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: |
@@ -105,23 +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 pad_haptics_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests ^
-            shader_recompiler_compute_tests ^
-            resource_tracking_tests scalar_provenance_tests ^
-            resource_materialization_tests shader_vertex_metadata_tests archive_file_tests avplayer_file_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|pad_haptics|ime_dialog|lru_cache|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|archive_file|avplayer_file|virtual_memory_allocation)$"
+          ctest --test-dir _Build/windows --output-on-failure -R "^(audio_out2_port|ime_dialog|virtual_memory_allocation)$"
 
       - name: Install
         shell: cmd
@@ -156,7 +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
@@ -167,9 +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: |
@@ -194,25 +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 pad_haptics_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests \
-              memory_tracker_tests page_manager_tests shader_recompiler_compute_tests \
-              resource_tracking_tests scalar_provenance_tests \
-              resource_materialization_tests shader_vertex_metadata_tests archive_file_tests avplayer_file_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|pad_haptics|ime_dialog|lru_cache|memory_tracker|page_manager|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|archive_file|avplayer_file|virtual_memory_allocation)$'
+            -R '^(audio_out2_port|ime_dialog|virtual_memory_allocation)$'
 
       - name: Install
         shell: bash
@@ -321,9 +302,8 @@ jobs:
             libxkbcommon-dev \
             libxrandr-dev \
             libxss-dev \
-            libxtst-dev \
             lld \
-            pkg-config \
+            ninja-build \
             wayland-protocols
 
       - name: Install Qt 6.10.3
@@ -335,9 +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: |
@@ -362,12 +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
@@ -386,16 +361,14 @@ jobs:
         run: |
           cmake --build _Build/linux \
             --target launcher page_manager_tests memory_tracker_tests \
-              audio_out2_port_tests pad_haptics_tests ime_dialog_tests lru_cache_tests virtual_memory_allocation_tests \
-              resource_tracking_tests scalar_provenance_tests \
-              resource_materialization_tests shader_vertex_metadata_tests archive_file_tests avplayer_file_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|pad_haptics|ime_dialog|lru_cache|memory_tracker|page_manager|resource_tracking|scalar_provenance|resource_materialization|shader_vertex_metadata|archive_file|avplayer_file|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

@@ -2,31 +2,10 @@ set(BUILD_SHARED_LIBS OFF CACHE BOOL "" FORCE)
 
 add_subdirectory(cpuinfo)
 
-include(FetchContent)
-
-# The Bluetooth DualSense transports speaker audio as fixed-size Opus packets.
-set(kyty_build_testing_saved "${BUILD_TESTING}")
-set(BUILD_TESTING OFF)
-set(OPUS_BUILD_PROGRAMS OFF CACHE BOOL "" FORCE)
-set(OPUS_INSTALL_PKG_CONFIG_MODULE OFF CACHE BOOL "" FORCE)
-set(OPUS_INSTALL_CMAKE_CONFIG_MODULE OFF CACHE BOOL "" FORCE)
-if(MSVC AND CMAKE_C_COMPILER_ID STREQUAL "Clang")
-	# Opus assumes MSVC accepts SSE4.1 intrinsics without per-file flags, but
-	# clang-cl rejects them in its baseline x64 files. Its AVX2 dispatch also
-	# depends on the SSE4.1 declarations, so retain the x64 SSE2 baseline only.
-	set(OPUS_X86_MAY_HAVE_SSE4_1 OFF CACHE BOOL "" FORCE)
-	set(OPUS_X86_MAY_HAVE_AVX2 OFF CACHE BOOL "" FORCE)
-endif()
-FetchContent_Declare(opus
-	GIT_REPOSITORY https://github.com/xiph/opus.git
-	GIT_TAG ddbe48383984d56acd9e1ab6a090c54ca6b735a6
-	GIT_SHALLOW FALSE)
-FetchContent_MakeAvailable(opus)
-set(BUILD_TESTING "${kyty_build_testing_saved}")
-install(FILES "${opus_SOURCE_DIR}/COPYING" DESTINATION "licenses/opus")
-
 # The Windows guest red-zone patcher needs a decoder/encoder and a small
 # trampoline assembler. Pin the same revisions used by shadPS4's implementation.
+include(FetchContent)
+
 set(XBYAK_BUILD_TESTS OFF CACHE BOOL "" FORCE)
 FetchContent_Declare(
 	xbyak
@@ -49,55 +28,8 @@ FetchContent_Declare(
 
 FetchContent_MakeAvailable(xbyak zydis)
 
-# ZArchive provides random-access reads for compressed game dumps. Build only
-# the reader/writer library and its zstd dependency; the command-line tool is
-# not part of KytyPS5's runtime distribution.
-set(ZSTD_BUILD_PROGRAMS OFF CACHE BOOL "" FORCE)
-set(ZSTD_BUILD_SHARED OFF CACHE BOOL "" FORCE)
-set(ZSTD_BUILD_STATIC ON CACHE BOOL "" FORCE)
-set(ZSTD_BUILD_TESTS OFF CACHE BOOL "" FORCE)
-set(ZSTD_LEGACY_SUPPORT OFF CACHE BOOL "" FORCE)
-set(ZSTD_MULTITHREAD_SUPPORT OFF CACHE BOOL "" FORCE)
-FetchContent_Declare(
-	zstd
-	GIT_REPOSITORY https://github.com/facebook/zstd.git
-	GIT_TAG f8745da6ff1ad1e7bab384bd1f9d742439278e99
-	GIT_SHALLOW FALSE
-	SOURCE_SUBDIR build/cmake
-)
-FetchContent_Declare(
-	zarchive_source
-	GIT_REPOSITORY https://github.com/Exzap/ZArchive.git
-	GIT_TAG d2c717730092c7bf8cbb033b12fd4001b7c4d932
-	GIT_SHALLOW FALSE
-	PATCH_COMMAND "${GIT_EXECUTABLE}" apply --ignore-whitespace
-		"${CMAKE_CURRENT_SOURCE_DIR}/patches/zarchive-reader.patch"
-	SOURCE_SUBDIR kyty-library-only
-)
-FetchContent_MakeAvailable(zstd zarchive_source)
-
-add_library(zarchive STATIC
-	"${zarchive_source_SOURCE_DIR}/src/zarchivereader.cpp"
-	"${zarchive_source_SOURCE_DIR}/src/zarchivewriter.cpp"
-	"${zarchive_source_SOURCE_DIR}/src/sha_256.c"
-)
-add_library(ZArchive::zarchive ALIAS zarchive)
-target_include_directories(zarchive SYSTEM PUBLIC "${zarchive_source_SOURCE_DIR}/include")
-target_include_directories(zarchive PRIVATE "${zstd_SOURCE_DIR}/lib")
-target_link_libraries(zarchive PRIVATE libzstd_static)
-target_compile_features(zarchive PRIVATE cxx_std_20)
-if(MSVC)
-	target_compile_options(zarchive PRIVATE /W0)
-else()
-	target_compile_options(zarchive PRIVATE -w)
-endif()
-
 set(SDL_SHARED OFF CACHE BOOL "" FORCE)
-if(APPLE)
-	# SDL's built-in iconv fallback does not support CP932.
-	set(SDL_SYSTEM_ICONV ON CACHE BOOL "Use system iconv for CP932 conversion" FORCE)
-endif()
-add_subdirectory(SDL3)
+add_subdirectory(SDL2)
 
 add_library(Vulkan-Headers INTERFACE)
 add_library(Vulkan::Headers ALIAS Vulkan-Headers)
@@ -154,9 +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)
@@ -188,7 +117,6 @@ endif()
 
 set(TRACY_STATIC ON CACHE BOOL "" FORCE)
 set(TRACY_ENABLE ON CACHE BOOL "" FORCE)
-set(TRACY_NO_CRASH_HANDLER ON CACHE BOOL "" FORCE)
 set(TRACY_ON_DEMAND ON CACHE BOOL "" FORCE)
 set(TRACY_DELAYED_INIT ON CACHE BOOL "" FORCE)
 set(TRACY_MANUAL_LIFETIME ON CACHE BOOL "" FORCE)

3rdparty/patches/zarchive-reader.patch

@@ -1,100 +0,0 @@
---- a/include/zarchive/zarchivecommon.h
-+++ b/include/zarchive/zarchivecommon.h
-@@ -256,7 +256,7 @@
- 
- 			bool IsWithinValidRange(uint64_t fileSize) const
- 			{
--				return (offset + size) <= fileSize;
-+				return offset <= fileSize && size <= fileSize - offset;
- 			}
- 		};
- 
---- a/src/zarchivereader.cpp
-+++ b/src/zarchivereader.cpp
-@@ -14,6 +14,7 @@
- 
- static bool _ifstream_readBytes(std::ifstream& file, uint64_t offset, void* buffer, uint32_t size)
- {
-+	file.clear();
- 	file.seekg(offset, std::ios_base::beg);
- 	file.read((char*)buffer, size);
- 	return file.gcount() == size;
-@@ -80,9 +81,27 @@
- 	// verify file tree
- 	if (fileTree[0].IsFile())
- 		return nullptr; // first entry must be root directory
--	auto rootName = GetName(nameTable, fileTree[0].GetNameOffset());
--	if (!rootName.empty())
-+	if (fileTree[0].GetNameOffset() != 0x7FFFFFFF)
- 		return nullptr; // root node must not have a name
-+	const uint64_t dataCapacity = (uint64_t)offsetRecords.size() * _ZARCHIVE::ENTRIES_PER_OFFSETRECORD * _ZARCHIVE::COMPRESSED_BLOCK_SIZE;
-+	for (size_t i = 0; i < fileTree.size(); ++i)
-+	{
-+		auto& entry = fileTree[i];
-+		if (i != 0)
-+		{
-+			auto name = GetName(nameTable, entry.GetNameOffset());
-+			if (name.empty() || name == "." || name == ".." || name.find_first_of("/\\") != std::string_view::npos || name.find('\0') != std::string_view::npos)
-+				return nullptr;
-+		}
-+		if (entry.IsFile())
-+		{
-+			if (entry.GetFileOffset() > dataCapacity || entry.GetFileSize() > dataCapacity - entry.GetFileOffset())
-+				return nullptr;
-+		}
-+		else if (entry.directoryRecord.nodeStartIndex > fileTree.size() ||
-+			entry.directoryRecord.count > fileTree.size() - entry.directoryRecord.nodeStartIndex)
-+			return nullptr;
-+	}
- 	// read meta data
- 	// todo
- 
-@@ -238,7 +257,7 @@
- 		uint32_t stepSize = std::min<uint32_t>(remainingBytes, _ZARCHIVE::COMPRESSED_BLOCK_SIZE - blockOffset);
- 		CacheBlock* block = GetCachedBlock(blockIdx);
- 		if (!block)
--			return 0;
-+			return bytesToRead - remainingBytes;
- 		std::memcpy(bufferU8, block->data + blockOffset, stepSize);
- 		rawReadOffset += stepSize;
- 		remainingBytes -= stepSize;
-@@ -325,14 +344,18 @@
- 	// determine offset and size of compressed block
- 	auto& record = m_offsetRecords[recordIndex];
- 	uint64_t offset = record.baseOffset;
-+	if (offset > m_compressedDataSize)
-+		return false;
- 	for (uint32_t i = 0; i < recordSubIndex; i++)
- 	{
--		offset += (uint64_t)record.size[i];
--		offset++;
-+		const uint64_t size = (uint64_t)record.size[i] + 1;
-+		if (size > m_compressedDataSize - offset)
-+			return false;
-+		offset += size;
- 	}
- 	uint32_t compressedSize = (uint32_t)record.size[recordSubIndex] + 1;
- 	// load file data
--	if ((offset + compressedSize) > m_compressedDataSize)
-+	if (compressedSize > m_compressedDataSize - offset)
- 		return false;
- 	offset += m_compressedDataOffset;
- 	if (compressedSize == _ZARCHIVE::COMPRESSED_BLOCK_SIZE)
-@@ -350,7 +373,7 @@
- // returns empty view on failure
- std::string_view ZArchiveReader::GetName(const std::vector<uint8_t>& nameTable, uint32_t nameOffset)
- {
--	if (nameOffset == 0x7FFFFFFF || nameOffset > nameTable.size())
-+	if (nameOffset == 0x7FFFFFFF || nameOffset >= nameTable.size())
- 		return "";
- 	// parse header
- 	uint16_t nameLength = nameTable[nameOffset] & 0x7F;
-@@ -359,7 +382,7 @@
- 		// extended 2-byte length
- 		if (nameOffset + 1 >= nameTable.size())
- 			return "";
--		nameLength |= ((uint16_t)nameTable[nameOffset] << 7);
-+		nameLength |= ((uint16_t)nameTable[nameOffset + 1] << 7);
- 		nameOffset += 2;
- 	}
- 	else

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.18...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 opus)
+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
@@ -407,26 +404,11 @@ add_executable(page_manager_tests EXCLUDE_FROM_ALL
 target_link_libraries(page_manager_tests common)
 target_include_directories(page_manager_tests PRIVATE ${inc_headers})
 
-add_executable(archive_file_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ArchiveFileTests.cpp"
-)
-target_link_libraries(archive_file_tests common ZArchive::zarchive)
-target_include_directories(archive_file_tests PRIVATE ${inc_headers})
-
-add_executable(avplayer_file_tests EXCLUDE_FROM_ALL "${KYTY_TESTS_DIR}/AvPlayerFileTests.cpp")
-target_link_libraries(avplayer_file_tests common FFmpeg::ffmpeg ZArchive::zarchive)
-target_include_directories(avplayer_file_tests PRIVATE ${inc_headers})
-
 add_executable(bit_array_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/BitArrayTests.cpp"
 )
 target_include_directories(bit_array_tests PRIVATE ${inc_headers})
 
-add_executable(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"
@@ -481,47 +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(audio_out_timing_tests EXCLUDE_FROM_ALL "${KYTY_TESTS_DIR}/AudioOutTimingTests.cpp")
-target_link_libraries(audio_out_timing_tests common fmt::fmt SDL3::SDL3-static)
-target_include_directories(audio_out_timing_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 opus)
-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 ${inc_headers})
-
-add_executable(png_enc_tests EXCLUDE_FROM_ALL "${KYTY_TESTS_DIR}/PngEncTests.cpp")
-target_link_libraries(png_enc_tests common fmt::fmt)
-target_include_directories(png_enc_tests PRIVATE ${inc_headers})
-
-add_executable(ces_tests EXCLUDE_FROM_ALL "${KYTY_TESTS_DIR}/CesTests.cpp")
-target_link_libraries(ces_tests common fmt::fmt SDL3::SDL3-static)
-target_include_directories(ces_tests PRIVATE ${inc_headers})
-
-add_executable(save_data_dialog_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/SaveDataDialogTests.cpp"
-	"${KYTY_SOURCE_DIR}/libs/dialog.cpp"
-)
-target_link_libraries(save_data_dialog_tests common fmt::fmt)
-target_include_directories(save_data_dialog_tests PRIVATE ${inc_headers})
-
-add_kyty_full_emulator_test(http_uri_parse_tests
-	"${KYTY_TESTS_DIR}/HttpUriParseTests.cpp")
-
 add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/EventQueueLifetimeTests.cpp"
 	"${KYTY_SOURCE_DIR}/kernel/eventQueue.cpp"
@@ -532,7 +473,6 @@ target_include_directories(event_queue_lifetime_tests PRIVATE ${inc_headers})
 
 add_kyty_full_emulator_test(kernel_file_system_tests
 	"${KYTY_TESTS_DIR}/KernelFileSystemTests.cpp")
-target_link_libraries(kernel_file_system_tests ZArchive::zarchive)
 
 add_executable(sync_on_address_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/SyncOnAddressTests.cpp"
@@ -599,10 +539,7 @@ if(BUILD_TESTING)
 	add_test(NAME image_page_table COMMAND $<TARGET_FILE:image_page_table_tests>)
 	add_test(NAME memory_tracker COMMAND $<TARGET_FILE:memory_tracker_tests>)
 	add_test(NAME page_manager COMMAND $<TARGET_FILE:page_manager_tests>)
-	add_test(NAME archive_file COMMAND $<TARGET_FILE:archive_file_tests>)
-	add_test(NAME avplayer_file COMMAND $<TARGET_FILE:avplayer_file_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>)
@@ -610,21 +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 audio_out_timing COMMAND $<TARGET_FILE:audio_out_timing_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 png_enc COMMAND $<TARGET_FILE:png_enc_tests>)
-	add_test(NAME ces COMMAND $<TARGET_FILE:ces_tests>)
-	add_test(NAME save_data_dialog COMMAND $<TARGET_FILE:save_data_dialog_tests>)
-	add_test(NAME http_uri_parse COMMAND $<TARGET_FILE:http_uri_parse_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
@@ -654,8 +579,6 @@ if(BUILD_TESTING)
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --compute-meta-clear-only)
 	add_test(NAME buffer_cache_dirty_gc
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --buffer-cache-gc-only)
-	add_test(NAME color_render_target_1d
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --rt1d-only)
 
 	if(WIN32)
 		# These tests still depend on the Windows multisample-depth path.
@@ -674,10 +597,7 @@ if(BUILD_TESTING)
 		image_page_table_tests
 		memory_tracker_tests
 		page_manager_tests
-		archive_file_tests
-		avplayer_file_tests
 		bit_array_tests
-		lru_cache_tests
 		shader_vertex_metadata_tests
 		resource_materialization_tests
 		resource_tracking_tests
@@ -687,32 +607,9 @@ if(BUILD_TESTING)
 		shader_recompiler_compute_tests
 		virtual_memory_allocation_tests
 		audio_out2_port_tests
-		audio_out_timing_tests
-		ngs2_sampler_tests
-		pad_haptics_tests
-		save_data_memory_tests
-		png_enc_tests
-		ces_tests
-		save_data_dialog_tests
-		http_uri_parse_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)
@@ -752,16 +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)
@@ -779,10 +667,6 @@ endif()
 install(TARGETS kyty_emulator DESTINATION .)
 if(KYTY_BUILD_LAUNCHER)
 	add_subdirectory("${KYTY_SOURCE_DIR}/launcher" "${CMAKE_CURRENT_BINARY_DIR}/launcher")
-
-	if(LINUX)
-		target_link_options(launcher PRIVATE $<$<CONFIG:Release>:-Wl,--strip-all>)
-	endif()
 endif()
 if(KYTY_CLANG_CL)
 	install(FILES "${KYTY_THIRD_PARTY_DIR}/winpthread/bin/libwinpthread-1.dll" DESTINATION .)

README.md

@@ -4,17 +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)
 
-**[Weekly updates](https://github.com/KytyPS5/KytyPS5/discussions/862)** — game progress, recent fixes and ongoing development.
-
-**[Development on Discord](https://discord.gg/UNrkMqGaBg)** — KytyPS5 development.
-
 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
@@ -24,12 +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
@@ -40,16 +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>
@@ -68,12 +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>UFC 5</strong><br>
-      <img src="docs/screenshots/ps5-06.png" width="300" alt="UFC 5 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>
@@ -83,7 +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
@@ -94,14 +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
 ```
 
@@ -136,12 +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`.
 
@@ -170,19 +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
@@ -197,50 +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`, `zstd`, and ZArchive
-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
@@ -333,21 +285,17 @@ open _Build/macos/install/KytyPS5.app  # or double-click in Finder
 ```
 
 On first launch, add one or more game folders in the global settings. The launcher searches those
-folders recursively for game directories containing `eboot.bin` and ZArchive (`.zar`) game dumps
-whose archive root contains `eboot.bin`. Select a detected game and run it from the game list.
-ZArchive dumps are mounted read-only and streamed directly; they do not need to be extracted first.
+folders recursively for game directories containing `eboot.bin`. Select a detected game and run it
+from the game list.
 
-The emulator can also be started directly with a legally obtained game directory, ELF file, or
-ZArchive dump:
+The emulator can also be started directly with a legally obtained game directory or ELF file:
 
 ```powershell
 .\_Build\windows\install\kyty_emulator.exe --game "D:\Games\ExampleGame"
-.\_Build\windows\install\kyty_emulator.exe --game "D:\Games\ExampleGame.zar"
 ```
 
 ```bash
 ./_Build/linux/install/kyty_emulator --game "/games/ExampleGame"
-./_Build/linux/install/kyty_emulator --game "/games/ExampleGame.zar"
 ```
 
 On macOS, the adjacent flat or app-bundled `libMoltenVK.dylib` is found automatically; no
@@ -391,6 +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/CMakeLists.txt

@@ -26,7 +26,6 @@ target_link_libraries(common
 		cpuinfo
 		kyty_git_version
 		spdlog::spdlog
-		ZArchive::zarchive
 )
 
 get_property(inc_headers TARGET common PROPERTY INCLUDE_DIRECTORIES)

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/archive.cpp

@@ -1,397 +0,0 @@
-#include "common/archive.h"
-
-#include "common/stringUtils.h"
-
-#include <algorithm>
-#include <condition_variable>
-#include <cstring>
-#include <mutex>
-#include <optional>
-#include <string_view>
-#include <thread>
-#include <unordered_map>
-#include <utility>
-#include <zarchive/zarchivereader.h>
-
-// POSIX hosts and MinGW's pthread-backed libstdc++ have cancellable condition waits.
-// MSVC/ClangCL and MinGW's native Win32 threading backend do not use pthread waits.
-#if !defined(_WIN32) || defined(_GLIBCXX_GCC_GTHR_POSIX_H)
-#include <pthread.h>
-#define KYTY_ARCHIVE_PTHREAD_CANCELLATION
-#endif
-
-namespace Common {
-
-// Backends expose immutable metadata and offset-based reads. Cursor management, path routing,
-// reader lifetime and the host-stack worker are shared by every archive format.
-class ArchiveReader {
-public:
-	struct Entry {
-		uint64_t id;
-		uint64_t size;
-		bool     is_file;
-	};
-
-	virtual ~ArchiveReader()                                                       = default;
-	virtual std::optional<Entry>        Find(std::string_view member)              = 0;
-	virtual std::vector<File::DirEntry> List(std::string_view member)              = 0;
-	virtual uint64_t Read(uint64_t id, uint64_t offset, uint32_t size, void* data) = 0;
-};
-
-namespace {
-
-class ZArchiveReaderBackend final: public ArchiveReader {
-public:
-	explicit ZArchiveReaderBackend(std::unique_ptr<ZArchiveReader> reader)
-	    : m_reader(std::move(reader)) {}
-
-	std::optional<Entry> Find(std::string_view member) override {
-		const auto node = m_reader->LookUp(member);
-		if (node == ZARCHIVE_INVALID_NODE) {
-			return {};
-		}
-		return Entry {node, m_reader->GetFileSize(node), m_reader->IsFile(node)};
-	}
-
-	std::vector<File::DirEntry> List(std::string_view member) override {
-		const auto node = m_reader->LookUp(member);
-		if (!m_reader->IsDirectory(node)) {
-			return {};
-		}
-		const auto count = m_reader->GetDirEntryCount(node);
-		if (count > (1u << 20u)) {
-			return {};
-		}
-		std::vector<File::DirEntry> result;
-		result.reserve(count);
-		for (uint32_t index = 0; index < count; ++index) {
-			ZArchiveReader::DirEntry entry {};
-			if (m_reader->GetDirEntry(node, index, entry)) {
-				result.push_back({std::string(entry.name), entry.isFile});
-			}
-		}
-		return result;
-	}
-
-	uint64_t Read(uint64_t id, uint64_t offset, uint32_t size, void* data) override {
-		// ZArchiveReader already serializes access to its stream and decompression cache.
-		return m_reader->ReadFromFile(static_cast<ZArchiveNodeHandle>(id), offset, size, data);
-	}
-
-private:
-	std::unique_ptr<ZArchiveReader> m_reader;
-};
-
-std::shared_ptr<ArchiveReader> OpenZArchive(const std::filesystem::path& path) {
-	auto reader = std::unique_ptr<ZArchiveReader>(ZArchiveReader::OpenFromFile(path));
-	return reader ? std::make_shared<ZArchiveReaderBackend>(std::move(reader)) : nullptr;
-}
-
-struct ArchiveFormat {
-	std::string_view extension;
-	std::shared_ptr<ArchiveReader> (*open)(const std::filesystem::path&);
-};
-
-constexpr ArchiveFormat Formats[] = {{".zar", OpenZArchive}};
-using NativeView                  = std::basic_string_view<std::filesystem::path::value_type>;
-
-const ArchiveFormat* FindFormat(NativeView path) {
-	for (const auto& format: Formats) {
-		if (path.size() < format.extension.size()) {
-			continue;
-		}
-		const auto extension = path.substr(path.size() - format.extension.size());
-		if (std::equal(
-		        extension.begin(), extension.end(), format.extension.begin(),
-		        [](auto a, char b) { return (a >= 'A' && a <= 'Z' ? a + ('a' - 'A') : a) == b; })) {
-			return &format;
-		}
-	}
-	return nullptr;
-}
-
-size_t FindMarker(NativeView path) {
-	// Native files use this check on every operation. The usual path has no marker and allocates
-	// nothing.
-	for (auto marker = path.find('!'); marker != NativeView::npos;
-	     marker      = path.find('!', marker + 1)) {
-		if ((marker + 1 == path.size() || path[marker + 1] == '/' || path[marker + 1] == '\\') &&
-		    FindFormat(path.substr(0, marker)) != nullptr) {
-			return marker;
-		}
-	}
-	return NativeView::npos;
-}
-
-struct ParsedPath {
-	std::filesystem::path archive;
-	std::string           member;
-};
-
-std::optional<ParsedPath> ParsePath(const std::filesystem::path& path) {
-	const auto& native = path.native();
-	const auto  marker = FindMarker(native);
-	if (marker == NativeView::npos) {
-		return {};
-	}
-	ParsedPath parsed {.archive = std::filesystem::path(native.substr(0, marker))};
-	auto       member = PathToGenericString(std::filesystem::path(native.substr(marker + 1)));
-	std::replace(member.begin(), member.end(), '\\', '/');
-	std::string_view remaining = member;
-	while (!remaining.empty()) {
-		const auto separator = remaining.find('/');
-		const auto component = remaining.substr(0, separator);
-		remaining.remove_prefix(separator == std::string_view::npos ? remaining.size() : separator + 1);
-		if (component.empty() || component == ".") {
-			continue;
-		}
-		if (component == "..") {
-			if (parsed.member.empty()) {
-				return {};
-			}
-			const auto separator = parsed.member.rfind('/');
-			parsed.member.resize(separator == std::string::npos ? 0 : separator);
-		} else {
-			if (!parsed.member.empty()) {
-				parsed.member += '/';
-			}
-			parsed.member += component;
-		}
-	}
-	return parsed;
-}
-
-std::mutex                                                              g_readers_mutex;
-std::unordered_map<std::filesystem::path, std::weak_ptr<ArchiveReader>> g_readers;
-
-// Guest stacks can be only 64 KiB. Decompress on an ordinary host stack, then copy into guest
-// memory on the caller so GPU-tracked writes use its fault handler. A single worker bounds host
-// threads and avoids duplicating the archive's own block cache. Requests live on waiting callers.
-class ArchiveIoThread {
-public:
-	static ArchiveIoThread& Instance() {
-		// Guest exit can run static destructors while other guest threads are still reading.
-		// Keep the worker alive until process termination, as with the guest threads themselves.
-		static auto* instance = new ArchiveIoThread;
-		return *instance;
-	}
-
-	uint64_t Read(ArchiveReader& reader, uint64_t id, uint64_t offset, uint32_t size, void* data) {
-		Request          request {reader, id, offset, size, data};
-		std::unique_lock lock(m_mutex);
-		if (m_last != nullptr) {
-			m_last->next = &request;
-		} else {
-			m_first = &request;
-		}
-		m_last = &request;
-		m_wake.notify_one();
-		request.ready.wait(lock, [&] { return request.done; });
-		return request.result;
-	}
-
-	KYTY_CLASS_NO_COPY(ArchiveIoThread);
-
-private:
-	struct Request {
-		ArchiveReader&          reader;
-		uint64_t                id;
-		uint64_t                offset;
-		uint32_t                size;
-		void*                   data;
-		Request*                next   = nullptr;
-		uint64_t                result = 0;
-		bool                    done   = false;
-		std::condition_variable ready {};
-	};
-
-	ArchiveIoThread() {
-		std::thread([this] { Loop(); }).detach();
-	}
-
-	void Loop() {
-		for (;;) {
-			Request* request;
-			{
-				std::unique_lock lock(m_mutex);
-				m_wake.wait(lock, [this] { return m_first != nullptr; });
-				request = m_first;
-				m_first = request->next;
-				if (m_first == nullptr) {
-					m_last = nullptr;
-				}
-			}
-			const auto result =
-			    request->reader.Read(request->id, request->offset, request->size, request->data);
-			{
-				std::lock_guard lock(m_mutex);
-				request->result = result;
-				request->done   = true;
-				// Notify before releasing the mutex: the waiting caller owns the condition
-				// variable.
-				request->ready.notify_one();
-			}
-		}
-	}
-
-	std::mutex              m_mutex;
-	std::condition_variable m_wake;
-	Request*                m_first = nullptr;
-	Request*                m_last  = nullptr;
-};
-
-} // namespace
-
-struct ArchiveFile::Private {
-	std::shared_ptr<ArchiveReader> reader;
-	ArchiveReader::Entry           entry;
-	uint64_t                       position = 0;
-};
-
-ArchiveFile::ArchiveFile(std::unique_ptr<Private> p): m_p(std::move(p)) {}
-ArchiveFile::~ArchiveFile() = default;
-
-uint64_t ArchiveFile::Size() const {
-	return m_p->entry.size;
-}
-uint64_t ArchiveFile::Tell() const {
-	return m_p->position;
-}
-bool ArchiveFile::Seek(uint64_t offset) {
-	m_p->position = offset;
-	return true;
-}
-
-void ArchiveFile::Read(void* data, uint32_t size, uint32_t* bytes_read) {
-#if defined(KYTY_ARCHIVE_PTHREAD_CANCELLATION)
-	// A pending cancellation must not destroy the waiting request or its TLS buffer while the
-	// worker still uses them. Restore the caller's state only after the copy is complete.
-	int cancellation_state = PTHREAD_CANCEL_ENABLE;
-	pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cancellation_state);
-#endif
-	constexpr uint32_t                chunk_size = 1u << 20u;
-	thread_local std::vector<uint8_t> chunk;
-	uint32_t                          read = 0;
-	if (data != nullptr && size != 0 && m_p->position < Size()) {
-		const auto requested =
-		    static_cast<uint32_t>(std::min<uint64_t>(size, Size() - m_p->position));
-		if (chunk.size() < std::min(chunk_size, requested)) {
-			chunk.resize(std::min(chunk_size, requested));
-		}
-		while (read < requested) {
-			const auto step = std::min(chunk_size, requested - read);
-			const auto got  = ArchiveIoThread::Instance().Read(*m_p->reader, m_p->entry.id,
-			                                                   m_p->position, step, chunk.data());
-			if (got == 0 || got > step) {
-				break;
-			}
-			std::memcpy(static_cast<uint8_t*>(data) + read, chunk.data(), got);
-			read += static_cast<uint32_t>(got);
-			m_p->position += got;
-			if (got < step) {
-				break;
-			}
-		}
-	}
-	if (bytes_read != nullptr) {
-		*bytes_read = read;
-	}
-#if defined(KYTY_ARCHIVE_PTHREAD_CANCELLATION)
-	pthread_setcancelstate(cancellation_state, nullptr);
-#endif
-}
-
-std::filesystem::path MakeArchivePath(const std::filesystem::path& archive,
-                                      const std::filesystem::path& member) {
-	auto root = archive;
-	root += "!";
-	return member.empty() ? root : root / member.relative_path();
-}
-
-bool IsSupportedArchive(const std::filesystem::path& path) {
-	return FindFormat(path.native()) != nullptr;
-}
-
-bool IsArchivePath(const std::filesystem::path& path) {
-	return FindMarker(path.native()) != NativeView::npos;
-}
-
-std::filesystem::path GetArchiveHostPath(const std::filesystem::path& path) {
-	const auto marker = FindMarker(path.native());
-	return marker == NativeView::npos ? std::filesystem::path {}
-	                                  : std::filesystem::path(path.native().substr(0, marker));
-}
-
-std::shared_ptr<ArchiveReader> OpenArchive(const std::filesystem::path& path) {
-	const auto  archive = IsArchivePath(path) ? GetArchiveHostPath(path) : path;
-	const auto* format  = FindFormat(archive.native());
-	if (format == nullptr) {
-		return {};
-	}
-	std::error_code error;
-	auto            key = std::filesystem::absolute(archive, error);
-	if (error) {
-		return {};
-	}
-	{
-		std::lock_guard lock(g_readers_mutex);
-		if (auto it = g_readers.find(key); it != g_readers.end()) {
-			if (auto reader = it->second.lock()) {
-				return reader;
-			}
-		}
-	}
-	// Do not hold the cache mutex while opening an unrelated archive from disk.
-	auto reader = format->open(archive);
-	if (reader == nullptr) {
-		return {};
-	}
-	std::lock_guard lock(g_readers_mutex);
-	std::erase_if(g_readers, [](const auto& entry) { return entry.second.expired(); });
-	auto& cached = g_readers[key];
-	if (auto existing = cached.lock()) {
-		return existing;
-	}
-	cached = reader;
-	return reader;
-}
-
-std::optional<File::Info> GetArchiveInfo(const std::filesystem::path& path) {
-	const auto parsed = ParsePath(path);
-	if (parsed) {
-		if (auto reader = OpenArchive(parsed->archive)) {
-			if (auto entry = reader->Find(parsed->member)) {
-				return File::Info {entry->is_file, entry->size};
-			}
-		}
-	}
-	return {};
-}
-
-std::vector<File::DirEntry> GetArchiveDirEntries(const std::filesystem::path& path) {
-	const auto parsed = ParsePath(path);
-	if (parsed) {
-		if (auto reader = OpenArchive(parsed->archive)) {
-			return reader->List(parsed->member);
-		}
-	}
-	return {};
-}
-
-std::unique_ptr<ArchiveFile> OpenArchiveFile(const std::filesystem::path& path) {
-	auto parsed = ParsePath(path);
-	if (!parsed) {
-		return {};
-	}
-	auto       reader = OpenArchive(parsed->archive);
-	const auto entry  = reader ? reader->Find(parsed->member) : std::nullopt;
-	if (!entry || !entry->is_file) {
-		return {};
-	}
-	auto p     = std::make_unique<ArchiveFile::Private>();
-	p->reader  = std::move(reader);
-	p->entry   = *entry;
-	return std::unique_ptr<ArchiveFile>(new ArchiveFile(std::move(p)));
-}
-
-} // namespace Common

src/common/archive.h

@@ -1,50 +0,0 @@
-#ifndef KYTY_COMMON_ARCHIVE_H_
-#define KYTY_COMMON_ARCHIVE_H_
-
-#include "common/file.h"
-
-#include <cstdint>
-#include <filesystem>
-#include <memory>
-#include <vector>
-
-namespace Common {
-
-class ArchiveReader;
-
-class ArchiveFile {
-public:
-	~ArchiveFile();
-
-	[[nodiscard]] uint64_t Size() const;
-	[[nodiscard]] uint64_t Tell() const;
-	bool                   Seek(uint64_t offset);
-	void                   Read(void* data, uint32_t size, uint32_t* bytes_read);
-
-	KYTY_CLASS_NO_COPY(ArchiveFile);
-
-private:
-	struct Private;
-	explicit ArchiveFile(std::unique_ptr<Private> p);
-	std::unique_ptr<Private> m_p;
-
-	friend std::unique_ptr<ArchiveFile> OpenArchiveFile(const std::filesystem::path& path);
-};
-
-// Virtual paths use "archive.ext!/member" and retain native host-path encoding.
-[[nodiscard]] std::filesystem::path       MakeArchivePath(const std::filesystem::path& archive,
-                                                          const std::filesystem::path& member = {});
-[[nodiscard]] bool                        IsSupportedArchive(const std::filesystem::path& path);
-[[nodiscard]] bool                        IsArchivePath(const std::filesystem::path& path);
-[[nodiscard]] std::optional<File::Info>   GetArchiveInfo(const std::filesystem::path& path);
-[[nodiscard]] std::vector<File::DirEntry> GetArchiveDirEntries(const std::filesystem::path& path);
-[[nodiscard]] std::filesystem::path       GetArchiveHostPath(const std::filesystem::path& path);
-[[nodiscard]] std::unique_ptr<ArchiveFile> OpenArchiveFile(const std::filesystem::path& path);
-
-// Retain this handle while a mount or metadata scan is active to reuse the index and block cache.
-// Accepts either a host archive filename or a virtual path inside the archive.
-[[nodiscard]] std::shared_ptr<ArchiveReader> OpenArchive(const std::filesystem::path& path);
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_ARCHIVE_H_ */

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

@@ -43,14 +43,6 @@ int32_t GetUserId() {
 	return g_config->user_id;
 }
 
-const std::string& GetAudioInputDevice() {
-	return g_config->audio_input_device;
-}
-
-const std::optional<ControllerColor>& GetControllerColor() {
-	return g_config->controller_color;
-}
-
 PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
@@ -63,14 +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);
 }
@@ -91,7 +75,7 @@ ShaderOptimizationType GetShaderOptimizationType() {
 	return g_config->shader_optimization_type;
 }
 
-LogDirection GetShaderLogDirection() {
+ShaderLogDirection GetShaderLogDirection() {
 	return g_config->shader_log_direction;
 }
 
@@ -111,7 +95,7 @@ bool GraphicsDebugDumpEnabled() {
 	return g_config->graphics_debug_dump_enabled;
 }
 
-LogDirection GetPrintfDirection() {
+OutputDirection GetPrintfDirection() {
 	return g_config->printf_direction;
 }
 
@@ -119,8 +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() {
@@ -139,10 +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

@@ -4,9 +4,7 @@
 #include "common/common.h"
 
 #include <cstddef>
-#include <array>
 #include <filesystem>
-#include <optional>
 #include <string>
 #include <vector>
 
@@ -23,12 +21,15 @@ 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 };
 
 using Keymap = std::vector<std::string>;
-using ControllerColor = std::array<uint8_t, 3>;
 
 constexpr uint32_t DEFAULT_CONSOLE_LANGUAGE = 1;
 constexpr uint32_t MAX_CONSOLE_LANGUAGE     = 29;
@@ -46,31 +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;
-	std::optional<ControllerColor> controller_color;
-	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;
@@ -84,20 +80,16 @@ uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
-const std::string& GetAudioInputDevice();
-const std::optional<ControllerColor>& GetControllerColor();
 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();
@@ -105,10 +97,10 @@ std::filesystem::path GetCommandBufferDumpFolder();
 
 bool GraphicsDebugDumpEnabled();
 
-LogDirection          GetPrintfDirection();
+OutputDirection       GetPrintfDirection();
 std::filesystem::path GetPrintfOutputFile();
 
-bool ProfilerEnabled();
+ProfilerDirection GetProfilerDirection();
 
 bool SpirvDebugPrintfEnabled();
 
@@ -116,7 +108,6 @@ bool GpuAssistedValidationEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
-bool TessellationEnabled();
 bool PlayGoHackEnabled();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();

src/common/file.cpp

@@ -1,6 +1,5 @@
 #include "common/file.h"
 
-#include "common/archive.h"
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/platform/sysFileIO.h"
@@ -62,8 +61,7 @@ SysFileTimeStruct DateTimeToFileTimeUtc(const DateTime& date_time) {
 } // namespace
 
 struct File::FilePrivate {
-	sys_file_t*                  f;
-	std::unique_ptr<ArchiveFile> archive;
+	sys_file_t* f;
 };
 
 File::File(): m_p(std::make_unique<FilePrivate>()) {
@@ -87,15 +85,11 @@ File::File(const std::filesystem::path& name, Mode mode): m_p(std::make_unique<F
 }
 
 bool File::IsInvalid() const {
-	return m_p->f == nullptr && m_p->archive == nullptr;
+	return m_p->f == nullptr;
 }
 
 bool File::Create(const std::filesystem::path& name) {
-	EXIT_IF(!IsInvalid());
-
-	if (IsArchivePath(name)) {
-		return false;
-	}
+	EXIT_IF(m_p->f != nullptr);
 
 	m_file_name = name;
 
@@ -110,16 +104,9 @@ bool File::Create(const std::filesystem::path& name) {
 }
 
 bool File::Open(const std::filesystem::path& name, Mode mode) {
-	EXIT_IF(!IsInvalid());
+	EXIT_IF(m_p->f != nullptr);
 
 	m_file_name = name;
-	if (IsArchivePath(name)) {
-		if (mode != Mode::Read) {
-			return false;
-		}
-		m_p->archive = OpenArchiveFile(name);
-		return m_p->archive != nullptr;
-	}
 
 	switch (mode) {
 		case Mode::Read: m_p->f = SysFileOpenR(name); break;
@@ -137,7 +124,7 @@ bool File::Open(const std::filesystem::path& name, Mode mode) {
 }
 
 bool File::OpenInMem(void* buf, uint32_t buf_size) {
-	EXIT_IF(!IsInvalid());
+	EXIT_IF(m_p->f != nullptr);
 
 	m_p->f = SysFileOpen(static_cast<uint8_t*>(buf), buf_size);
 
@@ -149,8 +136,12 @@ 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(!IsInvalid());
+	EXIT_IF(m_p->f != nullptr);
 
 	m_p->f = SysFileCreate();
 
@@ -163,7 +154,6 @@ bool File::CreateInMem() {
 }
 
 void File::Close() {
-	m_p->archive.reset();
 	if (m_p->f != nullptr) {
 		SysFileClose(m_p->f);
 		m_p->f = nullptr;
@@ -171,11 +161,7 @@ void File::Close() {
 }
 
 uint64_t File::Size() const {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return m_p->archive->Size();
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	if (m_p->f == nullptr) {
 		return 0;
@@ -185,76 +171,41 @@ uint64_t File::Size() const {
 }
 
 uint64_t File::Remaining() const {
-	EXIT_IF(IsInvalid());
+	EXIT_IF(m_p->f == nullptr);
 
-	const auto size     = Size();
-	const auto position = Tell();
-	return position < size ? size - position : 0;
+	return Size() - Tell();
 }
 
 uint64_t File::Size(const std::filesystem::path& name) {
-	if (IsArchivePath(name)) {
-		const auto info = GetArchiveInfo(name);
-		return info ? info->size : 0;
-	}
 	return SysFileSize(name);
 }
 
-std::optional<File::Info> File::GetInfo(const std::filesystem::path& name) {
-	if (IsArchivePath(name)) {
-		return GetArchiveInfo(name);
-	}
-	Info info {};
-	return SysFileGetInfo(name, &info.is_file, &info.size) ? std::optional(info) : std::nullopt;
-}
-
 bool File::Seek(uint64_t offset) {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return m_p->archive->Seek(offset);
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	return SysFileSeek(*m_p->f, offset);
 }
 
 bool File::Truncate(uint64_t size) {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return false;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	return SysFileTruncate(*m_p->f, size);
 }
 
 bool File::Unlink() {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return false;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	return SysFileUnlink(*m_p->f, m_file_name);
 }
 
 uint64_t File::Tell() const {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return m_p->archive->Tell();
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	return SysFileTell(*m_p->f);
 }
 
 void File::Read(void* data, uint32_t size, uint32_t* bytes_read) {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		m_p->archive->Read(data, size, bytes_read);
-		return;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	if (m_p->f != nullptr) {
 		SysFileRead(data, size, *m_p->f, bytes_read);
@@ -262,14 +213,7 @@ void File::Read(void* data, uint32_t size, uint32_t* bytes_read) {
 }
 
 void File::Write(const void* data, uint32_t size, uint32_t* bytes_written) {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		if (bytes_written != nullptr) {
-			*bytes_written = 0;
-		}
-		return;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	SysFileWrite(data, size, *m_p->f, bytes_written);
 }
@@ -305,120 +249,103 @@ 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) {
-	if (IsArchivePath(path)) {
-		const auto info = GetArchiveInfo(path);
-		return info && !info->is_file;
-	}
-	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) {
-	if (IsArchivePath(name)) {
-		const auto info = GetArchiveInfo(name);
-		return info && info->is_file;
-	}
 	return SysFileIsFileExisting(name);
 }
 
-bool File::CreateDirectory(const std::filesystem::path& path) {
-	if (IsArchivePath(path)) {
-		return false;
-	}
-    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) {
-	if (IsArchivePath(path)) {
-		return false;
-	}
-    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) {
-	if (IsArchivePath(path)) {
-		return false;
+	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+
+	std::vector<std::string> list = Common::Split(real_path, "/");
+
+	std::string p;
+
+	for (uint32_t si = 0; si < list.size(); si++) {
+		const std::string& s = list[si];
+
+		if (si != 0 || Common::StartsWith(real_path, "/")) {
+			p += "/";
+		}
+
+		p += s;
+
+		if (IsDirectoryExisting(p)) {
+			continue;
+		}
+
+		if (!CreateDirectory(p)) // @suppress("Invalid arguments")
+		{
+			return false;
+		}
 	}
-    const auto normalized_path =
-        WithoutTrailingSeparator(path);
 
-    auto current_path = normalized_path.root_path();
-
-    for (const auto& component:
-         normalized_path.relative_path()) {
-        current_path /= component;
-
-        if (IsDirectoryExisting(current_path)) {
-            continue;
-        }
-
-        if (!CreateDirectory(current_path)) {
-            return false;
-        }
-    }
-
-    return true;
+	return true;
 }
 
 bool File::DeleteDirectories(const std::filesystem::path& path) {
-	if (IsArchivePath(path)) {
-		return false;
+	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+
+	std::vector<std::string> list = Common::Split(real_path, "/");
+
+	std::string              p;
+	std::vector<std::string> list2;
+
+	for (uint32_t si = 0; si < list.size(); si++) {
+		const std::string& s = list[si];
+
+		if (si != 0 || Common::StartsWith(real_path, "/")) {
+			p += "/";
+		}
+
+		p += s;
+
+		list2.push_back(p);
 	}
-    const auto normalized_path =
-        WithoutTrailingSeparator(path);
 
-    std::vector<std::filesystem::path> directories;
-    auto current_path = normalized_path.root_path();
+	uint32_t num = list2.size();
 
-    for (const auto& component:
-         normalized_path.relative_path()) {
-        current_path /= component;
-        directories.push_back(current_path);
-    }
+	for (uint32_t si = num - 1; si < num; si--) {
+		if (!DeleteDirectory(list2[si])) {
+			return false;
+		}
+	}
 
-    for (auto it = directories.rbegin();
-         it != directories.rend(); ++it) {
-        if (!DeleteDirectory(*it)) {
-            return false;
-        }
-    }
-
-    return true;
+	return true;
 }
 
 bool File::DeleteFile(
     const std::filesystem::path& name) // @suppress("Member declaration not found")
 {
-	if (IsArchivePath(name)) {
-		return false;
-	}
 	return SysFileDeleteFile(name);
 }
 
 bool File::Flush() {
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		return true;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	return SysFileFlush(*m_p->f);
 }
 
-std::vector<std::byte> File::ReadWholeBuffer() {
+ByteBuffer File::ReadWholeBuffer() {
 	EXIT_IF(IsInvalid());
 	EXIT_IF(Tell() != 0);
 
@@ -426,18 +353,26 @@ 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) {
-	if (IsArchivePath(name)) {
-		return GetLastAccessTimeUTC(GetArchiveHostPath(name));
-	}
 	SysTimeStruct t {};
 	SysFileToSystemTimeUtc(SysFileGetLastAccessTimeUtc(name), t);
 
@@ -449,9 +384,6 @@ DateTime File::GetLastAccessTimeUTC(const std::filesystem::path& name) {
 }
 
 DateTime File::GetLastWriteTimeUTC(const std::filesystem::path& name) {
-	if (IsArchivePath(name)) {
-		return GetLastWriteTimeUTC(GetArchiveHostPath(name));
-	}
 	SysTimeStruct t {};
 	SysFileToSystemTimeUtc(SysFileGetLastWriteTimeUtc(name), t);
 
@@ -466,10 +398,6 @@ void File::GetLastAccessAndWriteTimeUTC(const std::filesystem::path& name, DateT
                                         DateTime* write) {
 	EXIT_IF(access == nullptr);
 	EXIT_IF(write == nullptr);
-	if (IsArchivePath(name)) {
-		GetLastAccessAndWriteTimeUTC(GetArchiveHostPath(name), access, write);
-		return;
-	}
 
 	SysTimeStruct     at {};
 	SysTimeStruct     wt {};
@@ -496,12 +424,7 @@ void File::GetLastAccessAndWriteTimeUTC(DateTime* access, DateTime* write) {
 	EXIT_IF(access == nullptr);
 	EXIT_IF(write == nullptr);
 
-	EXIT_IF(IsInvalid());
-
-	if (m_p->archive != nullptr) {
-		GetLastAccessAndWriteTimeUTC(m_file_name, access, write);
-		return;
-	}
+	EXIT_IF(m_p->f == nullptr);
 
 	if (m_p->f == nullptr) {
 		*access = DateTime();
@@ -531,35 +454,68 @@ void File::GetLastAccessAndWriteTimeUTC(DateTime* access, DateTime* write) {
 }
 
 bool File::SetLastAccessTimeUTC(const std::filesystem::path& name, const DateTime& dt) {
-	if (IsArchivePath(name)) {
-		return false;
-	}
 	auto f = DateTimeToFileTimeUtc(dt);
 	return SysFileSetLastAccessTimeUtc(name, f);
 }
 
 bool File::SetLastWriteTimeUTC(const std::filesystem::path& name, const DateTime& dt) {
-	if (IsArchivePath(name)) {
-		return false;
-	}
 	auto f = DateTimeToFileTimeUtc(dt);
 	return SysFileSetLastWriteTimeUtc(name, f);
 }
 
 bool File::SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name, const DateTime& access,
                                         const DateTime& write) {
-	if (IsArchivePath(name)) {
-		return false;
-	}
 	auto af = DateTimeToFileTimeUtc(access);
 	auto wf = DateTimeToFileTimeUtc(write);
 	return SysFileSetLastAccessAndWriteTimeUtc(name, af, wf);
 }
 
-std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& path) {
-	if (IsArchivePath(path)) {
-		return GetArchiveDirEntries(path);
+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;
 
 	SysFileGetDents(path, files);
@@ -567,8 +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;
@@ -577,18 +538,12 @@ std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& pat
 bool File::CopyFile(const std::filesystem::path& src,
                     const std::filesystem::path& dst) // @suppress("Member declaration not found")
 {
-	if (IsArchivePath(src) || IsArchivePath(dst)) {
-		return false;
-	}
 	return SysFileCopyFile(src, dst);
 }
 
 bool File::RenameFile(const std::filesystem::path& src,
                       const std::filesystem::path& dst) // @suppress("Member declaration not found")
 {
-	if (IsArchivePath(src) || IsArchivePath(dst)) {
-		return false;
-	}
 	if (IsFileExisting(dst)) {
 		DeleteFile(dst); // @suppress("Invalid arguments")
 	}
@@ -597,9 +552,6 @@ bool File::RenameFile(const std::filesystem::path& src,
 }
 
 void File::RemoveReadonly(const std::filesystem::path& name) {
-	if (IsArchivePath(name)) {
-		return;
-	}
 	SysFileRemoveReadonly(name);
 }
 

src/common/file.h

@@ -1,13 +1,12 @@
 #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 <optional>
 #include <string>
 #include <vector>
 
@@ -31,14 +30,18 @@ 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;
 	};
-	struct Info {
-		bool     is_file;
-		uint64_t size;
-	};
 
 	File();
 	explicit File(const std::filesystem::path& name);
@@ -48,6 +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();
@@ -69,15 +73,16 @@ 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);
-	static std::optional<Info> GetInfo(const std::filesystem::path& name);
 
 	static bool IsDirectoryExisting(const std::filesystem::path& path);
 	static bool IsFileExisting(const std::filesystem::path& name);
@@ -96,6 +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;
@@ -44,8 +52,8 @@ 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        SysFileGetInfo(const std::filesystem::path& name, bool* is_file, uint64_t* size);
 bool        SysFileIsError(sys_file_t& f); // NOLINT(google-runtime-references)
 bool        SysFileIsDirectoryExisting(const std::filesystem::path& path);
 bool        SysFileIsFileExisting(const std::filesystem::path& name);
@@ -68,6 +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>
@@ -54,7 +53,7 @@ struct sys_file_t {
 #endif
 
 static std::filesystem::path get_internal_name(const std::filesystem::path& name) {
-	return name.empty() || name.is_absolute() ? name : (std::filesystem::path(".") / name);
+	return name.is_absolute() ? name : (std::filesystem::path(".") / name);
 }
 
 // Pass access-pattern hints to the host.
@@ -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;
 
@@ -297,21 +300,10 @@ uint64_t SysFileSize(sys_file_t& f) {
 	return 0;
 }
 
-bool SysFileGetInfo(const std::filesystem::path& name, bool* is_file, uint64_t* size) {
-	const auto path = get_internal_name(name);
-	struct stat info {};
-	if (stat(path.c_str(), &info) != 0) {
-		return false;
-	}
-	*is_file = !S_ISDIR(info.st_mode);
-	*size = *is_file ? static_cast<uint64_t>(info.st_size) : 0;
-	return true;
-}
-
 uint64_t SysFileSize(const std::filesystem::path& file_name) {
-	bool is_file;
-	uint64_t size = 0;
-	SysFileGetInfo(file_name, &is_file, &size);
+	sys_file_t* f    = SysFileOpenR(file_name);
+	uint64_t    size = SysFileSize(*f);
+	SysFileClose(f);
 	return size;
 }
 
@@ -364,15 +356,29 @@ bool SysFileIsError(sys_file_t& f) {
 }
 
 bool SysFileIsDirectoryExisting(const std::filesystem::path& path) {
-	bool is_file;
-	uint64_t size;
-	return SysFileGetInfo(path, &is_file, &size) && !is_file;
+	auto real_name     = get_internal_name(path);
+	auto real_name_str = real_name.string();
+
+	struct stat s {};
+
+	if (0 != stat(real_name_str.c_str(), &s)) {
+		return false;
+	}
+
+	return S_ISDIR(s.st_mode); // NOLINT
 }
 
 bool SysFileIsFileExisting(const std::filesystem::path& name) {
-	bool is_file;
-	uint64_t size;
-	return SysFileGetInfo(name, &is_file, &size) && is_file;
+	auto real_name     = get_internal_name(name);
+	auto real_name_str = real_name.string();
+
+	struct stat s {};
+
+	if (0 != stat(real_name_str.c_str(), &s)) {
+		return false;
+	}
+
+	return !S_ISDIR(s.st_mode); // NOLINT
 }
 
 bool SysFileCreateDirectory(const std::filesystem::path& path) {
@@ -607,6 +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;
 
@@ -266,27 +269,19 @@ uint64_t SysFileSize(sys_file_t& f) {
 	return 0;
 }
 
-bool SysFileGetInfo(const std::filesystem::path& name, bool* is_file, uint64_t* size) {
-	const bool directory_path = !name.empty() && !name.has_filename() && name != name.root_path();
-	WIN32_FILE_ATTRIBUTE_DATA info {};
-	if (GetFileAttributesExW(directory_path ? name.parent_path().c_str() : name.c_str(),
-	                         GetFileExInfoStandard, &info) == 0) {
-		return false;
-	}
-	const bool file = (info.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) == 0;
-	if (directory_path && file) {
-		return false;
-	}
-	*is_file = file;
-	*size = *is_file ? ((static_cast<uint64_t>(info.nFileSizeHigh) << 32u) | info.nFileSizeLow) : 0;
-	return true;
-}
-
 uint64_t SysFileSize(const std::filesystem::path& file_name) {
-	bool is_file;
-	uint64_t size = 0;
-	SysFileGetInfo(file_name, &is_file, &size);
-	return size;
+	LARGE_INTEGER             s;
+	WIN32_FILE_ATTRIBUTE_DATA a;
+
+	auto wide = file_name.wstring();
+	if (GetFileAttributesExW(wide.c_str(), GetFileExInfoStandard, &a) == 0) {
+		return 0;
+	}
+
+	s.HighPart = static_cast<LONG>(a.nFileSizeHigh);
+	s.LowPart  = a.nFileSizeLow;
+
+	return s.QuadPart;
 }
 
 bool SysFileTruncate(sys_file_t& f, uint64_t size) {
@@ -354,15 +349,17 @@ bool SysFileIsError(sys_file_t& f) {
 }
 
 bool SysFileIsDirectoryExisting(const std::filesystem::path& path) {
-	bool is_file;
-	uint64_t size;
-	return SysFileGetInfo(path, &is_file, &size) && !is_file;
+	auto  wide = path.wstring();
+	DWORD a    = GetFileAttributesW(wide.c_str());
+	return a != INVALID_FILE_ATTRIBUTES &&
+	       ((a & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) != 0u);
 }
 
 bool SysFileIsFileExisting(const std::filesystem::path& name) {
-	bool is_file;
-	uint64_t size;
-	return SysFileGetInfo(name, &is_file, &size) && is_file;
+	auto  wide = name.wstring();
+	DWORD a    = GetFileAttributesW(wide.c_str());
+	return a != INVALID_FILE_ATTRIBUTES &&
+	       ((a & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) == 0u);
 }
 
 bool SysFileCreateDirectory(const std::filesystem::path& path) {
@@ -482,11 +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;
@@ -500,7 +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

@@ -2,13 +2,52 @@
 
 #include "common/emulatorConfig.h"
 
+#include <algorithm>
 #include <common/TracyProtocol.hpp>
 #include <common/TracyVersion.hpp>
 #include <cstdio>
 #include <tracy/Tracy.hpp>
+#include <vector>
+
+namespace {
+
+thread_local std::vector<Profiler::ScopedBlock*> g_block_stack;
+
+void RemoveBlock(Profiler::ScopedBlock* block) {
+	auto block_it = std::find(g_block_stack.rbegin(), g_block_stack.rend(), block);
+	if (block_it != g_block_stack.rend()) {
+		g_block_stack.erase(std::next(block_it).base());
+	}
+}
+
+} // namespace
 
 namespace Profiler {
 
+ScopedBlock::ScopedBlock(const tracy::SourceLocationData* source_location) {
+	if (tracy::ProfilerAvailable()) {
+		m_zone.emplace(source_location, TRACY_CALLSTACK, true);
+		g_block_stack.push_back(this);
+	}
+}
+
+ScopedBlock::~ScopedBlock() {
+	End();
+}
+
+void ScopedBlock::End() {
+	if (m_zone.has_value()) {
+		m_zone.reset();
+		RemoveBlock(this);
+	}
+}
+
+void EndBlock() {
+	if (!g_block_stack.empty()) {
+		g_block_stack.back()->End();
+	}
+}
+
 void SetThreadName(const char* name) {
 	if (tracy::ProfilerAvailable() && name != nullptr) {
 		tracy::SetThreadName(name);
@@ -16,13 +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/profiler.h

@@ -4,6 +4,7 @@
 #include "common/common.h"
 
 #include <cstdint>
+#include <optional>
 #include <tracy/Tracy.hpp> // IWYU pragma: export
 
 namespace profiler::colors {
@@ -22,6 +23,22 @@ inline constexpr uint32_t DeepOrangeA200 = 0xff6e40;
 
 namespace Profiler {
 
+class ScopedBlock {
+public:
+	explicit ScopedBlock(const tracy::SourceLocationData* source_location);
+	ScopedBlock(const ScopedBlock&)            = delete;
+	ScopedBlock& operator=(const ScopedBlock&) = delete;
+	ScopedBlock(ScopedBlock&&)                 = delete;
+	ScopedBlock& operator=(ScopedBlock&&)      = delete;
+	~ScopedBlock();
+
+	void End();
+
+private:
+	std::optional<tracy::ScopedZone> m_zone;
+};
+
+void EndBlock();
 void SetThreadName(const char* name);
 
 void Initialize();
@@ -49,11 +66,19 @@ struct Lifecycle {
 	    kyty_profiler_source_location_, line) {name, TracyFunction, TracyFile,                     \
 	                                           static_cast<uint32_t>(line),                        \
 	                                           KYTY_PROFILER_COLOR_OR_DEFAULT(0, __VA_ARGS__)};    \
-	tracy::ScopedZone KYTY_PROFILER_CONCAT(kyty_profiler_block_, line)(                           \
-	    &KYTY_PROFILER_CONCAT(kyty_profiler_source_location_, line),                              \
-	    TRACY_CALLSTACK, tracy::ProfilerAvailable())
+	Profiler::ScopedBlock KYTY_PROFILER_CONCAT(kyty_profiler_block_, line)(                        \
+	    &KYTY_PROFILER_CONCAT(kyty_profiler_source_location_, line))
 
-#define KYTY_PROFILER_FUNCTION(...) KYTY_PROFILER_BLOCK(nullptr __VA_OPT__(, ) __VA_ARGS__)
+#define KYTY_PROFILER_FUNCTION(...) KYTY_PROFILER_FUNCTION_IMPL(__LINE__ __VA_OPT__(, ) __VA_ARGS__)
+#define KYTY_PROFILER_FUNCTION_IMPL(line, ...)                                                     \
+	static constexpr tracy::SourceLocationData KYTY_PROFILER_CONCAT(                               \
+	    kyty_profiler_source_location_, line) {nullptr, TracyFunction, TracyFile,                  \
+	                                           static_cast<uint32_t>(line),                        \
+	                                           KYTY_PROFILER_COLOR_OR_DEFAULT(0, __VA_ARGS__)};    \
+	Profiler::ScopedBlock KYTY_PROFILER_CONCAT(kyty_profiler_block_, line)(                        \
+	    &KYTY_PROFILER_CONCAT(kyty_profiler_source_location_, line))
+
+#define KYTY_PROFILER_END_BLOCK Profiler::EndBlock()
 
 #define KYTY_PROFILER_THREAD(name) Profiler::SetThreadName(name)
 

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

@@ -2,11 +2,13 @@
 
 #include "common/assert.h"
 
+#include <algorithm>
 #include <atomic>
 #include <cerrno>
 #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
@@ -25,7 +27,8 @@
 #ifdef KYTY_WIN_CS
 #include <windows.h> // IWYU pragma: keep
 // IWYU pragma: no_include <winbase.h>
-constexpr DWORD KYTY_CS_SPIN_COUNT = 4000;
+constexpr DWORD    KYTY_CS_SPIN_COUNT          = 4000;
+constexpr uint64_t KYTY_SLEEP_SPIN_LIMIT_100NS = 500; // 50 us
 
 #ifndef CREATE_WAITABLE_TIMER_HIGH_RESOLUTION
 #define CREATE_WAITABLE_TIMER_HIGH_RESOLUTION 0x00000002
@@ -36,6 +39,30 @@ static void SleepHighResolution100ns(uint64_t units_100ns) {
 		return;
 	}
 
+	// Keep spinning only where a kernel transition is
+	// likely to cost more than the requested delay; ordinary millisecond sleeps use the
+	// per-thread high-resolution waitable timer below.
+	if (units_100ns <= KYTY_SLEEP_SPIN_LIMIT_100NS) {
+		LARGE_INTEGER frequency {};
+		LARGE_INTEGER start {};
+		if (QueryPerformanceFrequency(&frequency) != 0 && QueryPerformanceCounter(&start) != 0 &&
+		    frequency.QuadPart > 0) {
+			const auto wait_ticks =
+			    static_cast<LONGLONG>((static_cast<long double>(units_100ns) *
+			                           static_cast<long double>(frequency.QuadPart)) /
+			                          10000000.0L);
+			const auto    deadline = start.QuadPart + std::max<LONGLONG>(wait_ticks, 1);
+			LARGE_INTEGER now {};
+			do {
+				if (QueryPerformanceCounter(&now) == 0) {
+					break;
+				}
+				YieldProcessor();
+			} while (now.QuadPart < deadline);
+			return;
+		}
+	}
+
 	thread_local HANDLE timer = CreateWaitableTimerExW(
 	    nullptr, nullptr, CREATE_WAITABLE_TIMER_HIGH_RESOLUTION, TIMER_ALL_ACCESS);
 	if (timer == nullptr) {
@@ -105,13 +132,14 @@ static SleepConditionVariableCS_func_t ResolveSleepConditionVariableCS() {
 #endif
 
 #ifdef KYTY_POSIX_HIGH_RES_SLEEP
-// An absolute deadline preserves the requested sleep across signal interruptions.
+// Spin for very short waits; use an absolute deadline for longer waits.
 static void SleepHighResolutionNanos(uint64_t nanos) {
 	if (nanos == 0) {
 		return;
 	}
 
 	constexpr uint64_t NANOS_PER_SEC = 1000000000;
+	constexpr uint64_t SPIN_LIMIT_NS = 50000; // below this a context switch dominates
 
 	timespec deadline {};
 	if (clock_gettime(CLOCK_MONOTONIC, &deadline) != 0) {
@@ -123,6 +151,17 @@ static void SleepHighResolutionNanos(uint64_t nanos) {
 	deadline.tv_sec += static_cast<time_t>(target_nsec / NANOS_PER_SEC);
 	deadline.tv_nsec = static_cast<long>(target_nsec % NANOS_PER_SEC);
 
+	if (nanos <= SPIN_LIMIT_NS) {
+		timespec now {};
+		do {
+			if (clock_gettime(CLOCK_MONOTONIC, &now) != 0) {
+				return;
+			}
+		} while (now.tv_sec < deadline.tv_sec ||
+		         (now.tv_sec == deadline.tv_sec && now.tv_nsec < deadline.tv_nsec));
+		return;
+	}
+
 	while (clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, &deadline, nullptr) == EINTR) {
 	}
 }
@@ -158,7 +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) {}
@@ -215,6 +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);
@@ -292,6 +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
@@ -332,6 +406,7 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
+	UnregisterCondWaiter(m_cond_var.get());
 }
 
 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -340,6 +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
@@ -354,6 +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;
 }
 
@@ -368,13 +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;
 };
@@ -62,6 +61,7 @@ public:
 
 	void            BufferInit();
 	void            BufferFlush();
+	void            BufferFlushAndWait();
 	void            BufferWait();
 	HW::Context&    GetCtx() { return m_ctx; }
 	HW::UserConfig& GetUcfg() { return m_ucfg; }
@@ -72,9 +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);
@@ -101,7 +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);
 
@@ -132,7 +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; }
 
@@ -146,10 +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/pm4Dispatch.cpp

@@ -140,8 +140,6 @@ constexpr auto MakeShaderDispatchTable() {
 	g_hw_sh_func[Pm4::COMPUTE_PGM_RSRC3]               = HwShSetCsRegisters;
 	g_hw_sh_func[Pm4::COMPUTE_PACE_ID]                 = HwShSetCsRegisters;
 	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_PS]  = HwShSetRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_PS] = HwShSetRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_PS] = HwShSetRegisters;
 	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS]  = HwShSetRegisters;
 	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = HwShSetRegisters;
 	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = HwShSetRegisters;

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);
 		}
 
@@ -2266,6 +2271,7 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 	const bool gl2_writeback = ((gcr_cntl & GcrGl2Writeback) != 0);
 
 	auto trigger_interrupt = [&]() {
+		bool queued = false;
 		switch (interrupt_selector) {
 			case 0x00:
 			case 0x03: break;
@@ -2273,10 +2279,13 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			case 0x02:
 			case 0x04:
 				cp.TriggerEopEventAtEndOfPipe(interrupt_context_id);
-				cp.BufferFlush();
+				queued = true;
 				break;
 			default: EXIT("unknown release_mem interrupt selector\n");
 		}
+		if (queued) {
+			cp.BufferFlush();
+		}
 	};
 
 	if (data_sel == 0 || interrupt_selector == 4) {
@@ -2314,9 +2323,7 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 		cp.WriteAtEndOfPipe32(cache_policy, event_write_dest, eop_event_type, cache_action,
 		                      event_index, event_source, dst_gpu_addr, static_cast<uint32_t>(value),
 		                      interrupt_selector, interrupt_context_id);
-		if (interrupt_selector == 0x01 || interrupt_selector == 0x02) {
-			cp.BufferFlush();
-		}
+		cp.BufferFlush();
 
 		return 7;
 	}
@@ -3622,29 +3629,12 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_PS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_PS;
-	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_PS; offset++) {
-		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-			cp.GetShCtx().SetPsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_PS,
-			                                   value);
-		};
-	}
-	for (uint32_t offset = 4; offset <= 5; offset++) {
-		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-			cp.GetShCtx().SetPsAuxiliaryTableAddress(cmd_offset - 4, value);
-		};
-	}
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PACE_ID_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
 	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);
 	};
@@ -3664,13 +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;
@@ -3829,25 +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

@@ -35,6 +35,7 @@ namespace Libs::Graphics {
 
 static thread_local CommandProcessor* g_current_processor = nullptr;
 static thread_local Pm4Execution*     g_current_execution = nullptr;
+static thread_local bool              g_gpu_mutex_owned   = false;
 static thread_local bool              g_gpu_thread        = false;
 static thread_local GuestGpu*         g_gpu_state         = nullptr;
 
@@ -53,16 +54,36 @@ struct DrawIndexedIndirectArgs {
 	uint32_t start_instance_location;
 };
 
+class GpuMutexLock final {
+public:
+	explicit GpuMutexLock(Common::Mutex& mutex): m_mutex(mutex) {
+		if (g_gpu_mutex_owned) {
+			EXIT("recursive GPU mutex acquisition\n");
+		}
+		g_gpu_mutex_owned = true;
+		m_mutex.Lock();
+	}
+	~GpuMutexLock() {
+		if (!g_gpu_mutex_owned) {
+			EXIT("invalid GPU mutex release\n");
+		}
+		m_mutex.Unlock();
+		g_gpu_mutex_owned = false;
+	}
+
+private:
+	Common::Mutex& m_mutex;
+};
+
 static bool GraphicsRunDebugDumpEnabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
-	       Config::GetPrintfDirection() != Config::LogDirection::Silent;
+	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
 }
 
 GuestGpu::GuestGpu(RenderContext& renderer): m_renderer(renderer) {
 	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
 	GraphicsInitJmpTables();
 	m_gfx_cp = std::make_unique<CommandProcessor>(renderer, 0);
-	m_gfx_cp->Reset();
 	m_thread = std::jthread(ThreadRun, this);
 }
 
@@ -139,16 +160,20 @@ void GuestGpu::Submit(std::span<const uint32_t> draw_commands,
 	if (draw_commands.empty()) {
 		return;
 	}
-	Submission submission;
+	GpuMutexLock lock(m_submission_mutex);
+	Submission   submission;
 	submission.type              = SubmissionType::Graphics;
 	submission.queue_id          = 0;
 	submission.commands          = draw_commands;
 	submission.constant_commands = constant_commands;
+	submission.reset_processor   = m_graphics_done;
+	m_graphics_done              = false;
 	Enqueue(std::move(submission));
 }
 
 void GuestGpu::SubmitCompute(uint32_t queue, std::span<const uint32_t> commands) {
 	EXIT_IF(commands.empty());
+	GpuMutexLock lock(m_submission_mutex);
 
 	EXIT_NOT_IMPLEMENTED(queue < ComputeQueueBase || queue >= ComputeQueueBase + ComputeQueueCount);
 
@@ -161,21 +186,22 @@ void GuestGpu::SubmitCompute(uint32_t queue, std::span<const uint32_t> commands)
 }
 
 void GuestGpu::SubmitFlipPreparation(uint64_t request_id) {
-	Submission submission;
+	GpuMutexLock lock(m_submission_mutex);
+	Submission   submission;
 	submission.type            = SubmissionType::FlipPreparation;
 	submission.queue_id        = 0;
+	submission.reset_processor = m_graphics_done;
 	submission.flip_request_id = request_id;
+	m_graphics_done            = false;
 	Enqueue(std::move(submission));
 }
 
-void GuestGpu::SuspendPoint() {
-	EXIT_IF(IsGpuThread() || CommandScheduler::InDeferredOperation());
-	// Do not hold a queue lock while waiting: asynchronous work may be needed to
-	// finish the preceding graphics frame. The first point returns immediately.
-	m_suspend_point_ready->acquire();
-	Submission submission;
-	submission.type = SubmissionType::SuspendPoint;
-	Enqueue(std::move(submission));
+void GuestGpu::Done() {
+	GpuMutexLock lock(m_submission_mutex);
+	if (!IsGpuThread()) {
+		WaitForIdle();
+	}
+	m_graphics_done = true;
 	m_done_num++;
 }
 
@@ -203,9 +229,6 @@ void CommandProcessor::Reset() {
 	m_context_state_pushed             = false;
 	m_index_type_and_size              = 0;
 	m_index_buffer_size                = 0;
-	m_index_base_addr                  = 0;
-	m_num_instances                    = 1;
-	m_predicate_skip                   = false;
 	m_user_data_marker                 = HW::UserSgprType::Unknown;
 	m_draw_indirect_args_base_addr     = 0;
 	m_dispatch_indirect_args_base_addr = 0;
@@ -245,6 +268,10 @@ void CommandProcessor::BufferFlush() {
 	GetScheduler().Flush();
 }
 
+void CommandProcessor::BufferFlushAndWait() {
+	GetScheduler().FlushAndWait();
+}
+
 void CommandProcessor::BufferWait() {
 	BufferInit();
 	GetScheduler().Finish();
@@ -359,12 +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,
@@ -404,7 +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,
@@ -432,7 +455,6 @@ void GuestGpu::Enqueue(Submission submission) {
 }
 
 void GuestGpu::WaitForIdle() {
-	EXIT_IF(IsGpuThread() || CommandScheduler::InDeferredOperation());
 	Common::LockGuard lock(m_queue_mutex);
 	while (m_processing || !m_commands.empty() || m_submission_count != 0) {
 		m_idle.Wait(&m_queue_mutex);
@@ -538,7 +560,11 @@ void GuestGpu::ThreadRun(void* data) {
 
 bool GuestGpu::Process(Submission& submission) {
 	const bool first_slice = !submission.started;
-	auto&      cp          = GetProcessor(submission.queue_id);
+	auto& cp = GetProcessor(submission.queue_id);
+
+	if (first_slice && submission.reset_processor) {
+		cp.Reset();
+	}
 
 	if (first_slice) {
 		submission.started = true;
@@ -577,11 +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;
 		}
@@ -603,25 +629,18 @@ 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;
-		case SubmissionType::SuspendPoint:
-			cp.EmitGlobalBarrier();
-			m_renderer.GetCommandScheduler().DeferPriorityOperation(
-			    [ready = m_suspend_point_ready] { ready->release(); });
-			cp.BufferFlush();
-			cp.Reset();
-			break;
 	}
 
 	return complete;
@@ -653,16 +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() {
@@ -670,13 +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;
@@ -752,19 +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 = {};
-		}
 	}
 }
 
@@ -799,61 +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:
-			// The wait selector applies only to Z-pass query readiness.
-			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);
 	}
@@ -885,6 +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,
@@ -896,6 +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) {
@@ -951,22 +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,
@@ -977,6 +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);
@@ -1009,8 +1061,12 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 	m_sh_ctx.SetCsWaveSize(Pm4::ComputeWaveSize(mode));
 
 	uint32_t frame_num = 0;
+	// uint32_t local_x   = 1;
+	// uint32_t local_y   = 1;
+	// uint32_t local_z   = 1;
 
 	{
+		CheckBuffer();
 		frame_num = m_renderer.GetGpu().GetFrameNum();
 		if (GraphicsRunDebugDumpEnabled()) {
 			static std::atomic<uint32_t> log_count {0};
@@ -1030,23 +1086,66 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 			}
 		}
 
+		const auto& cs = m_sh_ctx.GetCs().cs_regs;
+		// local_x        = std::max(cs.num_thread_x, 1u);
+		// local_y        = std::max(cs.num_thread_y, 1u);
+		// local_z        = std::max(cs.num_thread_z, 1u);
+		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);
 	}
+
+	/*constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
+	auto               group_count = [](uint32_t threads, uint32_t group_size) {
+	    return (threads == 0
+	                ? 0u
+	                : (threads + std::max(group_size, 1u) - 1u) / std::max(group_size, 1u));
+	};
+
+	auto groups_x = thread_group_x;
+	auto groups_y = thread_group_y;
+	auto groups_z = thread_group_z;
+	if ((mode & DispatchInitiatorUseThreadDimensions) != 0) {
+	    groups_x = group_count(thread_group_x, local_x);
+	    groups_y = group_count(thread_group_y, local_y);
+	    groups_z = group_count(thread_group_z, local_z);
+	}
+
+	const uint64_t invocations =
+	    static_cast<uint64_t>(groups_x) * groups_y * groups_z * local_x * local_y * local_z;
+	if (invocations != 0) {
+	    BufferFlushAndWait();
+	}*/
 }
 
-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;
 	}
@@ -1068,7 +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);
@@ -1099,7 +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;
 			}
@@ -1116,17 +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);
 		}
 	};
 
@@ -1138,7 +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);
@@ -1151,9 +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;
@@ -1161,9 +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));
@@ -1171,18 +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);
 					}
 				};
 
@@ -1197,7 +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;
@@ -1205,6 +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);
@@ -1220,7 +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;
@@ -1252,6 +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;
 	}
 
@@ -1281,6 +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 {};
@@ -1297,6 +1433,8 @@ void CommandProcessor::EmitGlobalBarrier() {
 }
 
 void CommandProcessor::TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id) {
+	CheckBuffer();
+
 	Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id, interrupt_context_id);
 }
 
@@ -1380,6 +1518,8 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index,
 }
 
 void CommandProcessor::Flip() {
+	CheckBuffer();
+
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n");
 	}
@@ -1393,7 +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"
@@ -1403,6 +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),
@@ -1413,7 +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"
@@ -1428,6 +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(
@@ -1437,7 +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");
 	}
@@ -1447,10 +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/graphicsRun.h

@@ -13,7 +13,6 @@
 #include <deque>
 #include <memory>
 #include <mutex>
-#include <semaphore>
 #include <span>
 #include <thread>
 
@@ -33,14 +32,11 @@ public:
 	void               SendCommandSync(Common::UniqueFunction<void>&& command);
 
 	// Submitted command memory is borrowed and must remain valid until GPU execution completes.
-	void Submit(std::span<const uint32_t> draw_commands,
-	            std::span<const uint32_t> constant_commands);
-	void SubmitCompute(uint32_t queue, std::span<const uint32_t> commands);
-	void SubmitFlipPreparation(uint64_t request_id);
-	// Insert an ordered graphics drain; only a previous suspend point can block the caller.
-	void SuspendPoint();
-	// Wait for guest command processing, including all compute queues (not native GPU completion).
-	void              WaitForIdle();
+	void              Submit(std::span<const uint32_t> draw_commands,
+	                         std::span<const uint32_t> constant_commands);
+	void              SubmitCompute(uint32_t queue, std::span<const uint32_t> commands);
+	void              SubmitFlipPreparation(uint64_t request_id);
+	void              Done();
 	[[nodiscard]] int GetFrameNum() const;
 
 	[[nodiscard]] static bool IsGpuThread() noexcept;
@@ -52,7 +48,7 @@ private:
 	static constexpr uint32_t ComputeQueueBase     = 0x20;
 	static constexpr uint32_t QueueCount           = 1 + ComputeQueueCount;
 
-	enum class SubmissionType { Graphics, Compute, FlipPreparation, SuspendPoint };
+	enum class SubmissionType { Graphics, Compute, FlipPreparation };
 
 	struct Submission {
 		SubmissionType            type     = SubmissionType::Graphics;
@@ -61,6 +57,7 @@ private:
 		std::span<const uint32_t> constant_commands;
 		Pm4Execution              command_execution;
 		Pm4Execution              constant_execution;
+		bool                      reset_processor   = false;
 		bool                      started           = false;
 		bool                      command_complete  = false;
 		bool                      constant_complete = false;
@@ -69,12 +66,14 @@ private:
 	};
 
 	void              Enqueue(Submission submission);
+	void              WaitForIdle();
 	void              ProcessCommands();
 	bool              Process(Submission& submission);
 	static void       ThreadRun(void* data);
 	CommandProcessor& GetProcessor(uint32_t queue_id);
 
 	RenderContext&                                 m_renderer;
+	Common::Mutex                                  m_submission_mutex;
 	Common::Mutex                                  m_queue_mutex;
 	std::mutex                                     m_shutdown_mutex;
 	Common::CondVar                                m_work_available;
@@ -85,12 +84,10 @@ private:
 	uint32_t                                       m_next_queue        = 0;
 	uint32_t                                       m_submission_count  = 0;
 	bool                                           m_processing        = false;
+	bool                                           m_graphics_done     = true;
 	bool                                           m_accepting         = true;
 	bool                                           m_stopping          = false;
 	bool                                           m_shutdown_complete = false;
-	// Completion callbacks can outlive GuestGpu during renderer shutdown.
-	std::shared_ptr<std::binary_semaphore> m_suspend_point_ready =
-	    std::make_shared<std::binary_semaphore>(1);
 
 	std::unique_ptr<CommandProcessor>                                m_gfx_cp;
 	std::array<std::unique_ptr<CommandProcessor>, ComputeQueueCount> m_compute_cp;

src/graphics/guest_gpu/hardwareContext.h

@@ -458,9 +458,6 @@ struct PsShaderResource2 {
 
 struct PsStageRegisters {
 	uint64_t          data_addr = 0;
-	uint64_t          user_data_addr = 0;
-	// Hades II uses PS offsets 4:5 for a byte-addressed descriptor table.
-	uint64_t          auxiliary_table_addr = 0;
 	PsShaderResource1 rsrc1;
 	PsShaderResource2 rsrc2;
 };
@@ -501,8 +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;
 };
@@ -939,12 +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;
@@ -974,17 +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;
@@ -1007,25 +988,19 @@ 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; }
 
 	void SetPsShaderBase(uint64_t addr) { m_ps.ps_regs.data_addr = addr; }
-	void SetPsUserDataAddress(uint32_t word, uint32_t value) {
-		SetUserDataAddressWord(m_ps.ps_regs.user_data_addr, word, value);
-	}
-	void SetPsAuxiliaryTableAddress(uint32_t word, uint32_t value) {
-		SetUserDataAddressWord(m_ps.ps_regs.auxiliary_table_addr, word, value);
-	}
 	void SetPsShaderResource1(const PsShaderResource1& rsrc1) { m_ps.ps_regs.rsrc1 = rsrc1; }
 	void SetPsShaderResource2(const PsShaderResource2& rsrc2) { m_ps.ps_regs.rsrc2 = rsrc2; }
 
@@ -1062,12 +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

@@ -697,8 +697,6 @@ constexpr uint32_t CX_NUM = 0x3FF + 1;
 
 /* Shader registers */
 
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_PS = 0x2;
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_HI_PS = 0x3;
 constexpr uint32_t SPI_SHADER_PACE_ID_PS            = 0x6;
 constexpr uint32_t SPI_GRAPHICS_SHADER_CONTROL_PS    = 0x7;
 constexpr uint32_t SPI_SHADER_PGM_LO_PS              = 0x8;
@@ -912,7 +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;
 
@@ -990,12 +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/regionDefinitions.h

@@ -3,7 +3,6 @@
 
 #include "common/bitArray.h"
 #include "common/common.h"
-#include "kernel/memory.h"
 
 #include <compare>
 
@@ -11,8 +10,7 @@ namespace Libs::Graphics {
 
 constexpr uint64_t TRACKER_PAGE_SIZE    = 4ull * 1024ull;
 constexpr uint64_t TRACKER_REGION_SIZE  = 4ull * 1024ull * 1024ull;
-constexpr uint64_t TRACKER_ADDRESS_SIZE = 1ull << 44u;
-constexpr uint64_t LOWER_ADDRESS_SIZE   = 1ull << 40u;
+constexpr uint64_t TRACKER_ADDRESS_SIZE = 1ull << 40u;
 constexpr size_t   TRACKER_REGION_PAGES = TRACKER_REGION_SIZE / TRACKER_PAGE_SIZE;
 
 struct GuestRange {
@@ -21,11 +19,8 @@ struct GuestRange {
 
 	[[nodiscard]] constexpr bool Empty() const noexcept { return address == 0 && size == 0; }
 	[[nodiscard]] constexpr bool Valid() const noexcept {
-		if (address == 0 || size == 0) return false;
-		if (address < LOWER_ADDRESS_SIZE) return size <= LOWER_ADDRESS_SIZE - address;
-		constexpr auto base = LibKernel::Memory::kExtendedMemoryBase;
-		constexpr auto length = LibKernel::Memory::kExtendedMemorySize;
-		return address >= base && address - base < length && size <= length - (address - base);
+		return address != 0 && size != 0 && address < TRACKER_ADDRESS_SIZE &&
+		       size <= TRACKER_ADDRESS_SIZE - address;
 	}
 	[[nodiscard]] constexpr bool     ValidOrEmpty() const noexcept { return Empty() || Valid(); }
 	[[nodiscard]] constexpr uint64_t End() const noexcept { return address + size; }

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__)
@@ -30,14 +29,21 @@ class TrackingSpinLock final {
 public:
 	void lock() noexcept {
 		const auto thread = CurrentThread();
+		if (m_owner.load(std::memory_order_relaxed) == thread) {
+			EXIT("recursive region tracking lock\n");
+		}
 		while (m_lock.test_and_set(std::memory_order_acquire)) {
-			EXIT_NOT_IMPLEMENTED(m_owner.load(std::memory_order_relaxed) == thread);
+			if (m_owner.load(std::memory_order_relaxed) == thread) {
+				EXIT("recursive region tracking lock while contended\n");
+			}
 			std::atomic_signal_fence(std::memory_order_seq_cst);
 		}
 		m_owner.store(thread, std::memory_order_relaxed);
 	}
 	void unlock() noexcept {
-		EXIT_NOT_IMPLEMENTED(m_owner.load(std::memory_order_relaxed) != CurrentThread());
+		if (m_owner.load(std::memory_order_relaxed) != CurrentThread()) {
+			EXIT("region tracking lock released by non-owner\n");
+		}
 		m_owner.store(0, std::memory_order_relaxed);
 		m_lock.clear(std::memory_order_release);
 	}
@@ -81,7 +87,8 @@ public:
 	template <DirtySource source>
 	[[nodiscard]] bool IsModified(uint64_t offset, uint64_t size) const {
 		const auto [start, end] = GetPageRange(m_cpu_addr + offset, size);
-		return GetBits<source>().FirstRangeFrom(start).first < end;
+		const auto& bits        = GetBits<source>();
+		return RegionBits(bits, start, end).Any();
 	}
 
 	template <DirtySource source, bool enable>
@@ -104,43 +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>
@@ -171,6 +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);
 }
@@ -55,8 +61,7 @@ template <bool insert>
 void BufferCache::ChangeRegister(BufferId id) {
 	auto& buffer = m_slot_buffers[id];
 	PageTable::PageRange pages {};
-	EXIT_IF(!(GuestRange {buffer.CpuAddress(), buffer.Size()}.Valid()) ||
-	        !PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages));
+	EXIT_IF(!PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages));
 	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
 		if constexpr (insert) {
 			m_page_table[page] = id;
@@ -65,7 +70,6 @@ void BufferCache::ChangeRegister(BufferId id) {
 		}
 	}
 	const auto size_pages = pages.last_exclusive - pages.first;
-	const auto table_offset = PageIndex(buffer.CpuAddress()) * sizeof(vk::DeviceAddress);
 	if constexpr (insert) {
 		const auto [it, inserted] = m_buffers.emplace(buffer.CpuAddress(), id);
 		(void)it;
@@ -77,7 +81,7 @@ void BufferCache::ChangeRegister(BufferId id) {
 		for (uint64_t i = 0; i < size_pages; ++i) {
 			addresses.push_back(buffer.BufferDeviceAddress() + (i << CACHING_PAGEBITS));
 		}
-		WriteDataBuffer(m_bda_pagetable_buffer, table_offset,
+		WriteDataBuffer(m_bda_pagetable_buffer, pages.first * sizeof(vk::DeviceAddress),
 		                addresses.data(), addresses.size() * sizeof(vk::DeviceAddress));
 	} else {
 		const auto found = m_buffers.find(buffer.CpuAddress());
@@ -86,7 +90,7 @@ void BufferCache::ChangeRegister(BufferId id) {
 		EXIT_IF(buffer.Size() > m_total_used_memory);
 		m_total_used_memory -= buffer.Size();
 		m_lru_cache.Free(buffer.lru_id);
-		m_bda_pagetable_buffer.Fill(table_offset,
+		m_bda_pagetable_buffer.Fill(pages.first * sizeof(vk::DeviceAddress),
 		                            size_pages * sizeof(vk::DeviceAddress), 0);
 		buffer.is_deleted = true;
 	}
@@ -99,7 +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);
@@ -110,84 +115,79 @@ void BufferCache::DeleteBuffer(BufferId id) {
 	}
 }
 
-template <bool async>
-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");
 	}
-
-	auto [mapped, offset] = m_download_buffer.Map(total_size, 64);
-	std::unique_ptr<Buffer> temporary;
-	if (mapped == nullptr) {
-		temporary = std::make_unique<Buffer>(m_graphics, m_scheduler, MemoryUsage::Download, 0,
-		                                     vk::BufferUsageFlagBits::eTransferDst, total_size);
-		mapped = temporary->Mapped().data();
-	} else {
-		m_download_buffer.Commit();
+	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");
 	}
-	const auto& download = temporary ? *temporary : m_download_buffer;
-	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(), download.Handle(),
-	                  static_cast<uint32_t>(copies.size()), copies.data());
-
-	auto after          = before;
-	after.srcAccessMask = vk::AccessFlagBits::eTransferWrite;
-	after.dstAccessMask = vk::AccessFlagBits::eHostRead;
-	after.buffer        = download.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);
-	auto publish = [this, mapped, offset, total_size, buffer_address,
-	                copies = std::move(copies), owner = std::move(temporary)] {
-		(owner ? *owner : 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);
 		}
+		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;
 	};
-	if constexpr (async) {
-		m_scheduler.DeferPriorityOperation(std::move(publish));
-	} else {
-		const auto tick = m_scheduler.CurrentTick();
-		m_scheduler.Wait(tick);
-		m_scheduler.WaitPriorityOperations(tick);
-		publish();
+	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);
 	}
-	return true;
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
@@ -199,7 +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()));
@@ -252,26 +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<false>(buffer, window_begin, window_end - window_begin)) {
-			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) {
@@ -322,17 +340,13 @@ 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, (vaddr < LOWER_ADDRESS_SIZE ? LOWER_ADDRESS_SIZE
-				                                       : LibKernel::Memory::kExtendedMemoryBase +
-				                                             LibKernel::Memory::kExtendedMemorySize) - end);
+				end += std::min(StreamLeapSize, PageTable::kAddressSpaceSize - end);
 			}
 			if (expands_right) {
-				const auto minimum = vaddr < LOWER_ADDRESS_SIZE
-				                         ? CACHING_PAGESIZE * 2
-				                         : LibKernel::Memory::kExtendedMemoryBase;
+				const auto minimum = CACHING_PAGESIZE * 2;
 				if (begin > minimum) {
 					begin -= std::min(StreamLeapSize, begin - minimum);
 				}
@@ -357,8 +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);
 
@@ -465,23 +479,23 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBuffer(uint64_t vaddr, uint64_t
 		const auto alignment = std::max<uint64_t>(
 		    m_graphics.physical_device_properties.limits.minUniformBufferOffsetAlignment, 1);
 		auto [mapped, offset] = m_stream_buffer.Map(size, alignment, false);
-		if (mapped != nullptr) {
-			std::memcpy(mapped, reinterpret_cast<const void*>(vaddr), size);
+		if (mapped != nullptr && Libs::LibKernel::Memory::TryReadBacking(vaddr, mapped, size)) {
 			m_stream_buffer.Commit();
 			return {&m_stream_buffer, offset};
 		}
 	}
 
-	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) {
@@ -502,7 +516,8 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, u
 	}
 
 	auto [staging, stage_offset] = m_staging_buffer.Map(size, 16);
-	if (staging == nullptr || !Libs::LibKernel::Memory::TryReadSparseBacking(vaddr, staging, size)) {
+	if (staging == nullptr || (!Libs::LibKernel::Memory::TryReadBacking(vaddr, staging, size) &&
+	                           !Libs::LibKernel::Memory::TryReadPrtBacking(vaddr, staging, size))) {
 		EXIT("BufferCache: failed to read mapped guest image backing\n");
 	}
 	m_staging_buffer.Commit();
@@ -532,7 +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);
 }
 
@@ -566,6 +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);
 }
 
@@ -609,7 +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);
@@ -620,7 +642,19 @@ void BufferCache::RunGarbageCollector() {
 			return false;
 		}
 		if (dirty) {
-			EXIT_NOT_IMPLEMENTED(!DownloadBufferMemory<true>(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());
@@ -632,10 +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

@@ -29,22 +29,10 @@ class BufferCache {
 public:
 	static constexpr uint32_t CACHING_PAGEBITS  = 14;
 	static constexpr uint64_t CACHING_PAGESIZE  = uint64_t {1} << CACHING_PAGEBITS;
-	static constexpr uint64_t CACHING_NUMPAGES  = (LOWER_ADDRESS_SIZE + LibKernel::Memory::kExtendedMemorySize) >> CACHING_PAGEBITS;
+	static constexpr uint64_t CACHING_NUMPAGES  = uint64_t {1} << (40 - CACHING_PAGEBITS);
 	static constexpr uint64_t BDA_PAGETABLE_SIZE =
 	    CACHING_NUMPAGES * sizeof(vk::DeviceAddress);
 
-	static constexpr uint64_t PageIndex(uint64_t address) {
-		return (address < LOWER_ADDRESS_SIZE
-		            ? address
-		            : address - LibKernel::Memory::kExtendedMemoryBase + LOWER_ADDRESS_SIZE) >>
-		       CACHING_PAGEBITS;
-	}
-	static constexpr uint64_t GuestAddress(uint64_t offset) {
-		return offset < LOWER_ADDRESS_SIZE
-		           ? offset
-		           : offset - LOWER_ADDRESS_SIZE + LibKernel::Memory::kExtendedMemoryBase;
-	}
-
 	BufferCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
 	            TextureCache& texture_cache);
 	~BufferCache();
@@ -86,11 +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;
@@ -100,8 +83,14 @@ private:
 		bool                has_stream_leap;
 	};
 
-	using PageTable = MultiLevelPageTable<BufferId, CACHING_PAGEBITS, 44, 20>;
+	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);
@@ -117,9 +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);
-	// Synchronous downloads publish before returning; asynchronous callers wait before reuse.
-	template <bool async>
-	[[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;
@@ -134,9 +139,8 @@ void FaultManager::ProcessFaultBuffer() {
 		const auto* faults = std::bit_cast<const uint64_t*>(mapped);
 		const auto  count  = static_cast<uint32_t>(faults[0]);
 		for (uint32_t index = 1; index <= count; ++index) {
-			const auto address = BufferCache::GuestAddress(faults[index]);
-			fault_ranges.Add(address, BufferCache::CACHING_PAGESIZE);
-			LOGF("Accessed non-GPU cached memory at 0x%016" PRIx64 "\n", address);
+			fault_ranges.Add(faults[index], BufferCache::CACHING_PAGESIZE);
+			LOGF("Accessed non-GPU cached memory at 0x%016" PRIx64 "\n", faults[index]);
 		}
 		fault_ranges.ForEach([this](uint64_t start, uint64_t end) {
 			EXIT_IF(end - start > std::numeric_limits<uint32_t>::max());

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/samplerCache.cpp

@@ -14,10 +14,10 @@ SamplerCache::~SamplerCache() {
 	}
 }
 
-vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r, bool integer_border) {
+vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r) {
 	Common::LockGuard lock(m_mutex);
 
-	const SamplerKey key {r.fields[0], r.fields[1], r.fields[2], r.fields[3], integer_border};
+	const SamplerKey key {r.fields[0], r.fields[1], r.fields[2], r.fields[3]};
 	if (auto iter = m_samplers.find(key); iter != m_samplers.end()) {
 		return iter->second;
 	}
@@ -55,9 +55,11 @@ vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r, bool intege
 			case Prospero::SamplerAnisoRatio::kTwo: aniso_ratio = 2.0f; break;
 			case Prospero::SamplerAnisoRatio::kFour: aniso_ratio = 4.0f; break;
 			case Prospero::SamplerAnisoRatio::kEight: aniso_ratio = 8.0f; break;
-			case Prospero::SamplerAnisoRatio::kSixteen:
-			// Reserved encodings 5-7 fall back to the maximum ratio.
-			default: aniso_ratio = 16.0f; break;
+			case Prospero::SamplerAnisoRatio::kSixteen: aniso_ratio = 16.0f; break;
+			default:
+				EXIT("unknown ratio: %d dwords=%08x,%08x,%08x,%08x\n",
+				     static_cast<int>(r.MaxAnisoRatio()), r.fields[0], r.fields[1], r.fields[2],
+				     r.fields[3]);
 		}
 	}
 
@@ -93,23 +95,23 @@ vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r, bool intege
 		return vk::SamplerAddressMode::eClampToBorder;
 	};
 
-	vk::BorderColor border = integer_border ? vk::BorderColor::eIntTransparentBlack
-	                                       : vk::BorderColor::eFloatTransparentBlack;
+	vk::BorderColor border = vk::BorderColor::eIntTransparentBlack;
 	switch (static_cast<Prospero::SamplerBorderColor>(r.BorderColorType())) {
-		case Prospero::SamplerBorderColor::kTransBlack: break;
+		case Prospero::SamplerBorderColor::kTransBlack:
+			border = vk::BorderColor::eIntTransparentBlack;
+			break;
 		case Prospero::SamplerBorderColor::kOpaqueBlack:
-			border = integer_border ? vk::BorderColor::eIntOpaqueBlack
-			                        : vk::BorderColor::eFloatOpaqueBlack;
+			border = vk::BorderColor::eIntOpaqueBlack;
 			break;
 		case Prospero::SamplerBorderColor::kOpaqueWhite:
-			border = integer_border ? vk::BorderColor::eIntOpaqueWhite
-			                        : vk::BorderColor::eFloatOpaqueWhite;
+			border = vk::BorderColor::eIntOpaqueWhite;
 			break;
 		case Prospero::SamplerBorderColor::kFromTable:
 			LOGF(
 			    "temporary: approximating table border color as transparent black, index = %" PRIu16
 			    "\n",
 			    r.BorderColorPtr());
+			border = vk::BorderColor::eIntTransparentBlack;
 			break;
 		default: EXIT("unknown border color: %d", static_cast<int>(r.BorderColorType()));
 	}
@@ -123,11 +125,8 @@ vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r, bool intege
 	sampler_info.addressModeU = to_vk_address_mode(r.ClampX());
 	sampler_info.addressModeV = to_vk_address_mode(r.ClampY());
 	sampler_info.addressModeW = to_vk_address_mode(r.ClampZ());
-    const auto max_lod_bias = m_graphics.GetPhysicalDeviceProperties().limits.maxSamplerLodBias;
-    sampler_info.mipLodBias = std::clamp(
-        static_cast<float>(static_cast<int16_t>((r.LodBias() ^ 0x2000u) - 0x2000u)) / 256.0f,
-        -max_lod_bias, max_lod_bias
-    );
+	sampler_info.mipLodBias =
+	    static_cast<float>(static_cast<int16_t>((r.LodBias() ^ 0x2000u) - 0x2000u)) / 256.0f;
 	sampler_info.anisotropyEnable        = (aniso ? VK_TRUE : VK_FALSE);
 	sampler_info.maxAnisotropy           = aniso_ratio;
 	sampler_info.compareEnable           = (r.DepthCompareFunc() != 0 ? VK_TRUE : VK_FALSE);

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

@@ -24,10 +24,10 @@ public:
 	~SamplerCache();
 	KYTY_CLASS_NO_COPY(SamplerCache);
 
-	vk::Sampler GetSampler(const ShaderSamplerResource& r, bool integer_border);
+	vk::Sampler GetSampler(const ShaderSamplerResource& r);
 
 private:
-	using SamplerKey = std::array<uint32_t, 5>;
+	using SamplerKey = std::array<uint32_t, 4>;
 
 	struct SamplerKeyHash {
 		std::size_t operator()(const SamplerKey& key) const {

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 {
@@ -91,24 +100,7 @@ private:
 	};
 
 	using ImageIds       = InlinePageOwnerList<ImageId, 16>;
-	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 44, 14>;
-
-	// Callers have validated the nonempty 44-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);
-			}
-		}
-	}
+	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
 
 	[[nodiscard]] ImageId     InsertImage(const ImageInfo& info);
 	[[nodiscard]] ImageId     GetNullImage(const ImageDesc& desc);
@@ -128,6 +120,7 @@ private:
 	[[nodiscard]] static bool SameBacking(const ImageInfo& cached, const ImageInfo& requested,
 	                                      bool exact_format);
 	[[nodiscard]] static BindingType UploadBinding(const Image& image);
+	[[nodiscard]] bool               SafeToDownload(const Image& image);
 
 	// Caller holds m_lock; it also serializes the per-image query epoch.
 	[[nodiscard]] ImageIds      FindImagesInRegion(uint64_t address, uint64_t size,
@@ -138,30 +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

@@ -17,10 +17,11 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
-#include <bit>
 
 namespace Libs::Graphics {
 
+static std::atomic<uint32_t> g_render_color_log_count = 0;
+
 static bool DccAlphaOnMsb(const HW::ColorInfo& info) {
 	switch (info.format) {
 		case Prospero::ChannelLayout::k10_10_10_2:
@@ -137,7 +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;
@@ -158,6 +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");
 	}
@@ -198,7 +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);
@@ -226,26 +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");
@@ -260,7 +277,21 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 		EXIT("render-target backing range is invalid\n");
 	}
 
-	auto& desc = r.desc;
+	const vk::Extent2D view_extent = {std::max(width >> rt.view.current_mip_level, 1u),
+	                                  std::max(height >> rt.view.current_mip_level, 1u)};
+
+	auto decision_log_id = g_render_color_log_count.fetch_add(1);
+	if (decision_log_id < 128) {
+		LOGF("RenderColorTarget: slot=%" PRIu32 " addr=0x%010" PRIx64 " size=0x%016" PRIx64
+		     " extent=%ux%ux%u view_mip=%u view_extent=%ux%u levels=%u pitch=%u"
+		     " fmt=0x%08" PRIx32 " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 " samples=%u tile=%s\n",
+		     rt_slot, rt.base.addr, backing_size, width, height, depth, rt.view.current_mip_level,
+		     view_extent.width, view_extent.height, levels, pitch,
+		     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+		     static_cast<uint32_t>(rt.info.channel_order), samples, tile ? "tiled" : "linear");
+	}
+
+	TextureCache::ImageDesc desc {};
 	desc.type              = TextureCache::BindingType::RenderTarget;
 	desc.info.data         = {rt.base.addr, backing_size};
 	desc.info.pixel_format = target_format.format;
@@ -273,25 +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) {
@@ -332,6 +353,7 @@ void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColor
 	desc.view_info.layer_count = view.layer_count;
 	desc.view_info.usage       = vk::ImageUsageFlagBits::eColorAttachment;
 	auto& texture_cache        = m_context.GetTextureCache();
+	r.desc                     = std::move(desc);
 	r.guest_mip_level          = rt.view.current_mip_level;
 	r.guest_array_layer        = view.base_layer;
 	r.image_id                 = texture_cache.FindImage(r.desc, exact_format);

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

@@ -1,7 +1,6 @@
 #include "graphics/host_gpu/renderer/image/image.h"
 
 #include "common/assert.h"
-#include "common/logging/log.h"
 #include "common/profiler.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
@@ -11,9 +10,7 @@
 
 #include <algorithm>
 #include <array>
-#include <atomic>
 #include <cstdint>
-#include <fmt/format.h>
 #include <xxhash.h>
 
 namespace Libs::Graphics {
@@ -51,29 +48,12 @@ namespace {
 }
 
 [[nodiscard]] vk::ImageUsageFlags ImageUsageFlags(GraphicContext& graphics, const ImageInfo& info) {
-	auto usage = vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst;
 	if (info.IsBlock()) {
-		usage |= vk::ImageUsageFlagBits::eSampled;
-		if (graphics.supports_block_texel_view) {
-			const auto storage = usage | vk::ImageUsageFlagBits::eStorage;
-			if (graphics.GetImageFormatProperties(info.pixel_format, HostImageType(info.type),
-			                                      vk::ImageTiling::eOptimal, storage,
-			                                      ImageCreateFlags(graphics, info),
-			                                      nullptr) == vk::Result::eSuccess) {
-				usage = storage;
-			} else {
-				static std::atomic_flag warned = ATOMIC_FLAG_INIT;
-				if (!warned.test_and_set(std::memory_order_relaxed)) {
-					Log::WriteToConsoleAndLog(fmt::format(
-					    "Warning: format {} does not support storage access; block-compressed "
-					    "textures written by the guest will not render.\n",
-					    vk::to_string(info.pixel_format)));
-				}
-			}
-		}
-		return usage;
+		return vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst |
+		       vk::ImageUsageFlagBits::eSampled;
 	}
 	const auto properties = graphics.GetFormatProperties(info.pixel_format);
+	auto       usage = vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst;
 	if (HasFormatFeature(properties, vk::FormatFeatureFlagBits::eSampledImage)) {
 		usage |= vk::ImageUsageFlagBits::eSampled;
 	}
@@ -647,11 +627,12 @@ void Validate(const ImageInfo& info) {
 			}
 			break;
 		case ImageMetadataKind::Dcc:
-		case ImageMetadataKind::Cmask:
-			if (!GuestRange {info.metadata.range.address,
-			                 std::max<uint64_t>(info.metadata.range.size, 1)}.Valid() ||
+			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;
 	}
@@ -709,21 +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 {};
@@ -428,20 +483,6 @@ inline constexpr std::array<VideoOutFormatPolicy, 7> VIDEO_OUT_FORMAT_POLICIES {
 	return true;
 }
 
-// Unlike a register clear, a DWORD fill repeats the same word across the entire pixel.
-// Keep this separate: the first register word alone cannot describe a 64-bit color.
-[[nodiscard]] inline bool DecodeColorDwordFill(vk::Format format, uint32_t packed,
-                                               vk::ClearColorValue& clear) {
-	if (format == vk::Format::eR16G16B16A16Sfloat) {
-		if (packed != 0) {
-			return false;
-		}
-		clear = vk::ClearColorValue {};
-		return true;
-	}
-	return DecodePackedColorClear(format, packed, clear);
-}
-
 [[nodiscard]] inline bool DecodePackedStencilClear(uint32_t packed, uint8_t& clear) {
 	const auto value = static_cast<uint8_t>(packed);
 	if (packed != static_cast<uint32_t>(value) * 0x01010101u) {

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/imageView.h

@@ -31,14 +31,20 @@ namespace ImageViewOps {
 	       ImageViewOps::IsFormatDepthCompatible(view_format);
 }
 
-[[nodiscard]] inline bool IsValidImageSwizzle(uint32_t swizzle,
-                                             uint32_t component_count = 4) noexcept {
+[[nodiscard]] inline bool IsValidImageSwizzle(uint32_t swizzle) noexcept {
 	if ((swizzle & ~0xfffu) != 0) {
 		return false;
 	}
 	for (uint32_t channel = 0; channel < 4; channel++) {
-		const auto selector = GetDstSel(swizzle, channel);
-		if (selector > 1 && (selector < 4 || selector >= 4 + component_count)) return false;
+		switch (GetDstSel(swizzle, channel)) {
+			case 0:
+			case 1:
+			case 4:
+			case 5:
+			case 6:
+			case 7: break;
+			default: return false;
+		}
 	}
 	return true;
 }
@@ -75,8 +81,15 @@ SelectSampledColorView(vk::Format image_format, vk::Format view_format, uint32_t
 [[nodiscard]] inline bool IsSupportedSampledDepthView(vk::Format image_format,
                                                       vk::Format view_format,
                                                       uint32_t   swizzle) noexcept {
-	return IsSupportedSampledDepthFormat(image_format, view_format) &&
-	       IsValidImageSwizzle(swizzle, 1);
+	if (!IsSupportedSampledDepthFormat(image_format, view_format)) {
+		return false;
+	}
+	switch (swizzle) {
+		case DstSel(4, 4, 4, 4):
+		case DstSel(4, 0, 0, 0):
+		case DstSel(4, 0, 0, 1): return true;
+		default: return false;
+	}
 }
 
 [[nodiscard]] inline bool

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"
@@ -88,6 +87,30 @@ TileManager::~TileManager() {
 	}
 }
 
+TileManager::Scratch TileManager::AllocateScratch(uint64_t size) {
+	EXIT_IF(size == 0);
+	vk::BufferCreateInfo create {};
+	create.size  = size;
+	create.usage = vk::BufferUsageFlagBits::eStorageBuffer | vk::BufferUsageFlagBits::eTransferSrc |
+	               vk::BufferUsageFlagBits::eTransferDst;
+
+	VmaAllocationCreateInfo allocate {};
+	allocate.usage       = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
+	VkBuffer      buffer = VK_NULL_HANDLE;
+	VmaAllocation memory = nullptr;
+	const auto    raw    = static_cast<VkBufferCreateInfo>(create);
+	RequireVulkanSuccess(static_cast<vk::Result>(vmaCreateBuffer(
+	                         m_graphics.allocator, &raw, &allocate, &buffer, &memory, nullptr)),
+	                     "allocate TileManager scratch buffer");
+	return {buffer, memory, size};
+}
+
+void TileManager::DeferDestroy(Scratch scratch) {
+	auto allocator = m_graphics.allocator;
+	m_scheduler.DeferOperation(
+	    [allocator, scratch] { vmaDestroyBuffer(allocator, scratch.buffer, scratch.allocation); });
+}
+
 void TileManager::Prepare(bool tile, uint64_t tiled_capacity, uint64_t linear_capacity,
                           std::span<const GpuTileInfo> infos, uint64_t source_base,
                           uint64_t target_base, std::vector<Dispatch>& dispatches) {
@@ -198,7 +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);
@@ -236,8 +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;
@@ -260,12 +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);
@@ -345,7 +372,8 @@ 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 = GetScratchBuffer(linear_capacity, tiled);
+	auto scratch = AllocateScratch((linear_capacity + 3u) & ~uint64_t {3});
+	DeferDestroy(scratch);
 	Record(tiled, tiled_offset, tiled_capacity, scratch.buffer, 0, scratch.size, dispatches,
 	       true);
 	return {scratch.buffer, 0, linear_capacity};
@@ -375,9 +403,11 @@ void TileManager::TileImage(Image& image, std::span<const vk::BufferImageCopy> r
 	    std::max<uint64_t>(limits.minStorageBufferOffsetAlignment, 4);
 	const uint64_t        target_base = tiled_offset & (descriptor_alignment - 1);
 	std::vector<Dispatch> dispatches;
-	// Reserve stream parameters before recording the image download and conversion.
+	// 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 = GetScratchBuffer(linear_capacity, tiled);
+	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};
 	if (transform == ColorTransform::SwapBgra16) {
@@ -387,29 +417,20 @@ void TileManager::TileImage(Image& image, std::span<const vk::BufferImageCopy> r
 	       dispatches, false);
 }
 
-TileManager::Result TileManager::GetScratchBuffer(uint64_t size, vk::Buffer input) {
-	EXIT_IF(size == 0);
-	size = Common::AlignUp(size, 4);
-	auto& buffer = m_scratch[m_scratch[0] && m_scratch[0]->Handle() == input ? 1 : 0];
-	if (!buffer || buffer->Size() < size) {
-		if (buffer) {
-			m_scheduler.DeferOperation([old = std::move(buffer)]() mutable { old.reset(); });
-		}
-		buffer = std::make_unique<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, 0,
-		    vk::BufferUsageFlagBits::eStorageBuffer | vk::BufferUsageFlagBits::eTransferSrc |
-		        vk::BufferUsageFlagBits::eTransferDst, size);
-	}
-	return {buffer->Handle(), 0, size};
+TileManager::Result TileManager::GetScratchBuffer(uint64_t size) {
+	auto scratch = AllocateScratch((size + 3u) & ~uint64_t {3});
+	DeferDestroy(scratch);
+	return {scratch.buffer, 0, scratch.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)};
 }
@@ -450,7 +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;
@@ -486,8 +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();
@@ -584,7 +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;
@@ -650,7 +681,9 @@ void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 
 TileManager::Result TileManager::SwapBgra16(Result input) {
 	EXIT_NOT_IMPLEMENTED(input.size == 0 || input.size % 8u != 0 || input.size / 8u > UINT32_MAX);
-	auto result = GetScratchBuffer(input.size, input.buffer);
+	auto output = AllocateScratch(input.size);
+	DeferDestroy(output);
+	Result result {output.buffer, 0, output.size};
 	SwapBgra16(input, result, static_cast<uint32_t>(input.size / 8u));
 	return result;
 }

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

@@ -6,14 +6,13 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <array>
-#include <memory>
 #include <span>
 #include <vector>
+#include <vk_mem_alloc.h>
 
 namespace Libs::Graphics {
 
 class CommandScheduler;
-class Buffer;
 class Image;
 class StreamBuffer;
 struct GraphicContext;
@@ -63,7 +62,7 @@ public:
 	~TileManager();
 	KYTY_CLASS_NO_COPY(TileManager);
 
-	// Consume scratch results before the next acquisition, or pass their buffer as input.
+	// The returned device-local buffer remains alive through the current scheduler tick.
 	[[nodiscard]] Result Detile(vk::Buffer tiled, uint64_t tiled_offset, uint64_t tiled_capacity,
 	                            uint64_t linear_capacity, std::span<const GpuTileInfo> infos);
 	void Tile(vk::Buffer linear, uint64_t linear_offset, uint64_t linear_capacity, vk::Buffer tiled,
@@ -72,7 +71,7 @@ public:
 	               uint64_t tiled_offset, uint64_t tiled_capacity, uint64_t linear_capacity,
 	               std::span<const GpuTileInfo> infos,
 	               ColorTransform               transform = ColorTransform::None);
-	[[nodiscard]] Result GetScratchBuffer(uint64_t size, vk::Buffer input = nullptr);
+	[[nodiscard]] Result GetScratchBuffer(uint64_t size);
 	void                 ConvertD16(Result source, Result target, D16Direction direction, bool d32,
 	                                const D16Layout& layout);
 	[[nodiscard]] Result SwapBgra16(Result input);
@@ -106,16 +105,23 @@ private:
 		uint32_t pipeline_slot = 0;
 		uint64_t params_offset = 0;
 	};
+	struct Scratch {
+		vk::Buffer    buffer     = nullptr;
+		VmaAllocation allocation = nullptr;
+		uint64_t      size       = 0;
+	};
 	struct StorageBinding {
 		vk::DescriptorBufferInfo info;
 		uint32_t                 base = 0;
 	};
 
+	[[nodiscard]] Scratch         AllocateScratch(uint64_t size);
 	[[nodiscard]] StorageBinding  BindStorage(Result buffer, uint64_t size) const;
 	[[nodiscard]] static uint32_t ConversionRows(uint64_t offset, uint64_t row_stride,
 	                                             uint64_t active, uint32_t remaining,
 	                                             uint64_t alignment, uint64_t max_range,
 	                                             uint32_t max_groups) noexcept;
+	void                          DeferDestroy(Scratch scratch);
 	void Prepare(bool tile, uint64_t tiled_capacity, uint64_t linear_capacity,
 	             std::span<const GpuTileInfo> infos, uint64_t source_base, uint64_t target_base,
 	             std::vector<Dispatch>& dispatches);
@@ -128,7 +134,6 @@ private:
 	GraphicContext&                         m_graphics;
 	CommandScheduler&                       m_scheduler;
 	StreamBuffer&                           m_stream_buffer;
-	std::array<std::unique_ptr<Buffer>, 2>   m_scratch;
 	vk::DescriptorSetLayout                 m_descriptor_layout = nullptr;
 	vk::PipelineLayout                      m_pipeline_layout   = nullptr;
 	std::array<vk::Pipeline, PipelineCount> m_pipelines {};

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

@@ -1,58 +0,0 @@
-#include "graphics/host_gpu/renderer/pipeline/blendMapping.h"
-
-#include "graphics/guest_gpu/hardwareContext.h"
-
-#include <initializer_list>
-
-namespace Libs::Graphics {
-
-bool BlendFactorIsDualSource(uint8_t factor) {
-	return factor >= static_cast<uint8_t>(Prospero::BlendFactor::kSrc1Color) &&
-	       factor <= static_cast<uint8_t>(Prospero::BlendFactor::kOneMinusSrc1Alpha);
-}
-
-namespace {
-
-bool BlendFactorIsConstantColor(uint8_t factor) {
-	return factor == static_cast<uint8_t>(Prospero::BlendFactor::kConstantColor) ||
-	       factor == static_cast<uint8_t>(Prospero::BlendFactor::kOneMinusConstantColor);
-}
-
-} // namespace
-
-BlendMappingSupport ClassifyBlendMapping(const HW::BlendControl&                blend,
-                                         const Prospero::ColorComponentMapping& mapping) {
-	// Color constants are not swizzled with the exports; scalar constant alpha is unaffected.
-	if (!mapping.IsIdentity() && (BlendFactorIsConstantColor(blend.color_srcblend) ||
-	                              BlendFactorIsConstantColor(blend.color_destblend))) {
-		return BlendMappingSupport::Unsupported;
-	}
-	if (mapping.Map(3) == 3u) {
-		return BlendMappingSupport::Direct;
-	}
-	// Moving alpha requires the same equation for all channels.
-	if (blend.separate_alpha_blend && (blend.alpha_srcblend != blend.color_srcblend ||
-	                                   blend.alpha_destblend != blend.color_destblend ||
-	                                   blend.alpha_comb_fcn != blend.color_comb_fcn)) {
-		return BlendMappingSupport::Unsupported;
-	}
-	auto support = BlendMappingSupport::Direct;
-	for (const auto factor: {blend.color_srcblend, blend.color_destblend}) {
-		if (BlendFactorIsDualSource(factor)) {
-			return BlendMappingSupport::Unsupported;
-		}
-		switch (static_cast<Prospero::BlendFactor>(factor)) {
-			case Prospero::BlendFactor::kSrcAlpha:
-			case Prospero::BlendFactor::kOneMinusSrcAlpha:
-				support = BlendMappingSupport::SourceAlpha;
-				break;
-			case Prospero::BlendFactor::kDstAlpha:
-			case Prospero::BlendFactor::kOneMinusDstAlpha:
-			case Prospero::BlendFactor::kSrcAlphaSaturate: return BlendMappingSupport::Unsupported;
-			default: break;
-		}
-	}
-	return support;
-}
-
-} // namespace Libs::Graphics

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

@@ -1,27 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_PIPELINE_BLENDMAPPING_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_PIPELINE_BLENDMAPPING_H_
-
-#include <cstdint>
-
-namespace Libs::Graphics {
-
-namespace HW {
-struct BlendControl;
-}
-namespace Prospero {
-struct ColorComponentMapping;
-}
-
-enum class BlendMappingSupport {
-	Direct,
-	SourceAlpha, // Requires logical alpha in the second blend source.
-	Unsupported,
-};
-
-bool                BlendFactorIsDualSource(uint8_t factor);
-BlendMappingSupport ClassifyBlendMapping(const HW::BlendControl&                blend,
-                                         const Prospero::ColorComponentMapping& mapping);
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_PIPELINE_BLENDMAPPING_H_

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());