",
+		                     image.info.data.address, depth.desc.view_info.base_layer,
+		                     depth.desc.view_info.layer_count);
 		EXIT_IF(image_view == nullptr || image.backing.samples != depth.desc.info.samples);
-		const auto draw_writes = depth.AttachmentWriteAspects();
-		vk::ImageAspectFlags sampled_aspects;
-		for (const auto* stage: stages) {
-			for (const auto& binding: stage->images) {
-				if (binding.image_id != depth.image_id ||
-				    binding.desc.type != TextureCache::BindingType::Texture) continue;
-				const auto native =
-				    std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
-				EXIT_IF(native == image.views.end());
-				sampled_aspects |= native->info.aspect;
-				feedback_aspects |= DepthFeedbackAspects(draw_writes, depth.desc.view_info,
-				                                         native->info);
-			}
+		if (attachment_samples == 0) {
+			attachment_samples = depth.desc.info.samples;
+		} else if (attachment_samples != depth.desc.info.samples) {
+			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n", attachment_samples,
+			     depth.desc.info.samples);
 		}
-		if (feedback_aspects && !m_context.GetGraphics().attachment_feedback_loop_enabled) {
+		const bool feedback = depth.depth_write_enable && pixel &&
+		    std::ranges::any_of(pixel->images, [&](const TextureBinding& binding) {
+			    if (binding.image_id != depth.image_id ||
+			        binding.desc.type != TextureCache::BindingType::Texture) {
+				    return false;
+			    }
+			    const auto native =
+			        std::ranges::find(image.views, binding.image_view, &CachedImageView::view);
+			    EXIT_IF(native == image.views.end());
+			    const auto& sampled = native->info;
+			    const auto& target = depth.desc.view_info;
+			    return (sampled.aspect & vk::ImageAspectFlagBits::eDepth) &&
+			           ImageRangeOverlaps(sampled.base_level, sampled.level_count,
+			                              target.base_level, target.level_count) &&
+			           ImageRangeOverlaps(sampled.base_layer, sampled.layer_count,
+			                              target.base_layer, target.layer_count);
+		    });
+		if (feedback && !m_context.GetGraphics().attachment_feedback_loop_enabled) {
 			EXIT("depth attachment feedback loop is not supported by the host\n");
 		}
-		auto layout = depth_attachment_layout(depth);
-		if (sampled_aspects & ~DepthReadableAspects(layout)) {
-			layout = m_context.GetGraphics().attachment_feedback_loop_enabled
-			             ? vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
-			             : vk::ImageLayout::eGeneral;
-		}
+		const auto layout = feedback ? vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
+		                             : depth_attachment_layout(depth);
 		// The attachment store writes even when guest depth/stencil tests do not.
 		const auto access = vk::AccessFlagBits2::eDepthStencilAttachmentRead |
 		                    vk::AccessFlagBits2::eDepthStencilAttachmentWrite;
@@ -565,6 +618,9 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
 		state.width        = limits.maxFramebufferWidth;
 		state.height       = limits.maxFramebufferHeight;
+	} else if (attachment_samples == 0 ||
+	           vulkan_sample_count(attachment_samples) == vk::SampleCountFlagBits {}) {
+		EXIT("render state has no valid attachments\n");
 	}
 	if (state.num_layers == std::numeric_limits<uint32_t>::max()) {
 		state.num_layers = 1;
@@ -579,7 +635,7 @@ static bool DrawHasActivePixelShader(const CommandBuffer& buffer) {
 	const auto& ctx              = buffer.GetRegisters();
 	const auto& sh_regs          = ctx.GetShaderRegisters();
 	const bool  has_color_output = (ctx.GetRenderTargetMask() & sh_regs.m_cbShaderMask) != 0;
-	return buffer.GetShaders().GetPs().ps_regs.data_addr != 0 &&
+	return ShaderAddressValid(buffer.GetShaders().GetPs().ps_regs.data_addr) &&
 	       (has_color_output || PixelShaderHasDepthOrCoverageSideEffects(sh_regs));
 }
 
@@ -606,6 +662,7 @@ static bool ConsumeMetadataColorOperation(const CommandBuffer& buffer) {
 }
 
 struct DrawEmitInfo {
+	bool     indexed       = false;
 	int32_t  vertex_offset = 0;
 	uint32_t first_vertex  = 0;
 	uint32_t first_instance = 0;
@@ -659,7 +716,6 @@ struct PreparedVertexBuffers {
 
 	std::array<vk::Buffer, MaxBuffers>     buffers {};
 	std::array<vk::DeviceSize, MaxBuffers> offsets {};
-	std::array<vk::DeviceSize, MaxBuffers> sizes {};
 	uint32_t                               count = 0;
 };
 
@@ -669,13 +725,11 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 	        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX);
 
 	// Collect the non-empty guest vertex ranges.
-	std::array<uint64_t, ShaderVertexInputInfo::RES_MAX>          sizes {};
 	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> ranges {};
 	uint32_t                                                      range_count = 0;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex, vs_input_info);
-		sizes[i]           = size;
 		if (size == 0) {
 			continue;
 		}
@@ -724,7 +778,7 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 	vk::Buffer null_buffer = nullptr;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
-		const auto  size   = sizes[i];
+		const auto  size   = VertexBufferDescriptorSize(vertex, vs_input_info);
 		if (size == 0) {
 			if (null_buffer == nullptr) {
 				null_buffer = cache.GetBuffer(NULL_BUFFER_ID).Handle();
@@ -746,7 +800,6 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 
 		prepared.buffers[i] = range->binding.first->Handle();
 		prepared.offsets[i] = range->binding.second + vertex.addr - range->base_address;
-		prepared.sizes[i]   = std::min(size, range->acquired_end - vertex.addr);
 		SetVulkanObjectNameF(
 		    buffer.GetContext().GetGraphics().device, prepared.buffers[i],
 		    "Kyty.VertexBuffer[slot={} guest=0x{:016x} size=0x{:x} stride={} records={}]", i,
@@ -762,7 +815,8 @@ static void SetDrawDebugPhase(CommandBuffer& buffer, uint64_t submit_id, const D
 	                    draw.instance_count, draw.first_instance);
 }
 
-static bool GetDrawTopology(const HW::UserConfig& ucfg, vk::PrimitiveTopology& topology) {
+static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
+                            vk::PrimitiveTopology& topology) {
 
 	topology = vk::PrimitiveTopology::ePointList;
 
@@ -784,13 +838,15 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, vk::PrimitiveTopology& t
 		case Prospero::PrimitiveType::kTriStrip:
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
-		case Prospero::PrimitiveType::kPatch:
-			if (!Config::TessellationEnabled()) return false;
-			[[fallthrough]];
 		case Prospero::PrimitiveType::kRectList:
-		case Prospero::PrimitiveType::kRectListLegacy:
 			topology = vk::PrimitiveTopology::ePatchList;
 			break;
+		case Prospero::PrimitiveType::kRectListLegacy:
+			if (!auto_draw) {
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
+			}
+			topology = vk::PrimitiveTopology::eTriangleStrip;
+			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
 			break;
@@ -846,7 +902,43 @@ static bool ResolvePrimitiveRestart(const CommandBuffer& buffer,
 	return false;
 }
 
-static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
+bool RenderExecutor::PrepareDrawRenderState(CommandBuffer& buffer, const DrawCallInfo& draw,
+                                            uint32_t            render_target_slice_offset,
+                                            bool log_setup_phases, DrawRenderState& state) {
+	auto& ctx = buffer.GetRegisters();
+
+	if (ResolveColorTargets(buffer, render_target_slice_offset)) {
+		return false;
+	}
+	if (log_setup_phases) {
+		LogDrawPhase(draw.name, "ResolveRenderColorTarget");
+	}
+	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
+		if (slot == 0 || (render_target_mask_slot(ctx.GetRenderTargetMask(), slot) != 0 &&
+		                  ctx.GetRenderTarget(slot).base.addr != 0)) {
+			ResolveRenderColorTarget(buffer, state.color_info[state.color_count],
+			                         render_target_slice_offset, slot);
+			if (state.color_info[state.color_count].image_id) {
+				state.color_count++;
+			}
+		}
+	}
+	if (log_setup_phases) {
+		LogDrawPhase(draw.name, "ResolveRenderDepthTarget");
+	}
+	ResolveRenderDepthTarget(buffer, state.depth_info);
+
+	state.ps_active       = DrawHasActivePixelShader(buffer);
+	if (state.color_count == 0 && !state.depth_info.image_id && !state.ps_active) {
+		LogFramebufferSkip(draw.name, state.color_info[0], state.depth_info, buffer,
+		                   draw.index_count, 0);
+		return false;
+	}
+
+	return true;
+}
+
+static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw, bool log_phases,
                            DrawRenderState& state) {
 	auto& ctx    = buffer.GetRegisters();
 	auto& sh_ctx = buffer.GetShaders();
@@ -859,61 +951,16 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
 	state.ps_input_info = {};
 	std::array<Prospero::ColorComponentMapping, RENDER_COLOR_ATTACHMENTS_MAX>
 	    target_export_mapping {};
-	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
-		const auto& rt = ctx.GetRenderTarget(slot);
-		if (rt.base.addr != 0 && render_target_mask_slot(ctx.GetRenderTargetMask(), slot) != 0) {
-			target_export_mapping[slot] =
-			    TextureGetRenderTargetFormat(rt.info.format, rt.info.channel_type,
-			                                 rt.info.channel_order)
-			        .export_mapping;
-		}
+	for (uint32_t i = 0; i < state.color_count; i++) {
+		target_export_mapping[state.color_info[i].target_slot] = state.color_info[i].export_mapping;
 	}
 	auto& pipeline_cache = buffer.GetContext().GetPipelineCache();
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "GetGraphicsPrograms");
+	if (log_phases) {
+		LogDrawPhase(draw.name, "GetGraphicsPrograms");
 	}
 	state.programs = pipeline_cache.GetGraphicsPrograms(
 	    vertex_shader_info, pixel_shader_info, shader_regs, ctx, buffer.GetUserConfig(),
-	    target_export_mapping, state.ps_active, state.vertex_info, state.ps_input_info);
-}
-
-bool RenderExecutor::PrepareDrawRenderState(CommandBuffer& buffer, const DrawCallInfo& draw,
-                                            uint32_t            render_target_slice_offset,
-	                                        DrawRenderState& state) {
-	state.ps_active = DrawHasActivePixelShader(buffer);
-	RefreshShaders(buffer, draw, state);
-	uint32_t mrt_mask = 0;
-	if (state.ps_active) {
-		for (const auto& output: state.ps_input_info.stage.program->info.outputs) {
-			if (output.kind == ShaderRecompiler::IR::StageOutputKind::Mrt) {
-				mrt_mask |= 1u << output.index;
-			}
-		}
-	}
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "ResolveRenderColorTarget");
-	}
-	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
-		if ((mrt_mask & (1u << slot)) != 0) {
-			ResolveRenderColorTarget(buffer, state.color_info[state.color_count],
-			                         render_target_slice_offset, slot);
-			if (state.color_info[state.color_count].image_id) {
-				state.color_count++;
-			}
-		}
-	}
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "ResolveRenderDepthTarget");
-	}
-	ResolveRenderDepthTarget(buffer, state.depth_info);
-
-	if (state.color_count == 0 && !state.depth_info.image_id && !state.ps_active) {
-		LogFramebufferSkip(draw.Name(), state.color_info[0], state.depth_info, buffer,
-		                   draw.index_count, 0);
-		return false;
-	}
-
-	return true;
+	    target_export_mapping, state.ps_active, state.vs_input_info, state.ps_input_info);
 }
 
 static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffer,
@@ -927,14 +974,17 @@ static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffe
 		auto& stream = buffer.GetContext().GetBufferCache().GetUtilityBuffer(MemoryUsage::Stream);
 		prepared.offset = stream.Copy(source.host_data, source.size, 16);
 		prepared.buffer = stream.Handle();
-		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
-		                     "Kyty.IndexBuffer[guest=transient size=0x{:x} type={}]", source.size,
-		                     static_cast<uint32_t>(source.type));
 	} else {
 		auto [buffer_ptr, offset] =
 		    buffer.GetContext().GetBufferCache().ObtainBuffer(source.address, source.size, false);
 		prepared.buffer = buffer_ptr->Handle();
 		prepared.offset = offset;
+	}
+	if (source.host_data != nullptr) {
+		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
+		                     "Kyty.IndexBuffer[guest=transient size=0x{:x} type={}]", source.size,
+		                     static_cast<uint32_t>(source.type));
+	} else {
 		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
 		                     "Kyty.IndexBuffer[guest=0x{:016x} size=0x{:x} type={}]",
 		                     source.address, source.size, static_cast<uint32_t>(source.type));
@@ -948,9 +998,8 @@ static void CommitVertexBuffers(vk::CommandBuffer            vk_buffer,
 		EXIT_IF(prepared.buffers[i] == nullptr);
 	}
 	if (prepared.count != 0) {
-		// Guest descriptor bounds must survive allocation merging in the cache.
-		vk_buffer.bindVertexBuffers2(0, prepared.count, prepared.buffers.data(),
-		                             prepared.offsets.data(), prepared.sizes.data(), nullptr);
+		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
+		                            prepared.offsets.data());
 	}
 }
 
@@ -962,26 +1011,28 @@ static void CommitIndexBuffer(vk::CommandBuffer vk_buffer, const PreparedIndexBu
 }
 
 static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo& draw,
-	                             const DrawRenderState& state, uint32_t index_type_and_size,
+                                 const DrawRenderState& state, bool always_log,
+                                 bool force_legacy_rect_log, uint32_t index_type_and_size,
                                  const void* index_addr) {
 	if (!graphics_debug_dump_enabled()) {
 		return;
 	}
 
-	if (!draw.IsIndexed() && !Prospero::IsRectList(buffer.GetUserConfig().GetPrimType())) {
+	if (!always_log && !force_legacy_rect_log) {
 		return;
 	}
 
 	if (state.ps_active) {
-		LogDrawTargetState(draw.Name(), state.color_info[0], state.depth_info, buffer,
+		LogDrawTargetState(draw.name, state.color_info[0], state.depth_info, buffer,
 		                   state.ps_input_info, draw.index_count, 0);
 	}
-	LogDrawInputState(buffer, state.color_info[0], state.vertex_info[0], index_type_and_size,
+	LogDrawInputState(buffer, state.color_info[0], state.vs_input_info, index_type_and_size,
 	                  draw.index_count, index_addr);
 }
 
 static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_buffer,
-                               const DrawCallInfo& draw, const DrawEmitInfo& emit) {
+                               const ShaderVertexInputInfo& vs_input_info, const DrawCallInfo& draw,
+                               const DrawEmitInfo& emit) {
 	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kPointList:
 		case Prospero::PrimitiveType::kLineList:
@@ -990,9 +1041,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
 		case Prospero::PrimitiveType::kRectList:
-		case Prospero::PrimitiveType::kRectListLegacy:
-		case Prospero::PrimitiveType::kPatch:
-			if (draw.IsIndexed()) {
+			if (emit.indexed) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
 				                      emit.first_instance);
 			} else {
@@ -1000,10 +1049,18 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 				               emit.first_instance);
 			}
 			break;
+		case Prospero::PrimitiveType::kRectListLegacy:
+			if (emit.indexed) {
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
+			}
+			// Sarah
+			EXIT_NOT_IMPLEMENTED(!(draw.index_count == 3 && vs_input_info.buffers_num == 0));
+			vk_buffer.draw(4, draw.instance_count, emit.first_vertex, emit.first_instance);
+			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			EXIT_NOT_IMPLEMENTED((draw.index_count & 0x3u) != 0);
 			for (uint32_t i = 0; i < draw.index_count; i += 4) {
-				if (draw.IsIndexed()) {
+				if (emit.indexed) {
 					vk_buffer.drawIndexed(4, draw.instance_count, i, emit.vertex_offset,
 					                      emit.first_instance);
 				} else {
@@ -1020,23 +1077,16 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
                                          const DrawCallInfo& draw, DrawRenderState& state,
                                          vk::PrimitiveTopology topology, const DrawEmitInfo& emit,
                                          const DrawIndexBufferSource& index_source,
-	                                     bool primitive_restart_enable) {
+                                         bool primitive_restart_enable, bool log_pipeline_phase,
+                                         bool set_bind_debug, bool set_auto_debug) {
 	auto& ucfg = buffer.GetUserConfig();
-	const auto vertex_stages =
-	    std::span {state.vertex_info.data(), state.programs.VertexStageCount()};
-	const bool mesh_active = state.vertex_info[0].stage.program->stage == ShaderType::Mesh;
+	const bool mesh_active = state.vs_input_info.stage.program->stage == ShaderType::Mesh;
 	uint32_t   mesh_groups = 0;
 	if (mesh_active) {
-		const auto& mesh = state.vertex_info[0].mesh;
-		static std::atomic_bool restart_warned = false;
-		if (primitive_restart_enable && !restart_warned.exchange(true, std::memory_order_relaxed)) {
-			std::printf("Warning: primitive restart is not implemented for mesh shaders; "
-			            "continuing draw (primitive=%u indexed=%u)\n",
-			            static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed());
-		}
-		if (mesh.primitives_per_group == 0) {
+		const auto& mesh = state.vs_input_info.mesh;
+		if (primitive_restart_enable || mesh.primitives_per_group == 0) {
 			EXIT("unsupported mesh draw: primitive=%u indexed=%u restart=%u\n",
-			     static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed(), primitive_restart_enable);
+			     static_cast<uint32_t>(ucfg.GetPrimType()), emit.indexed, primitive_restart_enable);
 		}
 		const auto primitives = mesh.InputPrimitiveCount(draw.index_count);
 		if (primitives == 0 || draw.instance_count == 0) {
@@ -1053,63 +1103,64 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		}
 	}
 
-	if (mesh_active && draw.IsIndexed()) {
+	if (mesh_active && emit.indexed) {
 		// Register the original guest indices for shader reads; PrepareGraphicsBindings
 		// synchronizes registered BDA ranges before any draw commands are committed.
 		(void)m_context.GetBufferCache().FindBuffer(
 		    index_source.address, static_cast<uint64_t>(draw.index_count) *
 		                              index_source.guest_element_size);
 	}
-	LogDrawPhase(draw.Name(), "PrepareBindings");
-	auto&                            bindings = m_graphics_bindings;
-	std::array<PreparedBindings*, 4> descriptor_stages {};
-	uint32_t                         stage_count = 0;
-	for (uint32_t i = 0; i < vertex_stages.size(); i++) {
-		PrepareBindings(state.vertex_info[i].stage, bindings.vertex[i]);
-		descriptor_stages[stage_count++] = &bindings.vertex[i];
-	}
-	if (state.ps_active) {
-		if (!bindings.pixel) bindings.pixel.emplace();
-		PrepareBindings(state.ps_input_info.stage, *bindings.pixel);
-		descriptor_stages[stage_count++] = &*bindings.pixel;
-	}
-	const auto stages = std::span {descriptor_stages.data(), stage_count};
-	PrepareGraphicsBindings(stages, std::span {state.color_info, state.color_count});
+	LogDrawPhase(draw.name, "PrepareBindings");
+	auto bindings = PrepareGraphicsBindings(state.vs_input_info.stage, state.ps_input_info.stage,
+	                                        state.ps_active);
 	PreparedVertexBuffers vertex_bindings;
 	PreparedIndexBuffer   index_binding;
 	if (!mesh_active) {
-		LogDrawPhase(draw.Name(), "PrepareVertexBuffers");
-		vertex_bindings = AcquireVertexBuffers(buffer, state.vertex_info[0]);
+		LogDrawPhase(draw.name, "PrepareVertexBuffers");
+		vertex_bindings = AcquireVertexBuffers(buffer, state.vs_input_info);
 		index_binding   = PrepareIndexBuffer(buffer, index_source);
 	}
-	if (draw.IsIndexed()) {
-		LogDrawPhase(draw.Name(), "CreatePipeline");
-	}
-	auto& pipeline = m_context.GetPipelineCache().GetGraphicsPipeline(
-	    std::span {state.color_info, state.color_count}, state.depth_info, vertex_stages, buffer,
-	    state.ps_active ? &state.ps_input_info : nullptr, topology, primitive_restart_enable,
-	    state.programs);
-	vk::ImageAspectFlags feedback_aspects;
 	const auto rendering =
 	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info,
-	                         feedback_aspects, stages);
+	                         bindings.pixel);
+
+	if (log_pipeline_phase) {
+		LogDrawPhase(draw.name, "CreatePipeline");
+	}
+	auto& pipeline = m_context.GetPipelineCache().CreateGraphicsPipeline(
+	    std::span {state.color_info, state.color_count}, state.depth_info, state.vs_input_info, buffer,
+	    state.ps_active ? &state.ps_input_info : nullptr, topology, primitive_restart_enable,
+	    state.programs.vertex, state.programs.pixel);
 
 	// Resource preparation above may synchronously finish and restart the scheduler. From this
 	// point onward, every operation targets the current command buffer and cannot touch guest
 	// memory.
 	auto vk_buffer = buffer.Handle();
-	SetDrawDebugPhase(buffer, submit_id, draw, draw.IsIndexed() ? 0x100u : 0x200u);
+	if (set_bind_debug) {
+		SetDrawDebugPhase(buffer, submit_id, draw, 0x100u);
+	}
+	if (set_auto_debug) {
+		SetDrawDebugPhase(buffer, submit_id, draw, 0x200u);
+	}
 	if (!mesh_active) {
 		CommitVertexBuffers(vk_buffer, vertex_bindings);
 	}
-	if (state.ps_active && !draw.IsIndexed()) {
-		SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
+	if (bindings.pixel.has_value()) {
+		if (set_auto_debug) {
+			SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
+		}
 	}
-	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline, stages);
+	std::array<PreparedBindings*, 2> descriptor_stages {&bindings.vertex, nullptr};
+	const size_t                     descriptor_stage_count = bindings.pixel.has_value() ? 2u : 1u;
+	if (bindings.pixel) {
+		descriptor_stages[1] = &*bindings.pixel;
+	}
+	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline,
+	               std::span {descriptor_stages.data(), descriptor_stage_count});
 	if (mesh_active) {
 		const uint32_t draw_data[] {
 		    draw.index_count,
-		    draw.IsIndexed() ? static_cast<uint32_t>(emit.vertex_offset) : emit.first_vertex,
+		    emit.indexed ? static_cast<uint32_t>(emit.vertex_offset) : emit.first_vertex,
 		    emit.first_instance, index_source.guest_element_size,
 		    static_cast<uint32_t>(index_source.address),
 		    static_cast<uint32_t>(index_source.address >> 32u)};
@@ -1122,34 +1173,38 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		CommitIndexBuffer(vk_buffer, index_binding);
 	}
 
-	SetGraphicsDynamicParams(buffer, vk_buffer, vertex_stages.back(), state.depth_info, rendering);
+	SetGraphicsDynamicParams(buffer, vk_buffer, state.vs_input_info, state.color_info,
+	                         state.color_count, state.depth_info);
 	if (m_context.GetGraphics().attachment_feedback_loop_enabled) {
-		vk_buffer.setAttachmentFeedbackLoopEnableEXT(feedback_aspects);
+		vk_buffer.setAttachmentFeedbackLoopEnableEXT(
+		    rendering.depth_stencil_attachment.image_layout ==
+		            vk::ImageLayout::eAttachmentFeedbackLoopOptimalEXT
+		        ? vk::ImageAspectFlags {vk::ImageAspectFlagBits::eDepth}
+		        : vk::ImageAspectFlags {});
 	}
 
-	LogDrawPhase(draw.Name(), "BeginRendering");
-	if (!draw.IsIndexed()) {
+	LogDrawPhase(draw.name, "BeginRendering");
+	if (set_auto_debug) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x400u);
 	}
 	m_context.GetCommandScheduler().BeginRendering(rendering);
 	vk_buffer.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline.pipeline);
-	if (!draw.IsIndexed()) {
+	if (set_auto_debug) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x500u);
 	}
 	if (mesh_active) {
 		vk_buffer.drawMeshTasksEXT(mesh_groups, draw.instance_count, 1);
 	} else {
-		EmitDrawPrimitives(ucfg, vk_buffer, draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, state.vs_input_info, draw, emit);
 	}
 
-	if (!draw.IsIndexed()) {
+	if (set_auto_debug) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x600u);
 	}
 	vk::PipelineStageFlags shader_write_stages = {};
-	for (const auto& stage: vertex_stages) {
-		if (HasShaderBufferWrites(stage.stage)) {
-			shader_write_stages |= ShaderPipelineStages(NativeShaderStage(stage.logical_stage));
-		}
+	if (HasShaderBufferWrites(state.vs_input_info.stage)) {
+		shader_write_stages |= mesh_active ? vk::PipelineStageFlagBits::eMeshShaderEXT
+		                                   : vk::PipelineStageFlagBits::eVertexShader;
 	}
 	if (state.ps_active && HasShaderBufferWrites(state.ps_input_info.stage)) {
 		shader_write_stages |= vk::PipelineStageFlagBits::eFragmentShader;
@@ -1158,8 +1213,8 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		m_context.GetCommandScheduler().EndRendering();
 		ShaderWriteBarrier(vk_buffer, shader_write_stages);
 	}
-	LogDrawPhase(draw.Name(), "DrawComplete");
-	if (!draw.IsIndexed()) {
+	LogDrawPhase(draw.name, "DrawComplete");
+	if (set_auto_debug) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x700u);
 	}
 }
@@ -1183,8 +1238,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer) ||
-	    ResolveColorTargets(buffer, args.render_target_slice_offset)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
 		ResetBindings();
 		return;
 	}
@@ -1194,7 +1248,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	}
 
 	if (graphics_debug_dump_enabled()) {
-		LOGF("GraphicsRenderDrawIndex():Shader:\n");
+		sh_print("GraphicsRenderDrawIndex():Shader:", sh_ctx);
 		uc_print("GraphicsRenderDrawIndex():UserConfig:", ucfg);
 		hw_print(buffer);
 
@@ -1215,7 +1269,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	hw_check(buffer);
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, topology)) {
+	if (!GetDrawTopology(ucfg, false, topology)) {
 		return;
 	}
 
@@ -1250,28 +1304,34 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		index_source.size      = expanded_indices.size() * sizeof(uint16_t);
 	}
 
-	const DrawCallInfo draw {CommandBufferDebugOp::DrawIndex, args.index_count,
+	const DrawCallInfo draw {"DrawIndex", CommandBufferDebugOp::DrawIndex, args.index_count,
 	                        args.instance_count, args.first_instance};
 	DrawRenderState state {};
-	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, state)) {
+	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, true,
+	                            state)) {
 		ResetBindings();
 		return;
 	}
 
-	LogDrawStateIfNeeded(buffer, draw, state, args.index_type_and_size,
+	RefreshShaders(buffer, draw, true, state);
+
+	LogDrawStateIfNeeded(buffer, draw, state, true, false, args.index_type_and_size,
 	                     args.index_addr);
 
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
-	const auto [vertex_offset, instance_offset] =
-	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
+	const auto vertex_offset =
+	    indirect
+	        ? args.base_vertex
+	        : ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) + args.base_vertex;
 
 	DrawEmitInfo emit {};
-	emit.vertex_offset  = vertex_offset + args.base_vertex;
-	emit.first_instance = instance_offset;
+	emit.indexed       = true;
+	emit.vertex_offset = vertex_offset;
+	emit.first_instance =
+	    indirect ? args.first_instance : ResolveInstanceOffset(state.vs_input_info);
 
 	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source,
-	                    primitive_restart);
+	                    primitive_restart, true, true, false);
 	ResetBindings();
 }
 
@@ -1294,8 +1354,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer) ||
-	    ResolveColorTargets(buffer, args.render_target_slice_offset)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
 		ResetBindings();
 		return;
 	}
@@ -1305,7 +1364,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	}
 
 	if (graphics_debug_dump_enabled()) {
-		LOGF("GraphicsRenderDrawIndexAuto():Shader:\n");
+		sh_print("GraphicsRenderDrawIndexAuto():Shader:", sh_ctx);
 		uc_print("GraphicsRenderDrawIndexAuto():UserConfig:", ucfg);
 		hw_print(buffer);
 
@@ -1321,23 +1380,26 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 
 	hw_check(buffer);
 
-	const DrawCallInfo draw {CommandBufferDebugOp::DrawIndexAuto,
+	const DrawCallInfo draw {"DrawIndexAuto", CommandBufferDebugOp::DrawIndexAuto,
 	                         args.vertex_count, args.instance_count, args.first_instance};
 
-	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, topology)) {
-		ResetBindings();
-		return;
-	}
 	DrawRenderState state {};
-	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, state)) {
+	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, false,
+	                            state)) {
 		ResetBindings();
 		return;
 	}
 
-	const bool rect_list = Prospero::IsRectList(ucfg.GetPrimType());
-	if (rect_list && state.vertex_info[0].buffers_num == 0 &&
-	    state.vertex_info[0].stage.program->param_export_mask == 0 &&
+	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
+	if (!GetDrawTopology(ucfg, true, topology)) {
+		ResetBindings();
+		return;
+	}
+	RefreshShaders(buffer, draw, false, state);
+
+	const bool rect_list = topology == vk::PrimitiveTopology::ePatchList;
+	if (rect_list && state.vs_input_info.buffers_num == 0 &&
+	    state.vs_input_info.stage.program->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: "
@@ -1349,18 +1411,23 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	LogDrawStateIfNeeded(buffer, draw, state, 0, nullptr);
+	LogDrawStateIfNeeded(buffer, draw, state, false,
+	                     ucfg.GetPrimType() == Prospero::PrimitiveType::kRectListLegacy, 0,
+	                     nullptr);
 
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
-	const auto [vertex_offset, instance_offset] =
-	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
+	const auto vertex_offset =
+	    indirect ? static_cast<int32_t>(args.first_vertex)
+	             : ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
+	                   static_cast<int32_t>(args.first_vertex);
 	DrawEmitInfo emit {};
-	emit.first_vertex = static_cast<uint32_t>(vertex_offset + static_cast<int32_t>(args.first_vertex));
-	emit.first_instance = instance_offset;
+	emit.first_vertex = static_cast<uint32_t>(vertex_offset);
+	emit.first_instance =
+	    indirect ? args.first_instance : ResolveInstanceOffset(state.vs_input_info);
 
 	DrawIndexBufferSource index_source {};
-	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false);
+	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false, false,
+	                    false, true);
 	ResetBindings();
 }
 

src/graphics/host_gpu/renderer/renderDraw.h

@@ -2,14 +2,14 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERDRAW_H_
 
 #include <cstdint>
-#include <utility>
 
 namespace Libs::Graphics {
 
 struct ShaderVertexInputInfo;
 
-[[nodiscard]] std::pair<int32_t, uint32_t>
-ResolveDrawOffsets(uint32_t index_offset, const ShaderVertexInputInfo& vs_input_info);
+[[nodiscard]] int32_t  ResolveVertexOffset(uint32_t                     index_offset,
+                                           const ShaderVertexInputInfo& vs_input_info);
+[[nodiscard]] uint32_t ResolveInstanceOffset(const ShaderVertexInputInfo& vs_input_info);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/renderer/renderTarget.h

@@ -5,6 +5,7 @@
 
 #include <array>
 #include <cstdint>
+#include <type_traits>
 
 namespace Libs::Graphics {
 
@@ -66,6 +67,41 @@ inline constexpr TargetViewInfo ResolveTargetViewInfo(uint32_t base_layer, uint3
 	        base_layer, last_layer - base_layer + 1u, last_layer + 1u};
 }
 
+#pragma pack(push, 1)
+
+struct PipelineStencilStaticState {
+	vk::StencilOp failOp      = vk::StencilOp::eKeep;
+	vk::StencilOp passOp      = vk::StencilOp::eKeep;
+	vk::StencilOp depthFailOp = vk::StencilOp::eKeep;
+	vk::CompareOp compareOp   = vk::CompareOp::eNever;
+};
+
+struct PipelineStencilDynamicState {
+	uint32_t compareMask = 0;
+	uint32_t writeMask   = 0;
+	uint32_t reference   = 0;
+};
+
+#pragma pack(pop)
+
+inline constexpr bool stencil_face_accesses_attachment(const PipelineStencilStaticState&  state,
+                                                       const PipelineStencilDynamicState& dynamic) {
+	return state.compareOp != vk::CompareOp::eAlways ||
+	       (dynamic.writeMask != 0 &&
+	        (state.failOp != vk::StencilOp::eKeep || state.passOp != vk::StencilOp::eKeep ||
+	         state.depthFailOp != vk::StencilOp::eKeep));
+}
+
+static_assert(std::is_trivially_copyable_v<PipelineStencilStaticState>);
+static_assert(std::is_standard_layout_v<PipelineStencilStaticState>);
+static_assert(alignof(PipelineStencilStaticState) == 1);
+static_assert(sizeof(PipelineStencilStaticState) ==
+              sizeof(vk::StencilOp) * 3 + sizeof(vk::CompareOp));
+static_assert(std::is_trivially_copyable_v<PipelineStencilDynamicState>);
+static_assert(std::is_standard_layout_v<PipelineStencilDynamicState>);
+static_assert(alignof(PipelineStencilDynamicState) == 1);
+static_assert(sizeof(PipelineStencilDynamicState) == sizeof(uint32_t) * 3);
+
 } // namespace Libs::Graphics
 
 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERTARGET_H_